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/sched.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/proc_fs.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/stddef.h>
  22 #include <linux/slab.h>
  23 #include <linux/random.h>
  24 #include <linux/jhash.h>
  25 #include <linux/err.h>
  26 #include <linux/percpu.h>
  27 #include <linux/moduleparam.h>
  28 #include <linux/notifier.h>
  29 #include <linux/kernel.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/socket.h>
  32 #include <linux/mm.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/rculist_nulls.h>
  35
  36 #include <net/netfilter/nf_conntrack.h>
  37 #include <net/netfilter/nf_conntrack_l3proto.h>
  38 #include <net/netfilter/nf_conntrack_l4proto.h>
  39 #include <net/netfilter/nf_conntrack_expect.h>
  40 #include <net/netfilter/nf_conntrack_helper.h>
  41 #include <net/netfilter/nf_conntrack_core.h>
  42 #include <net/netfilter/nf_conntrack_extend.h>
  43 #include <net/netfilter/nf_conntrack_acct.h>
  44 #include <net/netfilter/nf_conntrack_ecache.h>
  45 #include <net/netfilter/nf_conntrack_zones.h>
  46 #include <net/netfilter/nf_nat.h>
  47 #include <net/netfilter/nf_nat_core.h>
  48
  49 #define NF_CONNTRACK_VERSION    "0.5.0"
  50
  51 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  52                                       enum nf_nat_manip_type manip,
  53                                       const struct nlattr *attr) __read_mostly;
  54 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  55
  56 DEFINE_SPINLOCK(nf_conntrack_lock);
  57 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  58
  59 unsigned int nf_conntrack_htable_size __read_mostly;
  60 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  61
  62 unsigned int nf_conntrack_max __read_mostly;
  63 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  64
  65 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  66 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  67
  68 unsigned int nf_conntrack_hash_rnd __read_mostly;
  69
  70 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  71 {
  72         unsigned int n;
  73
  74         /* The direction must be ignored, so we hash everything up to the
  75          * destination ports (which is a multiple of 4) and treat the last
  76          * three bytes manually.
  77          */
  78         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  79         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  80                       (((__force __u16)tuple->dst.u.all << 16) |
  81                       tuple->dst.protonum));
  82 }
  83
  84 static u32 __hash_bucket(u32 hash, unsigned int size)
  85 {
  86         return ((u64)hash * size) >> 32;
  87 }
  88
  89 static u32 hash_bucket(u32 hash, const struct net *net)
  90 {
  91         return __hash_bucket(hash, net->ct.htable_size);
  92 }
  93
  94 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  95                                   u16 zone, unsigned int size)
  96 {
  97         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
  98 }
  99
 100 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 101                                        const struct nf_conntrack_tuple *tuple)
 102 {
 103         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 104 }
 105
 106 bool
 107 nf_ct_get_tuple(const struct sk_buff *skb,
 108                 unsigned int nhoff,
 109                 unsigned int dataoff,
 110                 u_int16_t l3num,
 111                 u_int8_t protonum,
 112                 struct nf_conntrack_tuple *tuple,
 113                 const struct nf_conntrack_l3proto *l3proto,
 114                 const struct nf_conntrack_l4proto *l4proto)
 115 {
 116         memset(tuple, 0, sizeof(*tuple));
 117
 118         tuple->src.l3num = l3num;
 119         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 120                 return false;
 121
 122         tuple->dst.protonum = protonum;
 123         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 124
 125         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 126 }
 127 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 128
 129 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 130                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 131 {
 132         struct nf_conntrack_l3proto *l3proto;
 133         struct nf_conntrack_l4proto *l4proto;
 134         unsigned int protoff;
 135         u_int8_t protonum;
 136         int ret;
 137
 138         rcu_read_lock();
 139
 140         l3proto = __nf_ct_l3proto_find(l3num);
 141         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 142         if (ret != NF_ACCEPT) {
 143                 rcu_read_unlock();
 144                 return false;
 145         }
 146
 147         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 148
 149         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 150                               l3proto, l4proto);
 151
 152         rcu_read_unlock();
 153         return ret;
 154 }
 155 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 156
 157 bool
 158 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 159                    const struct nf_conntrack_tuple *orig,
 160                    const struct nf_conntrack_l3proto *l3proto,
 161                    const struct nf_conntrack_l4proto *l4proto)
 162 {
 163         memset(inverse, 0, sizeof(*inverse));
 164
 165         inverse->src.l3num = orig->src.l3num;
 166         if (l3proto->invert_tuple(inverse, orig) == 0)
 167                 return false;
 168
 169         inverse->dst.dir = !orig->dst.dir;
 170
 171         inverse->dst.protonum = orig->dst.protonum;
 172         return l4proto->invert_tuple(inverse, orig);
 173 }
 174 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 175
 176 static void
 177 clean_from_lists(struct nf_conn *ct)
 178 {
 179         pr_debug("clean_from_lists(%p)\n", ct);
 180         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 181         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 182
 183         /* Destroy all pending expectations */
 184         nf_ct_remove_expectations(ct);
 185 }
 186
 187 static void
 188 destroy_conntrack(struct nf_conntrack *nfct)
 189 {
 190         struct nf_conn *ct = (struct nf_conn *)nfct;
 191         struct net *net = nf_ct_net(ct);
 192         struct nf_conntrack_l4proto *l4proto;
 193
 194         pr_debug("destroy_conntrack(%p)\n", ct);
 195         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 196         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 197
 198         /* To make sure we don't get any weird locking issues here:
 199          * destroy_conntrack() MUST NOT be called with a write lock
 200          * to nf_conntrack_lock!!! -HW */
 201         rcu_read_lock();
 202         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 203         if (l4proto && l4proto->destroy)
 204                 l4proto->destroy(ct);
 205
 206         rcu_read_unlock();
 207
 208         spin_lock_bh(&nf_conntrack_lock);
 209         /* Expectations will have been removed in clean_from_lists,
 210          * except TFTP can create an expectation on the first packet,
 211          * before connection is in the list, so we need to clean here,
 212          * too. */
 213         nf_ct_remove_expectations(ct);
 214
 215         /* We overload first tuple to link into unconfirmed list. */
 216         if (!nf_ct_is_confirmed(ct)) {
 217                 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 218                 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 219         }
 220
 221         NF_CT_STAT_INC(net, delete);
 222         spin_unlock_bh(&nf_conntrack_lock);
 223
 224         if (ct->master)
 225                 nf_ct_put(ct->master);
 226
 227         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 228         nf_conntrack_free(ct);
 229 }
 230
 231 void nf_ct_delete_from_lists(struct nf_conn *ct)
 232 {
 233         struct net *net = nf_ct_net(ct);
 234
 235         nf_ct_helper_destroy(ct);
 236         spin_lock_bh(&nf_conntrack_lock);
 237         /* Inside lock so preempt is disabled on module removal path.
 238          * Otherwise we can get spurious warnings. */
 239         NF_CT_STAT_INC(net, delete_list);
 240         clean_from_lists(ct);
 241         spin_unlock_bh(&nf_conntrack_lock);
 242 }
 243 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
 244
 245 static void death_by_event(unsigned long ul_conntrack)
 246 {
 247         struct nf_conn *ct = (void *)ul_conntrack;
 248         struct net *net = nf_ct_net(ct);
 249
 250         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 251                 /* bad luck, let's retry again */
 252                 ct->timeout.expires = jiffies +
 253                         (random32() % net->ct.sysctl_events_retry_timeout);
 254                 add_timer(&ct->timeout);
 255                 return;
 256         }
 257         /* we've got the event delivered, now it's dying */
 258         set_bit(IPS_DYING_BIT, &ct->status);
 259         spin_lock(&nf_conntrack_lock);
 260         hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 261         spin_unlock(&nf_conntrack_lock);
 262         nf_ct_put(ct);
 263 }
 264
 265 void nf_ct_insert_dying_list(struct nf_conn *ct)
 266 {
 267         struct net *net = nf_ct_net(ct);
 268
 269         /* add this conntrack to the dying list */
 270         spin_lock_bh(&nf_conntrack_lock);
 271         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 272                              &net->ct.dying);
 273         spin_unlock_bh(&nf_conntrack_lock);
 274         /* set a new timer to retry event delivery */
 275         setup_timer(&ct->timeout, death_by_event, (unsigned long)ct);
 276         ct->timeout.expires = jiffies +
 277                 (random32() % net->ct.sysctl_events_retry_timeout);
 278         add_timer(&ct->timeout);
 279 }
 280 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list);
 281
 282 static void death_by_timeout(unsigned long ul_conntrack)
 283 {
 284         struct nf_conn *ct = (void *)ul_conntrack;
 285
 286         if (!test_bit(IPS_DYING_BIT, &ct->status) &&
 287             unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
 288                 /* destroy event was not delivered */
 289                 nf_ct_delete_from_lists(ct);
 290                 nf_ct_insert_dying_list(ct);
 291                 return;
 292         }
 293         set_bit(IPS_DYING_BIT, &ct->status);
 294         nf_ct_delete_from_lists(ct);
 295         nf_ct_put(ct);
 296 }
 297
 298 /*
 299  * Warning :
 300  * - Caller must take a reference on returned object
 301  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 302  * OR
 303  * - Caller must lock nf_conntrack_lock before calling this function
 304  */
 305 static struct nf_conntrack_tuple_hash *
 306 ____nf_conntrack_find(struct net *net, u16 zone,
 307                       const struct nf_conntrack_tuple *tuple, u32 hash)
 308 {
 309         struct nf_conntrack_tuple_hash *h;
 310         struct hlist_nulls_node *n;
 311         unsigned int bucket = hash_bucket(hash, net);
 312
 313         /* Disable BHs the entire time since we normally need to disable them
 314          * at least once for the stats anyway.
 315          */
 316         local_bh_disable();
 317 begin:
 318         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 319                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 320                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 321                         NF_CT_STAT_INC(net, found);
 322                         local_bh_enable();
 323                         return h;
 324                 }
 325                 NF_CT_STAT_INC(net, searched);
 326         }
 327         /*
 328          * if the nulls value we got at the end of this lookup is
 329          * not the expected one, we must restart lookup.
 330          * We probably met an item that was moved to another chain.
 331          */
 332         if (get_nulls_value(n) != bucket) {
 333                 NF_CT_STAT_INC(net, search_restart);
 334                 goto begin;
 335         }
 336         local_bh_enable();
 337
 338         return NULL;
 339 }
 340
 341 struct nf_conntrack_tuple_hash *
 342 __nf_conntrack_find(struct net *net, u16 zone,
 343                     const struct nf_conntrack_tuple *tuple)
 344 {
 345         return ____nf_conntrack_find(net, zone, tuple,
 346                                      hash_conntrack_raw(tuple, zone));
 347 }
 348 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 349
 350 /* Find a connection corresponding to a tuple. */
 351 static struct nf_conntrack_tuple_hash *
 352 __nf_conntrack_find_get(struct net *net, u16 zone,
 353                         const struct nf_conntrack_tuple *tuple, u32 hash)
 354 {
 355         struct nf_conntrack_tuple_hash *h;
 356         struct nf_conn *ct;
 357
 358         rcu_read_lock();
 359 begin:
 360         h = ____nf_conntrack_find(net, zone, tuple, hash);
 361         if (h) {
 362                 ct = nf_ct_tuplehash_to_ctrack(h);
 363                 if (unlikely(nf_ct_is_dying(ct) ||
 364                              !atomic_inc_not_zero(&ct->ct_general.use)))
 365                         h = NULL;
 366                 else {
 367                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 368                                      nf_ct_zone(ct) != zone)) {
 369                                 nf_ct_put(ct);
 370                                 goto begin;
 371                         }
 372                 }
 373         }
 374         rcu_read_unlock();
 375
 376         return h;
 377 }
 378
 379 struct nf_conntrack_tuple_hash *
 380 nf_conntrack_find_get(struct net *net, u16 zone,
 381                       const struct nf_conntrack_tuple *tuple)
 382 {
 383         return __nf_conntrack_find_get(net, zone, tuple,
 384                                        hash_conntrack_raw(tuple, zone));
 385 }
 386 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 387
 388 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 389                                        unsigned int hash,
 390                                        unsigned int repl_hash)
 391 {
 392         struct net *net = nf_ct_net(ct);
 393
 394         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 395                            &net->ct.hash[hash]);
 396         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 397                            &net->ct.hash[repl_hash]);
 398 }
 399
 400 void nf_conntrack_hash_insert(struct nf_conn *ct)
 401 {
 402         struct net *net = nf_ct_net(ct);
 403         unsigned int hash, repl_hash;
 404         u16 zone;
 405
 406         zone = nf_ct_zone(ct);
 407         hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 408         repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 409
 410         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 411 }
 412 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 413
 414 /* Confirm a connection given skb; places it in hash table */
 415 int
 416 __nf_conntrack_confirm(struct sk_buff *skb)
 417 {
 418         unsigned int hash, repl_hash;
 419         struct nf_conntrack_tuple_hash *h;
 420         struct nf_conn *ct;
 421         struct nf_conn_help *help;
 422         struct hlist_nulls_node *n;
 423         enum ip_conntrack_info ctinfo;
 424         struct net *net;
 425         u16 zone;
 426
 427         ct = nf_ct_get(skb, &ctinfo);
 428         net = nf_ct_net(ct);
 429
 430         /* ipt_REJECT uses nf_conntrack_attach to attach related
 431            ICMP/TCP RST packets in other direction.  Actual packet
 432            which created connection will be IP_CT_NEW or for an
 433            expected connection, IP_CT_RELATED. */
 434         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 435                 return NF_ACCEPT;
 436
 437         zone = nf_ct_zone(ct);
 438         /* reuse the hash saved before */
 439         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 440         hash = hash_bucket(hash, net);
 441         repl_hash = hash_conntrack(net, zone,
 442                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 443
 444         /* We're not in hash table, and we refuse to set up related
 445            connections for unconfirmed conns.  But packet copies and
 446            REJECT will give spurious warnings here. */
 447         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 448
 449         /* No external references means noone else could have
 450            confirmed us. */
 451         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 452         pr_debug("Confirming conntrack %p\n", ct);
 453
 454         spin_lock_bh(&nf_conntrack_lock);
 455
 456         /* We have to check the DYING flag inside the lock to prevent
 457            a race against nf_ct_get_next_corpse() possibly called from
 458            user context, else we insert an already 'dead' hash, blocking
 459            further use of that particular connection -JM */
 460
 461         if (unlikely(nf_ct_is_dying(ct))) {
 462                 spin_unlock_bh(&nf_conntrack_lock);
 463                 return NF_ACCEPT;
 464         }
 465
 466         /* See if there's one in the list already, including reverse:
 467            NAT could have grabbed it without realizing, since we're
 468            not in the hash.  If there is, we lost race. */
 469         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 470                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 471                                       &h->tuple) &&
 472                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 473                         goto out;
 474         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 475                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 476                                       &h->tuple) &&
 477                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 478                         goto out;
 479
 480         /* Remove from unconfirmed list */
 481         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 482
 483         /* Timer relative to confirmation time, not original
 484            setting time, otherwise we'd get timer wrap in
 485            weird delay cases. */
 486         ct->timeout.expires += jiffies;
 487         add_timer(&ct->timeout);
 488         atomic_inc(&ct->ct_general.use);
 489         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 490
 491         /* Since the lookup is lockless, hash insertion must be done after
 492          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 493          * guarantee that no other CPU can find the conntrack before the above
 494          * stores are visible.
 495          */
 496         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 497         NF_CT_STAT_INC(net, insert);
 498         spin_unlock_bh(&nf_conntrack_lock);
 499
 500         help = nfct_help(ct);
 501         if (help && help->helper)
 502                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 503
 504         nf_conntrack_event_cache(master_ct(ct) ?
 505                                  IPCT_RELATED : IPCT_NEW, ct);
 506         return NF_ACCEPT;
 507
 508 out:
 509         NF_CT_STAT_INC(net, insert_failed);
 510         spin_unlock_bh(&nf_conntrack_lock);
 511         return NF_DROP;
 512 }
 513 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 514
 515 /* Returns true if a connection correspondings to the tuple (required
 516    for NAT). */
 517 int
 518 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 519                          const struct nf_conn *ignored_conntrack)
 520 {
 521         struct net *net = nf_ct_net(ignored_conntrack);
 522         struct nf_conntrack_tuple_hash *h;
 523         struct hlist_nulls_node *n;
 524         struct nf_conn *ct;
 525         u16 zone = nf_ct_zone(ignored_conntrack);
 526         unsigned int hash = hash_conntrack(net, zone, tuple);
 527
 528         /* Disable BHs the entire time since we need to disable them at
 529          * least once for the stats anyway.
 530          */
 531         rcu_read_lock_bh();
 532         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 533                 ct = nf_ct_tuplehash_to_ctrack(h);
 534                 if (ct != ignored_conntrack &&
 535                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 536                     nf_ct_zone(ct) == zone) {
 537                         NF_CT_STAT_INC(net, found);
 538                         rcu_read_unlock_bh();
 539                         return 1;
 540                 }
 541                 NF_CT_STAT_INC(net, searched);
 542         }
 543         rcu_read_unlock_bh();
 544
 545         return 0;
 546 }
 547 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 548
 549 #define NF_CT_EVICTION_RANGE    8
 550
 551 /* There's a small race here where we may free a just-assured
 552    connection.  Too bad: we're in trouble anyway. */
 553 static noinline int early_drop(struct net *net, unsigned int hash)
 554 {
 555         /* Use oldest entry, which is roughly LRU */
 556         struct nf_conntrack_tuple_hash *h;
 557         struct nf_conn *ct = NULL, *tmp;
 558         struct hlist_nulls_node *n;
 559         unsigned int i, cnt = 0;
 560         int dropped = 0;
 561
 562         rcu_read_lock();
 563         for (i = 0; i < net->ct.htable_size; i++) {
 564                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 565                                          hnnode) {
 566                         tmp = nf_ct_tuplehash_to_ctrack(h);
 567                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 568                                 ct = tmp;
 569                         cnt++;
 570                 }
 571
 572                 if (ct != NULL) {
 573                         if (likely(!nf_ct_is_dying(ct) &&
 574                                    atomic_inc_not_zero(&ct->ct_general.use)))
 575                                 break;
 576                         else
 577                                 ct = NULL;
 578                 }
 579
 580                 if (cnt >= NF_CT_EVICTION_RANGE)
 581                         break;
 582
 583                 hash = (hash + 1) % net->ct.htable_size;
 584         }
 585         rcu_read_unlock();
 586
 587         if (!ct)
 588                 return dropped;
 589
 590         if (del_timer(&ct->timeout)) {
 591                 death_by_timeout((unsigned long)ct);
 592                 dropped = 1;
 593                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 594         }
 595         nf_ct_put(ct);
 596         return dropped;
 597 }
 598
 599 void init_nf_conntrack_hash_rnd(void)
 600 {
 601         unsigned int rand;
 602
 603         /*
 604          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 605          * Because there isn't enough entropy when system initializing,
 606          * and we initialize it as late as possible.
 607          */
 608         do {
 609                 get_random_bytes(&rand, sizeof(rand));
 610         } while (!rand);
 611         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 612 }
 613
 614 static struct nf_conn *
 615 __nf_conntrack_alloc(struct net *net, u16 zone,
 616                      const struct nf_conntrack_tuple *orig,
 617                      const struct nf_conntrack_tuple *repl,
 618                      gfp_t gfp, u32 hash)
 619 {
 620         struct nf_conn *ct;
 621
 622         if (unlikely(!nf_conntrack_hash_rnd)) {
 623                 init_nf_conntrack_hash_rnd();
 624                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 625                 hash = hash_conntrack_raw(orig, zone);
 626         }
 627
 628         /* We don't want any race condition at early drop stage */
 629         atomic_inc(&net->ct.count);
 630
 631         if (nf_conntrack_max &&
 632             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 633                 if (!early_drop(net, hash_bucket(hash, net))) {
 634                         atomic_dec(&net->ct.count);
 635                         if (net_ratelimit())
 636                                 printk(KERN_WARNING
 637                                        "nf_conntrack: table full, dropping"
 638                                        " packet.\n");
 639                         return ERR_PTR(-ENOMEM);
 640                 }
 641         }
 642
 643         /*
 644          * Do not use kmem_cache_zalloc(), as this cache uses
 645          * SLAB_DESTROY_BY_RCU.
 646          */
 647         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 648         if (ct == NULL) {
 649                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
 650                 atomic_dec(&net->ct.count);
 651                 return ERR_PTR(-ENOMEM);
 652         }
 653         /*
 654          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 655          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 656          */
 657         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 658                sizeof(*ct) - offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 659         spin_lock_init(&ct->lock);
 660         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 661         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 662         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 663         /* save hash for reusing when confirming */
 664         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 665         /* Don't set timer yet: wait for confirmation */
 666         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 667         write_pnet(&ct->ct_net, net);
 668 #ifdef CONFIG_NF_CONNTRACK_ZONES
 669         if (zone) {
 670                 struct nf_conntrack_zone *nf_ct_zone;
 671
 672                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 673                 if (!nf_ct_zone)
 674                         goto out_free;
 675                 nf_ct_zone->id = zone;
 676         }
 677 #endif
 678         /*
 679          * changes to lookup keys must be done before setting refcnt to 1
 680          */
 681         smp_wmb();
 682         atomic_set(&ct->ct_general.use, 1);
 683         return ct;
 684
 685 #ifdef CONFIG_NF_CONNTRACK_ZONES
 686 out_free:
 687         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 688         return ERR_PTR(-ENOMEM);
 689 #endif
 690 }
 691
 692 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 693                                    const struct nf_conntrack_tuple *orig,
 694                                    const struct nf_conntrack_tuple *repl,
 695                                    gfp_t gfp)
 696 {
 697         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 698 }
 699 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 700
 701 void nf_conntrack_free(struct nf_conn *ct)
 702 {
 703         struct net *net = nf_ct_net(ct);
 704
 705         nf_ct_ext_destroy(ct);
 706         atomic_dec(&net->ct.count);
 707         nf_ct_ext_free(ct);
 708         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 709 }
 710 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 711
 712 /* Allocate a new conntrack: we return -ENOMEM if classification
 713    failed due to stress.  Otherwise it really is unclassifiable. */
 714 static struct nf_conntrack_tuple_hash *
 715 init_conntrack(struct net *net, struct nf_conn *tmpl,
 716                const struct nf_conntrack_tuple *tuple,
 717                struct nf_conntrack_l3proto *l3proto,
 718                struct nf_conntrack_l4proto *l4proto,
 719                struct sk_buff *skb,
 720                unsigned int dataoff, u32 hash)
 721 {
 722         struct nf_conn *ct;
 723         struct nf_conn_help *help;
 724         struct nf_conntrack_tuple repl_tuple;
 725         struct nf_conntrack_ecache *ecache;
 726         struct nf_conntrack_expect *exp;
 727         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 728
 729         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 730                 pr_debug("Can't invert tuple.\n");
 731                 return NULL;
 732         }
 733
 734         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 735                                   hash);
 736         if (IS_ERR(ct)) {
 737                 pr_debug("Can't allocate conntrack.\n");
 738                 return (struct nf_conntrack_tuple_hash *)ct;
 739         }
 740
 741         if (!l4proto->new(ct, skb, dataoff)) {
 742                 nf_conntrack_free(ct);
 743                 pr_debug("init conntrack: can't track with proto module\n");
 744                 return NULL;
 745         }
 746
 747         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 748
 749         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 750         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 751                                  ecache ? ecache->expmask : 0,
 752                              GFP_ATOMIC);
 753
 754         spin_lock_bh(&nf_conntrack_lock);
 755         exp = nf_ct_find_expectation(net, zone, tuple);
 756         if (exp) {
 757                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 758                          ct, exp);
 759                 /* Welcome, Mr. Bond.  We've been expecting you... */
 760                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 761                 ct->master = exp->master;
 762                 if (exp->helper) {
 763                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 764                         if (help)
 765                                 rcu_assign_pointer(help->helper, exp->helper);
 766                 }
 767
 768 #ifdef CONFIG_NF_CONNTRACK_MARK
 769                 ct->mark = exp->master->mark;
 770 #endif
 771 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 772                 ct->secmark = exp->master->secmark;
 773 #endif
 774                 nf_conntrack_get(&ct->master->ct_general);
 775                 NF_CT_STAT_INC(net, expect_new);
 776         } else {
 777                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 778                 NF_CT_STAT_INC(net, new);
 779         }
 780
 781         /* Overload tuple linked list to put us in unconfirmed list. */
 782         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 783                        &net->ct.unconfirmed);
 784
 785         spin_unlock_bh(&nf_conntrack_lock);
 786
 787         if (exp) {
 788                 if (exp->expectfn)
 789                         exp->expectfn(ct, exp);
 790                 nf_ct_expect_put(exp);
 791         }
 792
 793         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 794 }
 795
 796 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 797 static inline struct nf_conn *
 798 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 799                   struct sk_buff *skb,
 800                   unsigned int dataoff,
 801                   u_int16_t l3num,
 802                   u_int8_t protonum,
 803                   struct nf_conntrack_l3proto *l3proto,
 804                   struct nf_conntrack_l4proto *l4proto,
 805                   int *set_reply,
 806                   enum ip_conntrack_info *ctinfo)
 807 {
 808         struct nf_conntrack_tuple tuple;
 809         struct nf_conntrack_tuple_hash *h;
 810         struct nf_conn *ct;
 811         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 812         u32 hash;
 813
 814         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 815                              dataoff, l3num, protonum, &tuple, l3proto,
 816                              l4proto)) {
 817                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 818                 return NULL;
 819         }
 820
 821         /* look for tuple match */
 822         hash = hash_conntrack_raw(&tuple, zone);
 823         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 824         if (!h) {
 825                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 826                                    skb, dataoff, hash);
 827                 if (!h)
 828                         return NULL;
 829                 if (IS_ERR(h))
 830                         return (void *)h;
 831         }
 832         ct = nf_ct_tuplehash_to_ctrack(h);
 833
 834         /* It exists; we have (non-exclusive) reference. */
 835         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 836                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 837                 /* Please set reply bit if this packet OK */
 838                 *set_reply = 1;
 839         } else {
 840                 /* Once we've had two way comms, always ESTABLISHED. */
 841                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 842                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 843                         *ctinfo = IP_CT_ESTABLISHED;
 844                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 845                         pr_debug("nf_conntrack_in: related packet for %p\n",
 846                                  ct);
 847                         *ctinfo = IP_CT_RELATED;
 848                 } else {
 849                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 850                         *ctinfo = IP_CT_NEW;
 851                 }
 852                 *set_reply = 0;
 853         }
 854         skb->nfct = &ct->ct_general;
 855         skb->nfctinfo = *ctinfo;
 856         return ct;
 857 }
 858
 859 unsigned int
 860 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 861                 struct sk_buff *skb)
 862 {
 863         struct nf_conn *ct, *tmpl = NULL;
 864         enum ip_conntrack_info ctinfo;
 865         struct nf_conntrack_l3proto *l3proto;
 866         struct nf_conntrack_l4proto *l4proto;
 867         unsigned int dataoff;
 868         u_int8_t protonum;
 869         int set_reply = 0;
 870         int ret;
 871
 872         if (skb->nfct) {
 873                 /* Previously seen (loopback or untracked)?  Ignore. */
 874                 tmpl = (struct nf_conn *)skb->nfct;
 875                 if (!nf_ct_is_template(tmpl)) {
 876                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 877                         return NF_ACCEPT;
 878                 }
 879                 skb->nfct = NULL;
 880         }
 881
 882         /* rcu_read_lock()ed by nf_hook_slow */
 883         l3proto = __nf_ct_l3proto_find(pf);
 884         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 885                                    &dataoff, &protonum);
 886         if (ret <= 0) {
 887                 pr_debug("not prepared to track yet or error occured\n");
 888                 NF_CT_STAT_INC_ATOMIC(net, error);
 889                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 890                 ret = -ret;
 891                 goto out;
 892         }
 893
 894         l4proto = __nf_ct_l4proto_find(pf, protonum);
 895
 896         /* It may be an special packet, error, unclean...
 897          * inverse of the return code tells to the netfilter
 898          * core what to do with the packet. */
 899         if (l4proto->error != NULL) {
 900                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 901                                      pf, hooknum);
 902                 if (ret <= 0) {
 903                         NF_CT_STAT_INC_ATOMIC(net, error);
 904                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 905                         ret = -ret;
 906                         goto out;
 907                 }
 908         }
 909
 910         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 911                                l3proto, l4proto, &set_reply, &ctinfo);
 912         if (!ct) {
 913                 /* Not valid part of a connection */
 914                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 915                 ret = NF_ACCEPT;
 916                 goto out;
 917         }
 918
 919         if (IS_ERR(ct)) {
 920                 /* Too stressed to deal. */
 921                 NF_CT_STAT_INC_ATOMIC(net, drop);
 922                 ret = NF_DROP;
 923                 goto out;
 924         }
 925
 926         NF_CT_ASSERT(skb->nfct);
 927
 928         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 929         if (ret <= 0) {
 930                 /* Invalid: inverse of the return code tells
 931                  * the netfilter core what to do */
 932                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 933                 nf_conntrack_put(skb->nfct);
 934                 skb->nfct = NULL;
 935                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 936                 if (ret == -NF_DROP)
 937                         NF_CT_STAT_INC_ATOMIC(net, drop);
 938                 ret = -ret;
 939                 goto out;
 940         }
 941
 942         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 943                 nf_conntrack_event_cache(IPCT_REPLY, ct);
 944 out:
 945         if (tmpl) {
 946                 /* Special case: we have to repeat this hook, assign the
 947                  * template again to this packet. We assume that this packet
 948                  * has no conntrack assigned. This is used by nf_ct_tcp. */
 949                 if (ret == NF_REPEAT)
 950                         skb->nfct = (struct nf_conntrack *)tmpl;
 951                 else
 952                         nf_ct_put(tmpl);
 953         }
 954
 955         return ret;
 956 }
 957 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 958
 959 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 960                           const struct nf_conntrack_tuple *orig)
 961 {
 962         bool ret;
 963
 964         rcu_read_lock();
 965         ret = nf_ct_invert_tuple(inverse, orig,
 966                                  __nf_ct_l3proto_find(orig->src.l3num),
 967                                  __nf_ct_l4proto_find(orig->src.l3num,
 968                                                       orig->dst.protonum));
 969         rcu_read_unlock();
 970         return ret;
 971 }
 972 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 973
 974 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 975    implicitly racy: see __nf_conntrack_confirm */
 976 void nf_conntrack_alter_reply(struct nf_conn *ct,
 977                               const struct nf_conntrack_tuple *newreply)
 978 {
 979         struct nf_conn_help *help = nfct_help(ct);
 980
 981         /* Should be unconfirmed, so not in hash table yet */
 982         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 983
 984         pr_debug("Altering reply tuple of %p to ", ct);
 985         nf_ct_dump_tuple(newreply);
 986
 987         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 988         if (ct->master || (help && !hlist_empty(&help->expectations)))
 989                 return;
 990
 991         rcu_read_lock();
 992         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
 993         rcu_read_unlock();
 994 }
 995 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 996
 997 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 998 void __nf_ct_refresh_acct(struct nf_conn *ct,
 999                           enum ip_conntrack_info ctinfo,
1000                           const struct sk_buff *skb,
1001                           unsigned long extra_jiffies,
1002                           int do_acct)
1003 {
1004         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1005         NF_CT_ASSERT(skb);
1006
1007         /* Only update if this is not a fixed timeout */
1008         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1009                 goto acct;
1010
1011         /* If not in hash table, timer will not be active yet */
1012         if (!nf_ct_is_confirmed(ct)) {
1013                 ct->timeout.expires = extra_jiffies;
1014         } else {
1015                 unsigned long newtime = jiffies + extra_jiffies;
1016
1017                 /* Only update the timeout if the new timeout is at least
1018                    HZ jiffies from the old timeout. Need del_timer for race
1019                    avoidance (may already be dying). */
1020                 if (newtime - ct->timeout.expires >= HZ)
1021                         mod_timer_pending(&ct->timeout, newtime);
1022         }
1023
1024 acct:
1025         if (do_acct) {
1026                 struct nf_conn_counter *acct;
1027
1028                 acct = nf_conn_acct_find(ct);
1029                 if (acct) {
1030                         spin_lock_bh(&ct->lock);
1031                         acct[CTINFO2DIR(ctinfo)].packets++;
1032                         acct[CTINFO2DIR(ctinfo)].bytes += skb->len;
1033                         spin_unlock_bh(&ct->lock);
1034                 }
1035         }
1036 }
1037 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1038
1039 bool __nf_ct_kill_acct(struct nf_conn *ct,
1040                        enum ip_conntrack_info ctinfo,
1041                        const struct sk_buff *skb,
1042                        int do_acct)
1043 {
1044         if (do_acct) {
1045                 struct nf_conn_counter *acct;
1046
1047                 acct = nf_conn_acct_find(ct);
1048                 if (acct) {
1049                         spin_lock_bh(&ct->lock);
1050                         acct[CTINFO2DIR(ctinfo)].packets++;
1051                         acct[CTINFO2DIR(ctinfo)].bytes +=
1052                                 skb->len - skb_network_offset(skb);
1053                         spin_unlock_bh(&ct->lock);
1054                 }
1055         }
1056
1057         if (del_timer(&ct->timeout)) {
1058                 ct->timeout.function((unsigned long)ct);
1059                 return true;
1060         }
1061         return false;
1062 }
1063 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1064
1065 #ifdef CONFIG_NF_CONNTRACK_ZONES
1066 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1067         .len    = sizeof(struct nf_conntrack_zone),
1068         .align  = __alignof__(struct nf_conntrack_zone),
1069         .id     = NF_CT_EXT_ZONE,
1070 };
1071 #endif
1072
1073 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1074
1075 #include <linux/netfilter/nfnetlink.h>
1076 #include <linux/netfilter/nfnetlink_conntrack.h>
1077 #include <linux/mutex.h>
1078
1079 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1080  * in ip_conntrack_core, since we don't want the protocols to autoload
1081  * or depend on ctnetlink */
1082 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1083                                const struct nf_conntrack_tuple *tuple)
1084 {
1085         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
1086         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
1087         return 0;
1088
1089 nla_put_failure:
1090         return -1;
1091 }
1092 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1093
1094 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1095         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1096         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1097 };
1098 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1099
1100 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1101                                struct nf_conntrack_tuple *t)
1102 {
1103         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1104                 return -EINVAL;
1105
1106         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1107         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1108
1109         return 0;
1110 }
1111 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1112
1113 int nf_ct_port_nlattr_tuple_size(void)
1114 {
1115         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1116 }
1117 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1118 #endif
1119
1120 /* Used by ipt_REJECT and ip6t_REJECT. */
1121 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1122 {
1123         struct nf_conn *ct;
1124         enum ip_conntrack_info ctinfo;
1125
1126         /* This ICMP is in reverse direction to the packet which caused it */
1127         ct = nf_ct_get(skb, &ctinfo);
1128         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1129                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
1130         else
1131                 ctinfo = IP_CT_RELATED;
1132
1133         /* Attach to new skbuff, and increment count */
1134         nskb->nfct = &ct->ct_general;
1135         nskb->nfctinfo = ctinfo;
1136         nf_conntrack_get(nskb->nfct);
1137 }
1138
1139 /* Bring out ya dead! */
1140 static struct nf_conn *
1141 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1142                 void *data, unsigned int *bucket)
1143 {
1144         struct nf_conntrack_tuple_hash *h;
1145         struct nf_conn *ct;
1146         struct hlist_nulls_node *n;
1147
1148         spin_lock_bh(&nf_conntrack_lock);
1149         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1150                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1151                         ct = nf_ct_tuplehash_to_ctrack(h);
1152                         if (iter(ct, data))
1153                                 goto found;
1154                 }
1155         }
1156         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1157                 ct = nf_ct_tuplehash_to_ctrack(h);
1158                 if (iter(ct, data))
1159                         set_bit(IPS_DYING_BIT, &ct->status);
1160         }
1161         spin_unlock_bh(&nf_conntrack_lock);
1162         return NULL;
1163 found:
1164         atomic_inc(&ct->ct_general.use);
1165         spin_unlock_bh(&nf_conntrack_lock);
1166         return ct;
1167 }
1168
1169 void nf_ct_iterate_cleanup(struct net *net,
1170                            int (*iter)(struct nf_conn *i, void *data),
1171                            void *data)
1172 {
1173         struct nf_conn *ct;
1174         unsigned int bucket = 0;
1175
1176         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1177                 /* Time to push up daises... */
1178                 if (del_timer(&ct->timeout))
1179                         death_by_timeout((unsigned long)ct);
1180                 /* ... else the timer will get him soon. */
1181
1182                 nf_ct_put(ct);
1183         }
1184 }
1185 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1186
1187 struct __nf_ct_flush_report {
1188         u32 pid;
1189         int report;
1190 };
1191
1192 static int kill_report(struct nf_conn *i, void *data)
1193 {
1194         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1195
1196         /* If we fail to deliver the event, death_by_timeout() will retry */
1197         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1198                                       fr->pid, fr->report) < 0)
1199                 return 1;
1200
1201         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1202         set_bit(IPS_DYING_BIT, &i->status);
1203         return 1;
1204 }
1205
1206 static int kill_all(struct nf_conn *i, void *data)
1207 {
1208         return 1;
1209 }
1210
1211 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1212 {
1213         if (vmalloced)
1214                 vfree(hash);
1215         else
1216                 free_pages((unsigned long)hash,
1217                            get_order(sizeof(struct hlist_head) * size));
1218 }
1219 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1220
1221 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1222 {
1223         struct __nf_ct_flush_report fr = {
1224                 .pid    = pid,
1225                 .report = report,
1226         };
1227         nf_ct_iterate_cleanup(net, kill_report, &fr);
1228 }
1229 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1230
1231 static void nf_ct_release_dying_list(struct net *net)
1232 {
1233         struct nf_conntrack_tuple_hash *h;
1234         struct nf_conn *ct;
1235         struct hlist_nulls_node *n;
1236
1237         spin_lock_bh(&nf_conntrack_lock);
1238         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1239                 ct = nf_ct_tuplehash_to_ctrack(h);
1240                 /* never fails to remove them, no listeners at this point */
1241                 nf_ct_kill(ct);
1242         }
1243         spin_unlock_bh(&nf_conntrack_lock);
1244 }
1245
1246 static int untrack_refs(void)
1247 {
1248         int cnt = 0, cpu;
1249
1250         for_each_possible_cpu(cpu) {
1251                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1252
1253                 cnt += atomic_read(&ct->ct_general.use) - 1;
1254         }
1255         return cnt;
1256 }
1257
1258 static void nf_conntrack_cleanup_init_net(void)
1259 {
1260         while (untrack_refs() > 0)
1261                 schedule();
1262
1263         nf_conntrack_helper_fini();
1264         nf_conntrack_proto_fini();
1265 #ifdef CONFIG_NF_CONNTRACK_ZONES
1266         nf_ct_extend_unregister(&nf_ct_zone_extend);
1267 #endif
1268 }
1269
1270 static void nf_conntrack_cleanup_net(struct net *net)
1271 {
1272  i_see_dead_people:
1273         nf_ct_iterate_cleanup(net, kill_all, NULL);
1274         nf_ct_release_dying_list(net);
1275         if (atomic_read(&net->ct.count) != 0) {
1276                 schedule();
1277                 goto i_see_dead_people;
1278         }
1279
1280         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1281                              net->ct.htable_size);
1282         nf_conntrack_ecache_fini(net);
1283         nf_conntrack_acct_fini(net);
1284         nf_conntrack_expect_fini(net);
1285         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1286         kfree(net->ct.slabname);
1287         free_percpu(net->ct.stat);
1288 }
1289
1290 /* Mishearing the voices in his head, our hero wonders how he's
1291    supposed to kill the mall. */
1292 void nf_conntrack_cleanup(struct net *net)
1293 {
1294         if (net_eq(net, &init_net))
1295                 rcu_assign_pointer(ip_ct_attach, NULL);
1296
1297         /* This makes sure all current packets have passed through
1298            netfilter framework.  Roll on, two-stage module
1299            delete... */
1300         synchronize_net();
1301
1302         nf_conntrack_cleanup_net(net);
1303
1304         if (net_eq(net, &init_net)) {
1305                 rcu_assign_pointer(nf_ct_destroy, NULL);
1306                 nf_conntrack_cleanup_init_net();
1307         }
1308 }
1309
1310 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1311 {
1312         struct hlist_nulls_head *hash;
1313         unsigned int nr_slots, i;
1314         size_t sz;
1315
1316         *vmalloced = 0;
1317
1318         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1319         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1320         sz = nr_slots * sizeof(struct hlist_nulls_head);
1321         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1322                                         get_order(sz));
1323         if (!hash) {
1324                 *vmalloced = 1;
1325                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1326                 hash = __vmalloc(sz, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1327                                  PAGE_KERNEL);
1328         }
1329
1330         if (hash && nulls)
1331                 for (i = 0; i < nr_slots; i++)
1332                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1333
1334         return hash;
1335 }
1336 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1337
1338 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1339 {
1340         int i, bucket, vmalloced, old_vmalloced;
1341         unsigned int hashsize, old_size;
1342         struct hlist_nulls_head *hash, *old_hash;
1343         struct nf_conntrack_tuple_hash *h;
1344         struct nf_conn *ct;
1345
1346         if (current->nsproxy->net_ns != &init_net)
1347                 return -EOPNOTSUPP;
1348
1349         /* On boot, we can set this without any fancy locking. */
1350         if (!nf_conntrack_htable_size)
1351                 return param_set_uint(val, kp);
1352
1353         hashsize = simple_strtoul(val, NULL, 0);
1354         if (!hashsize)
1355                 return -EINVAL;
1356
1357         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1358         if (!hash)
1359                 return -ENOMEM;
1360
1361         /* Lookups in the old hash might happen in parallel, which means we
1362          * might get false negatives during connection lookup. New connections
1363          * created because of a false negative won't make it into the hash
1364          * though since that required taking the lock.
1365          */
1366         spin_lock_bh(&nf_conntrack_lock);
1367         for (i = 0; i < init_net.ct.htable_size; i++) {
1368                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1369                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1370                                         struct nf_conntrack_tuple_hash, hnnode);
1371                         ct = nf_ct_tuplehash_to_ctrack(h);
1372                         hlist_nulls_del_rcu(&h->hnnode);
1373                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1374                                                   hashsize);
1375                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1376                 }
1377         }
1378         old_size = init_net.ct.htable_size;
1379         old_vmalloced = init_net.ct.hash_vmalloc;
1380         old_hash = init_net.ct.hash;
1381
1382         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1383         init_net.ct.hash_vmalloc = vmalloced;
1384         init_net.ct.hash = hash;
1385         spin_unlock_bh(&nf_conntrack_lock);
1386
1387         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1388         return 0;
1389 }
1390 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1391
1392 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1393                   &nf_conntrack_htable_size, 0600);
1394
1395 void nf_ct_untracked_status_or(unsigned long bits)
1396 {
1397         int cpu;
1398
1399         for_each_possible_cpu(cpu)
1400                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1401 }
1402 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1403
1404 static int nf_conntrack_init_init_net(void)
1405 {
1406         int max_factor = 8;
1407         int ret, cpu;
1408
1409         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1410          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1411         if (!nf_conntrack_htable_size) {
1412                 nf_conntrack_htable_size
1413                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1414                            / sizeof(struct hlist_head));
1415                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1416                         nf_conntrack_htable_size = 16384;
1417                 if (nf_conntrack_htable_size < 32)
1418                         nf_conntrack_htable_size = 32;
1419
1420                 /* Use a max. factor of four by default to get the same max as
1421                  * with the old struct list_heads. When a table size is given
1422                  * we use the old value of 8 to avoid reducing the max.
1423                  * entries. */
1424                 max_factor = 4;
1425         }
1426         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1427
1428         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1429                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1430                nf_conntrack_max);
1431
1432         ret = nf_conntrack_proto_init();
1433         if (ret < 0)
1434                 goto err_proto;
1435
1436         ret = nf_conntrack_helper_init();
1437         if (ret < 0)
1438                 goto err_helper;
1439
1440 #ifdef CONFIG_NF_CONNTRACK_ZONES
1441         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1442         if (ret < 0)
1443                 goto err_extend;
1444 #endif
1445         /* Set up fake conntrack: to never be deleted, not in any hashes */
1446         for_each_possible_cpu(cpu) {
1447                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1448                 write_pnet(&ct->ct_net, &init_net);
1449                 atomic_set(&ct->ct_general.use, 1);
1450         }
1451         /*  - and look it like as a confirmed connection */
1452         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1453         return 0;
1454
1455 #ifdef CONFIG_NF_CONNTRACK_ZONES
1456 err_extend:
1457         nf_conntrack_helper_fini();
1458 #endif
1459 err_helper:
1460         nf_conntrack_proto_fini();
1461 err_proto:
1462         return ret;
1463 }
1464
1465 /*
1466  * We need to use special "null" values, not used in hash table
1467  */
1468 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1469 #define DYING_NULLS_VAL         ((1<<30)+1)
1470
1471 static int nf_conntrack_init_net(struct net *net)
1472 {
1473         int ret;
1474
1475         atomic_set(&net->ct.count, 0);
1476         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1477         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1478         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1479         if (!net->ct.stat) {
1480                 ret = -ENOMEM;
1481                 goto err_stat;
1482         }
1483
1484         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1485         if (!net->ct.slabname) {
1486                 ret = -ENOMEM;
1487                 goto err_slabname;
1488         }
1489
1490         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1491                                                         sizeof(struct nf_conn), 0,
1492                                                         SLAB_DESTROY_BY_RCU, NULL);
1493         if (!net->ct.nf_conntrack_cachep) {
1494                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1495                 ret = -ENOMEM;
1496                 goto err_cache;
1497         }
1498
1499         net->ct.htable_size = nf_conntrack_htable_size;
1500         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size,
1501                                              &net->ct.hash_vmalloc, 1);
1502         if (!net->ct.hash) {
1503                 ret = -ENOMEM;
1504                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1505                 goto err_hash;
1506         }
1507         ret = nf_conntrack_expect_init(net);
1508         if (ret < 0)
1509                 goto err_expect;
1510         ret = nf_conntrack_acct_init(net);
1511         if (ret < 0)
1512                 goto err_acct;
1513         ret = nf_conntrack_ecache_init(net);
1514         if (ret < 0)
1515                 goto err_ecache;
1516
1517         return 0;
1518
1519 err_ecache:
1520         nf_conntrack_acct_fini(net);
1521 err_acct:
1522         nf_conntrack_expect_fini(net);
1523 err_expect:
1524         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1525                              net->ct.htable_size);
1526 err_hash:
1527         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1528 err_cache:
1529         kfree(net->ct.slabname);
1530 err_slabname:
1531         free_percpu(net->ct.stat);
1532 err_stat:
1533         return ret;
1534 }
1535
1536 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1537                         enum ip_conntrack_dir dir,
1538                         u32 seq);
1539 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);
1540
1541 int nf_conntrack_init(struct net *net)
1542 {
1543         int ret;
1544
1545         if (net_eq(net, &init_net)) {
1546                 ret = nf_conntrack_init_init_net();
1547                 if (ret < 0)
1548                         goto out_init_net;
1549         }
1550         ret = nf_conntrack_init_net(net);
1551         if (ret < 0)
1552                 goto out_net;
1553
1554         if (net_eq(net, &init_net)) {
1555                 /* For use by REJECT target */
1556                 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1557                 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1558
1559                 /* Howto get NAT offsets */
1560                 rcu_assign_pointer(nf_ct_nat_offset, NULL);
1561         }
1562         return 0;
1563
1564 out_net:
1565         if (net_eq(net, &init_net))
1566                 nf_conntrack_cleanup_init_net();
1567 out_init_net:
1568         return ret;
1569 }