net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  */
  14
  15 #include <linux/types.h>
  16 #include <linux/netfilter.h>
  17 #include <linux/module.h>
  18 #include <linux/sched.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/proc_fs.h>
  21 #include <linux/vmalloc.h>
  22 #include <linux/stddef.h>
  23 #include <linux/slab.h>
  24 #include <linux/random.h>
  25 #include <linux/jhash.h>
  26 #include <linux/err.h>
  27 #include <linux/percpu.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/notifier.h>
  30 #include <linux/kernel.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/socket.h>
  33 #include <linux/mm.h>
  34 #include <linux/nsproxy.h>
  35 #include <linux/rculist_nulls.h>
  36
  37 #include <net/netfilter/nf_conntrack.h>
  38 #include <net/netfilter/nf_conntrack_l3proto.h>
  39 #include <net/netfilter/nf_conntrack_l4proto.h>
  40 #include <net/netfilter/nf_conntrack_expect.h>
  41 #include <net/netfilter/nf_conntrack_helper.h>
  42 #include <net/netfilter/nf_conntrack_seqadj.h>
  43 #include <net/netfilter/nf_conntrack_core.h>
  44 #include <net/netfilter/nf_conntrack_extend.h>
  45 #include <net/netfilter/nf_conntrack_acct.h>
  46 #include <net/netfilter/nf_conntrack_ecache.h>
  47 #include <net/netfilter/nf_conntrack_zones.h>
  48 #include <net/netfilter/nf_conntrack_timestamp.h>
  49 #include <net/netfilter/nf_conntrack_timeout.h>
  50 #include <net/netfilter/nf_conntrack_labels.h>
  51 #include <net/netfilter/nf_conntrack_synproxy.h>
  52 #include <net/netfilter/nf_nat.h>
  53 #include <net/netfilter/nf_nat_core.h>
  54 #include <net/netfilter/nf_nat_helper.h>
  55
  56 #define NF_CONNTRACK_VERSION    "0.5.0"
  57
  58 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  59                                       enum nf_nat_manip_type manip,
  60                                       const struct nlattr *attr) __read_mostly;
  61 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  62
  63 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
  64                               struct nf_conn *ct,
  65                               enum ip_conntrack_info ctinfo,
  66                               unsigned int protoff);
  67 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
  68
  69 DEFINE_SPINLOCK(nf_conntrack_lock);
  70 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  71
  72 unsigned int nf_conntrack_htable_size __read_mostly;
  73 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  74
  75 unsigned int nf_conntrack_max __read_mostly;
  76 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  77
  78 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  79 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  80
  81 unsigned int nf_conntrack_hash_rnd __read_mostly;
  82 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
  83
  84 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  85 {
  86         unsigned int n;
  87
  88         /* The direction must be ignored, so we hash everything up to the
  89          * destination ports (which is a multiple of 4) and treat the last
  90          * three bytes manually.
  91          */
  92         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  93         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  94                       (((__force __u16)tuple->dst.u.all << 16) |
  95                       tuple->dst.protonum));
  96 }
  97
  98 static u32 __hash_bucket(u32 hash, unsigned int size)
  99 {
 100         return ((u64)hash * size) >> 32;
 101 }
 102
 103 static u32 hash_bucket(u32 hash, const struct net *net)
 104 {
 105         return __hash_bucket(hash, net->ct.htable_size);
 106 }
 107
 108 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
 109                                   u16 zone, unsigned int size)
 110 {
 111         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 112 }
 113
 114 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 115                                        const struct nf_conntrack_tuple *tuple)
 116 {
 117         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 118 }
 119
 120 bool
 121 nf_ct_get_tuple(const struct sk_buff *skb,
 122                 unsigned int nhoff,
 123                 unsigned int dataoff,
 124                 u_int16_t l3num,
 125                 u_int8_t protonum,
 126                 struct nf_conntrack_tuple *tuple,
 127                 const struct nf_conntrack_l3proto *l3proto,
 128                 const struct nf_conntrack_l4proto *l4proto)
 129 {
 130         memset(tuple, 0, sizeof(*tuple));
 131
 132         tuple->src.l3num = l3num;
 133         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 134                 return false;
 135
 136         tuple->dst.protonum = protonum;
 137         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 138
 139         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 140 }
 141 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 142
 143 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 144                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 145 {
 146         struct nf_conntrack_l3proto *l3proto;
 147         struct nf_conntrack_l4proto *l4proto;
 148         unsigned int protoff;
 149         u_int8_t protonum;
 150         int ret;
 151
 152         rcu_read_lock();
 153
 154         l3proto = __nf_ct_l3proto_find(l3num);
 155         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 156         if (ret != NF_ACCEPT) {
 157                 rcu_read_unlock();
 158                 return false;
 159         }
 160
 161         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 162
 163         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 164                               l3proto, l4proto);
 165
 166         rcu_read_unlock();
 167         return ret;
 168 }
 169 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 170
 171 bool
 172 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 173                    const struct nf_conntrack_tuple *orig,
 174                    const struct nf_conntrack_l3proto *l3proto,
 175                    const struct nf_conntrack_l4proto *l4proto)
 176 {
 177         memset(inverse, 0, sizeof(*inverse));
 178
 179         inverse->src.l3num = orig->src.l3num;
 180         if (l3proto->invert_tuple(inverse, orig) == 0)
 181                 return false;
 182
 183         inverse->dst.dir = !orig->dst.dir;
 184
 185         inverse->dst.protonum = orig->dst.protonum;
 186         return l4proto->invert_tuple(inverse, orig);
 187 }
 188 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 189
 190 static void
 191 clean_from_lists(struct nf_conn *ct)
 192 {
 193         pr_debug("clean_from_lists(%p)\n", ct);
 194         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 195         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 196
 197         /* Destroy all pending expectations */
 198         nf_ct_remove_expectations(ct);
 199 }
 200
 201 static void
 202 destroy_conntrack(struct nf_conntrack *nfct)
 203 {
 204         struct nf_conn *ct = (struct nf_conn *)nfct;
 205         struct net *net = nf_ct_net(ct);
 206         struct nf_conntrack_l4proto *l4proto;
 207
 208         pr_debug("destroy_conntrack(%p)\n", ct);
 209         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 210         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 211
 212         /* To make sure we don't get any weird locking issues here:
 213          * destroy_conntrack() MUST NOT be called with a write lock
 214          * to nf_conntrack_lock!!! -HW */
 215         rcu_read_lock();
 216         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 217         if (l4proto && l4proto->destroy)
 218                 l4proto->destroy(ct);
 219
 220         rcu_read_unlock();
 221
 222         spin_lock_bh(&nf_conntrack_lock);
 223         /* Expectations will have been removed in clean_from_lists,
 224          * except TFTP can create an expectation on the first packet,
 225          * before connection is in the list, so we need to clean here,
 226          * too. */
 227         nf_ct_remove_expectations(ct);
 228
 229         /* We overload first tuple to link into unconfirmed or dying list.*/
 230         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 231         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 232
 233         NF_CT_STAT_INC(net, delete);
 234         spin_unlock_bh(&nf_conntrack_lock);
 235
 236         if (ct->master)
 237                 nf_ct_put(ct->master);
 238
 239         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 240         nf_conntrack_free(ct);
 241 }
 242
 243 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 244 {
 245         struct net *net = nf_ct_net(ct);
 246
 247         nf_ct_helper_destroy(ct);
 248         spin_lock_bh(&nf_conntrack_lock);
 249         /* Inside lock so preempt is disabled on module removal path.
 250          * Otherwise we can get spurious warnings. */
 251         NF_CT_STAT_INC(net, delete_list);
 252         clean_from_lists(ct);
 253         /* add this conntrack to the dying list */
 254         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 255                              &net->ct.dying);
 256         spin_unlock_bh(&nf_conntrack_lock);
 257 }
 258
 259 static void death_by_event(unsigned long ul_conntrack)
 260 {
 261         struct nf_conn *ct = (void *)ul_conntrack;
 262         struct net *net = nf_ct_net(ct);
 263         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 264
 265         BUG_ON(ecache == NULL);
 266
 267         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 268                 /* bad luck, let's retry again */
 269                 ecache->timeout.expires = jiffies +
 270                         (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 271                 add_timer(&ecache->timeout);
 272                 return;
 273         }
 274         /* we've got the event delivered, now it's dying */
 275         set_bit(IPS_DYING_BIT, &ct->status);
 276         nf_ct_put(ct);
 277 }
 278
 279 static void nf_ct_dying_timeout(struct nf_conn *ct)
 280 {
 281         struct net *net = nf_ct_net(ct);
 282         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 283
 284         BUG_ON(ecache == NULL);
 285
 286         /* set a new timer to retry event delivery */
 287         setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
 288         ecache->timeout.expires = jiffies +
 289                 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 290         add_timer(&ecache->timeout);
 291 }
 292
 293 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 294 {
 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 (!nf_ct_is_dying(ct) &&
 302             unlikely(nf_conntrack_event_report(IPCT_DESTROY, ct,
 303             portid, report) < 0)) {
 304                 /* destroy event was not delivered */
 305                 nf_ct_delete_from_lists(ct);
 306                 nf_ct_dying_timeout(ct);
 307                 return false;
 308         }
 309         set_bit(IPS_DYING_BIT, &ct->status);
 310         nf_ct_delete_from_lists(ct);
 311         nf_ct_put(ct);
 312         return true;
 313 }
 314 EXPORT_SYMBOL_GPL(nf_ct_delete);
 315
 316 static void death_by_timeout(unsigned long ul_conntrack)
 317 {
 318         nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
 319 }
 320
 321 /*
 322  * Warning :
 323  * - Caller must take a reference on returned object
 324  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 325  * OR
 326  * - Caller must lock nf_conntrack_lock before calling this function
 327  */
 328 static struct nf_conntrack_tuple_hash *
 329 ____nf_conntrack_find(struct net *net, u16 zone,
 330                       const struct nf_conntrack_tuple *tuple, u32 hash)
 331 {
 332         struct nf_conntrack_tuple_hash *h;
 333         struct hlist_nulls_node *n;
 334         unsigned int bucket = hash_bucket(hash, net);
 335
 336         /* Disable BHs the entire time since we normally need to disable them
 337          * at least once for the stats anyway.
 338          */
 339         local_bh_disable();
 340 begin:
 341         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 342                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 343                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 344                         NF_CT_STAT_INC(net, found);
 345                         local_bh_enable();
 346                         return h;
 347                 }
 348                 NF_CT_STAT_INC(net, searched);
 349         }
 350         /*
 351          * if the nulls value we got at the end of this lookup is
 352          * not the expected one, we must restart lookup.
 353          * We probably met an item that was moved to another chain.
 354          */
 355         if (get_nulls_value(n) != bucket) {
 356                 NF_CT_STAT_INC(net, search_restart);
 357                 goto begin;
 358         }
 359         local_bh_enable();
 360
 361         return NULL;
 362 }
 363
 364 struct nf_conntrack_tuple_hash *
 365 __nf_conntrack_find(struct net *net, u16 zone,
 366                     const struct nf_conntrack_tuple *tuple)
 367 {
 368         return ____nf_conntrack_find(net, zone, tuple,
 369                                      hash_conntrack_raw(tuple, zone));
 370 }
 371 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 372
 373 /* Find a connection corresponding to a tuple. */
 374 static struct nf_conntrack_tuple_hash *
 375 __nf_conntrack_find_get(struct net *net, u16 zone,
 376                         const struct nf_conntrack_tuple *tuple, u32 hash)
 377 {
 378         struct nf_conntrack_tuple_hash *h;
 379         struct nf_conn *ct;
 380
 381         rcu_read_lock();
 382 begin:
 383         h = ____nf_conntrack_find(net, zone, tuple, hash);
 384         if (h) {
 385                 ct = nf_ct_tuplehash_to_ctrack(h);
 386                 if (unlikely(nf_ct_is_dying(ct) ||
 387                              !atomic_inc_not_zero(&ct->ct_general.use)))
 388                         h = NULL;
 389                 else {
 390                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 391                                      nf_ct_zone(ct) != zone)) {
 392                                 nf_ct_put(ct);
 393                                 goto begin;
 394                         }
 395                 }
 396         }
 397         rcu_read_unlock();
 398
 399         return h;
 400 }
 401
 402 struct nf_conntrack_tuple_hash *
 403 nf_conntrack_find_get(struct net *net, u16 zone,
 404                       const struct nf_conntrack_tuple *tuple)
 405 {
 406         return __nf_conntrack_find_get(net, zone, tuple,
 407                                        hash_conntrack_raw(tuple, zone));
 408 }
 409 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 410
 411 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 412                                        unsigned int hash,
 413                                        unsigned int repl_hash)
 414 {
 415         struct net *net = nf_ct_net(ct);
 416
 417         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 418                            &net->ct.hash[hash]);
 419         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 420                            &net->ct.hash[repl_hash]);
 421 }
 422
 423 int
 424 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 425 {
 426         struct net *net = nf_ct_net(ct);
 427         unsigned int hash, repl_hash;
 428         struct nf_conntrack_tuple_hash *h;
 429         struct hlist_nulls_node *n;
 430         u16 zone;
 431
 432         zone = nf_ct_zone(ct);
 433         hash = hash_conntrack(net, zone,
 434                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 435         repl_hash = hash_conntrack(net, zone,
 436                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 437
 438         spin_lock_bh(&nf_conntrack_lock);
 439
 440         /* See if there's one in the list already, including reverse */
 441         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 442                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 443                                       &h->tuple) &&
 444                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 445                         goto out;
 446         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 447                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 448                                       &h->tuple) &&
 449                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 450                         goto out;
 451
 452         add_timer(&ct->timeout);
 453         nf_conntrack_get(&ct->ct_general);
 454         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 455         NF_CT_STAT_INC(net, insert);
 456         spin_unlock_bh(&nf_conntrack_lock);
 457
 458         return 0;
 459
 460 out:
 461         NF_CT_STAT_INC(net, insert_failed);
 462         spin_unlock_bh(&nf_conntrack_lock);
 463         return -EEXIST;
 464 }
 465 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 466
 467 /* Confirm a connection given skb; places it in hash table */
 468 int
 469 __nf_conntrack_confirm(struct sk_buff *skb)
 470 {
 471         unsigned int hash, repl_hash;
 472         struct nf_conntrack_tuple_hash *h;
 473         struct nf_conn *ct;
 474         struct nf_conn_help *help;
 475         struct nf_conn_tstamp *tstamp;
 476         struct hlist_nulls_node *n;
 477         enum ip_conntrack_info ctinfo;
 478         struct net *net;
 479         u16 zone;
 480
 481         ct = nf_ct_get(skb, &ctinfo);
 482         net = nf_ct_net(ct);
 483
 484         /* ipt_REJECT uses nf_conntrack_attach to attach related
 485            ICMP/TCP RST packets in other direction.  Actual packet
 486            which created connection will be IP_CT_NEW or for an
 487            expected connection, IP_CT_RELATED. */
 488         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 489                 return NF_ACCEPT;
 490
 491         zone = nf_ct_zone(ct);
 492         /* reuse the hash saved before */
 493         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 494         hash = hash_bucket(hash, net);
 495         repl_hash = hash_conntrack(net, zone,
 496                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 497
 498         /* We're not in hash table, and we refuse to set up related
 499            connections for unconfirmed conns.  But packet copies and
 500            REJECT will give spurious warnings here. */
 501         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 502
 503         /* No external references means no one else could have
 504            confirmed us. */
 505         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 506         pr_debug("Confirming conntrack %p\n", ct);
 507
 508         spin_lock_bh(&nf_conntrack_lock);
 509
 510         /* We have to check the DYING flag inside the lock to prevent
 511            a race against nf_ct_get_next_corpse() possibly called from
 512            user context, else we insert an already 'dead' hash, blocking
 513            further use of that particular connection -JM */
 514
 515         if (unlikely(nf_ct_is_dying(ct))) {
 516                 spin_unlock_bh(&nf_conntrack_lock);
 517                 return NF_ACCEPT;
 518         }
 519
 520         /* See if there's one in the list already, including reverse:
 521            NAT could have grabbed it without realizing, since we're
 522            not in the hash.  If there is, we lost race. */
 523         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 524                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 525                                       &h->tuple) &&
 526                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 527                         goto out;
 528         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 529                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 530                                       &h->tuple) &&
 531                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 532                         goto out;
 533
 534         /* Remove from unconfirmed list */
 535         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 536
 537         /* Timer relative to confirmation time, not original
 538            setting time, otherwise we'd get timer wrap in
 539            weird delay cases. */
 540         ct->timeout.expires += jiffies;
 541         add_timer(&ct->timeout);
 542         atomic_inc(&ct->ct_general.use);
 543         ct->status |= IPS_CONFIRMED;
 544
 545         /* set conntrack timestamp, if enabled. */
 546         tstamp = nf_conn_tstamp_find(ct);
 547         if (tstamp) {
 548                 if (skb->tstamp.tv64 == 0)
 549                         __net_timestamp(skb);
 550
 551                 tstamp->start = ktime_to_ns(skb->tstamp);
 552         }
 553         /* Since the lookup is lockless, hash insertion must be done after
 554          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 555          * guarantee that no other CPU can find the conntrack before the above
 556          * stores are visible.
 557          */
 558         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 559         NF_CT_STAT_INC(net, insert);
 560         spin_unlock_bh(&nf_conntrack_lock);
 561
 562         help = nfct_help(ct);
 563         if (help && help->helper)
 564                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 565
 566         nf_conntrack_event_cache(master_ct(ct) ?
 567                                  IPCT_RELATED : IPCT_NEW, ct);
 568         return NF_ACCEPT;
 569
 570 out:
 571         NF_CT_STAT_INC(net, insert_failed);
 572         spin_unlock_bh(&nf_conntrack_lock);
 573         return NF_DROP;
 574 }
 575 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 576
 577 /* Returns true if a connection correspondings to the tuple (required
 578    for NAT). */
 579 int
 580 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 581                          const struct nf_conn *ignored_conntrack)
 582 {
 583         struct net *net = nf_ct_net(ignored_conntrack);
 584         struct nf_conntrack_tuple_hash *h;
 585         struct hlist_nulls_node *n;
 586         struct nf_conn *ct;
 587         u16 zone = nf_ct_zone(ignored_conntrack);
 588         unsigned int hash = hash_conntrack(net, zone, tuple);
 589
 590         /* Disable BHs the entire time since we need to disable them at
 591          * least once for the stats anyway.
 592          */
 593         rcu_read_lock_bh();
 594         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 595                 ct = nf_ct_tuplehash_to_ctrack(h);
 596                 if (ct != ignored_conntrack &&
 597                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 598                     nf_ct_zone(ct) == zone) {
 599                         NF_CT_STAT_INC(net, found);
 600                         rcu_read_unlock_bh();
 601                         return 1;
 602                 }
 603                 NF_CT_STAT_INC(net, searched);
 604         }
 605         rcu_read_unlock_bh();
 606
 607         return 0;
 608 }
 609 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 610
 611 #define NF_CT_EVICTION_RANGE    8
 612
 613 /* There's a small race here where we may free a just-assured
 614    connection.  Too bad: we're in trouble anyway. */
 615 static noinline int early_drop(struct net *net, unsigned int hash)
 616 {
 617         /* Use oldest entry, which is roughly LRU */
 618         struct nf_conntrack_tuple_hash *h;
 619         struct nf_conn *ct = NULL, *tmp;
 620         struct hlist_nulls_node *n;
 621         unsigned int i, cnt = 0;
 622         int dropped = 0;
 623
 624         rcu_read_lock();
 625         for (i = 0; i < net->ct.htable_size; i++) {
 626                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 627                                          hnnode) {
 628                         tmp = nf_ct_tuplehash_to_ctrack(h);
 629                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 630                                 ct = tmp;
 631                         cnt++;
 632                 }
 633
 634                 if (ct != NULL) {
 635                         if (likely(!nf_ct_is_dying(ct) &&
 636                                    atomic_inc_not_zero(&ct->ct_general.use)))
 637                                 break;
 638                         else
 639                                 ct = NULL;
 640                 }
 641
 642                 if (cnt >= NF_CT_EVICTION_RANGE)
 643                         break;
 644
 645                 hash = (hash + 1) % net->ct.htable_size;
 646         }
 647         rcu_read_unlock();
 648
 649         if (!ct)
 650                 return dropped;
 651
 652         if (del_timer(&ct->timeout)) {
 653                 if (nf_ct_delete(ct, 0, 0)) {
 654                         dropped = 1;
 655                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 656                 }
 657         }
 658         nf_ct_put(ct);
 659         return dropped;
 660 }
 661
 662 void init_nf_conntrack_hash_rnd(void)
 663 {
 664         unsigned int rand;
 665
 666         /*
 667          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 668          * Because there isn't enough entropy when system initializing,
 669          * and we initialize it as late as possible.
 670          */
 671         do {
 672                 get_random_bytes(&rand, sizeof(rand));
 673         } while (!rand);
 674         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 675 }
 676
 677 static struct nf_conn *
 678 __nf_conntrack_alloc(struct net *net, u16 zone,
 679                      const struct nf_conntrack_tuple *orig,
 680                      const struct nf_conntrack_tuple *repl,
 681                      gfp_t gfp, u32 hash)
 682 {
 683         struct nf_conn *ct;
 684
 685         if (unlikely(!nf_conntrack_hash_rnd)) {
 686                 init_nf_conntrack_hash_rnd();
 687                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 688                 hash = hash_conntrack_raw(orig, zone);
 689         }
 690
 691         /* We don't want any race condition at early drop stage */
 692         atomic_inc(&net->ct.count);
 693
 694         if (nf_conntrack_max &&
 695             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 696                 if (!early_drop(net, hash_bucket(hash, net))) {
 697                         atomic_dec(&net->ct.count);
 698                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
 699                         return ERR_PTR(-ENOMEM);
 700                 }
 701         }
 702
 703         /*
 704          * Do not use kmem_cache_zalloc(), as this cache uses
 705          * SLAB_DESTROY_BY_RCU.
 706          */
 707         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 708         if (ct == NULL) {
 709                 atomic_dec(&net->ct.count);
 710                 return ERR_PTR(-ENOMEM);
 711         }
 712         /*
 713          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 714          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 715          */
 716         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 717                offsetof(struct nf_conn, proto) -
 718                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 719         spin_lock_init(&ct->lock);
 720         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 721         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 722         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 723         /* save hash for reusing when confirming */
 724         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 725         /* Don't set timer yet: wait for confirmation */
 726         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 727         write_pnet(&ct->ct_net, net);
 728 #ifdef CONFIG_NF_CONNTRACK_ZONES
 729         if (zone) {
 730                 struct nf_conntrack_zone *nf_ct_zone;
 731
 732                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 733                 if (!nf_ct_zone)
 734                         goto out_free;
 735                 nf_ct_zone->id = zone;
 736         }
 737 #endif
 738         /*
 739          * changes to lookup keys must be done before setting refcnt to 1
 740          */
 741         smp_wmb();
 742         atomic_set(&ct->ct_general.use, 1);
 743         return ct;
 744
 745 #ifdef CONFIG_NF_CONNTRACK_ZONES
 746 out_free:
 747         atomic_dec(&net->ct.count);
 748         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 749         return ERR_PTR(-ENOMEM);
 750 #endif
 751 }
 752
 753 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 754                                    const struct nf_conntrack_tuple *orig,
 755                                    const struct nf_conntrack_tuple *repl,
 756                                    gfp_t gfp)
 757 {
 758         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 759 }
 760 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 761
 762 void nf_conntrack_free(struct nf_conn *ct)
 763 {
 764         struct net *net = nf_ct_net(ct);
 765
 766         nf_ct_ext_destroy(ct);
 767         atomic_dec(&net->ct.count);
 768         nf_ct_ext_free(ct);
 769         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 770 }
 771 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 772
 773
 774 /* Allocate a new conntrack: we return -ENOMEM if classification
 775    failed due to stress.  Otherwise it really is unclassifiable. */
 776 static struct nf_conntrack_tuple_hash *
 777 init_conntrack(struct net *net, struct nf_conn *tmpl,
 778                const struct nf_conntrack_tuple *tuple,
 779                struct nf_conntrack_l3proto *l3proto,
 780                struct nf_conntrack_l4proto *l4proto,
 781                struct sk_buff *skb,
 782                unsigned int dataoff, u32 hash)
 783 {
 784         struct nf_conn *ct;
 785         struct nf_conn_help *help;
 786         struct nf_conntrack_tuple repl_tuple;
 787         struct nf_conntrack_ecache *ecache;
 788         struct nf_conntrack_expect *exp;
 789         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 790         struct nf_conn_timeout *timeout_ext;
 791         unsigned int *timeouts;
 792
 793         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 794                 pr_debug("Can't invert tuple.\n");
 795                 return NULL;
 796         }
 797
 798         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 799                                   hash);
 800         if (IS_ERR(ct))
 801                 return (struct nf_conntrack_tuple_hash *)ct;
 802
 803         if (tmpl && nfct_synproxy(tmpl)) {
 804                 nfct_seqadj_ext_add(ct);
 805                 nfct_synproxy_ext_add(ct);
 806         }
 807
 808         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 809         if (timeout_ext)
 810                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 811         else
 812                 timeouts = l4proto->get_timeouts(net);
 813
 814         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 815                 nf_conntrack_free(ct);
 816                 pr_debug("init conntrack: can't track with proto module\n");
 817                 return NULL;
 818         }
 819
 820         if (timeout_ext)
 821                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 822
 823         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 824         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 825         nf_ct_labels_ext_add(ct);
 826
 827         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 828         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 829                                  ecache ? ecache->expmask : 0,
 830                              GFP_ATOMIC);
 831
 832         spin_lock_bh(&nf_conntrack_lock);
 833         exp = nf_ct_find_expectation(net, zone, tuple);
 834         if (exp) {
 835                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 836                          ct, exp);
 837                 /* Welcome, Mr. Bond.  We've been expecting you... */
 838                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 839                 ct->master = exp->master;
 840                 if (exp->helper) {
 841                         help = nf_ct_helper_ext_add(ct, exp->helper,
 842                                                     GFP_ATOMIC);
 843                         if (help)
 844                                 rcu_assign_pointer(help->helper, exp->helper);
 845                 }
 846
 847 #ifdef CONFIG_NF_CONNTRACK_MARK
 848                 ct->mark = exp->master->mark;
 849 #endif
 850 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 851                 ct->secmark = exp->master->secmark;
 852 #endif
 853                 nf_conntrack_get(&ct->master->ct_general);
 854                 NF_CT_STAT_INC(net, expect_new);
 855         } else {
 856                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 857                 NF_CT_STAT_INC(net, new);
 858         }
 859
 860         /* Overload tuple linked list to put us in unconfirmed list. */
 861         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 862                        &net->ct.unconfirmed);
 863
 864         spin_unlock_bh(&nf_conntrack_lock);
 865
 866         if (exp) {
 867                 if (exp->expectfn)
 868                         exp->expectfn(ct, exp);
 869                 nf_ct_expect_put(exp);
 870         }
 871
 872         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 873 }
 874
 875 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 876 static inline struct nf_conn *
 877 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 878                   struct sk_buff *skb,
 879                   unsigned int dataoff,
 880                   u_int16_t l3num,
 881                   u_int8_t protonum,
 882                   struct nf_conntrack_l3proto *l3proto,
 883                   struct nf_conntrack_l4proto *l4proto,
 884                   int *set_reply,
 885                   enum ip_conntrack_info *ctinfo)
 886 {
 887         struct nf_conntrack_tuple tuple;
 888         struct nf_conntrack_tuple_hash *h;
 889         struct nf_conn *ct;
 890         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 891         u32 hash;
 892
 893         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 894                              dataoff, l3num, protonum, &tuple, l3proto,
 895                              l4proto)) {
 896                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 897                 return NULL;
 898         }
 899
 900         /* look for tuple match */
 901         hash = hash_conntrack_raw(&tuple, zone);
 902         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 903         if (!h) {
 904                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 905                                    skb, dataoff, hash);
 906                 if (!h)
 907                         return NULL;
 908                 if (IS_ERR(h))
 909                         return (void *)h;
 910         }
 911         ct = nf_ct_tuplehash_to_ctrack(h);
 912
 913         /* It exists; we have (non-exclusive) reference. */
 914         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 915                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 916                 /* Please set reply bit if this packet OK */
 917                 *set_reply = 1;
 918         } else {
 919                 /* Once we've had two way comms, always ESTABLISHED. */
 920                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 921                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 922                         *ctinfo = IP_CT_ESTABLISHED;
 923                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 924                         pr_debug("nf_conntrack_in: related packet for %p\n",
 925                                  ct);
 926                         *ctinfo = IP_CT_RELATED;
 927                 } else {
 928                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 929                         *ctinfo = IP_CT_NEW;
 930                 }
 931                 *set_reply = 0;
 932         }
 933         skb->nfct = &ct->ct_general;
 934         skb->nfctinfo = *ctinfo;
 935         return ct;
 936 }
 937
 938 unsigned int
 939 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 940                 struct sk_buff *skb)
 941 {
 942         struct nf_conn *ct, *tmpl = NULL;
 943         enum ip_conntrack_info ctinfo;
 944         struct nf_conntrack_l3proto *l3proto;
 945         struct nf_conntrack_l4proto *l4proto;
 946         unsigned int *timeouts;
 947         unsigned int dataoff;
 948         u_int8_t protonum;
 949         int set_reply = 0;
 950         int ret;
 951
 952         if (skb->nfct) {
 953                 /* Previously seen (loopback or untracked)?  Ignore. */
 954                 tmpl = (struct nf_conn *)skb->nfct;
 955                 if (!nf_ct_is_template(tmpl)) {
 956                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 957                         return NF_ACCEPT;
 958                 }
 959                 skb->nfct = NULL;
 960         }
 961
 962         /* rcu_read_lock()ed by nf_hook_slow */
 963         l3proto = __nf_ct_l3proto_find(pf);
 964         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 965                                    &dataoff, &protonum);
 966         if (ret <= 0) {
 967                 pr_debug("not prepared to track yet or error occurred\n");
 968                 NF_CT_STAT_INC_ATOMIC(net, error);
 969                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 970                 ret = -ret;
 971                 goto out;
 972         }
 973
 974         l4proto = __nf_ct_l4proto_find(pf, protonum);
 975
 976         /* It may be an special packet, error, unclean...
 977          * inverse of the return code tells to the netfilter
 978          * core what to do with the packet. */
 979         if (l4proto->error != NULL) {
 980                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 981                                      pf, hooknum);
 982                 if (ret <= 0) {
 983                         NF_CT_STAT_INC_ATOMIC(net, error);
 984                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 985                         ret = -ret;
 986                         goto out;
 987                 }
 988                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 989                 if (skb->nfct)
 990                         goto out;
 991         }
 992
 993         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 994                                l3proto, l4proto, &set_reply, &ctinfo);
 995         if (!ct) {
 996                 /* Not valid part of a connection */
 997                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 998                 ret = NF_ACCEPT;
 999                 goto out;
1000         }
1001
1002         if (IS_ERR(ct)) {
1003                 /* Too stressed to deal. */
1004                 NF_CT_STAT_INC_ATOMIC(net, drop);
1005                 ret = NF_DROP;
1006                 goto out;
1007         }
1008
1009         NF_CT_ASSERT(skb->nfct);
1010
1011         /* Decide what timeout policy we want to apply to this flow. */
1012         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1013
1014         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1015         if (ret <= 0) {
1016                 /* Invalid: inverse of the return code tells
1017                  * the netfilter core what to do */
1018                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1019                 nf_conntrack_put(skb->nfct);
1020                 skb->nfct = NULL;
1021                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1022                 if (ret == -NF_DROP)
1023                         NF_CT_STAT_INC_ATOMIC(net, drop);
1024                 ret = -ret;
1025                 goto out;
1026         }
1027
1028         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1029                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1030 out:
1031         if (tmpl) {
1032                 /* Special case: we have to repeat this hook, assign the
1033                  * template again to this packet. We assume that this packet
1034                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1035                 if (ret == NF_REPEAT)
1036                         skb->nfct = (struct nf_conntrack *)tmpl;
1037                 else
1038                         nf_ct_put(tmpl);
1039         }
1040
1041         return ret;
1042 }
1043 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1044
1045 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1046                           const struct nf_conntrack_tuple *orig)
1047 {
1048         bool ret;
1049
1050         rcu_read_lock();
1051         ret = nf_ct_invert_tuple(inverse, orig,
1052                                  __nf_ct_l3proto_find(orig->src.l3num),
1053                                  __nf_ct_l4proto_find(orig->src.l3num,
1054                                                       orig->dst.protonum));
1055         rcu_read_unlock();
1056         return ret;
1057 }
1058 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1059
1060 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1061    implicitly racy: see __nf_conntrack_confirm */
1062 void nf_conntrack_alter_reply(struct nf_conn *ct,
1063                               const struct nf_conntrack_tuple *newreply)
1064 {
1065         struct nf_conn_help *help = nfct_help(ct);
1066
1067         /* Should be unconfirmed, so not in hash table yet */
1068         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1069
1070         pr_debug("Altering reply tuple of %p to ", ct);
1071         nf_ct_dump_tuple(newreply);
1072
1073         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1074         if (ct->master || (help && !hlist_empty(&help->expectations)))
1075                 return;
1076
1077         rcu_read_lock();
1078         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1079         rcu_read_unlock();
1080 }
1081 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1082
1083 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1084 void __nf_ct_refresh_acct(struct nf_conn *ct,
1085                           enum ip_conntrack_info ctinfo,
1086                           const struct sk_buff *skb,
1087                           unsigned long extra_jiffies,
1088                           int do_acct)
1089 {
1090         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1091         NF_CT_ASSERT(skb);
1092
1093         /* Only update if this is not a fixed timeout */
1094         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1095                 goto acct;
1096
1097         /* If not in hash table, timer will not be active yet */
1098         if (!nf_ct_is_confirmed(ct)) {
1099                 ct->timeout.expires = extra_jiffies;
1100         } else {
1101                 unsigned long newtime = jiffies + extra_jiffies;
1102
1103                 /* Only update the timeout if the new timeout is at least
1104                    HZ jiffies from the old timeout. Need del_timer for race
1105                    avoidance (may already be dying). */
1106                 if (newtime - ct->timeout.expires >= HZ)
1107                         mod_timer_pending(&ct->timeout, newtime);
1108         }
1109
1110 acct:
1111         if (do_acct) {
1112                 struct nf_conn_counter *acct;
1113
1114                 acct = nf_conn_acct_find(ct);
1115                 if (acct) {
1116                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1117                         atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes);
1118                 }
1119         }
1120 }
1121 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1122
1123 bool __nf_ct_kill_acct(struct nf_conn *ct,
1124                        enum ip_conntrack_info ctinfo,
1125                        const struct sk_buff *skb,
1126                        int do_acct)
1127 {
1128         if (do_acct) {
1129                 struct nf_conn_counter *acct;
1130
1131                 acct = nf_conn_acct_find(ct);
1132                 if (acct) {
1133                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1134                         atomic64_add(skb->len - skb_network_offset(skb),
1135                                      &acct[CTINFO2DIR(ctinfo)].bytes);
1136                 }
1137         }
1138
1139         if (del_timer(&ct->timeout)) {
1140                 ct->timeout.function((unsigned long)ct);
1141                 return true;
1142         }
1143         return false;
1144 }
1145 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1146
1147 #ifdef CONFIG_NF_CONNTRACK_ZONES
1148 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1149         .len    = sizeof(struct nf_conntrack_zone),
1150         .align  = __alignof__(struct nf_conntrack_zone),
1151         .id     = NF_CT_EXT_ZONE,
1152 };
1153 #endif
1154
1155 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1156
1157 #include <linux/netfilter/nfnetlink.h>
1158 #include <linux/netfilter/nfnetlink_conntrack.h>
1159 #include <linux/mutex.h>
1160
1161 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1162  * in ip_conntrack_core, since we don't want the protocols to autoload
1163  * or depend on ctnetlink */
1164 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1165                                const struct nf_conntrack_tuple *tuple)
1166 {
1167         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1168             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1169                 goto nla_put_failure;
1170         return 0;
1171
1172 nla_put_failure:
1173         return -1;
1174 }
1175 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1176
1177 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1178         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1179         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1180 };
1181 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1182
1183 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1184                                struct nf_conntrack_tuple *t)
1185 {
1186         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1187                 return -EINVAL;
1188
1189         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1190         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1191
1192         return 0;
1193 }
1194 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1195
1196 int nf_ct_port_nlattr_tuple_size(void)
1197 {
1198         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1199 }
1200 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1201 #endif
1202
1203 /* Used by ipt_REJECT and ip6t_REJECT. */
1204 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1205 {
1206         struct nf_conn *ct;
1207         enum ip_conntrack_info ctinfo;
1208
1209         /* This ICMP is in reverse direction to the packet which caused it */
1210         ct = nf_ct_get(skb, &ctinfo);
1211         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1212                 ctinfo = IP_CT_RELATED_REPLY;
1213         else
1214                 ctinfo = IP_CT_RELATED;
1215
1216         /* Attach to new skbuff, and increment count */
1217         nskb->nfct = &ct->ct_general;
1218         nskb->nfctinfo = ctinfo;
1219         nf_conntrack_get(nskb->nfct);
1220 }
1221
1222 /* Bring out ya dead! */
1223 static struct nf_conn *
1224 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1225                 void *data, unsigned int *bucket)
1226 {
1227         struct nf_conntrack_tuple_hash *h;
1228         struct nf_conn *ct;
1229         struct hlist_nulls_node *n;
1230
1231         spin_lock_bh(&nf_conntrack_lock);
1232         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1233                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1234                         if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1235                                 continue;
1236                         ct = nf_ct_tuplehash_to_ctrack(h);
1237                         if (iter(ct, data))
1238                                 goto found;
1239                 }
1240         }
1241         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1242                 ct = nf_ct_tuplehash_to_ctrack(h);
1243                 if (iter(ct, data))
1244                         set_bit(IPS_DYING_BIT, &ct->status);
1245         }
1246         spin_unlock_bh(&nf_conntrack_lock);
1247         return NULL;
1248 found:
1249         atomic_inc(&ct->ct_general.use);
1250         spin_unlock_bh(&nf_conntrack_lock);
1251         return ct;
1252 }
1253
1254 void nf_ct_iterate_cleanup(struct net *net,
1255                            int (*iter)(struct nf_conn *i, void *data),
1256                            void *data, u32 portid, int report)
1257 {
1258         struct nf_conn *ct;
1259         unsigned int bucket = 0;
1260
1261         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1262                 /* Time to push up daises... */
1263                 if (del_timer(&ct->timeout))
1264                         nf_ct_delete(ct, portid, report);
1265
1266                 /* ... else the timer will get him soon. */
1267
1268                 nf_ct_put(ct);
1269         }
1270 }
1271 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1272
1273 static int kill_all(struct nf_conn *i, void *data)
1274 {
1275         return 1;
1276 }
1277
1278 void nf_ct_free_hashtable(void *hash, unsigned int size)
1279 {
1280         if (is_vmalloc_addr(hash))
1281                 vfree(hash);
1282         else
1283                 free_pages((unsigned long)hash,
1284                            get_order(sizeof(struct hlist_head) * size));
1285 }
1286 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1287
1288 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1289 {
1290         nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1291 }
1292 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1293
1294 static void nf_ct_release_dying_list(struct net *net)
1295 {
1296         struct nf_conntrack_tuple_hash *h;
1297         struct nf_conn *ct;
1298         struct hlist_nulls_node *n;
1299
1300         spin_lock_bh(&nf_conntrack_lock);
1301         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1302                 ct = nf_ct_tuplehash_to_ctrack(h);
1303                 /* never fails to remove them, no listeners at this point */
1304                 nf_ct_kill(ct);
1305         }
1306         spin_unlock_bh(&nf_conntrack_lock);
1307 }
1308
1309 static int untrack_refs(void)
1310 {
1311         int cnt = 0, cpu;
1312
1313         for_each_possible_cpu(cpu) {
1314                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1315
1316                 cnt += atomic_read(&ct->ct_general.use) - 1;
1317         }
1318         return cnt;
1319 }
1320
1321 void nf_conntrack_cleanup_start(void)
1322 {
1323         RCU_INIT_POINTER(ip_ct_attach, NULL);
1324 }
1325
1326 void nf_conntrack_cleanup_end(void)
1327 {
1328         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1329         while (untrack_refs() > 0)
1330                 schedule();
1331
1332 #ifdef CONFIG_NF_CONNTRACK_ZONES
1333         nf_ct_extend_unregister(&nf_ct_zone_extend);
1334 #endif
1335         nf_conntrack_proto_fini();
1336         nf_conntrack_seqadj_fini();
1337         nf_conntrack_labels_fini();
1338         nf_conntrack_helper_fini();
1339         nf_conntrack_timeout_fini();
1340         nf_conntrack_ecache_fini();
1341         nf_conntrack_tstamp_fini();
1342         nf_conntrack_acct_fini();
1343         nf_conntrack_expect_fini();
1344 }
1345
1346 /*
1347  * Mishearing the voices in his head, our hero wonders how he's
1348  * supposed to kill the mall.
1349  */
1350 void nf_conntrack_cleanup_net(struct net *net)
1351 {
1352         LIST_HEAD(single);
1353
1354         list_add(&net->exit_list, &single);
1355         nf_conntrack_cleanup_net_list(&single);
1356 }
1357
1358 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1359 {
1360         int busy;
1361         struct net *net;
1362
1363         /*
1364          * This makes sure all current packets have passed through
1365          *  netfilter framework.  Roll on, two-stage module
1366          *  delete...
1367          */
1368         synchronize_net();
1369 i_see_dead_people:
1370         busy = 0;
1371         list_for_each_entry(net, net_exit_list, exit_list) {
1372                 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1373                 nf_ct_release_dying_list(net);
1374                 if (atomic_read(&net->ct.count) != 0)
1375                         busy = 1;
1376         }
1377         if (busy) {
1378                 schedule();
1379                 goto i_see_dead_people;
1380         }
1381
1382         list_for_each_entry(net, net_exit_list, exit_list) {
1383                 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1384                 nf_conntrack_proto_pernet_fini(net);
1385                 nf_conntrack_helper_pernet_fini(net);
1386                 nf_conntrack_ecache_pernet_fini(net);
1387                 nf_conntrack_tstamp_pernet_fini(net);
1388                 nf_conntrack_acct_pernet_fini(net);
1389                 nf_conntrack_expect_pernet_fini(net);
1390                 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1391                 kfree(net->ct.slabname);
1392                 free_percpu(net->ct.stat);
1393         }
1394 }
1395
1396 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1397 {
1398         struct hlist_nulls_head *hash;
1399         unsigned int nr_slots, i;
1400         size_t sz;
1401
1402         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1403         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1404         sz = nr_slots * sizeof(struct hlist_nulls_head);
1405         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1406                                         get_order(sz));
1407         if (!hash) {
1408                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1409                 hash = vzalloc(sz);
1410         }
1411
1412         if (hash && nulls)
1413                 for (i = 0; i < nr_slots; i++)
1414                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1415
1416         return hash;
1417 }
1418 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1419
1420 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1421 {
1422         int i, bucket, rc;
1423         unsigned int hashsize, old_size;
1424         struct hlist_nulls_head *hash, *old_hash;
1425         struct nf_conntrack_tuple_hash *h;
1426         struct nf_conn *ct;
1427
1428         if (current->nsproxy->net_ns != &init_net)
1429                 return -EOPNOTSUPP;
1430
1431         /* On boot, we can set this without any fancy locking. */
1432         if (!nf_conntrack_htable_size)
1433                 return param_set_uint(val, kp);
1434
1435         rc = kstrtouint(val, 0, &hashsize);
1436         if (rc)
1437                 return rc;
1438         if (!hashsize)
1439                 return -EINVAL;
1440
1441         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1442         if (!hash)
1443                 return -ENOMEM;
1444
1445         /* Lookups in the old hash might happen in parallel, which means we
1446          * might get false negatives during connection lookup. New connections
1447          * created because of a false negative won't make it into the hash
1448          * though since that required taking the lock.
1449          */
1450         spin_lock_bh(&nf_conntrack_lock);
1451         for (i = 0; i < init_net.ct.htable_size; i++) {
1452                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1453                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1454                                         struct nf_conntrack_tuple_hash, hnnode);
1455                         ct = nf_ct_tuplehash_to_ctrack(h);
1456                         hlist_nulls_del_rcu(&h->hnnode);
1457                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1458                                                   hashsize);
1459                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1460                 }
1461         }
1462         old_size = init_net.ct.htable_size;
1463         old_hash = init_net.ct.hash;
1464
1465         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1466         init_net.ct.hash = hash;
1467         spin_unlock_bh(&nf_conntrack_lock);
1468
1469         nf_ct_free_hashtable(old_hash, old_size);
1470         return 0;
1471 }
1472 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1473
1474 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1475                   &nf_conntrack_htable_size, 0600);
1476
1477 void nf_ct_untracked_status_or(unsigned long bits)
1478 {
1479         int cpu;
1480
1481         for_each_possible_cpu(cpu)
1482                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1483 }
1484 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1485
1486 int nf_conntrack_init_start(void)
1487 {
1488         int max_factor = 8;
1489         int ret, cpu;
1490
1491         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1492          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1493         if (!nf_conntrack_htable_size) {
1494                 nf_conntrack_htable_size
1495                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1496                            / sizeof(struct hlist_head));
1497                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1498                         nf_conntrack_htable_size = 16384;
1499                 if (nf_conntrack_htable_size < 32)
1500                         nf_conntrack_htable_size = 32;
1501
1502                 /* Use a max. factor of four by default to get the same max as
1503                  * with the old struct list_heads. When a table size is given
1504                  * we use the old value of 8 to avoid reducing the max.
1505                  * entries. */
1506                 max_factor = 4;
1507         }
1508         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1509
1510         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1511                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1512                nf_conntrack_max);
1513
1514         ret = nf_conntrack_expect_init();
1515         if (ret < 0)
1516                 goto err_expect;
1517
1518         ret = nf_conntrack_acct_init();
1519         if (ret < 0)
1520                 goto err_acct;
1521
1522         ret = nf_conntrack_tstamp_init();
1523         if (ret < 0)
1524                 goto err_tstamp;
1525
1526         ret = nf_conntrack_ecache_init();
1527         if (ret < 0)
1528                 goto err_ecache;
1529
1530         ret = nf_conntrack_timeout_init();
1531         if (ret < 0)
1532                 goto err_timeout;
1533
1534         ret = nf_conntrack_helper_init();
1535         if (ret < 0)
1536                 goto err_helper;
1537
1538         ret = nf_conntrack_labels_init();
1539         if (ret < 0)
1540                 goto err_labels;
1541
1542         ret = nf_conntrack_seqadj_init();
1543         if (ret < 0)
1544                 goto err_seqadj;
1545
1546 #ifdef CONFIG_NF_CONNTRACK_ZONES
1547         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1548         if (ret < 0)
1549                 goto err_extend;
1550 #endif
1551         ret = nf_conntrack_proto_init();
1552         if (ret < 0)
1553                 goto err_proto;
1554
1555         /* Set up fake conntrack: to never be deleted, not in any hashes */
1556         for_each_possible_cpu(cpu) {
1557                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1558                 write_pnet(&ct->ct_net, &init_net);
1559                 atomic_set(&ct->ct_general.use, 1);
1560         }
1561         /*  - and look it like as a confirmed connection */
1562         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1563         return 0;
1564
1565 err_proto:
1566 #ifdef CONFIG_NF_CONNTRACK_ZONES
1567         nf_ct_extend_unregister(&nf_ct_zone_extend);
1568 err_extend:
1569 #endif
1570         nf_conntrack_seqadj_fini();
1571 err_seqadj:
1572         nf_conntrack_labels_fini();
1573 err_labels:
1574         nf_conntrack_helper_fini();
1575 err_helper:
1576         nf_conntrack_timeout_fini();
1577 err_timeout:
1578         nf_conntrack_ecache_fini();
1579 err_ecache:
1580         nf_conntrack_tstamp_fini();
1581 err_tstamp:
1582         nf_conntrack_acct_fini();
1583 err_acct:
1584         nf_conntrack_expect_fini();
1585 err_expect:
1586         return ret;
1587 }
1588
1589 void nf_conntrack_init_end(void)
1590 {
1591         /* For use by REJECT target */
1592         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1593         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1594 }
1595
1596 /*
1597  * We need to use special "null" values, not used in hash table
1598  */
1599 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1600 #define DYING_NULLS_VAL         ((1<<30)+1)
1601 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1602
1603 int nf_conntrack_init_net(struct net *net)
1604 {
1605         int ret;
1606
1607         atomic_set(&net->ct.count, 0);
1608         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1609         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1610         INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1611         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1612         if (!net->ct.stat) {
1613                 ret = -ENOMEM;
1614                 goto err_stat;
1615         }
1616
1617         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1618         if (!net->ct.slabname) {
1619                 ret = -ENOMEM;
1620                 goto err_slabname;
1621         }
1622
1623         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1624                                                         sizeof(struct nf_conn), 0,
1625                                                         SLAB_DESTROY_BY_RCU, NULL);
1626         if (!net->ct.nf_conntrack_cachep) {
1627                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1628                 ret = -ENOMEM;
1629                 goto err_cache;
1630         }
1631
1632         net->ct.htable_size = nf_conntrack_htable_size;
1633         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1634         if (!net->ct.hash) {
1635                 ret = -ENOMEM;
1636                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1637                 goto err_hash;
1638         }
1639         ret = nf_conntrack_expect_pernet_init(net);
1640         if (ret < 0)
1641                 goto err_expect;
1642         ret = nf_conntrack_acct_pernet_init(net);
1643         if (ret < 0)
1644                 goto err_acct;
1645         ret = nf_conntrack_tstamp_pernet_init(net);
1646         if (ret < 0)
1647                 goto err_tstamp;
1648         ret = nf_conntrack_ecache_pernet_init(net);
1649         if (ret < 0)
1650                 goto err_ecache;
1651         ret = nf_conntrack_helper_pernet_init(net);
1652         if (ret < 0)
1653                 goto err_helper;
1654         ret = nf_conntrack_proto_pernet_init(net);
1655         if (ret < 0)
1656                 goto err_proto;
1657         return 0;
1658
1659 err_proto:
1660         nf_conntrack_helper_pernet_fini(net);
1661 err_helper:
1662         nf_conntrack_ecache_pernet_fini(net);
1663 err_ecache:
1664         nf_conntrack_tstamp_pernet_fini(net);
1665 err_tstamp:
1666         nf_conntrack_acct_pernet_fini(net);
1667 err_acct:
1668         nf_conntrack_expect_pernet_fini(net);
1669 err_expect:
1670         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1671 err_hash:
1672         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1673 err_cache:
1674         kfree(net->ct.slabname);
1675 err_slabname:
1676         free_percpu(net->ct.stat);
1677 err_stat:
1678         return ret;
1679 }