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_nat.h>
  49 #include <net/netfilter/nf_nat_core.h>
  50
  51 #define NF_CONNTRACK_VERSION    "0.5.0"
  52
  53 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  54                                       enum nf_nat_manip_type manip,
  55                                       const struct nlattr *attr) __read_mostly;
  56 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  57
  58 DEFINE_SPINLOCK(nf_conntrack_lock);
  59 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  60
  61 unsigned int nf_conntrack_htable_size __read_mostly;
  62 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  63
  64 unsigned int nf_conntrack_max __read_mostly;
  65 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  66
  67 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  68 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  69
  70 unsigned int nf_conntrack_hash_rnd __read_mostly;
  71 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
  72
  73 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  74 {
  75         unsigned int n;
  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         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  83                       (((__force __u16)tuple->dst.u.all << 16) |
  84                       tuple->dst.protonum));
  85 }
  86
  87 static u32 __hash_bucket(u32 hash, unsigned int size)
  88 {
  89         return ((u64)hash * size) >> 32;
  90 }
  91
  92 static u32 hash_bucket(u32 hash, const struct net *net)
  93 {
  94         return __hash_bucket(hash, net->ct.htable_size);
  95 }
  96
  97 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  98                                   u16 zone, unsigned int size)
  99 {
 100         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 101 }
 102
 103 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 104                                        const struct nf_conntrack_tuple *tuple)
 105 {
 106         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 107 }
 108
 109 bool
 110 nf_ct_get_tuple(const struct sk_buff *skb,
 111                 unsigned int nhoff,
 112                 unsigned int dataoff,
 113                 u_int16_t l3num,
 114                 u_int8_t protonum,
 115                 struct nf_conntrack_tuple *tuple,
 116                 const struct nf_conntrack_l3proto *l3proto,
 117                 const struct nf_conntrack_l4proto *l4proto)
 118 {
 119         memset(tuple, 0, sizeof(*tuple));
 120
 121         tuple->src.l3num = l3num;
 122         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 123                 return false;
 124
 125         tuple->dst.protonum = protonum;
 126         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 127
 128         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 129 }
 130 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 131
 132 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 133                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 134 {
 135         struct nf_conntrack_l3proto *l3proto;
 136         struct nf_conntrack_l4proto *l4proto;
 137         unsigned int protoff;
 138         u_int8_t protonum;
 139         int ret;
 140
 141         rcu_read_lock();
 142
 143         l3proto = __nf_ct_l3proto_find(l3num);
 144         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 145         if (ret != NF_ACCEPT) {
 146                 rcu_read_unlock();
 147                 return false;
 148         }
 149
 150         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 151
 152         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 153                               l3proto, l4proto);
 154
 155         rcu_read_unlock();
 156         return ret;
 157 }
 158 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 159
 160 bool
 161 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 162                    const struct nf_conntrack_tuple *orig,
 163                    const struct nf_conntrack_l3proto *l3proto,
 164                    const struct nf_conntrack_l4proto *l4proto)
 165 {
 166         memset(inverse, 0, sizeof(*inverse));
 167
 168         inverse->src.l3num = orig->src.l3num;
 169         if (l3proto->invert_tuple(inverse, orig) == 0)
 170                 return false;
 171
 172         inverse->dst.dir = !orig->dst.dir;
 173
 174         inverse->dst.protonum = orig->dst.protonum;
 175         return l4proto->invert_tuple(inverse, orig);
 176 }
 177 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 178
 179 static void
 180 clean_from_lists(struct nf_conn *ct)
 181 {
 182         pr_debug("clean_from_lists(%p)\n", ct);
 183         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 184         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 185
 186         /* Destroy all pending expectations */
 187         nf_ct_remove_expectations(ct);
 188 }
 189
 190 static void
 191 destroy_conntrack(struct nf_conntrack *nfct)
 192 {
 193         struct nf_conn *ct = (struct nf_conn *)nfct;
 194         struct net *net = nf_ct_net(ct);
 195         struct nf_conntrack_l4proto *l4proto;
 196
 197         pr_debug("destroy_conntrack(%p)\n", ct);
 198         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 199         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 200
 201         /* To make sure we don't get any weird locking issues here:
 202          * destroy_conntrack() MUST NOT be called with a write lock
 203          * to nf_conntrack_lock!!! -HW */
 204         rcu_read_lock();
 205         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 206         if (l4proto && l4proto->destroy)
 207                 l4proto->destroy(ct);
 208
 209         rcu_read_unlock();
 210
 211         spin_lock_bh(&nf_conntrack_lock);
 212         /* Expectations will have been removed in clean_from_lists,
 213          * except TFTP can create an expectation on the first packet,
 214          * before connection is in the list, so we need to clean here,
 215          * too. */
 216         nf_ct_remove_expectations(ct);
 217
 218         /* We overload first tuple to link into unconfirmed list. */
 219         if (!nf_ct_is_confirmed(ct)) {
 220                 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 221                 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 222         }
 223
 224         NF_CT_STAT_INC(net, delete);
 225         spin_unlock_bh(&nf_conntrack_lock);
 226
 227         if (ct->master)
 228                 nf_ct_put(ct->master);
 229
 230         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 231         nf_conntrack_free(ct);
 232 }
 233
 234 void nf_ct_delete_from_lists(struct nf_conn *ct)
 235 {
 236         struct net *net = nf_ct_net(ct);
 237
 238         nf_ct_helper_destroy(ct);
 239         spin_lock_bh(&nf_conntrack_lock);
 240         /* Inside lock so preempt is disabled on module removal path.
 241          * Otherwise we can get spurious warnings. */
 242         NF_CT_STAT_INC(net, delete_list);
 243         clean_from_lists(ct);
 244         spin_unlock_bh(&nf_conntrack_lock);
 245 }
 246 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
 247
 248 static void death_by_event(unsigned long ul_conntrack)
 249 {
 250         struct nf_conn *ct = (void *)ul_conntrack;
 251         struct net *net = nf_ct_net(ct);
 252
 253         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 254                 /* bad luck, let's retry again */
 255                 ct->timeout.expires = jiffies +
 256                         (random32() % net->ct.sysctl_events_retry_timeout);
 257                 add_timer(&ct->timeout);
 258                 return;
 259         }
 260         /* we've got the event delivered, now it's dying */
 261         set_bit(IPS_DYING_BIT, &ct->status);
 262         spin_lock(&nf_conntrack_lock);
 263         hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 264         spin_unlock(&nf_conntrack_lock);
 265         nf_ct_put(ct);
 266 }
 267
 268 void nf_ct_insert_dying_list(struct nf_conn *ct)
 269 {
 270         struct net *net = nf_ct_net(ct);
 271
 272         /* add this conntrack to the dying list */
 273         spin_lock_bh(&nf_conntrack_lock);
 274         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 275                              &net->ct.dying);
 276         spin_unlock_bh(&nf_conntrack_lock);
 277         /* set a new timer to retry event delivery */
 278         setup_timer(&ct->timeout, death_by_event, (unsigned long)ct);
 279         ct->timeout.expires = jiffies +
 280                 (random32() % net->ct.sysctl_events_retry_timeout);
 281         add_timer(&ct->timeout);
 282 }
 283 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list);
 284
 285 static void death_by_timeout(unsigned long ul_conntrack)
 286 {
 287         struct nf_conn *ct = (void *)ul_conntrack;
 288         struct nf_conn_tstamp *tstamp;
 289
 290         tstamp = nf_conn_tstamp_find(ct);
 291         if (tstamp && tstamp->stop == 0)
 292                 tstamp->stop = ktime_to_ns(ktime_get_real());
 293
 294         if (!test_bit(IPS_DYING_BIT, &ct->status) &&
 295             unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
 296                 /* destroy event was not delivered */
 297                 nf_ct_delete_from_lists(ct);
 298                 nf_ct_insert_dying_list(ct);
 299                 return;
 300         }
 301         set_bit(IPS_DYING_BIT, &ct->status);
 302         nf_ct_delete_from_lists(ct);
 303         nf_ct_put(ct);
 304 }
 305
 306 /*
 307  * Warning :
 308  * - Caller must take a reference on returned object
 309  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 310  * OR
 311  * - Caller must lock nf_conntrack_lock before calling this function
 312  */
 313 static struct nf_conntrack_tuple_hash *
 314 ____nf_conntrack_find(struct net *net, u16 zone,
 315                       const struct nf_conntrack_tuple *tuple, u32 hash)
 316 {
 317         struct nf_conntrack_tuple_hash *h;
 318         struct hlist_nulls_node *n;
 319         unsigned int bucket = hash_bucket(hash, net);
 320
 321         /* Disable BHs the entire time since we normally need to disable them
 322          * at least once for the stats anyway.
 323          */
 324         local_bh_disable();
 325 begin:
 326         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 327                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 328                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 329                         NF_CT_STAT_INC(net, found);
 330                         local_bh_enable();
 331                         return h;
 332                 }
 333                 NF_CT_STAT_INC(net, searched);
 334         }
 335         /*
 336          * if the nulls value we got at the end of this lookup is
 337          * not the expected one, we must restart lookup.
 338          * We probably met an item that was moved to another chain.
 339          */
 340         if (get_nulls_value(n) != bucket) {
 341                 NF_CT_STAT_INC(net, search_restart);
 342                 goto begin;
 343         }
 344         local_bh_enable();
 345
 346         return NULL;
 347 }
 348
 349 struct nf_conntrack_tuple_hash *
 350 __nf_conntrack_find(struct net *net, u16 zone,
 351                     const struct nf_conntrack_tuple *tuple)
 352 {
 353         return ____nf_conntrack_find(net, zone, tuple,
 354                                      hash_conntrack_raw(tuple, zone));
 355 }
 356 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 357
 358 /* Find a connection corresponding to a tuple. */
 359 static struct nf_conntrack_tuple_hash *
 360 __nf_conntrack_find_get(struct net *net, u16 zone,
 361                         const struct nf_conntrack_tuple *tuple, u32 hash)
 362 {
 363         struct nf_conntrack_tuple_hash *h;
 364         struct nf_conn *ct;
 365
 366         rcu_read_lock();
 367 begin:
 368         h = ____nf_conntrack_find(net, zone, tuple, hash);
 369         if (h) {
 370                 ct = nf_ct_tuplehash_to_ctrack(h);
 371                 if (unlikely(nf_ct_is_dying(ct) ||
 372                              !atomic_inc_not_zero(&ct->ct_general.use)))
 373                         h = NULL;
 374                 else {
 375                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 376                                      nf_ct_zone(ct) != zone)) {
 377                                 nf_ct_put(ct);
 378                                 goto begin;
 379                         }
 380                 }
 381         }
 382         rcu_read_unlock();
 383
 384         return h;
 385 }
 386
 387 struct nf_conntrack_tuple_hash *
 388 nf_conntrack_find_get(struct net *net, u16 zone,
 389                       const struct nf_conntrack_tuple *tuple)
 390 {
 391         return __nf_conntrack_find_get(net, zone, tuple,
 392                                        hash_conntrack_raw(tuple, zone));
 393 }
 394 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 395
 396 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 397                                        unsigned int hash,
 398                                        unsigned int repl_hash)
 399 {
 400         struct net *net = nf_ct_net(ct);
 401
 402         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 403                            &net->ct.hash[hash]);
 404         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 405                            &net->ct.hash[repl_hash]);
 406 }
 407
 408 int
 409 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 410 {
 411         struct net *net = nf_ct_net(ct);
 412         unsigned int hash, repl_hash;
 413         struct nf_conntrack_tuple_hash *h;
 414         struct hlist_nulls_node *n;
 415         u16 zone;
 416
 417         zone = nf_ct_zone(ct);
 418         hash = hash_conntrack(net, zone,
 419                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 420         repl_hash = hash_conntrack(net, zone,
 421                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 422
 423         spin_lock_bh(&nf_conntrack_lock);
 424
 425         /* See if there's one in the list already, including reverse */
 426         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 427                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 428                                       &h->tuple) &&
 429                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 430                         goto out;
 431         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 432                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 433                                       &h->tuple) &&
 434                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 435                         goto out;
 436
 437         add_timer(&ct->timeout);
 438         nf_conntrack_get(&ct->ct_general);
 439         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 440         NF_CT_STAT_INC(net, insert);
 441         spin_unlock_bh(&nf_conntrack_lock);
 442
 443         return 0;
 444
 445 out:
 446         NF_CT_STAT_INC(net, insert_failed);
 447         spin_unlock_bh(&nf_conntrack_lock);
 448         return -EEXIST;
 449 }
 450 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 451
 452 /* Confirm a connection given skb; places it in hash table */
 453 int
 454 __nf_conntrack_confirm(struct sk_buff *skb)
 455 {
 456         unsigned int hash, repl_hash;
 457         struct nf_conntrack_tuple_hash *h;
 458         struct nf_conn *ct;
 459         struct nf_conn_help *help;
 460         struct nf_conn_tstamp *tstamp;
 461         struct hlist_nulls_node *n;
 462         enum ip_conntrack_info ctinfo;
 463         struct net *net;
 464         u16 zone;
 465
 466         ct = nf_ct_get(skb, &ctinfo);
 467         net = nf_ct_net(ct);
 468
 469         /* ipt_REJECT uses nf_conntrack_attach to attach related
 470            ICMP/TCP RST packets in other direction.  Actual packet
 471            which created connection will be IP_CT_NEW or for an
 472            expected connection, IP_CT_RELATED. */
 473         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 474                 return NF_ACCEPT;
 475
 476         zone = nf_ct_zone(ct);
 477         /* reuse the hash saved before */
 478         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 479         hash = hash_bucket(hash, net);
 480         repl_hash = hash_conntrack(net, zone,
 481                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 482
 483         /* We're not in hash table, and we refuse to set up related
 484            connections for unconfirmed conns.  But packet copies and
 485            REJECT will give spurious warnings here. */
 486         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 487
 488         /* No external references means no one else could have
 489            confirmed us. */
 490         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 491         pr_debug("Confirming conntrack %p\n", ct);
 492
 493         spin_lock_bh(&nf_conntrack_lock);
 494
 495         /* We have to check the DYING flag inside the lock to prevent
 496            a race against nf_ct_get_next_corpse() possibly called from
 497            user context, else we insert an already 'dead' hash, blocking
 498            further use of that particular connection -JM */
 499
 500         if (unlikely(nf_ct_is_dying(ct))) {
 501                 spin_unlock_bh(&nf_conntrack_lock);
 502                 return NF_ACCEPT;
 503         }
 504
 505         /* See if there's one in the list already, including reverse:
 506            NAT could have grabbed it without realizing, since we're
 507            not in the hash.  If there is, we lost race. */
 508         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 509                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 510                                       &h->tuple) &&
 511                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 512                         goto out;
 513         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 514                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 515                                       &h->tuple) &&
 516                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 517                         goto out;
 518
 519         /* Remove from unconfirmed list */
 520         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 521
 522         /* Timer relative to confirmation time, not original
 523            setting time, otherwise we'd get timer wrap in
 524            weird delay cases. */
 525         ct->timeout.expires += jiffies;
 526         add_timer(&ct->timeout);
 527         atomic_inc(&ct->ct_general.use);
 528         ct->status |= IPS_CONFIRMED;
 529
 530         /* set conntrack timestamp, if enabled. */
 531         tstamp = nf_conn_tstamp_find(ct);
 532         if (tstamp) {
 533                 if (skb->tstamp.tv64 == 0)
 534                         __net_timestamp((struct sk_buff *)skb);
 535
 536                 tstamp->start = ktime_to_ns(skb->tstamp);
 537         }
 538         /* Since the lookup is lockless, hash insertion must be done after
 539          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 540          * guarantee that no other CPU can find the conntrack before the above
 541          * stores are visible.
 542          */
 543         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 544         NF_CT_STAT_INC(net, insert);
 545         spin_unlock_bh(&nf_conntrack_lock);
 546
 547         help = nfct_help(ct);
 548         if (help && help->helper)
 549                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 550
 551         nf_conntrack_event_cache(master_ct(ct) ?
 552                                  IPCT_RELATED : IPCT_NEW, ct);
 553         return NF_ACCEPT;
 554
 555 out:
 556         NF_CT_STAT_INC(net, insert_failed);
 557         spin_unlock_bh(&nf_conntrack_lock);
 558         return NF_DROP;
 559 }
 560 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 561
 562 /* Returns true if a connection correspondings to the tuple (required
 563    for NAT). */
 564 int
 565 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 566                          const struct nf_conn *ignored_conntrack)
 567 {
 568         struct net *net = nf_ct_net(ignored_conntrack);
 569         struct nf_conntrack_tuple_hash *h;
 570         struct hlist_nulls_node *n;
 571         struct nf_conn *ct;
 572         u16 zone = nf_ct_zone(ignored_conntrack);
 573         unsigned int hash = hash_conntrack(net, zone, tuple);
 574
 575         /* Disable BHs the entire time since we need to disable them at
 576          * least once for the stats anyway.
 577          */
 578         rcu_read_lock_bh();
 579         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 580                 ct = nf_ct_tuplehash_to_ctrack(h);
 581                 if (ct != ignored_conntrack &&
 582                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 583                     nf_ct_zone(ct) == zone) {
 584                         NF_CT_STAT_INC(net, found);
 585                         rcu_read_unlock_bh();
 586                         return 1;
 587                 }
 588                 NF_CT_STAT_INC(net, searched);
 589         }
 590         rcu_read_unlock_bh();
 591
 592         return 0;
 593 }
 594 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 595
 596 #define NF_CT_EVICTION_RANGE    8
 597
 598 /* There's a small race here where we may free a just-assured
 599    connection.  Too bad: we're in trouble anyway. */
 600 static noinline int early_drop(struct net *net, unsigned int hash)
 601 {
 602         /* Use oldest entry, which is roughly LRU */
 603         struct nf_conntrack_tuple_hash *h;
 604         struct nf_conn *ct = NULL, *tmp;
 605         struct hlist_nulls_node *n;
 606         unsigned int i, cnt = 0;
 607         int dropped = 0;
 608
 609         rcu_read_lock();
 610         for (i = 0; i < net->ct.htable_size; i++) {
 611                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 612                                          hnnode) {
 613                         tmp = nf_ct_tuplehash_to_ctrack(h);
 614                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 615                                 ct = tmp;
 616                         cnt++;
 617                 }
 618
 619                 if (ct != NULL) {
 620                         if (likely(!nf_ct_is_dying(ct) &&
 621                                    atomic_inc_not_zero(&ct->ct_general.use)))
 622                                 break;
 623                         else
 624                                 ct = NULL;
 625                 }
 626
 627                 if (cnt >= NF_CT_EVICTION_RANGE)
 628                         break;
 629
 630                 hash = (hash + 1) % net->ct.htable_size;
 631         }
 632         rcu_read_unlock();
 633
 634         if (!ct)
 635                 return dropped;
 636
 637         if (del_timer(&ct->timeout)) {
 638                 death_by_timeout((unsigned long)ct);
 639                 /* Check if we indeed killed this entry. Reliable event
 640                    delivery may have inserted it into the dying list. */
 641                 if (test_bit(IPS_DYING_BIT, &ct->status)) {
 642                         dropped = 1;
 643                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 644                 }
 645         }
 646         nf_ct_put(ct);
 647         return dropped;
 648 }
 649
 650 void init_nf_conntrack_hash_rnd(void)
 651 {
 652         unsigned int rand;
 653
 654         /*
 655          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 656          * Because there isn't enough entropy when system initializing,
 657          * and we initialize it as late as possible.
 658          */
 659         do {
 660                 get_random_bytes(&rand, sizeof(rand));
 661         } while (!rand);
 662         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 663 }
 664
 665 static struct nf_conn *
 666 __nf_conntrack_alloc(struct net *net, u16 zone,
 667                      const struct nf_conntrack_tuple *orig,
 668                      const struct nf_conntrack_tuple *repl,
 669                      gfp_t gfp, u32 hash)
 670 {
 671         struct nf_conn *ct;
 672
 673         if (unlikely(!nf_conntrack_hash_rnd)) {
 674                 init_nf_conntrack_hash_rnd();
 675                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 676                 hash = hash_conntrack_raw(orig, zone);
 677         }
 678
 679         /* We don't want any race condition at early drop stage */
 680         atomic_inc(&net->ct.count);
 681
 682         if (nf_conntrack_max &&
 683             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 684                 if (!early_drop(net, hash_bucket(hash, net))) {
 685                         atomic_dec(&net->ct.count);
 686                         if (net_ratelimit())
 687                                 printk(KERN_WARNING
 688                                        "nf_conntrack: table full, dropping"
 689                                        " packet.\n");
 690                         return ERR_PTR(-ENOMEM);
 691                 }
 692         }
 693
 694         /*
 695          * Do not use kmem_cache_zalloc(), as this cache uses
 696          * SLAB_DESTROY_BY_RCU.
 697          */
 698         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 699         if (ct == NULL) {
 700                 atomic_dec(&net->ct.count);
 701                 return ERR_PTR(-ENOMEM);
 702         }
 703         /*
 704          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 705          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 706          */
 707         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 708                offsetof(struct nf_conn, proto) -
 709                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 710         spin_lock_init(&ct->lock);
 711         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 712         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 713         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 714         /* save hash for reusing when confirming */
 715         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 716         /* Don't set timer yet: wait for confirmation */
 717         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 718         write_pnet(&ct->ct_net, net);
 719 #ifdef CONFIG_NF_CONNTRACK_ZONES
 720         if (zone) {
 721                 struct nf_conntrack_zone *nf_ct_zone;
 722
 723                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 724                 if (!nf_ct_zone)
 725                         goto out_free;
 726                 nf_ct_zone->id = zone;
 727         }
 728 #endif
 729         /*
 730          * changes to lookup keys must be done before setting refcnt to 1
 731          */
 732         smp_wmb();
 733         atomic_set(&ct->ct_general.use, 1);
 734         return ct;
 735
 736 #ifdef CONFIG_NF_CONNTRACK_ZONES
 737 out_free:
 738         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 739         return ERR_PTR(-ENOMEM);
 740 #endif
 741 }
 742
 743 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 744                                    const struct nf_conntrack_tuple *orig,
 745                                    const struct nf_conntrack_tuple *repl,
 746                                    gfp_t gfp)
 747 {
 748         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 749 }
 750 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 751
 752 void nf_conntrack_free(struct nf_conn *ct)
 753 {
 754         struct net *net = nf_ct_net(ct);
 755
 756         nf_ct_ext_destroy(ct);
 757         atomic_dec(&net->ct.count);
 758         nf_ct_ext_free(ct);
 759         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 760 }
 761 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 762
 763 /* Allocate a new conntrack: we return -ENOMEM if classification
 764    failed due to stress.  Otherwise it really is unclassifiable. */
 765 static struct nf_conntrack_tuple_hash *
 766 init_conntrack(struct net *net, struct nf_conn *tmpl,
 767                const struct nf_conntrack_tuple *tuple,
 768                struct nf_conntrack_l3proto *l3proto,
 769                struct nf_conntrack_l4proto *l4proto,
 770                struct sk_buff *skb,
 771                unsigned int dataoff, u32 hash)
 772 {
 773         struct nf_conn *ct;
 774         struct nf_conn_help *help;
 775         struct nf_conntrack_tuple repl_tuple;
 776         struct nf_conntrack_ecache *ecache;
 777         struct nf_conntrack_expect *exp;
 778         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 779         struct nf_conn_timeout *timeout_ext;
 780         unsigned int *timeouts;
 781
 782         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 783                 pr_debug("Can't invert tuple.\n");
 784                 return NULL;
 785         }
 786
 787         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 788                                   hash);
 789         if (IS_ERR(ct))
 790                 return (struct nf_conntrack_tuple_hash *)ct;
 791
 792         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 793         if (timeout_ext)
 794                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 795         else
 796                 timeouts = l4proto->get_timeouts(net);
 797
 798         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 799                 nf_conntrack_free(ct);
 800                 pr_debug("init conntrack: can't track with proto module\n");
 801                 return NULL;
 802         }
 803
 804         if (timeout_ext)
 805                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 806
 807         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 808         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 809
 810         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 811         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 812                                  ecache ? ecache->expmask : 0,
 813                              GFP_ATOMIC);
 814
 815         spin_lock_bh(&nf_conntrack_lock);
 816         exp = nf_ct_find_expectation(net, zone, tuple);
 817         if (exp) {
 818                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 819                          ct, exp);
 820                 /* Welcome, Mr. Bond.  We've been expecting you... */
 821                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 822                 ct->master = exp->master;
 823                 if (exp->helper) {
 824                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 825                         if (help)
 826                                 rcu_assign_pointer(help->helper, exp->helper);
 827                 }
 828
 829 #ifdef CONFIG_NF_CONNTRACK_MARK
 830                 ct->mark = exp->master->mark;
 831 #endif
 832 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 833                 ct->secmark = exp->master->secmark;
 834 #endif
 835                 nf_conntrack_get(&ct->master->ct_general);
 836                 NF_CT_STAT_INC(net, expect_new);
 837         } else {
 838                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 839                 NF_CT_STAT_INC(net, new);
 840         }
 841
 842         /* Overload tuple linked list to put us in unconfirmed list. */
 843         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 844                        &net->ct.unconfirmed);
 845
 846         spin_unlock_bh(&nf_conntrack_lock);
 847
 848         if (exp) {
 849                 if (exp->expectfn)
 850                         exp->expectfn(ct, exp);
 851                 nf_ct_expect_put(exp);
 852         }
 853
 854         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 855 }
 856
 857 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 858 static inline struct nf_conn *
 859 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 860                   struct sk_buff *skb,
 861                   unsigned int dataoff,
 862                   u_int16_t l3num,
 863                   u_int8_t protonum,
 864                   struct nf_conntrack_l3proto *l3proto,
 865                   struct nf_conntrack_l4proto *l4proto,
 866                   int *set_reply,
 867                   enum ip_conntrack_info *ctinfo)
 868 {
 869         struct nf_conntrack_tuple tuple;
 870         struct nf_conntrack_tuple_hash *h;
 871         struct nf_conn *ct;
 872         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 873         u32 hash;
 874
 875         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 876                              dataoff, l3num, protonum, &tuple, l3proto,
 877                              l4proto)) {
 878                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 879                 return NULL;
 880         }
 881
 882         /* look for tuple match */
 883         hash = hash_conntrack_raw(&tuple, zone);
 884         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 885         if (!h) {
 886                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 887                                    skb, dataoff, hash);
 888                 if (!h)
 889                         return NULL;
 890                 if (IS_ERR(h))
 891                         return (void *)h;
 892         }
 893         ct = nf_ct_tuplehash_to_ctrack(h);
 894
 895         /* It exists; we have (non-exclusive) reference. */
 896         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 897                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 898                 /* Please set reply bit if this packet OK */
 899                 *set_reply = 1;
 900         } else {
 901                 /* Once we've had two way comms, always ESTABLISHED. */
 902                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 903                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 904                         *ctinfo = IP_CT_ESTABLISHED;
 905                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 906                         pr_debug("nf_conntrack_in: related packet for %p\n",
 907                                  ct);
 908                         *ctinfo = IP_CT_RELATED;
 909                 } else {
 910                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 911                         *ctinfo = IP_CT_NEW;
 912                 }
 913                 *set_reply = 0;
 914         }
 915         skb->nfct = &ct->ct_general;
 916         skb->nfctinfo = *ctinfo;
 917         return ct;
 918 }
 919
 920 unsigned int
 921 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 922                 struct sk_buff *skb)
 923 {
 924         struct nf_conn *ct, *tmpl = NULL;
 925         enum ip_conntrack_info ctinfo;
 926         struct nf_conntrack_l3proto *l3proto;
 927         struct nf_conntrack_l4proto *l4proto;
 928         struct nf_conn_timeout *timeout_ext;
 929         unsigned int *timeouts;
 930         unsigned int dataoff;
 931         u_int8_t protonum;
 932         int set_reply = 0;
 933         int ret;
 934
 935         if (skb->nfct) {
 936                 /* Previously seen (loopback or untracked)?  Ignore. */
 937                 tmpl = (struct nf_conn *)skb->nfct;
 938                 if (!nf_ct_is_template(tmpl)) {
 939                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 940                         return NF_ACCEPT;
 941                 }
 942                 skb->nfct = NULL;
 943         }
 944
 945         /* rcu_read_lock()ed by nf_hook_slow */
 946         l3proto = __nf_ct_l3proto_find(pf);
 947         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 948                                    &dataoff, &protonum);
 949         if (ret <= 0) {
 950                 pr_debug("not prepared to track yet or error occurred\n");
 951                 NF_CT_STAT_INC_ATOMIC(net, error);
 952                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 953                 ret = -ret;
 954                 goto out;
 955         }
 956
 957         l4proto = __nf_ct_l4proto_find(pf, protonum);
 958
 959         /* It may be an special packet, error, unclean...
 960          * inverse of the return code tells to the netfilter
 961          * core what to do with the packet. */
 962         if (l4proto->error != NULL) {
 963                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 964                                      pf, hooknum);
 965                 if (ret <= 0) {
 966                         NF_CT_STAT_INC_ATOMIC(net, error);
 967                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 968                         ret = -ret;
 969                         goto out;
 970                 }
 971                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 972                 if (skb->nfct)
 973                         goto out;
 974         }
 975
 976         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 977                                l3proto, l4proto, &set_reply, &ctinfo);
 978         if (!ct) {
 979                 /* Not valid part of a connection */
 980                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 981                 ret = NF_ACCEPT;
 982                 goto out;
 983         }
 984
 985         if (IS_ERR(ct)) {
 986                 /* Too stressed to deal. */
 987                 NF_CT_STAT_INC_ATOMIC(net, drop);
 988                 ret = NF_DROP;
 989                 goto out;
 990         }
 991
 992         NF_CT_ASSERT(skb->nfct);
 993
 994         /* Decide what timeout policy we want to apply to this flow. */
 995         timeout_ext = nf_ct_timeout_find(ct);
 996         if (timeout_ext)
 997                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 998         else
 999                 timeouts = l4proto->get_timeouts(net);
1000
1001         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1002         if (ret <= 0) {
1003                 /* Invalid: inverse of the return code tells
1004                  * the netfilter core what to do */
1005                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1006                 nf_conntrack_put(skb->nfct);
1007                 skb->nfct = NULL;
1008                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1009                 if (ret == -NF_DROP)
1010                         NF_CT_STAT_INC_ATOMIC(net, drop);
1011                 ret = -ret;
1012                 goto out;
1013         }
1014
1015         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1016                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1017 out:
1018         if (tmpl) {
1019                 /* Special case: we have to repeat this hook, assign the
1020                  * template again to this packet. We assume that this packet
1021                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1022                 if (ret == NF_REPEAT)
1023                         skb->nfct = (struct nf_conntrack *)tmpl;
1024                 else
1025                         nf_ct_put(tmpl);
1026         }
1027
1028         return ret;
1029 }
1030 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1031
1032 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1033                           const struct nf_conntrack_tuple *orig)
1034 {
1035         bool ret;
1036
1037         rcu_read_lock();
1038         ret = nf_ct_invert_tuple(inverse, orig,
1039                                  __nf_ct_l3proto_find(orig->src.l3num),
1040                                  __nf_ct_l4proto_find(orig->src.l3num,
1041                                                       orig->dst.protonum));
1042         rcu_read_unlock();
1043         return ret;
1044 }
1045 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1046
1047 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1048    implicitly racy: see __nf_conntrack_confirm */
1049 void nf_conntrack_alter_reply(struct nf_conn *ct,
1050                               const struct nf_conntrack_tuple *newreply)
1051 {
1052         struct nf_conn_help *help = nfct_help(ct);
1053
1054         /* Should be unconfirmed, so not in hash table yet */
1055         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1056
1057         pr_debug("Altering reply tuple of %p to ", ct);
1058         nf_ct_dump_tuple(newreply);
1059
1060         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1061         if (ct->master || (help && !hlist_empty(&help->expectations)))
1062                 return;
1063
1064         rcu_read_lock();
1065         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1066         rcu_read_unlock();
1067 }
1068 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1069
1070 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1071 void __nf_ct_refresh_acct(struct nf_conn *ct,
1072                           enum ip_conntrack_info ctinfo,
1073                           const struct sk_buff *skb,
1074                           unsigned long extra_jiffies,
1075                           int do_acct)
1076 {
1077         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1078         NF_CT_ASSERT(skb);
1079
1080         /* Only update if this is not a fixed timeout */
1081         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1082                 goto acct;
1083
1084         /* If not in hash table, timer will not be active yet */
1085         if (!nf_ct_is_confirmed(ct)) {
1086                 ct->timeout.expires = extra_jiffies;
1087         } else {
1088                 unsigned long newtime = jiffies + extra_jiffies;
1089
1090                 /* Only update the timeout if the new timeout is at least
1091                    HZ jiffies from the old timeout. Need del_timer for race
1092                    avoidance (may already be dying). */
1093                 if (newtime - ct->timeout.expires >= HZ)
1094                         mod_timer_pending(&ct->timeout, newtime);
1095         }
1096
1097 acct:
1098         if (do_acct) {
1099                 struct nf_conn_counter *acct;
1100
1101                 acct = nf_conn_acct_find(ct);
1102                 if (acct) {
1103                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1104                         atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes);
1105                 }
1106         }
1107 }
1108 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1109
1110 bool __nf_ct_kill_acct(struct nf_conn *ct,
1111                        enum ip_conntrack_info ctinfo,
1112                        const struct sk_buff *skb,
1113                        int do_acct)
1114 {
1115         if (do_acct) {
1116                 struct nf_conn_counter *acct;
1117
1118                 acct = nf_conn_acct_find(ct);
1119                 if (acct) {
1120                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1121                         atomic64_add(skb->len - skb_network_offset(skb),
1122                                      &acct[CTINFO2DIR(ctinfo)].bytes);
1123                 }
1124         }
1125
1126         if (del_timer(&ct->timeout)) {
1127                 ct->timeout.function((unsigned long)ct);
1128                 return true;
1129         }
1130         return false;
1131 }
1132 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1133
1134 #ifdef CONFIG_NF_CONNTRACK_ZONES
1135 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1136         .len    = sizeof(struct nf_conntrack_zone),
1137         .align  = __alignof__(struct nf_conntrack_zone),
1138         .id     = NF_CT_EXT_ZONE,
1139 };
1140 #endif
1141
1142 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1143
1144 #include <linux/netfilter/nfnetlink.h>
1145 #include <linux/netfilter/nfnetlink_conntrack.h>
1146 #include <linux/mutex.h>
1147
1148 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1149  * in ip_conntrack_core, since we don't want the protocols to autoload
1150  * or depend on ctnetlink */
1151 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1152                                const struct nf_conntrack_tuple *tuple)
1153 {
1154         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
1155         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
1156         return 0;
1157
1158 nla_put_failure:
1159         return -1;
1160 }
1161 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1162
1163 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1164         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1165         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1166 };
1167 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1168
1169 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1170                                struct nf_conntrack_tuple *t)
1171 {
1172         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1173                 return -EINVAL;
1174
1175         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1176         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1177
1178         return 0;
1179 }
1180 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1181
1182 int nf_ct_port_nlattr_tuple_size(void)
1183 {
1184         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1185 }
1186 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1187 #endif
1188
1189 /* Used by ipt_REJECT and ip6t_REJECT. */
1190 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1191 {
1192         struct nf_conn *ct;
1193         enum ip_conntrack_info ctinfo;
1194
1195         /* This ICMP is in reverse direction to the packet which caused it */
1196         ct = nf_ct_get(skb, &ctinfo);
1197         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1198                 ctinfo = IP_CT_RELATED_REPLY;
1199         else
1200                 ctinfo = IP_CT_RELATED;
1201
1202         /* Attach to new skbuff, and increment count */
1203         nskb->nfct = &ct->ct_general;
1204         nskb->nfctinfo = ctinfo;
1205         nf_conntrack_get(nskb->nfct);
1206 }
1207
1208 /* Bring out ya dead! */
1209 static struct nf_conn *
1210 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1211                 void *data, unsigned int *bucket)
1212 {
1213         struct nf_conntrack_tuple_hash *h;
1214         struct nf_conn *ct;
1215         struct hlist_nulls_node *n;
1216
1217         spin_lock_bh(&nf_conntrack_lock);
1218         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1219                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1220                         ct = nf_ct_tuplehash_to_ctrack(h);
1221                         if (iter(ct, data))
1222                                 goto found;
1223                 }
1224         }
1225         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1226                 ct = nf_ct_tuplehash_to_ctrack(h);
1227                 if (iter(ct, data))
1228                         set_bit(IPS_DYING_BIT, &ct->status);
1229         }
1230         spin_unlock_bh(&nf_conntrack_lock);
1231         return NULL;
1232 found:
1233         atomic_inc(&ct->ct_general.use);
1234         spin_unlock_bh(&nf_conntrack_lock);
1235         return ct;
1236 }
1237
1238 void nf_ct_iterate_cleanup(struct net *net,
1239                            int (*iter)(struct nf_conn *i, void *data),
1240                            void *data)
1241 {
1242         struct nf_conn *ct;
1243         unsigned int bucket = 0;
1244
1245         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1246                 /* Time to push up daises... */
1247                 if (del_timer(&ct->timeout))
1248                         death_by_timeout((unsigned long)ct);
1249                 /* ... else the timer will get him soon. */
1250
1251                 nf_ct_put(ct);
1252         }
1253 }
1254 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1255
1256 struct __nf_ct_flush_report {
1257         u32 pid;
1258         int report;
1259 };
1260
1261 static int kill_report(struct nf_conn *i, void *data)
1262 {
1263         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1264         struct nf_conn_tstamp *tstamp;
1265
1266         tstamp = nf_conn_tstamp_find(i);
1267         if (tstamp && tstamp->stop == 0)
1268                 tstamp->stop = ktime_to_ns(ktime_get_real());
1269
1270         /* If we fail to deliver the event, death_by_timeout() will retry */
1271         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1272                                       fr->pid, fr->report) < 0)
1273                 return 1;
1274
1275         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1276         set_bit(IPS_DYING_BIT, &i->status);
1277         return 1;
1278 }
1279
1280 static int kill_all(struct nf_conn *i, void *data)
1281 {
1282         return 1;
1283 }
1284
1285 void nf_ct_free_hashtable(void *hash, unsigned int size)
1286 {
1287         if (is_vmalloc_addr(hash))
1288                 vfree(hash);
1289         else
1290                 free_pages((unsigned long)hash,
1291                            get_order(sizeof(struct hlist_head) * size));
1292 }
1293 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1294
1295 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1296 {
1297         struct __nf_ct_flush_report fr = {
1298                 .pid    = pid,
1299                 .report = report,
1300         };
1301         nf_ct_iterate_cleanup(net, kill_report, &fr);
1302 }
1303 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1304
1305 static void nf_ct_release_dying_list(struct net *net)
1306 {
1307         struct nf_conntrack_tuple_hash *h;
1308         struct nf_conn *ct;
1309         struct hlist_nulls_node *n;
1310
1311         spin_lock_bh(&nf_conntrack_lock);
1312         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1313                 ct = nf_ct_tuplehash_to_ctrack(h);
1314                 /* never fails to remove them, no listeners at this point */
1315                 nf_ct_kill(ct);
1316         }
1317         spin_unlock_bh(&nf_conntrack_lock);
1318 }
1319
1320 static int untrack_refs(void)
1321 {
1322         int cnt = 0, cpu;
1323
1324         for_each_possible_cpu(cpu) {
1325                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1326
1327                 cnt += atomic_read(&ct->ct_general.use) - 1;
1328         }
1329         return cnt;
1330 }
1331
1332 static void nf_conntrack_cleanup_init_net(void)
1333 {
1334         while (untrack_refs() > 0)
1335                 schedule();
1336
1337         nf_conntrack_helper_fini();
1338         nf_conntrack_proto_fini();
1339 #ifdef CONFIG_NF_CONNTRACK_ZONES
1340         nf_ct_extend_unregister(&nf_ct_zone_extend);
1341 #endif
1342 }
1343
1344 static void nf_conntrack_cleanup_net(struct net *net)
1345 {
1346  i_see_dead_people:
1347         nf_ct_iterate_cleanup(net, kill_all, NULL);
1348         nf_ct_release_dying_list(net);
1349         if (atomic_read(&net->ct.count) != 0) {
1350                 schedule();
1351                 goto i_see_dead_people;
1352         }
1353
1354         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1355         nf_conntrack_timeout_fini(net);
1356         nf_conntrack_ecache_fini(net);
1357         nf_conntrack_tstamp_fini(net);
1358         nf_conntrack_acct_fini(net);
1359         nf_conntrack_expect_fini(net);
1360         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1361         kfree(net->ct.slabname);
1362         free_percpu(net->ct.stat);
1363 }
1364
1365 /* Mishearing the voices in his head, our hero wonders how he's
1366    supposed to kill the mall. */
1367 void nf_conntrack_cleanup(struct net *net)
1368 {
1369         if (net_eq(net, &init_net))
1370                 RCU_INIT_POINTER(ip_ct_attach, NULL);
1371
1372         /* This makes sure all current packets have passed through
1373            netfilter framework.  Roll on, two-stage module
1374            delete... */
1375         synchronize_net();
1376
1377         nf_conntrack_cleanup_net(net);
1378
1379         if (net_eq(net, &init_net)) {
1380                 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1381                 nf_conntrack_cleanup_init_net();
1382         }
1383 }
1384
1385 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1386 {
1387         struct hlist_nulls_head *hash;
1388         unsigned int nr_slots, i;
1389         size_t sz;
1390
1391         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1392         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1393         sz = nr_slots * sizeof(struct hlist_nulls_head);
1394         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1395                                         get_order(sz));
1396         if (!hash) {
1397                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1398                 hash = vzalloc(sz);
1399         }
1400
1401         if (hash && nulls)
1402                 for (i = 0; i < nr_slots; i++)
1403                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1404
1405         return hash;
1406 }
1407 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1408
1409 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1410 {
1411         int i, bucket;
1412         unsigned int hashsize, old_size;
1413         struct hlist_nulls_head *hash, *old_hash;
1414         struct nf_conntrack_tuple_hash *h;
1415         struct nf_conn *ct;
1416
1417         if (current->nsproxy->net_ns != &init_net)
1418                 return -EOPNOTSUPP;
1419
1420         /* On boot, we can set this without any fancy locking. */
1421         if (!nf_conntrack_htable_size)
1422                 return param_set_uint(val, kp);
1423
1424         hashsize = simple_strtoul(val, NULL, 0);
1425         if (!hashsize)
1426                 return -EINVAL;
1427
1428         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1429         if (!hash)
1430                 return -ENOMEM;
1431
1432         /* Lookups in the old hash might happen in parallel, which means we
1433          * might get false negatives during connection lookup. New connections
1434          * created because of a false negative won't make it into the hash
1435          * though since that required taking the lock.
1436          */
1437         spin_lock_bh(&nf_conntrack_lock);
1438         for (i = 0; i < init_net.ct.htable_size; i++) {
1439                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1440                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1441                                         struct nf_conntrack_tuple_hash, hnnode);
1442                         ct = nf_ct_tuplehash_to_ctrack(h);
1443                         hlist_nulls_del_rcu(&h->hnnode);
1444                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1445                                                   hashsize);
1446                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1447                 }
1448         }
1449         old_size = init_net.ct.htable_size;
1450         old_hash = init_net.ct.hash;
1451
1452         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1453         init_net.ct.hash = hash;
1454         spin_unlock_bh(&nf_conntrack_lock);
1455
1456         nf_ct_free_hashtable(old_hash, old_size);
1457         return 0;
1458 }
1459 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1460
1461 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1462                   &nf_conntrack_htable_size, 0600);
1463
1464 void nf_ct_untracked_status_or(unsigned long bits)
1465 {
1466         int cpu;
1467
1468         for_each_possible_cpu(cpu)
1469                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1470 }
1471 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1472
1473 static int nf_conntrack_init_init_net(void)
1474 {
1475         int max_factor = 8;
1476         int ret, cpu;
1477
1478         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1479          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1480         if (!nf_conntrack_htable_size) {
1481                 nf_conntrack_htable_size
1482                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1483                            / sizeof(struct hlist_head));
1484                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1485                         nf_conntrack_htable_size = 16384;
1486                 if (nf_conntrack_htable_size < 32)
1487                         nf_conntrack_htable_size = 32;
1488
1489                 /* Use a max. factor of four by default to get the same max as
1490                  * with the old struct list_heads. When a table size is given
1491                  * we use the old value of 8 to avoid reducing the max.
1492                  * entries. */
1493                 max_factor = 4;
1494         }
1495         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1496
1497         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1498                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1499                nf_conntrack_max);
1500
1501         ret = nf_conntrack_proto_init();
1502         if (ret < 0)
1503                 goto err_proto;
1504
1505         ret = nf_conntrack_helper_init();
1506         if (ret < 0)
1507                 goto err_helper;
1508
1509 #ifdef CONFIG_NF_CONNTRACK_ZONES
1510         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1511         if (ret < 0)
1512                 goto err_extend;
1513 #endif
1514         /* Set up fake conntrack: to never be deleted, not in any hashes */
1515         for_each_possible_cpu(cpu) {
1516                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1517                 write_pnet(&ct->ct_net, &init_net);
1518                 atomic_set(&ct->ct_general.use, 1);
1519         }
1520         /*  - and look it like as a confirmed connection */
1521         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1522         return 0;
1523
1524 #ifdef CONFIG_NF_CONNTRACK_ZONES
1525 err_extend:
1526         nf_conntrack_helper_fini();
1527 #endif
1528 err_helper:
1529         nf_conntrack_proto_fini();
1530 err_proto:
1531         return ret;
1532 }
1533
1534 /*
1535  * We need to use special "null" values, not used in hash table
1536  */
1537 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1538 #define DYING_NULLS_VAL         ((1<<30)+1)
1539
1540 static int nf_conntrack_init_net(struct net *net)
1541 {
1542         int ret;
1543
1544         atomic_set(&net->ct.count, 0);
1545         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1546         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1547         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1548         if (!net->ct.stat) {
1549                 ret = -ENOMEM;
1550                 goto err_stat;
1551         }
1552
1553         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1554         if (!net->ct.slabname) {
1555                 ret = -ENOMEM;
1556                 goto err_slabname;
1557         }
1558
1559         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1560                                                         sizeof(struct nf_conn), 0,
1561                                                         SLAB_DESTROY_BY_RCU, NULL);
1562         if (!net->ct.nf_conntrack_cachep) {
1563                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1564                 ret = -ENOMEM;
1565                 goto err_cache;
1566         }
1567
1568         net->ct.htable_size = nf_conntrack_htable_size;
1569         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1570         if (!net->ct.hash) {
1571                 ret = -ENOMEM;
1572                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1573                 goto err_hash;
1574         }
1575         ret = nf_conntrack_expect_init(net);
1576         if (ret < 0)
1577                 goto err_expect;
1578         ret = nf_conntrack_acct_init(net);
1579         if (ret < 0)
1580                 goto err_acct;
1581         ret = nf_conntrack_tstamp_init(net);
1582         if (ret < 0)
1583                 goto err_tstamp;
1584         ret = nf_conntrack_ecache_init(net);
1585         if (ret < 0)
1586                 goto err_ecache;
1587         ret = nf_conntrack_timeout_init(net);
1588         if (ret < 0)
1589                 goto err_timeout;
1590
1591         return 0;
1592
1593 err_timeout:
1594         nf_conntrack_timeout_fini(net);
1595 err_ecache:
1596         nf_conntrack_tstamp_fini(net);
1597 err_tstamp:
1598         nf_conntrack_acct_fini(net);
1599 err_acct:
1600         nf_conntrack_expect_fini(net);
1601 err_expect:
1602         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1603 err_hash:
1604         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1605 err_cache:
1606         kfree(net->ct.slabname);
1607 err_slabname:
1608         free_percpu(net->ct.stat);
1609 err_stat:
1610         return ret;
1611 }
1612
1613 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1614                         enum ip_conntrack_dir dir,
1615                         u32 seq);
1616 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);
1617
1618 int nf_conntrack_init(struct net *net)
1619 {
1620         int ret;
1621
1622         if (net_eq(net, &init_net)) {
1623                 ret = nf_conntrack_init_init_net();
1624                 if (ret < 0)
1625                         goto out_init_net;
1626         }
1627         ret = nf_conntrack_init_net(net);
1628         if (ret < 0)
1629                 goto out_net;
1630
1631         if (net_eq(net, &init_net)) {
1632                 /* For use by REJECT target */
1633                 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1634                 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1635
1636                 /* Howto get NAT offsets */
1637                 RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
1638         }
1639         return 0;
1640
1641 out_net:
1642         if (net_eq(net, &init_net))
1643                 nf_conntrack_cleanup_init_net();
1644 out_init_net:
1645         return ret;
1646 }