net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License version 2 as
  11  * published by the Free Software Foundation.
  12  */
  13
  14 #include <linux/types.h>
  15 #include <linux/netfilter.h>
  16 #include <linux/module.h>
  17 #include <linux/sched.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/proc_fs.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/stddef.h>
  22 #include <linux/slab.h>
  23 #include <linux/random.h>
  24 #include <linux/jhash.h>
  25 #include <linux/err.h>
  26 #include <linux/percpu.h>
  27 #include <linux/moduleparam.h>
  28 #include <linux/notifier.h>
  29 #include <linux/kernel.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/socket.h>
  32 #include <linux/mm.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/rculist_nulls.h>
  35
  36 #include <net/netfilter/nf_conntrack.h>
  37 #include <net/netfilter/nf_conntrack_l3proto.h>
  38 #include <net/netfilter/nf_conntrack_l4proto.h>
  39 #include <net/netfilter/nf_conntrack_expect.h>
  40 #include <net/netfilter/nf_conntrack_helper.h>
  41 #include <net/netfilter/nf_conntrack_core.h>
  42 #include <net/netfilter/nf_conntrack_extend.h>
  43 #include <net/netfilter/nf_conntrack_acct.h>
  44 #include <net/netfilter/nf_conntrack_ecache.h>
  45 #include <net/netfilter/nf_conntrack_zones.h>
  46 #include <net/netfilter/nf_conntrack_timestamp.h>
  47 #include <net/netfilter/nf_conntrack_timeout.h>
  48 #include <net/netfilter/nf_conntrack_labels.h>
  49 #include <net/netfilter/nf_nat.h>
  50 #include <net/netfilter/nf_nat_core.h>
  51 #include <net/netfilter/nf_nat_helper.h>
  52
  53 #define NF_CONNTRACK_VERSION    "0.5.0"
  54
  55 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  56                                       enum nf_nat_manip_type manip,
  57                                       const struct nlattr *attr) __read_mostly;
  58 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  59
  60 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
  61                               struct nf_conn *ct,
  62                               enum ip_conntrack_info ctinfo,
  63                               unsigned int protoff);
  64 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
  65
  66 DEFINE_SPINLOCK(nf_conntrack_lock);
  67 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  68
  69 unsigned int nf_conntrack_htable_size __read_mostly;
  70 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  71
  72 unsigned int nf_conntrack_max __read_mostly;
  73 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  74
  75 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  76 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  77
  78 unsigned int nf_conntrack_hash_rnd __read_mostly;
  79 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
  80
  81 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  82 {
  83         unsigned int n;
  84
  85         /* The direction must be ignored, so we hash everything up to the
  86          * destination ports (which is a multiple of 4) and treat the last
  87          * three bytes manually.
  88          */
  89         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  90         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  91                       (((__force __u16)tuple->dst.u.all << 16) |
  92                       tuple->dst.protonum));
  93 }
  94
  95 static u32 __hash_bucket(u32 hash, unsigned int size)
  96 {
  97         return ((u64)hash * size) >> 32;
  98 }
  99
 100 static u32 hash_bucket(u32 hash, const struct net *net)
 101 {
 102         return __hash_bucket(hash, net->ct.htable_size);
 103 }
 104
 105 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
 106                                   u16 zone, unsigned int size)
 107 {
 108         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 109 }
 110
 111 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 112                                        const struct nf_conntrack_tuple *tuple)
 113 {
 114         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 115 }
 116
 117 bool
 118 nf_ct_get_tuple(const struct sk_buff *skb,
 119                 unsigned int nhoff,
 120                 unsigned int dataoff,
 121                 u_int16_t l3num,
 122                 u_int8_t protonum,
 123                 struct nf_conntrack_tuple *tuple,
 124                 const struct nf_conntrack_l3proto *l3proto,
 125                 const struct nf_conntrack_l4proto *l4proto)
 126 {
 127         memset(tuple, 0, sizeof(*tuple));
 128
 129         tuple->src.l3num = l3num;
 130         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 131                 return false;
 132
 133         tuple->dst.protonum = protonum;
 134         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 135
 136         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 137 }
 138 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 139
 140 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 141                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 142 {
 143         struct nf_conntrack_l3proto *l3proto;
 144         struct nf_conntrack_l4proto *l4proto;
 145         unsigned int protoff;
 146         u_int8_t protonum;
 147         int ret;
 148
 149         rcu_read_lock();
 150
 151         l3proto = __nf_ct_l3proto_find(l3num);
 152         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 153         if (ret != NF_ACCEPT) {
 154                 rcu_read_unlock();
 155                 return false;
 156         }
 157
 158         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 159
 160         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 161                               l3proto, l4proto);
 162
 163         rcu_read_unlock();
 164         return ret;
 165 }
 166 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 167
 168 bool
 169 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 170                    const struct nf_conntrack_tuple *orig,
 171                    const struct nf_conntrack_l3proto *l3proto,
 172                    const struct nf_conntrack_l4proto *l4proto)
 173 {
 174         memset(inverse, 0, sizeof(*inverse));
 175
 176         inverse->src.l3num = orig->src.l3num;
 177         if (l3proto->invert_tuple(inverse, orig) == 0)
 178                 return false;
 179
 180         inverse->dst.dir = !orig->dst.dir;
 181
 182         inverse->dst.protonum = orig->dst.protonum;
 183         return l4proto->invert_tuple(inverse, orig);
 184 }
 185 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 186
 187 static void
 188 clean_from_lists(struct nf_conn *ct)
 189 {
 190         pr_debug("clean_from_lists(%p)\n", ct);
 191         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 192         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 193
 194         /* Destroy all pending expectations */
 195         nf_ct_remove_expectations(ct);
 196 }
 197
 198 static void
 199 destroy_conntrack(struct nf_conntrack *nfct)
 200 {
 201         struct nf_conn *ct = (struct nf_conn *)nfct;
 202         struct net *net = nf_ct_net(ct);
 203         struct nf_conntrack_l4proto *l4proto;
 204
 205         pr_debug("destroy_conntrack(%p)\n", ct);
 206         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 207         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 208
 209         /* To make sure we don't get any weird locking issues here:
 210          * destroy_conntrack() MUST NOT be called with a write lock
 211          * to nf_conntrack_lock!!! -HW */
 212         rcu_read_lock();
 213         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 214         if (l4proto && l4proto->destroy)
 215                 l4proto->destroy(ct);
 216
 217         rcu_read_unlock();
 218
 219         spin_lock_bh(&nf_conntrack_lock);
 220         /* Expectations will have been removed in clean_from_lists,
 221          * except TFTP can create an expectation on the first packet,
 222          * before connection is in the list, so we need to clean here,
 223          * too. */
 224         nf_ct_remove_expectations(ct);
 225
 226         /* We overload first tuple to link into unconfirmed or dying list.*/
 227         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 228         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 229
 230         NF_CT_STAT_INC(net, delete);
 231         spin_unlock_bh(&nf_conntrack_lock);
 232
 233         if (ct->master)
 234                 nf_ct_put(ct->master);
 235
 236         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 237         nf_conntrack_free(ct);
 238 }
 239
 240 void nf_ct_delete_from_lists(struct nf_conn *ct)
 241 {
 242         struct net *net = nf_ct_net(ct);
 243
 244         nf_ct_helper_destroy(ct);
 245         spin_lock_bh(&nf_conntrack_lock);
 246         /* Inside lock so preempt is disabled on module removal path.
 247          * Otherwise we can get spurious warnings. */
 248         NF_CT_STAT_INC(net, delete_list);
 249         clean_from_lists(ct);
 250         /* add this conntrack to the dying list */
 251         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 252                              &net->ct.dying);
 253         spin_unlock_bh(&nf_conntrack_lock);
 254 }
 255 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
 256
 257 static void death_by_event(unsigned long ul_conntrack)
 258 {
 259         struct nf_conn *ct = (void *)ul_conntrack;
 260         struct net *net = nf_ct_net(ct);
 261         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 262
 263         BUG_ON(ecache == NULL);
 264
 265         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 266                 /* bad luck, let's retry again */
 267                 ecache->timeout.expires = jiffies +
 268                         (random32() % net->ct.sysctl_events_retry_timeout);
 269                 add_timer(&ecache->timeout);
 270                 return;
 271         }
 272         /* we've got the event delivered, now it's dying */
 273         set_bit(IPS_DYING_BIT, &ct->status);
 274         nf_ct_put(ct);
 275 }
 276
 277 void nf_ct_dying_timeout(struct nf_conn *ct)
 278 {
 279         struct net *net = nf_ct_net(ct);
 280         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 281
 282         BUG_ON(ecache == NULL);
 283
 284         /* set a new timer to retry event delivery */
 285         setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
 286         ecache->timeout.expires = jiffies +
 287                 (random32() % net->ct.sysctl_events_retry_timeout);
 288         add_timer(&ecache->timeout);
 289 }
 290 EXPORT_SYMBOL_GPL(nf_ct_dying_timeout);
 291
 292 static void death_by_timeout(unsigned long ul_conntrack)
 293 {
 294         struct nf_conn *ct = (void *)ul_conntrack;
 295         struct nf_conn_tstamp *tstamp;
 296
 297         tstamp = nf_conn_tstamp_find(ct);
 298         if (tstamp && tstamp->stop == 0)
 299                 tstamp->stop = ktime_to_ns(ktime_get_real());
 300
 301         if (!test_bit(IPS_DYING_BIT, &ct->status) &&
 302             unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
 303                 /* destroy event was not delivered */
 304                 nf_ct_delete_from_lists(ct);
 305                 nf_ct_dying_timeout(ct);
 306                 return;
 307         }
 308         set_bit(IPS_DYING_BIT, &ct->status);
 309         nf_ct_delete_from_lists(ct);
 310         nf_ct_put(ct);
 311 }
 312
 313 /*
 314  * Warning :
 315  * - Caller must take a reference on returned object
 316  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 317  * OR
 318  * - Caller must lock nf_conntrack_lock before calling this function
 319  */
 320 static struct nf_conntrack_tuple_hash *
 321 ____nf_conntrack_find(struct net *net, u16 zone,
 322                       const struct nf_conntrack_tuple *tuple, u32 hash)
 323 {
 324         struct nf_conntrack_tuple_hash *h;
 325         struct hlist_nulls_node *n;
 326         unsigned int bucket = hash_bucket(hash, net);
 327
 328         /* Disable BHs the entire time since we normally need to disable them
 329          * at least once for the stats anyway.
 330          */
 331         local_bh_disable();
 332 begin:
 333         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 334                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 335                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 336                         NF_CT_STAT_INC(net, found);
 337                         local_bh_enable();
 338                         return h;
 339                 }
 340                 NF_CT_STAT_INC(net, searched);
 341         }
 342         /*
 343          * if the nulls value we got at the end of this lookup is
 344          * not the expected one, we must restart lookup.
 345          * We probably met an item that was moved to another chain.
 346          */
 347         if (get_nulls_value(n) != bucket) {
 348                 NF_CT_STAT_INC(net, search_restart);
 349                 goto begin;
 350         }
 351         local_bh_enable();
 352
 353         return NULL;
 354 }
 355
 356 struct nf_conntrack_tuple_hash *
 357 __nf_conntrack_find(struct net *net, u16 zone,
 358                     const struct nf_conntrack_tuple *tuple)
 359 {
 360         return ____nf_conntrack_find(net, zone, tuple,
 361                                      hash_conntrack_raw(tuple, zone));
 362 }
 363 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 364
 365 /* Find a connection corresponding to a tuple. */
 366 static struct nf_conntrack_tuple_hash *
 367 __nf_conntrack_find_get(struct net *net, u16 zone,
 368                         const struct nf_conntrack_tuple *tuple, u32 hash)
 369 {
 370         struct nf_conntrack_tuple_hash *h;
 371         struct nf_conn *ct;
 372
 373         rcu_read_lock();
 374 begin:
 375         h = ____nf_conntrack_find(net, zone, tuple, hash);
 376         if (h) {
 377                 ct = nf_ct_tuplehash_to_ctrack(h);
 378                 if (unlikely(nf_ct_is_dying(ct) ||
 379                              !atomic_inc_not_zero(&ct->ct_general.use)))
 380                         h = NULL;
 381                 else {
 382                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 383                                      nf_ct_zone(ct) != zone)) {
 384                                 nf_ct_put(ct);
 385                                 goto begin;
 386                         }
 387                 }
 388         }
 389         rcu_read_unlock();
 390
 391         return h;
 392 }
 393
 394 struct nf_conntrack_tuple_hash *
 395 nf_conntrack_find_get(struct net *net, u16 zone,
 396                       const struct nf_conntrack_tuple *tuple)
 397 {
 398         return __nf_conntrack_find_get(net, zone, tuple,
 399                                        hash_conntrack_raw(tuple, zone));
 400 }
 401 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 402
 403 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 404                                        unsigned int hash,
 405                                        unsigned int repl_hash)
 406 {
 407         struct net *net = nf_ct_net(ct);
 408
 409         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 410                            &net->ct.hash[hash]);
 411         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 412                            &net->ct.hash[repl_hash]);
 413 }
 414
 415 int
 416 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 417 {
 418         struct net *net = nf_ct_net(ct);
 419         unsigned int hash, repl_hash;
 420         struct nf_conntrack_tuple_hash *h;
 421         struct hlist_nulls_node *n;
 422         u16 zone;
 423
 424         zone = nf_ct_zone(ct);
 425         hash = hash_conntrack(net, zone,
 426                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 427         repl_hash = hash_conntrack(net, zone,
 428                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 429
 430         spin_lock_bh(&nf_conntrack_lock);
 431
 432         /* See if there's one in the list already, including reverse */
 433         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 434                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 435                                       &h->tuple) &&
 436                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 437                         goto out;
 438         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 439                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 440                                       &h->tuple) &&
 441                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 442                         goto out;
 443
 444         add_timer(&ct->timeout);
 445         nf_conntrack_get(&ct->ct_general);
 446         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 447         NF_CT_STAT_INC(net, insert);
 448         spin_unlock_bh(&nf_conntrack_lock);
 449
 450         return 0;
 451
 452 out:
 453         NF_CT_STAT_INC(net, insert_failed);
 454         spin_unlock_bh(&nf_conntrack_lock);
 455         return -EEXIST;
 456 }
 457 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 458
 459 /* Confirm a connection given skb; places it in hash table */
 460 int
 461 __nf_conntrack_confirm(struct sk_buff *skb)
 462 {
 463         unsigned int hash, repl_hash;
 464         struct nf_conntrack_tuple_hash *h;
 465         struct nf_conn *ct;
 466         struct nf_conn_help *help;
 467         struct nf_conn_tstamp *tstamp;
 468         struct hlist_nulls_node *n;
 469         enum ip_conntrack_info ctinfo;
 470         struct net *net;
 471         u16 zone;
 472
 473         ct = nf_ct_get(skb, &ctinfo);
 474         net = nf_ct_net(ct);
 475
 476         /* ipt_REJECT uses nf_conntrack_attach to attach related
 477            ICMP/TCP RST packets in other direction.  Actual packet
 478            which created connection will be IP_CT_NEW or for an
 479            expected connection, IP_CT_RELATED. */
 480         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 481                 return NF_ACCEPT;
 482
 483         zone = nf_ct_zone(ct);
 484         /* reuse the hash saved before */
 485         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 486         hash = hash_bucket(hash, net);
 487         repl_hash = hash_conntrack(net, zone,
 488                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 489
 490         /* We're not in hash table, and we refuse to set up related
 491            connections for unconfirmed conns.  But packet copies and
 492            REJECT will give spurious warnings here. */
 493         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 494
 495         /* No external references means no one else could have
 496            confirmed us. */
 497         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 498         pr_debug("Confirming conntrack %p\n", ct);
 499
 500         spin_lock_bh(&nf_conntrack_lock);
 501
 502         /* We have to check the DYING flag inside the lock to prevent
 503            a race against nf_ct_get_next_corpse() possibly called from
 504            user context, else we insert an already 'dead' hash, blocking
 505            further use of that particular connection -JM */
 506
 507         if (unlikely(nf_ct_is_dying(ct))) {
 508                 spin_unlock_bh(&nf_conntrack_lock);
 509                 return NF_ACCEPT;
 510         }
 511
 512         /* See if there's one in the list already, including reverse:
 513            NAT could have grabbed it without realizing, since we're
 514            not in the hash.  If there is, we lost race. */
 515         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 516                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 517                                       &h->tuple) &&
 518                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 519                         goto out;
 520         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 521                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 522                                       &h->tuple) &&
 523                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 524                         goto out;
 525
 526         /* Remove from unconfirmed list */
 527         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 528
 529         /* Timer relative to confirmation time, not original
 530            setting time, otherwise we'd get timer wrap in
 531            weird delay cases. */
 532         ct->timeout.expires += jiffies;
 533         add_timer(&ct->timeout);
 534         atomic_inc(&ct->ct_general.use);
 535         ct->status |= IPS_CONFIRMED;
 536
 537         /* set conntrack timestamp, if enabled. */
 538         tstamp = nf_conn_tstamp_find(ct);
 539         if (tstamp) {
 540                 if (skb->tstamp.tv64 == 0)
 541                         __net_timestamp(skb);
 542
 543                 tstamp->start = ktime_to_ns(skb->tstamp);
 544         }
 545         /* Since the lookup is lockless, hash insertion must be done after
 546          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 547          * guarantee that no other CPU can find the conntrack before the above
 548          * stores are visible.
 549          */
 550         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 551         NF_CT_STAT_INC(net, insert);
 552         spin_unlock_bh(&nf_conntrack_lock);
 553
 554         help = nfct_help(ct);
 555         if (help && help->helper)
 556                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 557
 558         nf_conntrack_event_cache(master_ct(ct) ?
 559                                  IPCT_RELATED : IPCT_NEW, ct);
 560         return NF_ACCEPT;
 561
 562 out:
 563         NF_CT_STAT_INC(net, insert_failed);
 564         spin_unlock_bh(&nf_conntrack_lock);
 565         return NF_DROP;
 566 }
 567 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 568
 569 /* Returns true if a connection correspondings to the tuple (required
 570    for NAT). */
 571 int
 572 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 573                          const struct nf_conn *ignored_conntrack)
 574 {
 575         struct net *net = nf_ct_net(ignored_conntrack);
 576         struct nf_conntrack_tuple_hash *h;
 577         struct hlist_nulls_node *n;
 578         struct nf_conn *ct;
 579         u16 zone = nf_ct_zone(ignored_conntrack);
 580         unsigned int hash = hash_conntrack(net, zone, tuple);
 581
 582         /* Disable BHs the entire time since we need to disable them at
 583          * least once for the stats anyway.
 584          */
 585         rcu_read_lock_bh();
 586         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 587                 ct = nf_ct_tuplehash_to_ctrack(h);
 588                 if (ct != ignored_conntrack &&
 589                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 590                     nf_ct_zone(ct) == zone) {
 591                         NF_CT_STAT_INC(net, found);
 592                         rcu_read_unlock_bh();
 593                         return 1;
 594                 }
 595                 NF_CT_STAT_INC(net, searched);
 596         }
 597         rcu_read_unlock_bh();
 598
 599         return 0;
 600 }
 601 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 602
 603 #define NF_CT_EVICTION_RANGE    8
 604
 605 /* There's a small race here where we may free a just-assured
 606    connection.  Too bad: we're in trouble anyway. */
 607 static noinline int early_drop(struct net *net, unsigned int hash)
 608 {
 609         /* Use oldest entry, which is roughly LRU */
 610         struct nf_conntrack_tuple_hash *h;
 611         struct nf_conn *ct = NULL, *tmp;
 612         struct hlist_nulls_node *n;
 613         unsigned int i, cnt = 0;
 614         int dropped = 0;
 615
 616         rcu_read_lock();
 617         for (i = 0; i < net->ct.htable_size; i++) {
 618                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 619                                          hnnode) {
 620                         tmp = nf_ct_tuplehash_to_ctrack(h);
 621                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 622                                 ct = tmp;
 623                         cnt++;
 624                 }
 625
 626                 if (ct != NULL) {
 627                         if (likely(!nf_ct_is_dying(ct) &&
 628                                    atomic_inc_not_zero(&ct->ct_general.use)))
 629                                 break;
 630                         else
 631                                 ct = NULL;
 632                 }
 633
 634                 if (cnt >= NF_CT_EVICTION_RANGE)
 635                         break;
 636
 637                 hash = (hash + 1) % net->ct.htable_size;
 638         }
 639         rcu_read_unlock();
 640
 641         if (!ct)
 642                 return dropped;
 643
 644         if (del_timer(&ct->timeout)) {
 645                 death_by_timeout((unsigned long)ct);
 646                 /* Check if we indeed killed this entry. Reliable event
 647                    delivery may have inserted it into the dying list. */
 648                 if (test_bit(IPS_DYING_BIT, &ct->status)) {
 649                         dropped = 1;
 650                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 651                 }
 652         }
 653         nf_ct_put(ct);
 654         return dropped;
 655 }
 656
 657 void init_nf_conntrack_hash_rnd(void)
 658 {
 659         unsigned int rand;
 660
 661         /*
 662          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 663          * Because there isn't enough entropy when system initializing,
 664          * and we initialize it as late as possible.
 665          */
 666         do {
 667                 get_random_bytes(&rand, sizeof(rand));
 668         } while (!rand);
 669         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 670 }
 671
 672 static struct nf_conn *
 673 __nf_conntrack_alloc(struct net *net, u16 zone,
 674                      const struct nf_conntrack_tuple *orig,
 675                      const struct nf_conntrack_tuple *repl,
 676                      gfp_t gfp, u32 hash)
 677 {
 678         struct nf_conn *ct;
 679
 680         if (unlikely(!nf_conntrack_hash_rnd)) {
 681                 init_nf_conntrack_hash_rnd();
 682                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 683                 hash = hash_conntrack_raw(orig, zone);
 684         }
 685
 686         /* We don't want any race condition at early drop stage */
 687         atomic_inc(&net->ct.count);
 688
 689         if (nf_conntrack_max &&
 690             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 691                 if (!early_drop(net, hash_bucket(hash, net))) {
 692                         atomic_dec(&net->ct.count);
 693                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
 694                         return ERR_PTR(-ENOMEM);
 695                 }
 696         }
 697
 698         /*
 699          * Do not use kmem_cache_zalloc(), as this cache uses
 700          * SLAB_DESTROY_BY_RCU.
 701          */
 702         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 703         if (ct == NULL) {
 704                 atomic_dec(&net->ct.count);
 705                 return ERR_PTR(-ENOMEM);
 706         }
 707         /*
 708          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 709          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 710          */
 711         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 712                offsetof(struct nf_conn, proto) -
 713                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 714         spin_lock_init(&ct->lock);
 715         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 716         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 717         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 718         /* save hash for reusing when confirming */
 719         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 720         /* Don't set timer yet: wait for confirmation */
 721         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 722         write_pnet(&ct->ct_net, net);
 723 #ifdef CONFIG_NF_CONNTRACK_ZONES
 724         if (zone) {
 725                 struct nf_conntrack_zone *nf_ct_zone;
 726
 727                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 728                 if (!nf_ct_zone)
 729                         goto out_free;
 730                 nf_ct_zone->id = zone;
 731         }
 732 #endif
 733         /*
 734          * changes to lookup keys must be done before setting refcnt to 1
 735          */
 736         smp_wmb();
 737         atomic_set(&ct->ct_general.use, 1);
 738         return ct;
 739
 740 #ifdef CONFIG_NF_CONNTRACK_ZONES
 741 out_free:
 742         atomic_dec(&net->ct.count);
 743         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 744         return ERR_PTR(-ENOMEM);
 745 #endif
 746 }
 747
 748 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 749                                    const struct nf_conntrack_tuple *orig,
 750                                    const struct nf_conntrack_tuple *repl,
 751                                    gfp_t gfp)
 752 {
 753         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 754 }
 755 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 756
 757 void nf_conntrack_free(struct nf_conn *ct)
 758 {
 759         struct net *net = nf_ct_net(ct);
 760
 761         nf_ct_ext_destroy(ct);
 762         atomic_dec(&net->ct.count);
 763         nf_ct_ext_free(ct);
 764         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 765 }
 766 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 767
 768
 769 /* Allocate a new conntrack: we return -ENOMEM if classification
 770    failed due to stress.  Otherwise it really is unclassifiable. */
 771 static struct nf_conntrack_tuple_hash *
 772 init_conntrack(struct net *net, struct nf_conn *tmpl,
 773                const struct nf_conntrack_tuple *tuple,
 774                struct nf_conntrack_l3proto *l3proto,
 775                struct nf_conntrack_l4proto *l4proto,
 776                struct sk_buff *skb,
 777                unsigned int dataoff, u32 hash)
 778 {
 779         struct nf_conn *ct;
 780         struct nf_conn_help *help;
 781         struct nf_conntrack_tuple repl_tuple;
 782         struct nf_conntrack_ecache *ecache;
 783         struct nf_conntrack_expect *exp;
 784         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 785         struct nf_conn_timeout *timeout_ext;
 786         unsigned int *timeouts;
 787
 788         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 789                 pr_debug("Can't invert tuple.\n");
 790                 return NULL;
 791         }
 792
 793         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 794                                   hash);
 795         if (IS_ERR(ct))
 796                 return (struct nf_conntrack_tuple_hash *)ct;
 797
 798         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 799         if (timeout_ext)
 800                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 801         else
 802                 timeouts = l4proto->get_timeouts(net);
 803
 804         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 805                 nf_conntrack_free(ct);
 806                 pr_debug("init conntrack: can't track with proto module\n");
 807                 return NULL;
 808         }
 809
 810         if (timeout_ext)
 811                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 812
 813         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 814         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 815         nf_ct_labels_ext_add(ct);
 816
 817         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 818         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 819                                  ecache ? ecache->expmask : 0,
 820                              GFP_ATOMIC);
 821
 822         spin_lock_bh(&nf_conntrack_lock);
 823         exp = nf_ct_find_expectation(net, zone, tuple);
 824         if (exp) {
 825                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 826                          ct, exp);
 827                 /* Welcome, Mr. Bond.  We've been expecting you... */
 828                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 829                 ct->master = exp->master;
 830                 if (exp->helper) {
 831                         help = nf_ct_helper_ext_add(ct, exp->helper,
 832                                                     GFP_ATOMIC);
 833                         if (help)
 834                                 rcu_assign_pointer(help->helper, exp->helper);
 835                 }
 836
 837 #ifdef CONFIG_NF_CONNTRACK_MARK
 838                 ct->mark = exp->master->mark;
 839 #endif
 840 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 841                 ct->secmark = exp->master->secmark;
 842 #endif
 843                 nf_conntrack_get(&ct->master->ct_general);
 844                 NF_CT_STAT_INC(net, expect_new);
 845         } else {
 846                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 847                 NF_CT_STAT_INC(net, new);
 848         }
 849
 850         /* Overload tuple linked list to put us in unconfirmed list. */
 851         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 852                        &net->ct.unconfirmed);
 853
 854         spin_unlock_bh(&nf_conntrack_lock);
 855
 856         if (exp) {
 857                 if (exp->expectfn)
 858                         exp->expectfn(ct, exp);
 859                 nf_ct_expect_put(exp);
 860         }
 861
 862         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 863 }
 864
 865 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 866 static inline struct nf_conn *
 867 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 868                   struct sk_buff *skb,
 869                   unsigned int dataoff,
 870                   u_int16_t l3num,
 871                   u_int8_t protonum,
 872                   struct nf_conntrack_l3proto *l3proto,
 873                   struct nf_conntrack_l4proto *l4proto,
 874                   int *set_reply,
 875                   enum ip_conntrack_info *ctinfo)
 876 {
 877         struct nf_conntrack_tuple tuple;
 878         struct nf_conntrack_tuple_hash *h;
 879         struct nf_conn *ct;
 880         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 881         u32 hash;
 882
 883         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 884                              dataoff, l3num, protonum, &tuple, l3proto,
 885                              l4proto)) {
 886                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 887                 return NULL;
 888         }
 889
 890         /* look for tuple match */
 891         hash = hash_conntrack_raw(&tuple, zone);
 892         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 893         if (!h) {
 894                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 895                                    skb, dataoff, hash);
 896                 if (!h)
 897                         return NULL;
 898                 if (IS_ERR(h))
 899                         return (void *)h;
 900         }
 901         ct = nf_ct_tuplehash_to_ctrack(h);
 902
 903         /* It exists; we have (non-exclusive) reference. */
 904         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 905                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 906                 /* Please set reply bit if this packet OK */
 907                 *set_reply = 1;
 908         } else {
 909                 /* Once we've had two way comms, always ESTABLISHED. */
 910                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 911                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 912                         *ctinfo = IP_CT_ESTABLISHED;
 913                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 914                         pr_debug("nf_conntrack_in: related packet for %p\n",
 915                                  ct);
 916                         *ctinfo = IP_CT_RELATED;
 917                 } else {
 918                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 919                         *ctinfo = IP_CT_NEW;
 920                 }
 921                 *set_reply = 0;
 922         }
 923         skb->nfct = &ct->ct_general;
 924         skb->nfctinfo = *ctinfo;
 925         return ct;
 926 }
 927
 928 unsigned int
 929 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 930                 struct sk_buff *skb)
 931 {
 932         struct nf_conn *ct, *tmpl = NULL;
 933         enum ip_conntrack_info ctinfo;
 934         struct nf_conntrack_l3proto *l3proto;
 935         struct nf_conntrack_l4proto *l4proto;
 936         unsigned int *timeouts;
 937         unsigned int dataoff;
 938         u_int8_t protonum;
 939         int set_reply = 0;
 940         int ret;
 941
 942         if (skb->nfct) {
 943                 /* Previously seen (loopback or untracked)?  Ignore. */
 944                 tmpl = (struct nf_conn *)skb->nfct;
 945                 if (!nf_ct_is_template(tmpl)) {
 946                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 947                         return NF_ACCEPT;
 948                 }
 949                 skb->nfct = NULL;
 950         }
 951
 952         /* rcu_read_lock()ed by nf_hook_slow */
 953         l3proto = __nf_ct_l3proto_find(pf);
 954         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 955                                    &dataoff, &protonum);
 956         if (ret <= 0) {
 957                 pr_debug("not prepared to track yet or error occurred\n");
 958                 NF_CT_STAT_INC_ATOMIC(net, error);
 959                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 960                 ret = -ret;
 961                 goto out;
 962         }
 963
 964         l4proto = __nf_ct_l4proto_find(pf, protonum);
 965
 966         /* It may be an special packet, error, unclean...
 967          * inverse of the return code tells to the netfilter
 968          * core what to do with the packet. */
 969         if (l4proto->error != NULL) {
 970                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 971                                      pf, hooknum);
 972                 if (ret <= 0) {
 973                         NF_CT_STAT_INC_ATOMIC(net, error);
 974                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 975                         ret = -ret;
 976                         goto out;
 977                 }
 978                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 979                 if (skb->nfct)
 980                         goto out;
 981         }
 982
 983         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 984                                l3proto, l4proto, &set_reply, &ctinfo);
 985         if (!ct) {
 986                 /* Not valid part of a connection */
 987                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 988                 ret = NF_ACCEPT;
 989                 goto out;
 990         }
 991
 992         if (IS_ERR(ct)) {
 993                 /* Too stressed to deal. */
 994                 NF_CT_STAT_INC_ATOMIC(net, drop);
 995                 ret = NF_DROP;
 996                 goto out;
 997         }
 998
 999         NF_CT_ASSERT(skb->nfct);
1000
1001         /* Decide what timeout policy we want to apply to this flow. */
1002         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1003
1004         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1005         if (ret <= 0) {
1006                 /* Invalid: inverse of the return code tells
1007                  * the netfilter core what to do */
1008                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1009                 nf_conntrack_put(skb->nfct);
1010                 skb->nfct = NULL;
1011                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1012                 if (ret == -NF_DROP)
1013                         NF_CT_STAT_INC_ATOMIC(net, drop);
1014                 ret = -ret;
1015                 goto out;
1016         }
1017
1018         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1019                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1020 out:
1021         if (tmpl) {
1022                 /* Special case: we have to repeat this hook, assign the
1023                  * template again to this packet. We assume that this packet
1024                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1025                 if (ret == NF_REPEAT)
1026                         skb->nfct = (struct nf_conntrack *)tmpl;
1027                 else
1028                         nf_ct_put(tmpl);
1029         }
1030
1031         return ret;
1032 }
1033 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1034
1035 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1036                           const struct nf_conntrack_tuple *orig)
1037 {
1038         bool ret;
1039
1040         rcu_read_lock();
1041         ret = nf_ct_invert_tuple(inverse, orig,
1042                                  __nf_ct_l3proto_find(orig->src.l3num),
1043                                  __nf_ct_l4proto_find(orig->src.l3num,
1044                                                       orig->dst.protonum));
1045         rcu_read_unlock();
1046         return ret;
1047 }
1048 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1049
1050 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1051    implicitly racy: see __nf_conntrack_confirm */
1052 void nf_conntrack_alter_reply(struct nf_conn *ct,
1053                               const struct nf_conntrack_tuple *newreply)
1054 {
1055         struct nf_conn_help *help = nfct_help(ct);
1056
1057         /* Should be unconfirmed, so not in hash table yet */
1058         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1059
1060         pr_debug("Altering reply tuple of %p to ", ct);
1061         nf_ct_dump_tuple(newreply);
1062
1063         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1064         if (ct->master || (help && !hlist_empty(&help->expectations)))
1065                 return;
1066
1067         rcu_read_lock();
1068         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1069         rcu_read_unlock();
1070 }
1071 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1072
1073 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1074 void __nf_ct_refresh_acct(struct nf_conn *ct,
1075                           enum ip_conntrack_info ctinfo,
1076                           const struct sk_buff *skb,
1077                           unsigned long extra_jiffies,
1078                           int do_acct)
1079 {
1080         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1081         NF_CT_ASSERT(skb);
1082
1083         /* Only update if this is not a fixed timeout */
1084         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1085                 goto acct;
1086
1087         /* If not in hash table, timer will not be active yet */
1088         if (!nf_ct_is_confirmed(ct)) {
1089                 ct->timeout.expires = extra_jiffies;
1090         } else {
1091                 unsigned long newtime = jiffies + extra_jiffies;
1092
1093                 /* Only update the timeout if the new timeout is at least
1094                    HZ jiffies from the old timeout. Need del_timer for race
1095                    avoidance (may already be dying). */
1096                 if (newtime - ct->timeout.expires >= HZ)
1097                         mod_timer_pending(&ct->timeout, newtime);
1098         }
1099
1100 acct:
1101         if (do_acct) {
1102                 struct nf_conn_counter *acct;
1103
1104                 acct = nf_conn_acct_find(ct);
1105                 if (acct) {
1106                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1107                         atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes);
1108                 }
1109         }
1110 }
1111 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1112
1113 bool __nf_ct_kill_acct(struct nf_conn *ct,
1114                        enum ip_conntrack_info ctinfo,
1115                        const struct sk_buff *skb,
1116                        int do_acct)
1117 {
1118         if (do_acct) {
1119                 struct nf_conn_counter *acct;
1120
1121                 acct = nf_conn_acct_find(ct);
1122                 if (acct) {
1123                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1124                         atomic64_add(skb->len - skb_network_offset(skb),
1125                                      &acct[CTINFO2DIR(ctinfo)].bytes);
1126                 }
1127         }
1128
1129         if (del_timer(&ct->timeout)) {
1130                 ct->timeout.function((unsigned long)ct);
1131                 return true;
1132         }
1133         return false;
1134 }
1135 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1136
1137 #ifdef CONFIG_NF_CONNTRACK_ZONES
1138 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1139         .len    = sizeof(struct nf_conntrack_zone),
1140         .align  = __alignof__(struct nf_conntrack_zone),
1141         .id     = NF_CT_EXT_ZONE,
1142 };
1143 #endif
1144
1145 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1146
1147 #include <linux/netfilter/nfnetlink.h>
1148 #include <linux/netfilter/nfnetlink_conntrack.h>
1149 #include <linux/mutex.h>
1150
1151 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1152  * in ip_conntrack_core, since we don't want the protocols to autoload
1153  * or depend on ctnetlink */
1154 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1155                                const struct nf_conntrack_tuple *tuple)
1156 {
1157         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1158             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1159                 goto nla_put_failure;
1160         return 0;
1161
1162 nla_put_failure:
1163         return -1;
1164 }
1165 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1166
1167 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1168         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1169         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1170 };
1171 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1172
1173 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1174                                struct nf_conntrack_tuple *t)
1175 {
1176         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1177                 return -EINVAL;
1178
1179         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1180         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1181
1182         return 0;
1183 }
1184 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1185
1186 int nf_ct_port_nlattr_tuple_size(void)
1187 {
1188         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1189 }
1190 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1191 #endif
1192
1193 /* Used by ipt_REJECT and ip6t_REJECT. */
1194 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1195 {
1196         struct nf_conn *ct;
1197         enum ip_conntrack_info ctinfo;
1198
1199         /* This ICMP is in reverse direction to the packet which caused it */
1200         ct = nf_ct_get(skb, &ctinfo);
1201         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1202                 ctinfo = IP_CT_RELATED_REPLY;
1203         else
1204                 ctinfo = IP_CT_RELATED;
1205
1206         /* Attach to new skbuff, and increment count */
1207         nskb->nfct = &ct->ct_general;
1208         nskb->nfctinfo = ctinfo;
1209         nf_conntrack_get(nskb->nfct);
1210 }
1211
1212 /* Bring out ya dead! */
1213 static struct nf_conn *
1214 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1215                 void *data, unsigned int *bucket)
1216 {
1217         struct nf_conntrack_tuple_hash *h;
1218         struct nf_conn *ct;
1219         struct hlist_nulls_node *n;
1220
1221         spin_lock_bh(&nf_conntrack_lock);
1222         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1223                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1224                         if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1225                                 continue;
1226                         ct = nf_ct_tuplehash_to_ctrack(h);
1227                         if (iter(ct, data))
1228                                 goto found;
1229                 }
1230         }
1231         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1232                 ct = nf_ct_tuplehash_to_ctrack(h);
1233                 if (iter(ct, data))
1234                         set_bit(IPS_DYING_BIT, &ct->status);
1235         }
1236         spin_unlock_bh(&nf_conntrack_lock);
1237         return NULL;
1238 found:
1239         atomic_inc(&ct->ct_general.use);
1240         spin_unlock_bh(&nf_conntrack_lock);
1241         return ct;
1242 }
1243
1244 void nf_ct_iterate_cleanup(struct net *net,
1245                            int (*iter)(struct nf_conn *i, void *data),
1246                            void *data)
1247 {
1248         struct nf_conn *ct;
1249         unsigned int bucket = 0;
1250
1251         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1252                 /* Time to push up daises... */
1253                 if (del_timer(&ct->timeout))
1254                         death_by_timeout((unsigned long)ct);
1255                 /* ... else the timer will get him soon. */
1256
1257                 nf_ct_put(ct);
1258         }
1259 }
1260 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1261
1262 struct __nf_ct_flush_report {
1263         u32 pid;
1264         int report;
1265 };
1266
1267 static int kill_report(struct nf_conn *i, void *data)
1268 {
1269         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1270         struct nf_conn_tstamp *tstamp;
1271
1272         tstamp = nf_conn_tstamp_find(i);
1273         if (tstamp && tstamp->stop == 0)
1274                 tstamp->stop = ktime_to_ns(ktime_get_real());
1275
1276         /* If we fail to deliver the event, death_by_timeout() will retry */
1277         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1278                                       fr->pid, fr->report) < 0)
1279                 return 1;
1280
1281         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1282         set_bit(IPS_DYING_BIT, &i->status);
1283         return 1;
1284 }
1285
1286 static int kill_all(struct nf_conn *i, void *data)
1287 {
1288         return 1;
1289 }
1290
1291 void nf_ct_free_hashtable(void *hash, unsigned int size)
1292 {
1293         if (is_vmalloc_addr(hash))
1294                 vfree(hash);
1295         else
1296                 free_pages((unsigned long)hash,
1297                            get_order(sizeof(struct hlist_head) * size));
1298 }
1299 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1300
1301 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1302 {
1303         struct __nf_ct_flush_report fr = {
1304                 .pid    = pid,
1305                 .report = report,
1306         };
1307         nf_ct_iterate_cleanup(net, kill_report, &fr);
1308 }
1309 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1310
1311 static void nf_ct_release_dying_list(struct net *net)
1312 {
1313         struct nf_conntrack_tuple_hash *h;
1314         struct nf_conn *ct;
1315         struct hlist_nulls_node *n;
1316
1317         spin_lock_bh(&nf_conntrack_lock);
1318         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1319                 ct = nf_ct_tuplehash_to_ctrack(h);
1320                 /* never fails to remove them, no listeners at this point */
1321                 nf_ct_kill(ct);
1322         }
1323         spin_unlock_bh(&nf_conntrack_lock);
1324 }
1325
1326 static int untrack_refs(void)
1327 {
1328         int cnt = 0, cpu;
1329
1330         for_each_possible_cpu(cpu) {
1331                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1332
1333                 cnt += atomic_read(&ct->ct_general.use) - 1;
1334         }
1335         return cnt;
1336 }
1337
1338 void nf_conntrack_cleanup_start(void)
1339 {
1340         RCU_INIT_POINTER(ip_ct_attach, NULL);
1341 }
1342
1343 void nf_conntrack_cleanup_end(void)
1344 {
1345         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1346         while (untrack_refs() > 0)
1347                 schedule();
1348
1349 #ifdef CONFIG_NF_CONNTRACK_ZONES
1350         nf_ct_extend_unregister(&nf_ct_zone_extend);
1351 #endif
1352         nf_conntrack_proto_fini();
1353         nf_conntrack_labels_fini();
1354         nf_conntrack_helper_fini();
1355         nf_conntrack_timeout_fini();
1356         nf_conntrack_ecache_fini();
1357         nf_conntrack_tstamp_fini();
1358         nf_conntrack_acct_fini();
1359         nf_conntrack_expect_fini();
1360 }
1361
1362 /*
1363  * Mishearing the voices in his head, our hero wonders how he's
1364  * supposed to kill the mall.
1365  */
1366 void nf_conntrack_cleanup_net(struct net *net)
1367 {
1368         LIST_HEAD(single);
1369
1370         list_add(&net->exit_list, &single);
1371         nf_conntrack_cleanup_net_list(&single);
1372 }
1373
1374 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1375 {
1376         int busy;
1377         struct net *net;
1378
1379         /*
1380          * This makes sure all current packets have passed through
1381          *  netfilter framework.  Roll on, two-stage module
1382          *  delete...
1383          */
1384         synchronize_net();
1385 i_see_dead_people:
1386         busy = 0;
1387         list_for_each_entry(net, net_exit_list, exit_list) {
1388                 nf_ct_iterate_cleanup(net, kill_all, NULL);
1389                 nf_ct_release_dying_list(net);
1390                 if (atomic_read(&net->ct.count) != 0)
1391                         busy = 1;
1392         }
1393         if (busy) {
1394                 schedule();
1395                 goto i_see_dead_people;
1396         }
1397
1398         list_for_each_entry(net, net_exit_list, exit_list) {
1399                 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1400                 nf_conntrack_proto_pernet_fini(net);
1401                 nf_conntrack_helper_pernet_fini(net);
1402                 nf_conntrack_ecache_pernet_fini(net);
1403                 nf_conntrack_tstamp_pernet_fini(net);
1404                 nf_conntrack_acct_pernet_fini(net);
1405                 nf_conntrack_expect_pernet_fini(net);
1406                 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1407                 kfree(net->ct.slabname);
1408                 free_percpu(net->ct.stat);
1409         }
1410 }
1411
1412 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1413 {
1414         struct hlist_nulls_head *hash;
1415         unsigned int nr_slots, i;
1416         size_t sz;
1417
1418         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1419         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1420         sz = nr_slots * sizeof(struct hlist_nulls_head);
1421         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1422                                         get_order(sz));
1423         if (!hash) {
1424                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1425                 hash = vzalloc(sz);
1426         }
1427
1428         if (hash && nulls)
1429                 for (i = 0; i < nr_slots; i++)
1430                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1431
1432         return hash;
1433 }
1434 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1435
1436 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1437 {
1438         int i, bucket, rc;
1439         unsigned int hashsize, old_size;
1440         struct hlist_nulls_head *hash, *old_hash;
1441         struct nf_conntrack_tuple_hash *h;
1442         struct nf_conn *ct;
1443
1444         if (current->nsproxy->net_ns != &init_net)
1445                 return -EOPNOTSUPP;
1446
1447         /* On boot, we can set this without any fancy locking. */
1448         if (!nf_conntrack_htable_size)
1449                 return param_set_uint(val, kp);
1450
1451         rc = kstrtouint(val, 0, &hashsize);
1452         if (rc)
1453                 return rc;
1454         if (!hashsize)
1455                 return -EINVAL;
1456
1457         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1458         if (!hash)
1459                 return -ENOMEM;
1460
1461         /* Lookups in the old hash might happen in parallel, which means we
1462          * might get false negatives during connection lookup. New connections
1463          * created because of a false negative won't make it into the hash
1464          * though since that required taking the lock.
1465          */
1466         spin_lock_bh(&nf_conntrack_lock);
1467         for (i = 0; i < init_net.ct.htable_size; i++) {
1468                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1469                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1470                                         struct nf_conntrack_tuple_hash, hnnode);
1471                         ct = nf_ct_tuplehash_to_ctrack(h);
1472                         hlist_nulls_del_rcu(&h->hnnode);
1473                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1474                                                   hashsize);
1475                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1476                 }
1477         }
1478         old_size = init_net.ct.htable_size;
1479         old_hash = init_net.ct.hash;
1480
1481         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1482         init_net.ct.hash = hash;
1483         spin_unlock_bh(&nf_conntrack_lock);
1484
1485         nf_ct_free_hashtable(old_hash, old_size);
1486         return 0;
1487 }
1488 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1489
1490 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1491                   &nf_conntrack_htable_size, 0600);
1492
1493 void nf_ct_untracked_status_or(unsigned long bits)
1494 {
1495         int cpu;
1496
1497         for_each_possible_cpu(cpu)
1498                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1499 }
1500 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1501
1502 int nf_conntrack_init_start(void)
1503 {
1504         int max_factor = 8;
1505         int ret, cpu;
1506
1507         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1508          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1509         if (!nf_conntrack_htable_size) {
1510                 nf_conntrack_htable_size
1511                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1512                            / sizeof(struct hlist_head));
1513                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1514                         nf_conntrack_htable_size = 16384;
1515                 if (nf_conntrack_htable_size < 32)
1516                         nf_conntrack_htable_size = 32;
1517
1518                 /* Use a max. factor of four by default to get the same max as
1519                  * with the old struct list_heads. When a table size is given
1520                  * we use the old value of 8 to avoid reducing the max.
1521                  * entries. */
1522                 max_factor = 4;
1523         }
1524         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1525
1526         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1527                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1528                nf_conntrack_max);
1529
1530         ret = nf_conntrack_expect_init();
1531         if (ret < 0)
1532                 goto err_expect;
1533
1534         ret = nf_conntrack_acct_init();
1535         if (ret < 0)
1536                 goto err_acct;
1537
1538         ret = nf_conntrack_tstamp_init();
1539         if (ret < 0)
1540                 goto err_tstamp;
1541
1542         ret = nf_conntrack_ecache_init();
1543         if (ret < 0)
1544                 goto err_ecache;
1545
1546         ret = nf_conntrack_timeout_init();
1547         if (ret < 0)
1548                 goto err_timeout;
1549
1550         ret = nf_conntrack_helper_init();
1551         if (ret < 0)
1552                 goto err_helper;
1553
1554         ret = nf_conntrack_labels_init();
1555         if (ret < 0)
1556                 goto err_labels;
1557
1558 #ifdef CONFIG_NF_CONNTRACK_ZONES
1559         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1560         if (ret < 0)
1561                 goto err_extend;
1562 #endif
1563         ret = nf_conntrack_proto_init();
1564         if (ret < 0)
1565                 goto err_proto;
1566
1567         /* Set up fake conntrack: to never be deleted, not in any hashes */
1568         for_each_possible_cpu(cpu) {
1569                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1570                 write_pnet(&ct->ct_net, &init_net);
1571                 atomic_set(&ct->ct_general.use, 1);
1572         }
1573         /*  - and look it like as a confirmed connection */
1574         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1575         return 0;
1576
1577 err_proto:
1578 #ifdef CONFIG_NF_CONNTRACK_ZONES
1579         nf_ct_extend_unregister(&nf_ct_zone_extend);
1580 err_extend:
1581 #endif
1582         nf_conntrack_labels_fini();
1583 err_labels:
1584         nf_conntrack_helper_fini();
1585 err_helper:
1586         nf_conntrack_timeout_fini();
1587 err_timeout:
1588         nf_conntrack_ecache_fini();
1589 err_ecache:
1590         nf_conntrack_tstamp_fini();
1591 err_tstamp:
1592         nf_conntrack_acct_fini();
1593 err_acct:
1594         nf_conntrack_expect_fini();
1595 err_expect:
1596         return ret;
1597 }
1598
1599 void nf_conntrack_init_end(void)
1600 {
1601         /* For use by REJECT target */
1602         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1603         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1604
1605         /* Howto get NAT offsets */
1606         RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
1607 }
1608
1609 /*
1610  * We need to use special "null" values, not used in hash table
1611  */
1612 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1613 #define DYING_NULLS_VAL         ((1<<30)+1)
1614 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1615
1616 int nf_conntrack_init_net(struct net *net)
1617 {
1618         int ret;
1619
1620         atomic_set(&net->ct.count, 0);
1621         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1622         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1623         INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1624         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1625         if (!net->ct.stat) {
1626                 ret = -ENOMEM;
1627                 goto err_stat;
1628         }
1629
1630         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1631         if (!net->ct.slabname) {
1632                 ret = -ENOMEM;
1633                 goto err_slabname;
1634         }
1635
1636         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1637                                                         sizeof(struct nf_conn), 0,
1638                                                         SLAB_DESTROY_BY_RCU, NULL);
1639         if (!net->ct.nf_conntrack_cachep) {
1640                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1641                 ret = -ENOMEM;
1642                 goto err_cache;
1643         }
1644
1645         net->ct.htable_size = nf_conntrack_htable_size;
1646         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1647         if (!net->ct.hash) {
1648                 ret = -ENOMEM;
1649                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1650                 goto err_hash;
1651         }
1652         ret = nf_conntrack_expect_pernet_init(net);
1653         if (ret < 0)
1654                 goto err_expect;
1655         ret = nf_conntrack_acct_pernet_init(net);
1656         if (ret < 0)
1657                 goto err_acct;
1658         ret = nf_conntrack_tstamp_pernet_init(net);
1659         if (ret < 0)
1660                 goto err_tstamp;
1661         ret = nf_conntrack_ecache_pernet_init(net);
1662         if (ret < 0)
1663                 goto err_ecache;
1664         ret = nf_conntrack_helper_pernet_init(net);
1665         if (ret < 0)
1666                 goto err_helper;
1667         ret = nf_conntrack_proto_pernet_init(net);
1668         if (ret < 0)
1669                 goto err_proto;
1670         return 0;
1671
1672 err_proto:
1673         nf_conntrack_helper_pernet_fini(net);
1674 err_helper:
1675         nf_conntrack_ecache_pernet_fini(net);
1676 err_ecache:
1677         nf_conntrack_tstamp_pernet_fini(net);
1678 err_tstamp:
1679         nf_conntrack_acct_pernet_fini(net);
1680 err_acct:
1681         nf_conntrack_expect_pernet_fini(net);
1682 err_expect:
1683         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1684 err_hash:
1685         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1686 err_cache:
1687         kfree(net->ct.slabname);
1688 err_slabname:
1689         free_percpu(net->ct.stat);
1690 err_stat:
1691         return ret;
1692 }
1693
1694 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1695                         enum ip_conntrack_dir dir,
1696                         u32 seq);
1697 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);