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 static inline bool
 311 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
 312                         const struct nf_conntrack_tuple *tuple,
 313                         u16 zone)
 314 {
 315         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 316
 317         /* A conntrack can be recreated with the equal tuple,
 318          * so we need to check that the conntrack is confirmed
 319          */
 320         return nf_ct_tuple_equal(tuple, &h->tuple) &&
 321                 nf_ct_zone(ct) == zone &&
 322                 nf_ct_is_confirmed(ct);
 323 }
 324
 325 /*
 326  * Warning :
 327  * - Caller must take a reference on returned object
 328  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 329  * OR
 330  * - Caller must lock nf_conntrack_lock before calling this function
 331  */
 332 static struct nf_conntrack_tuple_hash *
 333 ____nf_conntrack_find(struct net *net, u16 zone,
 334                       const struct nf_conntrack_tuple *tuple, u32 hash)
 335 {
 336         struct nf_conntrack_tuple_hash *h;
 337         struct hlist_nulls_node *n;
 338         unsigned int bucket = hash_bucket(hash, net);
 339
 340         /* Disable BHs the entire time since we normally need to disable them
 341          * at least once for the stats anyway.
 342          */
 343         local_bh_disable();
 344 begin:
 345         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 346                 if (nf_ct_key_equal(h, tuple, zone)) {
 347                         NF_CT_STAT_INC(net, found);
 348                         local_bh_enable();
 349                         return h;
 350                 }
 351                 NF_CT_STAT_INC(net, searched);
 352         }
 353         /*
 354          * if the nulls value we got at the end of this lookup is
 355          * not the expected one, we must restart lookup.
 356          * We probably met an item that was moved to another chain.
 357          */
 358         if (get_nulls_value(n) != bucket) {
 359                 NF_CT_STAT_INC(net, search_restart);
 360                 goto begin;
 361         }
 362         local_bh_enable();
 363
 364         return NULL;
 365 }
 366
 367 struct nf_conntrack_tuple_hash *
 368 __nf_conntrack_find(struct net *net, u16 zone,
 369                     const struct nf_conntrack_tuple *tuple)
 370 {
 371         return ____nf_conntrack_find(net, zone, tuple,
 372                                      hash_conntrack_raw(tuple, zone));
 373 }
 374 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 375
 376 /* Find a connection corresponding to a tuple. */
 377 static struct nf_conntrack_tuple_hash *
 378 __nf_conntrack_find_get(struct net *net, u16 zone,
 379                         const struct nf_conntrack_tuple *tuple, u32 hash)
 380 {
 381         struct nf_conntrack_tuple_hash *h;
 382         struct nf_conn *ct;
 383
 384         rcu_read_lock();
 385 begin:
 386         h = ____nf_conntrack_find(net, zone, tuple, hash);
 387         if (h) {
 388                 ct = nf_ct_tuplehash_to_ctrack(h);
 389                 if (unlikely(nf_ct_is_dying(ct) ||
 390                              !atomic_inc_not_zero(&ct->ct_general.use)))
 391                         h = NULL;
 392                 else {
 393                         if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
 394                                 nf_ct_put(ct);
 395                                 goto begin;
 396                         }
 397                 }
 398         }
 399         rcu_read_unlock();
 400
 401         return h;
 402 }
 403
 404 struct nf_conntrack_tuple_hash *
 405 nf_conntrack_find_get(struct net *net, u16 zone,
 406                       const struct nf_conntrack_tuple *tuple)
 407 {
 408         return __nf_conntrack_find_get(net, zone, tuple,
 409                                        hash_conntrack_raw(tuple, zone));
 410 }
 411 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 412
 413 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 414                                        unsigned int hash,
 415                                        unsigned int repl_hash)
 416 {
 417         struct net *net = nf_ct_net(ct);
 418
 419         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 420                            &net->ct.hash[hash]);
 421         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 422                            &net->ct.hash[repl_hash]);
 423 }
 424
 425 void nf_conntrack_hash_insert(struct nf_conn *ct)
 426 {
 427         struct net *net = nf_ct_net(ct);
 428         unsigned int hash, repl_hash;
 429         u16 zone;
 430
 431         zone = nf_ct_zone(ct);
 432         hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 433         repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 434
 435         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 436 }
 437 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 438
 439 /* Confirm a connection given skb; places it in hash table */
 440 int
 441 __nf_conntrack_confirm(struct sk_buff *skb)
 442 {
 443         unsigned int hash, repl_hash;
 444         struct nf_conntrack_tuple_hash *h;
 445         struct nf_conn *ct;
 446         struct nf_conn_help *help;
 447         struct nf_conn_tstamp *tstamp;
 448         struct hlist_nulls_node *n;
 449         enum ip_conntrack_info ctinfo;
 450         struct net *net;
 451         u16 zone;
 452
 453         ct = nf_ct_get(skb, &ctinfo);
 454         net = nf_ct_net(ct);
 455
 456         /* ipt_REJECT uses nf_conntrack_attach to attach related
 457            ICMP/TCP RST packets in other direction.  Actual packet
 458            which created connection will be IP_CT_NEW or for an
 459            expected connection, IP_CT_RELATED. */
 460         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 461                 return NF_ACCEPT;
 462
 463         zone = nf_ct_zone(ct);
 464         /* reuse the hash saved before */
 465         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 466         hash = hash_bucket(hash, net);
 467         repl_hash = hash_conntrack(net, zone,
 468                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 469
 470         /* We're not in hash table, and we refuse to set up related
 471            connections for unconfirmed conns.  But packet copies and
 472            REJECT will give spurious warnings here. */
 473         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 474
 475         /* No external references means no one else could have
 476            confirmed us. */
 477         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 478         pr_debug("Confirming conntrack %p\n", ct);
 479
 480         spin_lock_bh(&nf_conntrack_lock);
 481
 482         /* We have to check the DYING flag inside the lock to prevent
 483            a race against nf_ct_get_next_corpse() possibly called from
 484            user context, else we insert an already 'dead' hash, blocking
 485            further use of that particular connection -JM */
 486
 487         if (unlikely(nf_ct_is_dying(ct))) {
 488                 spin_unlock_bh(&nf_conntrack_lock);
 489                 return NF_ACCEPT;
 490         }
 491
 492         /* See if there's one in the list already, including reverse:
 493            NAT could have grabbed it without realizing, since we're
 494            not in the hash.  If there is, we lost race. */
 495         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 496                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 497                                       &h->tuple) &&
 498                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 499                         goto out;
 500         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 501                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 502                                       &h->tuple) &&
 503                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 504                         goto out;
 505
 506         /* Remove from unconfirmed list */
 507         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 508
 509         /* Timer relative to confirmation time, not original
 510            setting time, otherwise we'd get timer wrap in
 511            weird delay cases. */
 512         ct->timeout.expires += jiffies;
 513         add_timer(&ct->timeout);
 514         atomic_inc(&ct->ct_general.use);
 515         ct->status |= IPS_CONFIRMED;
 516
 517         /* set conntrack timestamp, if enabled. */
 518         tstamp = nf_conn_tstamp_find(ct);
 519         if (tstamp) {
 520                 if (skb->tstamp.tv64 == 0)
 521                         __net_timestamp((struct sk_buff *)skb);
 522
 523                 tstamp->start = ktime_to_ns(skb->tstamp);
 524         }
 525         /* Since the lookup is lockless, hash insertion must be done after
 526          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 527          * guarantee that no other CPU can find the conntrack before the above
 528          * stores are visible.
 529          */
 530         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 531         NF_CT_STAT_INC(net, insert);
 532         spin_unlock_bh(&nf_conntrack_lock);
 533
 534         help = nfct_help(ct);
 535         if (help && help->helper)
 536                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 537
 538         nf_conntrack_event_cache(master_ct(ct) ?
 539                                  IPCT_RELATED : IPCT_NEW, ct);
 540         return NF_ACCEPT;
 541
 542 out:
 543         NF_CT_STAT_INC(net, insert_failed);
 544         spin_unlock_bh(&nf_conntrack_lock);
 545         return NF_DROP;
 546 }
 547 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 548
 549 /* Returns true if a connection correspondings to the tuple (required
 550    for NAT). */
 551 int
 552 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 553                          const struct nf_conn *ignored_conntrack)
 554 {
 555         struct net *net = nf_ct_net(ignored_conntrack);
 556         struct nf_conntrack_tuple_hash *h;
 557         struct hlist_nulls_node *n;
 558         struct nf_conn *ct;
 559         u16 zone = nf_ct_zone(ignored_conntrack);
 560         unsigned int hash = hash_conntrack(net, zone, tuple);
 561
 562         /* Disable BHs the entire time since we need to disable them at
 563          * least once for the stats anyway.
 564          */
 565         rcu_read_lock_bh();
 566  begin:
 567         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 568                 ct = nf_ct_tuplehash_to_ctrack(h);
 569                 if (ct != ignored_conntrack &&
 570                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 571                     nf_ct_zone(ct) == zone) {
 572                         NF_CT_STAT_INC(net, found);
 573                         rcu_read_unlock_bh();
 574                         return 1;
 575                 }
 576                 NF_CT_STAT_INC(net, searched);
 577         }
 578
 579         if (get_nulls_value(n) != hash) {
 580                 NF_CT_STAT_INC(net, search_restart);
 581                 goto begin;
 582         }
 583
 584         rcu_read_unlock_bh();
 585
 586         return 0;
 587 }
 588 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 589
 590 #define NF_CT_EVICTION_RANGE    8
 591
 592 /* There's a small race here where we may free a just-assured
 593    connection.  Too bad: we're in trouble anyway. */
 594 static noinline int early_drop(struct net *net, unsigned int hash)
 595 {
 596         /* Use oldest entry, which is roughly LRU */
 597         struct nf_conntrack_tuple_hash *h;
 598         struct nf_conn *ct = NULL, *tmp;
 599         struct hlist_nulls_node *n;
 600         unsigned int i, cnt = 0;
 601         int dropped = 0;
 602
 603         rcu_read_lock();
 604         for (i = 0; i < net->ct.htable_size; i++) {
 605                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 606                                          hnnode) {
 607                         tmp = nf_ct_tuplehash_to_ctrack(h);
 608                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 609                                 ct = tmp;
 610                         cnt++;
 611                 }
 612
 613                 if (ct != NULL) {
 614                         if (likely(!nf_ct_is_dying(ct) &&
 615                                    atomic_inc_not_zero(&ct->ct_general.use)))
 616                                 break;
 617                         else
 618                                 ct = NULL;
 619                 }
 620
 621                 if (cnt >= NF_CT_EVICTION_RANGE)
 622                         break;
 623
 624                 hash = (hash + 1) % net->ct.htable_size;
 625         }
 626         rcu_read_unlock();
 627
 628         if (!ct)
 629                 return dropped;
 630
 631         if (del_timer(&ct->timeout)) {
 632                 death_by_timeout((unsigned long)ct);
 633                 dropped = 1;
 634                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 635         }
 636         nf_ct_put(ct);
 637         return dropped;
 638 }
 639
 640 void init_nf_conntrack_hash_rnd(void)
 641 {
 642         unsigned int rand;
 643
 644         /*
 645          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 646          * Because there isn't enough entropy when system initializing,
 647          * and we initialize it as late as possible.
 648          */
 649         do {
 650                 get_random_bytes(&rand, sizeof(rand));
 651         } while (!rand);
 652         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 653 }
 654
 655 static struct nf_conn *
 656 __nf_conntrack_alloc(struct net *net, u16 zone,
 657                      const struct nf_conntrack_tuple *orig,
 658                      const struct nf_conntrack_tuple *repl,
 659                      gfp_t gfp, u32 hash)
 660 {
 661         struct nf_conn *ct;
 662
 663         if (unlikely(!nf_conntrack_hash_rnd)) {
 664                 init_nf_conntrack_hash_rnd();
 665                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 666                 hash = hash_conntrack_raw(orig, zone);
 667         }
 668
 669         /* We don't want any race condition at early drop stage */
 670         atomic_inc(&net->ct.count);
 671
 672         if (nf_conntrack_max &&
 673             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 674                 if (!early_drop(net, hash_bucket(hash, net))) {
 675                         atomic_dec(&net->ct.count);
 676                         if (net_ratelimit())
 677                                 printk(KERN_WARNING
 678                                        "nf_conntrack: table full, dropping"
 679                                        " packet.\n");
 680                         return ERR_PTR(-ENOMEM);
 681                 }
 682         }
 683
 684         /*
 685          * Do not use kmem_cache_zalloc(), as this cache uses
 686          * SLAB_DESTROY_BY_RCU.
 687          */
 688         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 689         if (ct == NULL) {
 690                 atomic_dec(&net->ct.count);
 691                 return ERR_PTR(-ENOMEM);
 692         }
 693         /*
 694          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 695          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 696          */
 697         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 698                offsetof(struct nf_conn, proto) -
 699                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 700         spin_lock_init(&ct->lock);
 701         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 702         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 703         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 704         /* save hash for reusing when confirming */
 705         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 706         /* Don't set timer yet: wait for confirmation */
 707         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 708         write_pnet(&ct->ct_net, net);
 709 #ifdef CONFIG_NF_CONNTRACK_ZONES
 710         if (zone) {
 711                 struct nf_conntrack_zone *nf_ct_zone;
 712
 713                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 714                 if (!nf_ct_zone)
 715                         goto out_free;
 716                 nf_ct_zone->id = zone;
 717         }
 718 #endif
 719         /*
 720          * changes to lookup keys must be done before setting refcnt to 1
 721          */
 722         smp_wmb();
 723         atomic_set(&ct->ct_general.use, 1);
 724         return ct;
 725
 726 #ifdef CONFIG_NF_CONNTRACK_ZONES
 727 out_free:
 728         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 729         return ERR_PTR(-ENOMEM);
 730 #endif
 731 }
 732
 733 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 734                                    const struct nf_conntrack_tuple *orig,
 735                                    const struct nf_conntrack_tuple *repl,
 736                                    gfp_t gfp)
 737 {
 738         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 739 }
 740 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 741
 742 void nf_conntrack_free(struct nf_conn *ct)
 743 {
 744         struct net *net = nf_ct_net(ct);
 745
 746         nf_ct_ext_destroy(ct);
 747         atomic_dec(&net->ct.count);
 748         nf_ct_ext_free(ct);
 749         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 750 }
 751 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 752
 753 /* Allocate a new conntrack: we return -ENOMEM if classification
 754    failed due to stress.  Otherwise it really is unclassifiable. */
 755 static struct nf_conntrack_tuple_hash *
 756 init_conntrack(struct net *net, struct nf_conn *tmpl,
 757                const struct nf_conntrack_tuple *tuple,
 758                struct nf_conntrack_l3proto *l3proto,
 759                struct nf_conntrack_l4proto *l4proto,
 760                struct sk_buff *skb,
 761                unsigned int dataoff, u32 hash)
 762 {
 763         struct nf_conn *ct;
 764         struct nf_conn_help *help;
 765         struct nf_conntrack_tuple repl_tuple;
 766         struct nf_conntrack_ecache *ecache;
 767         struct nf_conntrack_expect *exp;
 768         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 769
 770         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 771                 pr_debug("Can't invert tuple.\n");
 772                 return NULL;
 773         }
 774
 775         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 776                                   hash);
 777         if (IS_ERR(ct))
 778                 return (struct nf_conntrack_tuple_hash *)ct;
 779
 780         if (!l4proto->new(ct, skb, dataoff)) {
 781                 nf_conntrack_free(ct);
 782                 pr_debug("init conntrack: can't track with proto module\n");
 783                 return NULL;
 784         }
 785
 786         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 787         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 788
 789         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 790         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 791                                  ecache ? ecache->expmask : 0,
 792                              GFP_ATOMIC);
 793
 794         spin_lock_bh(&nf_conntrack_lock);
 795         exp = nf_ct_find_expectation(net, zone, tuple);
 796         if (exp) {
 797                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 798                          ct, exp);
 799                 /* Welcome, Mr. Bond.  We've been expecting you... */
 800                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 801                 ct->master = exp->master;
 802                 if (exp->helper) {
 803                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 804                         if (help)
 805                                 rcu_assign_pointer(help->helper, exp->helper);
 806                 }
 807
 808 #ifdef CONFIG_NF_CONNTRACK_MARK
 809                 ct->mark = exp->master->mark;
 810 #endif
 811 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 812                 ct->secmark = exp->master->secmark;
 813 #endif
 814                 nf_conntrack_get(&ct->master->ct_general);
 815                 NF_CT_STAT_INC(net, expect_new);
 816         } else {
 817                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 818                 NF_CT_STAT_INC(net, new);
 819         }
 820
 821         /* Overload tuple linked list to put us in unconfirmed list. */
 822         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 823                        &net->ct.unconfirmed);
 824
 825         spin_unlock_bh(&nf_conntrack_lock);
 826
 827         if (exp) {
 828                 if (exp->expectfn)
 829                         exp->expectfn(ct, exp);
 830                 nf_ct_expect_put(exp);
 831         }
 832
 833         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 834 }
 835
 836 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 837 static inline struct nf_conn *
 838 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 839                   struct sk_buff *skb,
 840                   unsigned int dataoff,
 841                   u_int16_t l3num,
 842                   u_int8_t protonum,
 843                   struct nf_conntrack_l3proto *l3proto,
 844                   struct nf_conntrack_l4proto *l4proto,
 845                   int *set_reply,
 846                   enum ip_conntrack_info *ctinfo)
 847 {
 848         struct nf_conntrack_tuple tuple;
 849         struct nf_conntrack_tuple_hash *h;
 850         struct nf_conn *ct;
 851         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 852         u32 hash;
 853
 854         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 855                              dataoff, l3num, protonum, &tuple, l3proto,
 856                              l4proto)) {
 857                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 858                 return NULL;
 859         }
 860
 861         /* look for tuple match */
 862         hash = hash_conntrack_raw(&tuple, zone);
 863         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 864         if (!h) {
 865                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 866                                    skb, dataoff, hash);
 867                 if (!h)
 868                         return NULL;
 869                 if (IS_ERR(h))
 870                         return (void *)h;
 871         }
 872         ct = nf_ct_tuplehash_to_ctrack(h);
 873
 874         /* It exists; we have (non-exclusive) reference. */
 875         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 876                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 877                 /* Please set reply bit if this packet OK */
 878                 *set_reply = 1;
 879         } else {
 880                 /* Once we've had two way comms, always ESTABLISHED. */
 881                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 882                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 883                         *ctinfo = IP_CT_ESTABLISHED;
 884                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 885                         pr_debug("nf_conntrack_in: related packet for %p\n",
 886                                  ct);
 887                         *ctinfo = IP_CT_RELATED;
 888                 } else {
 889                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 890                         *ctinfo = IP_CT_NEW;
 891                 }
 892                 *set_reply = 0;
 893         }
 894         skb->nfct = &ct->ct_general;
 895         skb->nfctinfo = *ctinfo;
 896         return ct;
 897 }
 898
 899 unsigned int
 900 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 901                 struct sk_buff *skb)
 902 {
 903         struct nf_conn *ct, *tmpl = NULL;
 904         enum ip_conntrack_info ctinfo;
 905         struct nf_conntrack_l3proto *l3proto;
 906         struct nf_conntrack_l4proto *l4proto;
 907         unsigned int dataoff;
 908         u_int8_t protonum;
 909         int set_reply = 0;
 910         int ret;
 911
 912         if (skb->nfct) {
 913                 /* Previously seen (loopback or untracked)?  Ignore. */
 914                 tmpl = (struct nf_conn *)skb->nfct;
 915                 if (!nf_ct_is_template(tmpl)) {
 916                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 917                         return NF_ACCEPT;
 918                 }
 919                 skb->nfct = NULL;
 920         }
 921
 922         /* rcu_read_lock()ed by nf_hook_slow */
 923         l3proto = __nf_ct_l3proto_find(pf);
 924         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 925                                    &dataoff, &protonum);
 926         if (ret <= 0) {
 927                 pr_debug("not prepared to track yet or error occurred\n");
 928                 NF_CT_STAT_INC_ATOMIC(net, error);
 929                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 930                 ret = -ret;
 931                 goto out;
 932         }
 933
 934         l4proto = __nf_ct_l4proto_find(pf, protonum);
 935
 936         /* It may be an special packet, error, unclean...
 937          * inverse of the return code tells to the netfilter
 938          * core what to do with the packet. */
 939         if (l4proto->error != NULL) {
 940                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 941                                      pf, hooknum);
 942                 if (ret <= 0) {
 943                         NF_CT_STAT_INC_ATOMIC(net, error);
 944                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 945                         ret = -ret;
 946                         goto out;
 947                 }
 948                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 949                 if (skb->nfct)
 950                         goto out;
 951         }
 952
 953         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 954                                l3proto, l4proto, &set_reply, &ctinfo);
 955         if (!ct) {
 956                 /* Not valid part of a connection */
 957                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 958                 ret = NF_ACCEPT;
 959                 goto out;
 960         }
 961
 962         if (IS_ERR(ct)) {
 963                 /* Too stressed to deal. */
 964                 NF_CT_STAT_INC_ATOMIC(net, drop);
 965                 ret = NF_DROP;
 966                 goto out;
 967         }
 968
 969         NF_CT_ASSERT(skb->nfct);
 970
 971         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 972         if (ret <= 0) {
 973                 /* Invalid: inverse of the return code tells
 974                  * the netfilter core what to do */
 975                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 976                 nf_conntrack_put(skb->nfct);
 977                 skb->nfct = NULL;
 978                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 979                 if (ret == -NF_DROP)
 980                         NF_CT_STAT_INC_ATOMIC(net, drop);
 981                 ret = -ret;
 982                 goto out;
 983         }
 984
 985         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 986                 nf_conntrack_event_cache(IPCT_REPLY, ct);
 987 out:
 988         if (tmpl) {
 989                 /* Special case: we have to repeat this hook, assign the
 990                  * template again to this packet. We assume that this packet
 991                  * has no conntrack assigned. This is used by nf_ct_tcp. */
 992                 if (ret == NF_REPEAT)
 993                         skb->nfct = (struct nf_conntrack *)tmpl;
 994                 else
 995                         nf_ct_put(tmpl);
 996         }
 997
 998         return ret;
 999 }
1000 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1001
1002 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1003                           const struct nf_conntrack_tuple *orig)
1004 {
1005         bool ret;
1006
1007         rcu_read_lock();
1008         ret = nf_ct_invert_tuple(inverse, orig,
1009                                  __nf_ct_l3proto_find(orig->src.l3num),
1010                                  __nf_ct_l4proto_find(orig->src.l3num,
1011                                                       orig->dst.protonum));
1012         rcu_read_unlock();
1013         return ret;
1014 }
1015 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1016
1017 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1018    implicitly racy: see __nf_conntrack_confirm */
1019 void nf_conntrack_alter_reply(struct nf_conn *ct,
1020                               const struct nf_conntrack_tuple *newreply)
1021 {
1022         struct nf_conn_help *help = nfct_help(ct);
1023
1024         /* Should be unconfirmed, so not in hash table yet */
1025         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1026
1027         pr_debug("Altering reply tuple of %p to ", ct);
1028         nf_ct_dump_tuple(newreply);
1029
1030         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1031         if (ct->master || (help && !hlist_empty(&help->expectations)))
1032                 return;
1033
1034         rcu_read_lock();
1035         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1036         rcu_read_unlock();
1037 }
1038 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1039
1040 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1041 void __nf_ct_refresh_acct(struct nf_conn *ct,
1042                           enum ip_conntrack_info ctinfo,
1043                           const struct sk_buff *skb,
1044                           unsigned long extra_jiffies,
1045                           int do_acct)
1046 {
1047         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1048         NF_CT_ASSERT(skb);
1049
1050         /* Only update if this is not a fixed timeout */
1051         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1052                 goto acct;
1053
1054         /* If not in hash table, timer will not be active yet */
1055         if (!nf_ct_is_confirmed(ct)) {
1056                 ct->timeout.expires = extra_jiffies;
1057         } else {
1058                 unsigned long newtime = jiffies + extra_jiffies;
1059
1060                 /* Only update the timeout if the new timeout is at least
1061                    HZ jiffies from the old timeout. Need del_timer for race
1062                    avoidance (may already be dying). */
1063                 if (newtime - ct->timeout.expires >= HZ)
1064                         mod_timer_pending(&ct->timeout, newtime);
1065         }
1066
1067 acct:
1068         if (do_acct) {
1069                 struct nf_conn_counter *acct;
1070
1071                 acct = nf_conn_acct_find(ct);
1072                 if (acct) {
1073                         spin_lock_bh(&ct->lock);
1074                         acct[CTINFO2DIR(ctinfo)].packets++;
1075                         acct[CTINFO2DIR(ctinfo)].bytes += skb->len;
1076                         spin_unlock_bh(&ct->lock);
1077                 }
1078         }
1079 }
1080 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1081
1082 bool __nf_ct_kill_acct(struct nf_conn *ct,
1083                        enum ip_conntrack_info ctinfo,
1084                        const struct sk_buff *skb,
1085                        int do_acct)
1086 {
1087         if (do_acct) {
1088                 struct nf_conn_counter *acct;
1089
1090                 acct = nf_conn_acct_find(ct);
1091                 if (acct) {
1092                         spin_lock_bh(&ct->lock);
1093                         acct[CTINFO2DIR(ctinfo)].packets++;
1094                         acct[CTINFO2DIR(ctinfo)].bytes +=
1095                                 skb->len - skb_network_offset(skb);
1096                         spin_unlock_bh(&ct->lock);
1097                 }
1098         }
1099
1100         if (del_timer(&ct->timeout)) {
1101                 ct->timeout.function((unsigned long)ct);
1102                 return true;
1103         }
1104         return false;
1105 }
1106 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1107
1108 #ifdef CONFIG_NF_CONNTRACK_ZONES
1109 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1110         .len    = sizeof(struct nf_conntrack_zone),
1111         .align  = __alignof__(struct nf_conntrack_zone),
1112         .id     = NF_CT_EXT_ZONE,
1113 };
1114 #endif
1115
1116 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1117
1118 #include <linux/netfilter/nfnetlink.h>
1119 #include <linux/netfilter/nfnetlink_conntrack.h>
1120 #include <linux/mutex.h>
1121
1122 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1123  * in ip_conntrack_core, since we don't want the protocols to autoload
1124  * or depend on ctnetlink */
1125 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1126                                const struct nf_conntrack_tuple *tuple)
1127 {
1128         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
1129         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
1130         return 0;
1131
1132 nla_put_failure:
1133         return -1;
1134 }
1135 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1136
1137 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1138         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1139         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1140 };
1141 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1142
1143 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1144                                struct nf_conntrack_tuple *t)
1145 {
1146         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1147                 return -EINVAL;
1148
1149         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1150         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1151
1152         return 0;
1153 }
1154 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1155
1156 int nf_ct_port_nlattr_tuple_size(void)
1157 {
1158         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1159 }
1160 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1161 #endif
1162
1163 /* Used by ipt_REJECT and ip6t_REJECT. */
1164 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1165 {
1166         struct nf_conn *ct;
1167         enum ip_conntrack_info ctinfo;
1168
1169         /* This ICMP is in reverse direction to the packet which caused it */
1170         ct = nf_ct_get(skb, &ctinfo);
1171         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1172                 ctinfo = IP_CT_RELATED_REPLY;
1173         else
1174                 ctinfo = IP_CT_RELATED;
1175
1176         /* Attach to new skbuff, and increment count */
1177         nskb->nfct = &ct->ct_general;
1178         nskb->nfctinfo = ctinfo;
1179         nf_conntrack_get(nskb->nfct);
1180 }
1181
1182 /* Bring out ya dead! */
1183 static struct nf_conn *
1184 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1185                 void *data, unsigned int *bucket)
1186 {
1187         struct nf_conntrack_tuple_hash *h;
1188         struct nf_conn *ct;
1189         struct hlist_nulls_node *n;
1190
1191         spin_lock_bh(&nf_conntrack_lock);
1192         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1193                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1194                         ct = nf_ct_tuplehash_to_ctrack(h);
1195                         if (iter(ct, data))
1196                                 goto found;
1197                 }
1198         }
1199         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1200                 ct = nf_ct_tuplehash_to_ctrack(h);
1201                 if (iter(ct, data))
1202                         set_bit(IPS_DYING_BIT, &ct->status);
1203         }
1204         spin_unlock_bh(&nf_conntrack_lock);
1205         return NULL;
1206 found:
1207         atomic_inc(&ct->ct_general.use);
1208         spin_unlock_bh(&nf_conntrack_lock);
1209         return ct;
1210 }
1211
1212 void nf_ct_iterate_cleanup(struct net *net,
1213                            int (*iter)(struct nf_conn *i, void *data),
1214                            void *data)
1215 {
1216         struct nf_conn *ct;
1217         unsigned int bucket = 0;
1218
1219         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1220                 /* Time to push up daises... */
1221                 if (del_timer(&ct->timeout))
1222                         death_by_timeout((unsigned long)ct);
1223                 /* ... else the timer will get him soon. */
1224
1225                 nf_ct_put(ct);
1226         }
1227 }
1228 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1229
1230 struct __nf_ct_flush_report {
1231         u32 pid;
1232         int report;
1233 };
1234
1235 static int kill_report(struct nf_conn *i, void *data)
1236 {
1237         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1238         struct nf_conn_tstamp *tstamp;
1239
1240         tstamp = nf_conn_tstamp_find(i);
1241         if (tstamp && tstamp->stop == 0)
1242                 tstamp->stop = ktime_to_ns(ktime_get_real());
1243
1244         /* If we fail to deliver the event, death_by_timeout() will retry */
1245         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1246                                       fr->pid, fr->report) < 0)
1247                 return 1;
1248
1249         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1250         set_bit(IPS_DYING_BIT, &i->status);
1251         return 1;
1252 }
1253
1254 static int kill_all(struct nf_conn *i, void *data)
1255 {
1256         return 1;
1257 }
1258
1259 void nf_ct_free_hashtable(void *hash, unsigned int size)
1260 {
1261         if (is_vmalloc_addr(hash))
1262                 vfree(hash);
1263         else
1264                 free_pages((unsigned long)hash,
1265                            get_order(sizeof(struct hlist_head) * size));
1266 }
1267 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1268
1269 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1270 {
1271         struct __nf_ct_flush_report fr = {
1272                 .pid    = pid,
1273                 .report = report,
1274         };
1275         nf_ct_iterate_cleanup(net, kill_report, &fr);
1276 }
1277 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1278
1279 static void nf_ct_release_dying_list(struct net *net)
1280 {
1281         struct nf_conntrack_tuple_hash *h;
1282         struct nf_conn *ct;
1283         struct hlist_nulls_node *n;
1284
1285         spin_lock_bh(&nf_conntrack_lock);
1286         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1287                 ct = nf_ct_tuplehash_to_ctrack(h);
1288                 /* never fails to remove them, no listeners at this point */
1289                 nf_ct_kill(ct);
1290         }
1291         spin_unlock_bh(&nf_conntrack_lock);
1292 }
1293
1294 static int untrack_refs(void)
1295 {
1296         int cnt = 0, cpu;
1297
1298         for_each_possible_cpu(cpu) {
1299                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1300
1301                 cnt += atomic_read(&ct->ct_general.use) - 1;
1302         }
1303         return cnt;
1304 }
1305
1306 static void nf_conntrack_cleanup_init_net(void)
1307 {
1308         while (untrack_refs() > 0)
1309                 schedule();
1310
1311         nf_conntrack_helper_fini();
1312         nf_conntrack_proto_fini();
1313 #ifdef CONFIG_NF_CONNTRACK_ZONES
1314         nf_ct_extend_unregister(&nf_ct_zone_extend);
1315 #endif
1316 }
1317
1318 static void nf_conntrack_cleanup_net(struct net *net)
1319 {
1320  i_see_dead_people:
1321         nf_ct_iterate_cleanup(net, kill_all, NULL);
1322         nf_ct_release_dying_list(net);
1323         if (atomic_read(&net->ct.count) != 0) {
1324                 schedule();
1325                 goto i_see_dead_people;
1326         }
1327
1328         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1329         nf_conntrack_ecache_fini(net);
1330         nf_conntrack_tstamp_fini(net);
1331         nf_conntrack_acct_fini(net);
1332         nf_conntrack_expect_fini(net);
1333         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1334         kfree(net->ct.slabname);
1335         free_percpu(net->ct.stat);
1336 }
1337
1338 /* Mishearing the voices in his head, our hero wonders how he's
1339    supposed to kill the mall. */
1340 void nf_conntrack_cleanup(struct net *net)
1341 {
1342         if (net_eq(net, &init_net))
1343                 RCU_INIT_POINTER(ip_ct_attach, NULL);
1344
1345         /* This makes sure all current packets have passed through
1346            netfilter framework.  Roll on, two-stage module
1347            delete... */
1348         synchronize_net();
1349
1350         nf_conntrack_cleanup_net(net);
1351
1352         if (net_eq(net, &init_net)) {
1353                 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1354                 nf_conntrack_cleanup_init_net();
1355         }
1356 }
1357
1358 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1359 {
1360         struct hlist_nulls_head *hash;
1361         unsigned int nr_slots, i;
1362         size_t sz;
1363
1364         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1365         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1366         sz = nr_slots * sizeof(struct hlist_nulls_head);
1367         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1368                                         get_order(sz));
1369         if (!hash) {
1370                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1371                 hash = __vmalloc(sz, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1372                                  PAGE_KERNEL);
1373         }
1374
1375         if (hash && nulls)
1376                 for (i = 0; i < nr_slots; i++)
1377                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1378
1379         return hash;
1380 }
1381 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1382
1383 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1384 {
1385         int i, bucket;
1386         unsigned int hashsize, old_size;
1387         struct hlist_nulls_head *hash, *old_hash;
1388         struct nf_conntrack_tuple_hash *h;
1389         struct nf_conn *ct;
1390
1391         if (current->nsproxy->net_ns != &init_net)
1392                 return -EOPNOTSUPP;
1393
1394         /* On boot, we can set this without any fancy locking. */
1395         if (!nf_conntrack_htable_size)
1396                 return param_set_uint(val, kp);
1397
1398         hashsize = simple_strtoul(val, NULL, 0);
1399         if (!hashsize)
1400                 return -EINVAL;
1401
1402         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1403         if (!hash)
1404                 return -ENOMEM;
1405
1406         /* Lookups in the old hash might happen in parallel, which means we
1407          * might get false negatives during connection lookup. New connections
1408          * created because of a false negative won't make it into the hash
1409          * though since that required taking the lock.
1410          */
1411         spin_lock_bh(&nf_conntrack_lock);
1412         for (i = 0; i < init_net.ct.htable_size; i++) {
1413                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1414                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1415                                         struct nf_conntrack_tuple_hash, hnnode);
1416                         ct = nf_ct_tuplehash_to_ctrack(h);
1417                         hlist_nulls_del_rcu(&h->hnnode);
1418                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1419                                                   hashsize);
1420                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1421                 }
1422         }
1423         old_size = init_net.ct.htable_size;
1424         old_hash = init_net.ct.hash;
1425
1426         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1427         init_net.ct.hash = hash;
1428         spin_unlock_bh(&nf_conntrack_lock);
1429
1430         nf_ct_free_hashtable(old_hash, old_size);
1431         return 0;
1432 }
1433 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1434
1435 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1436                   &nf_conntrack_htable_size, 0600);
1437
1438 void nf_ct_untracked_status_or(unsigned long bits)
1439 {
1440         int cpu;
1441
1442         for_each_possible_cpu(cpu)
1443                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1444 }
1445 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1446
1447 static int nf_conntrack_init_init_net(void)
1448 {
1449         int max_factor = 8;
1450         int ret, cpu;
1451
1452         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1453          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1454         if (!nf_conntrack_htable_size) {
1455                 nf_conntrack_htable_size
1456                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1457                            / sizeof(struct hlist_head));
1458                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1459                         nf_conntrack_htable_size = 16384;
1460                 if (nf_conntrack_htable_size < 32)
1461                         nf_conntrack_htable_size = 32;
1462
1463                 /* Use a max. factor of four by default to get the same max as
1464                  * with the old struct list_heads. When a table size is given
1465                  * we use the old value of 8 to avoid reducing the max.
1466                  * entries. */
1467                 max_factor = 4;
1468         }
1469         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1470
1471         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1472                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1473                nf_conntrack_max);
1474
1475         ret = nf_conntrack_proto_init();
1476         if (ret < 0)
1477                 goto err_proto;
1478
1479         ret = nf_conntrack_helper_init();
1480         if (ret < 0)
1481                 goto err_helper;
1482
1483 #ifdef CONFIG_NF_CONNTRACK_ZONES
1484         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1485         if (ret < 0)
1486                 goto err_extend;
1487 #endif
1488         /* Set up fake conntrack: to never be deleted, not in any hashes */
1489         for_each_possible_cpu(cpu) {
1490                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1491                 write_pnet(&ct->ct_net, &init_net);
1492                 atomic_set(&ct->ct_general.use, 1);
1493         }
1494         /*  - and look it like as a confirmed connection */
1495         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1496         return 0;
1497
1498 #ifdef CONFIG_NF_CONNTRACK_ZONES
1499 err_extend:
1500         nf_conntrack_helper_fini();
1501 #endif
1502 err_helper:
1503         nf_conntrack_proto_fini();
1504 err_proto:
1505         return ret;
1506 }
1507
1508 /*
1509  * We need to use special "null" values, not used in hash table
1510  */
1511 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1512 #define DYING_NULLS_VAL         ((1<<30)+1)
1513
1514 static int nf_conntrack_init_net(struct net *net)
1515 {
1516         static atomic64_t unique_id;
1517         int ret;
1518
1519         atomic_set(&net->ct.count, 0);
1520         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1521         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1522         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1523         if (!net->ct.stat) {
1524                 ret = -ENOMEM;
1525                 goto err_stat;
1526         }
1527
1528         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%llu",
1529                                 (u64)atomic64_inc_return(&unique_id));
1530         if (!net->ct.slabname) {
1531                 ret = -ENOMEM;
1532                 goto err_slabname;
1533         }
1534
1535         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1536                                                         sizeof(struct nf_conn), 0,
1537                                                         SLAB_DESTROY_BY_RCU, NULL);
1538         if (!net->ct.nf_conntrack_cachep) {
1539                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1540                 ret = -ENOMEM;
1541                 goto err_cache;
1542         }
1543
1544         net->ct.htable_size = nf_conntrack_htable_size;
1545         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1546         if (!net->ct.hash) {
1547                 ret = -ENOMEM;
1548                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1549                 goto err_hash;
1550         }
1551         ret = nf_conntrack_expect_init(net);
1552         if (ret < 0)
1553                 goto err_expect;
1554         ret = nf_conntrack_acct_init(net);
1555         if (ret < 0)
1556                 goto err_acct;
1557         ret = nf_conntrack_tstamp_init(net);
1558         if (ret < 0)
1559                 goto err_tstamp;
1560         ret = nf_conntrack_ecache_init(net);
1561         if (ret < 0)
1562                 goto err_ecache;
1563
1564         return 0;
1565
1566 err_ecache:
1567         nf_conntrack_tstamp_fini(net);
1568 err_tstamp:
1569         nf_conntrack_acct_fini(net);
1570 err_acct:
1571         nf_conntrack_expect_fini(net);
1572 err_expect:
1573         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1574 err_hash:
1575         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1576 err_cache:
1577         kfree(net->ct.slabname);
1578 err_slabname:
1579         free_percpu(net->ct.stat);
1580 err_stat:
1581         return ret;
1582 }
1583
1584 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1585                         enum ip_conntrack_dir dir,
1586                         u32 seq);
1587 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);
1588
1589 int nf_conntrack_init(struct net *net)
1590 {
1591         int ret;
1592
1593         if (net_eq(net, &init_net)) {
1594                 ret = nf_conntrack_init_init_net();
1595                 if (ret < 0)
1596                         goto out_init_net;
1597         }
1598         ret = nf_conntrack_init_net(net);
1599         if (ret < 0)
1600                 goto out_net;
1601
1602         if (net_eq(net, &init_net)) {
1603                 /* For use by REJECT target */
1604                 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1605                 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1606
1607                 /* Howto get NAT offsets */
1608                 RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
1609         }
1610         return 0;
1611
1612 out_net:
1613         if (net_eq(net, &init_net))
1614                 nf_conntrack_cleanup_init_net();
1615 out_init_net:
1616         return ret;
1617 }