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