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  * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  */
  14
  15 #include <linux/types.h>
  16 #include <linux/netfilter.h>
  17 #include <linux/module.h>
  18 #include <linux/sched.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/proc_fs.h>
  21 #include <linux/vmalloc.h>
  22 #include <linux/stddef.h>
  23 #include <linux/slab.h>
  24 #include <linux/random.h>
  25 #include <linux/jhash.h>
  26 #include <linux/err.h>
  27 #include <linux/percpu.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/notifier.h>
  30 #include <linux/kernel.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/socket.h>
  33 #include <linux/mm.h>
  34 #include <linux/nsproxy.h>
  35 #include <linux/rculist_nulls.h>
  36
  37 #include <net/netfilter/nf_conntrack.h>
  38 #include <net/netfilter/nf_conntrack_l3proto.h>
  39 #include <net/netfilter/nf_conntrack_l4proto.h>
  40 #include <net/netfilter/nf_conntrack_expect.h>
  41 #include <net/netfilter/nf_conntrack_helper.h>
  42 #include <net/netfilter/nf_conntrack_core.h>
  43 #include <net/netfilter/nf_conntrack_extend.h>
  44 #include <net/netfilter/nf_conntrack_acct.h>
  45 #include <net/netfilter/nf_conntrack_ecache.h>
  46 #include <net/netfilter/nf_conntrack_zones.h>
  47 #include <net/netfilter/nf_conntrack_timestamp.h>
  48 #include <net/netfilter/nf_conntrack_timeout.h>
  49 #include <net/netfilter/nf_conntrack_labels.h>
  50 #include <net/netfilter/nf_nat.h>
  51 #include <net/netfilter/nf_nat_core.h>
  52 #include <net/netfilter/nf_nat_helper.h>
  53
  54 #define NF_CONNTRACK_VERSION    "0.5.0"
  55
  56 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  57                                       enum nf_nat_manip_type manip,
  58                                       const struct nlattr *attr) __read_mostly;
  59 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  60
  61 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
  62                               struct nf_conn *ct,
  63                               enum ip_conntrack_info ctinfo,
  64                               unsigned int protoff);
  65 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
  66
  67 DEFINE_SPINLOCK(nf_conntrack_lock);
  68 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  69
  70 unsigned int nf_conntrack_htable_size __read_mostly;
  71 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  72
  73 unsigned int nf_conntrack_max __read_mostly;
  74 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  75
  76 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  77 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  78
  79 unsigned int nf_conntrack_hash_rnd __read_mostly;
  80 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
  81
  82 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  83 {
  84         unsigned int n;
  85
  86         /* The direction must be ignored, so we hash everything up to the
  87          * destination ports (which is a multiple of 4) and treat the last
  88          * three bytes manually.
  89          */
  90         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  91         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  92                       (((__force __u16)tuple->dst.u.all << 16) |
  93                       tuple->dst.protonum));
  94 }
  95
  96 static u32 __hash_bucket(u32 hash, unsigned int size)
  97 {
  98         return ((u64)hash * size) >> 32;
  99 }
 100
 101 static u32 hash_bucket(u32 hash, const struct net *net)
 102 {
 103         return __hash_bucket(hash, net->ct.htable_size);
 104 }
 105
 106 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
 107                                   u16 zone, unsigned int size)
 108 {
 109         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 110 }
 111
 112 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 113                                        const struct nf_conntrack_tuple *tuple)
 114 {
 115         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 116 }
 117
 118 bool
 119 nf_ct_get_tuple(const struct sk_buff *skb,
 120                 unsigned int nhoff,
 121                 unsigned int dataoff,
 122                 u_int16_t l3num,
 123                 u_int8_t protonum,
 124                 struct nf_conntrack_tuple *tuple,
 125                 const struct nf_conntrack_l3proto *l3proto,
 126                 const struct nf_conntrack_l4proto *l4proto)
 127 {
 128         memset(tuple, 0, sizeof(*tuple));
 129
 130         tuple->src.l3num = l3num;
 131         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 132                 return false;
 133
 134         tuple->dst.protonum = protonum;
 135         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 136
 137         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 138 }
 139 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 140
 141 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 142                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 143 {
 144         struct nf_conntrack_l3proto *l3proto;
 145         struct nf_conntrack_l4proto *l4proto;
 146         unsigned int protoff;
 147         u_int8_t protonum;
 148         int ret;
 149
 150         rcu_read_lock();
 151
 152         l3proto = __nf_ct_l3proto_find(l3num);
 153         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 154         if (ret != NF_ACCEPT) {
 155                 rcu_read_unlock();
 156                 return false;
 157         }
 158
 159         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 160
 161         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 162                               l3proto, l4proto);
 163
 164         rcu_read_unlock();
 165         return ret;
 166 }
 167 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 168
 169 bool
 170 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 171                    const struct nf_conntrack_tuple *orig,
 172                    const struct nf_conntrack_l3proto *l3proto,
 173                    const struct nf_conntrack_l4proto *l4proto)
 174 {
 175         memset(inverse, 0, sizeof(*inverse));
 176
 177         inverse->src.l3num = orig->src.l3num;
 178         if (l3proto->invert_tuple(inverse, orig) == 0)
 179                 return false;
 180
 181         inverse->dst.dir = !orig->dst.dir;
 182
 183         inverse->dst.protonum = orig->dst.protonum;
 184         return l4proto->invert_tuple(inverse, orig);
 185 }
 186 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 187
 188 static void
 189 clean_from_lists(struct nf_conn *ct)
 190 {
 191         pr_debug("clean_from_lists(%p)\n", ct);
 192         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 193         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 194
 195         /* Destroy all pending expectations */
 196         nf_ct_remove_expectations(ct);
 197 }
 198
 199 static void
 200 destroy_conntrack(struct nf_conntrack *nfct)
 201 {
 202         struct nf_conn *ct = (struct nf_conn *)nfct;
 203         struct net *net = nf_ct_net(ct);
 204         struct nf_conntrack_l4proto *l4proto;
 205
 206         pr_debug("destroy_conntrack(%p)\n", ct);
 207         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 208         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 209
 210         /* To make sure we don't get any weird locking issues here:
 211          * destroy_conntrack() MUST NOT be called with a write lock
 212          * to nf_conntrack_lock!!! -HW */
 213         rcu_read_lock();
 214         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 215         if (l4proto && l4proto->destroy)
 216                 l4proto->destroy(ct);
 217
 218         rcu_read_unlock();
 219
 220         spin_lock_bh(&nf_conntrack_lock);
 221         /* Expectations will have been removed in clean_from_lists,
 222          * except TFTP can create an expectation on the first packet,
 223          * before connection is in the list, so we need to clean here,
 224          * too. */
 225         nf_ct_remove_expectations(ct);
 226
 227         /* We overload first tuple to link into unconfirmed or dying list.*/
 228         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 229         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 230
 231         NF_CT_STAT_INC(net, delete);
 232         spin_unlock_bh(&nf_conntrack_lock);
 233
 234         if (ct->master)
 235                 nf_ct_put(ct->master);
 236
 237         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 238         nf_conntrack_free(ct);
 239 }
 240
 241 void nf_ct_delete_from_lists(struct nf_conn *ct)
 242 {
 243         struct net *net = nf_ct_net(ct);
 244
 245         nf_ct_helper_destroy(ct);
 246         spin_lock_bh(&nf_conntrack_lock);
 247         /* Inside lock so preempt is disabled on module removal path.
 248          * Otherwise we can get spurious warnings. */
 249         NF_CT_STAT_INC(net, delete_list);
 250         clean_from_lists(ct);
 251         /* add this conntrack to the dying list */
 252         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 253                              &net->ct.dying);
 254         spin_unlock_bh(&nf_conntrack_lock);
 255 }
 256 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
 257
 258 static void death_by_event(unsigned long ul_conntrack)
 259 {
 260         struct nf_conn *ct = (void *)ul_conntrack;
 261         struct net *net = nf_ct_net(ct);
 262         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 263
 264         BUG_ON(ecache == NULL);
 265
 266         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 267                 /* bad luck, let's retry again */
 268                 ecache->timeout.expires = jiffies +
 269                         (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 270                 add_timer(&ecache->timeout);
 271                 return;
 272         }
 273         /* we've got the event delivered, now it's dying */
 274         set_bit(IPS_DYING_BIT, &ct->status);
 275         nf_ct_put(ct);
 276 }
 277
 278 void nf_ct_dying_timeout(struct nf_conn *ct)
 279 {
 280         struct net *net = nf_ct_net(ct);
 281         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 282
 283         BUG_ON(ecache == NULL);
 284
 285         /* set a new timer to retry event delivery */
 286         setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
 287         ecache->timeout.expires = jiffies +
 288                 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 289         add_timer(&ecache->timeout);
 290 }
 291 EXPORT_SYMBOL_GPL(nf_ct_dying_timeout);
 292
 293 static void death_by_timeout(unsigned long ul_conntrack)
 294 {
 295         struct nf_conn *ct = (void *)ul_conntrack;
 296         struct nf_conn_tstamp *tstamp;
 297
 298         tstamp = nf_conn_tstamp_find(ct);
 299         if (tstamp && tstamp->stop == 0)
 300                 tstamp->stop = ktime_to_ns(ktime_get_real());
 301
 302         if (!test_bit(IPS_DYING_BIT, &ct->status) &&
 303             unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
 304                 /* destroy event was not delivered */
 305                 nf_ct_delete_from_lists(ct);
 306                 nf_ct_dying_timeout(ct);
 307                 return;
 308         }
 309         set_bit(IPS_DYING_BIT, &ct->status);
 310         nf_ct_delete_from_lists(ct);
 311         nf_ct_put(ct);
 312 }
 313
 314 /*
 315  * Warning :
 316  * - Caller must take a reference on returned object
 317  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 318  * OR
 319  * - Caller must lock nf_conntrack_lock before calling this function
 320  */
 321 static struct nf_conntrack_tuple_hash *
 322 ____nf_conntrack_find(struct net *net, u16 zone,
 323                       const struct nf_conntrack_tuple *tuple, u32 hash)
 324 {
 325         struct nf_conntrack_tuple_hash *h;
 326         struct hlist_nulls_node *n;
 327         unsigned int bucket = hash_bucket(hash, net);
 328
 329         /* Disable BHs the entire time since we normally need to disable them
 330          * at least once for the stats anyway.
 331          */
 332         local_bh_disable();
 333 begin:
 334         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 335                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 336                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 337                         NF_CT_STAT_INC(net, found);
 338                         local_bh_enable();
 339                         return h;
 340                 }
 341                 NF_CT_STAT_INC(net, searched);
 342         }
 343         /*
 344          * if the nulls value we got at the end of this lookup is
 345          * not the expected one, we must restart lookup.
 346          * We probably met an item that was moved to another chain.
 347          */
 348         if (get_nulls_value(n) != bucket) {
 349                 NF_CT_STAT_INC(net, search_restart);
 350                 goto begin;
 351         }
 352         local_bh_enable();
 353
 354         return NULL;
 355 }
 356
 357 struct nf_conntrack_tuple_hash *
 358 __nf_conntrack_find(struct net *net, u16 zone,
 359                     const struct nf_conntrack_tuple *tuple)
 360 {
 361         return ____nf_conntrack_find(net, zone, tuple,
 362                                      hash_conntrack_raw(tuple, zone));
 363 }
 364 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 365
 366 /* Find a connection corresponding to a tuple. */
 367 static struct nf_conntrack_tuple_hash *
 368 __nf_conntrack_find_get(struct net *net, u16 zone,
 369                         const struct nf_conntrack_tuple *tuple, u32 hash)
 370 {
 371         struct nf_conntrack_tuple_hash *h;
 372         struct nf_conn *ct;
 373
 374         rcu_read_lock();
 375 begin:
 376         h = ____nf_conntrack_find(net, zone, tuple, hash);
 377         if (h) {
 378                 ct = nf_ct_tuplehash_to_ctrack(h);
 379                 if (unlikely(nf_ct_is_dying(ct) ||
 380                              !atomic_inc_not_zero(&ct->ct_general.use)))
 381                         h = NULL;
 382                 else {
 383                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 384                                      nf_ct_zone(ct) != zone)) {
 385                                 nf_ct_put(ct);
 386                                 goto begin;
 387                         }
 388                 }
 389         }
 390         rcu_read_unlock();
 391
 392         return h;
 393 }
 394
 395 struct nf_conntrack_tuple_hash *
 396 nf_conntrack_find_get(struct net *net, u16 zone,
 397                       const struct nf_conntrack_tuple *tuple)
 398 {
 399         return __nf_conntrack_find_get(net, zone, tuple,
 400                                        hash_conntrack_raw(tuple, zone));
 401 }
 402 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 403
 404 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 405                                        unsigned int hash,
 406                                        unsigned int repl_hash)
 407 {
 408         struct net *net = nf_ct_net(ct);
 409
 410         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 411                            &net->ct.hash[hash]);
 412         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 413                            &net->ct.hash[repl_hash]);
 414 }
 415
 416 int
 417 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 418 {
 419         struct net *net = nf_ct_net(ct);
 420         unsigned int hash, repl_hash;
 421         struct nf_conntrack_tuple_hash *h;
 422         struct hlist_nulls_node *n;
 423         u16 zone;
 424
 425         zone = nf_ct_zone(ct);
 426         hash = hash_conntrack(net, zone,
 427                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 428         repl_hash = hash_conntrack(net, zone,
 429                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 430
 431         spin_lock_bh(&nf_conntrack_lock);
 432
 433         /* See if there's one in the list already, including reverse */
 434         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 435                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 436                                       &h->tuple) &&
 437                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 438                         goto out;
 439         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 440                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 441                                       &h->tuple) &&
 442                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 443                         goto out;
 444
 445         add_timer(&ct->timeout);
 446         nf_conntrack_get(&ct->ct_general);
 447         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 448         NF_CT_STAT_INC(net, insert);
 449         spin_unlock_bh(&nf_conntrack_lock);
 450
 451         return 0;
 452
 453 out:
 454         NF_CT_STAT_INC(net, insert_failed);
 455         spin_unlock_bh(&nf_conntrack_lock);
 456         return -EEXIST;
 457 }
 458 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 459
 460 /* Confirm a connection given skb; places it in hash table */
 461 int
 462 __nf_conntrack_confirm(struct sk_buff *skb)
 463 {
 464         unsigned int hash, repl_hash;
 465         struct nf_conntrack_tuple_hash *h;
 466         struct nf_conn *ct;
 467         struct nf_conn_help *help;
 468         struct nf_conn_tstamp *tstamp;
 469         struct hlist_nulls_node *n;
 470         enum ip_conntrack_info ctinfo;
 471         struct net *net;
 472         u16 zone;
 473
 474         ct = nf_ct_get(skb, &ctinfo);
 475         net = nf_ct_net(ct);
 476
 477         /* ipt_REJECT uses nf_conntrack_attach to attach related
 478            ICMP/TCP RST packets in other direction.  Actual packet
 479            which created connection will be IP_CT_NEW or for an
 480            expected connection, IP_CT_RELATED. */
 481         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 482                 return NF_ACCEPT;
 483
 484         zone = nf_ct_zone(ct);
 485         /* reuse the hash saved before */
 486         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 487         hash = hash_bucket(hash, net);
 488         repl_hash = hash_conntrack(net, zone,
 489                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 490
 491         /* We're not in hash table, and we refuse to set up related
 492            connections for unconfirmed conns.  But packet copies and
 493            REJECT will give spurious warnings here. */
 494         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 495
 496         /* No external references means no one else could have
 497            confirmed us. */
 498         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 499         pr_debug("Confirming conntrack %p\n", ct);
 500
 501         spin_lock_bh(&nf_conntrack_lock);
 502
 503         /* We have to check the DYING flag inside the lock to prevent
 504            a race against nf_ct_get_next_corpse() possibly called from
 505            user context, else we insert an already 'dead' hash, blocking
 506            further use of that particular connection -JM */
 507
 508         if (unlikely(nf_ct_is_dying(ct))) {
 509                 spin_unlock_bh(&nf_conntrack_lock);
 510                 return NF_ACCEPT;
 511         }
 512
 513         /* See if there's one in the list already, including reverse:
 514            NAT could have grabbed it without realizing, since we're
 515            not in the hash.  If there is, we lost race. */
 516         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 517                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 518                                       &h->tuple) &&
 519                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 520                         goto out;
 521         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 522                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 523                                       &h->tuple) &&
 524                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 525                         goto out;
 526
 527         /* Remove from unconfirmed list */
 528         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 529
 530         /* Timer relative to confirmation time, not original
 531            setting time, otherwise we'd get timer wrap in
 532            weird delay cases. */
 533         ct->timeout.expires += jiffies;
 534         add_timer(&ct->timeout);
 535         atomic_inc(&ct->ct_general.use);
 536         ct->status |= IPS_CONFIRMED;
 537
 538         /* set conntrack timestamp, if enabled. */
 539         tstamp = nf_conn_tstamp_find(ct);
 540         if (tstamp) {
 541                 if (skb->tstamp.tv64 == 0)
 542                         __net_timestamp(skb);
 543
 544                 tstamp->start = ktime_to_ns(skb->tstamp);
 545         }
 546         /* Since the lookup is lockless, hash insertion must be done after
 547          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 548          * guarantee that no other CPU can find the conntrack before the above
 549          * stores are visible.
 550          */
 551         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 552         NF_CT_STAT_INC(net, insert);
 553         spin_unlock_bh(&nf_conntrack_lock);
 554
 555         help = nfct_help(ct);
 556         if (help && help->helper)
 557                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 558
 559         nf_conntrack_event_cache(master_ct(ct) ?
 560                                  IPCT_RELATED : IPCT_NEW, ct);
 561         return NF_ACCEPT;
 562
 563 out:
 564         NF_CT_STAT_INC(net, insert_failed);
 565         spin_unlock_bh(&nf_conntrack_lock);
 566         return NF_DROP;
 567 }
 568 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 569
 570 /* Returns true if a connection correspondings to the tuple (required
 571    for NAT). */
 572 int
 573 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 574                          const struct nf_conn *ignored_conntrack)
 575 {
 576         struct net *net = nf_ct_net(ignored_conntrack);
 577         struct nf_conntrack_tuple_hash *h;
 578         struct hlist_nulls_node *n;
 579         struct nf_conn *ct;
 580         u16 zone = nf_ct_zone(ignored_conntrack);
 581         unsigned int hash = hash_conntrack(net, zone, tuple);
 582
 583         /* Disable BHs the entire time since we need to disable them at
 584          * least once for the stats anyway.
 585          */
 586         rcu_read_lock_bh();
 587         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 588                 ct = nf_ct_tuplehash_to_ctrack(h);
 589                 if (ct != ignored_conntrack &&
 590                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 591                     nf_ct_zone(ct) == zone) {
 592                         NF_CT_STAT_INC(net, found);
 593                         rcu_read_unlock_bh();
 594                         return 1;
 595                 }
 596                 NF_CT_STAT_INC(net, searched);
 597         }
 598         rcu_read_unlock_bh();
 599
 600         return 0;
 601 }
 602 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 603
 604 #define NF_CT_EVICTION_RANGE    8
 605
 606 /* There's a small race here where we may free a just-assured
 607    connection.  Too bad: we're in trouble anyway. */
 608 static noinline int early_drop(struct net *net, unsigned int hash)
 609 {
 610         /* Use oldest entry, which is roughly LRU */
 611         struct nf_conntrack_tuple_hash *h;
 612         struct nf_conn *ct = NULL, *tmp;
 613         struct hlist_nulls_node *n;
 614         unsigned int i, cnt = 0;
 615         int dropped = 0;
 616
 617         rcu_read_lock();
 618         for (i = 0; i < net->ct.htable_size; i++) {
 619                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 620                                          hnnode) {
 621                         tmp = nf_ct_tuplehash_to_ctrack(h);
 622                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 623                                 ct = tmp;
 624                         cnt++;
 625                 }
 626
 627                 if (ct != NULL) {
 628                         if (likely(!nf_ct_is_dying(ct) &&
 629                                    atomic_inc_not_zero(&ct->ct_general.use)))
 630                                 break;
 631                         else
 632                                 ct = NULL;
 633                 }
 634
 635                 if (cnt >= NF_CT_EVICTION_RANGE)
 636                         break;
 637
 638                 hash = (hash + 1) % net->ct.htable_size;
 639         }
 640         rcu_read_unlock();
 641
 642         if (!ct)
 643                 return dropped;
 644
 645         if (del_timer(&ct->timeout)) {
 646                 death_by_timeout((unsigned long)ct);
 647                 /* Check if we indeed killed this entry. Reliable event
 648                    delivery may have inserted it into the dying list. */
 649                 if (test_bit(IPS_DYING_BIT, &ct->status)) {
 650                         dropped = 1;
 651                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 652                 }
 653         }
 654         nf_ct_put(ct);
 655         return dropped;
 656 }
 657
 658 void init_nf_conntrack_hash_rnd(void)
 659 {
 660         unsigned int rand;
 661
 662         /*
 663          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 664          * Because there isn't enough entropy when system initializing,
 665          * and we initialize it as late as possible.
 666          */
 667         do {
 668                 get_random_bytes(&rand, sizeof(rand));
 669         } while (!rand);
 670         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 671 }
 672
 673 static struct nf_conn *
 674 __nf_conntrack_alloc(struct net *net, u16 zone,
 675                      const struct nf_conntrack_tuple *orig,
 676                      const struct nf_conntrack_tuple *repl,
 677                      gfp_t gfp, u32 hash)
 678 {
 679         struct nf_conn *ct;
 680
 681         if (unlikely(!nf_conntrack_hash_rnd)) {
 682                 init_nf_conntrack_hash_rnd();
 683                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 684                 hash = hash_conntrack_raw(orig, zone);
 685         }
 686
 687         /* We don't want any race condition at early drop stage */
 688         atomic_inc(&net->ct.count);
 689
 690         if (nf_conntrack_max &&
 691             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 692                 if (!early_drop(net, hash_bucket(hash, net))) {
 693                         atomic_dec(&net->ct.count);
 694                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
 695                         return ERR_PTR(-ENOMEM);
 696                 }
 697         }
 698
 699         /*
 700          * Do not use kmem_cache_zalloc(), as this cache uses
 701          * SLAB_DESTROY_BY_RCU.
 702          */
 703         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 704         if (ct == NULL) {
 705                 atomic_dec(&net->ct.count);
 706                 return ERR_PTR(-ENOMEM);
 707         }
 708         /*
 709          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 710          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 711          */
 712         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 713                offsetof(struct nf_conn, proto) -
 714                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 715         spin_lock_init(&ct->lock);
 716         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 717         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 718         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 719         /* save hash for reusing when confirming */
 720         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 721         /* Don't set timer yet: wait for confirmation */
 722         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 723         write_pnet(&ct->ct_net, net);
 724 #ifdef CONFIG_NF_CONNTRACK_ZONES
 725         if (zone) {
 726                 struct nf_conntrack_zone *nf_ct_zone;
 727
 728                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 729                 if (!nf_ct_zone)
 730                         goto out_free;
 731                 nf_ct_zone->id = zone;
 732         }
 733 #endif
 734         /*
 735          * changes to lookup keys must be done before setting refcnt to 1
 736          */
 737         smp_wmb();
 738         atomic_set(&ct->ct_general.use, 1);
 739         return ct;
 740
 741 #ifdef CONFIG_NF_CONNTRACK_ZONES
 742 out_free:
 743         atomic_dec(&net->ct.count);
 744         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 745         return ERR_PTR(-ENOMEM);
 746 #endif
 747 }
 748
 749 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 750                                    const struct nf_conntrack_tuple *orig,
 751                                    const struct nf_conntrack_tuple *repl,
 752                                    gfp_t gfp)
 753 {
 754         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 755 }
 756 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 757
 758 void nf_conntrack_free(struct nf_conn *ct)
 759 {
 760         struct net *net = nf_ct_net(ct);
 761
 762         nf_ct_ext_destroy(ct);
 763         atomic_dec(&net->ct.count);
 764         nf_ct_ext_free(ct);
 765         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 766 }
 767 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 768
 769
 770 /* Allocate a new conntrack: we return -ENOMEM if classification
 771    failed due to stress.  Otherwise it really is unclassifiable. */
 772 static struct nf_conntrack_tuple_hash *
 773 init_conntrack(struct net *net, struct nf_conn *tmpl,
 774                const struct nf_conntrack_tuple *tuple,
 775                struct nf_conntrack_l3proto *l3proto,
 776                struct nf_conntrack_l4proto *l4proto,
 777                struct sk_buff *skb,
 778                unsigned int dataoff, u32 hash)
 779 {
 780         struct nf_conn *ct;
 781         struct nf_conn_help *help;
 782         struct nf_conntrack_tuple repl_tuple;
 783         struct nf_conntrack_ecache *ecache;
 784         struct nf_conntrack_expect *exp;
 785         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 786         struct nf_conn_timeout *timeout_ext;
 787         unsigned int *timeouts;
 788
 789         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 790                 pr_debug("Can't invert tuple.\n");
 791                 return NULL;
 792         }
 793
 794         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 795                                   hash);
 796         if (IS_ERR(ct))
 797                 return (struct nf_conntrack_tuple_hash *)ct;
 798
 799         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 800         if (timeout_ext)
 801                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 802         else
 803                 timeouts = l4proto->get_timeouts(net);
 804
 805         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 806                 nf_conntrack_free(ct);
 807                 pr_debug("init conntrack: can't track with proto module\n");
 808                 return NULL;
 809         }
 810
 811         if (timeout_ext)
 812                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 813
 814         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 815         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 816         nf_ct_labels_ext_add(ct);
 817
 818         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 819         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 820                                  ecache ? ecache->expmask : 0,
 821                              GFP_ATOMIC);
 822
 823         spin_lock_bh(&nf_conntrack_lock);
 824         exp = nf_ct_find_expectation(net, zone, tuple);
 825         if (exp) {
 826                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 827                          ct, exp);
 828                 /* Welcome, Mr. Bond.  We've been expecting you... */
 829                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 830                 ct->master = exp->master;
 831                 if (exp->helper) {
 832                         help = nf_ct_helper_ext_add(ct, exp->helper,
 833                                                     GFP_ATOMIC);
 834                         if (help)
 835                                 rcu_assign_pointer(help->helper, exp->helper);
 836                 }
 837
 838 #ifdef CONFIG_NF_CONNTRACK_MARK
 839                 ct->mark = exp->master->mark;
 840 #endif
 841 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 842                 ct->secmark = exp->master->secmark;
 843 #endif
 844                 nf_conntrack_get(&ct->master->ct_general);
 845                 NF_CT_STAT_INC(net, expect_new);
 846         } else {
 847                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 848                 NF_CT_STAT_INC(net, new);
 849         }
 850
 851         /* Overload tuple linked list to put us in unconfirmed list. */
 852         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 853                        &net->ct.unconfirmed);
 854
 855         spin_unlock_bh(&nf_conntrack_lock);
 856
 857         if (exp) {
 858                 if (exp->expectfn)
 859                         exp->expectfn(ct, exp);
 860                 nf_ct_expect_put(exp);
 861         }
 862
 863         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 864 }
 865
 866 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 867 static inline struct nf_conn *
 868 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 869                   struct sk_buff *skb,
 870                   unsigned int dataoff,
 871                   u_int16_t l3num,
 872                   u_int8_t protonum,
 873                   struct nf_conntrack_l3proto *l3proto,
 874                   struct nf_conntrack_l4proto *l4proto,
 875                   int *set_reply,
 876                   enum ip_conntrack_info *ctinfo)
 877 {
 878         struct nf_conntrack_tuple tuple;
 879         struct nf_conntrack_tuple_hash *h;
 880         struct nf_conn *ct;
 881         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 882         u32 hash;
 883
 884         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 885                              dataoff, l3num, protonum, &tuple, l3proto,
 886                              l4proto)) {
 887                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 888                 return NULL;
 889         }
 890
 891         /* look for tuple match */
 892         hash = hash_conntrack_raw(&tuple, zone);
 893         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 894         if (!h) {
 895                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 896                                    skb, dataoff, hash);
 897                 if (!h)
 898                         return NULL;
 899                 if (IS_ERR(h))
 900                         return (void *)h;
 901         }
 902         ct = nf_ct_tuplehash_to_ctrack(h);
 903
 904         /* It exists; we have (non-exclusive) reference. */
 905         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 906                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 907                 /* Please set reply bit if this packet OK */
 908                 *set_reply = 1;
 909         } else {
 910                 /* Once we've had two way comms, always ESTABLISHED. */
 911                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 912                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 913                         *ctinfo = IP_CT_ESTABLISHED;
 914                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 915                         pr_debug("nf_conntrack_in: related packet for %p\n",
 916                                  ct);
 917                         *ctinfo = IP_CT_RELATED;
 918                 } else {
 919                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 920                         *ctinfo = IP_CT_NEW;
 921                 }
 922                 *set_reply = 0;
 923         }
 924         skb->nfct = &ct->ct_general;
 925         skb->nfctinfo = *ctinfo;
 926         return ct;
 927 }
 928
 929 unsigned int
 930 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 931                 struct sk_buff *skb)
 932 {
 933         struct nf_conn *ct, *tmpl = NULL;
 934         enum ip_conntrack_info ctinfo;
 935         struct nf_conntrack_l3proto *l3proto;
 936         struct nf_conntrack_l4proto *l4proto;
 937         unsigned int *timeouts;
 938         unsigned int dataoff;
 939         u_int8_t protonum;
 940         int set_reply = 0;
 941         int ret;
 942
 943         if (skb->nfct) {
 944                 /* Previously seen (loopback or untracked)?  Ignore. */
 945                 tmpl = (struct nf_conn *)skb->nfct;
 946                 if (!nf_ct_is_template(tmpl)) {
 947                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 948                         return NF_ACCEPT;
 949                 }
 950                 skb->nfct = NULL;
 951         }
 952
 953         /* rcu_read_lock()ed by nf_hook_slow */
 954         l3proto = __nf_ct_l3proto_find(pf);
 955         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 956                                    &dataoff, &protonum);
 957         if (ret <= 0) {
 958                 pr_debug("not prepared to track yet or error occurred\n");
 959                 NF_CT_STAT_INC_ATOMIC(net, error);
 960                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 961                 ret = -ret;
 962                 goto out;
 963         }
 964
 965         l4proto = __nf_ct_l4proto_find(pf, protonum);
 966
 967         /* It may be an special packet, error, unclean...
 968          * inverse of the return code tells to the netfilter
 969          * core what to do with the packet. */
 970         if (l4proto->error != NULL) {
 971                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 972                                      pf, hooknum);
 973                 if (ret <= 0) {
 974                         NF_CT_STAT_INC_ATOMIC(net, error);
 975                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 976                         ret = -ret;
 977                         goto out;
 978                 }
 979                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 980                 if (skb->nfct)
 981                         goto out;
 982         }
 983
 984         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 985                                l3proto, l4proto, &set_reply, &ctinfo);
 986         if (!ct) {
 987                 /* Not valid part of a connection */
 988                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 989                 ret = NF_ACCEPT;
 990                 goto out;
 991         }
 992
 993         if (IS_ERR(ct)) {
 994                 /* Too stressed to deal. */
 995                 NF_CT_STAT_INC_ATOMIC(net, drop);
 996                 ret = NF_DROP;
 997                 goto out;
 998         }
 999
1000         NF_CT_ASSERT(skb->nfct);
1001
1002         /* Decide what timeout policy we want to apply to this flow. */
1003         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1004
1005         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1006         if (ret <= 0) {
1007                 /* Invalid: inverse of the return code tells
1008                  * the netfilter core what to do */
1009                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1010                 nf_conntrack_put(skb->nfct);
1011                 skb->nfct = NULL;
1012                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1013                 if (ret == -NF_DROP)
1014                         NF_CT_STAT_INC_ATOMIC(net, drop);
1015                 ret = -ret;
1016                 goto out;
1017         }
1018
1019         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1020                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1021 out:
1022         if (tmpl) {
1023                 /* Special case: we have to repeat this hook, assign the
1024                  * template again to this packet. We assume that this packet
1025                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1026                 if (ret == NF_REPEAT)
1027                         skb->nfct = (struct nf_conntrack *)tmpl;
1028                 else
1029                         nf_ct_put(tmpl);
1030         }
1031
1032         return ret;
1033 }
1034 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1035
1036 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1037                           const struct nf_conntrack_tuple *orig)
1038 {
1039         bool ret;
1040
1041         rcu_read_lock();
1042         ret = nf_ct_invert_tuple(inverse, orig,
1043                                  __nf_ct_l3proto_find(orig->src.l3num),
1044                                  __nf_ct_l4proto_find(orig->src.l3num,
1045                                                       orig->dst.protonum));
1046         rcu_read_unlock();
1047         return ret;
1048 }
1049 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1050
1051 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1052    implicitly racy: see __nf_conntrack_confirm */
1053 void nf_conntrack_alter_reply(struct nf_conn *ct,
1054                               const struct nf_conntrack_tuple *newreply)
1055 {
1056         struct nf_conn_help *help = nfct_help(ct);
1057
1058         /* Should be unconfirmed, so not in hash table yet */
1059         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1060
1061         pr_debug("Altering reply tuple of %p to ", ct);
1062         nf_ct_dump_tuple(newreply);
1063
1064         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1065         if (ct->master || (help && !hlist_empty(&help->expectations)))
1066                 return;
1067
1068         rcu_read_lock();
1069         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1070         rcu_read_unlock();
1071 }
1072 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1073
1074 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1075 void __nf_ct_refresh_acct(struct nf_conn *ct,
1076                           enum ip_conntrack_info ctinfo,
1077                           const struct sk_buff *skb,
1078                           unsigned long extra_jiffies,
1079                           int do_acct)
1080 {
1081         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1082         NF_CT_ASSERT(skb);
1083
1084         /* Only update if this is not a fixed timeout */
1085         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1086                 goto acct;
1087
1088         /* If not in hash table, timer will not be active yet */
1089         if (!nf_ct_is_confirmed(ct)) {
1090                 ct->timeout.expires = extra_jiffies;
1091         } else {
1092                 unsigned long newtime = jiffies + extra_jiffies;
1093
1094                 /* Only update the timeout if the new timeout is at least
1095                    HZ jiffies from the old timeout. Need del_timer for race
1096                    avoidance (may already be dying). */
1097                 if (newtime - ct->timeout.expires >= HZ)
1098                         mod_timer_pending(&ct->timeout, newtime);
1099         }
1100
1101 acct:
1102         if (do_acct) {
1103                 struct nf_conn_counter *acct;
1104
1105                 acct = nf_conn_acct_find(ct);
1106                 if (acct) {
1107                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1108                         atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes);
1109                 }
1110         }
1111 }
1112 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1113
1114 bool __nf_ct_kill_acct(struct nf_conn *ct,
1115                        enum ip_conntrack_info ctinfo,
1116                        const struct sk_buff *skb,
1117                        int do_acct)
1118 {
1119         if (do_acct) {
1120                 struct nf_conn_counter *acct;
1121
1122                 acct = nf_conn_acct_find(ct);
1123                 if (acct) {
1124                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1125                         atomic64_add(skb->len - skb_network_offset(skb),
1126                                      &acct[CTINFO2DIR(ctinfo)].bytes);
1127                 }
1128         }
1129
1130         if (del_timer(&ct->timeout)) {
1131                 ct->timeout.function((unsigned long)ct);
1132                 return true;
1133         }
1134         return false;
1135 }
1136 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1137
1138 #ifdef CONFIG_NF_CONNTRACK_ZONES
1139 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1140         .len    = sizeof(struct nf_conntrack_zone),
1141         .align  = __alignof__(struct nf_conntrack_zone),
1142         .id     = NF_CT_EXT_ZONE,
1143 };
1144 #endif
1145
1146 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1147
1148 #include <linux/netfilter/nfnetlink.h>
1149 #include <linux/netfilter/nfnetlink_conntrack.h>
1150 #include <linux/mutex.h>
1151
1152 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1153  * in ip_conntrack_core, since we don't want the protocols to autoload
1154  * or depend on ctnetlink */
1155 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1156                                const struct nf_conntrack_tuple *tuple)
1157 {
1158         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1159             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1160                 goto nla_put_failure;
1161         return 0;
1162
1163 nla_put_failure:
1164         return -1;
1165 }
1166 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1167
1168 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1169         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1170         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1171 };
1172 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1173
1174 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1175                                struct nf_conntrack_tuple *t)
1176 {
1177         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1178                 return -EINVAL;
1179
1180         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1181         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1182
1183         return 0;
1184 }
1185 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1186
1187 int nf_ct_port_nlattr_tuple_size(void)
1188 {
1189         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1190 }
1191 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1192 #endif
1193
1194 /* Used by ipt_REJECT and ip6t_REJECT. */
1195 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1196 {
1197         struct nf_conn *ct;
1198         enum ip_conntrack_info ctinfo;
1199
1200         /* This ICMP is in reverse direction to the packet which caused it */
1201         ct = nf_ct_get(skb, &ctinfo);
1202         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1203                 ctinfo = IP_CT_RELATED_REPLY;
1204         else
1205                 ctinfo = IP_CT_RELATED;
1206
1207         /* Attach to new skbuff, and increment count */
1208         nskb->nfct = &ct->ct_general;
1209         nskb->nfctinfo = ctinfo;
1210         nf_conntrack_get(nskb->nfct);
1211 }
1212
1213 /* Bring out ya dead! */
1214 static struct nf_conn *
1215 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1216                 void *data, unsigned int *bucket)
1217 {
1218         struct nf_conntrack_tuple_hash *h;
1219         struct nf_conn *ct;
1220         struct hlist_nulls_node *n;
1221
1222         spin_lock_bh(&nf_conntrack_lock);
1223         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1224                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1225                         if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1226                                 continue;
1227                         ct = nf_ct_tuplehash_to_ctrack(h);
1228                         if (iter(ct, data))
1229                                 goto found;
1230                 }
1231         }
1232         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1233                 ct = nf_ct_tuplehash_to_ctrack(h);
1234                 if (iter(ct, data))
1235                         set_bit(IPS_DYING_BIT, &ct->status);
1236         }
1237         spin_unlock_bh(&nf_conntrack_lock);
1238         return NULL;
1239 found:
1240         atomic_inc(&ct->ct_general.use);
1241         spin_unlock_bh(&nf_conntrack_lock);
1242         return ct;
1243 }
1244
1245 void nf_ct_iterate_cleanup(struct net *net,
1246                            int (*iter)(struct nf_conn *i, void *data),
1247                            void *data)
1248 {
1249         struct nf_conn *ct;
1250         unsigned int bucket = 0;
1251
1252         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1253                 /* Time to push up daises... */
1254                 if (del_timer(&ct->timeout))
1255                         death_by_timeout((unsigned long)ct);
1256                 /* ... else the timer will get him soon. */
1257
1258                 nf_ct_put(ct);
1259         }
1260 }
1261 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1262
1263 struct __nf_ct_flush_report {
1264         u32 portid;
1265         int report;
1266 };
1267
1268 static int kill_report(struct nf_conn *i, void *data)
1269 {
1270         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1271         struct nf_conn_tstamp *tstamp;
1272
1273         tstamp = nf_conn_tstamp_find(i);
1274         if (tstamp && tstamp->stop == 0)
1275                 tstamp->stop = ktime_to_ns(ktime_get_real());
1276
1277         /* If we fail to deliver the event, death_by_timeout() will retry */
1278         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1279                                       fr->portid, fr->report) < 0)
1280                 return 1;
1281
1282         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1283         set_bit(IPS_DYING_BIT, &i->status);
1284         return 1;
1285 }
1286
1287 static int kill_all(struct nf_conn *i, void *data)
1288 {
1289         return 1;
1290 }
1291
1292 void nf_ct_free_hashtable(void *hash, unsigned int size)
1293 {
1294         if (is_vmalloc_addr(hash))
1295                 vfree(hash);
1296         else
1297                 free_pages((unsigned long)hash,
1298                            get_order(sizeof(struct hlist_head) * size));
1299 }
1300 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1301
1302 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1303 {
1304         struct __nf_ct_flush_report fr = {
1305                 .portid = portid,
1306                 .report = report,
1307         };
1308         nf_ct_iterate_cleanup(net, kill_report, &fr);
1309 }
1310 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1311
1312 static void nf_ct_release_dying_list(struct net *net)
1313 {
1314         struct nf_conntrack_tuple_hash *h;
1315         struct nf_conn *ct;
1316         struct hlist_nulls_node *n;
1317
1318         spin_lock_bh(&nf_conntrack_lock);
1319         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1320                 ct = nf_ct_tuplehash_to_ctrack(h);
1321                 /* never fails to remove them, no listeners at this point */
1322                 nf_ct_kill(ct);
1323         }
1324         spin_unlock_bh(&nf_conntrack_lock);
1325 }
1326
1327 static int untrack_refs(void)
1328 {
1329         int cnt = 0, cpu;
1330
1331         for_each_possible_cpu(cpu) {
1332                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1333
1334                 cnt += atomic_read(&ct->ct_general.use) - 1;
1335         }
1336         return cnt;
1337 }
1338
1339 void nf_conntrack_cleanup_start(void)
1340 {
1341         RCU_INIT_POINTER(ip_ct_attach, NULL);
1342 }
1343
1344 void nf_conntrack_cleanup_end(void)
1345 {
1346         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1347         while (untrack_refs() > 0)
1348                 schedule();
1349
1350 #ifdef CONFIG_NF_CONNTRACK_ZONES
1351         nf_ct_extend_unregister(&nf_ct_zone_extend);
1352 #endif
1353         nf_conntrack_proto_fini();
1354         nf_conntrack_labels_fini();
1355         nf_conntrack_helper_fini();
1356         nf_conntrack_timeout_fini();
1357         nf_conntrack_ecache_fini();
1358         nf_conntrack_tstamp_fini();
1359         nf_conntrack_acct_fini();
1360         nf_conntrack_expect_fini();
1361 }
1362
1363 /*
1364  * Mishearing the voices in his head, our hero wonders how he's
1365  * supposed to kill the mall.
1366  */
1367 void nf_conntrack_cleanup_net(struct net *net)
1368 {
1369         LIST_HEAD(single);
1370
1371         list_add(&net->exit_list, &single);
1372         nf_conntrack_cleanup_net_list(&single);
1373 }
1374
1375 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1376 {
1377         int busy;
1378         struct net *net;
1379
1380         /*
1381          * This makes sure all current packets have passed through
1382          *  netfilter framework.  Roll on, two-stage module
1383          *  delete...
1384          */
1385         synchronize_net();
1386 i_see_dead_people:
1387         busy = 0;
1388         list_for_each_entry(net, net_exit_list, exit_list) {
1389                 nf_ct_iterate_cleanup(net, kill_all, NULL);
1390                 nf_ct_release_dying_list(net);
1391                 if (atomic_read(&net->ct.count) != 0)
1392                         busy = 1;
1393         }
1394         if (busy) {
1395                 schedule();
1396                 goto i_see_dead_people;
1397         }
1398
1399         list_for_each_entry(net, net_exit_list, exit_list) {
1400                 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1401                 nf_conntrack_proto_pernet_fini(net);
1402                 nf_conntrack_helper_pernet_fini(net);
1403                 nf_conntrack_ecache_pernet_fini(net);
1404                 nf_conntrack_tstamp_pernet_fini(net);
1405                 nf_conntrack_acct_pernet_fini(net);
1406                 nf_conntrack_expect_pernet_fini(net);
1407                 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1408                 kfree(net->ct.slabname);
1409                 free_percpu(net->ct.stat);
1410         }
1411 }
1412
1413 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1414 {
1415         struct hlist_nulls_head *hash;
1416         unsigned int nr_slots, i;
1417         size_t sz;
1418
1419         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1420         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1421         sz = nr_slots * sizeof(struct hlist_nulls_head);
1422         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1423                                         get_order(sz));
1424         if (!hash) {
1425                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1426                 hash = vzalloc(sz);
1427         }
1428
1429         if (hash && nulls)
1430                 for (i = 0; i < nr_slots; i++)
1431                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1432
1433         return hash;
1434 }
1435 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1436
1437 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1438 {
1439         int i, bucket, rc;
1440         unsigned int hashsize, old_size;
1441         struct hlist_nulls_head *hash, *old_hash;
1442         struct nf_conntrack_tuple_hash *h;
1443         struct nf_conn *ct;
1444
1445         if (current->nsproxy->net_ns != &init_net)
1446                 return -EOPNOTSUPP;
1447
1448         /* On boot, we can set this without any fancy locking. */
1449         if (!nf_conntrack_htable_size)
1450                 return param_set_uint(val, kp);
1451
1452         rc = kstrtouint(val, 0, &hashsize);
1453         if (rc)
1454                 return rc;
1455         if (!hashsize)
1456                 return -EINVAL;
1457
1458         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1459         if (!hash)
1460                 return -ENOMEM;
1461
1462         /* Lookups in the old hash might happen in parallel, which means we
1463          * might get false negatives during connection lookup. New connections
1464          * created because of a false negative won't make it into the hash
1465          * though since that required taking the lock.
1466          */
1467         spin_lock_bh(&nf_conntrack_lock);
1468         for (i = 0; i < init_net.ct.htable_size; i++) {
1469                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1470                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1471                                         struct nf_conntrack_tuple_hash, hnnode);
1472                         ct = nf_ct_tuplehash_to_ctrack(h);
1473                         hlist_nulls_del_rcu(&h->hnnode);
1474                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1475                                                   hashsize);
1476                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1477                 }
1478         }
1479         old_size = init_net.ct.htable_size;
1480         old_hash = init_net.ct.hash;
1481
1482         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1483         init_net.ct.hash = hash;
1484         spin_unlock_bh(&nf_conntrack_lock);
1485
1486         nf_ct_free_hashtable(old_hash, old_size);
1487         return 0;
1488 }
1489 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1490
1491 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1492                   &nf_conntrack_htable_size, 0600);
1493
1494 void nf_ct_untracked_status_or(unsigned long bits)
1495 {
1496         int cpu;
1497
1498         for_each_possible_cpu(cpu)
1499                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1500 }
1501 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1502
1503 int nf_conntrack_init_start(void)
1504 {
1505         int max_factor = 8;
1506         int ret, cpu;
1507
1508         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1509          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1510         if (!nf_conntrack_htable_size) {
1511                 nf_conntrack_htable_size
1512                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1513                            / sizeof(struct hlist_head));
1514                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1515                         nf_conntrack_htable_size = 16384;
1516                 if (nf_conntrack_htable_size < 32)
1517                         nf_conntrack_htable_size = 32;
1518
1519                 /* Use a max. factor of four by default to get the same max as
1520                  * with the old struct list_heads. When a table size is given
1521                  * we use the old value of 8 to avoid reducing the max.
1522                  * entries. */
1523                 max_factor = 4;
1524         }
1525         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1526
1527         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1528                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1529                nf_conntrack_max);
1530
1531         ret = nf_conntrack_expect_init();
1532         if (ret < 0)
1533                 goto err_expect;
1534
1535         ret = nf_conntrack_acct_init();
1536         if (ret < 0)
1537                 goto err_acct;
1538
1539         ret = nf_conntrack_tstamp_init();
1540         if (ret < 0)
1541                 goto err_tstamp;
1542
1543         ret = nf_conntrack_ecache_init();
1544         if (ret < 0)
1545                 goto err_ecache;
1546
1547         ret = nf_conntrack_timeout_init();
1548         if (ret < 0)
1549                 goto err_timeout;
1550
1551         ret = nf_conntrack_helper_init();
1552         if (ret < 0)
1553                 goto err_helper;
1554
1555         ret = nf_conntrack_labels_init();
1556         if (ret < 0)
1557                 goto err_labels;
1558
1559 #ifdef CONFIG_NF_CONNTRACK_ZONES
1560         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1561         if (ret < 0)
1562                 goto err_extend;
1563 #endif
1564         ret = nf_conntrack_proto_init();
1565         if (ret < 0)
1566                 goto err_proto;
1567
1568         /* Set up fake conntrack: to never be deleted, not in any hashes */
1569         for_each_possible_cpu(cpu) {
1570                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1571                 write_pnet(&ct->ct_net, &init_net);
1572                 atomic_set(&ct->ct_general.use, 1);
1573         }
1574         /*  - and look it like as a confirmed connection */
1575         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1576         return 0;
1577
1578 err_proto:
1579 #ifdef CONFIG_NF_CONNTRACK_ZONES
1580         nf_ct_extend_unregister(&nf_ct_zone_extend);
1581 err_extend:
1582 #endif
1583         nf_conntrack_labels_fini();
1584 err_labels:
1585         nf_conntrack_helper_fini();
1586 err_helper:
1587         nf_conntrack_timeout_fini();
1588 err_timeout:
1589         nf_conntrack_ecache_fini();
1590 err_ecache:
1591         nf_conntrack_tstamp_fini();
1592 err_tstamp:
1593         nf_conntrack_acct_fini();
1594 err_acct:
1595         nf_conntrack_expect_fini();
1596 err_expect:
1597         return ret;
1598 }
1599
1600 void nf_conntrack_init_end(void)
1601 {
1602         /* For use by REJECT target */
1603         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1604         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1605
1606         /* Howto get NAT offsets */
1607         RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
1608 }
1609
1610 /*
1611  * We need to use special "null" values, not used in hash table
1612  */
1613 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1614 #define DYING_NULLS_VAL         ((1<<30)+1)
1615 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1616
1617 int nf_conntrack_init_net(struct net *net)
1618 {
1619         int ret;
1620
1621         atomic_set(&net->ct.count, 0);
1622         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1623         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1624         INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1625         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1626         if (!net->ct.stat) {
1627                 ret = -ENOMEM;
1628                 goto err_stat;
1629         }
1630
1631         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1632         if (!net->ct.slabname) {
1633                 ret = -ENOMEM;
1634                 goto err_slabname;
1635         }
1636
1637         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1638                                                         sizeof(struct nf_conn), 0,
1639                                                         SLAB_DESTROY_BY_RCU, NULL);
1640         if (!net->ct.nf_conntrack_cachep) {
1641                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1642                 ret = -ENOMEM;
1643                 goto err_cache;
1644         }
1645
1646         net->ct.htable_size = nf_conntrack_htable_size;
1647         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1648         if (!net->ct.hash) {
1649                 ret = -ENOMEM;
1650                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1651                 goto err_hash;
1652         }
1653         ret = nf_conntrack_expect_pernet_init(net);
1654         if (ret < 0)
1655                 goto err_expect;
1656         ret = nf_conntrack_acct_pernet_init(net);
1657         if (ret < 0)
1658                 goto err_acct;
1659         ret = nf_conntrack_tstamp_pernet_init(net);
1660         if (ret < 0)
1661                 goto err_tstamp;
1662         ret = nf_conntrack_ecache_pernet_init(net);
1663         if (ret < 0)
1664                 goto err_ecache;
1665         ret = nf_conntrack_helper_pernet_init(net);
1666         if (ret < 0)
1667                 goto err_helper;
1668         ret = nf_conntrack_proto_pernet_init(net);
1669         if (ret < 0)
1670                 goto err_proto;
1671         return 0;
1672
1673 err_proto:
1674         nf_conntrack_helper_pernet_fini(net);
1675 err_helper:
1676         nf_conntrack_ecache_pernet_fini(net);
1677 err_ecache:
1678         nf_conntrack_tstamp_pernet_fini(net);
1679 err_tstamp:
1680         nf_conntrack_acct_pernet_fini(net);
1681 err_acct:
1682         nf_conntrack_expect_pernet_fini(net);
1683 err_expect:
1684         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1685 err_hash:
1686         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1687 err_cache:
1688         kfree(net->ct.slabname);
1689 err_slabname:
1690         free_percpu(net->ct.stat);
1691 err_stat:
1692         return ret;
1693 }
1694
1695 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1696                         enum ip_conntrack_dir dir,
1697                         u32 seq);
1698 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);