Pileus Git - ~andy/linux/blob - net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License version 2 as
  11  * published by the Free Software Foundation.
  12  */
  13
  14 #include <linux/types.h>
  15 #include <linux/netfilter.h>
  16 #include <linux/module.h>
  17 #include <linux/sched.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/proc_fs.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/stddef.h>
  22 #include <linux/slab.h>
  23 #include <linux/random.h>
  24 #include <linux/jhash.h>
  25 #include <linux/err.h>
  26 #include <linux/percpu.h>
  27 #include <linux/moduleparam.h>
  28 #include <linux/notifier.h>
  29 #include <linux/kernel.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/socket.h>
  32 #include <linux/mm.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/rculist_nulls.h>
  35
  36 #include <net/netfilter/nf_conntrack.h>
  37 #include <net/netfilter/nf_conntrack_l3proto.h>
  38 #include <net/netfilter/nf_conntrack_l4proto.h>
  39 #include <net/netfilter/nf_conntrack_expect.h>
  40 #include <net/netfilter/nf_conntrack_helper.h>
  41 #include <net/netfilter/nf_conntrack_core.h>
  42 #include <net/netfilter/nf_conntrack_extend.h>
  43 #include <net/netfilter/nf_conntrack_acct.h>
  44 #include <net/netfilter/nf_conntrack_ecache.h>
  45 #include <net/netfilter/nf_conntrack_zones.h>
  46 #include <net/netfilter/nf_conntrack_timestamp.h>
  47 #include <net/netfilter/nf_conntrack_timeout.h>
  48 #include <net/netfilter/nf_conntrack_labels.h>
  49 #include <net/netfilter/nf_nat.h>
  50 #include <net/netfilter/nf_nat_core.h>
  51
  52 #define NF_CONNTRACK_VERSION    "0.5.0"
  53
  54 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  55                                       enum nf_nat_manip_type manip,
  56                                       const struct nlattr *attr) __read_mostly;
  57 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  58
  59 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
  60                               struct nf_conn *ct,
  61                               enum ip_conntrack_info ctinfo,
  62                               unsigned int protoff);
  63 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
  64
  65 DEFINE_SPINLOCK(nf_conntrack_lock);
  66 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  67
  68 unsigned int nf_conntrack_htable_size __read_mostly;
  69 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  70
  71 unsigned int nf_conntrack_max __read_mostly;
  72 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  73
  74 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  75 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  76
  77 unsigned int nf_conntrack_hash_rnd __read_mostly;
  78 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
  79
  80 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  81 {
  82         unsigned int n;
  83
  84         /* The direction must be ignored, so we hash everything up to the
  85          * destination ports (which is a multiple of 4) and treat the last
  86          * three bytes manually.
  87          */
  88         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  89         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  90                       (((__force __u16)tuple->dst.u.all << 16) |
  91                       tuple->dst.protonum));
  92 }
  93
  94 static u32 __hash_bucket(u32 hash, unsigned int size)
  95 {
  96         return ((u64)hash * size) >> 32;
  97 }
  98
  99 static u32 hash_bucket(u32 hash, const struct net *net)
 100 {
 101         return __hash_bucket(hash, net->ct.htable_size);
 102 }
 103
 104 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
 105                                   u16 zone, unsigned int size)
 106 {
 107         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 108 }
 109
 110 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 111                                        const struct nf_conntrack_tuple *tuple)
 112 {
 113         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 114 }
 115
 116 bool
 117 nf_ct_get_tuple(const struct sk_buff *skb,
 118                 unsigned int nhoff,
 119                 unsigned int dataoff,
 120                 u_int16_t l3num,
 121                 u_int8_t protonum,
 122                 struct nf_conntrack_tuple *tuple,
 123                 const struct nf_conntrack_l3proto *l3proto,
 124                 const struct nf_conntrack_l4proto *l4proto)
 125 {
 126         memset(tuple, 0, sizeof(*tuple));
 127
 128         tuple->src.l3num = l3num;
 129         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 130                 return false;
 131
 132         tuple->dst.protonum = protonum;
 133         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 134
 135         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 136 }
 137 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 138
 139 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 140                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 141 {
 142         struct nf_conntrack_l3proto *l3proto;
 143         struct nf_conntrack_l4proto *l4proto;
 144         unsigned int protoff;
 145         u_int8_t protonum;
 146         int ret;
 147
 148         rcu_read_lock();
 149
 150         l3proto = __nf_ct_l3proto_find(l3num);
 151         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 152         if (ret != NF_ACCEPT) {
 153                 rcu_read_unlock();
 154                 return false;
 155         }
 156
 157         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 158
 159         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 160                               l3proto, l4proto);
 161
 162         rcu_read_unlock();
 163         return ret;
 164 }
 165 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 166
 167 bool
 168 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 169                    const struct nf_conntrack_tuple *orig,
 170                    const struct nf_conntrack_l3proto *l3proto,
 171                    const struct nf_conntrack_l4proto *l4proto)
 172 {
 173         memset(inverse, 0, sizeof(*inverse));
 174
 175         inverse->src.l3num = orig->src.l3num;
 176         if (l3proto->invert_tuple(inverse, orig) == 0)
 177                 return false;
 178
 179         inverse->dst.dir = !orig->dst.dir;
 180
 181         inverse->dst.protonum = orig->dst.protonum;
 182         return l4proto->invert_tuple(inverse, orig);
 183 }
 184 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 185
 186 static void
 187 clean_from_lists(struct nf_conn *ct)
 188 {
 189         pr_debug("clean_from_lists(%p)\n", ct);
 190         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 191         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 192
 193         /* Destroy all pending expectations */
 194         nf_ct_remove_expectations(ct);
 195 }
 196
 197 static void
 198 destroy_conntrack(struct nf_conntrack *nfct)
 199 {
 200         struct nf_conn *ct = (struct nf_conn *)nfct;
 201         struct net *net = nf_ct_net(ct);
 202         struct nf_conntrack_l4proto *l4proto;
 203
 204         pr_debug("destroy_conntrack(%p)\n", ct);
 205         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 206         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 207
 208         /* To make sure we don't get any weird locking issues here:
 209          * destroy_conntrack() MUST NOT be called with a write lock
 210          * to nf_conntrack_lock!!! -HW */
 211         rcu_read_lock();
 212         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 213         if (l4proto && l4proto->destroy)
 214                 l4proto->destroy(ct);
 215
 216         rcu_read_unlock();
 217
 218         spin_lock_bh(&nf_conntrack_lock);
 219         /* Expectations will have been removed in clean_from_lists,
 220          * except TFTP can create an expectation on the first packet,
 221          * before connection is in the list, so we need to clean here,
 222          * too. */
 223         nf_ct_remove_expectations(ct);
 224
 225         /* We overload first tuple to link into unconfirmed or dying list.*/
 226         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 227         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 228
 229         NF_CT_STAT_INC(net, delete);
 230         spin_unlock_bh(&nf_conntrack_lock);
 231
 232         if (ct->master)
 233                 nf_ct_put(ct->master);
 234
 235         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 236         nf_conntrack_free(ct);
 237 }
 238
 239 void nf_ct_delete_from_lists(struct nf_conn *ct)
 240 {
 241         struct net *net = nf_ct_net(ct);
 242
 243         nf_ct_helper_destroy(ct);
 244         spin_lock_bh(&nf_conntrack_lock);
 245         /* Inside lock so preempt is disabled on module removal path.
 246          * Otherwise we can get spurious warnings. */
 247         NF_CT_STAT_INC(net, delete_list);
 248         clean_from_lists(ct);
 249         /* add this conntrack to the dying list */
 250         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 251                              &net->ct.dying);
 252         spin_unlock_bh(&nf_conntrack_lock);
 253 }
 254 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
 255
 256 static void death_by_event(unsigned long ul_conntrack)
 257 {
 258         struct nf_conn *ct = (void *)ul_conntrack;
 259         struct net *net = nf_ct_net(ct);
 260         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 261
 262         BUG_ON(ecache == NULL);
 263
 264         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 265                 /* bad luck, let's retry again */
 266                 ecache->timeout.expires = jiffies +
 267                         (random32() % net->ct.sysctl_events_retry_timeout);
 268                 add_timer(&ecache->timeout);
 269                 return;
 270         }
 271         /* we've got the event delivered, now it's dying */
 272         set_bit(IPS_DYING_BIT, &ct->status);
 273         nf_ct_put(ct);
 274 }
 275
 276 void nf_ct_dying_timeout(struct nf_conn *ct)
 277 {
 278         struct net *net = nf_ct_net(ct);
 279         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 280
 281         BUG_ON(ecache == NULL);
 282
 283         /* set a new timer to retry event delivery */
 284         setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
 285         ecache->timeout.expires = jiffies +
 286                 (random32() % net->ct.sysctl_events_retry_timeout);
 287         add_timer(&ecache->timeout);
 288 }
 289 EXPORT_SYMBOL_GPL(nf_ct_dying_timeout);
 290
 291 static void death_by_timeout(unsigned long ul_conntrack)
 292 {
 293         struct nf_conn *ct = (void *)ul_conntrack;
 294         struct nf_conn_tstamp *tstamp;
 295
 296         tstamp = nf_conn_tstamp_find(ct);
 297         if (tstamp && tstamp->stop == 0)
 298                 tstamp->stop = ktime_to_ns(ktime_get_real());
 299
 300         if (!test_bit(IPS_DYING_BIT, &ct->status) &&
 301             unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
 302                 /* destroy event was not delivered */
 303                 nf_ct_delete_from_lists(ct);
 304                 nf_ct_dying_timeout(ct);
 305                 return;
 306         }
 307         set_bit(IPS_DYING_BIT, &ct->status);
 308         nf_ct_delete_from_lists(ct);
 309         nf_ct_put(ct);
 310 }
 311
 312 /*
 313  * Warning :
 314  * - Caller must take a reference on returned object
 315  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 316  * OR
 317  * - Caller must lock nf_conntrack_lock before calling this function
 318  */
 319 static struct nf_conntrack_tuple_hash *
 320 ____nf_conntrack_find(struct net *net, u16 zone,
 321                       const struct nf_conntrack_tuple *tuple, u32 hash)
 322 {
 323         struct nf_conntrack_tuple_hash *h;
 324         struct hlist_nulls_node *n;
 325         unsigned int bucket = hash_bucket(hash, net);
 326
 327         /* Disable BHs the entire time since we normally need to disable them
 328          * at least once for the stats anyway.
 329          */
 330         local_bh_disable();
 331 begin:
 332         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 333                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 334                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 335                         NF_CT_STAT_INC(net, found);
 336                         local_bh_enable();
 337                         return h;
 338                 }
 339                 NF_CT_STAT_INC(net, searched);
 340         }
 341         /*
 342          * if the nulls value we got at the end of this lookup is
 343          * not the expected one, we must restart lookup.
 344          * We probably met an item that was moved to another chain.
 345          */
 346         if (get_nulls_value(n) != bucket) {
 347                 NF_CT_STAT_INC(net, search_restart);
 348                 goto begin;
 349         }
 350         local_bh_enable();
 351
 352         return NULL;
 353 }
 354
 355 struct nf_conntrack_tuple_hash *
 356 __nf_conntrack_find(struct net *net, u16 zone,
 357                     const struct nf_conntrack_tuple *tuple)
 358 {
 359         return ____nf_conntrack_find(net, zone, tuple,
 360                                      hash_conntrack_raw(tuple, zone));
 361 }
 362 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 363
 364 /* Find a connection corresponding to a tuple. */
 365 static struct nf_conntrack_tuple_hash *
 366 __nf_conntrack_find_get(struct net *net, u16 zone,
 367                         const struct nf_conntrack_tuple *tuple, u32 hash)
 368 {
 369         struct nf_conntrack_tuple_hash *h;
 370         struct nf_conn *ct;
 371
 372         rcu_read_lock();
 373 begin:
 374         h = ____nf_conntrack_find(net, zone, tuple, hash);
 375         if (h) {
 376                 ct = nf_ct_tuplehash_to_ctrack(h);
 377                 if (unlikely(nf_ct_is_dying(ct) ||
 378                              !atomic_inc_not_zero(&ct->ct_general.use)))
 379                         h = NULL;
 380                 else {
 381                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 382                                      nf_ct_zone(ct) != zone)) {
 383                                 nf_ct_put(ct);
 384                                 goto begin;
 385                         }
 386                 }
 387         }
 388         rcu_read_unlock();
 389
 390         return h;
 391 }
 392
 393 struct nf_conntrack_tuple_hash *
 394 nf_conntrack_find_get(struct net *net, u16 zone,
 395                       const struct nf_conntrack_tuple *tuple)
 396 {
 397         return __nf_conntrack_find_get(net, zone, tuple,
 398                                        hash_conntrack_raw(tuple, zone));
 399 }
 400 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 401
 402 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 403                                        unsigned int hash,
 404                                        unsigned int repl_hash)
 405 {
 406         struct net *net = nf_ct_net(ct);
 407
 408         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 409                            &net->ct.hash[hash]);
 410         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 411                            &net->ct.hash[repl_hash]);
 412 }
 413
 414 int
 415 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 416 {
 417         struct net *net = nf_ct_net(ct);
 418         unsigned int hash, repl_hash;
 419         struct nf_conntrack_tuple_hash *h;
 420         struct hlist_nulls_node *n;
 421         u16 zone;
 422
 423         zone = nf_ct_zone(ct);
 424         hash = hash_conntrack(net, zone,
 425                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 426         repl_hash = hash_conntrack(net, zone,
 427                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 428
 429         spin_lock_bh(&nf_conntrack_lock);
 430
 431         /* See if there's one in the list already, including reverse */
 432         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 433                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 434                                       &h->tuple) &&
 435                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 436                         goto out;
 437         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 438                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 439                                       &h->tuple) &&
 440                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 441                         goto out;
 442
 443         add_timer(&ct->timeout);
 444         nf_conntrack_get(&ct->ct_general);
 445         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 446         NF_CT_STAT_INC(net, insert);
 447         spin_unlock_bh(&nf_conntrack_lock);
 448
 449         return 0;
 450
 451 out:
 452         NF_CT_STAT_INC(net, insert_failed);
 453         spin_unlock_bh(&nf_conntrack_lock);
 454         return -EEXIST;
 455 }
 456 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 457
 458 /* Confirm a connection given skb; places it in hash table */
 459 int
 460 __nf_conntrack_confirm(struct sk_buff *skb)
 461 {
 462         unsigned int hash, repl_hash;
 463         struct nf_conntrack_tuple_hash *h;
 464         struct nf_conn *ct;
 465         struct nf_conn_help *help;
 466         struct nf_conn_tstamp *tstamp;
 467         struct hlist_nulls_node *n;
 468         enum ip_conntrack_info ctinfo;
 469         struct net *net;
 470         u16 zone;
 471
 472         ct = nf_ct_get(skb, &ctinfo);
 473         net = nf_ct_net(ct);
 474
 475         /* ipt_REJECT uses nf_conntrack_attach to attach related
 476            ICMP/TCP RST packets in other direction.  Actual packet
 477            which created connection will be IP_CT_NEW or for an
 478            expected connection, IP_CT_RELATED. */
 479         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 480                 return NF_ACCEPT;
 481
 482         zone = nf_ct_zone(ct);
 483         /* reuse the hash saved before */
 484         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 485         hash = hash_bucket(hash, net);
 486         repl_hash = hash_conntrack(net, zone,
 487                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 488
 489         /* We're not in hash table, and we refuse to set up related
 490            connections for unconfirmed conns.  But packet copies and
 491            REJECT will give spurious warnings here. */
 492         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 493
 494         /* No external references means no one else could have
 495            confirmed us. */
 496         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 497         pr_debug("Confirming conntrack %p\n", ct);
 498
 499         spin_lock_bh(&nf_conntrack_lock);
 500
 501         /* We have to check the DYING flag inside the lock to prevent
 502            a race against nf_ct_get_next_corpse() possibly called from
 503            user context, else we insert an already 'dead' hash, blocking
 504            further use of that particular connection -JM */
 505
 506         if (unlikely(nf_ct_is_dying(ct))) {
 507                 spin_unlock_bh(&nf_conntrack_lock);
 508                 return NF_ACCEPT;
 509         }
 510
 511         /* See if there's one in the list already, including reverse:
 512            NAT could have grabbed it without realizing, since we're
 513            not in the hash.  If there is, we lost race. */
 514         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 515                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 516                                       &h->tuple) &&
 517                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 518                         goto out;
 519         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 520                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 521                                       &h->tuple) &&
 522                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 523                         goto out;
 524
 525         /* Remove from unconfirmed list */
 526         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 527
 528         /* Timer relative to confirmation time, not original
 529            setting time, otherwise we'd get timer wrap in
 530            weird delay cases. */
 531         ct->timeout.expires += jiffies;
 532         add_timer(&ct->timeout);
 533         atomic_inc(&ct->ct_general.use);
 534         ct->status |= IPS_CONFIRMED;
 535
 536         /* set conntrack timestamp, if enabled. */
 537         tstamp = nf_conn_tstamp_find(ct);
 538         if (tstamp) {
 539                 if (skb->tstamp.tv64 == 0)
 540                         __net_timestamp(skb);
 541
 542                 tstamp->start = ktime_to_ns(skb->tstamp);
 543         }
 544         /* Since the lookup is lockless, hash insertion must be done after
 545          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 546          * guarantee that no other CPU can find the conntrack before the above
 547          * stores are visible.
 548          */
 549         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 550         NF_CT_STAT_INC(net, insert);
 551         spin_unlock_bh(&nf_conntrack_lock);
 552
 553         help = nfct_help(ct);
 554         if (help && help->helper)
 555                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 556
 557         nf_conntrack_event_cache(master_ct(ct) ?
 558                                  IPCT_RELATED : IPCT_NEW, ct);
 559         return NF_ACCEPT;
 560
 561 out:
 562         NF_CT_STAT_INC(net, insert_failed);
 563         spin_unlock_bh(&nf_conntrack_lock);
 564         return NF_DROP;
 565 }
 566 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 567
 568 /* Returns true if a connection correspondings to the tuple (required
 569    for NAT). */
 570 int
 571 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 572                          const struct nf_conn *ignored_conntrack)
 573 {
 574         struct net *net = nf_ct_net(ignored_conntrack);
 575         struct nf_conntrack_tuple_hash *h;
 576         struct hlist_nulls_node *n;
 577         struct nf_conn *ct;
 578         u16 zone = nf_ct_zone(ignored_conntrack);
 579         unsigned int hash = hash_conntrack(net, zone, tuple);
 580
 581         /* Disable BHs the entire time since we need to disable them at
 582          * least once for the stats anyway.
 583          */
 584         rcu_read_lock_bh();
 585         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 586                 ct = nf_ct_tuplehash_to_ctrack(h);
 587                 if (ct != ignored_conntrack &&
 588                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 589                     nf_ct_zone(ct) == zone) {
 590                         NF_CT_STAT_INC(net, found);
 591                         rcu_read_unlock_bh();
 592                         return 1;
 593                 }
 594                 NF_CT_STAT_INC(net, searched);
 595         }
 596         rcu_read_unlock_bh();
 597
 598         return 0;
 599 }
 600 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 601
 602 #define NF_CT_EVICTION_RANGE    8
 603
 604 /* There's a small race here where we may free a just-assured
 605    connection.  Too bad: we're in trouble anyway. */
 606 static noinline int early_drop(struct net *net, unsigned int hash)
 607 {
 608         /* Use oldest entry, which is roughly LRU */
 609         struct nf_conntrack_tuple_hash *h;
 610         struct nf_conn *ct = NULL, *tmp;
 611         struct hlist_nulls_node *n;
 612         unsigned int i, cnt = 0;
 613         int dropped = 0;
 614
 615         rcu_read_lock();
 616         for (i = 0; i < net->ct.htable_size; i++) {
 617                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 618                                          hnnode) {
 619                         tmp = nf_ct_tuplehash_to_ctrack(h);
 620                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 621                                 ct = tmp;
 622                         cnt++;
 623                 }
 624
 625                 if (ct != NULL) {
 626                         if (likely(!nf_ct_is_dying(ct) &&
 627                                    atomic_inc_not_zero(&ct->ct_general.use)))
 628                                 break;
 629                         else
 630                                 ct = NULL;
 631                 }
 632
 633                 if (cnt >= NF_CT_EVICTION_RANGE)
 634                         break;
 635
 636                 hash = (hash + 1) % net->ct.htable_size;
 637         }
 638         rcu_read_unlock();
 639
 640         if (!ct)
 641                 return dropped;
 642
 643         if (del_timer(&ct->timeout)) {
 644                 death_by_timeout((unsigned long)ct);
 645                 /* Check if we indeed killed this entry. Reliable event
 646                    delivery may have inserted it into the dying list. */
 647                 if (test_bit(IPS_DYING_BIT, &ct->status)) {
 648                         dropped = 1;
 649                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 650                 }
 651         }
 652         nf_ct_put(ct);
 653         return dropped;
 654 }
 655
 656 void init_nf_conntrack_hash_rnd(void)
 657 {
 658         unsigned int rand;
 659
 660         /*
 661          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 662          * Because there isn't enough entropy when system initializing,
 663          * and we initialize it as late as possible.
 664          */
 665         do {
 666                 get_random_bytes(&rand, sizeof(rand));
 667         } while (!rand);
 668         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 669 }
 670
 671 static struct nf_conn *
 672 __nf_conntrack_alloc(struct net *net, u16 zone,
 673                      const struct nf_conntrack_tuple *orig,
 674                      const struct nf_conntrack_tuple *repl,
 675                      gfp_t gfp, u32 hash)
 676 {
 677         struct nf_conn *ct;
 678
 679         if (unlikely(!nf_conntrack_hash_rnd)) {
 680                 init_nf_conntrack_hash_rnd();
 681                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 682                 hash = hash_conntrack_raw(orig, zone);
 683         }
 684
 685         /* We don't want any race condition at early drop stage */
 686         atomic_inc(&net->ct.count);
 687
 688         if (nf_conntrack_max &&
 689             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 690                 if (!early_drop(net, hash_bucket(hash, net))) {
 691                         atomic_dec(&net->ct.count);
 692                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
 693                         return ERR_PTR(-ENOMEM);
 694                 }
 695         }
 696
 697         /*
 698          * Do not use kmem_cache_zalloc(), as this cache uses
 699          * SLAB_DESTROY_BY_RCU.
 700          */
 701         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 702         if (ct == NULL) {
 703                 atomic_dec(&net->ct.count);
 704                 return ERR_PTR(-ENOMEM);
 705         }
 706         /*
 707          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 708          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 709          */
 710         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 711                offsetof(struct nf_conn, proto) -
 712                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 713         spin_lock_init(&ct->lock);
 714         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 715         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 716         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 717         /* save hash for reusing when confirming */
 718         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 719         /* Don't set timer yet: wait for confirmation */
 720         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 721         write_pnet(&ct->ct_net, net);
 722 #ifdef CONFIG_NF_CONNTRACK_ZONES
 723         if (zone) {
 724                 struct nf_conntrack_zone *nf_ct_zone;
 725
 726                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 727                 if (!nf_ct_zone)
 728                         goto out_free;
 729                 nf_ct_zone->id = zone;
 730         }
 731 #endif
 732         /*
 733          * changes to lookup keys must be done before setting refcnt to 1
 734          */
 735         smp_wmb();
 736         atomic_set(&ct->ct_general.use, 1);
 737         return ct;
 738
 739 #ifdef CONFIG_NF_CONNTRACK_ZONES
 740 out_free:
 741         atomic_dec(&net->ct.count);
 742         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 743         return ERR_PTR(-ENOMEM);
 744 #endif
 745 }
 746
 747 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 748                                    const struct nf_conntrack_tuple *orig,
 749                                    const struct nf_conntrack_tuple *repl,
 750                                    gfp_t gfp)
 751 {
 752         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 753 }
 754 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 755
 756 void nf_conntrack_free(struct nf_conn *ct)
 757 {
 758         struct net *net = nf_ct_net(ct);
 759
 760         nf_ct_ext_destroy(ct);
 761         atomic_dec(&net->ct.count);
 762         nf_ct_ext_free(ct);
 763         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 764 }
 765 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 766
 767
 768 /* Allocate a new conntrack: we return -ENOMEM if classification
 769    failed due to stress.  Otherwise it really is unclassifiable. */
 770 static struct nf_conntrack_tuple_hash *
 771 init_conntrack(struct net *net, struct nf_conn *tmpl,
 772                const struct nf_conntrack_tuple *tuple,
 773                struct nf_conntrack_l3proto *l3proto,
 774                struct nf_conntrack_l4proto *l4proto,
 775                struct sk_buff *skb,
 776                unsigned int dataoff, u32 hash)
 777 {
 778         struct nf_conn *ct;
 779         struct nf_conn_help *help;
 780         struct nf_conntrack_tuple repl_tuple;
 781         struct nf_conntrack_ecache *ecache;
 782         struct nf_conntrack_expect *exp;
 783         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 784         struct nf_conn_timeout *timeout_ext;
 785         unsigned int *timeouts;
 786
 787         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 788                 pr_debug("Can't invert tuple.\n");
 789                 return NULL;
 790         }
 791
 792         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 793                                   hash);
 794         if (IS_ERR(ct))
 795                 return (struct nf_conntrack_tuple_hash *)ct;
 796
 797         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 798         if (timeout_ext)
 799                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 800         else
 801                 timeouts = l4proto->get_timeouts(net);
 802
 803         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 804                 nf_conntrack_free(ct);
 805                 pr_debug("init conntrack: can't track with proto module\n");
 806                 return NULL;
 807         }
 808
 809         if (timeout_ext)
 810                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 811
 812         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 813         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 814         nf_ct_labels_ext_add(ct);
 815
 816         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 817         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 818                                  ecache ? ecache->expmask : 0,
 819                              GFP_ATOMIC);
 820
 821         spin_lock_bh(&nf_conntrack_lock);
 822         exp = nf_ct_find_expectation(net, zone, tuple);
 823         if (exp) {
 824                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 825                          ct, exp);
 826                 /* Welcome, Mr. Bond.  We've been expecting you... */
 827                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 828                 ct->master = exp->master;
 829                 if (exp->helper) {
 830                         help = nf_ct_helper_ext_add(ct, exp->helper,
 831                                                     GFP_ATOMIC);
 832                         if (help)
 833                                 rcu_assign_pointer(help->helper, exp->helper);
 834                 }
 835
 836 #ifdef CONFIG_NF_CONNTRACK_MARK
 837                 ct->mark = exp->master->mark;
 838 #endif
 839 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 840                 ct->secmark = exp->master->secmark;
 841 #endif
 842                 nf_conntrack_get(&ct->master->ct_general);
 843                 NF_CT_STAT_INC(net, expect_new);
 844         } else {
 845                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 846                 NF_CT_STAT_INC(net, new);
 847         }
 848
 849         /* Overload tuple linked list to put us in unconfirmed list. */
 850         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 851                        &net->ct.unconfirmed);
 852
 853         spin_unlock_bh(&nf_conntrack_lock);
 854
 855         if (exp) {
 856                 if (exp->expectfn)
 857                         exp->expectfn(ct, exp);
 858                 nf_ct_expect_put(exp);
 859         }
 860
 861         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 862 }
 863
 864 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 865 static inline struct nf_conn *
 866 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 867                   struct sk_buff *skb,
 868                   unsigned int dataoff,
 869                   u_int16_t l3num,
 870                   u_int8_t protonum,
 871                   struct nf_conntrack_l3proto *l3proto,
 872                   struct nf_conntrack_l4proto *l4proto,
 873                   int *set_reply,
 874                   enum ip_conntrack_info *ctinfo)
 875 {
 876         struct nf_conntrack_tuple tuple;
 877         struct nf_conntrack_tuple_hash *h;
 878         struct nf_conn *ct;
 879         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 880         u32 hash;
 881
 882         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 883                              dataoff, l3num, protonum, &tuple, l3proto,
 884                              l4proto)) {
 885                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 886                 return NULL;
 887         }
 888
 889         /* look for tuple match */
 890         hash = hash_conntrack_raw(&tuple, zone);
 891         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 892         if (!h) {
 893                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 894                                    skb, dataoff, hash);
 895                 if (!h)
 896                         return NULL;
 897                 if (IS_ERR(h))
 898                         return (void *)h;
 899         }
 900         ct = nf_ct_tuplehash_to_ctrack(h);
 901
 902         /* It exists; we have (non-exclusive) reference. */
 903         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 904                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 905                 /* Please set reply bit if this packet OK */
 906                 *set_reply = 1;
 907         } else {
 908                 /* Once we've had two way comms, always ESTABLISHED. */
 909                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 910                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 911                         *ctinfo = IP_CT_ESTABLISHED;
 912                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 913                         pr_debug("nf_conntrack_in: related packet for %p\n",
 914                                  ct);
 915                         *ctinfo = IP_CT_RELATED;
 916                 } else {
 917                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 918                         *ctinfo = IP_CT_NEW;
 919                 }
 920                 *set_reply = 0;
 921         }
 922         skb->nfct = &ct->ct_general;
 923         skb->nfctinfo = *ctinfo;
 924         return ct;
 925 }
 926
 927 unsigned int
 928 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 929                 struct sk_buff *skb)
 930 {
 931         struct nf_conn *ct, *tmpl = NULL;
 932         enum ip_conntrack_info ctinfo;
 933         struct nf_conntrack_l3proto *l3proto;
 934         struct nf_conntrack_l4proto *l4proto;
 935         unsigned int *timeouts;
 936         unsigned int dataoff;
 937         u_int8_t protonum;
 938         int set_reply = 0;
 939         int ret;
 940
 941         if (skb->nfct) {
 942                 /* Previously seen (loopback or untracked)?  Ignore. */
 943                 tmpl = (struct nf_conn *)skb->nfct;
 944                 if (!nf_ct_is_template(tmpl)) {
 945                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 946                         return NF_ACCEPT;
 947                 }
 948                 skb->nfct = NULL;
 949         }
 950
 951         /* rcu_read_lock()ed by nf_hook_slow */
 952         l3proto = __nf_ct_l3proto_find(pf);
 953         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 954                                    &dataoff, &protonum);
 955         if (ret <= 0) {
 956                 pr_debug("not prepared to track yet or error occurred\n");
 957                 NF_CT_STAT_INC_ATOMIC(net, error);
 958                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 959                 ret = -ret;
 960                 goto out;
 961         }
 962
 963         l4proto = __nf_ct_l4proto_find(pf, protonum);
 964
 965         /* It may be an special packet, error, unclean...
 966          * inverse of the return code tells to the netfilter
 967          * core what to do with the packet. */
 968         if (l4proto->error != NULL) {
 969                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 970                                      pf, hooknum);
 971                 if (ret <= 0) {
 972                         NF_CT_STAT_INC_ATOMIC(net, error);
 973                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 974                         ret = -ret;
 975                         goto out;
 976                 }
 977                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 978                 if (skb->nfct)
 979                         goto out;
 980         }
 981
 982         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 983                                l3proto, l4proto, &set_reply, &ctinfo);
 984         if (!ct) {
 985                 /* Not valid part of a connection */
 986                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 987                 ret = NF_ACCEPT;
 988                 goto out;
 989         }
 990
 991         if (IS_ERR(ct)) {
 992                 /* Too stressed to deal. */
 993                 NF_CT_STAT_INC_ATOMIC(net, drop);
 994                 ret = NF_DROP;
 995                 goto out;
 996         }
 997
 998         NF_CT_ASSERT(skb->nfct);
 999
1000         /* Decide what timeout policy we want to apply to this flow. */
1001         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1002
1003         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1004         if (ret <= 0) {
1005                 /* Invalid: inverse of the return code tells
1006                  * the netfilter core what to do */
1007                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1008                 nf_conntrack_put(skb->nfct);
1009                 skb->nfct = NULL;
1010                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1011                 if (ret == -NF_DROP)
1012                         NF_CT_STAT_INC_ATOMIC(net, drop);
1013                 ret = -ret;
1014                 goto out;
1015         }
1016
1017         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1018                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1019 out:
1020         if (tmpl) {
1021                 /* Special case: we have to repeat this hook, assign the
1022                  * template again to this packet. We assume that this packet
1023                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1024                 if (ret == NF_REPEAT)
1025                         skb->nfct = (struct nf_conntrack *)tmpl;
1026                 else
1027                         nf_ct_put(tmpl);
1028         }
1029
1030         return ret;
1031 }
1032 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1033
1034 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1035                           const struct nf_conntrack_tuple *orig)
1036 {
1037         bool ret;
1038
1039         rcu_read_lock();
1040         ret = nf_ct_invert_tuple(inverse, orig,
1041                                  __nf_ct_l3proto_find(orig->src.l3num),
1042                                  __nf_ct_l4proto_find(orig->src.l3num,
1043                                                       orig->dst.protonum));
1044         rcu_read_unlock();
1045         return ret;
1046 }
1047 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1048
1049 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1050    implicitly racy: see __nf_conntrack_confirm */
1051 void nf_conntrack_alter_reply(struct nf_conn *ct,
1052                               const struct nf_conntrack_tuple *newreply)
1053 {
1054         struct nf_conn_help *help = nfct_help(ct);
1055
1056         /* Should be unconfirmed, so not in hash table yet */
1057         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1058
1059         pr_debug("Altering reply tuple of %p to ", ct);
1060         nf_ct_dump_tuple(newreply);
1061
1062         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1063         if (ct->master || (help && !hlist_empty(&help->expectations)))
1064                 return;
1065
1066         rcu_read_lock();
1067         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1068         rcu_read_unlock();
1069 }
1070 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1071
1072 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1073 void __nf_ct_refresh_acct(struct nf_conn *ct,
1074                           enum ip_conntrack_info ctinfo,
1075                           const struct sk_buff *skb,
1076                           unsigned long extra_jiffies,
1077                           int do_acct)
1078 {
1079         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1080         NF_CT_ASSERT(skb);
1081
1082         /* Only update if this is not a fixed timeout */
1083         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1084                 goto acct;
1085
1086         /* If not in hash table, timer will not be active yet */
1087         if (!nf_ct_is_confirmed(ct)) {
1088                 ct->timeout.expires = extra_jiffies;
1089         } else {
1090                 unsigned long newtime = jiffies + extra_jiffies;
1091
1092                 /* Only update the timeout if the new timeout is at least
1093                    HZ jiffies from the old timeout. Need del_timer for race
1094                    avoidance (may already be dying). */
1095                 if (newtime - ct->timeout.expires >= HZ)
1096                         mod_timer_pending(&ct->timeout, newtime);
1097         }
1098
1099 acct:
1100         if (do_acct) {
1101                 struct nf_conn_counter *acct;
1102
1103                 acct = nf_conn_acct_find(ct);
1104                 if (acct) {
1105                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1106                         atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes);
1107                 }
1108         }
1109 }
1110 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1111
1112 bool __nf_ct_kill_acct(struct nf_conn *ct,
1113                        enum ip_conntrack_info ctinfo,
1114                        const struct sk_buff *skb,
1115                        int do_acct)
1116 {
1117         if (do_acct) {
1118                 struct nf_conn_counter *acct;
1119
1120                 acct = nf_conn_acct_find(ct);
1121                 if (acct) {
1122                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1123                         atomic64_add(skb->len - skb_network_offset(skb),
1124                                      &acct[CTINFO2DIR(ctinfo)].bytes);
1125                 }
1126         }
1127
1128         if (del_timer(&ct->timeout)) {
1129                 ct->timeout.function((unsigned long)ct);
1130                 return true;
1131         }
1132         return false;
1133 }
1134 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1135
1136 #ifdef CONFIG_NF_CONNTRACK_ZONES
1137 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1138         .len    = sizeof(struct nf_conntrack_zone),
1139         .align  = __alignof__(struct nf_conntrack_zone),
1140         .id     = NF_CT_EXT_ZONE,
1141 };
1142 #endif
1143
1144 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1145
1146 #include <linux/netfilter/nfnetlink.h>
1147 #include <linux/netfilter/nfnetlink_conntrack.h>
1148 #include <linux/mutex.h>
1149
1150 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1151  * in ip_conntrack_core, since we don't want the protocols to autoload
1152  * or depend on ctnetlink */
1153 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1154                                const struct nf_conntrack_tuple *tuple)
1155 {
1156         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1157             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1158                 goto nla_put_failure;
1159         return 0;
1160
1161 nla_put_failure:
1162         return -1;
1163 }
1164 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1165
1166 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1167         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1168         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1169 };
1170 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1171
1172 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1173                                struct nf_conntrack_tuple *t)
1174 {
1175         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1176                 return -EINVAL;
1177
1178         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1179         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1180
1181         return 0;
1182 }
1183 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1184
1185 int nf_ct_port_nlattr_tuple_size(void)
1186 {
1187         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1188 }
1189 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1190 #endif
1191
1192 /* Used by ipt_REJECT and ip6t_REJECT. */
1193 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1194 {
1195         struct nf_conn *ct;
1196         enum ip_conntrack_info ctinfo;
1197
1198         /* This ICMP is in reverse direction to the packet which caused it */
1199         ct = nf_ct_get(skb, &ctinfo);
1200         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1201                 ctinfo = IP_CT_RELATED_REPLY;
1202         else
1203                 ctinfo = IP_CT_RELATED;
1204
1205         /* Attach to new skbuff, and increment count */
1206         nskb->nfct = &ct->ct_general;
1207         nskb->nfctinfo = ctinfo;
1208         nf_conntrack_get(nskb->nfct);
1209 }
1210
1211 /* Bring out ya dead! */
1212 static struct nf_conn *
1213 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1214                 void *data, unsigned int *bucket)
1215 {
1216         struct nf_conntrack_tuple_hash *h;
1217         struct nf_conn *ct;
1218         struct hlist_nulls_node *n;
1219
1220         spin_lock_bh(&nf_conntrack_lock);
1221         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1222                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1223                         if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1224                                 continue;
1225                         ct = nf_ct_tuplehash_to_ctrack(h);
1226                         if (iter(ct, data))
1227                                 goto found;
1228                 }
1229         }
1230         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1231                 ct = nf_ct_tuplehash_to_ctrack(h);
1232                 if (iter(ct, data))
1233                         set_bit(IPS_DYING_BIT, &ct->status);
1234         }
1235         spin_unlock_bh(&nf_conntrack_lock);
1236         return NULL;
1237 found:
1238         atomic_inc(&ct->ct_general.use);
1239         spin_unlock_bh(&nf_conntrack_lock);
1240         return ct;
1241 }
1242
1243 void nf_ct_iterate_cleanup(struct net *net,
1244                            int (*iter)(struct nf_conn *i, void *data),
1245                            void *data)
1246 {
1247         struct nf_conn *ct;
1248         unsigned int bucket = 0;
1249
1250         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1251                 /* Time to push up daises... */
1252                 if (del_timer(&ct->timeout))
1253                         death_by_timeout((unsigned long)ct);
1254                 /* ... else the timer will get him soon. */
1255
1256                 nf_ct_put(ct);
1257         }
1258 }
1259 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1260
1261 struct __nf_ct_flush_report {
1262         u32 pid;
1263         int report;
1264 };
1265
1266 static int kill_report(struct nf_conn *i, void *data)
1267 {
1268         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1269         struct nf_conn_tstamp *tstamp;
1270
1271         tstamp = nf_conn_tstamp_find(i);
1272         if (tstamp && tstamp->stop == 0)
1273                 tstamp->stop = ktime_to_ns(ktime_get_real());
1274
1275         /* If we fail to deliver the event, death_by_timeout() will retry */
1276         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1277                                       fr->pid, fr->report) < 0)
1278                 return 1;
1279
1280         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1281         set_bit(IPS_DYING_BIT, &i->status);
1282         return 1;
1283 }
1284
1285 static int kill_all(struct nf_conn *i, void *data)
1286 {
1287         return 1;
1288 }
1289
1290 void nf_ct_free_hashtable(void *hash, unsigned int size)
1291 {
1292         if (is_vmalloc_addr(hash))
1293                 vfree(hash);
1294         else
1295                 free_pages((unsigned long)hash,
1296                            get_order(sizeof(struct hlist_head) * size));
1297 }
1298 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1299
1300 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1301 {
1302         struct __nf_ct_flush_report fr = {
1303                 .pid    = pid,
1304                 .report = report,
1305         };
1306         nf_ct_iterate_cleanup(net, kill_report, &fr);
1307 }
1308 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1309
1310 static void nf_ct_release_dying_list(struct net *net)
1311 {
1312         struct nf_conntrack_tuple_hash *h;
1313         struct nf_conn *ct;
1314         struct hlist_nulls_node *n;
1315
1316         spin_lock_bh(&nf_conntrack_lock);
1317         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1318                 ct = nf_ct_tuplehash_to_ctrack(h);
1319                 /* never fails to remove them, no listeners at this point */
1320                 nf_ct_kill(ct);
1321         }
1322         spin_unlock_bh(&nf_conntrack_lock);
1323 }
1324
1325 static int untrack_refs(void)
1326 {
1327         int cnt = 0, cpu;
1328
1329         for_each_possible_cpu(cpu) {
1330                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1331
1332                 cnt += atomic_read(&ct->ct_general.use) - 1;
1333         }
1334         return cnt;
1335 }
1336
1337 void nf_conntrack_cleanup_start(void)
1338 {
1339         RCU_INIT_POINTER(ip_ct_attach, NULL);
1340 }
1341
1342 void nf_conntrack_cleanup_end(void)
1343 {
1344         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1345         while (untrack_refs() > 0)
1346                 schedule();
1347
1348 #ifdef CONFIG_NF_CONNTRACK_ZONES
1349         nf_ct_extend_unregister(&nf_ct_zone_extend);
1350 #endif
1351         nf_conntrack_proto_fini();
1352         nf_conntrack_labels_fini();
1353         nf_conntrack_helper_fini();
1354         nf_conntrack_timeout_fini();
1355         nf_conntrack_ecache_fini();
1356         nf_conntrack_tstamp_fini();
1357         nf_conntrack_acct_fini();
1358         nf_conntrack_expect_fini();
1359 }
1360
1361 /*
1362  * Mishearing the voices in his head, our hero wonders how he's
1363  * supposed to kill the mall.
1364  */
1365 void nf_conntrack_cleanup_net(struct net *net)
1366 {
1367         /*
1368          * This makes sure all current packets have passed through
1369          *  netfilter framework.  Roll on, two-stage module
1370          *  delete...
1371          */
1372         synchronize_net();
1373  i_see_dead_people:
1374         nf_ct_iterate_cleanup(net, kill_all, NULL);
1375         nf_ct_release_dying_list(net);
1376         if (atomic_read(&net->ct.count) != 0) {
1377                 schedule();
1378                 goto i_see_dead_people;
1379         }
1380
1381         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1382         nf_conntrack_proto_pernet_fini(net);
1383         nf_conntrack_helper_pernet_fini(net);
1384         nf_conntrack_ecache_pernet_fini(net);
1385         nf_conntrack_tstamp_pernet_fini(net);
1386         nf_conntrack_acct_pernet_fini(net);
1387         nf_conntrack_expect_pernet_fini(net);
1388         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1389         kfree(net->ct.slabname);
1390         free_percpu(net->ct.stat);
1391 }
1392
1393 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1394 {
1395         struct hlist_nulls_head *hash;
1396         unsigned int nr_slots, i;
1397         size_t sz;
1398
1399         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1400         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1401         sz = nr_slots * sizeof(struct hlist_nulls_head);
1402         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1403                                         get_order(sz));
1404         if (!hash) {
1405                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1406                 hash = vzalloc(sz);
1407         }
1408
1409         if (hash && nulls)
1410                 for (i = 0; i < nr_slots; i++)
1411                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1412
1413         return hash;
1414 }
1415 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1416
1417 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1418 {
1419         int i, bucket, rc;
1420         unsigned int hashsize, old_size;
1421         struct hlist_nulls_head *hash, *old_hash;
1422         struct nf_conntrack_tuple_hash *h;
1423         struct nf_conn *ct;
1424
1425         if (current->nsproxy->net_ns != &init_net)
1426                 return -EOPNOTSUPP;
1427
1428         /* On boot, we can set this without any fancy locking. */
1429         if (!nf_conntrack_htable_size)
1430                 return param_set_uint(val, kp);
1431
1432         rc = kstrtouint(val, 0, &hashsize);
1433         if (rc)
1434                 return rc;
1435         if (!hashsize)
1436                 return -EINVAL;
1437
1438         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1439         if (!hash)
1440                 return -ENOMEM;
1441
1442         /* Lookups in the old hash might happen in parallel, which means we
1443          * might get false negatives during connection lookup. New connections
1444          * created because of a false negative won't make it into the hash
1445          * though since that required taking the lock.
1446          */
1447         spin_lock_bh(&nf_conntrack_lock);
1448         for (i = 0; i < init_net.ct.htable_size; i++) {
1449                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1450                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1451                                         struct nf_conntrack_tuple_hash, hnnode);
1452                         ct = nf_ct_tuplehash_to_ctrack(h);
1453                         hlist_nulls_del_rcu(&h->hnnode);
1454                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1455                                                   hashsize);
1456                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1457                 }
1458         }
1459         old_size = init_net.ct.htable_size;
1460         old_hash = init_net.ct.hash;
1461
1462         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1463         init_net.ct.hash = hash;
1464         spin_unlock_bh(&nf_conntrack_lock);
1465
1466         nf_ct_free_hashtable(old_hash, old_size);
1467         return 0;
1468 }
1469 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1470
1471 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1472                   &nf_conntrack_htable_size, 0600);
1473
1474 void nf_ct_untracked_status_or(unsigned long bits)
1475 {
1476         int cpu;
1477
1478         for_each_possible_cpu(cpu)
1479                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1480 }
1481 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1482
1483 int nf_conntrack_init_start(void)
1484 {
1485         int max_factor = 8;
1486         int ret, cpu;
1487
1488         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1489          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1490         if (!nf_conntrack_htable_size) {
1491                 nf_conntrack_htable_size
1492                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1493                            / sizeof(struct hlist_head));
1494                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1495                         nf_conntrack_htable_size = 16384;
1496                 if (nf_conntrack_htable_size < 32)
1497                         nf_conntrack_htable_size = 32;
1498
1499                 /* Use a max. factor of four by default to get the same max as
1500                  * with the old struct list_heads. When a table size is given
1501                  * we use the old value of 8 to avoid reducing the max.
1502                  * entries. */
1503                 max_factor = 4;
1504         }
1505         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1506
1507         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1508                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1509                nf_conntrack_max);
1510
1511         ret = nf_conntrack_expect_init();
1512         if (ret < 0)
1513                 goto err_expect;
1514
1515         ret = nf_conntrack_acct_init();
1516         if (ret < 0)
1517                 goto err_acct;
1518
1519         ret = nf_conntrack_tstamp_init();
1520         if (ret < 0)
1521                 goto err_tstamp;
1522
1523         ret = nf_conntrack_ecache_init();
1524         if (ret < 0)
1525                 goto err_ecache;
1526
1527         ret = nf_conntrack_timeout_init();
1528         if (ret < 0)
1529                 goto err_timeout;
1530
1531         ret = nf_conntrack_helper_init();
1532         if (ret < 0)
1533                 goto err_helper;
1534
1535         ret = nf_conntrack_labels_init();
1536         if (ret < 0)
1537                 goto err_labels;
1538
1539 #ifdef CONFIG_NF_CONNTRACK_ZONES
1540         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1541         if (ret < 0)
1542                 goto err_extend;
1543 #endif
1544         ret = nf_conntrack_proto_init();
1545         if (ret < 0)
1546                 goto err_proto;
1547
1548         /* Set up fake conntrack: to never be deleted, not in any hashes */
1549         for_each_possible_cpu(cpu) {
1550                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1551                 write_pnet(&ct->ct_net, &init_net);
1552                 atomic_set(&ct->ct_general.use, 1);
1553         }
1554         /*  - and look it like as a confirmed connection */
1555         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1556         return 0;
1557
1558 err_proto:
1559 #ifdef CONFIG_NF_CONNTRACK_ZONES
1560         nf_ct_extend_unregister(&nf_ct_zone_extend);
1561 err_extend:
1562 #endif
1563         nf_conntrack_labels_fini();
1564 err_labels:
1565         nf_conntrack_helper_fini();
1566 err_helper:
1567         nf_conntrack_timeout_fini();
1568 err_timeout:
1569         nf_conntrack_ecache_fini();
1570 err_ecache:
1571         nf_conntrack_tstamp_fini();
1572 err_tstamp:
1573         nf_conntrack_acct_fini();
1574 err_acct:
1575         nf_conntrack_expect_fini();
1576 err_expect:
1577         return ret;
1578 }
1579
1580 void nf_conntrack_init_end(void)
1581 {
1582         /* For use by REJECT target */
1583         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1584         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1585
1586         /* Howto get NAT offsets */
1587         RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
1588 }
1589
1590 /*
1591  * We need to use special "null" values, not used in hash table
1592  */
1593 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1594 #define DYING_NULLS_VAL         ((1<<30)+1)
1595 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1596
1597 int nf_conntrack_init_net(struct net *net)
1598 {
1599         int ret;
1600
1601         atomic_set(&net->ct.count, 0);
1602         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1603         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1604         INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1605         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1606         if (!net->ct.stat) {
1607                 ret = -ENOMEM;
1608                 goto err_stat;
1609         }
1610
1611         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1612         if (!net->ct.slabname) {
1613                 ret = -ENOMEM;
1614                 goto err_slabname;
1615         }
1616
1617         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1618                                                         sizeof(struct nf_conn), 0,
1619                                                         SLAB_DESTROY_BY_RCU, NULL);
1620         if (!net->ct.nf_conntrack_cachep) {
1621                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1622                 ret = -ENOMEM;
1623                 goto err_cache;
1624         }
1625
1626         net->ct.htable_size = nf_conntrack_htable_size;
1627         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1628         if (!net->ct.hash) {
1629                 ret = -ENOMEM;
1630                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1631                 goto err_hash;
1632         }
1633         ret = nf_conntrack_expect_pernet_init(net);
1634         if (ret < 0)
1635                 goto err_expect;
1636         ret = nf_conntrack_acct_pernet_init(net);
1637         if (ret < 0)
1638                 goto err_acct;
1639         ret = nf_conntrack_tstamp_pernet_init(net);
1640         if (ret < 0)
1641                 goto err_tstamp;
1642         ret = nf_conntrack_ecache_pernet_init(net);
1643         if (ret < 0)
1644                 goto err_ecache;
1645         ret = nf_conntrack_helper_pernet_init(net);
1646         if (ret < 0)
1647                 goto err_helper;
1648         ret = nf_conntrack_proto_pernet_init(net);
1649         if (ret < 0)
1650                 goto err_proto;
1651         return 0;
1652
1653 err_proto:
1654         nf_conntrack_helper_pernet_fini(net);
1655 err_helper:
1656         nf_conntrack_ecache_pernet_fini(net);
1657 err_ecache:
1658         nf_conntrack_tstamp_pernet_fini(net);
1659 err_tstamp:
1660         nf_conntrack_acct_pernet_fini(net);
1661 err_acct:
1662         nf_conntrack_expect_pernet_fini(net);
1663 err_expect:
1664         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1665 err_hash:
1666         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1667 err_cache:
1668         kfree(net->ct.slabname);
1669 err_slabname:
1670         free_percpu(net->ct.stat);
1671 err_stat:
1672         return ret;
1673 }
1674
1675 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1676                         enum ip_conntrack_dir dir,
1677                         u32 seq);
1678 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);