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