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         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 554                 ct = nf_ct_tuplehash_to_ctrack(h);
 555                 if (ct != ignored_conntrack &&
 556                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 557                     nf_ct_zone(ct) == zone) {
 558                         NF_CT_STAT_INC(net, found);
 559                         rcu_read_unlock_bh();
 560                         return 1;
 561                 }
 562                 NF_CT_STAT_INC(net, searched);
 563         }
 564         rcu_read_unlock_bh();
 565
 566         return 0;
 567 }
 568 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 569
 570 #define NF_CT_EVICTION_RANGE    8
 571
 572 /* There's a small race here where we may free a just-assured
 573    connection.  Too bad: we're in trouble anyway. */
 574 static noinline int early_drop(struct net *net, unsigned int hash)
 575 {
 576         /* Use oldest entry, which is roughly LRU */
 577         struct nf_conntrack_tuple_hash *h;
 578         struct nf_conn *ct = NULL, *tmp;
 579         struct hlist_nulls_node *n;
 580         unsigned int i, cnt = 0;
 581         int dropped = 0;
 582
 583         rcu_read_lock();
 584         for (i = 0; i < net->ct.htable_size; i++) {
 585                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 586                                          hnnode) {
 587                         tmp = nf_ct_tuplehash_to_ctrack(h);
 588                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 589                                 ct = tmp;
 590                         cnt++;
 591                 }
 592
 593                 if (ct != NULL) {
 594                         if (likely(!nf_ct_is_dying(ct) &&
 595                                    atomic_inc_not_zero(&ct->ct_general.use)))
 596                                 break;
 597                         else
 598                                 ct = NULL;
 599                 }
 600
 601                 if (cnt >= NF_CT_EVICTION_RANGE)
 602                         break;
 603
 604                 hash = (hash + 1) % net->ct.htable_size;
 605         }
 606         rcu_read_unlock();
 607
 608         if (!ct)
 609                 return dropped;
 610
 611         if (del_timer(&ct->timeout)) {
 612                 death_by_timeout((unsigned long)ct);
 613                 dropped = 1;
 614                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 615         }
 616         nf_ct_put(ct);
 617         return dropped;
 618 }
 619
 620 void init_nf_conntrack_hash_rnd(void)
 621 {
 622         unsigned int rand;
 623
 624         /*
 625          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 626          * Because there isn't enough entropy when system initializing,
 627          * and we initialize it as late as possible.
 628          */
 629         do {
 630                 get_random_bytes(&rand, sizeof(rand));
 631         } while (!rand);
 632         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 633 }
 634
 635 static struct nf_conn *
 636 __nf_conntrack_alloc(struct net *net, u16 zone,
 637                      const struct nf_conntrack_tuple *orig,
 638                      const struct nf_conntrack_tuple *repl,
 639                      gfp_t gfp, u32 hash)
 640 {
 641         struct nf_conn *ct;
 642
 643         if (unlikely(!nf_conntrack_hash_rnd)) {
 644                 init_nf_conntrack_hash_rnd();
 645                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 646                 hash = hash_conntrack_raw(orig, zone);
 647         }
 648
 649         /* We don't want any race condition at early drop stage */
 650         atomic_inc(&net->ct.count);
 651
 652         if (nf_conntrack_max &&
 653             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 654                 if (!early_drop(net, hash_bucket(hash, net))) {
 655                         atomic_dec(&net->ct.count);
 656                         if (net_ratelimit())
 657                                 printk(KERN_WARNING
 658                                        "nf_conntrack: table full, dropping"
 659                                        " packet.\n");
 660                         return ERR_PTR(-ENOMEM);
 661                 }
 662         }
 663
 664         /*
 665          * Do not use kmem_cache_zalloc(), as this cache uses
 666          * SLAB_DESTROY_BY_RCU.
 667          */
 668         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 669         if (ct == NULL) {
 670                 atomic_dec(&net->ct.count);
 671                 return ERR_PTR(-ENOMEM);
 672         }
 673         /*
 674          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 675          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 676          */
 677         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 678                offsetof(struct nf_conn, proto) -
 679                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 680         spin_lock_init(&ct->lock);
 681         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 682         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 683         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 684         /* save hash for reusing when confirming */
 685         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 686         /* Don't set timer yet: wait for confirmation */
 687         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 688         write_pnet(&ct->ct_net, net);
 689 #ifdef CONFIG_NF_CONNTRACK_ZONES
 690         if (zone) {
 691                 struct nf_conntrack_zone *nf_ct_zone;
 692
 693                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 694                 if (!nf_ct_zone)
 695                         goto out_free;
 696                 nf_ct_zone->id = zone;
 697         }
 698 #endif
 699         /*
 700          * changes to lookup keys must be done before setting refcnt to 1
 701          */
 702         smp_wmb();
 703         atomic_set(&ct->ct_general.use, 1);
 704         return ct;
 705
 706 #ifdef CONFIG_NF_CONNTRACK_ZONES
 707 out_free:
 708         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 709         return ERR_PTR(-ENOMEM);
 710 #endif
 711 }
 712
 713 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 714                                    const struct nf_conntrack_tuple *orig,
 715                                    const struct nf_conntrack_tuple *repl,
 716                                    gfp_t gfp)
 717 {
 718         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 719 }
 720 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 721
 722 void nf_conntrack_free(struct nf_conn *ct)
 723 {
 724         struct net *net = nf_ct_net(ct);
 725
 726         nf_ct_ext_destroy(ct);
 727         atomic_dec(&net->ct.count);
 728         nf_ct_ext_free(ct);
 729         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 730 }
 731 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 732
 733 /* Allocate a new conntrack: we return -ENOMEM if classification
 734    failed due to stress.  Otherwise it really is unclassifiable. */
 735 static struct nf_conntrack_tuple_hash *
 736 init_conntrack(struct net *net, struct nf_conn *tmpl,
 737                const struct nf_conntrack_tuple *tuple,
 738                struct nf_conntrack_l3proto *l3proto,
 739                struct nf_conntrack_l4proto *l4proto,
 740                struct sk_buff *skb,
 741                unsigned int dataoff, u32 hash)
 742 {
 743         struct nf_conn *ct;
 744         struct nf_conn_help *help;
 745         struct nf_conntrack_tuple repl_tuple;
 746         struct nf_conntrack_ecache *ecache;
 747         struct nf_conntrack_expect *exp;
 748         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 749
 750         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 751                 pr_debug("Can't invert tuple.\n");
 752                 return NULL;
 753         }
 754
 755         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 756                                   hash);
 757         if (IS_ERR(ct))
 758                 return (struct nf_conntrack_tuple_hash *)ct;
 759
 760         if (!l4proto->new(ct, skb, dataoff)) {
 761                 nf_conntrack_free(ct);
 762                 pr_debug("init conntrack: can't track with proto module\n");
 763                 return NULL;
 764         }
 765
 766         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 767         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 768
 769         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 770         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 771                                  ecache ? ecache->expmask : 0,
 772                              GFP_ATOMIC);
 773
 774         spin_lock_bh(&nf_conntrack_lock);
 775         exp = nf_ct_find_expectation(net, zone, tuple);
 776         if (exp) {
 777                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 778                          ct, exp);
 779                 /* Welcome, Mr. Bond.  We've been expecting you... */
 780                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 781                 ct->master = exp->master;
 782                 if (exp->helper) {
 783                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 784                         if (help)
 785                                 rcu_assign_pointer(help->helper, exp->helper);
 786                 }
 787
 788 #ifdef CONFIG_NF_CONNTRACK_MARK
 789                 ct->mark = exp->master->mark;
 790 #endif
 791 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 792                 ct->secmark = exp->master->secmark;
 793 #endif
 794                 nf_conntrack_get(&ct->master->ct_general);
 795                 NF_CT_STAT_INC(net, expect_new);
 796         } else {
 797                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 798                 NF_CT_STAT_INC(net, new);
 799         }
 800
 801         /* Overload tuple linked list to put us in unconfirmed list. */
 802         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 803                        &net->ct.unconfirmed);
 804
 805         spin_unlock_bh(&nf_conntrack_lock);
 806
 807         if (exp) {
 808                 if (exp->expectfn)
 809                         exp->expectfn(ct, exp);
 810                 nf_ct_expect_put(exp);
 811         }
 812
 813         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 814 }
 815
 816 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 817 static inline struct nf_conn *
 818 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 819                   struct sk_buff *skb,
 820                   unsigned int dataoff,
 821                   u_int16_t l3num,
 822                   u_int8_t protonum,
 823                   struct nf_conntrack_l3proto *l3proto,
 824                   struct nf_conntrack_l4proto *l4proto,
 825                   int *set_reply,
 826                   enum ip_conntrack_info *ctinfo)
 827 {
 828         struct nf_conntrack_tuple tuple;
 829         struct nf_conntrack_tuple_hash *h;
 830         struct nf_conn *ct;
 831         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 832         u32 hash;
 833
 834         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 835                              dataoff, l3num, protonum, &tuple, l3proto,
 836                              l4proto)) {
 837                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 838                 return NULL;
 839         }
 840
 841         /* look for tuple match */
 842         hash = hash_conntrack_raw(&tuple, zone);
 843         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 844         if (!h) {
 845                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 846                                    skb, dataoff, hash);
 847                 if (!h)
 848                         return NULL;
 849                 if (IS_ERR(h))
 850                         return (void *)h;
 851         }
 852         ct = nf_ct_tuplehash_to_ctrack(h);
 853
 854         /* It exists; we have (non-exclusive) reference. */
 855         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 856                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 857                 /* Please set reply bit if this packet OK */
 858                 *set_reply = 1;
 859         } else {
 860                 /* Once we've had two way comms, always ESTABLISHED. */
 861                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 862                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 863                         *ctinfo = IP_CT_ESTABLISHED;
 864                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 865                         pr_debug("nf_conntrack_in: related packet for %p\n",
 866                                  ct);
 867                         *ctinfo = IP_CT_RELATED;
 868                 } else {
 869                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 870                         *ctinfo = IP_CT_NEW;
 871                 }
 872                 *set_reply = 0;
 873         }
 874         skb->nfct = &ct->ct_general;
 875         skb->nfctinfo = *ctinfo;
 876         return ct;
 877 }
 878
 879 unsigned int
 880 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 881                 struct sk_buff *skb)
 882 {
 883         struct nf_conn *ct, *tmpl = NULL;
 884         enum ip_conntrack_info ctinfo;
 885         struct nf_conntrack_l3proto *l3proto;
 886         struct nf_conntrack_l4proto *l4proto;
 887         unsigned int dataoff;
 888         u_int8_t protonum;
 889         int set_reply = 0;
 890         int ret;
 891
 892         if (skb->nfct) {
 893                 /* Previously seen (loopback or untracked)?  Ignore. */
 894                 tmpl = (struct nf_conn *)skb->nfct;
 895                 if (!nf_ct_is_template(tmpl)) {
 896                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 897                         return NF_ACCEPT;
 898                 }
 899                 skb->nfct = NULL;
 900         }
 901
 902         /* rcu_read_lock()ed by nf_hook_slow */
 903         l3proto = __nf_ct_l3proto_find(pf);
 904         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 905                                    &dataoff, &protonum);
 906         if (ret <= 0) {
 907                 pr_debug("not prepared to track yet or error occurred\n");
 908                 NF_CT_STAT_INC_ATOMIC(net, error);
 909                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 910                 ret = -ret;
 911                 goto out;
 912         }
 913
 914         l4proto = __nf_ct_l4proto_find(pf, protonum);
 915
 916         /* It may be an special packet, error, unclean...
 917          * inverse of the return code tells to the netfilter
 918          * core what to do with the packet. */
 919         if (l4proto->error != NULL) {
 920                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 921                                      pf, hooknum);
 922                 if (ret <= 0) {
 923                         NF_CT_STAT_INC_ATOMIC(net, error);
 924                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 925                         ret = -ret;
 926                         goto out;
 927                 }
 928                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 929                 if (skb->nfct)
 930                         goto out;
 931         }
 932
 933         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 934                                l3proto, l4proto, &set_reply, &ctinfo);
 935         if (!ct) {
 936                 /* Not valid part of a connection */
 937                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 938                 ret = NF_ACCEPT;
 939                 goto out;
 940         }
 941
 942         if (IS_ERR(ct)) {
 943                 /* Too stressed to deal. */
 944                 NF_CT_STAT_INC_ATOMIC(net, drop);
 945                 ret = NF_DROP;
 946                 goto out;
 947         }
 948
 949         NF_CT_ASSERT(skb->nfct);
 950
 951         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 952         if (ret <= 0) {
 953                 /* Invalid: inverse of the return code tells
 954                  * the netfilter core what to do */
 955                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 956                 nf_conntrack_put(skb->nfct);
 957                 skb->nfct = NULL;
 958                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 959                 if (ret == -NF_DROP)
 960                         NF_CT_STAT_INC_ATOMIC(net, drop);
 961                 ret = -ret;
 962                 goto out;
 963         }
 964
 965         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 966                 nf_conntrack_event_cache(IPCT_REPLY, ct);
 967 out:
 968         if (tmpl) {
 969                 /* Special case: we have to repeat this hook, assign the
 970                  * template again to this packet. We assume that this packet
 971                  * has no conntrack assigned. This is used by nf_ct_tcp. */
 972                 if (ret == NF_REPEAT)
 973                         skb->nfct = (struct nf_conntrack *)tmpl;
 974                 else
 975                         nf_ct_put(tmpl);
 976         }
 977
 978         return ret;
 979 }
 980 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 981
 982 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 983                           const struct nf_conntrack_tuple *orig)
 984 {
 985         bool ret;
 986
 987         rcu_read_lock();
 988         ret = nf_ct_invert_tuple(inverse, orig,
 989                                  __nf_ct_l3proto_find(orig->src.l3num),
 990                                  __nf_ct_l4proto_find(orig->src.l3num,
 991                                                       orig->dst.protonum));
 992         rcu_read_unlock();
 993         return ret;
 994 }
 995 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 996
 997 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 998    implicitly racy: see __nf_conntrack_confirm */
 999 void nf_conntrack_alter_reply(struct nf_conn *ct,
1000                               const struct nf_conntrack_tuple *newreply)
1001 {
1002         struct nf_conn_help *help = nfct_help(ct);
1003
1004         /* Should be unconfirmed, so not in hash table yet */
1005         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1006
1007         pr_debug("Altering reply tuple of %p to ", ct);
1008         nf_ct_dump_tuple(newreply);
1009
1010         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1011         if (ct->master || (help && !hlist_empty(&help->expectations)))
1012                 return;
1013
1014         rcu_read_lock();
1015         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1016         rcu_read_unlock();
1017 }
1018 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1019
1020 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1021 void __nf_ct_refresh_acct(struct nf_conn *ct,
1022                           enum ip_conntrack_info ctinfo,
1023                           const struct sk_buff *skb,
1024                           unsigned long extra_jiffies,
1025                           int do_acct)
1026 {
1027         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1028         NF_CT_ASSERT(skb);
1029
1030         /* Only update if this is not a fixed timeout */
1031         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1032                 goto acct;
1033
1034         /* If not in hash table, timer will not be active yet */
1035         if (!nf_ct_is_confirmed(ct)) {
1036                 ct->timeout.expires = extra_jiffies;
1037         } else {
1038                 unsigned long newtime = jiffies + extra_jiffies;
1039
1040                 /* Only update the timeout if the new timeout is at least
1041                    HZ jiffies from the old timeout. Need del_timer for race
1042                    avoidance (may already be dying). */
1043                 if (newtime - ct->timeout.expires >= HZ)
1044                         mod_timer_pending(&ct->timeout, newtime);
1045         }
1046
1047 acct:
1048         if (do_acct) {
1049                 struct nf_conn_counter *acct;
1050
1051                 acct = nf_conn_acct_find(ct);
1052                 if (acct) {
1053                         spin_lock_bh(&ct->lock);
1054                         acct[CTINFO2DIR(ctinfo)].packets++;
1055                         acct[CTINFO2DIR(ctinfo)].bytes += skb->len;
1056                         spin_unlock_bh(&ct->lock);
1057                 }
1058         }
1059 }
1060 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1061
1062 bool __nf_ct_kill_acct(struct nf_conn *ct,
1063                        enum ip_conntrack_info ctinfo,
1064                        const struct sk_buff *skb,
1065                        int do_acct)
1066 {
1067         if (do_acct) {
1068                 struct nf_conn_counter *acct;
1069
1070                 acct = nf_conn_acct_find(ct);
1071                 if (acct) {
1072                         spin_lock_bh(&ct->lock);
1073                         acct[CTINFO2DIR(ctinfo)].packets++;
1074                         acct[CTINFO2DIR(ctinfo)].bytes +=
1075                                 skb->len - skb_network_offset(skb);
1076                         spin_unlock_bh(&ct->lock);
1077                 }
1078         }
1079
1080         if (del_timer(&ct->timeout)) {
1081                 ct->timeout.function((unsigned long)ct);
1082                 return true;
1083         }
1084         return false;
1085 }
1086 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1087
1088 #ifdef CONFIG_NF_CONNTRACK_ZONES
1089 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1090         .len    = sizeof(struct nf_conntrack_zone),
1091         .align  = __alignof__(struct nf_conntrack_zone),
1092         .id     = NF_CT_EXT_ZONE,
1093 };
1094 #endif
1095
1096 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1097
1098 #include <linux/netfilter/nfnetlink.h>
1099 #include <linux/netfilter/nfnetlink_conntrack.h>
1100 #include <linux/mutex.h>
1101
1102 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1103  * in ip_conntrack_core, since we don't want the protocols to autoload
1104  * or depend on ctnetlink */
1105 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1106                                const struct nf_conntrack_tuple *tuple)
1107 {
1108         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
1109         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
1110         return 0;
1111
1112 nla_put_failure:
1113         return -1;
1114 }
1115 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1116
1117 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1118         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1119         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1120 };
1121 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1122
1123 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1124                                struct nf_conntrack_tuple *t)
1125 {
1126         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1127                 return -EINVAL;
1128
1129         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1130         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1131
1132         return 0;
1133 }
1134 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1135
1136 int nf_ct_port_nlattr_tuple_size(void)
1137 {
1138         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1139 }
1140 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1141 #endif
1142
1143 /* Used by ipt_REJECT and ip6t_REJECT. */
1144 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1145 {
1146         struct nf_conn *ct;
1147         enum ip_conntrack_info ctinfo;
1148
1149         /* This ICMP is in reverse direction to the packet which caused it */
1150         ct = nf_ct_get(skb, &ctinfo);
1151         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1152                 ctinfo = IP_CT_RELATED_REPLY;
1153         else
1154                 ctinfo = IP_CT_RELATED;
1155
1156         /* Attach to new skbuff, and increment count */
1157         nskb->nfct = &ct->ct_general;
1158         nskb->nfctinfo = ctinfo;
1159         nf_conntrack_get(nskb->nfct);
1160 }
1161
1162 /* Bring out ya dead! */
1163 static struct nf_conn *
1164 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1165                 void *data, unsigned int *bucket)
1166 {
1167         struct nf_conntrack_tuple_hash *h;
1168         struct nf_conn *ct;
1169         struct hlist_nulls_node *n;
1170
1171         spin_lock_bh(&nf_conntrack_lock);
1172         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1173                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1174                         ct = nf_ct_tuplehash_to_ctrack(h);
1175                         if (iter(ct, data))
1176                                 goto found;
1177                 }
1178         }
1179         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1180                 ct = nf_ct_tuplehash_to_ctrack(h);
1181                 if (iter(ct, data))
1182                         set_bit(IPS_DYING_BIT, &ct->status);
1183         }
1184         spin_unlock_bh(&nf_conntrack_lock);
1185         return NULL;
1186 found:
1187         atomic_inc(&ct->ct_general.use);
1188         spin_unlock_bh(&nf_conntrack_lock);
1189         return ct;
1190 }
1191
1192 void nf_ct_iterate_cleanup(struct net *net,
1193                            int (*iter)(struct nf_conn *i, void *data),
1194                            void *data)
1195 {
1196         struct nf_conn *ct;
1197         unsigned int bucket = 0;
1198
1199         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1200                 /* Time to push up daises... */
1201                 if (del_timer(&ct->timeout))
1202                         death_by_timeout((unsigned long)ct);
1203                 /* ... else the timer will get him soon. */
1204
1205                 nf_ct_put(ct);
1206         }
1207 }
1208 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1209
1210 struct __nf_ct_flush_report {
1211         u32 pid;
1212         int report;
1213 };
1214
1215 static int kill_report(struct nf_conn *i, void *data)
1216 {
1217         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1218         struct nf_conn_tstamp *tstamp;
1219
1220         tstamp = nf_conn_tstamp_find(i);
1221         if (tstamp && tstamp->stop == 0)
1222                 tstamp->stop = ktime_to_ns(ktime_get_real());
1223
1224         /* If we fail to deliver the event, death_by_timeout() will retry */
1225         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1226                                       fr->pid, fr->report) < 0)
1227                 return 1;
1228
1229         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1230         set_bit(IPS_DYING_BIT, &i->status);
1231         return 1;
1232 }
1233
1234 static int kill_all(struct nf_conn *i, void *data)
1235 {
1236         return 1;
1237 }
1238
1239 void nf_ct_free_hashtable(void *hash, unsigned int size)
1240 {
1241         if (is_vmalloc_addr(hash))
1242                 vfree(hash);
1243         else
1244                 free_pages((unsigned long)hash,
1245                            get_order(sizeof(struct hlist_head) * size));
1246 }
1247 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1248
1249 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1250 {
1251         struct __nf_ct_flush_report fr = {
1252                 .pid    = pid,
1253                 .report = report,
1254         };
1255         nf_ct_iterate_cleanup(net, kill_report, &fr);
1256 }
1257 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1258
1259 static void nf_ct_release_dying_list(struct net *net)
1260 {
1261         struct nf_conntrack_tuple_hash *h;
1262         struct nf_conn *ct;
1263         struct hlist_nulls_node *n;
1264
1265         spin_lock_bh(&nf_conntrack_lock);
1266         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1267                 ct = nf_ct_tuplehash_to_ctrack(h);
1268                 /* never fails to remove them, no listeners at this point */
1269                 nf_ct_kill(ct);
1270         }
1271         spin_unlock_bh(&nf_conntrack_lock);
1272 }
1273
1274 static int untrack_refs(void)
1275 {
1276         int cnt = 0, cpu;
1277
1278         for_each_possible_cpu(cpu) {
1279                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1280
1281                 cnt += atomic_read(&ct->ct_general.use) - 1;
1282         }
1283         return cnt;
1284 }
1285
1286 static void nf_conntrack_cleanup_init_net(void)
1287 {
1288         while (untrack_refs() > 0)
1289                 schedule();
1290
1291         nf_conntrack_helper_fini();
1292         nf_conntrack_proto_fini();
1293 #ifdef CONFIG_NF_CONNTRACK_ZONES
1294         nf_ct_extend_unregister(&nf_ct_zone_extend);
1295 #endif
1296 }
1297
1298 static void nf_conntrack_cleanup_net(struct net *net)
1299 {
1300  i_see_dead_people:
1301         nf_ct_iterate_cleanup(net, kill_all, NULL);
1302         nf_ct_release_dying_list(net);
1303         if (atomic_read(&net->ct.count) != 0) {
1304                 schedule();
1305                 goto i_see_dead_people;
1306         }
1307
1308         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1309         nf_conntrack_ecache_fini(net);
1310         nf_conntrack_tstamp_fini(net);
1311         nf_conntrack_acct_fini(net);
1312         nf_conntrack_expect_fini(net);
1313         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1314         kfree(net->ct.slabname);
1315         free_percpu(net->ct.stat);
1316 }
1317
1318 /* Mishearing the voices in his head, our hero wonders how he's
1319    supposed to kill the mall. */
1320 void nf_conntrack_cleanup(struct net *net)
1321 {
1322         if (net_eq(net, &init_net))
1323                 RCU_INIT_POINTER(ip_ct_attach, NULL);
1324
1325         /* This makes sure all current packets have passed through
1326            netfilter framework.  Roll on, two-stage module
1327            delete... */
1328         synchronize_net();
1329
1330         nf_conntrack_cleanup_net(net);
1331
1332         if (net_eq(net, &init_net)) {
1333                 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1334                 nf_conntrack_cleanup_init_net();
1335         }
1336 }
1337
1338 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1339 {
1340         struct hlist_nulls_head *hash;
1341         unsigned int nr_slots, i;
1342         size_t sz;
1343
1344         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1345         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1346         sz = nr_slots * sizeof(struct hlist_nulls_head);
1347         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1348                                         get_order(sz));
1349         if (!hash) {
1350                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1351                 hash = __vmalloc(sz, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1352                                  PAGE_KERNEL);
1353         }
1354
1355         if (hash && nulls)
1356                 for (i = 0; i < nr_slots; i++)
1357                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1358
1359         return hash;
1360 }
1361 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1362
1363 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1364 {
1365         int i, bucket;
1366         unsigned int hashsize, old_size;
1367         struct hlist_nulls_head *hash, *old_hash;
1368         struct nf_conntrack_tuple_hash *h;
1369         struct nf_conn *ct;
1370
1371         if (current->nsproxy->net_ns != &init_net)
1372                 return -EOPNOTSUPP;
1373
1374         /* On boot, we can set this without any fancy locking. */
1375         if (!nf_conntrack_htable_size)
1376                 return param_set_uint(val, kp);
1377
1378         hashsize = simple_strtoul(val, NULL, 0);
1379         if (!hashsize)
1380                 return -EINVAL;
1381
1382         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1383         if (!hash)
1384                 return -ENOMEM;
1385
1386         /* Lookups in the old hash might happen in parallel, which means we
1387          * might get false negatives during connection lookup. New connections
1388          * created because of a false negative won't make it into the hash
1389          * though since that required taking the lock.
1390          */
1391         spin_lock_bh(&nf_conntrack_lock);
1392         for (i = 0; i < init_net.ct.htable_size; i++) {
1393                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1394                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1395                                         struct nf_conntrack_tuple_hash, hnnode);
1396                         ct = nf_ct_tuplehash_to_ctrack(h);
1397                         hlist_nulls_del_rcu(&h->hnnode);
1398                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1399                                                   hashsize);
1400                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1401                 }
1402         }
1403         old_size = init_net.ct.htable_size;
1404         old_hash = init_net.ct.hash;
1405
1406         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1407         init_net.ct.hash = hash;
1408         spin_unlock_bh(&nf_conntrack_lock);
1409
1410         nf_ct_free_hashtable(old_hash, old_size);
1411         return 0;
1412 }
1413 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1414
1415 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1416                   &nf_conntrack_htable_size, 0600);
1417
1418 void nf_ct_untracked_status_or(unsigned long bits)
1419 {
1420         int cpu;
1421
1422         for_each_possible_cpu(cpu)
1423                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1424 }
1425 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1426
1427 static int nf_conntrack_init_init_net(void)
1428 {
1429         int max_factor = 8;
1430         int ret, cpu;
1431
1432         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1433          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1434         if (!nf_conntrack_htable_size) {
1435                 nf_conntrack_htable_size
1436                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1437                            / sizeof(struct hlist_head));
1438                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1439                         nf_conntrack_htable_size = 16384;
1440                 if (nf_conntrack_htable_size < 32)
1441                         nf_conntrack_htable_size = 32;
1442
1443                 /* Use a max. factor of four by default to get the same max as
1444                  * with the old struct list_heads. When a table size is given
1445                  * we use the old value of 8 to avoid reducing the max.
1446                  * entries. */
1447                 max_factor = 4;
1448         }
1449         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1450
1451         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1452                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1453                nf_conntrack_max);
1454
1455         ret = nf_conntrack_proto_init();
1456         if (ret < 0)
1457                 goto err_proto;
1458
1459         ret = nf_conntrack_helper_init();
1460         if (ret < 0)
1461                 goto err_helper;
1462
1463 #ifdef CONFIG_NF_CONNTRACK_ZONES
1464         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1465         if (ret < 0)
1466                 goto err_extend;
1467 #endif
1468         /* Set up fake conntrack: to never be deleted, not in any hashes */
1469         for_each_possible_cpu(cpu) {
1470                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1471                 write_pnet(&ct->ct_net, &init_net);
1472                 atomic_set(&ct->ct_general.use, 1);
1473         }
1474         /*  - and look it like as a confirmed connection */
1475         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1476         return 0;
1477
1478 #ifdef CONFIG_NF_CONNTRACK_ZONES
1479 err_extend:
1480         nf_conntrack_helper_fini();
1481 #endif
1482 err_helper:
1483         nf_conntrack_proto_fini();
1484 err_proto:
1485         return ret;
1486 }
1487
1488 /*
1489  * We need to use special "null" values, not used in hash table
1490  */
1491 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1492 #define DYING_NULLS_VAL         ((1<<30)+1)
1493
1494 static int nf_conntrack_init_net(struct net *net)
1495 {
1496         int ret;
1497
1498         atomic_set(&net->ct.count, 0);
1499         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1500         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1501         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1502         if (!net->ct.stat) {
1503                 ret = -ENOMEM;
1504                 goto err_stat;
1505         }
1506
1507         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1508         if (!net->ct.slabname) {
1509                 ret = -ENOMEM;
1510                 goto err_slabname;
1511         }
1512
1513         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1514                                                         sizeof(struct nf_conn), 0,
1515                                                         SLAB_DESTROY_BY_RCU, NULL);
1516         if (!net->ct.nf_conntrack_cachep) {
1517                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1518                 ret = -ENOMEM;
1519                 goto err_cache;
1520         }
1521
1522         net->ct.htable_size = nf_conntrack_htable_size;
1523         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1524         if (!net->ct.hash) {
1525                 ret = -ENOMEM;
1526                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1527                 goto err_hash;
1528         }
1529         ret = nf_conntrack_expect_init(net);
1530         if (ret < 0)
1531                 goto err_expect;
1532         ret = nf_conntrack_acct_init(net);
1533         if (ret < 0)
1534                 goto err_acct;
1535         ret = nf_conntrack_tstamp_init(net);
1536         if (ret < 0)
1537                 goto err_tstamp;
1538         ret = nf_conntrack_ecache_init(net);
1539         if (ret < 0)
1540                 goto err_ecache;
1541
1542         return 0;
1543
1544 err_ecache:
1545         nf_conntrack_tstamp_fini(net);
1546 err_tstamp:
1547         nf_conntrack_acct_fini(net);
1548 err_acct:
1549         nf_conntrack_expect_fini(net);
1550 err_expect:
1551         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1552 err_hash:
1553         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1554 err_cache:
1555         kfree(net->ct.slabname);
1556 err_slabname:
1557         free_percpu(net->ct.stat);
1558 err_stat:
1559         return ret;
1560 }
1561
1562 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1563                         enum ip_conntrack_dir dir,
1564                         u32 seq);
1565 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);
1566
1567 int nf_conntrack_init(struct net *net)
1568 {
1569         int ret;
1570
1571         if (net_eq(net, &init_net)) {
1572                 ret = nf_conntrack_init_init_net();
1573                 if (ret < 0)
1574                         goto out_init_net;
1575         }
1576         ret = nf_conntrack_init_net(net);
1577         if (ret < 0)
1578                 goto out_net;
1579
1580         if (net_eq(net, &init_net)) {
1581                 /* For use by REJECT target */
1582                 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1583                 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1584
1585                 /* Howto get NAT offsets */
1586                 RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
1587         }
1588         return 0;
1589
1590 out_net:
1591         if (net_eq(net, &init_net))
1592                 nf_conntrack_cleanup_init_net();
1593 out_init_net:
1594         return ret;
1595 }