SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License version 2 as
  11  * published by the Free Software Foundation.
  12  */
  13
  14 #include <linux/types.h>
  15 #include <linux/netfilter.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/proc_fs.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/stddef.h>
  21 #include <linux/slab.h>
  22 #include <linux/random.h>
  23 #include <linux/jhash.h>
  24 #include <linux/err.h>
  25 #include <linux/percpu.h>
  26 #include <linux/moduleparam.h>
  27 #include <linux/notifier.h>
  28 #include <linux/kernel.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/socket.h>
  31 #include <linux/mm.h>
  32 #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         /* To make sure we don't get any weird locking issues here:
 187          * destroy_conntrack() MUST NOT be called with a write lock
 188          * to nf_conntrack_lock!!! -HW */
 189         rcu_read_lock();
 190         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 191         if (l4proto && l4proto->destroy)
 192                 l4proto->destroy(ct);
 193
 194         rcu_read_unlock();
 195
 196         spin_lock_bh(&nf_conntrack_lock);
 197         /* Expectations will have been removed in clean_from_lists,
 198          * except TFTP can create an expectation on the first packet,
 199          * before connection is in the list, so we need to clean here,
 200          * too. */
 201         nf_ct_remove_expectations(ct);
 202
 203         /* We overload first tuple to link into unconfirmed list. */
 204         if (!nf_ct_is_confirmed(ct)) {
 205                 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 206                 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 207         }
 208
 209         NF_CT_STAT_INC(net, delete);
 210         spin_unlock_bh(&nf_conntrack_lock);
 211
 212         if (ct->master)
 213                 nf_ct_put(ct->master);
 214
 215         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 216         nf_conntrack_free(ct);
 217 }
 218
 219 void nf_ct_delete_from_lists(struct nf_conn *ct)
 220 {
 221         struct net *net = nf_ct_net(ct);
 222
 223         nf_ct_helper_destroy(ct);
 224         spin_lock_bh(&nf_conntrack_lock);
 225         /* Inside lock so preempt is disabled on module removal path.
 226          * Otherwise we can get spurious warnings. */
 227         NF_CT_STAT_INC(net, delete_list);
 228         clean_from_lists(ct);
 229         spin_unlock_bh(&nf_conntrack_lock);
 230 }
 231 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
 232
 233 static void death_by_event(unsigned long ul_conntrack)
 234 {
 235         struct nf_conn *ct = (void *)ul_conntrack;
 236         struct net *net = nf_ct_net(ct);
 237
 238         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 239                 /* bad luck, let's retry again */
 240                 ct->timeout.expires = jiffies +
 241                         (random32() % net->ct.sysctl_events_retry_timeout);
 242                 add_timer(&ct->timeout);
 243                 return;
 244         }
 245         /* we've got the event delivered, now it's dying */
 246         set_bit(IPS_DYING_BIT, &ct->status);
 247         spin_lock(&nf_conntrack_lock);
 248         hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 249         spin_unlock(&nf_conntrack_lock);
 250         nf_ct_put(ct);
 251 }
 252
 253 void nf_ct_insert_dying_list(struct nf_conn *ct)
 254 {
 255         struct net *net = nf_ct_net(ct);
 256
 257         /* add this conntrack to the dying list */
 258         spin_lock_bh(&nf_conntrack_lock);
 259         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 260                              &net->ct.dying);
 261         spin_unlock_bh(&nf_conntrack_lock);
 262         /* set a new timer to retry event delivery */
 263         setup_timer(&ct->timeout, death_by_event, (unsigned long)ct);
 264         ct->timeout.expires = jiffies +
 265                 (random32() % net->ct.sysctl_events_retry_timeout);
 266         add_timer(&ct->timeout);
 267 }
 268 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list);
 269
 270 static void death_by_timeout(unsigned long ul_conntrack)
 271 {
 272         struct nf_conn *ct = (void *)ul_conntrack;
 273
 274         if (!test_bit(IPS_DYING_BIT, &ct->status) &&
 275             unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
 276                 /* destroy event was not delivered */
 277                 nf_ct_delete_from_lists(ct);
 278                 nf_ct_insert_dying_list(ct);
 279                 return;
 280         }
 281         set_bit(IPS_DYING_BIT, &ct->status);
 282         nf_ct_delete_from_lists(ct);
 283         nf_ct_put(ct);
 284 }
 285
 286 /*
 287  * Warning :
 288  * - Caller must take a reference on returned object
 289  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 290  * OR
 291  * - Caller must lock nf_conntrack_lock before calling this function
 292  */
 293 struct nf_conntrack_tuple_hash *
 294 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
 295 {
 296         struct nf_conntrack_tuple_hash *h;
 297         struct hlist_nulls_node *n;
 298         unsigned int hash = hash_conntrack(tuple);
 299
 300         /* Disable BHs the entire time since we normally need to disable them
 301          * at least once for the stats anyway.
 302          */
 303         local_bh_disable();
 304 begin:
 305         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 306                 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
 307                         NF_CT_STAT_INC(net, found);
 308                         local_bh_enable();
 309                         return h;
 310                 }
 311                 NF_CT_STAT_INC(net, searched);
 312         }
 313         /*
 314          * if the nulls value we got at the end of this lookup is
 315          * not the expected one, we must restart lookup.
 316          * We probably met an item that was moved to another chain.
 317          */
 318         if (get_nulls_value(n) != hash)
 319                 goto begin;
 320         local_bh_enable();
 321
 322         return NULL;
 323 }
 324 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 325
 326 /* Find a connection corresponding to a tuple. */
 327 struct nf_conntrack_tuple_hash *
 328 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
 329 {
 330         struct nf_conntrack_tuple_hash *h;
 331         struct nf_conn *ct;
 332
 333         rcu_read_lock();
 334 begin:
 335         h = __nf_conntrack_find(net, tuple);
 336         if (h) {
 337                 ct = nf_ct_tuplehash_to_ctrack(h);
 338                 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 339                         h = NULL;
 340                 else {
 341                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) {
 342                                 nf_ct_put(ct);
 343                                 goto begin;
 344                         }
 345                 }
 346         }
 347         rcu_read_unlock();
 348
 349         return h;
 350 }
 351 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 352
 353 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 354                                        unsigned int hash,
 355                                        unsigned int repl_hash)
 356 {
 357         struct net *net = nf_ct_net(ct);
 358
 359         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 360                            &net->ct.hash[hash]);
 361         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 362                            &net->ct.hash[repl_hash]);
 363 }
 364
 365 void nf_conntrack_hash_insert(struct nf_conn *ct)
 366 {
 367         unsigned int hash, repl_hash;
 368
 369         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 370         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 371
 372         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 373 }
 374 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 375
 376 /* Confirm a connection given skb; places it in hash table */
 377 int
 378 __nf_conntrack_confirm(struct sk_buff *skb)
 379 {
 380         unsigned int hash, repl_hash;
 381         struct nf_conntrack_tuple_hash *h;
 382         struct nf_conn *ct;
 383         struct nf_conn_help *help;
 384         struct hlist_nulls_node *n;
 385         enum ip_conntrack_info ctinfo;
 386         struct net *net;
 387
 388         ct = nf_ct_get(skb, &ctinfo);
 389         net = nf_ct_net(ct);
 390
 391         /* ipt_REJECT uses nf_conntrack_attach to attach related
 392            ICMP/TCP RST packets in other direction.  Actual packet
 393            which created connection will be IP_CT_NEW or for an
 394            expected connection, IP_CT_RELATED. */
 395         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 396                 return NF_ACCEPT;
 397
 398         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 399         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 400
 401         /* We're not in hash table, and we refuse to set up related
 402            connections for unconfirmed conns.  But packet copies and
 403            REJECT will give spurious warnings here. */
 404         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 405
 406         /* No external references means noone else could have
 407            confirmed us. */
 408         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 409         pr_debug("Confirming conntrack %p\n", ct);
 410
 411         spin_lock_bh(&nf_conntrack_lock);
 412
 413         /* See if there's one in the list already, including reverse:
 414            NAT could have grabbed it without realizing, since we're
 415            not in the hash.  If there is, we lost race. */
 416         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 417                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 418                                       &h->tuple))
 419                         goto out;
 420         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 421                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 422                                       &h->tuple))
 423                         goto out;
 424
 425         /* Remove from unconfirmed list */
 426         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 427
 428         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 429         /* Timer relative to confirmation time, not original
 430            setting time, otherwise we'd get timer wrap in
 431            weird delay cases. */
 432         ct->timeout.expires += jiffies;
 433         add_timer(&ct->timeout);
 434         atomic_inc(&ct->ct_general.use);
 435         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 436         NF_CT_STAT_INC(net, insert);
 437         spin_unlock_bh(&nf_conntrack_lock);
 438         help = nfct_help(ct);
 439         if (help && help->helper)
 440                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 441
 442         nf_conntrack_event_cache(master_ct(ct) ?
 443                                  IPCT_RELATED : IPCT_NEW, ct);
 444         return NF_ACCEPT;
 445
 446 out:
 447         NF_CT_STAT_INC(net, insert_failed);
 448         spin_unlock_bh(&nf_conntrack_lock);
 449         return NF_DROP;
 450 }
 451 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 452
 453 /* Returns true if a connection correspondings to the tuple (required
 454    for NAT). */
 455 int
 456 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 457                          const struct nf_conn *ignored_conntrack)
 458 {
 459         struct net *net = nf_ct_net(ignored_conntrack);
 460         struct nf_conntrack_tuple_hash *h;
 461         struct hlist_nulls_node *n;
 462         unsigned int hash = hash_conntrack(tuple);
 463
 464         /* Disable BHs the entire time since we need to disable them at
 465          * least once for the stats anyway.
 466          */
 467         rcu_read_lock_bh();
 468         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 469                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 470                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 471                         NF_CT_STAT_INC(net, found);
 472                         rcu_read_unlock_bh();
 473                         return 1;
 474                 }
 475                 NF_CT_STAT_INC(net, searched);
 476         }
 477         rcu_read_unlock_bh();
 478
 479         return 0;
 480 }
 481 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 482
 483 #define NF_CT_EVICTION_RANGE    8
 484
 485 /* There's a small race here where we may free a just-assured
 486    connection.  Too bad: we're in trouble anyway. */
 487 static noinline int early_drop(struct net *net, unsigned int hash)
 488 {
 489         /* Use oldest entry, which is roughly LRU */
 490         struct nf_conntrack_tuple_hash *h;
 491         struct nf_conn *ct = NULL, *tmp;
 492         struct hlist_nulls_node *n;
 493         unsigned int i, cnt = 0;
 494         int dropped = 0;
 495
 496         rcu_read_lock();
 497         for (i = 0; i < nf_conntrack_htable_size; i++) {
 498                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 499                                          hnnode) {
 500                         tmp = nf_ct_tuplehash_to_ctrack(h);
 501                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 502                                 ct = tmp;
 503                         cnt++;
 504                 }
 505
 506                 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 507                         ct = NULL;
 508                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
 509                         break;
 510                 hash = (hash + 1) % nf_conntrack_htable_size;
 511         }
 512         rcu_read_unlock();
 513
 514         if (!ct)
 515                 return dropped;
 516
 517         if (del_timer(&ct->timeout)) {
 518                 death_by_timeout((unsigned long)ct);
 519                 dropped = 1;
 520                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 521         }
 522         nf_ct_put(ct);
 523         return dropped;
 524 }
 525
 526 struct nf_conn *nf_conntrack_alloc(struct net *net,
 527                                    const struct nf_conntrack_tuple *orig,
 528                                    const struct nf_conntrack_tuple *repl,
 529                                    gfp_t gfp)
 530 {
 531         struct nf_conn *ct;
 532
 533         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 534                 get_random_bytes(&nf_conntrack_hash_rnd,
 535                                 sizeof(nf_conntrack_hash_rnd));
 536                 nf_conntrack_hash_rnd_initted = 1;
 537         }
 538
 539         /* We don't want any race condition at early drop stage */
 540         atomic_inc(&net->ct.count);
 541
 542         if (nf_conntrack_max &&
 543             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 544                 unsigned int hash = hash_conntrack(orig);
 545                 if (!early_drop(net, hash)) {
 546                         atomic_dec(&net->ct.count);
 547                         if (net_ratelimit())
 548                                 printk(KERN_WARNING
 549                                        "nf_conntrack: table full, dropping"
 550                                        " packet.\n");
 551                         return ERR_PTR(-ENOMEM);
 552                 }
 553         }
 554
 555         ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
 556         if (ct == NULL) {
 557                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
 558                 atomic_dec(&net->ct.count);
 559                 return ERR_PTR(-ENOMEM);
 560         }
 561
 562         spin_lock_init(&ct->lock);
 563         atomic_set(&ct->ct_general.use, 1);
 564         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 565         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 566         /* Don't set timer yet: wait for confirmation */
 567         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 568 #ifdef CONFIG_NET_NS
 569         ct->ct_net = net;
 570 #endif
 571
 572         return ct;
 573 }
 574 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 575
 576 void nf_conntrack_free(struct nf_conn *ct)
 577 {
 578         struct net *net = nf_ct_net(ct);
 579
 580         nf_ct_ext_destroy(ct);
 581         atomic_dec(&net->ct.count);
 582         nf_ct_ext_free(ct);
 583         kmem_cache_free(nf_conntrack_cachep, ct);
 584 }
 585 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 586
 587 /* Allocate a new conntrack: we return -ENOMEM if classification
 588    failed due to stress.  Otherwise it really is unclassifiable. */
 589 static struct nf_conntrack_tuple_hash *
 590 init_conntrack(struct net *net,
 591                const struct nf_conntrack_tuple *tuple,
 592                struct nf_conntrack_l3proto *l3proto,
 593                struct nf_conntrack_l4proto *l4proto,
 594                struct sk_buff *skb,
 595                unsigned int dataoff)
 596 {
 597         struct nf_conn *ct;
 598         struct nf_conn_help *help;
 599         struct nf_conntrack_tuple repl_tuple;
 600         struct nf_conntrack_expect *exp;
 601
 602         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 603                 pr_debug("Can't invert tuple.\n");
 604                 return NULL;
 605         }
 606
 607         ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
 608         if (IS_ERR(ct)) {
 609                 pr_debug("Can't allocate conntrack.\n");
 610                 return (struct nf_conntrack_tuple_hash *)ct;
 611         }
 612
 613         if (!l4proto->new(ct, skb, dataoff)) {
 614                 nf_conntrack_free(ct);
 615                 pr_debug("init conntrack: can't track with proto module\n");
 616                 return NULL;
 617         }
 618
 619         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 620         nf_ct_ecache_ext_add(ct, GFP_ATOMIC);
 621
 622         spin_lock_bh(&nf_conntrack_lock);
 623         exp = nf_ct_find_expectation(net, tuple);
 624         if (exp) {
 625                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 626                          ct, exp);
 627                 /* Welcome, Mr. Bond.  We've been expecting you... */
 628                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 629                 ct->master = exp->master;
 630                 if (exp->helper) {
 631                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 632                         if (help)
 633                                 rcu_assign_pointer(help->helper, exp->helper);
 634                 }
 635
 636 #ifdef CONFIG_NF_CONNTRACK_MARK
 637                 ct->mark = exp->master->mark;
 638 #endif
 639 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 640                 ct->secmark = exp->master->secmark;
 641 #endif
 642                 nf_conntrack_get(&ct->master->ct_general);
 643                 NF_CT_STAT_INC(net, expect_new);
 644         } else {
 645                 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
 646                 NF_CT_STAT_INC(net, new);
 647         }
 648
 649         /* Overload tuple linked list to put us in unconfirmed list. */
 650         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 651                        &net->ct.unconfirmed);
 652
 653         spin_unlock_bh(&nf_conntrack_lock);
 654
 655         if (exp) {
 656                 if (exp->expectfn)
 657                         exp->expectfn(ct, exp);
 658                 nf_ct_expect_put(exp);
 659         }
 660
 661         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 662 }
 663
 664 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 665 static inline struct nf_conn *
 666 resolve_normal_ct(struct net *net,
 667                   struct sk_buff *skb,
 668                   unsigned int dataoff,
 669                   u_int16_t l3num,
 670                   u_int8_t protonum,
 671                   struct nf_conntrack_l3proto *l3proto,
 672                   struct nf_conntrack_l4proto *l4proto,
 673                   int *set_reply,
 674                   enum ip_conntrack_info *ctinfo)
 675 {
 676         struct nf_conntrack_tuple tuple;
 677         struct nf_conntrack_tuple_hash *h;
 678         struct nf_conn *ct;
 679
 680         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 681                              dataoff, l3num, protonum, &tuple, l3proto,
 682                              l4proto)) {
 683                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 684                 return NULL;
 685         }
 686
 687         /* look for tuple match */
 688         h = nf_conntrack_find_get(net, &tuple);
 689         if (!h) {
 690                 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
 691                 if (!h)
 692                         return NULL;
 693                 if (IS_ERR(h))
 694                         return (void *)h;
 695         }
 696         ct = nf_ct_tuplehash_to_ctrack(h);
 697
 698         /* It exists; we have (non-exclusive) reference. */
 699         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 700                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 701                 /* Please set reply bit if this packet OK */
 702                 *set_reply = 1;
 703         } else {
 704                 /* Once we've had two way comms, always ESTABLISHED. */
 705                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 706                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 707                         *ctinfo = IP_CT_ESTABLISHED;
 708                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 709                         pr_debug("nf_conntrack_in: related packet for %p\n",
 710                                  ct);
 711                         *ctinfo = IP_CT_RELATED;
 712                 } else {
 713                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 714                         *ctinfo = IP_CT_NEW;
 715                 }
 716                 *set_reply = 0;
 717         }
 718         skb->nfct = &ct->ct_general;
 719         skb->nfctinfo = *ctinfo;
 720         return ct;
 721 }
 722
 723 unsigned int
 724 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 725                 struct sk_buff *skb)
 726 {
 727         struct nf_conn *ct;
 728         enum ip_conntrack_info ctinfo;
 729         struct nf_conntrack_l3proto *l3proto;
 730         struct nf_conntrack_l4proto *l4proto;
 731         unsigned int dataoff;
 732         u_int8_t protonum;
 733         int set_reply = 0;
 734         int ret;
 735
 736         /* Previously seen (loopback or untracked)?  Ignore. */
 737         if (skb->nfct) {
 738                 NF_CT_STAT_INC_ATOMIC(net, ignore);
 739                 return NF_ACCEPT;
 740         }
 741
 742         /* rcu_read_lock()ed by nf_hook_slow */
 743         l3proto = __nf_ct_l3proto_find(pf);
 744         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 745                                    &dataoff, &protonum);
 746         if (ret <= 0) {
 747                 pr_debug("not prepared to track yet or error occured\n");
 748                 NF_CT_STAT_INC_ATOMIC(net, error);
 749                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 750                 return -ret;
 751         }
 752
 753         l4proto = __nf_ct_l4proto_find(pf, protonum);
 754
 755         /* It may be an special packet, error, unclean...
 756          * inverse of the return code tells to the netfilter
 757          * core what to do with the packet. */
 758         if (l4proto->error != NULL) {
 759                 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
 760                 if (ret <= 0) {
 761                         NF_CT_STAT_INC_ATOMIC(net, error);
 762                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 763                         return -ret;
 764                 }
 765         }
 766
 767         ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
 768                                l3proto, l4proto, &set_reply, &ctinfo);
 769         if (!ct) {
 770                 /* Not valid part of a connection */
 771                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 772                 return NF_ACCEPT;
 773         }
 774
 775         if (IS_ERR(ct)) {
 776                 /* Too stressed to deal. */
 777                 NF_CT_STAT_INC_ATOMIC(net, drop);
 778                 return NF_DROP;
 779         }
 780
 781         NF_CT_ASSERT(skb->nfct);
 782
 783         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 784         if (ret <= 0) {
 785                 /* Invalid: inverse of the return code tells
 786                  * the netfilter core what to do */
 787                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 788                 nf_conntrack_put(skb->nfct);
 789                 skb->nfct = NULL;
 790                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 791                 if (ret == -NF_DROP)
 792                         NF_CT_STAT_INC_ATOMIC(net, drop);
 793                 return -ret;
 794         }
 795
 796         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 797                 nf_conntrack_event_cache(IPCT_STATUS, ct);
 798
 799         return ret;
 800 }
 801 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 802
 803 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 804                           const struct nf_conntrack_tuple *orig)
 805 {
 806         bool ret;
 807
 808         rcu_read_lock();
 809         ret = nf_ct_invert_tuple(inverse, orig,
 810                                  __nf_ct_l3proto_find(orig->src.l3num),
 811                                  __nf_ct_l4proto_find(orig->src.l3num,
 812                                                       orig->dst.protonum));
 813         rcu_read_unlock();
 814         return ret;
 815 }
 816 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 817
 818 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 819    implicitly racy: see __nf_conntrack_confirm */
 820 void nf_conntrack_alter_reply(struct nf_conn *ct,
 821                               const struct nf_conntrack_tuple *newreply)
 822 {
 823         struct nf_conn_help *help = nfct_help(ct);
 824
 825         /* Should be unconfirmed, so not in hash table yet */
 826         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 827
 828         pr_debug("Altering reply tuple of %p to ", ct);
 829         nf_ct_dump_tuple(newreply);
 830
 831         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 832         if (ct->master || (help && !hlist_empty(&help->expectations)))
 833                 return;
 834
 835         rcu_read_lock();
 836         __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
 837         rcu_read_unlock();
 838 }
 839 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 840
 841 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 842 void __nf_ct_refresh_acct(struct nf_conn *ct,
 843                           enum ip_conntrack_info ctinfo,
 844                           const struct sk_buff *skb,
 845                           unsigned long extra_jiffies,
 846                           int do_acct)
 847 {
 848         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 849         NF_CT_ASSERT(skb);
 850
 851         /* Only update if this is not a fixed timeout */
 852         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
 853                 goto acct;
 854
 855         /* If not in hash table, timer will not be active yet */
 856         if (!nf_ct_is_confirmed(ct)) {
 857                 ct->timeout.expires = extra_jiffies;
 858         } else {
 859                 unsigned long newtime = jiffies + extra_jiffies;
 860
 861                 /* Only update the timeout if the new timeout is at least
 862                    HZ jiffies from the old timeout. Need del_timer for race
 863                    avoidance (may already be dying). */
 864                 if (newtime - ct->timeout.expires >= HZ)
 865                         mod_timer_pending(&ct->timeout, newtime);
 866         }
 867
 868 acct:
 869         if (do_acct) {
 870                 struct nf_conn_counter *acct;
 871
 872                 acct = nf_conn_acct_find(ct);
 873                 if (acct) {
 874                         spin_lock_bh(&ct->lock);
 875                         acct[CTINFO2DIR(ctinfo)].packets++;
 876                         acct[CTINFO2DIR(ctinfo)].bytes +=
 877                                 skb->len - skb_network_offset(skb);
 878                         spin_unlock_bh(&ct->lock);
 879                 }
 880         }
 881 }
 882 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 883
 884 bool __nf_ct_kill_acct(struct nf_conn *ct,
 885                        enum ip_conntrack_info ctinfo,
 886                        const struct sk_buff *skb,
 887                        int do_acct)
 888 {
 889         if (do_acct) {
 890                 struct nf_conn_counter *acct;
 891
 892                 acct = nf_conn_acct_find(ct);
 893                 if (acct) {
 894                         spin_lock_bh(&ct->lock);
 895                         acct[CTINFO2DIR(ctinfo)].packets++;
 896                         acct[CTINFO2DIR(ctinfo)].bytes +=
 897                                 skb->len - skb_network_offset(skb);
 898                         spin_unlock_bh(&ct->lock);
 899                 }
 900         }
 901
 902         if (del_timer(&ct->timeout)) {
 903                 ct->timeout.function((unsigned long)ct);
 904                 return true;
 905         }
 906         return false;
 907 }
 908 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
 909
 910 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 911
 912 #include <linux/netfilter/nfnetlink.h>
 913 #include <linux/netfilter/nfnetlink_conntrack.h>
 914 #include <linux/mutex.h>
 915
 916 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 917  * in ip_conntrack_core, since we don't want the protocols to autoload
 918  * or depend on ctnetlink */
 919 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
 920                                const struct nf_conntrack_tuple *tuple)
 921 {
 922         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
 923         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
 924         return 0;
 925
 926 nla_put_failure:
 927         return -1;
 928 }
 929 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
 930
 931 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
 932         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
 933         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
 934 };
 935 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
 936
 937 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
 938                                struct nf_conntrack_tuple *t)
 939 {
 940         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
 941                 return -EINVAL;
 942
 943         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
 944         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
 945
 946         return 0;
 947 }
 948 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
 949
 950 int nf_ct_port_nlattr_tuple_size(void)
 951 {
 952         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
 953 }
 954 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
 955 #endif
 956
 957 /* Used by ipt_REJECT and ip6t_REJECT. */
 958 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 959 {
 960         struct nf_conn *ct;
 961         enum ip_conntrack_info ctinfo;
 962
 963         /* This ICMP is in reverse direction to the packet which caused it */
 964         ct = nf_ct_get(skb, &ctinfo);
 965         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 966                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 967         else
 968                 ctinfo = IP_CT_RELATED;
 969
 970         /* Attach to new skbuff, and increment count */
 971         nskb->nfct = &ct->ct_general;
 972         nskb->nfctinfo = ctinfo;
 973         nf_conntrack_get(nskb->nfct);
 974 }
 975
 976 /* Bring out ya dead! */
 977 static struct nf_conn *
 978 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
 979                 void *data, unsigned int *bucket)
 980 {
 981         struct nf_conntrack_tuple_hash *h;
 982         struct nf_conn *ct;
 983         struct hlist_nulls_node *n;
 984
 985         spin_lock_bh(&nf_conntrack_lock);
 986         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 987                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
 988                         ct = nf_ct_tuplehash_to_ctrack(h);
 989                         if (iter(ct, data))
 990                                 goto found;
 991                 }
 992         }
 993         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
 994                 ct = nf_ct_tuplehash_to_ctrack(h);
 995                 if (iter(ct, data))
 996                         set_bit(IPS_DYING_BIT, &ct->status);
 997         }
 998         spin_unlock_bh(&nf_conntrack_lock);
 999         return NULL;
1000 found:
1001         atomic_inc(&ct->ct_general.use);
1002         spin_unlock_bh(&nf_conntrack_lock);
1003         return ct;
1004 }
1005
1006 void nf_ct_iterate_cleanup(struct net *net,
1007                            int (*iter)(struct nf_conn *i, void *data),
1008                            void *data)
1009 {
1010         struct nf_conn *ct;
1011         unsigned int bucket = 0;
1012
1013         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1014                 /* Time to push up daises... */
1015                 if (del_timer(&ct->timeout))
1016                         death_by_timeout((unsigned long)ct);
1017                 /* ... else the timer will get him soon. */
1018
1019                 nf_ct_put(ct);
1020         }
1021 }
1022 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1023
1024 struct __nf_ct_flush_report {
1025         u32 pid;
1026         int report;
1027 };
1028
1029 static int kill_report(struct nf_conn *i, void *data)
1030 {
1031         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1032
1033         /* If we fail to deliver the event, death_by_timeout() will retry */
1034         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1035                                       fr->pid, fr->report) < 0)
1036                 return 1;
1037
1038         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1039         set_bit(IPS_DYING_BIT, &i->status);
1040         return 1;
1041 }
1042
1043 static int kill_all(struct nf_conn *i, void *data)
1044 {
1045         return 1;
1046 }
1047
1048 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1049 {
1050         if (vmalloced)
1051                 vfree(hash);
1052         else
1053                 free_pages((unsigned long)hash,
1054                            get_order(sizeof(struct hlist_head) * size));
1055 }
1056 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1057
1058 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1059 {
1060         struct __nf_ct_flush_report fr = {
1061                 .pid    = pid,
1062                 .report = report,
1063         };
1064         nf_ct_iterate_cleanup(net, kill_report, &fr);
1065 }
1066 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1067
1068 static void nf_ct_release_dying_list(void)
1069 {
1070         struct nf_conntrack_tuple_hash *h;
1071         struct nf_conn *ct;
1072         struct hlist_nulls_node *n;
1073
1074         spin_lock_bh(&nf_conntrack_lock);
1075         hlist_nulls_for_each_entry(h, n, &init_net.ct.dying, hnnode) {
1076                 ct = nf_ct_tuplehash_to_ctrack(h);
1077                 /* never fails to remove them, no listeners at this point */
1078                 nf_ct_kill(ct);
1079         }
1080         spin_unlock_bh(&nf_conntrack_lock);
1081 }
1082
1083 static void nf_conntrack_cleanup_init_net(void)
1084 {
1085         nf_conntrack_helper_fini();
1086         nf_conntrack_proto_fini();
1087         kmem_cache_destroy(nf_conntrack_cachep);
1088 }
1089
1090 static void nf_conntrack_cleanup_net(struct net *net)
1091 {
1092  i_see_dead_people:
1093         nf_ct_iterate_cleanup(net, kill_all, NULL);
1094         nf_ct_release_dying_list();
1095         if (atomic_read(&net->ct.count) != 0) {
1096                 schedule();
1097                 goto i_see_dead_people;
1098         }
1099         /* wait until all references to nf_conntrack_untracked are dropped */
1100         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1101                 schedule();
1102
1103         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1104                              nf_conntrack_htable_size);
1105         nf_conntrack_ecache_fini(net);
1106         nf_conntrack_acct_fini(net);
1107         nf_conntrack_expect_fini(net);
1108         free_percpu(net->ct.stat);
1109 }
1110
1111 /* Mishearing the voices in his head, our hero wonders how he's
1112    supposed to kill the mall. */
1113 void nf_conntrack_cleanup(struct net *net)
1114 {
1115         if (net_eq(net, &init_net))
1116                 rcu_assign_pointer(ip_ct_attach, NULL);
1117
1118         /* This makes sure all current packets have passed through
1119            netfilter framework.  Roll on, two-stage module
1120            delete... */
1121         synchronize_net();
1122
1123         nf_conntrack_cleanup_net(net);
1124
1125         if (net_eq(net, &init_net)) {
1126                 rcu_assign_pointer(nf_ct_destroy, NULL);
1127                 nf_conntrack_cleanup_init_net();
1128         }
1129 }
1130
1131 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1132 {
1133         struct hlist_nulls_head *hash;
1134         unsigned int nr_slots, i;
1135         size_t sz;
1136
1137         *vmalloced = 0;
1138
1139         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1140         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1141         sz = nr_slots * sizeof(struct hlist_nulls_head);
1142         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1143                                         get_order(sz));
1144         if (!hash) {
1145                 *vmalloced = 1;
1146                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1147                 hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
1148         }
1149
1150         if (hash && nulls)
1151                 for (i = 0; i < nr_slots; i++)
1152                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1153
1154         return hash;
1155 }
1156 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1157
1158 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1159 {
1160         int i, bucket, vmalloced, old_vmalloced;
1161         unsigned int hashsize, old_size;
1162         int rnd;
1163         struct hlist_nulls_head *hash, *old_hash;
1164         struct nf_conntrack_tuple_hash *h;
1165
1166         /* On boot, we can set this without any fancy locking. */
1167         if (!nf_conntrack_htable_size)
1168                 return param_set_uint(val, kp);
1169
1170         hashsize = simple_strtoul(val, NULL, 0);
1171         if (!hashsize)
1172                 return -EINVAL;
1173
1174         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1175         if (!hash)
1176                 return -ENOMEM;
1177
1178         /* We have to rehahs for the new table anyway, so we also can
1179          * use a newrandom seed */
1180         get_random_bytes(&rnd, sizeof(rnd));
1181
1182         /* Lookups in the old hash might happen in parallel, which means we
1183          * might get false negatives during connection lookup. New connections
1184          * created because of a false negative won't make it into the hash
1185          * though since that required taking the lock.
1186          */
1187         spin_lock_bh(&nf_conntrack_lock);
1188         for (i = 0; i < nf_conntrack_htable_size; i++) {
1189                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1190                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1191                                         struct nf_conntrack_tuple_hash, hnnode);
1192                         hlist_nulls_del_rcu(&h->hnnode);
1193                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1194                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1195                 }
1196         }
1197         old_size = nf_conntrack_htable_size;
1198         old_vmalloced = init_net.ct.hash_vmalloc;
1199         old_hash = init_net.ct.hash;
1200
1201         nf_conntrack_htable_size = hashsize;
1202         init_net.ct.hash_vmalloc = vmalloced;
1203         init_net.ct.hash = hash;
1204         nf_conntrack_hash_rnd = rnd;
1205         spin_unlock_bh(&nf_conntrack_lock);
1206
1207         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1208         return 0;
1209 }
1210 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1211
1212 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1213                   &nf_conntrack_htable_size, 0600);
1214
1215 static int nf_conntrack_init_init_net(void)
1216 {
1217         int max_factor = 8;
1218         int ret;
1219
1220         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1221          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1222         if (!nf_conntrack_htable_size) {
1223                 nf_conntrack_htable_size
1224                         = (((num_physpages << PAGE_SHIFT) / 16384)
1225                            / sizeof(struct hlist_head));
1226                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1227                         nf_conntrack_htable_size = 16384;
1228                 if (nf_conntrack_htable_size < 32)
1229                         nf_conntrack_htable_size = 32;
1230
1231                 /* Use a max. factor of four by default to get the same max as
1232                  * with the old struct list_heads. When a table size is given
1233                  * we use the old value of 8 to avoid reducing the max.
1234                  * entries. */
1235                 max_factor = 4;
1236         }
1237         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1238
1239         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1240                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1241                nf_conntrack_max);
1242
1243         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1244                                                 sizeof(struct nf_conn),
1245                                                 0, SLAB_DESTROY_BY_RCU, NULL);
1246         if (!nf_conntrack_cachep) {
1247                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1248                 ret = -ENOMEM;
1249                 goto err_cache;
1250         }
1251
1252         ret = nf_conntrack_proto_init();
1253         if (ret < 0)
1254                 goto err_proto;
1255
1256         ret = nf_conntrack_helper_init();
1257         if (ret < 0)
1258                 goto err_helper;
1259
1260         return 0;
1261
1262 err_helper:
1263         nf_conntrack_proto_fini();
1264 err_proto:
1265         kmem_cache_destroy(nf_conntrack_cachep);
1266 err_cache:
1267         return ret;
1268 }
1269
1270 /*
1271  * We need to use special "null" values, not used in hash table
1272  */
1273 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1274 #define DYING_NULLS_VAL         ((1<<30)+1)
1275
1276 static int nf_conntrack_init_net(struct net *net)
1277 {
1278         int ret;
1279
1280         atomic_set(&net->ct.count, 0);
1281         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1282         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1283         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1284         if (!net->ct.stat) {
1285                 ret = -ENOMEM;
1286                 goto err_stat;
1287         }
1288         net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1289                                              &net->ct.hash_vmalloc, 1);
1290         if (!net->ct.hash) {
1291                 ret = -ENOMEM;
1292                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1293                 goto err_hash;
1294         }
1295         ret = nf_conntrack_expect_init(net);
1296         if (ret < 0)
1297                 goto err_expect;
1298         ret = nf_conntrack_acct_init(net);
1299         if (ret < 0)
1300                 goto err_acct;
1301         ret = nf_conntrack_ecache_init(net);
1302         if (ret < 0)
1303                 goto err_ecache;
1304
1305         /* Set up fake conntrack:
1306             - to never be deleted, not in any hashes */
1307 #ifdef CONFIG_NET_NS
1308         nf_conntrack_untracked.ct_net = &init_net;
1309 #endif
1310         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1311         /*  - and look it like as a confirmed connection */
1312         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1313
1314         return 0;
1315
1316 err_ecache:
1317         nf_conntrack_acct_fini(net);
1318 err_acct:
1319         nf_conntrack_expect_fini(net);
1320 err_expect:
1321         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1322                              nf_conntrack_htable_size);
1323 err_hash:
1324         free_percpu(net->ct.stat);
1325 err_stat:
1326         return ret;
1327 }
1328
1329 int nf_conntrack_init(struct net *net)
1330 {
1331         int ret;
1332
1333         if (net_eq(net, &init_net)) {
1334                 ret = nf_conntrack_init_init_net();
1335                 if (ret < 0)
1336                         goto out_init_net;
1337         }
1338         ret = nf_conntrack_init_net(net);
1339         if (ret < 0)
1340                 goto out_net;
1341
1342         if (net_eq(net, &init_net)) {
1343                 /* For use by REJECT target */
1344                 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1345                 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1346         }
1347         return 0;
1348
1349 out_net:
1350         if (net_eq(net, &init_net))
1351                 nf_conntrack_cleanup_init_net();
1352 out_init_net:
1353         return ret;
1354 }