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