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_conntrack_timestamp.h>
  47 #include <net/netfilter/nf_nat.h>
  48 #include <net/netfilter/nf_nat_core.h>
  49
  50 #define NF_CONNTRACK_VERSION    "0.5.0"
  51
  52 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  53                                       enum nf_nat_manip_type manip,
  54                                       const struct nlattr *attr) __read_mostly;
  55 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  56
  57 DEFINE_SPINLOCK(nf_conntrack_lock);
  58 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  59
  60 unsigned int nf_conntrack_htable_size __read_mostly;
  61 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  62
  63 unsigned int nf_conntrack_max __read_mostly;
  64 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  65
  66 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  67 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  68
  69 unsigned int nf_conntrack_hash_rnd __read_mostly;
  70
  71 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  72 {
  73         unsigned int n;
  74
  75         /* The direction must be ignored, so we hash everything up to the
  76          * destination ports (which is a multiple of 4) and treat the last
  77          * three bytes manually.
  78          */
  79         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  80         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  81                       (((__force __u16)tuple->dst.u.all << 16) |
  82                       tuple->dst.protonum));
  83 }
  84
  85 static u32 __hash_bucket(u32 hash, unsigned int size)
  86 {
  87         return ((u64)hash * size) >> 32;
  88 }
  89
  90 static u32 hash_bucket(u32 hash, const struct net *net)
  91 {
  92         return __hash_bucket(hash, net->ct.htable_size);
  93 }
  94
  95 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  96                                   u16 zone, unsigned int size)
  97 {
  98         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
  99 }
 100
 101 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 102                                        const struct nf_conntrack_tuple *tuple)
 103 {
 104         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 105 }
 106
 107 bool
 108 nf_ct_get_tuple(const struct sk_buff *skb,
 109                 unsigned int nhoff,
 110                 unsigned int dataoff,
 111                 u_int16_t l3num,
 112                 u_int8_t protonum,
 113                 struct nf_conntrack_tuple *tuple,
 114                 const struct nf_conntrack_l3proto *l3proto,
 115                 const struct nf_conntrack_l4proto *l4proto)
 116 {
 117         memset(tuple, 0, sizeof(*tuple));
 118
 119         tuple->src.l3num = l3num;
 120         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 121                 return false;
 122
 123         tuple->dst.protonum = protonum;
 124         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 125
 126         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 127 }
 128 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 129
 130 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 131                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 132 {
 133         struct nf_conntrack_l3proto *l3proto;
 134         struct nf_conntrack_l4proto *l4proto;
 135         unsigned int protoff;
 136         u_int8_t protonum;
 137         int ret;
 138
 139         rcu_read_lock();
 140
 141         l3proto = __nf_ct_l3proto_find(l3num);
 142         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 143         if (ret != NF_ACCEPT) {
 144                 rcu_read_unlock();
 145                 return false;
 146         }
 147
 148         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 149
 150         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 151                               l3proto, l4proto);
 152
 153         rcu_read_unlock();
 154         return ret;
 155 }
 156 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 157
 158 bool
 159 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 160                    const struct nf_conntrack_tuple *orig,
 161                    const struct nf_conntrack_l3proto *l3proto,
 162                    const struct nf_conntrack_l4proto *l4proto)
 163 {
 164         memset(inverse, 0, sizeof(*inverse));
 165
 166         inverse->src.l3num = orig->src.l3num;
 167         if (l3proto->invert_tuple(inverse, orig) == 0)
 168                 return false;
 169
 170         inverse->dst.dir = !orig->dst.dir;
 171
 172         inverse->dst.protonum = orig->dst.protonum;
 173         return l4proto->invert_tuple(inverse, orig);
 174 }
 175 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 176
 177 static void
 178 clean_from_lists(struct nf_conn *ct)
 179 {
 180         pr_debug("clean_from_lists(%p)\n", ct);
 181         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 182         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 183
 184         /* Destroy all pending expectations */
 185         nf_ct_remove_expectations(ct);
 186 }
 187
 188 static void
 189 destroy_conntrack(struct nf_conntrack *nfct)
 190 {
 191         struct nf_conn *ct = (struct nf_conn *)nfct;
 192         struct net *net = nf_ct_net(ct);
 193         struct nf_conntrack_l4proto *l4proto;
 194
 195         pr_debug("destroy_conntrack(%p)\n", ct);
 196         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 197         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 198
 199         /* To make sure we don't get any weird locking issues here:
 200          * destroy_conntrack() MUST NOT be called with a write lock
 201          * to nf_conntrack_lock!!! -HW */
 202         rcu_read_lock();
 203         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 204         if (l4proto && l4proto->destroy)
 205                 l4proto->destroy(ct);
 206
 207         rcu_read_unlock();
 208
 209         spin_lock_bh(&nf_conntrack_lock);
 210         /* Expectations will have been removed in clean_from_lists,
 211          * except TFTP can create an expectation on the first packet,
 212          * before connection is in the list, so we need to clean here,
 213          * too. */
 214         nf_ct_remove_expectations(ct);
 215
 216         /* We overload first tuple to link into unconfirmed list. */
 217         if (!nf_ct_is_confirmed(ct)) {
 218                 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 219                 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 220         }
 221
 222         NF_CT_STAT_INC(net, delete);
 223         spin_unlock_bh(&nf_conntrack_lock);
 224
 225         if (ct->master)
 226                 nf_ct_put(ct->master);
 227
 228         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 229         nf_conntrack_free(ct);
 230 }
 231
 232 void nf_ct_delete_from_lists(struct nf_conn *ct)
 233 {
 234         struct net *net = nf_ct_net(ct);
 235
 236         nf_ct_helper_destroy(ct);
 237         spin_lock_bh(&nf_conntrack_lock);
 238         /* Inside lock so preempt is disabled on module removal path.
 239          * Otherwise we can get spurious warnings. */
 240         NF_CT_STAT_INC(net, delete_list);
 241         clean_from_lists(ct);
 242         spin_unlock_bh(&nf_conntrack_lock);
 243 }
 244 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
 245
 246 static void death_by_event(unsigned long ul_conntrack)
 247 {
 248         struct nf_conn *ct = (void *)ul_conntrack;
 249         struct net *net = nf_ct_net(ct);
 250         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 251
 252         BUG_ON(ecache == NULL);
 253
 254         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 255                 /* bad luck, let's retry again */
 256                 ecache->timeout.expires = jiffies +
 257                         (random32() % net->ct.sysctl_events_retry_timeout);
 258                 add_timer(&ecache->timeout);
 259                 return;
 260         }
 261         /* we've got the event delivered, now it's dying */
 262         set_bit(IPS_DYING_BIT, &ct->status);
 263         spin_lock(&nf_conntrack_lock);
 264         hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 265         spin_unlock(&nf_conntrack_lock);
 266         nf_ct_put(ct);
 267 }
 268
 269 void nf_ct_insert_dying_list(struct nf_conn *ct)
 270 {
 271         struct net *net = nf_ct_net(ct);
 272         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 273
 274         BUG_ON(ecache == NULL);
 275
 276         /* add this conntrack to the dying list */
 277         spin_lock_bh(&nf_conntrack_lock);
 278         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 279                              &net->ct.dying);
 280         spin_unlock_bh(&nf_conntrack_lock);
 281         /* set a new timer to retry event delivery */
 282         setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
 283         ecache->timeout.expires = jiffies +
 284                 (random32() % net->ct.sysctl_events_retry_timeout);
 285         add_timer(&ecache->timeout);
 286 }
 287 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list);
 288
 289 static void death_by_timeout(unsigned long ul_conntrack)
 290 {
 291         struct nf_conn *ct = (void *)ul_conntrack;
 292         struct nf_conn_tstamp *tstamp;
 293
 294         tstamp = nf_conn_tstamp_find(ct);
 295         if (tstamp && tstamp->stop == 0)
 296                 tstamp->stop = ktime_to_ns(ktime_get_real());
 297
 298         if (!test_bit(IPS_DYING_BIT, &ct->status) &&
 299             unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
 300                 /* destroy event was not delivered */
 301                 nf_ct_delete_from_lists(ct);
 302                 nf_ct_insert_dying_list(ct);
 303                 return;
 304         }
 305         set_bit(IPS_DYING_BIT, &ct->status);
 306         nf_ct_delete_from_lists(ct);
 307         nf_ct_put(ct);
 308 }
 309
 310 /*
 311  * Warning :
 312  * - Caller must take a reference on returned object
 313  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 314  * OR
 315  * - Caller must lock nf_conntrack_lock before calling this function
 316  */
 317 static struct nf_conntrack_tuple_hash *
 318 ____nf_conntrack_find(struct net *net, u16 zone,
 319                       const struct nf_conntrack_tuple *tuple, u32 hash)
 320 {
 321         struct nf_conntrack_tuple_hash *h;
 322         struct hlist_nulls_node *n;
 323         unsigned int bucket = hash_bucket(hash, net);
 324
 325         /* Disable BHs the entire time since we normally need to disable them
 326          * at least once for the stats anyway.
 327          */
 328         local_bh_disable();
 329 begin:
 330         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 331                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 332                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 333                         NF_CT_STAT_INC(net, found);
 334                         local_bh_enable();
 335                         return h;
 336                 }
 337                 NF_CT_STAT_INC(net, searched);
 338         }
 339         /*
 340          * if the nulls value we got at the end of this lookup is
 341          * not the expected one, we must restart lookup.
 342          * We probably met an item that was moved to another chain.
 343          */
 344         if (get_nulls_value(n) != bucket) {
 345                 NF_CT_STAT_INC(net, search_restart);
 346                 goto begin;
 347         }
 348         local_bh_enable();
 349
 350         return NULL;
 351 }
 352
 353 struct nf_conntrack_tuple_hash *
 354 __nf_conntrack_find(struct net *net, u16 zone,
 355                     const struct nf_conntrack_tuple *tuple)
 356 {
 357         return ____nf_conntrack_find(net, zone, tuple,
 358                                      hash_conntrack_raw(tuple, zone));
 359 }
 360 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 361
 362 /* Find a connection corresponding to a tuple. */
 363 static struct nf_conntrack_tuple_hash *
 364 __nf_conntrack_find_get(struct net *net, u16 zone,
 365                         const struct nf_conntrack_tuple *tuple, u32 hash)
 366 {
 367         struct nf_conntrack_tuple_hash *h;
 368         struct nf_conn *ct;
 369
 370         rcu_read_lock();
 371 begin:
 372         h = ____nf_conntrack_find(net, zone, tuple, hash);
 373         if (h) {
 374                 ct = nf_ct_tuplehash_to_ctrack(h);
 375                 if (unlikely(nf_ct_is_dying(ct) ||
 376                              !atomic_inc_not_zero(&ct->ct_general.use)))
 377                         h = NULL;
 378                 else {
 379                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 380                                      nf_ct_zone(ct) != zone)) {
 381                                 nf_ct_put(ct);
 382                                 goto begin;
 383                         }
 384                 }
 385         }
 386         rcu_read_unlock();
 387
 388         return h;
 389 }
 390
 391 struct nf_conntrack_tuple_hash *
 392 nf_conntrack_find_get(struct net *net, u16 zone,
 393                       const struct nf_conntrack_tuple *tuple)
 394 {
 395         return __nf_conntrack_find_get(net, zone, tuple,
 396                                        hash_conntrack_raw(tuple, zone));
 397 }
 398 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 399
 400 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 401                                        unsigned int hash,
 402                                        unsigned int repl_hash)
 403 {
 404         struct net *net = nf_ct_net(ct);
 405
 406         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 407                            &net->ct.hash[hash]);
 408         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 409                            &net->ct.hash[repl_hash]);
 410 }
 411
 412 void nf_conntrack_hash_insert(struct nf_conn *ct)
 413 {
 414         struct net *net = nf_ct_net(ct);
 415         unsigned int hash, repl_hash;
 416         u16 zone;
 417
 418         zone = nf_ct_zone(ct);
 419         hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 420         repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 421
 422         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 423 }
 424 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 425
 426 /* Confirm a connection given skb; places it in hash table */
 427 int
 428 __nf_conntrack_confirm(struct sk_buff *skb)
 429 {
 430         unsigned int hash, repl_hash;
 431         struct nf_conntrack_tuple_hash *h;
 432         struct nf_conn *ct;
 433         struct nf_conn_help *help;
 434         struct nf_conn_tstamp *tstamp;
 435         struct hlist_nulls_node *n;
 436         enum ip_conntrack_info ctinfo;
 437         struct net *net;
 438         u16 zone;
 439
 440         ct = nf_ct_get(skb, &ctinfo);
 441         net = nf_ct_net(ct);
 442
 443         /* ipt_REJECT uses nf_conntrack_attach to attach related
 444            ICMP/TCP RST packets in other direction.  Actual packet
 445            which created connection will be IP_CT_NEW or for an
 446            expected connection, IP_CT_RELATED. */
 447         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 448                 return NF_ACCEPT;
 449
 450         zone = nf_ct_zone(ct);
 451         /* reuse the hash saved before */
 452         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 453         hash = hash_bucket(hash, net);
 454         repl_hash = hash_conntrack(net, zone,
 455                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 456
 457         /* We're not in hash table, and we refuse to set up related
 458            connections for unconfirmed conns.  But packet copies and
 459            REJECT will give spurious warnings here. */
 460         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 461
 462         /* No external references means no one else could have
 463            confirmed us. */
 464         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 465         pr_debug("Confirming conntrack %p\n", ct);
 466
 467         spin_lock_bh(&nf_conntrack_lock);
 468
 469         /* We have to check the DYING flag inside the lock to prevent
 470            a race against nf_ct_get_next_corpse() possibly called from
 471            user context, else we insert an already 'dead' hash, blocking
 472            further use of that particular connection -JM */
 473
 474         if (unlikely(nf_ct_is_dying(ct))) {
 475                 spin_unlock_bh(&nf_conntrack_lock);
 476                 return NF_ACCEPT;
 477         }
 478
 479         /* See if there's one in the list already, including reverse:
 480            NAT could have grabbed it without realizing, since we're
 481            not in the hash.  If there is, we lost race. */
 482         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 483                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 484                                       &h->tuple) &&
 485                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 486                         goto out;
 487         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 488                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 489                                       &h->tuple) &&
 490                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 491                         goto out;
 492
 493         /* Remove from unconfirmed list */
 494         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 495
 496         /* Timer relative to confirmation time, not original
 497            setting time, otherwise we'd get timer wrap in
 498            weird delay cases. */
 499         ct->timeout.expires += jiffies;
 500         add_timer(&ct->timeout);
 501         atomic_inc(&ct->ct_general.use);
 502         ct->status |= IPS_CONFIRMED;
 503
 504         /* set conntrack timestamp, if enabled. */
 505         tstamp = nf_conn_tstamp_find(ct);
 506         if (tstamp) {
 507                 if (skb->tstamp.tv64 == 0)
 508                         __net_timestamp((struct sk_buff *)skb);
 509
 510                 tstamp->start = ktime_to_ns(skb->tstamp);
 511         }
 512         /* Since the lookup is lockless, hash insertion must be done after
 513          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 514          * guarantee that no other CPU can find the conntrack before the above
 515          * stores are visible.
 516          */
 517         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 518         NF_CT_STAT_INC(net, insert);
 519         spin_unlock_bh(&nf_conntrack_lock);
 520
 521         help = nfct_help(ct);
 522         if (help && help->helper)
 523                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 524
 525         nf_conntrack_event_cache(master_ct(ct) ?
 526                                  IPCT_RELATED : IPCT_NEW, ct);
 527         return NF_ACCEPT;
 528
 529 out:
 530         NF_CT_STAT_INC(net, insert_failed);
 531         spin_unlock_bh(&nf_conntrack_lock);
 532         return NF_DROP;
 533 }
 534 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 535
 536 /* Returns true if a connection correspondings to the tuple (required
 537    for NAT). */
 538 int
 539 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 540                          const struct nf_conn *ignored_conntrack)
 541 {
 542         struct net *net = nf_ct_net(ignored_conntrack);
 543         struct nf_conntrack_tuple_hash *h;
 544         struct hlist_nulls_node *n;
 545         struct nf_conn *ct;
 546         u16 zone = nf_ct_zone(ignored_conntrack);
 547         unsigned int hash = hash_conntrack(net, zone, tuple);
 548
 549         /* Disable BHs the entire time since we need to disable them at
 550          * least once for the stats anyway.
 551          */
 552         rcu_read_lock_bh();
 553  begin:
 554         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 555                 ct = nf_ct_tuplehash_to_ctrack(h);
 556                 if (ct != ignored_conntrack &&
 557                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 558                     nf_ct_zone(ct) == zone) {
 559                         NF_CT_STAT_INC(net, found);
 560                         rcu_read_unlock_bh();
 561                         return 1;
 562                 }
 563                 NF_CT_STAT_INC(net, searched);
 564         }
 565
 566         if (get_nulls_value(n) != hash) {
 567                 NF_CT_STAT_INC(net, search_restart);
 568                 goto begin;
 569         }
 570
 571         rcu_read_unlock_bh();
 572
 573         return 0;
 574 }
 575 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 576
 577 #define NF_CT_EVICTION_RANGE    8
 578
 579 /* There's a small race here where we may free a just-assured
 580    connection.  Too bad: we're in trouble anyway. */
 581 static noinline int early_drop(struct net *net, unsigned int hash)
 582 {
 583         /* Use oldest entry, which is roughly LRU */
 584         struct nf_conntrack_tuple_hash *h;
 585         struct nf_conn *ct = NULL, *tmp;
 586         struct hlist_nulls_node *n;
 587         unsigned int i, cnt = 0;
 588         int dropped = 0;
 589
 590         rcu_read_lock();
 591         for (i = 0; i < net->ct.htable_size; i++) {
 592                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 593                                          hnnode) {
 594                         tmp = nf_ct_tuplehash_to_ctrack(h);
 595                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 596                                 ct = tmp;
 597                         cnt++;
 598                 }
 599
 600                 if (ct != NULL) {
 601                         if (likely(!nf_ct_is_dying(ct) &&
 602                                    atomic_inc_not_zero(&ct->ct_general.use)))
 603                                 break;
 604                         else
 605                                 ct = NULL;
 606                 }
 607
 608                 if (cnt >= NF_CT_EVICTION_RANGE)
 609                         break;
 610
 611                 hash = (hash + 1) % net->ct.htable_size;
 612         }
 613         rcu_read_unlock();
 614
 615         if (!ct)
 616                 return dropped;
 617
 618         if (del_timer(&ct->timeout)) {
 619                 death_by_timeout((unsigned long)ct);
 620                 dropped = 1;
 621                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 622         }
 623         nf_ct_put(ct);
 624         return dropped;
 625 }
 626
 627 void init_nf_conntrack_hash_rnd(void)
 628 {
 629         unsigned int rand;
 630
 631         /*
 632          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 633          * Because there isn't enough entropy when system initializing,
 634          * and we initialize it as late as possible.
 635          */
 636         do {
 637                 get_random_bytes(&rand, sizeof(rand));
 638         } while (!rand);
 639         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 640 }
 641
 642 static struct nf_conn *
 643 __nf_conntrack_alloc(struct net *net, u16 zone,
 644                      const struct nf_conntrack_tuple *orig,
 645                      const struct nf_conntrack_tuple *repl,
 646                      gfp_t gfp, u32 hash)
 647 {
 648         struct nf_conn *ct;
 649
 650         if (unlikely(!nf_conntrack_hash_rnd)) {
 651                 init_nf_conntrack_hash_rnd();
 652                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 653                 hash = hash_conntrack_raw(orig, zone);
 654         }
 655
 656         /* We don't want any race condition at early drop stage */
 657         atomic_inc(&net->ct.count);
 658
 659         if (nf_conntrack_max &&
 660             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 661                 if (!early_drop(net, hash_bucket(hash, net))) {
 662                         atomic_dec(&net->ct.count);
 663                         if (net_ratelimit())
 664                                 printk(KERN_WARNING
 665                                        "nf_conntrack: table full, dropping"
 666                                        " packet.\n");
 667                         return ERR_PTR(-ENOMEM);
 668                 }
 669         }
 670
 671         /*
 672          * Do not use kmem_cache_zalloc(), as this cache uses
 673          * SLAB_DESTROY_BY_RCU.
 674          */
 675         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 676         if (ct == NULL) {
 677                 atomic_dec(&net->ct.count);
 678                 return ERR_PTR(-ENOMEM);
 679         }
 680         /*
 681          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 682          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 683          */
 684         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 685                offsetof(struct nf_conn, proto) -
 686                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 687         spin_lock_init(&ct->lock);
 688         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 689         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 690         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 691         /* save hash for reusing when confirming */
 692         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 693         /* Don't set timer yet: wait for confirmation */
 694         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 695         write_pnet(&ct->ct_net, net);
 696 #ifdef CONFIG_NF_CONNTRACK_ZONES
 697         if (zone) {
 698                 struct nf_conntrack_zone *nf_ct_zone;
 699
 700                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 701                 if (!nf_ct_zone)
 702                         goto out_free;
 703                 nf_ct_zone->id = zone;
 704         }
 705 #endif
 706         /*
 707          * changes to lookup keys must be done before setting refcnt to 1
 708          */
 709         smp_wmb();
 710         atomic_set(&ct->ct_general.use, 1);
 711         return ct;
 712
 713 #ifdef CONFIG_NF_CONNTRACK_ZONES
 714 out_free:
 715         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 716         return ERR_PTR(-ENOMEM);
 717 #endif
 718 }
 719
 720 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 721                                    const struct nf_conntrack_tuple *orig,
 722                                    const struct nf_conntrack_tuple *repl,
 723                                    gfp_t gfp)
 724 {
 725         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 726 }
 727 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 728
 729 void nf_conntrack_free(struct nf_conn *ct)
 730 {
 731         struct net *net = nf_ct_net(ct);
 732
 733         nf_ct_ext_destroy(ct);
 734         atomic_dec(&net->ct.count);
 735         nf_ct_ext_free(ct);
 736         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 737 }
 738 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 739
 740 /* Allocate a new conntrack: we return -ENOMEM if classification
 741    failed due to stress.  Otherwise it really is unclassifiable. */
 742 static struct nf_conntrack_tuple_hash *
 743 init_conntrack(struct net *net, struct nf_conn *tmpl,
 744                const struct nf_conntrack_tuple *tuple,
 745                struct nf_conntrack_l3proto *l3proto,
 746                struct nf_conntrack_l4proto *l4proto,
 747                struct sk_buff *skb,
 748                unsigned int dataoff, u32 hash)
 749 {
 750         struct nf_conn *ct;
 751         struct nf_conn_help *help;
 752         struct nf_conntrack_tuple repl_tuple;
 753         struct nf_conntrack_ecache *ecache;
 754         struct nf_conntrack_expect *exp;
 755         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 756
 757         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 758                 pr_debug("Can't invert tuple.\n");
 759                 return NULL;
 760         }
 761
 762         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 763                                   hash);
 764         if (IS_ERR(ct))
 765                 return (struct nf_conntrack_tuple_hash *)ct;
 766
 767         if (!l4proto->new(ct, skb, dataoff)) {
 768                 nf_conntrack_free(ct);
 769                 pr_debug("init conntrack: can't track with proto module\n");
 770                 return NULL;
 771         }
 772
 773         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 774         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 775
 776         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 777         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 778                                  ecache ? ecache->expmask : 0,
 779                              GFP_ATOMIC);
 780
 781         spin_lock_bh(&nf_conntrack_lock);
 782         exp = nf_ct_find_expectation(net, zone, tuple);
 783         if (exp) {
 784                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 785                          ct, exp);
 786                 /* Welcome, Mr. Bond.  We've been expecting you... */
 787                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 788                 ct->master = exp->master;
 789                 if (exp->helper) {
 790                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 791                         if (help)
 792                                 rcu_assign_pointer(help->helper, exp->helper);
 793                 }
 794
 795 #ifdef CONFIG_NF_CONNTRACK_MARK
 796                 ct->mark = exp->master->mark;
 797 #endif
 798 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 799                 ct->secmark = exp->master->secmark;
 800 #endif
 801                 nf_conntrack_get(&ct->master->ct_general);
 802                 NF_CT_STAT_INC(net, expect_new);
 803         } else {
 804                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 805                 NF_CT_STAT_INC(net, new);
 806         }
 807
 808         /* Overload tuple linked list to put us in unconfirmed list. */
 809         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 810                        &net->ct.unconfirmed);
 811
 812         spin_unlock_bh(&nf_conntrack_lock);
 813
 814         if (exp) {
 815                 if (exp->expectfn)
 816                         exp->expectfn(ct, exp);
 817                 nf_ct_expect_put(exp);
 818         }
 819
 820         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 821 }
 822
 823 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 824 static inline struct nf_conn *
 825 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 826                   struct sk_buff *skb,
 827                   unsigned int dataoff,
 828                   u_int16_t l3num,
 829                   u_int8_t protonum,
 830                   struct nf_conntrack_l3proto *l3proto,
 831                   struct nf_conntrack_l4proto *l4proto,
 832                   int *set_reply,
 833                   enum ip_conntrack_info *ctinfo)
 834 {
 835         struct nf_conntrack_tuple tuple;
 836         struct nf_conntrack_tuple_hash *h;
 837         struct nf_conn *ct;
 838         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 839         u32 hash;
 840
 841         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 842                              dataoff, l3num, protonum, &tuple, l3proto,
 843                              l4proto)) {
 844                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 845                 return NULL;
 846         }
 847
 848         /* look for tuple match */
 849         hash = hash_conntrack_raw(&tuple, zone);
 850         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 851         if (!h) {
 852                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 853                                    skb, dataoff, hash);
 854                 if (!h)
 855                         return NULL;
 856                 if (IS_ERR(h))
 857                         return (void *)h;
 858         }
 859         ct = nf_ct_tuplehash_to_ctrack(h);
 860
 861         /* It exists; we have (non-exclusive) reference. */
 862         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 863                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 864                 /* Please set reply bit if this packet OK */
 865                 *set_reply = 1;
 866         } else {
 867                 /* Once we've had two way comms, always ESTABLISHED. */
 868                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 869                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 870                         *ctinfo = IP_CT_ESTABLISHED;
 871                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 872                         pr_debug("nf_conntrack_in: related packet for %p\n",
 873                                  ct);
 874                         *ctinfo = IP_CT_RELATED;
 875                 } else {
 876                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 877                         *ctinfo = IP_CT_NEW;
 878                 }
 879                 *set_reply = 0;
 880         }
 881         skb->nfct = &ct->ct_general;
 882         skb->nfctinfo = *ctinfo;
 883         return ct;
 884 }
 885
 886 unsigned int
 887 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 888                 struct sk_buff *skb)
 889 {
 890         struct nf_conn *ct, *tmpl = NULL;
 891         enum ip_conntrack_info ctinfo;
 892         struct nf_conntrack_l3proto *l3proto;
 893         struct nf_conntrack_l4proto *l4proto;
 894         unsigned int dataoff;
 895         u_int8_t protonum;
 896         int set_reply = 0;
 897         int ret;
 898
 899         if (skb->nfct) {
 900                 /* Previously seen (loopback or untracked)?  Ignore. */
 901                 tmpl = (struct nf_conn *)skb->nfct;
 902                 if (!nf_ct_is_template(tmpl)) {
 903                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 904                         return NF_ACCEPT;
 905                 }
 906                 skb->nfct = NULL;
 907         }
 908
 909         /* rcu_read_lock()ed by nf_hook_slow */
 910         l3proto = __nf_ct_l3proto_find(pf);
 911         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 912                                    &dataoff, &protonum);
 913         if (ret <= 0) {
 914                 pr_debug("not prepared to track yet or error occurred\n");
 915                 NF_CT_STAT_INC_ATOMIC(net, error);
 916                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 917                 ret = -ret;
 918                 goto out;
 919         }
 920
 921         l4proto = __nf_ct_l4proto_find(pf, protonum);
 922
 923         /* It may be an special packet, error, unclean...
 924          * inverse of the return code tells to the netfilter
 925          * core what to do with the packet. */
 926         if (l4proto->error != NULL) {
 927                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 928                                      pf, hooknum);
 929                 if (ret <= 0) {
 930                         NF_CT_STAT_INC_ATOMIC(net, error);
 931                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 932                         ret = -ret;
 933                         goto out;
 934                 }
 935                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 936                 if (skb->nfct)
 937                         goto out;
 938         }
 939
 940         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 941                                l3proto, l4proto, &set_reply, &ctinfo);
 942         if (!ct) {
 943                 /* Not valid part of a connection */
 944                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 945                 ret = NF_ACCEPT;
 946                 goto out;
 947         }
 948
 949         if (IS_ERR(ct)) {
 950                 /* Too stressed to deal. */
 951                 NF_CT_STAT_INC_ATOMIC(net, drop);
 952                 ret = NF_DROP;
 953                 goto out;
 954         }
 955
 956         NF_CT_ASSERT(skb->nfct);
 957
 958         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 959         if (ret <= 0) {
 960                 /* Invalid: inverse of the return code tells
 961                  * the netfilter core what to do */
 962                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 963                 nf_conntrack_put(skb->nfct);
 964                 skb->nfct = NULL;
 965                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 966                 if (ret == -NF_DROP)
 967                         NF_CT_STAT_INC_ATOMIC(net, drop);
 968                 ret = -ret;
 969                 goto out;
 970         }
 971
 972         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 973                 nf_conntrack_event_cache(IPCT_REPLY, ct);
 974 out:
 975         if (tmpl) {
 976                 /* Special case: we have to repeat this hook, assign the
 977                  * template again to this packet. We assume that this packet
 978                  * has no conntrack assigned. This is used by nf_ct_tcp. */
 979                 if (ret == NF_REPEAT)
 980                         skb->nfct = (struct nf_conntrack *)tmpl;
 981                 else
 982                         nf_ct_put(tmpl);
 983         }
 984
 985         return ret;
 986 }
 987 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 988
 989 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 990                           const struct nf_conntrack_tuple *orig)
 991 {
 992         bool ret;
 993
 994         rcu_read_lock();
 995         ret = nf_ct_invert_tuple(inverse, orig,
 996                                  __nf_ct_l3proto_find(orig->src.l3num),
 997                                  __nf_ct_l4proto_find(orig->src.l3num,
 998                                                       orig->dst.protonum));
 999         rcu_read_unlock();
1000         return ret;
1001 }
1002 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1003
1004 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1005    implicitly racy: see __nf_conntrack_confirm */
1006 void nf_conntrack_alter_reply(struct nf_conn *ct,
1007                               const struct nf_conntrack_tuple *newreply)
1008 {
1009         struct nf_conn_help *help = nfct_help(ct);
1010
1011         /* Should be unconfirmed, so not in hash table yet */
1012         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1013
1014         pr_debug("Altering reply tuple of %p to ", ct);
1015         nf_ct_dump_tuple(newreply);
1016
1017         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1018         if (ct->master || (help && !hlist_empty(&help->expectations)))
1019                 return;
1020
1021         rcu_read_lock();
1022         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1023         rcu_read_unlock();
1024 }
1025 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1026
1027 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1028 void __nf_ct_refresh_acct(struct nf_conn *ct,
1029                           enum ip_conntrack_info ctinfo,
1030                           const struct sk_buff *skb,
1031                           unsigned long extra_jiffies,
1032                           int do_acct)
1033 {
1034         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1035         NF_CT_ASSERT(skb);
1036
1037         /* Only update if this is not a fixed timeout */
1038         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1039                 goto acct;
1040
1041         /* If not in hash table, timer will not be active yet */
1042         if (!nf_ct_is_confirmed(ct)) {
1043                 ct->timeout.expires = extra_jiffies;
1044         } else {
1045                 unsigned long newtime = jiffies + extra_jiffies;
1046
1047                 /* Only update the timeout if the new timeout is at least
1048                    HZ jiffies from the old timeout. Need del_timer for race
1049                    avoidance (may already be dying). */
1050                 if (newtime - ct->timeout.expires >= HZ)
1051                         mod_timer_pending(&ct->timeout, newtime);
1052         }
1053
1054 acct:
1055         if (do_acct) {
1056                 struct nf_conn_counter *acct;
1057
1058                 acct = nf_conn_acct_find(ct);
1059                 if (acct) {
1060                         spin_lock_bh(&ct->lock);
1061                         acct[CTINFO2DIR(ctinfo)].packets++;
1062                         acct[CTINFO2DIR(ctinfo)].bytes += skb->len;
1063                         spin_unlock_bh(&ct->lock);
1064                 }
1065         }
1066 }
1067 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1068
1069 bool __nf_ct_kill_acct(struct nf_conn *ct,
1070                        enum ip_conntrack_info ctinfo,
1071                        const struct sk_buff *skb,
1072                        int do_acct)
1073 {
1074         if (do_acct) {
1075                 struct nf_conn_counter *acct;
1076
1077                 acct = nf_conn_acct_find(ct);
1078                 if (acct) {
1079                         spin_lock_bh(&ct->lock);
1080                         acct[CTINFO2DIR(ctinfo)].packets++;
1081                         acct[CTINFO2DIR(ctinfo)].bytes +=
1082                                 skb->len - skb_network_offset(skb);
1083                         spin_unlock_bh(&ct->lock);
1084                 }
1085         }
1086
1087         if (del_timer(&ct->timeout)) {
1088                 ct->timeout.function((unsigned long)ct);
1089                 return true;
1090         }
1091         return false;
1092 }
1093 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1094
1095 #ifdef CONFIG_NF_CONNTRACK_ZONES
1096 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1097         .len    = sizeof(struct nf_conntrack_zone),
1098         .align  = __alignof__(struct nf_conntrack_zone),
1099         .id     = NF_CT_EXT_ZONE,
1100 };
1101 #endif
1102
1103 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1104
1105 #include <linux/netfilter/nfnetlink.h>
1106 #include <linux/netfilter/nfnetlink_conntrack.h>
1107 #include <linux/mutex.h>
1108
1109 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1110  * in ip_conntrack_core, since we don't want the protocols to autoload
1111  * or depend on ctnetlink */
1112 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1113                                const struct nf_conntrack_tuple *tuple)
1114 {
1115         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
1116         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
1117         return 0;
1118
1119 nla_put_failure:
1120         return -1;
1121 }
1122 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1123
1124 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1125         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1126         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1127 };
1128 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1129
1130 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1131                                struct nf_conntrack_tuple *t)
1132 {
1133         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1134                 return -EINVAL;
1135
1136         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1137         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1138
1139         return 0;
1140 }
1141 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1142
1143 int nf_ct_port_nlattr_tuple_size(void)
1144 {
1145         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1146 }
1147 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1148 #endif
1149
1150 /* Used by ipt_REJECT and ip6t_REJECT. */
1151 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1152 {
1153         struct nf_conn *ct;
1154         enum ip_conntrack_info ctinfo;
1155
1156         /* This ICMP is in reverse direction to the packet which caused it */
1157         ct = nf_ct_get(skb, &ctinfo);
1158         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1159                 ctinfo = IP_CT_RELATED_REPLY;
1160         else
1161                 ctinfo = IP_CT_RELATED;
1162
1163         /* Attach to new skbuff, and increment count */
1164         nskb->nfct = &ct->ct_general;
1165         nskb->nfctinfo = ctinfo;
1166         nf_conntrack_get(nskb->nfct);
1167 }
1168
1169 /* Bring out ya dead! */
1170 static struct nf_conn *
1171 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1172                 void *data, unsigned int *bucket)
1173 {
1174         struct nf_conntrack_tuple_hash *h;
1175         struct nf_conn *ct;
1176         struct hlist_nulls_node *n;
1177
1178         spin_lock_bh(&nf_conntrack_lock);
1179         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1180                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1181                         ct = nf_ct_tuplehash_to_ctrack(h);
1182                         if (iter(ct, data))
1183                                 goto found;
1184                 }
1185         }
1186         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1187                 ct = nf_ct_tuplehash_to_ctrack(h);
1188                 if (iter(ct, data))
1189                         set_bit(IPS_DYING_BIT, &ct->status);
1190         }
1191         spin_unlock_bh(&nf_conntrack_lock);
1192         return NULL;
1193 found:
1194         atomic_inc(&ct->ct_general.use);
1195         spin_unlock_bh(&nf_conntrack_lock);
1196         return ct;
1197 }
1198
1199 void nf_ct_iterate_cleanup(struct net *net,
1200                            int (*iter)(struct nf_conn *i, void *data),
1201                            void *data)
1202 {
1203         struct nf_conn *ct;
1204         unsigned int bucket = 0;
1205
1206         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1207                 /* Time to push up daises... */
1208                 if (del_timer(&ct->timeout))
1209                         death_by_timeout((unsigned long)ct);
1210                 /* ... else the timer will get him soon. */
1211
1212                 nf_ct_put(ct);
1213         }
1214 }
1215 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1216
1217 struct __nf_ct_flush_report {
1218         u32 pid;
1219         int report;
1220 };
1221
1222 static int kill_report(struct nf_conn *i, void *data)
1223 {
1224         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1225         struct nf_conn_tstamp *tstamp;
1226
1227         tstamp = nf_conn_tstamp_find(i);
1228         if (tstamp && tstamp->stop == 0)
1229                 tstamp->stop = ktime_to_ns(ktime_get_real());
1230
1231         /* If we fail to deliver the event, death_by_timeout() will retry */
1232         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1233                                       fr->pid, fr->report) < 0)
1234                 return 1;
1235
1236         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1237         set_bit(IPS_DYING_BIT, &i->status);
1238         return 1;
1239 }
1240
1241 static int kill_all(struct nf_conn *i, void *data)
1242 {
1243         return 1;
1244 }
1245
1246 void nf_ct_free_hashtable(void *hash, unsigned int size)
1247 {
1248         if (is_vmalloc_addr(hash))
1249                 vfree(hash);
1250         else
1251                 free_pages((unsigned long)hash,
1252                            get_order(sizeof(struct hlist_head) * size));
1253 }
1254 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1255
1256 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1257 {
1258         struct __nf_ct_flush_report fr = {
1259                 .pid    = pid,
1260                 .report = report,
1261         };
1262         nf_ct_iterate_cleanup(net, kill_report, &fr);
1263 }
1264 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1265
1266 static void nf_ct_release_dying_list(struct net *net)
1267 {
1268         struct nf_conntrack_tuple_hash *h;
1269         struct nf_conn *ct;
1270         struct hlist_nulls_node *n;
1271
1272         spin_lock_bh(&nf_conntrack_lock);
1273         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1274                 ct = nf_ct_tuplehash_to_ctrack(h);
1275                 /* never fails to remove them, no listeners at this point */
1276                 nf_ct_kill(ct);
1277         }
1278         spin_unlock_bh(&nf_conntrack_lock);
1279 }
1280
1281 static int untrack_refs(void)
1282 {
1283         int cnt = 0, cpu;
1284
1285         for_each_possible_cpu(cpu) {
1286                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1287
1288                 cnt += atomic_read(&ct->ct_general.use) - 1;
1289         }
1290         return cnt;
1291 }
1292
1293 static void nf_conntrack_cleanup_init_net(void)
1294 {
1295         while (untrack_refs() > 0)
1296                 schedule();
1297
1298         nf_conntrack_helper_fini();
1299         nf_conntrack_proto_fini();
1300 #ifdef CONFIG_NF_CONNTRACK_ZONES
1301         nf_ct_extend_unregister(&nf_ct_zone_extend);
1302 #endif
1303 }
1304
1305 static void nf_conntrack_cleanup_net(struct net *net)
1306 {
1307  i_see_dead_people:
1308         nf_ct_iterate_cleanup(net, kill_all, NULL);
1309         nf_ct_release_dying_list(net);
1310         if (atomic_read(&net->ct.count) != 0) {
1311                 schedule();
1312                 goto i_see_dead_people;
1313         }
1314
1315         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1316         nf_conntrack_ecache_fini(net);
1317         nf_conntrack_tstamp_fini(net);
1318         nf_conntrack_acct_fini(net);
1319         nf_conntrack_expect_fini(net);
1320         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1321         kfree(net->ct.slabname);
1322         free_percpu(net->ct.stat);
1323 }
1324
1325 /* Mishearing the voices in his head, our hero wonders how he's
1326    supposed to kill the mall. */
1327 void nf_conntrack_cleanup(struct net *net)
1328 {
1329         if (net_eq(net, &init_net))
1330                 RCU_INIT_POINTER(ip_ct_attach, NULL);
1331
1332         /* This makes sure all current packets have passed through
1333            netfilter framework.  Roll on, two-stage module
1334            delete... */
1335         synchronize_net();
1336
1337         nf_conntrack_cleanup_net(net);
1338
1339         if (net_eq(net, &init_net)) {
1340                 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1341                 nf_conntrack_cleanup_init_net();
1342         }
1343 }
1344
1345 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1346 {
1347         struct hlist_nulls_head *hash;
1348         unsigned int nr_slots, i;
1349         size_t sz;
1350
1351         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1352         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1353         sz = nr_slots * sizeof(struct hlist_nulls_head);
1354         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1355                                         get_order(sz));
1356         if (!hash) {
1357                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1358                 hash = __vmalloc(sz, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1359                                  PAGE_KERNEL);
1360         }
1361
1362         if (hash && nulls)
1363                 for (i = 0; i < nr_slots; i++)
1364                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1365
1366         return hash;
1367 }
1368 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1369
1370 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1371 {
1372         int i, bucket;
1373         unsigned int hashsize, old_size;
1374         struct hlist_nulls_head *hash, *old_hash;
1375         struct nf_conntrack_tuple_hash *h;
1376         struct nf_conn *ct;
1377
1378         if (current->nsproxy->net_ns != &init_net)
1379                 return -EOPNOTSUPP;
1380
1381         /* On boot, we can set this without any fancy locking. */
1382         if (!nf_conntrack_htable_size)
1383                 return param_set_uint(val, kp);
1384
1385         hashsize = simple_strtoul(val, NULL, 0);
1386         if (!hashsize)
1387                 return -EINVAL;
1388
1389         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1390         if (!hash)
1391                 return -ENOMEM;
1392
1393         /* Lookups in the old hash might happen in parallel, which means we
1394          * might get false negatives during connection lookup. New connections
1395          * created because of a false negative won't make it into the hash
1396          * though since that required taking the lock.
1397          */
1398         spin_lock_bh(&nf_conntrack_lock);
1399         for (i = 0; i < init_net.ct.htable_size; i++) {
1400                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1401                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1402                                         struct nf_conntrack_tuple_hash, hnnode);
1403                         ct = nf_ct_tuplehash_to_ctrack(h);
1404                         hlist_nulls_del_rcu(&h->hnnode);
1405                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1406                                                   hashsize);
1407                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1408                 }
1409         }
1410         old_size = init_net.ct.htable_size;
1411         old_hash = init_net.ct.hash;
1412
1413         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1414         init_net.ct.hash = hash;
1415         spin_unlock_bh(&nf_conntrack_lock);
1416
1417         nf_ct_free_hashtable(old_hash, old_size);
1418         return 0;
1419 }
1420 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1421
1422 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1423                   &nf_conntrack_htable_size, 0600);
1424
1425 void nf_ct_untracked_status_or(unsigned long bits)
1426 {
1427         int cpu;
1428
1429         for_each_possible_cpu(cpu)
1430                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1431 }
1432 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1433
1434 static int nf_conntrack_init_init_net(void)
1435 {
1436         int max_factor = 8;
1437         int ret, cpu;
1438
1439         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1440          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1441         if (!nf_conntrack_htable_size) {
1442                 nf_conntrack_htable_size
1443                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1444                            / sizeof(struct hlist_head));
1445                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1446                         nf_conntrack_htable_size = 16384;
1447                 if (nf_conntrack_htable_size < 32)
1448                         nf_conntrack_htable_size = 32;
1449
1450                 /* Use a max. factor of four by default to get the same max as
1451                  * with the old struct list_heads. When a table size is given
1452                  * we use the old value of 8 to avoid reducing the max.
1453                  * entries. */
1454                 max_factor = 4;
1455         }
1456         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1457
1458         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1459                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1460                nf_conntrack_max);
1461
1462         ret = nf_conntrack_proto_init();
1463         if (ret < 0)
1464                 goto err_proto;
1465
1466         ret = nf_conntrack_helper_init();
1467         if (ret < 0)
1468                 goto err_helper;
1469
1470 #ifdef CONFIG_NF_CONNTRACK_ZONES
1471         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1472         if (ret < 0)
1473                 goto err_extend;
1474 #endif
1475         /* Set up fake conntrack: to never be deleted, not in any hashes */
1476         for_each_possible_cpu(cpu) {
1477                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1478                 write_pnet(&ct->ct_net, &init_net);
1479                 atomic_set(&ct->ct_general.use, 1);
1480         }
1481         /*  - and look it like as a confirmed connection */
1482         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1483         return 0;
1484
1485 #ifdef CONFIG_NF_CONNTRACK_ZONES
1486 err_extend:
1487         nf_conntrack_helper_fini();
1488 #endif
1489 err_helper:
1490         nf_conntrack_proto_fini();
1491 err_proto:
1492         return ret;
1493 }
1494
1495 /*
1496  * We need to use special "null" values, not used in hash table
1497  */
1498 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1499 #define DYING_NULLS_VAL         ((1<<30)+1)
1500
1501 static int nf_conntrack_init_net(struct net *net)
1502 {
1503         static atomic64_t unique_id;
1504         int ret;
1505
1506         atomic_set(&net->ct.count, 0);
1507         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1508         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1509         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1510         if (!net->ct.stat) {
1511                 ret = -ENOMEM;
1512                 goto err_stat;
1513         }
1514
1515         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%llu",
1516                                 (u64)atomic64_inc_return(&unique_id));
1517         if (!net->ct.slabname) {
1518                 ret = -ENOMEM;
1519                 goto err_slabname;
1520         }
1521
1522         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1523                                                         sizeof(struct nf_conn), 0,
1524                                                         SLAB_DESTROY_BY_RCU, NULL);
1525         if (!net->ct.nf_conntrack_cachep) {
1526                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1527                 ret = -ENOMEM;
1528                 goto err_cache;
1529         }
1530
1531         net->ct.htable_size = nf_conntrack_htable_size;
1532         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1533         if (!net->ct.hash) {
1534                 ret = -ENOMEM;
1535                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1536                 goto err_hash;
1537         }
1538         ret = nf_conntrack_expect_init(net);
1539         if (ret < 0)
1540                 goto err_expect;
1541         ret = nf_conntrack_acct_init(net);
1542         if (ret < 0)
1543                 goto err_acct;
1544         ret = nf_conntrack_tstamp_init(net);
1545         if (ret < 0)
1546                 goto err_tstamp;
1547         ret = nf_conntrack_ecache_init(net);
1548         if (ret < 0)
1549                 goto err_ecache;
1550
1551         return 0;
1552
1553 err_ecache:
1554         nf_conntrack_tstamp_fini(net);
1555 err_tstamp:
1556         nf_conntrack_acct_fini(net);
1557 err_acct:
1558         nf_conntrack_expect_fini(net);
1559 err_expect:
1560         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1561 err_hash:
1562         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1563 err_cache:
1564         kfree(net->ct.slabname);
1565 err_slabname:
1566         free_percpu(net->ct.stat);
1567 err_stat:
1568         return ret;
1569 }
1570
1571 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1572                         enum ip_conntrack_dir dir,
1573                         u32 seq);
1574 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);
1575
1576 int nf_conntrack_init(struct net *net)
1577 {
1578         int ret;
1579
1580         if (net_eq(net, &init_net)) {
1581                 ret = nf_conntrack_init_init_net();
1582                 if (ret < 0)
1583                         goto out_init_net;
1584         }
1585         ret = nf_conntrack_init_net(net);
1586         if (ret < 0)
1587                 goto out_net;
1588
1589         if (net_eq(net, &init_net)) {
1590                 /* For use by REJECT target */
1591                 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1592                 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1593
1594                 /* Howto get NAT offsets */
1595                 RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
1596         }
1597         return 0;
1598
1599 out_net:
1600         if (net_eq(net, &init_net))
1601                 nf_conntrack_cleanup_init_net();
1602 out_init_net:
1603         return ret;
1604 }