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_seqadj.h>
  43 #include <net/netfilter/nf_conntrack_core.h>
  44 #include <net/netfilter/nf_conntrack_extend.h>
  45 #include <net/netfilter/nf_conntrack_acct.h>
  46 #include <net/netfilter/nf_conntrack_ecache.h>
  47 #include <net/netfilter/nf_conntrack_zones.h>
  48 #include <net/netfilter/nf_conntrack_timestamp.h>
  49 #include <net/netfilter/nf_conntrack_timeout.h>
  50 #include <net/netfilter/nf_conntrack_labels.h>
  51 #include <net/netfilter/nf_conntrack_synproxy.h>
  52 #include <net/netfilter/nf_nat.h>
  53 #include <net/netfilter/nf_nat_core.h>
  54 #include <net/netfilter/nf_nat_helper.h>
  55
  56 #define NF_CONNTRACK_VERSION    "0.5.0"
  57
  58 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  59                                       enum nf_nat_manip_type manip,
  60                                       const struct nlattr *attr) __read_mostly;
  61 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  62
  63 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
  64 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
  65
  66 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
  67 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
  68
  69 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
  70 {
  71         h1 %= CONNTRACK_LOCKS;
  72         h2 %= CONNTRACK_LOCKS;
  73         spin_unlock(&nf_conntrack_locks[h1]);
  74         if (h1 != h2)
  75                 spin_unlock(&nf_conntrack_locks[h2]);
  76 }
  77
  78 /* return true if we need to recompute hashes (in case hash table was resized) */
  79 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
  80                                      unsigned int h2, unsigned int sequence)
  81 {
  82         h1 %= CONNTRACK_LOCKS;
  83         h2 %= CONNTRACK_LOCKS;
  84         if (h1 <= h2) {
  85                 spin_lock(&nf_conntrack_locks[h1]);
  86                 if (h1 != h2)
  87                         spin_lock_nested(&nf_conntrack_locks[h2],
  88                                          SINGLE_DEPTH_NESTING);
  89         } else {
  90                 spin_lock(&nf_conntrack_locks[h2]);
  91                 spin_lock_nested(&nf_conntrack_locks[h1],
  92                                  SINGLE_DEPTH_NESTING);
  93         }
  94         if (read_seqcount_retry(&net->ct.generation, sequence)) {
  95                 nf_conntrack_double_unlock(h1, h2);
  96                 return true;
  97         }
  98         return false;
  99 }
 100
 101 static void nf_conntrack_all_lock(void)
 102 {
 103         int i;
 104
 105         for (i = 0; i < CONNTRACK_LOCKS; i++)
 106                 spin_lock_nested(&nf_conntrack_locks[i], i);
 107 }
 108
 109 static void nf_conntrack_all_unlock(void)
 110 {
 111         int i;
 112
 113         for (i = 0; i < CONNTRACK_LOCKS; i++)
 114                 spin_unlock(&nf_conntrack_locks[i]);
 115 }
 116
 117 unsigned int nf_conntrack_htable_size __read_mostly;
 118 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
 119
 120 unsigned int nf_conntrack_max __read_mostly;
 121 EXPORT_SYMBOL_GPL(nf_conntrack_max);
 122
 123 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
 124 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
 125
 126 unsigned int nf_conntrack_hash_rnd __read_mostly;
 127 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
 128
 129 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple)
 130 {
 131         unsigned int n;
 132
 133         /* The direction must be ignored, so we hash everything up to the
 134          * destination ports (which is a multiple of 4) and treat the last
 135          * three bytes manually.
 136          */
 137         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
 138         return jhash2((u32 *)tuple, n, nf_conntrack_hash_rnd ^
 139                       (((__force __u16)tuple->dst.u.all << 16) |
 140                       tuple->dst.protonum));
 141 }
 142
 143 static u32 __hash_bucket(u32 hash, unsigned int size)
 144 {
 145         return reciprocal_scale(hash, size);
 146 }
 147
 148 static u32 hash_bucket(u32 hash, const struct net *net)
 149 {
 150         return __hash_bucket(hash, net->ct.htable_size);
 151 }
 152
 153 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
 154                                   unsigned int size)
 155 {
 156         return __hash_bucket(hash_conntrack_raw(tuple), size);
 157 }
 158
 159 static inline u_int32_t hash_conntrack(const struct net *net,
 160                                        const struct nf_conntrack_tuple *tuple)
 161 {
 162         return __hash_conntrack(tuple, net->ct.htable_size);
 163 }
 164
 165 bool
 166 nf_ct_get_tuple(const struct sk_buff *skb,
 167                 unsigned int nhoff,
 168                 unsigned int dataoff,
 169                 u_int16_t l3num,
 170                 u_int8_t protonum,
 171                 struct nf_conntrack_tuple *tuple,
 172                 const struct nf_conntrack_l3proto *l3proto,
 173                 const struct nf_conntrack_l4proto *l4proto)
 174 {
 175         memset(tuple, 0, sizeof(*tuple));
 176
 177         tuple->src.l3num = l3num;
 178         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 179                 return false;
 180
 181         tuple->dst.protonum = protonum;
 182         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 183
 184         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 185 }
 186 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 187
 188 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 189                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 190 {
 191         struct nf_conntrack_l3proto *l3proto;
 192         struct nf_conntrack_l4proto *l4proto;
 193         unsigned int protoff;
 194         u_int8_t protonum;
 195         int ret;
 196
 197         rcu_read_lock();
 198
 199         l3proto = __nf_ct_l3proto_find(l3num);
 200         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 201         if (ret != NF_ACCEPT) {
 202                 rcu_read_unlock();
 203                 return false;
 204         }
 205
 206         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 207
 208         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 209                               l3proto, l4proto);
 210
 211         rcu_read_unlock();
 212         return ret;
 213 }
 214 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 215
 216 bool
 217 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 218                    const struct nf_conntrack_tuple *orig,
 219                    const struct nf_conntrack_l3proto *l3proto,
 220                    const struct nf_conntrack_l4proto *l4proto)
 221 {
 222         memset(inverse, 0, sizeof(*inverse));
 223
 224         inverse->src.l3num = orig->src.l3num;
 225         if (l3proto->invert_tuple(inverse, orig) == 0)
 226                 return false;
 227
 228         inverse->dst.dir = !orig->dst.dir;
 229
 230         inverse->dst.protonum = orig->dst.protonum;
 231         return l4proto->invert_tuple(inverse, orig);
 232 }
 233 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 234
 235 static void
 236 clean_from_lists(struct nf_conn *ct)
 237 {
 238         pr_debug("clean_from_lists(%p)\n", ct);
 239         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 240         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 241
 242         /* Destroy all pending expectations */
 243         nf_ct_remove_expectations(ct);
 244 }
 245
 246 /* must be called with local_bh_disable */
 247 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
 248 {
 249         struct ct_pcpu *pcpu;
 250
 251         /* add this conntrack to the (per cpu) dying list */
 252         ct->cpu = smp_processor_id();
 253         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 254
 255         spin_lock(&pcpu->lock);
 256         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 257                              &pcpu->dying);
 258         spin_unlock(&pcpu->lock);
 259 }
 260
 261 /* must be called with local_bh_disable */
 262 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
 263 {
 264         struct ct_pcpu *pcpu;
 265
 266         /* add this conntrack to the (per cpu) unconfirmed list */
 267         ct->cpu = smp_processor_id();
 268         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 269
 270         spin_lock(&pcpu->lock);
 271         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 272                              &pcpu->unconfirmed);
 273         spin_unlock(&pcpu->lock);
 274 }
 275
 276 /* must be called with local_bh_disable */
 277 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
 278 {
 279         struct ct_pcpu *pcpu;
 280
 281         /* We overload first tuple to link into unconfirmed or dying list.*/
 282         pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
 283
 284         spin_lock(&pcpu->lock);
 285         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 286         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 287         spin_unlock(&pcpu->lock);
 288 }
 289
 290 /* Released via destroy_conntrack() */
 291 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
 292                                  const struct nf_conntrack_zone *zone,
 293                                  gfp_t flags)
 294 {
 295         struct nf_conn *tmpl;
 296
 297         tmpl = kzalloc(sizeof(*tmpl), flags);
 298         if (tmpl == NULL)
 299                 return NULL;
 300
 301         tmpl->status = IPS_TEMPLATE;
 302         write_pnet(&tmpl->ct_net, net);
 303
 304         if (nf_ct_zone_add(tmpl, flags, zone) < 0)
 305                 goto out_free;
 306
 307         atomic_set(&tmpl->ct_general.use, 0);
 308
 309         return tmpl;
 310 out_free:
 311         kfree(tmpl);
 312         return NULL;
 313 }
 314 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
 315
 316 void nf_ct_tmpl_free(struct nf_conn *tmpl)
 317 {
 318         nf_ct_ext_destroy(tmpl);
 319         nf_ct_ext_free(tmpl);
 320         kfree(tmpl);
 321 }
 322 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
 323
 324 static void
 325 destroy_conntrack(struct nf_conntrack *nfct)
 326 {
 327         struct nf_conn *ct = (struct nf_conn *)nfct;
 328         struct net *net = nf_ct_net(ct);
 329         struct nf_conntrack_l4proto *l4proto;
 330
 331         pr_debug("destroy_conntrack(%p)\n", ct);
 332         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 333         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 334
 335         if (unlikely(nf_ct_is_template(ct))) {
 336                 nf_ct_tmpl_free(ct);
 337                 return;
 338         }
 339         rcu_read_lock();
 340         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 341         if (l4proto && l4proto->destroy)
 342                 l4proto->destroy(ct);
 343
 344         rcu_read_unlock();
 345
 346         local_bh_disable();
 347         /* Expectations will have been removed in clean_from_lists,
 348          * except TFTP can create an expectation on the first packet,
 349          * before connection is in the list, so we need to clean here,
 350          * too.
 351          */
 352         nf_ct_remove_expectations(ct);
 353
 354         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 355
 356         NF_CT_STAT_INC(net, delete);
 357         local_bh_enable();
 358
 359         if (ct->master)
 360                 nf_ct_put(ct->master);
 361
 362         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 363         nf_conntrack_free(ct);
 364 }
 365
 366 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 367 {
 368         struct net *net = nf_ct_net(ct);
 369         unsigned int hash, reply_hash;
 370         unsigned int sequence;
 371
 372         nf_ct_helper_destroy(ct);
 373
 374         local_bh_disable();
 375         do {
 376                 sequence = read_seqcount_begin(&net->ct.generation);
 377                 hash = hash_conntrack(net,
 378                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 379                 reply_hash = hash_conntrack(net,
 380                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 381         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 382
 383         clean_from_lists(ct);
 384         nf_conntrack_double_unlock(hash, reply_hash);
 385
 386         nf_ct_add_to_dying_list(ct);
 387
 388         NF_CT_STAT_INC(net, delete_list);
 389         local_bh_enable();
 390 }
 391
 392 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 393 {
 394         struct nf_conn_tstamp *tstamp;
 395
 396         tstamp = nf_conn_tstamp_find(ct);
 397         if (tstamp && tstamp->stop == 0)
 398                 tstamp->stop = ktime_get_real_ns();
 399
 400         if (nf_ct_is_dying(ct))
 401                 goto delete;
 402
 403         if (nf_conntrack_event_report(IPCT_DESTROY, ct,
 404                                     portid, report) < 0) {
 405                 /* destroy event was not delivered */
 406                 nf_ct_delete_from_lists(ct);
 407                 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
 408                 return false;
 409         }
 410
 411         nf_conntrack_ecache_work(nf_ct_net(ct));
 412         set_bit(IPS_DYING_BIT, &ct->status);
 413  delete:
 414         nf_ct_delete_from_lists(ct);
 415         nf_ct_put(ct);
 416         return true;
 417 }
 418 EXPORT_SYMBOL_GPL(nf_ct_delete);
 419
 420 static void death_by_timeout(unsigned long ul_conntrack)
 421 {
 422         nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
 423 }
 424
 425 static inline bool
 426 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
 427                 const struct nf_conntrack_tuple *tuple,
 428                 const struct nf_conntrack_zone *zone)
 429 {
 430         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 431
 432         /* A conntrack can be recreated with the equal tuple,
 433          * so we need to check that the conntrack is confirmed
 434          */
 435         return nf_ct_tuple_equal(tuple, &h->tuple) &&
 436                nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
 437                nf_ct_is_confirmed(ct);
 438 }
 439
 440 /*
 441  * Warning :
 442  * - Caller must take a reference on returned object
 443  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 444  */
 445 static struct nf_conntrack_tuple_hash *
 446 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
 447                       const struct nf_conntrack_tuple *tuple, u32 hash)
 448 {
 449         struct nf_conntrack_tuple_hash *h;
 450         struct hlist_nulls_node *n;
 451         unsigned int bucket = hash_bucket(hash, net);
 452
 453         /* Disable BHs the entire time since we normally need to disable them
 454          * at least once for the stats anyway.
 455          */
 456         local_bh_disable();
 457 begin:
 458         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 459                 if (nf_ct_key_equal(h, tuple, zone)) {
 460                         NF_CT_STAT_INC(net, found);
 461                         local_bh_enable();
 462                         return h;
 463                 }
 464                 NF_CT_STAT_INC(net, searched);
 465         }
 466         /*
 467          * if the nulls value we got at the end of this lookup is
 468          * not the expected one, we must restart lookup.
 469          * We probably met an item that was moved to another chain.
 470          */
 471         if (get_nulls_value(n) != bucket) {
 472                 NF_CT_STAT_INC(net, search_restart);
 473                 goto begin;
 474         }
 475         local_bh_enable();
 476
 477         return NULL;
 478 }
 479
 480 /* Find a connection corresponding to a tuple. */
 481 static struct nf_conntrack_tuple_hash *
 482 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 483                         const struct nf_conntrack_tuple *tuple, u32 hash)
 484 {
 485         struct nf_conntrack_tuple_hash *h;
 486         struct nf_conn *ct;
 487
 488         rcu_read_lock();
 489 begin:
 490         h = ____nf_conntrack_find(net, zone, tuple, hash);
 491         if (h) {
 492                 ct = nf_ct_tuplehash_to_ctrack(h);
 493                 if (unlikely(nf_ct_is_dying(ct) ||
 494                              !atomic_inc_not_zero(&ct->ct_general.use)))
 495                         h = NULL;
 496                 else {
 497                         if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
 498                                 nf_ct_put(ct);
 499                                 goto begin;
 500                         }
 501                 }
 502         }
 503         rcu_read_unlock();
 504
 505         return h;
 506 }
 507
 508 struct nf_conntrack_tuple_hash *
 509 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
 510                       const struct nf_conntrack_tuple *tuple)
 511 {
 512         return __nf_conntrack_find_get(net, zone, tuple,
 513                                        hash_conntrack_raw(tuple));
 514 }
 515 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 516
 517 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 518                                        unsigned int hash,
 519                                        unsigned int reply_hash)
 520 {
 521         struct net *net = nf_ct_net(ct);
 522
 523         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 524                            &net->ct.hash[hash]);
 525         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 526                            &net->ct.hash[reply_hash]);
 527 }
 528
 529 int
 530 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 531 {
 532         const struct nf_conntrack_zone *zone;
 533         struct net *net = nf_ct_net(ct);
 534         unsigned int hash, reply_hash;
 535         struct nf_conntrack_tuple_hash *h;
 536         struct hlist_nulls_node *n;
 537         unsigned int sequence;
 538
 539         zone = nf_ct_zone(ct);
 540
 541         local_bh_disable();
 542         do {
 543                 sequence = read_seqcount_begin(&net->ct.generation);
 544                 hash = hash_conntrack(net,
 545                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 546                 reply_hash = hash_conntrack(net,
 547                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 548         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 549
 550         /* See if there's one in the list already, including reverse */
 551         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 552                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 553                                       &h->tuple) &&
 554                     nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
 555                                      NF_CT_DIRECTION(h)))
 556                         goto out;
 557         hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
 558                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 559                                       &h->tuple) &&
 560                     nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
 561                                      NF_CT_DIRECTION(h)))
 562                         goto out;
 563
 564         add_timer(&ct->timeout);
 565         smp_wmb();
 566         /* The caller holds a reference to this object */
 567         atomic_set(&ct->ct_general.use, 2);
 568         __nf_conntrack_hash_insert(ct, hash, reply_hash);
 569         nf_conntrack_double_unlock(hash, reply_hash);
 570         NF_CT_STAT_INC(net, insert);
 571         local_bh_enable();
 572         return 0;
 573
 574 out:
 575         nf_conntrack_double_unlock(hash, reply_hash);
 576         NF_CT_STAT_INC(net, insert_failed);
 577         local_bh_enable();
 578         return -EEXIST;
 579 }
 580 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 581
 582 /* Confirm a connection given skb; places it in hash table */
 583 int
 584 __nf_conntrack_confirm(struct sk_buff *skb)
 585 {
 586         const struct nf_conntrack_zone *zone;
 587         unsigned int hash, reply_hash;
 588         struct nf_conntrack_tuple_hash *h;
 589         struct nf_conn *ct;
 590         struct nf_conn_help *help;
 591         struct nf_conn_tstamp *tstamp;
 592         struct hlist_nulls_node *n;
 593         enum ip_conntrack_info ctinfo;
 594         struct net *net;
 595         unsigned int sequence;
 596
 597         ct = nf_ct_get(skb, &ctinfo);
 598         net = nf_ct_net(ct);
 599
 600         /* ipt_REJECT uses nf_conntrack_attach to attach related
 601            ICMP/TCP RST packets in other direction.  Actual packet
 602            which created connection will be IP_CT_NEW or for an
 603            expected connection, IP_CT_RELATED. */
 604         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 605                 return NF_ACCEPT;
 606
 607         zone = nf_ct_zone(ct);
 608         local_bh_disable();
 609
 610         do {
 611                 sequence = read_seqcount_begin(&net->ct.generation);
 612                 /* reuse the hash saved before */
 613                 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 614                 hash = hash_bucket(hash, net);
 615                 reply_hash = hash_conntrack(net,
 616                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 617
 618         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 619
 620         /* We're not in hash table, and we refuse to set up related
 621          * connections for unconfirmed conns.  But packet copies and
 622          * REJECT will give spurious warnings here.
 623          */
 624         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 625
 626         /* No external references means no one else could have
 627          * confirmed us.
 628          */
 629         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 630         pr_debug("Confirming conntrack %p\n", ct);
 631         /* We have to check the DYING flag after unlink to prevent
 632          * a race against nf_ct_get_next_corpse() possibly called from
 633          * user context, else we insert an already 'dead' hash, blocking
 634          * further use of that particular connection -JM.
 635          */
 636         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 637
 638         if (unlikely(nf_ct_is_dying(ct)))
 639                 goto out;
 640
 641         /* See if there's one in the list already, including reverse:
 642            NAT could have grabbed it without realizing, since we're
 643            not in the hash.  If there is, we lost race. */
 644         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 645                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 646                                       &h->tuple) &&
 647                     nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
 648                                      NF_CT_DIRECTION(h)))
 649                         goto out;
 650         hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
 651                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 652                                       &h->tuple) &&
 653                     nf_ct_zone_equal(nf_ct_tuplehash_to_ctrack(h), zone,
 654                                      NF_CT_DIRECTION(h)))
 655                         goto out;
 656
 657         /* Timer relative to confirmation time, not original
 658            setting time, otherwise we'd get timer wrap in
 659            weird delay cases. */
 660         ct->timeout.expires += jiffies;
 661         add_timer(&ct->timeout);
 662         atomic_inc(&ct->ct_general.use);
 663         ct->status |= IPS_CONFIRMED;
 664
 665         /* set conntrack timestamp, if enabled. */
 666         tstamp = nf_conn_tstamp_find(ct);
 667         if (tstamp) {
 668                 if (skb->tstamp.tv64 == 0)
 669                         __net_timestamp(skb);
 670
 671                 tstamp->start = ktime_to_ns(skb->tstamp);
 672         }
 673         /* Since the lookup is lockless, hash insertion must be done after
 674          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 675          * guarantee that no other CPU can find the conntrack before the above
 676          * stores are visible.
 677          */
 678         __nf_conntrack_hash_insert(ct, hash, reply_hash);
 679         nf_conntrack_double_unlock(hash, reply_hash);
 680         NF_CT_STAT_INC(net, insert);
 681         local_bh_enable();
 682
 683         help = nfct_help(ct);
 684         if (help && help->helper)
 685                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 686
 687         nf_conntrack_event_cache(master_ct(ct) ?
 688                                  IPCT_RELATED : IPCT_NEW, ct);
 689         return NF_ACCEPT;
 690
 691 out:
 692         nf_ct_add_to_dying_list(ct);
 693         nf_conntrack_double_unlock(hash, reply_hash);
 694         NF_CT_STAT_INC(net, insert_failed);
 695         local_bh_enable();
 696         return NF_DROP;
 697 }
 698 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 699
 700 /* Returns true if a connection correspondings to the tuple (required
 701    for NAT). */
 702 int
 703 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 704                          const struct nf_conn *ignored_conntrack)
 705 {
 706         struct net *net = nf_ct_net(ignored_conntrack);
 707         const struct nf_conntrack_zone *zone;
 708         struct nf_conntrack_tuple_hash *h;
 709         struct hlist_nulls_node *n;
 710         struct nf_conn *ct;
 711         unsigned int hash;
 712
 713         zone = nf_ct_zone(ignored_conntrack);
 714         hash = hash_conntrack(net, tuple);
 715
 716         /* Disable BHs the entire time since we need to disable them at
 717          * least once for the stats anyway.
 718          */
 719         rcu_read_lock_bh();
 720         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 721                 ct = nf_ct_tuplehash_to_ctrack(h);
 722                 if (ct != ignored_conntrack &&
 723                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 724                     nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h))) {
 725                         NF_CT_STAT_INC(net, found);
 726                         rcu_read_unlock_bh();
 727                         return 1;
 728                 }
 729                 NF_CT_STAT_INC(net, searched);
 730         }
 731         rcu_read_unlock_bh();
 732
 733         return 0;
 734 }
 735 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 736
 737 #define NF_CT_EVICTION_RANGE    8
 738
 739 /* There's a small race here where we may free a just-assured
 740    connection.  Too bad: we're in trouble anyway. */
 741 static noinline int early_drop(struct net *net, unsigned int _hash)
 742 {
 743         /* Use oldest entry, which is roughly LRU */
 744         struct nf_conntrack_tuple_hash *h;
 745         struct nf_conn *ct = NULL, *tmp;
 746         struct hlist_nulls_node *n;
 747         unsigned int i = 0, cnt = 0;
 748         int dropped = 0;
 749         unsigned int hash, sequence;
 750         spinlock_t *lockp;
 751
 752         local_bh_disable();
 753 restart:
 754         sequence = read_seqcount_begin(&net->ct.generation);
 755         hash = hash_bucket(_hash, net);
 756         for (; i < net->ct.htable_size; i++) {
 757                 lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS];
 758                 spin_lock(lockp);
 759                 if (read_seqcount_retry(&net->ct.generation, sequence)) {
 760                         spin_unlock(lockp);
 761                         goto restart;
 762                 }
 763                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 764                                          hnnode) {
 765                         tmp = nf_ct_tuplehash_to_ctrack(h);
 766                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status) &&
 767                             !nf_ct_is_dying(tmp) &&
 768                             atomic_inc_not_zero(&tmp->ct_general.use)) {
 769                                 ct = tmp;
 770                                 break;
 771                         }
 772                         cnt++;
 773                 }
 774
 775                 hash = (hash + 1) % net->ct.htable_size;
 776                 spin_unlock(lockp);
 777
 778                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
 779                         break;
 780
 781         }
 782         local_bh_enable();
 783
 784         if (!ct)
 785                 return dropped;
 786
 787         if (del_timer(&ct->timeout)) {
 788                 if (nf_ct_delete(ct, 0, 0)) {
 789                         dropped = 1;
 790                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 791                 }
 792         }
 793         nf_ct_put(ct);
 794         return dropped;
 795 }
 796
 797 void init_nf_conntrack_hash_rnd(void)
 798 {
 799         unsigned int rand;
 800
 801         /*
 802          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 803          * Because there isn't enough entropy when system initializing,
 804          * and we initialize it as late as possible.
 805          */
 806         do {
 807                 get_random_bytes(&rand, sizeof(rand));
 808         } while (!rand);
 809         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 810 }
 811
 812 static struct nf_conn *
 813 __nf_conntrack_alloc(struct net *net,
 814                      const struct nf_conntrack_zone *zone,
 815                      const struct nf_conntrack_tuple *orig,
 816                      const struct nf_conntrack_tuple *repl,
 817                      gfp_t gfp, u32 hash)
 818 {
 819         struct nf_conn *ct;
 820
 821         if (unlikely(!nf_conntrack_hash_rnd)) {
 822                 init_nf_conntrack_hash_rnd();
 823                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 824                 hash = hash_conntrack_raw(orig);
 825         }
 826
 827         /* We don't want any race condition at early drop stage */
 828         atomic_inc(&net->ct.count);
 829
 830         if (nf_conntrack_max &&
 831             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 832                 if (!early_drop(net, hash)) {
 833                         atomic_dec(&net->ct.count);
 834                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
 835                         return ERR_PTR(-ENOMEM);
 836                 }
 837         }
 838
 839         /*
 840          * Do not use kmem_cache_zalloc(), as this cache uses
 841          * SLAB_DESTROY_BY_RCU.
 842          */
 843         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 844         if (ct == NULL)
 845                 goto out;
 846
 847         spin_lock_init(&ct->lock);
 848         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 849         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 850         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 851         /* save hash for reusing when confirming */
 852         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 853         ct->status = 0;
 854         /* Don't set timer yet: wait for confirmation */
 855         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 856         write_pnet(&ct->ct_net, net);
 857         memset(&ct->__nfct_init_offset[0], 0,
 858                offsetof(struct nf_conn, proto) -
 859                offsetof(struct nf_conn, __nfct_init_offset[0]));
 860
 861         if (zone && nf_ct_zone_add(ct, GFP_ATOMIC, zone) < 0)
 862                 goto out_free;
 863
 864         /* Because we use RCU lookups, we set ct_general.use to zero before
 865          * this is inserted in any list.
 866          */
 867         atomic_set(&ct->ct_general.use, 0);
 868         return ct;
 869 out_free:
 870         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 871 out:
 872         atomic_dec(&net->ct.count);
 873         return ERR_PTR(-ENOMEM);
 874 }
 875
 876 struct nf_conn *nf_conntrack_alloc(struct net *net,
 877                                    const struct nf_conntrack_zone *zone,
 878                                    const struct nf_conntrack_tuple *orig,
 879                                    const struct nf_conntrack_tuple *repl,
 880                                    gfp_t gfp)
 881 {
 882         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 883 }
 884 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 885
 886 void nf_conntrack_free(struct nf_conn *ct)
 887 {
 888         struct net *net = nf_ct_net(ct);
 889
 890         /* A freed object has refcnt == 0, that's
 891          * the golden rule for SLAB_DESTROY_BY_RCU
 892          */
 893         NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
 894
 895         nf_ct_ext_destroy(ct);
 896         nf_ct_ext_free(ct);
 897         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 898         smp_mb__before_atomic();
 899         atomic_dec(&net->ct.count);
 900 }
 901 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 902
 903
 904 /* Allocate a new conntrack: we return -ENOMEM if classification
 905    failed due to stress.  Otherwise it really is unclassifiable. */
 906 static struct nf_conntrack_tuple_hash *
 907 init_conntrack(struct net *net, struct nf_conn *tmpl,
 908                const struct nf_conntrack_tuple *tuple,
 909                struct nf_conntrack_l3proto *l3proto,
 910                struct nf_conntrack_l4proto *l4proto,
 911                struct sk_buff *skb,
 912                unsigned int dataoff, u32 hash)
 913 {
 914         struct nf_conn *ct;
 915         struct nf_conn_help *help;
 916         struct nf_conntrack_tuple repl_tuple;
 917         struct nf_conntrack_ecache *ecache;
 918         struct nf_conntrack_expect *exp = NULL;
 919         const struct nf_conntrack_zone *zone;
 920         struct nf_conn_timeout *timeout_ext;
 921         struct nf_conntrack_zone tmp;
 922         unsigned int *timeouts;
 923
 924         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 925                 pr_debug("Can't invert tuple.\n");
 926                 return NULL;
 927         }
 928
 929         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
 930         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 931                                   hash);
 932         if (IS_ERR(ct))
 933                 return (struct nf_conntrack_tuple_hash *)ct;
 934
 935         if (tmpl && nfct_synproxy(tmpl)) {
 936                 nfct_seqadj_ext_add(ct);
 937                 nfct_synproxy_ext_add(ct);
 938         }
 939
 940         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 941         if (timeout_ext)
 942                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 943         else
 944                 timeouts = l4proto->get_timeouts(net);
 945
 946         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 947                 nf_conntrack_free(ct);
 948                 pr_debug("init conntrack: can't track with proto module\n");
 949                 return NULL;
 950         }
 951
 952         if (timeout_ext)
 953                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 954
 955         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 956         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 957         nf_ct_labels_ext_add(ct);
 958
 959         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 960         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 961                                  ecache ? ecache->expmask : 0,
 962                              GFP_ATOMIC);
 963
 964         local_bh_disable();
 965         if (net->ct.expect_count) {
 966                 spin_lock(&nf_conntrack_expect_lock);
 967                 exp = nf_ct_find_expectation(net, zone, tuple);
 968                 if (exp) {
 969                         pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 970                                  ct, exp);
 971                         /* Welcome, Mr. Bond.  We've been expecting you... */
 972                         __set_bit(IPS_EXPECTED_BIT, &ct->status);
 973                         /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
 974                         ct->master = exp->master;
 975                         if (exp->helper) {
 976                                 help = nf_ct_helper_ext_add(ct, exp->helper,
 977                                                             GFP_ATOMIC);
 978                                 if (help)
 979                                         rcu_assign_pointer(help->helper, exp->helper);
 980                         }
 981
 982 #ifdef CONFIG_NF_CONNTRACK_MARK
 983                         ct->mark = exp->master->mark;
 984 #endif
 985 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 986                         ct->secmark = exp->master->secmark;
 987 #endif
 988                         NF_CT_STAT_INC(net, expect_new);
 989                 }
 990                 spin_unlock(&nf_conntrack_expect_lock);
 991         }
 992         if (!exp) {
 993                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 994                 NF_CT_STAT_INC(net, new);
 995         }
 996
 997         /* Now it is inserted into the unconfirmed list, bump refcount */
 998         nf_conntrack_get(&ct->ct_general);
 999         nf_ct_add_to_unconfirmed_list(ct);
1000
1001         local_bh_enable();
1002
1003         if (exp) {
1004                 if (exp->expectfn)
1005                         exp->expectfn(ct, exp);
1006                 nf_ct_expect_put(exp);
1007         }
1008
1009         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1010 }
1011
1012 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1013 static inline struct nf_conn *
1014 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1015                   struct sk_buff *skb,
1016                   unsigned int dataoff,
1017                   u_int16_t l3num,
1018                   u_int8_t protonum,
1019                   struct nf_conntrack_l3proto *l3proto,
1020                   struct nf_conntrack_l4proto *l4proto,
1021                   int *set_reply,
1022                   enum ip_conntrack_info *ctinfo)
1023 {
1024         const struct nf_conntrack_zone *zone;
1025         struct nf_conntrack_tuple tuple;
1026         struct nf_conntrack_tuple_hash *h;
1027         struct nf_conntrack_zone tmp;
1028         struct nf_conn *ct;
1029         u32 hash;
1030
1031         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1032                              dataoff, l3num, protonum, &tuple, l3proto,
1033                              l4proto)) {
1034                 pr_debug("resolve_normal_ct: Can't get tuple\n");
1035                 return NULL;
1036         }
1037
1038         /* look for tuple match */
1039         zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1040         hash = hash_conntrack_raw(&tuple);
1041         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1042         if (!h) {
1043                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1044                                    skb, dataoff, hash);
1045                 if (!h)
1046                         return NULL;
1047                 if (IS_ERR(h))
1048                         return (void *)h;
1049         }
1050         ct = nf_ct_tuplehash_to_ctrack(h);
1051
1052         /* It exists; we have (non-exclusive) reference. */
1053         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1054                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1055                 /* Please set reply bit if this packet OK */
1056                 *set_reply = 1;
1057         } else {
1058                 /* Once we've had two way comms, always ESTABLISHED. */
1059                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1060                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
1061                         *ctinfo = IP_CT_ESTABLISHED;
1062                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1063                         pr_debug("nf_conntrack_in: related packet for %p\n",
1064                                  ct);
1065                         *ctinfo = IP_CT_RELATED;
1066                 } else {
1067                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
1068                         *ctinfo = IP_CT_NEW;
1069                 }
1070                 *set_reply = 0;
1071         }
1072         skb->nfct = &ct->ct_general;
1073         skb->nfctinfo = *ctinfo;
1074         return ct;
1075 }
1076
1077 unsigned int
1078 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1079                 struct sk_buff *skb)
1080 {
1081         struct nf_conn *ct, *tmpl = NULL;
1082         enum ip_conntrack_info ctinfo;
1083         struct nf_conntrack_l3proto *l3proto;
1084         struct nf_conntrack_l4proto *l4proto;
1085         unsigned int *timeouts;
1086         unsigned int dataoff;
1087         u_int8_t protonum;
1088         int set_reply = 0;
1089         int ret;
1090
1091         if (skb->nfct) {
1092                 /* Previously seen (loopback or untracked)?  Ignore. */
1093                 tmpl = (struct nf_conn *)skb->nfct;
1094                 if (!nf_ct_is_template(tmpl)) {
1095                         NF_CT_STAT_INC_ATOMIC(net, ignore);
1096                         return NF_ACCEPT;
1097                 }
1098                 skb->nfct = NULL;
1099         }
1100
1101         /* rcu_read_lock()ed by nf_hook_slow */
1102         l3proto = __nf_ct_l3proto_find(pf);
1103         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1104                                    &dataoff, &protonum);
1105         if (ret <= 0) {
1106                 pr_debug("not prepared to track yet or error occurred\n");
1107                 NF_CT_STAT_INC_ATOMIC(net, error);
1108                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1109                 ret = -ret;
1110                 goto out;
1111         }
1112
1113         l4proto = __nf_ct_l4proto_find(pf, protonum);
1114
1115         /* It may be an special packet, error, unclean...
1116          * inverse of the return code tells to the netfilter
1117          * core what to do with the packet. */
1118         if (l4proto->error != NULL) {
1119                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1120                                      pf, hooknum);
1121                 if (ret <= 0) {
1122                         NF_CT_STAT_INC_ATOMIC(net, error);
1123                         NF_CT_STAT_INC_ATOMIC(net, invalid);
1124                         ret = -ret;
1125                         goto out;
1126                 }
1127                 /* ICMP[v6] protocol trackers may assign one conntrack. */
1128                 if (skb->nfct)
1129                         goto out;
1130         }
1131
1132         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1133                                l3proto, l4proto, &set_reply, &ctinfo);
1134         if (!ct) {
1135                 /* Not valid part of a connection */
1136                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1137                 ret = NF_ACCEPT;
1138                 goto out;
1139         }
1140
1141         if (IS_ERR(ct)) {
1142                 /* Too stressed to deal. */
1143                 NF_CT_STAT_INC_ATOMIC(net, drop);
1144                 ret = NF_DROP;
1145                 goto out;
1146         }
1147
1148         NF_CT_ASSERT(skb->nfct);
1149
1150         /* Decide what timeout policy we want to apply to this flow. */
1151         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1152
1153         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1154         if (ret <= 0) {
1155                 /* Invalid: inverse of the return code tells
1156                  * the netfilter core what to do */
1157                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1158                 nf_conntrack_put(skb->nfct);
1159                 skb->nfct = NULL;
1160                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1161                 if (ret == -NF_DROP)
1162                         NF_CT_STAT_INC_ATOMIC(net, drop);
1163                 ret = -ret;
1164                 goto out;
1165         }
1166
1167         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1168                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1169 out:
1170         if (tmpl) {
1171                 /* Special case: we have to repeat this hook, assign the
1172                  * template again to this packet. We assume that this packet
1173                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1174                 if (ret == NF_REPEAT)
1175                         skb->nfct = (struct nf_conntrack *)tmpl;
1176                 else
1177                         nf_ct_put(tmpl);
1178         }
1179
1180         return ret;
1181 }
1182 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1183
1184 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1185                           const struct nf_conntrack_tuple *orig)
1186 {
1187         bool ret;
1188
1189         rcu_read_lock();
1190         ret = nf_ct_invert_tuple(inverse, orig,
1191                                  __nf_ct_l3proto_find(orig->src.l3num),
1192                                  __nf_ct_l4proto_find(orig->src.l3num,
1193                                                       orig->dst.protonum));
1194         rcu_read_unlock();
1195         return ret;
1196 }
1197 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1198
1199 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1200    implicitly racy: see __nf_conntrack_confirm */
1201 void nf_conntrack_alter_reply(struct nf_conn *ct,
1202                               const struct nf_conntrack_tuple *newreply)
1203 {
1204         struct nf_conn_help *help = nfct_help(ct);
1205
1206         /* Should be unconfirmed, so not in hash table yet */
1207         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1208
1209         pr_debug("Altering reply tuple of %p to ", ct);
1210         nf_ct_dump_tuple(newreply);
1211
1212         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1213         if (ct->master || (help && !hlist_empty(&help->expectations)))
1214                 return;
1215
1216         rcu_read_lock();
1217         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1218         rcu_read_unlock();
1219 }
1220 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1221
1222 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1223 void __nf_ct_refresh_acct(struct nf_conn *ct,
1224                           enum ip_conntrack_info ctinfo,
1225                           const struct sk_buff *skb,
1226                           unsigned long extra_jiffies,
1227                           int do_acct)
1228 {
1229         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1230         NF_CT_ASSERT(skb);
1231
1232         /* Only update if this is not a fixed timeout */
1233         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1234                 goto acct;
1235
1236         /* If not in hash table, timer will not be active yet */
1237         if (!nf_ct_is_confirmed(ct)) {
1238                 ct->timeout.expires = extra_jiffies;
1239         } else {
1240                 unsigned long newtime = jiffies + extra_jiffies;
1241
1242                 /* Only update the timeout if the new timeout is at least
1243                    HZ jiffies from the old timeout. Need del_timer for race
1244                    avoidance (may already be dying). */
1245                 if (newtime - ct->timeout.expires >= HZ)
1246                         mod_timer_pending(&ct->timeout, newtime);
1247         }
1248
1249 acct:
1250         if (do_acct) {
1251                 struct nf_conn_acct *acct;
1252
1253                 acct = nf_conn_acct_find(ct);
1254                 if (acct) {
1255                         struct nf_conn_counter *counter = acct->counter;
1256
1257                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1258                         atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1259                 }
1260         }
1261 }
1262 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1263
1264 bool __nf_ct_kill_acct(struct nf_conn *ct,
1265                        enum ip_conntrack_info ctinfo,
1266                        const struct sk_buff *skb,
1267                        int do_acct)
1268 {
1269         if (do_acct) {
1270                 struct nf_conn_acct *acct;
1271
1272                 acct = nf_conn_acct_find(ct);
1273                 if (acct) {
1274                         struct nf_conn_counter *counter = acct->counter;
1275
1276                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1277                         atomic64_add(skb->len - skb_network_offset(skb),
1278                                      &counter[CTINFO2DIR(ctinfo)].bytes);
1279                 }
1280         }
1281
1282         if (del_timer(&ct->timeout)) {
1283                 ct->timeout.function((unsigned long)ct);
1284                 return true;
1285         }
1286         return false;
1287 }
1288 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1289
1290 #ifdef CONFIG_NF_CONNTRACK_ZONES
1291 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1292         .len    = sizeof(struct nf_conntrack_zone),
1293         .align  = __alignof__(struct nf_conntrack_zone),
1294         .id     = NF_CT_EXT_ZONE,
1295 };
1296 #endif
1297
1298 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1299
1300 #include <linux/netfilter/nfnetlink.h>
1301 #include <linux/netfilter/nfnetlink_conntrack.h>
1302 #include <linux/mutex.h>
1303
1304 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1305  * in ip_conntrack_core, since we don't want the protocols to autoload
1306  * or depend on ctnetlink */
1307 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1308                                const struct nf_conntrack_tuple *tuple)
1309 {
1310         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1311             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1312                 goto nla_put_failure;
1313         return 0;
1314
1315 nla_put_failure:
1316         return -1;
1317 }
1318 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1319
1320 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1321         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1322         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1323 };
1324 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1325
1326 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1327                                struct nf_conntrack_tuple *t)
1328 {
1329         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1330                 return -EINVAL;
1331
1332         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1333         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1334
1335         return 0;
1336 }
1337 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1338
1339 int nf_ct_port_nlattr_tuple_size(void)
1340 {
1341         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1342 }
1343 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1344 #endif
1345
1346 /* Used by ipt_REJECT and ip6t_REJECT. */
1347 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1348 {
1349         struct nf_conn *ct;
1350         enum ip_conntrack_info ctinfo;
1351
1352         /* This ICMP is in reverse direction to the packet which caused it */
1353         ct = nf_ct_get(skb, &ctinfo);
1354         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1355                 ctinfo = IP_CT_RELATED_REPLY;
1356         else
1357                 ctinfo = IP_CT_RELATED;
1358
1359         /* Attach to new skbuff, and increment count */
1360         nskb->nfct = &ct->ct_general;
1361         nskb->nfctinfo = ctinfo;
1362         nf_conntrack_get(nskb->nfct);
1363 }
1364
1365 /* Bring out ya dead! */
1366 static struct nf_conn *
1367 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1368                 void *data, unsigned int *bucket)
1369 {
1370         struct nf_conntrack_tuple_hash *h;
1371         struct nf_conn *ct;
1372         struct hlist_nulls_node *n;
1373         int cpu;
1374         spinlock_t *lockp;
1375
1376         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1377                 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1378                 local_bh_disable();
1379                 spin_lock(lockp);
1380                 if (*bucket < net->ct.htable_size) {
1381                         hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1382                                 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1383                                         continue;
1384                                 ct = nf_ct_tuplehash_to_ctrack(h);
1385                                 if (iter(ct, data))
1386                                         goto found;
1387                         }
1388                 }
1389                 spin_unlock(lockp);
1390                 local_bh_enable();
1391         }
1392
1393         for_each_possible_cpu(cpu) {
1394                 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1395
1396                 spin_lock_bh(&pcpu->lock);
1397                 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1398                         ct = nf_ct_tuplehash_to_ctrack(h);
1399                         if (iter(ct, data))
1400                                 set_bit(IPS_DYING_BIT, &ct->status);
1401                 }
1402                 spin_unlock_bh(&pcpu->lock);
1403         }
1404         return NULL;
1405 found:
1406         atomic_inc(&ct->ct_general.use);
1407         spin_unlock(lockp);
1408         local_bh_enable();
1409         return ct;
1410 }
1411
1412 void nf_ct_iterate_cleanup(struct net *net,
1413                            int (*iter)(struct nf_conn *i, void *data),
1414                            void *data, u32 portid, int report)
1415 {
1416         struct nf_conn *ct;
1417         unsigned int bucket = 0;
1418
1419         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1420                 /* Time to push up daises... */
1421                 if (del_timer(&ct->timeout))
1422                         nf_ct_delete(ct, portid, report);
1423
1424                 /* ... else the timer will get him soon. */
1425
1426                 nf_ct_put(ct);
1427         }
1428 }
1429 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1430
1431 static int kill_all(struct nf_conn *i, void *data)
1432 {
1433         return 1;
1434 }
1435
1436 void nf_ct_free_hashtable(void *hash, unsigned int size)
1437 {
1438         if (is_vmalloc_addr(hash))
1439                 vfree(hash);
1440         else
1441                 free_pages((unsigned long)hash,
1442                            get_order(sizeof(struct hlist_head) * size));
1443 }
1444 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1445
1446 static int untrack_refs(void)
1447 {
1448         int cnt = 0, cpu;
1449
1450         for_each_possible_cpu(cpu) {
1451                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1452
1453                 cnt += atomic_read(&ct->ct_general.use) - 1;
1454         }
1455         return cnt;
1456 }
1457
1458 void nf_conntrack_cleanup_start(void)
1459 {
1460         RCU_INIT_POINTER(ip_ct_attach, NULL);
1461 }
1462
1463 void nf_conntrack_cleanup_end(void)
1464 {
1465         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1466         while (untrack_refs() > 0)
1467                 schedule();
1468
1469 #ifdef CONFIG_NF_CONNTRACK_ZONES
1470         nf_ct_extend_unregister(&nf_ct_zone_extend);
1471 #endif
1472         nf_conntrack_proto_fini();
1473         nf_conntrack_seqadj_fini();
1474         nf_conntrack_labels_fini();
1475         nf_conntrack_helper_fini();
1476         nf_conntrack_timeout_fini();
1477         nf_conntrack_ecache_fini();
1478         nf_conntrack_tstamp_fini();
1479         nf_conntrack_acct_fini();
1480         nf_conntrack_expect_fini();
1481 }
1482
1483 /*
1484  * Mishearing the voices in his head, our hero wonders how he's
1485  * supposed to kill the mall.
1486  */
1487 void nf_conntrack_cleanup_net(struct net *net)
1488 {
1489         LIST_HEAD(single);
1490
1491         list_add(&net->exit_list, &single);
1492         nf_conntrack_cleanup_net_list(&single);
1493 }
1494
1495 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1496 {
1497         int busy;
1498         struct net *net;
1499
1500         /*
1501          * This makes sure all current packets have passed through
1502          *  netfilter framework.  Roll on, two-stage module
1503          *  delete...
1504          */
1505         synchronize_net();
1506 i_see_dead_people:
1507         busy = 0;
1508         list_for_each_entry(net, net_exit_list, exit_list) {
1509                 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1510                 if (atomic_read(&net->ct.count) != 0)
1511                         busy = 1;
1512         }
1513         if (busy) {
1514                 schedule();
1515                 goto i_see_dead_people;
1516         }
1517
1518         list_for_each_entry(net, net_exit_list, exit_list) {
1519                 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1520                 nf_conntrack_proto_pernet_fini(net);
1521                 nf_conntrack_helper_pernet_fini(net);
1522                 nf_conntrack_ecache_pernet_fini(net);
1523                 nf_conntrack_tstamp_pernet_fini(net);
1524                 nf_conntrack_acct_pernet_fini(net);
1525                 nf_conntrack_expect_pernet_fini(net);
1526                 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1527                 kfree(net->ct.slabname);
1528                 free_percpu(net->ct.stat);
1529                 free_percpu(net->ct.pcpu_lists);
1530         }
1531 }
1532
1533 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1534 {
1535         struct hlist_nulls_head *hash;
1536         unsigned int nr_slots, i;
1537         size_t sz;
1538
1539         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1540         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1541         sz = nr_slots * sizeof(struct hlist_nulls_head);
1542         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1543                                         get_order(sz));
1544         if (!hash)
1545                 hash = vzalloc(sz);
1546
1547         if (hash && nulls)
1548                 for (i = 0; i < nr_slots; i++)
1549                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1550
1551         return hash;
1552 }
1553 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1554
1555 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1556 {
1557         int i, bucket, rc;
1558         unsigned int hashsize, old_size;
1559         struct hlist_nulls_head *hash, *old_hash;
1560         struct nf_conntrack_tuple_hash *h;
1561         struct nf_conn *ct;
1562
1563         if (current->nsproxy->net_ns != &init_net)
1564                 return -EOPNOTSUPP;
1565
1566         /* On boot, we can set this without any fancy locking. */
1567         if (!nf_conntrack_htable_size)
1568                 return param_set_uint(val, kp);
1569
1570         rc = kstrtouint(val, 0, &hashsize);
1571         if (rc)
1572                 return rc;
1573         if (!hashsize)
1574                 return -EINVAL;
1575
1576         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1577         if (!hash)
1578                 return -ENOMEM;
1579
1580         local_bh_disable();
1581         nf_conntrack_all_lock();
1582         write_seqcount_begin(&init_net.ct.generation);
1583
1584         /* Lookups in the old hash might happen in parallel, which means we
1585          * might get false negatives during connection lookup. New connections
1586          * created because of a false negative won't make it into the hash
1587          * though since that required taking the locks.
1588          */
1589
1590         for (i = 0; i < init_net.ct.htable_size; i++) {
1591                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1592                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1593                                         struct nf_conntrack_tuple_hash, hnnode);
1594                         ct = nf_ct_tuplehash_to_ctrack(h);
1595                         hlist_nulls_del_rcu(&h->hnnode);
1596                         bucket = __hash_conntrack(&h->tuple, hashsize);
1597                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1598                 }
1599         }
1600         old_size = init_net.ct.htable_size;
1601         old_hash = init_net.ct.hash;
1602
1603         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1604         init_net.ct.hash = hash;
1605
1606         write_seqcount_end(&init_net.ct.generation);
1607         nf_conntrack_all_unlock();
1608         local_bh_enable();
1609
1610         nf_ct_free_hashtable(old_hash, old_size);
1611         return 0;
1612 }
1613 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1614
1615 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1616                   &nf_conntrack_htable_size, 0600);
1617
1618 void nf_ct_untracked_status_or(unsigned long bits)
1619 {
1620         int cpu;
1621
1622         for_each_possible_cpu(cpu)
1623                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1624 }
1625 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1626
1627 int nf_conntrack_init_start(void)
1628 {
1629         int max_factor = 8;
1630         int i, ret, cpu;
1631
1632         for (i = 0; i < CONNTRACK_LOCKS; i++)
1633                 spin_lock_init(&nf_conntrack_locks[i]);
1634
1635         if (!nf_conntrack_htable_size) {
1636                 /* Idea from tcp.c: use 1/16384 of memory.
1637                  * On i386: 32MB machine has 512 buckets.
1638                  * >= 1GB machines have 16384 buckets.
1639                  * >= 4GB machines have 65536 buckets.
1640                  */
1641                 nf_conntrack_htable_size
1642                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1643                            / sizeof(struct hlist_head));
1644                 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1645                         nf_conntrack_htable_size = 65536;
1646                 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1647                         nf_conntrack_htable_size = 16384;
1648                 if (nf_conntrack_htable_size < 32)
1649                         nf_conntrack_htable_size = 32;
1650
1651                 /* Use a max. factor of four by default to get the same max as
1652                  * with the old struct list_heads. When a table size is given
1653                  * we use the old value of 8 to avoid reducing the max.
1654                  * entries. */
1655                 max_factor = 4;
1656         }
1657         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1658
1659         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1660                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1661                nf_conntrack_max);
1662
1663         ret = nf_conntrack_expect_init();
1664         if (ret < 0)
1665                 goto err_expect;
1666
1667         ret = nf_conntrack_acct_init();
1668         if (ret < 0)
1669                 goto err_acct;
1670
1671         ret = nf_conntrack_tstamp_init();
1672         if (ret < 0)
1673                 goto err_tstamp;
1674
1675         ret = nf_conntrack_ecache_init();
1676         if (ret < 0)
1677                 goto err_ecache;
1678
1679         ret = nf_conntrack_timeout_init();
1680         if (ret < 0)
1681                 goto err_timeout;
1682
1683         ret = nf_conntrack_helper_init();
1684         if (ret < 0)
1685                 goto err_helper;
1686
1687         ret = nf_conntrack_labels_init();
1688         if (ret < 0)
1689                 goto err_labels;
1690
1691         ret = nf_conntrack_seqadj_init();
1692         if (ret < 0)
1693                 goto err_seqadj;
1694
1695 #ifdef CONFIG_NF_CONNTRACK_ZONES
1696         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1697         if (ret < 0)
1698                 goto err_extend;
1699 #endif
1700         ret = nf_conntrack_proto_init();
1701         if (ret < 0)
1702                 goto err_proto;
1703
1704         /* Set up fake conntrack: to never be deleted, not in any hashes */
1705         for_each_possible_cpu(cpu) {
1706                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1707                 write_pnet(&ct->ct_net, &init_net);
1708                 atomic_set(&ct->ct_general.use, 1);
1709         }
1710         /*  - and look it like as a confirmed connection */
1711         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1712         return 0;
1713
1714 err_proto:
1715 #ifdef CONFIG_NF_CONNTRACK_ZONES
1716         nf_ct_extend_unregister(&nf_ct_zone_extend);
1717 err_extend:
1718 #endif
1719         nf_conntrack_seqadj_fini();
1720 err_seqadj:
1721         nf_conntrack_labels_fini();
1722 err_labels:
1723         nf_conntrack_helper_fini();
1724 err_helper:
1725         nf_conntrack_timeout_fini();
1726 err_timeout:
1727         nf_conntrack_ecache_fini();
1728 err_ecache:
1729         nf_conntrack_tstamp_fini();
1730 err_tstamp:
1731         nf_conntrack_acct_fini();
1732 err_acct:
1733         nf_conntrack_expect_fini();
1734 err_expect:
1735         return ret;
1736 }
1737
1738 void nf_conntrack_init_end(void)
1739 {
1740         /* For use by REJECT target */
1741         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1742         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1743 }
1744
1745 /*
1746  * We need to use special "null" values, not used in hash table
1747  */
1748 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1749 #define DYING_NULLS_VAL         ((1<<30)+1)
1750 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1751
1752 int nf_conntrack_init_net(struct net *net)
1753 {
1754         int ret = -ENOMEM;
1755         int cpu;
1756
1757         atomic_set(&net->ct.count, 0);
1758         seqcount_init(&net->ct.generation);
1759
1760         net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1761         if (!net->ct.pcpu_lists)
1762                 goto err_stat;
1763
1764         for_each_possible_cpu(cpu) {
1765                 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1766
1767                 spin_lock_init(&pcpu->lock);
1768                 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1769                 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1770         }
1771
1772         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1773         if (!net->ct.stat)
1774                 goto err_pcpu_lists;
1775
1776         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1777         if (!net->ct.slabname)
1778                 goto err_slabname;
1779
1780         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1781                                                         sizeof(struct nf_conn), 0,
1782                                                         SLAB_DESTROY_BY_RCU, NULL);
1783         if (!net->ct.nf_conntrack_cachep) {
1784                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1785                 goto err_cache;
1786         }
1787
1788         net->ct.htable_size = nf_conntrack_htable_size;
1789         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1790         if (!net->ct.hash) {
1791                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1792                 goto err_hash;
1793         }
1794         ret = nf_conntrack_expect_pernet_init(net);
1795         if (ret < 0)
1796                 goto err_expect;
1797         ret = nf_conntrack_acct_pernet_init(net);
1798         if (ret < 0)
1799                 goto err_acct;
1800         ret = nf_conntrack_tstamp_pernet_init(net);
1801         if (ret < 0)
1802                 goto err_tstamp;
1803         ret = nf_conntrack_ecache_pernet_init(net);
1804         if (ret < 0)
1805                 goto err_ecache;
1806         ret = nf_conntrack_helper_pernet_init(net);
1807         if (ret < 0)
1808                 goto err_helper;
1809         ret = nf_conntrack_proto_pernet_init(net);
1810         if (ret < 0)
1811                 goto err_proto;
1812         return 0;
1813
1814 err_proto:
1815         nf_conntrack_helper_pernet_fini(net);
1816 err_helper:
1817         nf_conntrack_ecache_pernet_fini(net);
1818 err_ecache:
1819         nf_conntrack_tstamp_pernet_fini(net);
1820 err_tstamp:
1821         nf_conntrack_acct_pernet_fini(net);
1822 err_acct:
1823         nf_conntrack_expect_pernet_fini(net);
1824 err_expect:
1825         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1826 err_hash:
1827         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1828 err_cache:
1829         kfree(net->ct.slabname);
1830 err_slabname:
1831         free_percpu(net->ct.stat);
1832 err_pcpu_lists:
1833         free_percpu(net->ct.pcpu_lists);
1834 err_stat:
1835         return ret;
1836 }