net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License version 2 as
  11  * published by the Free Software Foundation.
  12  */
  13
  14 #include <linux/types.h>
  15 #include <linux/netfilter.h>
  16 #include <linux/module.h>
  17 #include <linux/sched.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/proc_fs.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/stddef.h>
  22 #include <linux/slab.h>
  23 #include <linux/random.h>
  24 #include <linux/jhash.h>
  25 #include <linux/err.h>
  26 #include <linux/percpu.h>
  27 #include <linux/moduleparam.h>
  28 #include <linux/notifier.h>
  29 #include <linux/kernel.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/socket.h>
  32 #include <linux/mm.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/rculist_nulls.h>
  35
  36 #include <net/netfilter/nf_conntrack.h>
  37 #include <net/netfilter/nf_conntrack_l3proto.h>
  38 #include <net/netfilter/nf_conntrack_l4proto.h>
  39 #include <net/netfilter/nf_conntrack_expect.h>
  40 #include <net/netfilter/nf_conntrack_helper.h>
  41 #include <net/netfilter/nf_conntrack_core.h>
  42 #include <net/netfilter/nf_conntrack_extend.h>
  43 #include <net/netfilter/nf_conntrack_acct.h>
  44 #include <net/netfilter/nf_conntrack_ecache.h>
  45 #include <net/netfilter/nf_conntrack_zones.h>
  46 #include <net/netfilter/nf_nat.h>
  47 #include <net/netfilter/nf_nat_core.h>
  48
  49 #define NF_CONNTRACK_VERSION    "0.5.0"
  50
  51 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  52                                       enum nf_nat_manip_type manip,
  53                                       const struct nlattr *attr) __read_mostly;
  54 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  55
  56 DEFINE_SPINLOCK(nf_conntrack_lock);
  57 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  58
  59 unsigned int nf_conntrack_htable_size __read_mostly;
  60 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  61
  62 unsigned int nf_conntrack_max __read_mostly;
  63 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  64
  65 struct nf_conn nf_conntrack_untracked __read_mostly;
  66 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
  67
  68 static int nf_conntrack_hash_rnd_initted;
  69 static unsigned int nf_conntrack_hash_rnd;
  70
  71 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  72                                   u16 zone, unsigned int size, unsigned int rnd)
  73 {
  74         unsigned int n;
  75         u_int32_t h;
  76
  77         /* The direction must be ignored, so we hash everything up to the
  78          * destination ports (which is a multiple of 4) and treat the last
  79          * three bytes manually.
  80          */
  81         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  82         h = jhash2((u32 *)tuple, n,
  83                    zone ^ rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
  84                                  tuple->dst.protonum));
  85
  86         return ((u64)h * size) >> 32;
  87 }
  88
  89 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
  90                                        const struct nf_conntrack_tuple *tuple)
  91 {
  92         return __hash_conntrack(tuple, zone, net->ct.htable_size,
  93                                 nf_conntrack_hash_rnd);
  94 }
  95
  96 bool
  97 nf_ct_get_tuple(const struct sk_buff *skb,
  98                 unsigned int nhoff,
  99                 unsigned int dataoff,
 100                 u_int16_t l3num,
 101                 u_int8_t protonum,
 102                 struct nf_conntrack_tuple *tuple,
 103                 const struct nf_conntrack_l3proto *l3proto,
 104                 const struct nf_conntrack_l4proto *l4proto)
 105 {
 106         memset(tuple, 0, sizeof(*tuple));
 107
 108         tuple->src.l3num = l3num;
 109         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 110                 return false;
 111
 112         tuple->dst.protonum = protonum;
 113         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 114
 115         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 116 }
 117 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 118
 119 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 120                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 121 {
 122         struct nf_conntrack_l3proto *l3proto;
 123         struct nf_conntrack_l4proto *l4proto;
 124         unsigned int protoff;
 125         u_int8_t protonum;
 126         int ret;
 127
 128         rcu_read_lock();
 129
 130         l3proto = __nf_ct_l3proto_find(l3num);
 131         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 132         if (ret != NF_ACCEPT) {
 133                 rcu_read_unlock();
 134                 return false;
 135         }
 136
 137         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 138
 139         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 140                               l3proto, l4proto);
 141
 142         rcu_read_unlock();
 143         return ret;
 144 }
 145 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 146
 147 bool
 148 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 149                    const struct nf_conntrack_tuple *orig,
 150                    const struct nf_conntrack_l3proto *l3proto,
 151                    const struct nf_conntrack_l4proto *l4proto)
 152 {
 153         memset(inverse, 0, sizeof(*inverse));
 154
 155         inverse->src.l3num = orig->src.l3num;
 156         if (l3proto->invert_tuple(inverse, orig) == 0)
 157                 return false;
 158
 159         inverse->dst.dir = !orig->dst.dir;
 160
 161         inverse->dst.protonum = orig->dst.protonum;
 162         return l4proto->invert_tuple(inverse, orig);
 163 }
 164 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 165
 166 static void
 167 clean_from_lists(struct nf_conn *ct)
 168 {
 169         pr_debug("clean_from_lists(%p)\n", ct);
 170         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 171         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 172
 173         /* Destroy all pending expectations */
 174         nf_ct_remove_expectations(ct);
 175 }
 176
 177 static void
 178 destroy_conntrack(struct nf_conntrack *nfct)
 179 {
 180         struct nf_conn *ct = (struct nf_conn *)nfct;
 181         struct net *net = nf_ct_net(ct);
 182         struct nf_conntrack_l4proto *l4proto;
 183
 184         pr_debug("destroy_conntrack(%p)\n", ct);
 185         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 186         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 187
 188         /* To make sure we don't get any weird locking issues here:
 189          * destroy_conntrack() MUST NOT be called with a write lock
 190          * to nf_conntrack_lock!!! -HW */
 191         rcu_read_lock();
 192         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 193         if (l4proto && l4proto->destroy)
 194                 l4proto->destroy(ct);
 195
 196         rcu_read_unlock();
 197
 198         spin_lock_bh(&nf_conntrack_lock);
 199         /* Expectations will have been removed in clean_from_lists,
 200          * except TFTP can create an expectation on the first packet,
 201          * before connection is in the list, so we need to clean here,
 202          * too. */
 203         nf_ct_remove_expectations(ct);
 204
 205         /* We overload first tuple to link into unconfirmed list. */
 206         if (!nf_ct_is_confirmed(ct)) {
 207                 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 208                 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 209         }
 210
 211         NF_CT_STAT_INC(net, delete);
 212         spin_unlock_bh(&nf_conntrack_lock);
 213
 214         if (ct->master)
 215                 nf_ct_put(ct->master);
 216
 217         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 218         nf_conntrack_free(ct);
 219 }
 220
 221 void nf_ct_delete_from_lists(struct nf_conn *ct)
 222 {
 223         struct net *net = nf_ct_net(ct);
 224
 225         nf_ct_helper_destroy(ct);
 226         spin_lock_bh(&nf_conntrack_lock);
 227         /* Inside lock so preempt is disabled on module removal path.
 228          * Otherwise we can get spurious warnings. */
 229         NF_CT_STAT_INC(net, delete_list);
 230         clean_from_lists(ct);
 231         spin_unlock_bh(&nf_conntrack_lock);
 232 }
 233 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
 234
 235 static void death_by_event(unsigned long ul_conntrack)
 236 {
 237         struct nf_conn *ct = (void *)ul_conntrack;
 238         struct net *net = nf_ct_net(ct);
 239
 240         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 241                 /* bad luck, let's retry again */
 242                 ct->timeout.expires = jiffies +
 243                         (random32() % net->ct.sysctl_events_retry_timeout);
 244                 add_timer(&ct->timeout);
 245                 return;
 246         }
 247         /* we've got the event delivered, now it's dying */
 248         set_bit(IPS_DYING_BIT, &ct->status);
 249         spin_lock(&nf_conntrack_lock);
 250         hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 251         spin_unlock(&nf_conntrack_lock);
 252         nf_ct_put(ct);
 253 }
 254
 255 void nf_ct_insert_dying_list(struct nf_conn *ct)
 256 {
 257         struct net *net = nf_ct_net(ct);
 258
 259         /* add this conntrack to the dying list */
 260         spin_lock_bh(&nf_conntrack_lock);
 261         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 262                              &net->ct.dying);
 263         spin_unlock_bh(&nf_conntrack_lock);
 264         /* set a new timer to retry event delivery */
 265         setup_timer(&ct->timeout, death_by_event, (unsigned long)ct);
 266         ct->timeout.expires = jiffies +
 267                 (random32() % net->ct.sysctl_events_retry_timeout);
 268         add_timer(&ct->timeout);
 269 }
 270 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list);
 271
 272 static void death_by_timeout(unsigned long ul_conntrack)
 273 {
 274         struct nf_conn *ct = (void *)ul_conntrack;
 275
 276         if (!test_bit(IPS_DYING_BIT, &ct->status) &&
 277             unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
 278                 /* destroy event was not delivered */
 279                 nf_ct_delete_from_lists(ct);
 280                 nf_ct_insert_dying_list(ct);
 281                 return;
 282         }
 283         set_bit(IPS_DYING_BIT, &ct->status);
 284         nf_ct_delete_from_lists(ct);
 285         nf_ct_put(ct);
 286 }
 287
 288 /*
 289  * Warning :
 290  * - Caller must take a reference on returned object
 291  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 292  * OR
 293  * - Caller must lock nf_conntrack_lock before calling this function
 294  */
 295 struct nf_conntrack_tuple_hash *
 296 __nf_conntrack_find(struct net *net, u16 zone,
 297                     const struct nf_conntrack_tuple *tuple)
 298 {
 299         struct nf_conntrack_tuple_hash *h;
 300         struct hlist_nulls_node *n;
 301         unsigned int hash = hash_conntrack(net, zone, tuple);
 302
 303         /* Disable BHs the entire time since we normally need to disable them
 304          * at least once for the stats anyway.
 305          */
 306         local_bh_disable();
 307 begin:
 308         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 309                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 310                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 311                         NF_CT_STAT_INC(net, found);
 312                         local_bh_enable();
 313                         return h;
 314                 }
 315                 NF_CT_STAT_INC(net, searched);
 316         }
 317         /*
 318          * if the nulls value we got at the end of this lookup is
 319          * not the expected one, we must restart lookup.
 320          * We probably met an item that was moved to another chain.
 321          */
 322         if (get_nulls_value(n) != hash) {
 323                 NF_CT_STAT_INC(net, search_restart);
 324                 goto begin;
 325         }
 326         local_bh_enable();
 327
 328         return NULL;
 329 }
 330 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 331
 332 /* Find a connection corresponding to a tuple. */
 333 struct nf_conntrack_tuple_hash *
 334 nf_conntrack_find_get(struct net *net, u16 zone,
 335                       const struct nf_conntrack_tuple *tuple)
 336 {
 337         struct nf_conntrack_tuple_hash *h;
 338         struct nf_conn *ct;
 339
 340         rcu_read_lock();
 341 begin:
 342         h = __nf_conntrack_find(net, zone, tuple);
 343         if (h) {
 344                 ct = nf_ct_tuplehash_to_ctrack(h);
 345                 if (unlikely(nf_ct_is_dying(ct) ||
 346                              !atomic_inc_not_zero(&ct->ct_general.use)))
 347                         h = NULL;
 348                 else {
 349                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 350                                      nf_ct_zone(ct) != zone)) {
 351                                 nf_ct_put(ct);
 352                                 goto begin;
 353                         }
 354                 }
 355         }
 356         rcu_read_unlock();
 357
 358         return h;
 359 }
 360 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 361
 362 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 363                                        unsigned int hash,
 364                                        unsigned int repl_hash)
 365 {
 366         struct net *net = nf_ct_net(ct);
 367
 368         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 369                            &net->ct.hash[hash]);
 370         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 371                            &net->ct.hash[repl_hash]);
 372 }
 373
 374 void nf_conntrack_hash_insert(struct nf_conn *ct)
 375 {
 376         struct net *net = nf_ct_net(ct);
 377         unsigned int hash, repl_hash;
 378         u16 zone;
 379
 380         zone = nf_ct_zone(ct);
 381         hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 382         repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 383
 384         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 385 }
 386 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 387
 388 /* Confirm a connection given skb; places it in hash table */
 389 int
 390 __nf_conntrack_confirm(struct sk_buff *skb)
 391 {
 392         unsigned int hash, repl_hash;
 393         struct nf_conntrack_tuple_hash *h;
 394         struct nf_conn *ct;
 395         struct nf_conn_help *help;
 396         struct hlist_nulls_node *n;
 397         enum ip_conntrack_info ctinfo;
 398         struct net *net;
 399         u16 zone;
 400
 401         ct = nf_ct_get(skb, &ctinfo);
 402         net = nf_ct_net(ct);
 403
 404         /* ipt_REJECT uses nf_conntrack_attach to attach related
 405            ICMP/TCP RST packets in other direction.  Actual packet
 406            which created connection will be IP_CT_NEW or for an
 407            expected connection, IP_CT_RELATED. */
 408         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 409                 return NF_ACCEPT;
 410
 411         zone = nf_ct_zone(ct);
 412         hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 413         repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 414
 415         /* We're not in hash table, and we refuse to set up related
 416            connections for unconfirmed conns.  But packet copies and
 417            REJECT will give spurious warnings here. */
 418         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 419
 420         /* No external references means noone else could have
 421            confirmed us. */
 422         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 423         pr_debug("Confirming conntrack %p\n", ct);
 424
 425         spin_lock_bh(&nf_conntrack_lock);
 426
 427         /* We have to check the DYING flag inside the lock to prevent
 428            a race against nf_ct_get_next_corpse() possibly called from
 429            user context, else we insert an already 'dead' hash, blocking
 430            further use of that particular connection -JM */
 431
 432         if (unlikely(nf_ct_is_dying(ct))) {
 433                 spin_unlock_bh(&nf_conntrack_lock);
 434                 return NF_ACCEPT;
 435         }
 436
 437         /* See if there's one in the list already, including reverse:
 438            NAT could have grabbed it without realizing, since we're
 439            not in the hash.  If there is, we lost race. */
 440         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 441                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 442                                       &h->tuple) &&
 443                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 444                         goto out;
 445         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 446                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 447                                       &h->tuple) &&
 448                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 449                         goto out;
 450
 451         /* Remove from unconfirmed list */
 452         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 453
 454         /* Timer relative to confirmation time, not original
 455            setting time, otherwise we'd get timer wrap in
 456            weird delay cases. */
 457         ct->timeout.expires += jiffies;
 458         add_timer(&ct->timeout);
 459         atomic_inc(&ct->ct_general.use);
 460         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 461
 462         /* Since the lookup is lockless, hash insertion must be done after
 463          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 464          * guarantee that no other CPU can find the conntrack before the above
 465          * stores are visible.
 466          */
 467         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 468         NF_CT_STAT_INC(net, insert);
 469         spin_unlock_bh(&nf_conntrack_lock);
 470
 471         help = nfct_help(ct);
 472         if (help && help->helper)
 473                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 474
 475         nf_conntrack_event_cache(master_ct(ct) ?
 476                                  IPCT_RELATED : IPCT_NEW, ct);
 477         return NF_ACCEPT;
 478
 479 out:
 480         NF_CT_STAT_INC(net, insert_failed);
 481         spin_unlock_bh(&nf_conntrack_lock);
 482         return NF_DROP;
 483 }
 484 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 485
 486 /* Returns true if a connection correspondings to the tuple (required
 487    for NAT). */
 488 int
 489 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 490                          const struct nf_conn *ignored_conntrack)
 491 {
 492         struct net *net = nf_ct_net(ignored_conntrack);
 493         struct nf_conntrack_tuple_hash *h;
 494         struct hlist_nulls_node *n;
 495         struct nf_conn *ct;
 496         u16 zone = nf_ct_zone(ignored_conntrack);
 497         unsigned int hash = hash_conntrack(net, zone, tuple);
 498
 499         /* Disable BHs the entire time since we need to disable them at
 500          * least once for the stats anyway.
 501          */
 502         rcu_read_lock_bh();
 503         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 504                 ct = nf_ct_tuplehash_to_ctrack(h);
 505                 if (ct != ignored_conntrack &&
 506                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 507                     nf_ct_zone(ct) == zone) {
 508                         NF_CT_STAT_INC(net, found);
 509                         rcu_read_unlock_bh();
 510                         return 1;
 511                 }
 512                 NF_CT_STAT_INC(net, searched);
 513         }
 514         rcu_read_unlock_bh();
 515
 516         return 0;
 517 }
 518 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 519
 520 #define NF_CT_EVICTION_RANGE    8
 521
 522 /* There's a small race here where we may free a just-assured
 523    connection.  Too bad: we're in trouble anyway. */
 524 static noinline int early_drop(struct net *net, unsigned int hash)
 525 {
 526         /* Use oldest entry, which is roughly LRU */
 527         struct nf_conntrack_tuple_hash *h;
 528         struct nf_conn *ct = NULL, *tmp;
 529         struct hlist_nulls_node *n;
 530         unsigned int i, cnt = 0;
 531         int dropped = 0;
 532
 533         rcu_read_lock();
 534         for (i = 0; i < net->ct.htable_size; i++) {
 535                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 536                                          hnnode) {
 537                         tmp = nf_ct_tuplehash_to_ctrack(h);
 538                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 539                                 ct = tmp;
 540                         cnt++;
 541                 }
 542
 543                 if (ct != NULL) {
 544                         if (likely(!nf_ct_is_dying(ct) &&
 545                                    atomic_inc_not_zero(&ct->ct_general.use)))
 546                                 break;
 547                         else
 548                                 ct = NULL;
 549                 }
 550
 551                 if (cnt >= NF_CT_EVICTION_RANGE)
 552                         break;
 553
 554                 hash = (hash + 1) % net->ct.htable_size;
 555         }
 556         rcu_read_unlock();
 557
 558         if (!ct)
 559                 return dropped;
 560
 561         if (del_timer(&ct->timeout)) {
 562                 death_by_timeout((unsigned long)ct);
 563                 dropped = 1;
 564                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 565         }
 566         nf_ct_put(ct);
 567         return dropped;
 568 }
 569
 570 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 571                                    const struct nf_conntrack_tuple *orig,
 572                                    const struct nf_conntrack_tuple *repl,
 573                                    gfp_t gfp)
 574 {
 575         struct nf_conn *ct;
 576
 577         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 578                 get_random_bytes(&nf_conntrack_hash_rnd,
 579                                 sizeof(nf_conntrack_hash_rnd));
 580                 nf_conntrack_hash_rnd_initted = 1;
 581         }
 582
 583         /* We don't want any race condition at early drop stage */
 584         atomic_inc(&net->ct.count);
 585
 586         if (nf_conntrack_max &&
 587             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 588                 unsigned int hash = hash_conntrack(net, zone, orig);
 589                 if (!early_drop(net, hash)) {
 590                         atomic_dec(&net->ct.count);
 591                         if (net_ratelimit())
 592                                 printk(KERN_WARNING
 593                                        "nf_conntrack: table full, dropping"
 594                                        " packet.\n");
 595                         return ERR_PTR(-ENOMEM);
 596                 }
 597         }
 598
 599         /*
 600          * Do not use kmem_cache_zalloc(), as this cache uses
 601          * SLAB_DESTROY_BY_RCU.
 602          */
 603         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 604         if (ct == NULL) {
 605                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
 606                 atomic_dec(&net->ct.count);
 607                 return ERR_PTR(-ENOMEM);
 608         }
 609         /*
 610          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 611          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 612          */
 613         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 614                sizeof(*ct) - offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 615         spin_lock_init(&ct->lock);
 616         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 617         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 618         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 619         ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev = NULL;
 620         /* Don't set timer yet: wait for confirmation */
 621         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 622 #ifdef CONFIG_NET_NS
 623         ct->ct_net = net;
 624 #endif
 625 #ifdef CONFIG_NF_CONNTRACK_ZONES
 626         if (zone) {
 627                 struct nf_conntrack_zone *nf_ct_zone;
 628
 629                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 630                 if (!nf_ct_zone)
 631                         goto out_free;
 632                 nf_ct_zone->id = zone;
 633         }
 634 #endif
 635         /*
 636          * changes to lookup keys must be done before setting refcnt to 1
 637          */
 638         smp_wmb();
 639         atomic_set(&ct->ct_general.use, 1);
 640         return ct;
 641
 642 #ifdef CONFIG_NF_CONNTRACK_ZONES
 643 out_free:
 644         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 645         return ERR_PTR(-ENOMEM);
 646 #endif
 647 }
 648 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 649
 650 void nf_conntrack_free(struct nf_conn *ct)
 651 {
 652         struct net *net = nf_ct_net(ct);
 653
 654         nf_ct_ext_destroy(ct);
 655         atomic_dec(&net->ct.count);
 656         nf_ct_ext_free(ct);
 657         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 658 }
 659 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 660
 661 /* Allocate a new conntrack: we return -ENOMEM if classification
 662    failed due to stress.  Otherwise it really is unclassifiable. */
 663 static struct nf_conntrack_tuple_hash *
 664 init_conntrack(struct net *net, struct nf_conn *tmpl,
 665                const struct nf_conntrack_tuple *tuple,
 666                struct nf_conntrack_l3proto *l3proto,
 667                struct nf_conntrack_l4proto *l4proto,
 668                struct sk_buff *skb,
 669                unsigned int dataoff)
 670 {
 671         struct nf_conn *ct;
 672         struct nf_conn_help *help;
 673         struct nf_conntrack_tuple repl_tuple;
 674         struct nf_conntrack_ecache *ecache;
 675         struct nf_conntrack_expect *exp;
 676         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 677
 678         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 679                 pr_debug("Can't invert tuple.\n");
 680                 return NULL;
 681         }
 682
 683         ct = nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC);
 684         if (IS_ERR(ct)) {
 685                 pr_debug("Can't allocate conntrack.\n");
 686                 return (struct nf_conntrack_tuple_hash *)ct;
 687         }
 688
 689         if (!l4proto->new(ct, skb, dataoff)) {
 690                 nf_conntrack_free(ct);
 691                 pr_debug("init conntrack: can't track with proto module\n");
 692                 return NULL;
 693         }
 694
 695         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 696
 697         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 698         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 699                                  ecache ? ecache->expmask : 0,
 700                              GFP_ATOMIC);
 701
 702         spin_lock_bh(&nf_conntrack_lock);
 703         exp = nf_ct_find_expectation(net, zone, tuple);
 704         if (exp) {
 705                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 706                          ct, exp);
 707                 /* Welcome, Mr. Bond.  We've been expecting you... */
 708                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 709                 ct->master = exp->master;
 710                 if (exp->helper) {
 711                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 712                         if (help)
 713                                 rcu_assign_pointer(help->helper, exp->helper);
 714                 }
 715
 716 #ifdef CONFIG_NF_CONNTRACK_MARK
 717                 ct->mark = exp->master->mark;
 718 #endif
 719 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 720                 ct->secmark = exp->master->secmark;
 721 #endif
 722                 nf_conntrack_get(&ct->master->ct_general);
 723                 NF_CT_STAT_INC(net, expect_new);
 724         } else {
 725                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 726                 NF_CT_STAT_INC(net, new);
 727         }
 728
 729         /* Overload tuple linked list to put us in unconfirmed list. */
 730         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 731                        &net->ct.unconfirmed);
 732
 733         spin_unlock_bh(&nf_conntrack_lock);
 734
 735         if (exp) {
 736                 if (exp->expectfn)
 737                         exp->expectfn(ct, exp);
 738                 nf_ct_expect_put(exp);
 739         }
 740
 741         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 742 }
 743
 744 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 745 static inline struct nf_conn *
 746 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 747                   struct sk_buff *skb,
 748                   unsigned int dataoff,
 749                   u_int16_t l3num,
 750                   u_int8_t protonum,
 751                   struct nf_conntrack_l3proto *l3proto,
 752                   struct nf_conntrack_l4proto *l4proto,
 753                   int *set_reply,
 754                   enum ip_conntrack_info *ctinfo)
 755 {
 756         struct nf_conntrack_tuple tuple;
 757         struct nf_conntrack_tuple_hash *h;
 758         struct nf_conn *ct;
 759         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 760
 761         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 762                              dataoff, l3num, protonum, &tuple, l3proto,
 763                              l4proto)) {
 764                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 765                 return NULL;
 766         }
 767
 768         /* look for tuple match */
 769         h = nf_conntrack_find_get(net, zone, &tuple);
 770         if (!h) {
 771                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 772                                    skb, dataoff);
 773                 if (!h)
 774                         return NULL;
 775                 if (IS_ERR(h))
 776                         return (void *)h;
 777         }
 778         ct = nf_ct_tuplehash_to_ctrack(h);
 779
 780         /* It exists; we have (non-exclusive) reference. */
 781         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 782                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 783                 /* Please set reply bit if this packet OK */
 784                 *set_reply = 1;
 785         } else {
 786                 /* Once we've had two way comms, always ESTABLISHED. */
 787                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 788                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 789                         *ctinfo = IP_CT_ESTABLISHED;
 790                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 791                         pr_debug("nf_conntrack_in: related packet for %p\n",
 792                                  ct);
 793                         *ctinfo = IP_CT_RELATED;
 794                 } else {
 795                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 796                         *ctinfo = IP_CT_NEW;
 797                 }
 798                 *set_reply = 0;
 799         }
 800         skb->nfct = &ct->ct_general;
 801         skb->nfctinfo = *ctinfo;
 802         return ct;
 803 }
 804
 805 unsigned int
 806 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 807                 struct sk_buff *skb)
 808 {
 809         struct nf_conn *ct, *tmpl = NULL;
 810         enum ip_conntrack_info ctinfo;
 811         struct nf_conntrack_l3proto *l3proto;
 812         struct nf_conntrack_l4proto *l4proto;
 813         unsigned int dataoff;
 814         u_int8_t protonum;
 815         int set_reply = 0;
 816         int ret;
 817
 818         if (skb->nfct) {
 819                 /* Previously seen (loopback or untracked)?  Ignore. */
 820                 tmpl = (struct nf_conn *)skb->nfct;
 821                 if (!nf_ct_is_template(tmpl)) {
 822                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 823                         return NF_ACCEPT;
 824                 }
 825                 skb->nfct = NULL;
 826         }
 827
 828         /* rcu_read_lock()ed by nf_hook_slow */
 829         l3proto = __nf_ct_l3proto_find(pf);
 830         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 831                                    &dataoff, &protonum);
 832         if (ret <= 0) {
 833                 pr_debug("not prepared to track yet or error occured\n");
 834                 NF_CT_STAT_INC_ATOMIC(net, error);
 835                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 836                 ret = -ret;
 837                 goto out;
 838         }
 839
 840         l4proto = __nf_ct_l4proto_find(pf, protonum);
 841
 842         /* It may be an special packet, error, unclean...
 843          * inverse of the return code tells to the netfilter
 844          * core what to do with the packet. */
 845         if (l4proto->error != NULL) {
 846                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 847                                      pf, hooknum);
 848                 if (ret <= 0) {
 849                         NF_CT_STAT_INC_ATOMIC(net, error);
 850                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 851                         ret = -ret;
 852                         goto out;
 853                 }
 854         }
 855
 856         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 857                                l3proto, l4proto, &set_reply, &ctinfo);
 858         if (!ct) {
 859                 /* Not valid part of a connection */
 860                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 861                 ret = NF_ACCEPT;
 862                 goto out;
 863         }
 864
 865         if (IS_ERR(ct)) {
 866                 /* Too stressed to deal. */
 867                 NF_CT_STAT_INC_ATOMIC(net, drop);
 868                 ret = NF_DROP;
 869                 goto out;
 870         }
 871
 872         NF_CT_ASSERT(skb->nfct);
 873
 874         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 875         if (ret <= 0) {
 876                 /* Invalid: inverse of the return code tells
 877                  * the netfilter core what to do */
 878                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 879                 nf_conntrack_put(skb->nfct);
 880                 skb->nfct = NULL;
 881                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 882                 if (ret == -NF_DROP)
 883                         NF_CT_STAT_INC_ATOMIC(net, drop);
 884                 ret = -ret;
 885                 goto out;
 886         }
 887
 888         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 889                 nf_conntrack_event_cache(IPCT_REPLY, ct);
 890 out:
 891         if (tmpl)
 892                 nf_ct_put(tmpl);
 893
 894         return ret;
 895 }
 896 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 897
 898 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 899                           const struct nf_conntrack_tuple *orig)
 900 {
 901         bool ret;
 902
 903         rcu_read_lock();
 904         ret = nf_ct_invert_tuple(inverse, orig,
 905                                  __nf_ct_l3proto_find(orig->src.l3num),
 906                                  __nf_ct_l4proto_find(orig->src.l3num,
 907                                                       orig->dst.protonum));
 908         rcu_read_unlock();
 909         return ret;
 910 }
 911 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 912
 913 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 914    implicitly racy: see __nf_conntrack_confirm */
 915 void nf_conntrack_alter_reply(struct nf_conn *ct,
 916                               const struct nf_conntrack_tuple *newreply)
 917 {
 918         struct nf_conn_help *help = nfct_help(ct);
 919
 920         /* Should be unconfirmed, so not in hash table yet */
 921         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 922
 923         pr_debug("Altering reply tuple of %p to ", ct);
 924         nf_ct_dump_tuple(newreply);
 925
 926         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 927         if (ct->master || (help && !hlist_empty(&help->expectations)))
 928                 return;
 929
 930         rcu_read_lock();
 931         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
 932         rcu_read_unlock();
 933 }
 934 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 935
 936 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 937 void __nf_ct_refresh_acct(struct nf_conn *ct,
 938                           enum ip_conntrack_info ctinfo,
 939                           const struct sk_buff *skb,
 940                           unsigned long extra_jiffies,
 941                           int do_acct)
 942 {
 943         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 944         NF_CT_ASSERT(skb);
 945
 946         /* Only update if this is not a fixed timeout */
 947         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
 948                 goto acct;
 949
 950         /* If not in hash table, timer will not be active yet */
 951         if (!nf_ct_is_confirmed(ct)) {
 952                 ct->timeout.expires = extra_jiffies;
 953         } else {
 954                 unsigned long newtime = jiffies + extra_jiffies;
 955
 956                 /* Only update the timeout if the new timeout is at least
 957                    HZ jiffies from the old timeout. Need del_timer for race
 958                    avoidance (may already be dying). */
 959                 if (newtime - ct->timeout.expires >= HZ)
 960                         mod_timer_pending(&ct->timeout, newtime);
 961         }
 962
 963 acct:
 964         if (do_acct) {
 965                 struct nf_conn_counter *acct;
 966
 967                 acct = nf_conn_acct_find(ct);
 968                 if (acct) {
 969                         spin_lock_bh(&ct->lock);
 970                         acct[CTINFO2DIR(ctinfo)].packets++;
 971                         acct[CTINFO2DIR(ctinfo)].bytes +=
 972                                 skb->len - skb_network_offset(skb);
 973                         spin_unlock_bh(&ct->lock);
 974                 }
 975         }
 976 }
 977 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 978
 979 bool __nf_ct_kill_acct(struct nf_conn *ct,
 980                        enum ip_conntrack_info ctinfo,
 981                        const struct sk_buff *skb,
 982                        int do_acct)
 983 {
 984         if (do_acct) {
 985                 struct nf_conn_counter *acct;
 986
 987                 acct = nf_conn_acct_find(ct);
 988                 if (acct) {
 989                         spin_lock_bh(&ct->lock);
 990                         acct[CTINFO2DIR(ctinfo)].packets++;
 991                         acct[CTINFO2DIR(ctinfo)].bytes +=
 992                                 skb->len - skb_network_offset(skb);
 993                         spin_unlock_bh(&ct->lock);
 994                 }
 995         }
 996
 997         if (del_timer(&ct->timeout)) {
 998                 ct->timeout.function((unsigned long)ct);
 999                 return true;
1000         }
1001         return false;
1002 }
1003 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1004
1005 #ifdef CONFIG_NF_CONNTRACK_ZONES
1006 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1007         .len    = sizeof(struct nf_conntrack_zone),
1008         .align  = __alignof__(struct nf_conntrack_zone),
1009         .id     = NF_CT_EXT_ZONE,
1010 };
1011 #endif
1012
1013 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1014
1015 #include <linux/netfilter/nfnetlink.h>
1016 #include <linux/netfilter/nfnetlink_conntrack.h>
1017 #include <linux/mutex.h>
1018
1019 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1020  * in ip_conntrack_core, since we don't want the protocols to autoload
1021  * or depend on ctnetlink */
1022 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1023                                const struct nf_conntrack_tuple *tuple)
1024 {
1025         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
1026         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
1027         return 0;
1028
1029 nla_put_failure:
1030         return -1;
1031 }
1032 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1033
1034 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1035         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1036         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1037 };
1038 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1039
1040 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1041                                struct nf_conntrack_tuple *t)
1042 {
1043         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1044                 return -EINVAL;
1045
1046         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1047         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1048
1049         return 0;
1050 }
1051 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1052
1053 int nf_ct_port_nlattr_tuple_size(void)
1054 {
1055         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1056 }
1057 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1058 #endif
1059
1060 /* Used by ipt_REJECT and ip6t_REJECT. */
1061 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1062 {
1063         struct nf_conn *ct;
1064         enum ip_conntrack_info ctinfo;
1065
1066         /* This ICMP is in reverse direction to the packet which caused it */
1067         ct = nf_ct_get(skb, &ctinfo);
1068         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1069                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
1070         else
1071                 ctinfo = IP_CT_RELATED;
1072
1073         /* Attach to new skbuff, and increment count */
1074         nskb->nfct = &ct->ct_general;
1075         nskb->nfctinfo = ctinfo;
1076         nf_conntrack_get(nskb->nfct);
1077 }
1078
1079 /* Bring out ya dead! */
1080 static struct nf_conn *
1081 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1082                 void *data, unsigned int *bucket)
1083 {
1084         struct nf_conntrack_tuple_hash *h;
1085         struct nf_conn *ct;
1086         struct hlist_nulls_node *n;
1087
1088         spin_lock_bh(&nf_conntrack_lock);
1089         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1090                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1091                         ct = nf_ct_tuplehash_to_ctrack(h);
1092                         if (iter(ct, data))
1093                                 goto found;
1094                 }
1095         }
1096         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1097                 ct = nf_ct_tuplehash_to_ctrack(h);
1098                 if (iter(ct, data))
1099                         set_bit(IPS_DYING_BIT, &ct->status);
1100         }
1101         spin_unlock_bh(&nf_conntrack_lock);
1102         return NULL;
1103 found:
1104         atomic_inc(&ct->ct_general.use);
1105         spin_unlock_bh(&nf_conntrack_lock);
1106         return ct;
1107 }
1108
1109 void nf_ct_iterate_cleanup(struct net *net,
1110                            int (*iter)(struct nf_conn *i, void *data),
1111                            void *data)
1112 {
1113         struct nf_conn *ct;
1114         unsigned int bucket = 0;
1115
1116         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1117                 /* Time to push up daises... */
1118                 if (del_timer(&ct->timeout))
1119                         death_by_timeout((unsigned long)ct);
1120                 /* ... else the timer will get him soon. */
1121
1122                 nf_ct_put(ct);
1123         }
1124 }
1125 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1126
1127 struct __nf_ct_flush_report {
1128         u32 pid;
1129         int report;
1130 };
1131
1132 static int kill_report(struct nf_conn *i, void *data)
1133 {
1134         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1135
1136         /* If we fail to deliver the event, death_by_timeout() will retry */
1137         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1138                                       fr->pid, fr->report) < 0)
1139                 return 1;
1140
1141         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1142         set_bit(IPS_DYING_BIT, &i->status);
1143         return 1;
1144 }
1145
1146 static int kill_all(struct nf_conn *i, void *data)
1147 {
1148         return 1;
1149 }
1150
1151 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1152 {
1153         if (vmalloced)
1154                 vfree(hash);
1155         else
1156                 free_pages((unsigned long)hash,
1157                            get_order(sizeof(struct hlist_head) * size));
1158 }
1159 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1160
1161 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1162 {
1163         struct __nf_ct_flush_report fr = {
1164                 .pid    = pid,
1165                 .report = report,
1166         };
1167         nf_ct_iterate_cleanup(net, kill_report, &fr);
1168 }
1169 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1170
1171 static void nf_ct_release_dying_list(struct net *net)
1172 {
1173         struct nf_conntrack_tuple_hash *h;
1174         struct nf_conn *ct;
1175         struct hlist_nulls_node *n;
1176
1177         spin_lock_bh(&nf_conntrack_lock);
1178         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1179                 ct = nf_ct_tuplehash_to_ctrack(h);
1180                 /* never fails to remove them, no listeners at this point */
1181                 nf_ct_kill(ct);
1182         }
1183         spin_unlock_bh(&nf_conntrack_lock);
1184 }
1185
1186 static void nf_conntrack_cleanup_init_net(void)
1187 {
1188         /* wait until all references to nf_conntrack_untracked are dropped */
1189         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1190                 schedule();
1191
1192         nf_conntrack_helper_fini();
1193         nf_conntrack_proto_fini();
1194 #ifdef CONFIG_NF_CONNTRACK_ZONES
1195         nf_ct_extend_unregister(&nf_ct_zone_extend);
1196 #endif
1197 }
1198
1199 static void nf_conntrack_cleanup_net(struct net *net)
1200 {
1201  i_see_dead_people:
1202         nf_ct_iterate_cleanup(net, kill_all, NULL);
1203         nf_ct_release_dying_list(net);
1204         if (atomic_read(&net->ct.count) != 0) {
1205                 schedule();
1206                 goto i_see_dead_people;
1207         }
1208
1209         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1210                              net->ct.htable_size);
1211         nf_conntrack_ecache_fini(net);
1212         nf_conntrack_acct_fini(net);
1213         nf_conntrack_expect_fini(net);
1214         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1215         kfree(net->ct.slabname);
1216         free_percpu(net->ct.stat);
1217 }
1218
1219 /* Mishearing the voices in his head, our hero wonders how he's
1220    supposed to kill the mall. */
1221 void nf_conntrack_cleanup(struct net *net)
1222 {
1223         if (net_eq(net, &init_net))
1224                 rcu_assign_pointer(ip_ct_attach, NULL);
1225
1226         /* This makes sure all current packets have passed through
1227            netfilter framework.  Roll on, two-stage module
1228            delete... */
1229         synchronize_net();
1230
1231         nf_conntrack_cleanup_net(net);
1232
1233         if (net_eq(net, &init_net)) {
1234                 rcu_assign_pointer(nf_ct_destroy, NULL);
1235                 nf_conntrack_cleanup_init_net();
1236         }
1237 }
1238
1239 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1240 {
1241         struct hlist_nulls_head *hash;
1242         unsigned int nr_slots, i;
1243         size_t sz;
1244
1245         *vmalloced = 0;
1246
1247         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1248         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1249         sz = nr_slots * sizeof(struct hlist_nulls_head);
1250         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1251                                         get_order(sz));
1252         if (!hash) {
1253                 *vmalloced = 1;
1254                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1255                 hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
1256         }
1257
1258         if (hash && nulls)
1259                 for (i = 0; i < nr_slots; i++)
1260                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1261
1262         return hash;
1263 }
1264 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1265
1266 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1267 {
1268         int i, bucket, vmalloced, old_vmalloced;
1269         unsigned int hashsize, old_size;
1270         struct hlist_nulls_head *hash, *old_hash;
1271         struct nf_conntrack_tuple_hash *h;
1272         struct nf_conn *ct;
1273
1274         if (current->nsproxy->net_ns != &init_net)
1275                 return -EOPNOTSUPP;
1276
1277         /* On boot, we can set this without any fancy locking. */
1278         if (!nf_conntrack_htable_size)
1279                 return param_set_uint(val, kp);
1280
1281         hashsize = simple_strtoul(val, NULL, 0);
1282         if (!hashsize)
1283                 return -EINVAL;
1284
1285         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1286         if (!hash)
1287                 return -ENOMEM;
1288
1289         /* Lookups in the old hash might happen in parallel, which means we
1290          * might get false negatives during connection lookup. New connections
1291          * created because of a false negative won't make it into the hash
1292          * though since that required taking the lock.
1293          */
1294         spin_lock_bh(&nf_conntrack_lock);
1295         for (i = 0; i < init_net.ct.htable_size; i++) {
1296                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1297                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1298                                         struct nf_conntrack_tuple_hash, hnnode);
1299                         ct = nf_ct_tuplehash_to_ctrack(h);
1300                         hlist_nulls_del_rcu(&h->hnnode);
1301                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1302                                                   hashsize,
1303                                                   nf_conntrack_hash_rnd);
1304                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1305                 }
1306         }
1307         old_size = init_net.ct.htable_size;
1308         old_vmalloced = init_net.ct.hash_vmalloc;
1309         old_hash = init_net.ct.hash;
1310
1311         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1312         init_net.ct.hash_vmalloc = vmalloced;
1313         init_net.ct.hash = hash;
1314         spin_unlock_bh(&nf_conntrack_lock);
1315
1316         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1317         return 0;
1318 }
1319 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1320
1321 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1322                   &nf_conntrack_htable_size, 0600);
1323
1324 static int nf_conntrack_init_init_net(void)
1325 {
1326         int max_factor = 8;
1327         int ret;
1328
1329         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1330          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1331         if (!nf_conntrack_htable_size) {
1332                 nf_conntrack_htable_size
1333                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1334                            / sizeof(struct hlist_head));
1335                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1336                         nf_conntrack_htable_size = 16384;
1337                 if (nf_conntrack_htable_size < 32)
1338                         nf_conntrack_htable_size = 32;
1339
1340                 /* Use a max. factor of four by default to get the same max as
1341                  * with the old struct list_heads. When a table size is given
1342                  * we use the old value of 8 to avoid reducing the max.
1343                  * entries. */
1344                 max_factor = 4;
1345         }
1346         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1347
1348         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1349                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1350                nf_conntrack_max);
1351
1352         ret = nf_conntrack_proto_init();
1353         if (ret < 0)
1354                 goto err_proto;
1355
1356         ret = nf_conntrack_helper_init();
1357         if (ret < 0)
1358                 goto err_helper;
1359
1360 #ifdef CONFIG_NF_CONNTRACK_ZONES
1361         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1362         if (ret < 0)
1363                 goto err_extend;
1364 #endif
1365         /* Set up fake conntrack: to never be deleted, not in any hashes */
1366 #ifdef CONFIG_NET_NS
1367         nf_conntrack_untracked.ct_net = &init_net;
1368 #endif
1369         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1370         /*  - and look it like as a confirmed connection */
1371         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1372
1373         return 0;
1374
1375 #ifdef CONFIG_NF_CONNTRACK_ZONES
1376 err_extend:
1377         nf_conntrack_helper_fini();
1378 #endif
1379 err_helper:
1380         nf_conntrack_proto_fini();
1381 err_proto:
1382         return ret;
1383 }
1384
1385 /*
1386  * We need to use special "null" values, not used in hash table
1387  */
1388 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1389 #define DYING_NULLS_VAL         ((1<<30)+1)
1390
1391 static int nf_conntrack_init_net(struct net *net)
1392 {
1393         int ret;
1394
1395         atomic_set(&net->ct.count, 0);
1396         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1397         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1398         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1399         if (!net->ct.stat) {
1400                 ret = -ENOMEM;
1401                 goto err_stat;
1402         }
1403
1404         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1405         if (!net->ct.slabname) {
1406                 ret = -ENOMEM;
1407                 goto err_slabname;
1408         }
1409
1410         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1411                                                         sizeof(struct nf_conn), 0,
1412                                                         SLAB_DESTROY_BY_RCU, NULL);
1413         if (!net->ct.nf_conntrack_cachep) {
1414                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1415                 ret = -ENOMEM;
1416                 goto err_cache;
1417         }
1418
1419         net->ct.htable_size = nf_conntrack_htable_size;
1420         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size,
1421                                              &net->ct.hash_vmalloc, 1);
1422         if (!net->ct.hash) {
1423                 ret = -ENOMEM;
1424                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1425                 goto err_hash;
1426         }
1427         ret = nf_conntrack_expect_init(net);
1428         if (ret < 0)
1429                 goto err_expect;
1430         ret = nf_conntrack_acct_init(net);
1431         if (ret < 0)
1432                 goto err_acct;
1433         ret = nf_conntrack_ecache_init(net);
1434         if (ret < 0)
1435                 goto err_ecache;
1436
1437         return 0;
1438
1439 err_ecache:
1440         nf_conntrack_acct_fini(net);
1441 err_acct:
1442         nf_conntrack_expect_fini(net);
1443 err_expect:
1444         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1445                              net->ct.htable_size);
1446 err_hash:
1447         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1448 err_cache:
1449         kfree(net->ct.slabname);
1450 err_slabname:
1451         free_percpu(net->ct.stat);
1452 err_stat:
1453         return ret;
1454 }
1455
1456 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1457                         enum ip_conntrack_dir dir,
1458                         u32 seq);
1459 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);
1460
1461 int nf_conntrack_init(struct net *net)
1462 {
1463         int ret;
1464
1465         if (net_eq(net, &init_net)) {
1466                 ret = nf_conntrack_init_init_net();
1467                 if (ret < 0)
1468                         goto out_init_net;
1469         }
1470         ret = nf_conntrack_init_net(net);
1471         if (ret < 0)
1472                 goto out_net;
1473
1474         if (net_eq(net, &init_net)) {
1475                 /* For use by REJECT target */
1476                 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1477                 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1478
1479                 /* Howto get NAT offsets */
1480                 rcu_assign_pointer(nf_ct_nat_offset, NULL);
1481         }
1482         return 0;
1483
1484 out_net:
1485         if (net_eq(net, &init_net))
1486                 nf_conntrack_cleanup_init_net();
1487 out_init_net:
1488         return ret;
1489 }