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         /* Timer relative to confirmation time, not original
 429            setting time, otherwise we'd get timer wrap in
 430            weird delay cases. */
 431         ct->timeout.expires += jiffies;
 432         add_timer(&ct->timeout);
 433         atomic_inc(&ct->ct_general.use);
 434         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 435
 436         /* Since the lookup is lockless, hash insertion must be done after
 437          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 438          * guarantee that no other CPU can find the conntrack before the above
 439          * stores are visible.
 440          */
 441         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 442         NF_CT_STAT_INC(net, insert);
 443         spin_unlock_bh(&nf_conntrack_lock);
 444
 445         help = nfct_help(ct);
 446         if (help && help->helper)
 447                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 448
 449         nf_conntrack_event_cache(master_ct(ct) ?
 450                                  IPCT_RELATED : IPCT_NEW, ct);
 451         return NF_ACCEPT;
 452
 453 out:
 454         NF_CT_STAT_INC(net, insert_failed);
 455         spin_unlock_bh(&nf_conntrack_lock);
 456         return NF_DROP;
 457 }
 458 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 459
 460 /* Returns true if a connection correspondings to the tuple (required
 461    for NAT). */
 462 int
 463 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 464                          const struct nf_conn *ignored_conntrack)
 465 {
 466         struct net *net = nf_ct_net(ignored_conntrack);
 467         struct nf_conntrack_tuple_hash *h;
 468         struct hlist_nulls_node *n;
 469         unsigned int hash = hash_conntrack(tuple);
 470
 471         /* Disable BHs the entire time since we need to disable them at
 472          * least once for the stats anyway.
 473          */
 474         rcu_read_lock_bh();
 475         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 476                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 477                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 478                         NF_CT_STAT_INC(net, found);
 479                         rcu_read_unlock_bh();
 480                         return 1;
 481                 }
 482                 NF_CT_STAT_INC(net, searched);
 483         }
 484         rcu_read_unlock_bh();
 485
 486         return 0;
 487 }
 488 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 489
 490 #define NF_CT_EVICTION_RANGE    8
 491
 492 /* There's a small race here where we may free a just-assured
 493    connection.  Too bad: we're in trouble anyway. */
 494 static noinline int early_drop(struct net *net, unsigned int hash)
 495 {
 496         /* Use oldest entry, which is roughly LRU */
 497         struct nf_conntrack_tuple_hash *h;
 498         struct nf_conn *ct = NULL, *tmp;
 499         struct hlist_nulls_node *n;
 500         unsigned int i, cnt = 0;
 501         int dropped = 0;
 502
 503         rcu_read_lock();
 504         for (i = 0; i < nf_conntrack_htable_size; i++) {
 505                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 506                                          hnnode) {
 507                         tmp = nf_ct_tuplehash_to_ctrack(h);
 508                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 509                                 ct = tmp;
 510                         cnt++;
 511                 }
 512
 513                 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 514                         ct = NULL;
 515                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
 516                         break;
 517                 hash = (hash + 1) % nf_conntrack_htable_size;
 518         }
 519         rcu_read_unlock();
 520
 521         if (!ct)
 522                 return dropped;
 523
 524         if (del_timer(&ct->timeout)) {
 525                 death_by_timeout((unsigned long)ct);
 526                 dropped = 1;
 527                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 528         }
 529         nf_ct_put(ct);
 530         return dropped;
 531 }
 532
 533 struct nf_conn *nf_conntrack_alloc(struct net *net,
 534                                    const struct nf_conntrack_tuple *orig,
 535                                    const struct nf_conntrack_tuple *repl,
 536                                    gfp_t gfp)
 537 {
 538         struct nf_conn *ct;
 539
 540         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 541                 get_random_bytes(&nf_conntrack_hash_rnd,
 542                                 sizeof(nf_conntrack_hash_rnd));
 543                 nf_conntrack_hash_rnd_initted = 1;
 544         }
 545
 546         /* We don't want any race condition at early drop stage */
 547         atomic_inc(&net->ct.count);
 548
 549         if (nf_conntrack_max &&
 550             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 551                 unsigned int hash = hash_conntrack(orig);
 552                 if (!early_drop(net, hash)) {
 553                         atomic_dec(&net->ct.count);
 554                         if (net_ratelimit())
 555                                 printk(KERN_WARNING
 556                                        "nf_conntrack: table full, dropping"
 557                                        " packet.\n");
 558                         return ERR_PTR(-ENOMEM);
 559                 }
 560         }
 561
 562         ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
 563         if (ct == NULL) {
 564                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
 565                 atomic_dec(&net->ct.count);
 566                 return ERR_PTR(-ENOMEM);
 567         }
 568
 569         spin_lock_init(&ct->lock);
 570         atomic_set(&ct->ct_general.use, 1);
 571         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 572         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 573         /* Don't set timer yet: wait for confirmation */
 574         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 575 #ifdef CONFIG_NET_NS
 576         ct->ct_net = net;
 577 #endif
 578
 579         return ct;
 580 }
 581 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 582
 583 void nf_conntrack_free(struct nf_conn *ct)
 584 {
 585         struct net *net = nf_ct_net(ct);
 586
 587         nf_ct_ext_destroy(ct);
 588         atomic_dec(&net->ct.count);
 589         nf_ct_ext_free(ct);
 590         kmem_cache_free(nf_conntrack_cachep, ct);
 591 }
 592 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 593
 594 /* Allocate a new conntrack: we return -ENOMEM if classification
 595    failed due to stress.  Otherwise it really is unclassifiable. */
 596 static struct nf_conntrack_tuple_hash *
 597 init_conntrack(struct net *net,
 598                const struct nf_conntrack_tuple *tuple,
 599                struct nf_conntrack_l3proto *l3proto,
 600                struct nf_conntrack_l4proto *l4proto,
 601                struct sk_buff *skb,
 602                unsigned int dataoff)
 603 {
 604         struct nf_conn *ct;
 605         struct nf_conn_help *help;
 606         struct nf_conntrack_tuple repl_tuple;
 607         struct nf_conntrack_expect *exp;
 608
 609         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 610                 pr_debug("Can't invert tuple.\n");
 611                 return NULL;
 612         }
 613
 614         ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
 615         if (IS_ERR(ct)) {
 616                 pr_debug("Can't allocate conntrack.\n");
 617                 return (struct nf_conntrack_tuple_hash *)ct;
 618         }
 619
 620         if (!l4proto->new(ct, skb, dataoff)) {
 621                 nf_conntrack_free(ct);
 622                 pr_debug("init conntrack: can't track with proto module\n");
 623                 return NULL;
 624         }
 625
 626         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 627         nf_ct_ecache_ext_add(ct, GFP_ATOMIC);
 628
 629         spin_lock_bh(&nf_conntrack_lock);
 630         exp = nf_ct_find_expectation(net, tuple);
 631         if (exp) {
 632                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 633                          ct, exp);
 634                 /* Welcome, Mr. Bond.  We've been expecting you... */
 635                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 636                 ct->master = exp->master;
 637                 if (exp->helper) {
 638                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 639                         if (help)
 640                                 rcu_assign_pointer(help->helper, exp->helper);
 641                 }
 642
 643 #ifdef CONFIG_NF_CONNTRACK_MARK
 644                 ct->mark = exp->master->mark;
 645 #endif
 646 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 647                 ct->secmark = exp->master->secmark;
 648 #endif
 649                 nf_conntrack_get(&ct->master->ct_general);
 650                 NF_CT_STAT_INC(net, expect_new);
 651         } else {
 652                 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
 653                 NF_CT_STAT_INC(net, new);
 654         }
 655
 656         /* Overload tuple linked list to put us in unconfirmed list. */
 657         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 658                        &net->ct.unconfirmed);
 659
 660         spin_unlock_bh(&nf_conntrack_lock);
 661
 662         if (exp) {
 663                 if (exp->expectfn)
 664                         exp->expectfn(ct, exp);
 665                 nf_ct_expect_put(exp);
 666         }
 667
 668         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 669 }
 670
 671 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 672 static inline struct nf_conn *
 673 resolve_normal_ct(struct net *net,
 674                   struct sk_buff *skb,
 675                   unsigned int dataoff,
 676                   u_int16_t l3num,
 677                   u_int8_t protonum,
 678                   struct nf_conntrack_l3proto *l3proto,
 679                   struct nf_conntrack_l4proto *l4proto,
 680                   int *set_reply,
 681                   enum ip_conntrack_info *ctinfo)
 682 {
 683         struct nf_conntrack_tuple tuple;
 684         struct nf_conntrack_tuple_hash *h;
 685         struct nf_conn *ct;
 686
 687         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 688                              dataoff, l3num, protonum, &tuple, l3proto,
 689                              l4proto)) {
 690                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 691                 return NULL;
 692         }
 693
 694         /* look for tuple match */
 695         h = nf_conntrack_find_get(net, &tuple);
 696         if (!h) {
 697                 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
 698                 if (!h)
 699                         return NULL;
 700                 if (IS_ERR(h))
 701                         return (void *)h;
 702         }
 703         ct = nf_ct_tuplehash_to_ctrack(h);
 704
 705         /* It exists; we have (non-exclusive) reference. */
 706         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 707                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 708                 /* Please set reply bit if this packet OK */
 709                 *set_reply = 1;
 710         } else {
 711                 /* Once we've had two way comms, always ESTABLISHED. */
 712                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 713                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 714                         *ctinfo = IP_CT_ESTABLISHED;
 715                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 716                         pr_debug("nf_conntrack_in: related packet for %p\n",
 717                                  ct);
 718                         *ctinfo = IP_CT_RELATED;
 719                 } else {
 720                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 721                         *ctinfo = IP_CT_NEW;
 722                 }
 723                 *set_reply = 0;
 724         }
 725         skb->nfct = &ct->ct_general;
 726         skb->nfctinfo = *ctinfo;
 727         return ct;
 728 }
 729
 730 unsigned int
 731 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 732                 struct sk_buff *skb)
 733 {
 734         struct nf_conn *ct;
 735         enum ip_conntrack_info ctinfo;
 736         struct nf_conntrack_l3proto *l3proto;
 737         struct nf_conntrack_l4proto *l4proto;
 738         unsigned int dataoff;
 739         u_int8_t protonum;
 740         int set_reply = 0;
 741         int ret;
 742
 743         /* Previously seen (loopback or untracked)?  Ignore. */
 744         if (skb->nfct) {
 745                 NF_CT_STAT_INC_ATOMIC(net, ignore);
 746                 return NF_ACCEPT;
 747         }
 748
 749         /* rcu_read_lock()ed by nf_hook_slow */
 750         l3proto = __nf_ct_l3proto_find(pf);
 751         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 752                                    &dataoff, &protonum);
 753         if (ret <= 0) {
 754                 pr_debug("not prepared to track yet or error occured\n");
 755                 NF_CT_STAT_INC_ATOMIC(net, error);
 756                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 757                 return -ret;
 758         }
 759
 760         l4proto = __nf_ct_l4proto_find(pf, protonum);
 761
 762         /* It may be an special packet, error, unclean...
 763          * inverse of the return code tells to the netfilter
 764          * core what to do with the packet. */
 765         if (l4proto->error != NULL) {
 766                 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
 767                 if (ret <= 0) {
 768                         NF_CT_STAT_INC_ATOMIC(net, error);
 769                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 770                         return -ret;
 771                 }
 772         }
 773
 774         ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
 775                                l3proto, l4proto, &set_reply, &ctinfo);
 776         if (!ct) {
 777                 /* Not valid part of a connection */
 778                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 779                 return NF_ACCEPT;
 780         }
 781
 782         if (IS_ERR(ct)) {
 783                 /* Too stressed to deal. */
 784                 NF_CT_STAT_INC_ATOMIC(net, drop);
 785                 return NF_DROP;
 786         }
 787
 788         NF_CT_ASSERT(skb->nfct);
 789
 790         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 791         if (ret <= 0) {
 792                 /* Invalid: inverse of the return code tells
 793                  * the netfilter core what to do */
 794                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 795                 nf_conntrack_put(skb->nfct);
 796                 skb->nfct = NULL;
 797                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 798                 if (ret == -NF_DROP)
 799                         NF_CT_STAT_INC_ATOMIC(net, drop);
 800                 return -ret;
 801         }
 802
 803         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 804                 nf_conntrack_event_cache(IPCT_STATUS, ct);
 805
 806         return ret;
 807 }
 808 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 809
 810 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 811                           const struct nf_conntrack_tuple *orig)
 812 {
 813         bool ret;
 814
 815         rcu_read_lock();
 816         ret = nf_ct_invert_tuple(inverse, orig,
 817                                  __nf_ct_l3proto_find(orig->src.l3num),
 818                                  __nf_ct_l4proto_find(orig->src.l3num,
 819                                                       orig->dst.protonum));
 820         rcu_read_unlock();
 821         return ret;
 822 }
 823 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 824
 825 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 826    implicitly racy: see __nf_conntrack_confirm */
 827 void nf_conntrack_alter_reply(struct nf_conn *ct,
 828                               const struct nf_conntrack_tuple *newreply)
 829 {
 830         struct nf_conn_help *help = nfct_help(ct);
 831
 832         /* Should be unconfirmed, so not in hash table yet */
 833         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 834
 835         pr_debug("Altering reply tuple of %p to ", ct);
 836         nf_ct_dump_tuple(newreply);
 837
 838         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 839         if (ct->master || (help && !hlist_empty(&help->expectations)))
 840                 return;
 841
 842         rcu_read_lock();
 843         __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
 844         rcu_read_unlock();
 845 }
 846 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 847
 848 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 849 void __nf_ct_refresh_acct(struct nf_conn *ct,
 850                           enum ip_conntrack_info ctinfo,
 851                           const struct sk_buff *skb,
 852                           unsigned long extra_jiffies,
 853                           int do_acct)
 854 {
 855         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 856         NF_CT_ASSERT(skb);
 857
 858         /* Only update if this is not a fixed timeout */
 859         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
 860                 goto acct;
 861
 862         /* If not in hash table, timer will not be active yet */
 863         if (!nf_ct_is_confirmed(ct)) {
 864                 ct->timeout.expires = extra_jiffies;
 865         } else {
 866                 unsigned long newtime = jiffies + extra_jiffies;
 867
 868                 /* Only update the timeout if the new timeout is at least
 869                    HZ jiffies from the old timeout. Need del_timer for race
 870                    avoidance (may already be dying). */
 871                 if (newtime - ct->timeout.expires >= HZ)
 872                         mod_timer_pending(&ct->timeout, newtime);
 873         }
 874
 875 acct:
 876         if (do_acct) {
 877                 struct nf_conn_counter *acct;
 878
 879                 acct = nf_conn_acct_find(ct);
 880                 if (acct) {
 881                         spin_lock_bh(&ct->lock);
 882                         acct[CTINFO2DIR(ctinfo)].packets++;
 883                         acct[CTINFO2DIR(ctinfo)].bytes +=
 884                                 skb->len - skb_network_offset(skb);
 885                         spin_unlock_bh(&ct->lock);
 886                 }
 887         }
 888 }
 889 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 890
 891 bool __nf_ct_kill_acct(struct nf_conn *ct,
 892                        enum ip_conntrack_info ctinfo,
 893                        const struct sk_buff *skb,
 894                        int do_acct)
 895 {
 896         if (do_acct) {
 897                 struct nf_conn_counter *acct;
 898
 899                 acct = nf_conn_acct_find(ct);
 900                 if (acct) {
 901                         spin_lock_bh(&ct->lock);
 902                         acct[CTINFO2DIR(ctinfo)].packets++;
 903                         acct[CTINFO2DIR(ctinfo)].bytes +=
 904                                 skb->len - skb_network_offset(skb);
 905                         spin_unlock_bh(&ct->lock);
 906                 }
 907         }
 908
 909         if (del_timer(&ct->timeout)) {
 910                 ct->timeout.function((unsigned long)ct);
 911                 return true;
 912         }
 913         return false;
 914 }
 915 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
 916
 917 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 918
 919 #include <linux/netfilter/nfnetlink.h>
 920 #include <linux/netfilter/nfnetlink_conntrack.h>
 921 #include <linux/mutex.h>
 922
 923 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 924  * in ip_conntrack_core, since we don't want the protocols to autoload
 925  * or depend on ctnetlink */
 926 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
 927                                const struct nf_conntrack_tuple *tuple)
 928 {
 929         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
 930         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
 931         return 0;
 932
 933 nla_put_failure:
 934         return -1;
 935 }
 936 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
 937
 938 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
 939         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
 940         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
 941 };
 942 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
 943
 944 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
 945                                struct nf_conntrack_tuple *t)
 946 {
 947         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
 948                 return -EINVAL;
 949
 950         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
 951         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
 952
 953         return 0;
 954 }
 955 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
 956
 957 int nf_ct_port_nlattr_tuple_size(void)
 958 {
 959         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
 960 }
 961 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
 962 #endif
 963
 964 /* Used by ipt_REJECT and ip6t_REJECT. */
 965 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 966 {
 967         struct nf_conn *ct;
 968         enum ip_conntrack_info ctinfo;
 969
 970         /* This ICMP is in reverse direction to the packet which caused it */
 971         ct = nf_ct_get(skb, &ctinfo);
 972         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 973                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 974         else
 975                 ctinfo = IP_CT_RELATED;
 976
 977         /* Attach to new skbuff, and increment count */
 978         nskb->nfct = &ct->ct_general;
 979         nskb->nfctinfo = ctinfo;
 980         nf_conntrack_get(nskb->nfct);
 981 }
 982
 983 /* Bring out ya dead! */
 984 static struct nf_conn *
 985 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
 986                 void *data, unsigned int *bucket)
 987 {
 988         struct nf_conntrack_tuple_hash *h;
 989         struct nf_conn *ct;
 990         struct hlist_nulls_node *n;
 991
 992         spin_lock_bh(&nf_conntrack_lock);
 993         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 994                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
 995                         ct = nf_ct_tuplehash_to_ctrack(h);
 996                         if (iter(ct, data))
 997                                 goto found;
 998                 }
 999         }
1000         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1001                 ct = nf_ct_tuplehash_to_ctrack(h);
1002                 if (iter(ct, data))
1003                         set_bit(IPS_DYING_BIT, &ct->status);
1004         }
1005         spin_unlock_bh(&nf_conntrack_lock);
1006         return NULL;
1007 found:
1008         atomic_inc(&ct->ct_general.use);
1009         spin_unlock_bh(&nf_conntrack_lock);
1010         return ct;
1011 }
1012
1013 void nf_ct_iterate_cleanup(struct net *net,
1014                            int (*iter)(struct nf_conn *i, void *data),
1015                            void *data)
1016 {
1017         struct nf_conn *ct;
1018         unsigned int bucket = 0;
1019
1020         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1021                 /* Time to push up daises... */
1022                 if (del_timer(&ct->timeout))
1023                         death_by_timeout((unsigned long)ct);
1024                 /* ... else the timer will get him soon. */
1025
1026                 nf_ct_put(ct);
1027         }
1028 }
1029 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1030
1031 struct __nf_ct_flush_report {
1032         u32 pid;
1033         int report;
1034 };
1035
1036 static int kill_report(struct nf_conn *i, void *data)
1037 {
1038         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1039
1040         /* If we fail to deliver the event, death_by_timeout() will retry */
1041         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1042                                       fr->pid, fr->report) < 0)
1043                 return 1;
1044
1045         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1046         set_bit(IPS_DYING_BIT, &i->status);
1047         return 1;
1048 }
1049
1050 static int kill_all(struct nf_conn *i, void *data)
1051 {
1052         return 1;
1053 }
1054
1055 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1056 {
1057         if (vmalloced)
1058                 vfree(hash);
1059         else
1060                 free_pages((unsigned long)hash,
1061                            get_order(sizeof(struct hlist_head) * size));
1062 }
1063 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1064
1065 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1066 {
1067         struct __nf_ct_flush_report fr = {
1068                 .pid    = pid,
1069                 .report = report,
1070         };
1071         nf_ct_iterate_cleanup(net, kill_report, &fr);
1072 }
1073 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1074
1075 static void nf_ct_release_dying_list(void)
1076 {
1077         struct nf_conntrack_tuple_hash *h;
1078         struct nf_conn *ct;
1079         struct hlist_nulls_node *n;
1080
1081         spin_lock_bh(&nf_conntrack_lock);
1082         hlist_nulls_for_each_entry(h, n, &init_net.ct.dying, hnnode) {
1083                 ct = nf_ct_tuplehash_to_ctrack(h);
1084                 /* never fails to remove them, no listeners at this point */
1085                 nf_ct_kill(ct);
1086         }
1087         spin_unlock_bh(&nf_conntrack_lock);
1088 }
1089
1090 static void nf_conntrack_cleanup_init_net(void)
1091 {
1092         nf_conntrack_helper_fini();
1093         nf_conntrack_proto_fini();
1094         kmem_cache_destroy(nf_conntrack_cachep);
1095 }
1096
1097 static void nf_conntrack_cleanup_net(struct net *net)
1098 {
1099  i_see_dead_people:
1100         nf_ct_iterate_cleanup(net, kill_all, NULL);
1101         nf_ct_release_dying_list();
1102         if (atomic_read(&net->ct.count) != 0) {
1103                 schedule();
1104                 goto i_see_dead_people;
1105         }
1106         /* wait until all references to nf_conntrack_untracked are dropped */
1107         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1108                 schedule();
1109
1110         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1111                              nf_conntrack_htable_size);
1112         nf_conntrack_ecache_fini(net);
1113         nf_conntrack_acct_fini(net);
1114         nf_conntrack_expect_fini(net);
1115         free_percpu(net->ct.stat);
1116 }
1117
1118 /* Mishearing the voices in his head, our hero wonders how he's
1119    supposed to kill the mall. */
1120 void nf_conntrack_cleanup(struct net *net)
1121 {
1122         if (net_eq(net, &init_net))
1123                 rcu_assign_pointer(ip_ct_attach, NULL);
1124
1125         /* This makes sure all current packets have passed through
1126            netfilter framework.  Roll on, two-stage module
1127            delete... */
1128         synchronize_net();
1129
1130         nf_conntrack_cleanup_net(net);
1131
1132         if (net_eq(net, &init_net)) {
1133                 rcu_assign_pointer(nf_ct_destroy, NULL);
1134                 nf_conntrack_cleanup_init_net();
1135         }
1136 }
1137
1138 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1139 {
1140         struct hlist_nulls_head *hash;
1141         unsigned int nr_slots, i;
1142         size_t sz;
1143
1144         *vmalloced = 0;
1145
1146         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1147         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1148         sz = nr_slots * sizeof(struct hlist_nulls_head);
1149         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1150                                         get_order(sz));
1151         if (!hash) {
1152                 *vmalloced = 1;
1153                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1154                 hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
1155         }
1156
1157         if (hash && nulls)
1158                 for (i = 0; i < nr_slots; i++)
1159                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1160
1161         return hash;
1162 }
1163 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1164
1165 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1166 {
1167         int i, bucket, vmalloced, old_vmalloced;
1168         unsigned int hashsize, old_size;
1169         int rnd;
1170         struct hlist_nulls_head *hash, *old_hash;
1171         struct nf_conntrack_tuple_hash *h;
1172
1173         /* On boot, we can set this without any fancy locking. */
1174         if (!nf_conntrack_htable_size)
1175                 return param_set_uint(val, kp);
1176
1177         hashsize = simple_strtoul(val, NULL, 0);
1178         if (!hashsize)
1179                 return -EINVAL;
1180
1181         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1182         if (!hash)
1183                 return -ENOMEM;
1184
1185         /* We have to rehahs for the new table anyway, so we also can
1186          * use a newrandom seed */
1187         get_random_bytes(&rnd, sizeof(rnd));
1188
1189         /* Lookups in the old hash might happen in parallel, which means we
1190          * might get false negatives during connection lookup. New connections
1191          * created because of a false negative won't make it into the hash
1192          * though since that required taking the lock.
1193          */
1194         spin_lock_bh(&nf_conntrack_lock);
1195         for (i = 0; i < nf_conntrack_htable_size; i++) {
1196                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1197                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1198                                         struct nf_conntrack_tuple_hash, hnnode);
1199                         hlist_nulls_del_rcu(&h->hnnode);
1200                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1201                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1202                 }
1203         }
1204         old_size = nf_conntrack_htable_size;
1205         old_vmalloced = init_net.ct.hash_vmalloc;
1206         old_hash = init_net.ct.hash;
1207
1208         nf_conntrack_htable_size = hashsize;
1209         init_net.ct.hash_vmalloc = vmalloced;
1210         init_net.ct.hash = hash;
1211         nf_conntrack_hash_rnd = rnd;
1212         spin_unlock_bh(&nf_conntrack_lock);
1213
1214         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1215         return 0;
1216 }
1217 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1218
1219 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1220                   &nf_conntrack_htable_size, 0600);
1221
1222 static int nf_conntrack_init_init_net(void)
1223 {
1224         int max_factor = 8;
1225         int ret;
1226
1227         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1228          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1229         if (!nf_conntrack_htable_size) {
1230                 nf_conntrack_htable_size
1231                         = (((num_physpages << PAGE_SHIFT) / 16384)
1232                            / sizeof(struct hlist_head));
1233                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1234                         nf_conntrack_htable_size = 16384;
1235                 if (nf_conntrack_htable_size < 32)
1236                         nf_conntrack_htable_size = 32;
1237
1238                 /* Use a max. factor of four by default to get the same max as
1239                  * with the old struct list_heads. When a table size is given
1240                  * we use the old value of 8 to avoid reducing the max.
1241                  * entries. */
1242                 max_factor = 4;
1243         }
1244         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1245
1246         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1247                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1248                nf_conntrack_max);
1249
1250         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1251                                                 sizeof(struct nf_conn),
1252                                                 0, SLAB_DESTROY_BY_RCU, NULL);
1253         if (!nf_conntrack_cachep) {
1254                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1255                 ret = -ENOMEM;
1256                 goto err_cache;
1257         }
1258
1259         ret = nf_conntrack_proto_init();
1260         if (ret < 0)
1261                 goto err_proto;
1262
1263         ret = nf_conntrack_helper_init();
1264         if (ret < 0)
1265                 goto err_helper;
1266
1267         return 0;
1268
1269 err_helper:
1270         nf_conntrack_proto_fini();
1271 err_proto:
1272         kmem_cache_destroy(nf_conntrack_cachep);
1273 err_cache:
1274         return ret;
1275 }
1276
1277 /*
1278  * We need to use special "null" values, not used in hash table
1279  */
1280 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1281 #define DYING_NULLS_VAL         ((1<<30)+1)
1282
1283 static int nf_conntrack_init_net(struct net *net)
1284 {
1285         int ret;
1286
1287         atomic_set(&net->ct.count, 0);
1288         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1289         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1290         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1291         if (!net->ct.stat) {
1292                 ret = -ENOMEM;
1293                 goto err_stat;
1294         }
1295         net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1296                                              &net->ct.hash_vmalloc, 1);
1297         if (!net->ct.hash) {
1298                 ret = -ENOMEM;
1299                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1300                 goto err_hash;
1301         }
1302         ret = nf_conntrack_expect_init(net);
1303         if (ret < 0)
1304                 goto err_expect;
1305         ret = nf_conntrack_acct_init(net);
1306         if (ret < 0)
1307                 goto err_acct;
1308         ret = nf_conntrack_ecache_init(net);
1309         if (ret < 0)
1310                 goto err_ecache;
1311
1312         /* Set up fake conntrack:
1313             - to never be deleted, not in any hashes */
1314 #ifdef CONFIG_NET_NS
1315         nf_conntrack_untracked.ct_net = &init_net;
1316 #endif
1317         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1318         /*  - and look it like as a confirmed connection */
1319         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1320
1321         return 0;
1322
1323 err_ecache:
1324         nf_conntrack_acct_fini(net);
1325 err_acct:
1326         nf_conntrack_expect_fini(net);
1327 err_expect:
1328         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1329                              nf_conntrack_htable_size);
1330 err_hash:
1331         free_percpu(net->ct.stat);
1332 err_stat:
1333         return ret;
1334 }
1335
1336 int nf_conntrack_init(struct net *net)
1337 {
1338         int ret;
1339
1340         if (net_eq(net, &init_net)) {
1341                 ret = nf_conntrack_init_init_net();
1342                 if (ret < 0)
1343                         goto out_init_net;
1344         }
1345         ret = nf_conntrack_init_net(net);
1346         if (ret < 0)
1347                 goto out_net;
1348
1349         if (net_eq(net, &init_net)) {
1350                 /* For use by REJECT target */
1351                 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1352                 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1353         }
1354         return 0;
1355
1356 out_net:
1357         if (net_eq(net, &init_net))
1358                 nf_conntrack_cleanup_init_net();
1359 out_init_net:
1360         return ret;
1361 }