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, u16 zone)
 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, zone ^ 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                                   u16 zone, unsigned int size)
 155 {
 156         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 157 }
 158
 159 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 160                                        const struct nf_conntrack_tuple *tuple)
 161 {
 162         return __hash_conntrack(tuple, zone, 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 static void
 291 destroy_conntrack(struct nf_conntrack *nfct)
 292 {
 293         struct nf_conn *ct = (struct nf_conn *)nfct;
 294         struct net *net = nf_ct_net(ct);
 295         struct nf_conntrack_l4proto *l4proto;
 296
 297         pr_debug("destroy_conntrack(%p)\n", ct);
 298         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 299         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 300
 301         rcu_read_lock();
 302         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 303         if (l4proto && l4proto->destroy)
 304                 l4proto->destroy(ct);
 305
 306         rcu_read_unlock();
 307
 308         local_bh_disable();
 309         /* Expectations will have been removed in clean_from_lists,
 310          * except TFTP can create an expectation on the first packet,
 311          * before connection is in the list, so we need to clean here,
 312          * too.
 313          */
 314         nf_ct_remove_expectations(ct);
 315
 316         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 317
 318         NF_CT_STAT_INC(net, delete);
 319         local_bh_enable();
 320
 321         if (ct->master)
 322                 nf_ct_put(ct->master);
 323
 324         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 325         nf_conntrack_free(ct);
 326 }
 327
 328 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 329 {
 330         struct net *net = nf_ct_net(ct);
 331         unsigned int hash, reply_hash;
 332         u16 zone = nf_ct_zone(ct);
 333         unsigned int sequence;
 334
 335         nf_ct_helper_destroy(ct);
 336
 337         local_bh_disable();
 338         do {
 339                 sequence = read_seqcount_begin(&net->ct.generation);
 340                 hash = hash_conntrack(net, zone,
 341                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 342                 reply_hash = hash_conntrack(net, zone,
 343                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 344         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 345
 346         clean_from_lists(ct);
 347         nf_conntrack_double_unlock(hash, reply_hash);
 348
 349         nf_ct_add_to_dying_list(ct);
 350
 351         NF_CT_STAT_INC(net, delete_list);
 352         local_bh_enable();
 353 }
 354
 355 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 356 {
 357         struct nf_conn_tstamp *tstamp;
 358
 359         tstamp = nf_conn_tstamp_find(ct);
 360         if (tstamp && tstamp->stop == 0)
 361                 tstamp->stop = ktime_get_real_ns();
 362
 363         if (nf_ct_is_dying(ct))
 364                 goto delete;
 365
 366         if (nf_conntrack_event_report(IPCT_DESTROY, ct,
 367                                     portid, report) < 0) {
 368                 /* destroy event was not delivered */
 369                 nf_ct_delete_from_lists(ct);
 370                 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
 371                 return false;
 372         }
 373
 374         nf_conntrack_ecache_work(nf_ct_net(ct));
 375         set_bit(IPS_DYING_BIT, &ct->status);
 376  delete:
 377         nf_ct_delete_from_lists(ct);
 378         nf_ct_put(ct);
 379         return true;
 380 }
 381 EXPORT_SYMBOL_GPL(nf_ct_delete);
 382
 383 static void death_by_timeout(unsigned long ul_conntrack)
 384 {
 385         nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
 386 }
 387
 388 static inline bool
 389 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
 390                         const struct nf_conntrack_tuple *tuple,
 391                         u16 zone)
 392 {
 393         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 394
 395         /* A conntrack can be recreated with the equal tuple,
 396          * so we need to check that the conntrack is confirmed
 397          */
 398         return nf_ct_tuple_equal(tuple, &h->tuple) &&
 399                 nf_ct_zone(ct) == zone &&
 400                 nf_ct_is_confirmed(ct);
 401 }
 402
 403 /*
 404  * Warning :
 405  * - Caller must take a reference on returned object
 406  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 407  */
 408 static struct nf_conntrack_tuple_hash *
 409 ____nf_conntrack_find(struct net *net, u16 zone,
 410                       const struct nf_conntrack_tuple *tuple, u32 hash)
 411 {
 412         struct nf_conntrack_tuple_hash *h;
 413         struct hlist_nulls_node *n;
 414         unsigned int bucket = hash_bucket(hash, net);
 415
 416         /* Disable BHs the entire time since we normally need to disable them
 417          * at least once for the stats anyway.
 418          */
 419         local_bh_disable();
 420 begin:
 421         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 422                 if (nf_ct_key_equal(h, tuple, zone)) {
 423                         NF_CT_STAT_INC(net, found);
 424                         local_bh_enable();
 425                         return h;
 426                 }
 427                 NF_CT_STAT_INC(net, searched);
 428         }
 429         /*
 430          * if the nulls value we got at the end of this lookup is
 431          * not the expected one, we must restart lookup.
 432          * We probably met an item that was moved to another chain.
 433          */
 434         if (get_nulls_value(n) != bucket) {
 435                 NF_CT_STAT_INC(net, search_restart);
 436                 goto begin;
 437         }
 438         local_bh_enable();
 439
 440         return NULL;
 441 }
 442
 443 /* Find a connection corresponding to a tuple. */
 444 static struct nf_conntrack_tuple_hash *
 445 __nf_conntrack_find_get(struct net *net, u16 zone,
 446                         const struct nf_conntrack_tuple *tuple, u32 hash)
 447 {
 448         struct nf_conntrack_tuple_hash *h;
 449         struct nf_conn *ct;
 450
 451         rcu_read_lock();
 452 begin:
 453         h = ____nf_conntrack_find(net, zone, tuple, hash);
 454         if (h) {
 455                 ct = nf_ct_tuplehash_to_ctrack(h);
 456                 if (unlikely(nf_ct_is_dying(ct) ||
 457                              !atomic_inc_not_zero(&ct->ct_general.use)))
 458                         h = NULL;
 459                 else {
 460                         if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
 461                                 nf_ct_put(ct);
 462                                 goto begin;
 463                         }
 464                 }
 465         }
 466         rcu_read_unlock();
 467
 468         return h;
 469 }
 470
 471 struct nf_conntrack_tuple_hash *
 472 nf_conntrack_find_get(struct net *net, u16 zone,
 473                       const struct nf_conntrack_tuple *tuple)
 474 {
 475         return __nf_conntrack_find_get(net, zone, tuple,
 476                                        hash_conntrack_raw(tuple, zone));
 477 }
 478 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 479
 480 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 481                                        unsigned int hash,
 482                                        unsigned int reply_hash)
 483 {
 484         struct net *net = nf_ct_net(ct);
 485
 486         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 487                            &net->ct.hash[hash]);
 488         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 489                            &net->ct.hash[reply_hash]);
 490 }
 491
 492 int
 493 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 494 {
 495         struct net *net = nf_ct_net(ct);
 496         unsigned int hash, reply_hash;
 497         struct nf_conntrack_tuple_hash *h;
 498         struct hlist_nulls_node *n;
 499         u16 zone;
 500         unsigned int sequence;
 501
 502         zone = nf_ct_zone(ct);
 503
 504         local_bh_disable();
 505         do {
 506                 sequence = read_seqcount_begin(&net->ct.generation);
 507                 hash = hash_conntrack(net, zone,
 508                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 509                 reply_hash = hash_conntrack(net, zone,
 510                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 511         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 512
 513         /* See if there's one in the list already, including reverse */
 514         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 515                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 516                                       &h->tuple) &&
 517                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 518                         goto out;
 519         hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
 520                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 521                                       &h->tuple) &&
 522                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 523                         goto out;
 524
 525         add_timer(&ct->timeout);
 526         smp_wmb();
 527         /* The caller holds a reference to this object */
 528         atomic_set(&ct->ct_general.use, 2);
 529         __nf_conntrack_hash_insert(ct, hash, reply_hash);
 530         nf_conntrack_double_unlock(hash, reply_hash);
 531         NF_CT_STAT_INC(net, insert);
 532         local_bh_enable();
 533         return 0;
 534
 535 out:
 536         nf_conntrack_double_unlock(hash, reply_hash);
 537         NF_CT_STAT_INC(net, insert_failed);
 538         local_bh_enable();
 539         return -EEXIST;
 540 }
 541 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 542
 543 /* deletion from this larval template list happens via nf_ct_put() */
 544 void nf_conntrack_tmpl_insert(struct net *net, struct nf_conn *tmpl)
 545 {
 546         struct ct_pcpu *pcpu;
 547
 548         __set_bit(IPS_TEMPLATE_BIT, &tmpl->status);
 549         __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
 550         nf_conntrack_get(&tmpl->ct_general);
 551
 552         /* add this conntrack to the (per cpu) tmpl list */
 553         local_bh_disable();
 554         tmpl->cpu = smp_processor_id();
 555         pcpu = per_cpu_ptr(nf_ct_net(tmpl)->ct.pcpu_lists, tmpl->cpu);
 556
 557         spin_lock(&pcpu->lock);
 558         /* Overload tuple linked list to put us in template list. */
 559         hlist_nulls_add_head_rcu(&tmpl->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 560                                  &pcpu->tmpl);
 561         spin_unlock_bh(&pcpu->lock);
 562 }
 563 EXPORT_SYMBOL_GPL(nf_conntrack_tmpl_insert);
 564
 565 /* Confirm a connection given skb; places it in hash table */
 566 int
 567 __nf_conntrack_confirm(struct sk_buff *skb)
 568 {
 569         unsigned int hash, reply_hash;
 570         struct nf_conntrack_tuple_hash *h;
 571         struct nf_conn *ct;
 572         struct nf_conn_help *help;
 573         struct nf_conn_tstamp *tstamp;
 574         struct hlist_nulls_node *n;
 575         enum ip_conntrack_info ctinfo;
 576         struct net *net;
 577         u16 zone;
 578         unsigned int sequence;
 579
 580         ct = nf_ct_get(skb, &ctinfo);
 581         net = nf_ct_net(ct);
 582
 583         /* ipt_REJECT uses nf_conntrack_attach to attach related
 584            ICMP/TCP RST packets in other direction.  Actual packet
 585            which created connection will be IP_CT_NEW or for an
 586            expected connection, IP_CT_RELATED. */
 587         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 588                 return NF_ACCEPT;
 589
 590         zone = nf_ct_zone(ct);
 591         local_bh_disable();
 592
 593         do {
 594                 sequence = read_seqcount_begin(&net->ct.generation);
 595                 /* reuse the hash saved before */
 596                 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 597                 hash = hash_bucket(hash, net);
 598                 reply_hash = hash_conntrack(net, zone,
 599                                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 600
 601         } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
 602
 603         /* We're not in hash table, and we refuse to set up related
 604          * connections for unconfirmed conns.  But packet copies and
 605          * REJECT will give spurious warnings here.
 606          */
 607         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 608
 609         /* No external references means no one else could have
 610          * confirmed us.
 611          */
 612         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 613         pr_debug("Confirming conntrack %p\n", ct);
 614         /* We have to check the DYING flag inside the lock to prevent
 615            a race against nf_ct_get_next_corpse() possibly called from
 616            user context, else we insert an already 'dead' hash, blocking
 617            further use of that particular connection -JM */
 618
 619         if (unlikely(nf_ct_is_dying(ct))) {
 620                 nf_conntrack_double_unlock(hash, reply_hash);
 621                 local_bh_enable();
 622                 return NF_ACCEPT;
 623         }
 624
 625         /* See if there's one in the list already, including reverse:
 626            NAT could have grabbed it without realizing, since we're
 627            not in the hash.  If there is, we lost race. */
 628         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 629                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 630                                       &h->tuple) &&
 631                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 632                         goto out;
 633         hlist_nulls_for_each_entry(h, n, &net->ct.hash[reply_hash], hnnode)
 634                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 635                                       &h->tuple) &&
 636                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 637                         goto out;
 638
 639         nf_ct_del_from_dying_or_unconfirmed_list(ct);
 640
 641         /* Timer relative to confirmation time, not original
 642            setting time, otherwise we'd get timer wrap in
 643            weird delay cases. */
 644         ct->timeout.expires += jiffies;
 645         add_timer(&ct->timeout);
 646         atomic_inc(&ct->ct_general.use);
 647         ct->status |= IPS_CONFIRMED;
 648
 649         /* set conntrack timestamp, if enabled. */
 650         tstamp = nf_conn_tstamp_find(ct);
 651         if (tstamp) {
 652                 if (skb->tstamp.tv64 == 0)
 653                         __net_timestamp(skb);
 654
 655                 tstamp->start = ktime_to_ns(skb->tstamp);
 656         }
 657         /* Since the lookup is lockless, hash insertion must be done after
 658          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 659          * guarantee that no other CPU can find the conntrack before the above
 660          * stores are visible.
 661          */
 662         __nf_conntrack_hash_insert(ct, hash, reply_hash);
 663         nf_conntrack_double_unlock(hash, reply_hash);
 664         NF_CT_STAT_INC(net, insert);
 665         local_bh_enable();
 666
 667         help = nfct_help(ct);
 668         if (help && help->helper)
 669                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 670
 671         nf_conntrack_event_cache(master_ct(ct) ?
 672                                  IPCT_RELATED : IPCT_NEW, ct);
 673         return NF_ACCEPT;
 674
 675 out:
 676         nf_conntrack_double_unlock(hash, reply_hash);
 677         NF_CT_STAT_INC(net, insert_failed);
 678         local_bh_enable();
 679         return NF_DROP;
 680 }
 681 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 682
 683 /* Returns true if a connection correspondings to the tuple (required
 684    for NAT). */
 685 int
 686 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 687                          const struct nf_conn *ignored_conntrack)
 688 {
 689         struct net *net = nf_ct_net(ignored_conntrack);
 690         struct nf_conntrack_tuple_hash *h;
 691         struct hlist_nulls_node *n;
 692         struct nf_conn *ct;
 693         u16 zone = nf_ct_zone(ignored_conntrack);
 694         unsigned int hash = hash_conntrack(net, zone, tuple);
 695
 696         /* Disable BHs the entire time since we need to disable them at
 697          * least once for the stats anyway.
 698          */
 699         rcu_read_lock_bh();
 700         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 701                 ct = nf_ct_tuplehash_to_ctrack(h);
 702                 if (ct != ignored_conntrack &&
 703                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 704                     nf_ct_zone(ct) == zone) {
 705                         NF_CT_STAT_INC(net, found);
 706                         rcu_read_unlock_bh();
 707                         return 1;
 708                 }
 709                 NF_CT_STAT_INC(net, searched);
 710         }
 711         rcu_read_unlock_bh();
 712
 713         return 0;
 714 }
 715 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 716
 717 #define NF_CT_EVICTION_RANGE    8
 718
 719 /* There's a small race here where we may free a just-assured
 720    connection.  Too bad: we're in trouble anyway. */
 721 static noinline int early_drop(struct net *net, unsigned int _hash)
 722 {
 723         /* Use oldest entry, which is roughly LRU */
 724         struct nf_conntrack_tuple_hash *h;
 725         struct nf_conn *ct = NULL, *tmp;
 726         struct hlist_nulls_node *n;
 727         unsigned int i = 0, cnt = 0;
 728         int dropped = 0;
 729         unsigned int hash, sequence;
 730         spinlock_t *lockp;
 731
 732         local_bh_disable();
 733 restart:
 734         sequence = read_seqcount_begin(&net->ct.generation);
 735         hash = hash_bucket(_hash, net);
 736         for (; i < net->ct.htable_size; i++) {
 737                 lockp = &nf_conntrack_locks[hash % CONNTRACK_LOCKS];
 738                 spin_lock(lockp);
 739                 if (read_seqcount_retry(&net->ct.generation, sequence)) {
 740                         spin_unlock(lockp);
 741                         goto restart;
 742                 }
 743                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 744                                          hnnode) {
 745                         tmp = nf_ct_tuplehash_to_ctrack(h);
 746                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status) &&
 747                             !nf_ct_is_dying(tmp) &&
 748                             atomic_inc_not_zero(&tmp->ct_general.use)) {
 749                                 ct = tmp;
 750                                 break;
 751                         }
 752                         cnt++;
 753                 }
 754
 755                 hash = (hash + 1) % net->ct.htable_size;
 756                 spin_unlock(lockp);
 757
 758                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
 759                         break;
 760
 761         }
 762         local_bh_enable();
 763
 764         if (!ct)
 765                 return dropped;
 766
 767         if (del_timer(&ct->timeout)) {
 768                 if (nf_ct_delete(ct, 0, 0)) {
 769                         dropped = 1;
 770                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 771                 }
 772         }
 773         nf_ct_put(ct);
 774         return dropped;
 775 }
 776
 777 void init_nf_conntrack_hash_rnd(void)
 778 {
 779         unsigned int rand;
 780
 781         /*
 782          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 783          * Because there isn't enough entropy when system initializing,
 784          * and we initialize it as late as possible.
 785          */
 786         do {
 787                 get_random_bytes(&rand, sizeof(rand));
 788         } while (!rand);
 789         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 790 }
 791
 792 static struct nf_conn *
 793 __nf_conntrack_alloc(struct net *net, u16 zone,
 794                      const struct nf_conntrack_tuple *orig,
 795                      const struct nf_conntrack_tuple *repl,
 796                      gfp_t gfp, u32 hash)
 797 {
 798         struct nf_conn *ct;
 799
 800         if (unlikely(!nf_conntrack_hash_rnd)) {
 801                 init_nf_conntrack_hash_rnd();
 802                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 803                 hash = hash_conntrack_raw(orig, zone);
 804         }
 805
 806         /* We don't want any race condition at early drop stage */
 807         atomic_inc(&net->ct.count);
 808
 809         if (nf_conntrack_max &&
 810             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 811                 if (!early_drop(net, hash)) {
 812                         atomic_dec(&net->ct.count);
 813                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
 814                         return ERR_PTR(-ENOMEM);
 815                 }
 816         }
 817
 818         /*
 819          * Do not use kmem_cache_zalloc(), as this cache uses
 820          * SLAB_DESTROY_BY_RCU.
 821          */
 822         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 823         if (ct == NULL) {
 824                 atomic_dec(&net->ct.count);
 825                 return ERR_PTR(-ENOMEM);
 826         }
 827         spin_lock_init(&ct->lock);
 828         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 829         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 830         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 831         /* save hash for reusing when confirming */
 832         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 833         ct->status = 0;
 834         /* Don't set timer yet: wait for confirmation */
 835         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 836         write_pnet(&ct->ct_net, net);
 837         memset(&ct->__nfct_init_offset[0], 0,
 838                offsetof(struct nf_conn, proto) -
 839                offsetof(struct nf_conn, __nfct_init_offset[0]));
 840 #ifdef CONFIG_NF_CONNTRACK_ZONES
 841         if (zone) {
 842                 struct nf_conntrack_zone *nf_ct_zone;
 843
 844                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 845                 if (!nf_ct_zone)
 846                         goto out_free;
 847                 nf_ct_zone->id = zone;
 848         }
 849 #endif
 850         /* Because we use RCU lookups, we set ct_general.use to zero before
 851          * this is inserted in any list.
 852          */
 853         atomic_set(&ct->ct_general.use, 0);
 854         return ct;
 855
 856 #ifdef CONFIG_NF_CONNTRACK_ZONES
 857 out_free:
 858         atomic_dec(&net->ct.count);
 859         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 860         return ERR_PTR(-ENOMEM);
 861 #endif
 862 }
 863
 864 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 865                                    const struct nf_conntrack_tuple *orig,
 866                                    const struct nf_conntrack_tuple *repl,
 867                                    gfp_t gfp)
 868 {
 869         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 870 }
 871 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 872
 873 void nf_conntrack_free(struct nf_conn *ct)
 874 {
 875         struct net *net = nf_ct_net(ct);
 876
 877         /* A freed object has refcnt == 0, that's
 878          * the golden rule for SLAB_DESTROY_BY_RCU
 879          */
 880         NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
 881
 882         nf_ct_ext_destroy(ct);
 883         nf_ct_ext_free(ct);
 884         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 885         smp_mb__before_atomic();
 886         atomic_dec(&net->ct.count);
 887 }
 888 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 889
 890
 891 /* Allocate a new conntrack: we return -ENOMEM if classification
 892    failed due to stress.  Otherwise it really is unclassifiable. */
 893 static struct nf_conntrack_tuple_hash *
 894 init_conntrack(struct net *net, struct nf_conn *tmpl,
 895                const struct nf_conntrack_tuple *tuple,
 896                struct nf_conntrack_l3proto *l3proto,
 897                struct nf_conntrack_l4proto *l4proto,
 898                struct sk_buff *skb,
 899                unsigned int dataoff, u32 hash)
 900 {
 901         struct nf_conn *ct;
 902         struct nf_conn_help *help;
 903         struct nf_conntrack_tuple repl_tuple;
 904         struct nf_conntrack_ecache *ecache;
 905         struct nf_conntrack_expect *exp = NULL;
 906         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 907         struct nf_conn_timeout *timeout_ext;
 908         unsigned int *timeouts;
 909
 910         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 911                 pr_debug("Can't invert tuple.\n");
 912                 return NULL;
 913         }
 914
 915         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 916                                   hash);
 917         if (IS_ERR(ct))
 918                 return (struct nf_conntrack_tuple_hash *)ct;
 919
 920         if (tmpl && nfct_synproxy(tmpl)) {
 921                 nfct_seqadj_ext_add(ct);
 922                 nfct_synproxy_ext_add(ct);
 923         }
 924
 925         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 926         if (timeout_ext)
 927                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 928         else
 929                 timeouts = l4proto->get_timeouts(net);
 930
 931         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 932                 nf_conntrack_free(ct);
 933                 pr_debug("init conntrack: can't track with proto module\n");
 934                 return NULL;
 935         }
 936
 937         if (timeout_ext)
 938                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 939
 940         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 941         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 942         nf_ct_labels_ext_add(ct);
 943
 944         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 945         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 946                                  ecache ? ecache->expmask : 0,
 947                              GFP_ATOMIC);
 948
 949         local_bh_disable();
 950         if (net->ct.expect_count) {
 951                 spin_lock(&nf_conntrack_expect_lock);
 952                 exp = nf_ct_find_expectation(net, zone, tuple);
 953                 if (exp) {
 954                         pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 955                                  ct, exp);
 956                         /* Welcome, Mr. Bond.  We've been expecting you... */
 957                         __set_bit(IPS_EXPECTED_BIT, &ct->status);
 958                         /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
 959                         ct->master = exp->master;
 960                         if (exp->helper) {
 961                                 help = nf_ct_helper_ext_add(ct, exp->helper,
 962                                                             GFP_ATOMIC);
 963                                 if (help)
 964                                         rcu_assign_pointer(help->helper, exp->helper);
 965                         }
 966
 967 #ifdef CONFIG_NF_CONNTRACK_MARK
 968                         ct->mark = exp->master->mark;
 969 #endif
 970 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 971                         ct->secmark = exp->master->secmark;
 972 #endif
 973                         NF_CT_STAT_INC(net, expect_new);
 974                 }
 975                 spin_unlock(&nf_conntrack_expect_lock);
 976         }
 977         if (!exp) {
 978                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 979                 NF_CT_STAT_INC(net, new);
 980         }
 981
 982         /* Now it is inserted into the unconfirmed list, bump refcount */
 983         nf_conntrack_get(&ct->ct_general);
 984         nf_ct_add_to_unconfirmed_list(ct);
 985
 986         local_bh_enable();
 987
 988         if (exp) {
 989                 if (exp->expectfn)
 990                         exp->expectfn(ct, exp);
 991                 nf_ct_expect_put(exp);
 992         }
 993
 994         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 995 }
 996
 997 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 998 static inline struct nf_conn *
 999 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1000                   struct sk_buff *skb,
1001                   unsigned int dataoff,
1002                   u_int16_t l3num,
1003                   u_int8_t protonum,
1004                   struct nf_conntrack_l3proto *l3proto,
1005                   struct nf_conntrack_l4proto *l4proto,
1006                   int *set_reply,
1007                   enum ip_conntrack_info *ctinfo)
1008 {
1009         struct nf_conntrack_tuple tuple;
1010         struct nf_conntrack_tuple_hash *h;
1011         struct nf_conn *ct;
1012         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
1013         u32 hash;
1014
1015         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1016                              dataoff, l3num, protonum, &tuple, l3proto,
1017                              l4proto)) {
1018                 pr_debug("resolve_normal_ct: Can't get tuple\n");
1019                 return NULL;
1020         }
1021
1022         /* look for tuple match */
1023         hash = hash_conntrack_raw(&tuple, zone);
1024         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1025         if (!h) {
1026                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1027                                    skb, dataoff, hash);
1028                 if (!h)
1029                         return NULL;
1030                 if (IS_ERR(h))
1031                         return (void *)h;
1032         }
1033         ct = nf_ct_tuplehash_to_ctrack(h);
1034
1035         /* It exists; we have (non-exclusive) reference. */
1036         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1037                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1038                 /* Please set reply bit if this packet OK */
1039                 *set_reply = 1;
1040         } else {
1041                 /* Once we've had two way comms, always ESTABLISHED. */
1042                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1043                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
1044                         *ctinfo = IP_CT_ESTABLISHED;
1045                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1046                         pr_debug("nf_conntrack_in: related packet for %p\n",
1047                                  ct);
1048                         *ctinfo = IP_CT_RELATED;
1049                 } else {
1050                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
1051                         *ctinfo = IP_CT_NEW;
1052                 }
1053                 *set_reply = 0;
1054         }
1055         skb->nfct = &ct->ct_general;
1056         skb->nfctinfo = *ctinfo;
1057         return ct;
1058 }
1059
1060 unsigned int
1061 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1062                 struct sk_buff *skb)
1063 {
1064         struct nf_conn *ct, *tmpl = NULL;
1065         enum ip_conntrack_info ctinfo;
1066         struct nf_conntrack_l3proto *l3proto;
1067         struct nf_conntrack_l4proto *l4proto;
1068         unsigned int *timeouts;
1069         unsigned int dataoff;
1070         u_int8_t protonum;
1071         int set_reply = 0;
1072         int ret;
1073
1074         if (skb->nfct) {
1075                 /* Previously seen (loopback or untracked)?  Ignore. */
1076                 tmpl = (struct nf_conn *)skb->nfct;
1077                 if (!nf_ct_is_template(tmpl)) {
1078                         NF_CT_STAT_INC_ATOMIC(net, ignore);
1079                         return NF_ACCEPT;
1080                 }
1081                 skb->nfct = NULL;
1082         }
1083
1084         /* rcu_read_lock()ed by nf_hook_slow */
1085         l3proto = __nf_ct_l3proto_find(pf);
1086         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1087                                    &dataoff, &protonum);
1088         if (ret <= 0) {
1089                 pr_debug("not prepared to track yet or error occurred\n");
1090                 NF_CT_STAT_INC_ATOMIC(net, error);
1091                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1092                 ret = -ret;
1093                 goto out;
1094         }
1095
1096         l4proto = __nf_ct_l4proto_find(pf, protonum);
1097
1098         /* It may be an special packet, error, unclean...
1099          * inverse of the return code tells to the netfilter
1100          * core what to do with the packet. */
1101         if (l4proto->error != NULL) {
1102                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1103                                      pf, hooknum);
1104                 if (ret <= 0) {
1105                         NF_CT_STAT_INC_ATOMIC(net, error);
1106                         NF_CT_STAT_INC_ATOMIC(net, invalid);
1107                         ret = -ret;
1108                         goto out;
1109                 }
1110                 /* ICMP[v6] protocol trackers may assign one conntrack. */
1111                 if (skb->nfct)
1112                         goto out;
1113         }
1114
1115         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1116                                l3proto, l4proto, &set_reply, &ctinfo);
1117         if (!ct) {
1118                 /* Not valid part of a connection */
1119                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1120                 ret = NF_ACCEPT;
1121                 goto out;
1122         }
1123
1124         if (IS_ERR(ct)) {
1125                 /* Too stressed to deal. */
1126                 NF_CT_STAT_INC_ATOMIC(net, drop);
1127                 ret = NF_DROP;
1128                 goto out;
1129         }
1130
1131         NF_CT_ASSERT(skb->nfct);
1132
1133         /* Decide what timeout policy we want to apply to this flow. */
1134         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1135
1136         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1137         if (ret <= 0) {
1138                 /* Invalid: inverse of the return code tells
1139                  * the netfilter core what to do */
1140                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1141                 nf_conntrack_put(skb->nfct);
1142                 skb->nfct = NULL;
1143                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1144                 if (ret == -NF_DROP)
1145                         NF_CT_STAT_INC_ATOMIC(net, drop);
1146                 ret = -ret;
1147                 goto out;
1148         }
1149
1150         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1151                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1152 out:
1153         if (tmpl) {
1154                 /* Special case: we have to repeat this hook, assign the
1155                  * template again to this packet. We assume that this packet
1156                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1157                 if (ret == NF_REPEAT)
1158                         skb->nfct = (struct nf_conntrack *)tmpl;
1159                 else
1160                         nf_ct_put(tmpl);
1161         }
1162
1163         return ret;
1164 }
1165 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1166
1167 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1168                           const struct nf_conntrack_tuple *orig)
1169 {
1170         bool ret;
1171
1172         rcu_read_lock();
1173         ret = nf_ct_invert_tuple(inverse, orig,
1174                                  __nf_ct_l3proto_find(orig->src.l3num),
1175                                  __nf_ct_l4proto_find(orig->src.l3num,
1176                                                       orig->dst.protonum));
1177         rcu_read_unlock();
1178         return ret;
1179 }
1180 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1181
1182 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1183    implicitly racy: see __nf_conntrack_confirm */
1184 void nf_conntrack_alter_reply(struct nf_conn *ct,
1185                               const struct nf_conntrack_tuple *newreply)
1186 {
1187         struct nf_conn_help *help = nfct_help(ct);
1188
1189         /* Should be unconfirmed, so not in hash table yet */
1190         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1191
1192         pr_debug("Altering reply tuple of %p to ", ct);
1193         nf_ct_dump_tuple(newreply);
1194
1195         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1196         if (ct->master || (help && !hlist_empty(&help->expectations)))
1197                 return;
1198
1199         rcu_read_lock();
1200         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1201         rcu_read_unlock();
1202 }
1203 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1204
1205 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1206 void __nf_ct_refresh_acct(struct nf_conn *ct,
1207                           enum ip_conntrack_info ctinfo,
1208                           const struct sk_buff *skb,
1209                           unsigned long extra_jiffies,
1210                           int do_acct)
1211 {
1212         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1213         NF_CT_ASSERT(skb);
1214
1215         /* Only update if this is not a fixed timeout */
1216         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1217                 goto acct;
1218
1219         /* If not in hash table, timer will not be active yet */
1220         if (!nf_ct_is_confirmed(ct)) {
1221                 ct->timeout.expires = extra_jiffies;
1222         } else {
1223                 unsigned long newtime = jiffies + extra_jiffies;
1224
1225                 /* Only update the timeout if the new timeout is at least
1226                    HZ jiffies from the old timeout. Need del_timer for race
1227                    avoidance (may already be dying). */
1228                 if (newtime - ct->timeout.expires >= HZ)
1229                         mod_timer_pending(&ct->timeout, newtime);
1230         }
1231
1232 acct:
1233         if (do_acct) {
1234                 struct nf_conn_acct *acct;
1235
1236                 acct = nf_conn_acct_find(ct);
1237                 if (acct) {
1238                         struct nf_conn_counter *counter = acct->counter;
1239
1240                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1241                         atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1242                 }
1243         }
1244 }
1245 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1246
1247 bool __nf_ct_kill_acct(struct nf_conn *ct,
1248                        enum ip_conntrack_info ctinfo,
1249                        const struct sk_buff *skb,
1250                        int do_acct)
1251 {
1252         if (do_acct) {
1253                 struct nf_conn_acct *acct;
1254
1255                 acct = nf_conn_acct_find(ct);
1256                 if (acct) {
1257                         struct nf_conn_counter *counter = acct->counter;
1258
1259                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1260                         atomic64_add(skb->len - skb_network_offset(skb),
1261                                      &counter[CTINFO2DIR(ctinfo)].bytes);
1262                 }
1263         }
1264
1265         if (del_timer(&ct->timeout)) {
1266                 ct->timeout.function((unsigned long)ct);
1267                 return true;
1268         }
1269         return false;
1270 }
1271 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1272
1273 #ifdef CONFIG_NF_CONNTRACK_ZONES
1274 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1275         .len    = sizeof(struct nf_conntrack_zone),
1276         .align  = __alignof__(struct nf_conntrack_zone),
1277         .id     = NF_CT_EXT_ZONE,
1278 };
1279 #endif
1280
1281 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1282
1283 #include <linux/netfilter/nfnetlink.h>
1284 #include <linux/netfilter/nfnetlink_conntrack.h>
1285 #include <linux/mutex.h>
1286
1287 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1288  * in ip_conntrack_core, since we don't want the protocols to autoload
1289  * or depend on ctnetlink */
1290 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1291                                const struct nf_conntrack_tuple *tuple)
1292 {
1293         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1294             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1295                 goto nla_put_failure;
1296         return 0;
1297
1298 nla_put_failure:
1299         return -1;
1300 }
1301 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1302
1303 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1304         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1305         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1306 };
1307 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1308
1309 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1310                                struct nf_conntrack_tuple *t)
1311 {
1312         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1313                 return -EINVAL;
1314
1315         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1316         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1317
1318         return 0;
1319 }
1320 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1321
1322 int nf_ct_port_nlattr_tuple_size(void)
1323 {
1324         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1325 }
1326 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1327 #endif
1328
1329 /* Used by ipt_REJECT and ip6t_REJECT. */
1330 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1331 {
1332         struct nf_conn *ct;
1333         enum ip_conntrack_info ctinfo;
1334
1335         /* This ICMP is in reverse direction to the packet which caused it */
1336         ct = nf_ct_get(skb, &ctinfo);
1337         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1338                 ctinfo = IP_CT_RELATED_REPLY;
1339         else
1340                 ctinfo = IP_CT_RELATED;
1341
1342         /* Attach to new skbuff, and increment count */
1343         nskb->nfct = &ct->ct_general;
1344         nskb->nfctinfo = ctinfo;
1345         nf_conntrack_get(nskb->nfct);
1346 }
1347
1348 /* Bring out ya dead! */
1349 static struct nf_conn *
1350 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1351                 void *data, unsigned int *bucket)
1352 {
1353         struct nf_conntrack_tuple_hash *h;
1354         struct nf_conn *ct;
1355         struct hlist_nulls_node *n;
1356         int cpu;
1357         spinlock_t *lockp;
1358
1359         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1360                 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1361                 local_bh_disable();
1362                 spin_lock(lockp);
1363                 if (*bucket < net->ct.htable_size) {
1364                         hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1365                                 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1366                                         continue;
1367                                 ct = nf_ct_tuplehash_to_ctrack(h);
1368                                 if (iter(ct, data))
1369                                         goto found;
1370                         }
1371                 }
1372                 spin_unlock(lockp);
1373                 local_bh_enable();
1374         }
1375
1376         for_each_possible_cpu(cpu) {
1377                 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1378
1379                 spin_lock_bh(&pcpu->lock);
1380                 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1381                         ct = nf_ct_tuplehash_to_ctrack(h);
1382                         if (iter(ct, data))
1383                                 set_bit(IPS_DYING_BIT, &ct->status);
1384                 }
1385                 spin_unlock_bh(&pcpu->lock);
1386         }
1387         return NULL;
1388 found:
1389         atomic_inc(&ct->ct_general.use);
1390         spin_unlock(lockp);
1391         local_bh_enable();
1392         return ct;
1393 }
1394
1395 void nf_ct_iterate_cleanup(struct net *net,
1396                            int (*iter)(struct nf_conn *i, void *data),
1397                            void *data, u32 portid, int report)
1398 {
1399         struct nf_conn *ct;
1400         unsigned int bucket = 0;
1401
1402         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1403                 /* Time to push up daises... */
1404                 if (del_timer(&ct->timeout))
1405                         nf_ct_delete(ct, portid, report);
1406
1407                 /* ... else the timer will get him soon. */
1408
1409                 nf_ct_put(ct);
1410         }
1411 }
1412 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1413
1414 static int kill_all(struct nf_conn *i, void *data)
1415 {
1416         return 1;
1417 }
1418
1419 void nf_ct_free_hashtable(void *hash, unsigned int size)
1420 {
1421         if (is_vmalloc_addr(hash))
1422                 vfree(hash);
1423         else
1424                 free_pages((unsigned long)hash,
1425                            get_order(sizeof(struct hlist_head) * size));
1426 }
1427 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1428
1429 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1430 {
1431         nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1432 }
1433 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1434
1435 static int untrack_refs(void)
1436 {
1437         int cnt = 0, cpu;
1438
1439         for_each_possible_cpu(cpu) {
1440                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1441
1442                 cnt += atomic_read(&ct->ct_general.use) - 1;
1443         }
1444         return cnt;
1445 }
1446
1447 void nf_conntrack_cleanup_start(void)
1448 {
1449         RCU_INIT_POINTER(ip_ct_attach, NULL);
1450 }
1451
1452 void nf_conntrack_cleanup_end(void)
1453 {
1454         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1455         while (untrack_refs() > 0)
1456                 schedule();
1457
1458 #ifdef CONFIG_NF_CONNTRACK_ZONES
1459         nf_ct_extend_unregister(&nf_ct_zone_extend);
1460 #endif
1461         nf_conntrack_proto_fini();
1462         nf_conntrack_seqadj_fini();
1463         nf_conntrack_labels_fini();
1464         nf_conntrack_helper_fini();
1465         nf_conntrack_timeout_fini();
1466         nf_conntrack_ecache_fini();
1467         nf_conntrack_tstamp_fini();
1468         nf_conntrack_acct_fini();
1469         nf_conntrack_expect_fini();
1470 }
1471
1472 /*
1473  * Mishearing the voices in his head, our hero wonders how he's
1474  * supposed to kill the mall.
1475  */
1476 void nf_conntrack_cleanup_net(struct net *net)
1477 {
1478         LIST_HEAD(single);
1479
1480         list_add(&net->exit_list, &single);
1481         nf_conntrack_cleanup_net_list(&single);
1482 }
1483
1484 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1485 {
1486         int busy;
1487         struct net *net;
1488
1489         /*
1490          * This makes sure all current packets have passed through
1491          *  netfilter framework.  Roll on, two-stage module
1492          *  delete...
1493          */
1494         synchronize_net();
1495 i_see_dead_people:
1496         busy = 0;
1497         list_for_each_entry(net, net_exit_list, exit_list) {
1498                 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1499                 if (atomic_read(&net->ct.count) != 0)
1500                         busy = 1;
1501         }
1502         if (busy) {
1503                 schedule();
1504                 goto i_see_dead_people;
1505         }
1506
1507         list_for_each_entry(net, net_exit_list, exit_list) {
1508                 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1509                 nf_conntrack_proto_pernet_fini(net);
1510                 nf_conntrack_helper_pernet_fini(net);
1511                 nf_conntrack_ecache_pernet_fini(net);
1512                 nf_conntrack_tstamp_pernet_fini(net);
1513                 nf_conntrack_acct_pernet_fini(net);
1514                 nf_conntrack_expect_pernet_fini(net);
1515                 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1516                 kfree(net->ct.slabname);
1517                 free_percpu(net->ct.stat);
1518                 free_percpu(net->ct.pcpu_lists);
1519         }
1520 }
1521
1522 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1523 {
1524         struct hlist_nulls_head *hash;
1525         unsigned int nr_slots, i;
1526         size_t sz;
1527
1528         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1529         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1530         sz = nr_slots * sizeof(struct hlist_nulls_head);
1531         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1532                                         get_order(sz));
1533         if (!hash) {
1534                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1535                 hash = vzalloc(sz);
1536         }
1537
1538         if (hash && nulls)
1539                 for (i = 0; i < nr_slots; i++)
1540                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1541
1542         return hash;
1543 }
1544 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1545
1546 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1547 {
1548         int i, bucket, rc;
1549         unsigned int hashsize, old_size;
1550         struct hlist_nulls_head *hash, *old_hash;
1551         struct nf_conntrack_tuple_hash *h;
1552         struct nf_conn *ct;
1553
1554         if (current->nsproxy->net_ns != &init_net)
1555                 return -EOPNOTSUPP;
1556
1557         /* On boot, we can set this without any fancy locking. */
1558         if (!nf_conntrack_htable_size)
1559                 return param_set_uint(val, kp);
1560
1561         rc = kstrtouint(val, 0, &hashsize);
1562         if (rc)
1563                 return rc;
1564         if (!hashsize)
1565                 return -EINVAL;
1566
1567         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1568         if (!hash)
1569                 return -ENOMEM;
1570
1571         local_bh_disable();
1572         nf_conntrack_all_lock();
1573         write_seqcount_begin(&init_net.ct.generation);
1574
1575         /* Lookups in the old hash might happen in parallel, which means we
1576          * might get false negatives during connection lookup. New connections
1577          * created because of a false negative won't make it into the hash
1578          * though since that required taking the locks.
1579          */
1580
1581         for (i = 0; i < init_net.ct.htable_size; i++) {
1582                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1583                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1584                                         struct nf_conntrack_tuple_hash, hnnode);
1585                         ct = nf_ct_tuplehash_to_ctrack(h);
1586                         hlist_nulls_del_rcu(&h->hnnode);
1587                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1588                                                   hashsize);
1589                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1590                 }
1591         }
1592         old_size = init_net.ct.htable_size;
1593         old_hash = init_net.ct.hash;
1594
1595         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1596         init_net.ct.hash = hash;
1597
1598         write_seqcount_end(&init_net.ct.generation);
1599         nf_conntrack_all_unlock();
1600         local_bh_enable();
1601
1602         nf_ct_free_hashtable(old_hash, old_size);
1603         return 0;
1604 }
1605 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1606
1607 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1608                   &nf_conntrack_htable_size, 0600);
1609
1610 void nf_ct_untracked_status_or(unsigned long bits)
1611 {
1612         int cpu;
1613
1614         for_each_possible_cpu(cpu)
1615                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1616 }
1617 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1618
1619 int nf_conntrack_init_start(void)
1620 {
1621         int max_factor = 8;
1622         int i, ret, cpu;
1623
1624         for (i = 0; i < CONNTRACK_LOCKS; i++)
1625                 spin_lock_init(&nf_conntrack_locks[i]);
1626
1627         if (!nf_conntrack_htable_size) {
1628                 /* Idea from tcp.c: use 1/16384 of memory.
1629                  * On i386: 32MB machine has 512 buckets.
1630                  * >= 1GB machines have 16384 buckets.
1631                  * >= 4GB machines have 65536 buckets.
1632                  */
1633                 nf_conntrack_htable_size
1634                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1635                            / sizeof(struct hlist_head));
1636                 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1637                         nf_conntrack_htable_size = 65536;
1638                 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1639                         nf_conntrack_htable_size = 16384;
1640                 if (nf_conntrack_htable_size < 32)
1641                         nf_conntrack_htable_size = 32;
1642
1643                 /* Use a max. factor of four by default to get the same max as
1644                  * with the old struct list_heads. When a table size is given
1645                  * we use the old value of 8 to avoid reducing the max.
1646                  * entries. */
1647                 max_factor = 4;
1648         }
1649         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1650
1651         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1652                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1653                nf_conntrack_max);
1654
1655         ret = nf_conntrack_expect_init();
1656         if (ret < 0)
1657                 goto err_expect;
1658
1659         ret = nf_conntrack_acct_init();
1660         if (ret < 0)
1661                 goto err_acct;
1662
1663         ret = nf_conntrack_tstamp_init();
1664         if (ret < 0)
1665                 goto err_tstamp;
1666
1667         ret = nf_conntrack_ecache_init();
1668         if (ret < 0)
1669                 goto err_ecache;
1670
1671         ret = nf_conntrack_timeout_init();
1672         if (ret < 0)
1673                 goto err_timeout;
1674
1675         ret = nf_conntrack_helper_init();
1676         if (ret < 0)
1677                 goto err_helper;
1678
1679         ret = nf_conntrack_labels_init();
1680         if (ret < 0)
1681                 goto err_labels;
1682
1683         ret = nf_conntrack_seqadj_init();
1684         if (ret < 0)
1685                 goto err_seqadj;
1686
1687 #ifdef CONFIG_NF_CONNTRACK_ZONES
1688         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1689         if (ret < 0)
1690                 goto err_extend;
1691 #endif
1692         ret = nf_conntrack_proto_init();
1693         if (ret < 0)
1694                 goto err_proto;
1695
1696         /* Set up fake conntrack: to never be deleted, not in any hashes */
1697         for_each_possible_cpu(cpu) {
1698                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1699                 write_pnet(&ct->ct_net, &init_net);
1700                 atomic_set(&ct->ct_general.use, 1);
1701         }
1702         /*  - and look it like as a confirmed connection */
1703         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1704         return 0;
1705
1706 err_proto:
1707 #ifdef CONFIG_NF_CONNTRACK_ZONES
1708         nf_ct_extend_unregister(&nf_ct_zone_extend);
1709 err_extend:
1710 #endif
1711         nf_conntrack_seqadj_fini();
1712 err_seqadj:
1713         nf_conntrack_labels_fini();
1714 err_labels:
1715         nf_conntrack_helper_fini();
1716 err_helper:
1717         nf_conntrack_timeout_fini();
1718 err_timeout:
1719         nf_conntrack_ecache_fini();
1720 err_ecache:
1721         nf_conntrack_tstamp_fini();
1722 err_tstamp:
1723         nf_conntrack_acct_fini();
1724 err_acct:
1725         nf_conntrack_expect_fini();
1726 err_expect:
1727         return ret;
1728 }
1729
1730 void nf_conntrack_init_end(void)
1731 {
1732         /* For use by REJECT target */
1733         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1734         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1735 }
1736
1737 /*
1738  * We need to use special "null" values, not used in hash table
1739  */
1740 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1741 #define DYING_NULLS_VAL         ((1<<30)+1)
1742 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1743
1744 int nf_conntrack_init_net(struct net *net)
1745 {
1746         int ret = -ENOMEM;
1747         int cpu;
1748
1749         atomic_set(&net->ct.count, 0);
1750         seqcount_init(&net->ct.generation);
1751
1752         net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1753         if (!net->ct.pcpu_lists)
1754                 goto err_stat;
1755
1756         for_each_possible_cpu(cpu) {
1757                 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1758
1759                 spin_lock_init(&pcpu->lock);
1760                 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1761                 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1762                 INIT_HLIST_NULLS_HEAD(&pcpu->tmpl, TEMPLATE_NULLS_VAL);
1763         }
1764
1765         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1766         if (!net->ct.stat)
1767                 goto err_pcpu_lists;
1768
1769         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1770         if (!net->ct.slabname)
1771                 goto err_slabname;
1772
1773         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1774                                                         sizeof(struct nf_conn), 0,
1775                                                         SLAB_DESTROY_BY_RCU, NULL);
1776         if (!net->ct.nf_conntrack_cachep) {
1777                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1778                 goto err_cache;
1779         }
1780
1781         net->ct.htable_size = nf_conntrack_htable_size;
1782         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1783         if (!net->ct.hash) {
1784                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1785                 goto err_hash;
1786         }
1787         ret = nf_conntrack_expect_pernet_init(net);
1788         if (ret < 0)
1789                 goto err_expect;
1790         ret = nf_conntrack_acct_pernet_init(net);
1791         if (ret < 0)
1792                 goto err_acct;
1793         ret = nf_conntrack_tstamp_pernet_init(net);
1794         if (ret < 0)
1795                 goto err_tstamp;
1796         ret = nf_conntrack_ecache_pernet_init(net);
1797         if (ret < 0)
1798                 goto err_ecache;
1799         ret = nf_conntrack_helper_pernet_init(net);
1800         if (ret < 0)
1801                 goto err_helper;
1802         ret = nf_conntrack_proto_pernet_init(net);
1803         if (ret < 0)
1804                 goto err_proto;
1805         return 0;
1806
1807 err_proto:
1808         nf_conntrack_helper_pernet_fini(net);
1809 err_helper:
1810         nf_conntrack_ecache_pernet_fini(net);
1811 err_ecache:
1812         nf_conntrack_tstamp_pernet_fini(net);
1813 err_tstamp:
1814         nf_conntrack_acct_pernet_fini(net);
1815 err_acct:
1816         nf_conntrack_expect_pernet_fini(net);
1817 err_expect:
1818         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1819 err_hash:
1820         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1821 err_cache:
1822         kfree(net->ct.slabname);
1823 err_slabname:
1824         free_percpu(net->ct.stat);
1825 err_pcpu_lists:
1826         free_percpu(net->ct.pcpu_lists);
1827 err_stat:
1828         return ret;
1829 }