net/netfilter/nf_conntrack_core.c

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