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 DEFINE_SPINLOCK(nf_conntrack_lock);
  64 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  65
  66 unsigned int nf_conntrack_htable_size __read_mostly;
  67 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  68
  69 unsigned int nf_conntrack_max __read_mostly;
  70 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  71
  72 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  73 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  74
  75 unsigned int nf_conntrack_hash_rnd __read_mostly;
  76 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
  77
  78 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  79 {
  80         unsigned int n;
  81
  82         /* The direction must be ignored, so we hash everything up to the
  83          * destination ports (which is a multiple of 4) and treat the last
  84          * three bytes manually.
  85          */
  86         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  87         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  88                       (((__force __u16)tuple->dst.u.all << 16) |
  89                       tuple->dst.protonum));
  90 }
  91
  92 static u32 __hash_bucket(u32 hash, unsigned int size)
  93 {
  94         return ((u64)hash * size) >> 32;
  95 }
  96
  97 static u32 hash_bucket(u32 hash, const struct net *net)
  98 {
  99         return __hash_bucket(hash, net->ct.htable_size);
 100 }
 101
 102 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
 103                                   u16 zone, unsigned int size)
 104 {
 105         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 106 }
 107
 108 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 109                                        const struct nf_conntrack_tuple *tuple)
 110 {
 111         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 112 }
 113
 114 bool
 115 nf_ct_get_tuple(const struct sk_buff *skb,
 116                 unsigned int nhoff,
 117                 unsigned int dataoff,
 118                 u_int16_t l3num,
 119                 u_int8_t protonum,
 120                 struct nf_conntrack_tuple *tuple,
 121                 const struct nf_conntrack_l3proto *l3proto,
 122                 const struct nf_conntrack_l4proto *l4proto)
 123 {
 124         memset(tuple, 0, sizeof(*tuple));
 125
 126         tuple->src.l3num = l3num;
 127         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 128                 return false;
 129
 130         tuple->dst.protonum = protonum;
 131         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 132
 133         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 134 }
 135 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 136
 137 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 138                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 139 {
 140         struct nf_conntrack_l3proto *l3proto;
 141         struct nf_conntrack_l4proto *l4proto;
 142         unsigned int protoff;
 143         u_int8_t protonum;
 144         int ret;
 145
 146         rcu_read_lock();
 147
 148         l3proto = __nf_ct_l3proto_find(l3num);
 149         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 150         if (ret != NF_ACCEPT) {
 151                 rcu_read_unlock();
 152                 return false;
 153         }
 154
 155         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 156
 157         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 158                               l3proto, l4proto);
 159
 160         rcu_read_unlock();
 161         return ret;
 162 }
 163 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 164
 165 bool
 166 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 167                    const struct nf_conntrack_tuple *orig,
 168                    const struct nf_conntrack_l3proto *l3proto,
 169                    const struct nf_conntrack_l4proto *l4proto)
 170 {
 171         memset(inverse, 0, sizeof(*inverse));
 172
 173         inverse->src.l3num = orig->src.l3num;
 174         if (l3proto->invert_tuple(inverse, orig) == 0)
 175                 return false;
 176
 177         inverse->dst.dir = !orig->dst.dir;
 178
 179         inverse->dst.protonum = orig->dst.protonum;
 180         return l4proto->invert_tuple(inverse, orig);
 181 }
 182 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 183
 184 static void
 185 clean_from_lists(struct nf_conn *ct)
 186 {
 187         pr_debug("clean_from_lists(%p)\n", ct);
 188         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 189         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 190
 191         /* Destroy all pending expectations */
 192         nf_ct_remove_expectations(ct);
 193 }
 194
 195 static void
 196 destroy_conntrack(struct nf_conntrack *nfct)
 197 {
 198         struct nf_conn *ct = (struct nf_conn *)nfct;
 199         struct net *net = nf_ct_net(ct);
 200         struct nf_conntrack_l4proto *l4proto;
 201
 202         pr_debug("destroy_conntrack(%p)\n", ct);
 203         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 204         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 205
 206         /* To make sure we don't get any weird locking issues here:
 207          * destroy_conntrack() MUST NOT be called with a write lock
 208          * to nf_conntrack_lock!!! -HW */
 209         rcu_read_lock();
 210         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 211         if (l4proto && l4proto->destroy)
 212                 l4proto->destroy(ct);
 213
 214         rcu_read_unlock();
 215
 216         spin_lock_bh(&nf_conntrack_lock);
 217         /* Expectations will have been removed in clean_from_lists,
 218          * except TFTP can create an expectation on the first packet,
 219          * before connection is in the list, so we need to clean here,
 220          * too. */
 221         nf_ct_remove_expectations(ct);
 222
 223         /* We overload first tuple to link into unconfirmed or dying list.*/
 224         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 225         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 226
 227         NF_CT_STAT_INC(net, delete);
 228         spin_unlock_bh(&nf_conntrack_lock);
 229
 230         if (ct->master)
 231                 nf_ct_put(ct->master);
 232
 233         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 234         nf_conntrack_free(ct);
 235 }
 236
 237 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 238 {
 239         struct net *net = nf_ct_net(ct);
 240
 241         nf_ct_helper_destroy(ct);
 242         spin_lock_bh(&nf_conntrack_lock);
 243         /* Inside lock so preempt is disabled on module removal path.
 244          * Otherwise we can get spurious warnings. */
 245         NF_CT_STAT_INC(net, delete_list);
 246         clean_from_lists(ct);
 247         /* add this conntrack to the dying list */
 248         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 249                              &net->ct.dying);
 250         spin_unlock_bh(&nf_conntrack_lock);
 251 }
 252
 253 static void death_by_event(unsigned long ul_conntrack)
 254 {
 255         struct nf_conn *ct = (void *)ul_conntrack;
 256         struct net *net = nf_ct_net(ct);
 257         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 258
 259         BUG_ON(ecache == NULL);
 260
 261         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 262                 /* bad luck, let's retry again */
 263                 ecache->timeout.expires = jiffies +
 264                         (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 265                 add_timer(&ecache->timeout);
 266                 return;
 267         }
 268         /* we've got the event delivered, now it's dying */
 269         set_bit(IPS_DYING_BIT, &ct->status);
 270         nf_ct_put(ct);
 271 }
 272
 273 static void nf_ct_dying_timeout(struct nf_conn *ct)
 274 {
 275         struct net *net = nf_ct_net(ct);
 276         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 277
 278         BUG_ON(ecache == NULL);
 279
 280         /* set a new timer to retry event delivery */
 281         setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
 282         ecache->timeout.expires = jiffies +
 283                 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 284         add_timer(&ecache->timeout);
 285 }
 286
 287 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 288 {
 289         struct nf_conn_tstamp *tstamp;
 290
 291         tstamp = nf_conn_tstamp_find(ct);
 292         if (tstamp && tstamp->stop == 0)
 293                 tstamp->stop = ktime_to_ns(ktime_get_real());
 294
 295         if (!nf_ct_is_dying(ct) &&
 296             unlikely(nf_conntrack_event_report(IPCT_DESTROY, ct,
 297             portid, report) < 0)) {
 298                 /* destroy event was not delivered */
 299                 nf_ct_delete_from_lists(ct);
 300                 nf_ct_dying_timeout(ct);
 301                 return false;
 302         }
 303         set_bit(IPS_DYING_BIT, &ct->status);
 304         nf_ct_delete_from_lists(ct);
 305         nf_ct_put(ct);
 306         return true;
 307 }
 308 EXPORT_SYMBOL_GPL(nf_ct_delete);
 309
 310 static void death_by_timeout(unsigned long ul_conntrack)
 311 {
 312         nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
 313 }
 314
 315 static inline bool
 316 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
 317                         const struct nf_conntrack_tuple *tuple,
 318                         u16 zone)
 319 {
 320         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 321
 322         /* A conntrack can be recreated with the equal tuple,
 323          * so we need to check that the conntrack is confirmed
 324          */
 325         return nf_ct_tuple_equal(tuple, &h->tuple) &&
 326                 nf_ct_zone(ct) == zone &&
 327                 nf_ct_is_confirmed(ct);
 328 }
 329
 330 /*
 331  * Warning :
 332  * - Caller must take a reference on returned object
 333  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 334  * OR
 335  * - Caller must lock nf_conntrack_lock before calling this function
 336  */
 337 static struct nf_conntrack_tuple_hash *
 338 ____nf_conntrack_find(struct net *net, u16 zone,
 339                       const struct nf_conntrack_tuple *tuple, u32 hash)
 340 {
 341         struct nf_conntrack_tuple_hash *h;
 342         struct hlist_nulls_node *n;
 343         unsigned int bucket = hash_bucket(hash, net);
 344
 345         /* Disable BHs the entire time since we normally need to disable them
 346          * at least once for the stats anyway.
 347          */
 348         local_bh_disable();
 349 begin:
 350         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 351                 if (nf_ct_key_equal(h, tuple, zone)) {
 352                         NF_CT_STAT_INC(net, found);
 353                         local_bh_enable();
 354                         return h;
 355                 }
 356                 NF_CT_STAT_INC(net, searched);
 357         }
 358         /*
 359          * if the nulls value we got at the end of this lookup is
 360          * not the expected one, we must restart lookup.
 361          * We probably met an item that was moved to another chain.
 362          */
 363         if (get_nulls_value(n) != bucket) {
 364                 NF_CT_STAT_INC(net, search_restart);
 365                 goto begin;
 366         }
 367         local_bh_enable();
 368
 369         return NULL;
 370 }
 371
 372 /* Find a connection corresponding to a tuple. */
 373 static struct nf_conntrack_tuple_hash *
 374 __nf_conntrack_find_get(struct net *net, u16 zone,
 375                         const struct nf_conntrack_tuple *tuple, u32 hash)
 376 {
 377         struct nf_conntrack_tuple_hash *h;
 378         struct nf_conn *ct;
 379
 380         rcu_read_lock();
 381 begin:
 382         h = ____nf_conntrack_find(net, zone, tuple, hash);
 383         if (h) {
 384                 ct = nf_ct_tuplehash_to_ctrack(h);
 385                 if (unlikely(nf_ct_is_dying(ct) ||
 386                              !atomic_inc_not_zero(&ct->ct_general.use)))
 387                         h = NULL;
 388                 else {
 389                         if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
 390                                 nf_ct_put(ct);
 391                                 goto begin;
 392                         }
 393                 }
 394         }
 395         rcu_read_unlock();
 396
 397         return h;
 398 }
 399
 400 struct nf_conntrack_tuple_hash *
 401 nf_conntrack_find_get(struct net *net, u16 zone,
 402                       const struct nf_conntrack_tuple *tuple)
 403 {
 404         return __nf_conntrack_find_get(net, zone, tuple,
 405                                        hash_conntrack_raw(tuple, zone));
 406 }
 407 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 408
 409 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 410                                        unsigned int hash,
 411                                        unsigned int repl_hash)
 412 {
 413         struct net *net = nf_ct_net(ct);
 414
 415         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 416                            &net->ct.hash[hash]);
 417         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 418                            &net->ct.hash[repl_hash]);
 419 }
 420
 421 int
 422 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 423 {
 424         struct net *net = nf_ct_net(ct);
 425         unsigned int hash, repl_hash;
 426         struct nf_conntrack_tuple_hash *h;
 427         struct hlist_nulls_node *n;
 428         u16 zone;
 429
 430         zone = nf_ct_zone(ct);
 431         hash = hash_conntrack(net, zone,
 432                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 433         repl_hash = hash_conntrack(net, zone,
 434                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 435
 436         spin_lock_bh(&nf_conntrack_lock);
 437
 438         /* See if there's one in the list already, including reverse */
 439         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 440                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 441                                       &h->tuple) &&
 442                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 443                         goto out;
 444         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 445                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 446                                       &h->tuple) &&
 447                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 448                         goto out;
 449
 450         add_timer(&ct->timeout);
 451         nf_conntrack_get(&ct->ct_general);
 452         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 453         NF_CT_STAT_INC(net, insert);
 454         spin_unlock_bh(&nf_conntrack_lock);
 455
 456         return 0;
 457
 458 out:
 459         NF_CT_STAT_INC(net, insert_failed);
 460         spin_unlock_bh(&nf_conntrack_lock);
 461         return -EEXIST;
 462 }
 463 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 464
 465 /* Confirm a connection given skb; places it in hash table */
 466 int
 467 __nf_conntrack_confirm(struct sk_buff *skb)
 468 {
 469         unsigned int hash, repl_hash;
 470         struct nf_conntrack_tuple_hash *h;
 471         struct nf_conn *ct;
 472         struct nf_conn_help *help;
 473         struct nf_conn_tstamp *tstamp;
 474         struct hlist_nulls_node *n;
 475         enum ip_conntrack_info ctinfo;
 476         struct net *net;
 477         u16 zone;
 478
 479         ct = nf_ct_get(skb, &ctinfo);
 480         net = nf_ct_net(ct);
 481
 482         /* ipt_REJECT uses nf_conntrack_attach to attach related
 483            ICMP/TCP RST packets in other direction.  Actual packet
 484            which created connection will be IP_CT_NEW or for an
 485            expected connection, IP_CT_RELATED. */
 486         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 487                 return NF_ACCEPT;
 488
 489         zone = nf_ct_zone(ct);
 490         /* reuse the hash saved before */
 491         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 492         hash = hash_bucket(hash, net);
 493         repl_hash = hash_conntrack(net, zone,
 494                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 495
 496         /* We're not in hash table, and we refuse to set up related
 497            connections for unconfirmed conns.  But packet copies and
 498            REJECT will give spurious warnings here. */
 499         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 500
 501         /* No external references means no one else could have
 502            confirmed us. */
 503         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 504         pr_debug("Confirming conntrack %p\n", ct);
 505
 506         spin_lock_bh(&nf_conntrack_lock);
 507
 508         /* We have to check the DYING flag inside the lock to prevent
 509            a race against nf_ct_get_next_corpse() possibly called from
 510            user context, else we insert an already 'dead' hash, blocking
 511            further use of that particular connection -JM */
 512
 513         if (unlikely(nf_ct_is_dying(ct))) {
 514                 spin_unlock_bh(&nf_conntrack_lock);
 515                 return NF_ACCEPT;
 516         }
 517
 518         /* See if there's one in the list already, including reverse:
 519            NAT could have grabbed it without realizing, since we're
 520            not in the hash.  If there is, we lost race. */
 521         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 522                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 523                                       &h->tuple) &&
 524                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 525                         goto out;
 526         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 527                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 528                                       &h->tuple) &&
 529                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 530                         goto out;
 531
 532         /* Remove from unconfirmed list */
 533         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 534
 535         /* Timer relative to confirmation time, not original
 536            setting time, otherwise we'd get timer wrap in
 537            weird delay cases. */
 538         ct->timeout.expires += jiffies;
 539         add_timer(&ct->timeout);
 540         atomic_inc(&ct->ct_general.use);
 541         ct->status |= IPS_CONFIRMED;
 542
 543         /* set conntrack timestamp, if enabled. */
 544         tstamp = nf_conn_tstamp_find(ct);
 545         if (tstamp) {
 546                 if (skb->tstamp.tv64 == 0)
 547                         __net_timestamp(skb);
 548
 549                 tstamp->start = ktime_to_ns(skb->tstamp);
 550         }
 551         /* Since the lookup is lockless, hash insertion must be done after
 552          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 553          * guarantee that no other CPU can find the conntrack before the above
 554          * stores are visible.
 555          */
 556         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 557         NF_CT_STAT_INC(net, insert);
 558         spin_unlock_bh(&nf_conntrack_lock);
 559
 560         help = nfct_help(ct);
 561         if (help && help->helper)
 562                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 563
 564         nf_conntrack_event_cache(master_ct(ct) ?
 565                                  IPCT_RELATED : IPCT_NEW, ct);
 566         return NF_ACCEPT;
 567
 568 out:
 569         NF_CT_STAT_INC(net, insert_failed);
 570         spin_unlock_bh(&nf_conntrack_lock);
 571         return NF_DROP;
 572 }
 573 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 574
 575 /* Returns true if a connection correspondings to the tuple (required
 576    for NAT). */
 577 int
 578 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 579                          const struct nf_conn *ignored_conntrack)
 580 {
 581         struct net *net = nf_ct_net(ignored_conntrack);
 582         struct nf_conntrack_tuple_hash *h;
 583         struct hlist_nulls_node *n;
 584         struct nf_conn *ct;
 585         u16 zone = nf_ct_zone(ignored_conntrack);
 586         unsigned int hash = hash_conntrack(net, zone, tuple);
 587
 588         /* Disable BHs the entire time since we need to disable them at
 589          * least once for the stats anyway.
 590          */
 591         rcu_read_lock_bh();
 592         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 593                 ct = nf_ct_tuplehash_to_ctrack(h);
 594                 if (ct != ignored_conntrack &&
 595                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 596                     nf_ct_zone(ct) == zone) {
 597                         NF_CT_STAT_INC(net, found);
 598                         rcu_read_unlock_bh();
 599                         return 1;
 600                 }
 601                 NF_CT_STAT_INC(net, searched);
 602         }
 603         rcu_read_unlock_bh();
 604
 605         return 0;
 606 }
 607 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 608
 609 #define NF_CT_EVICTION_RANGE    8
 610
 611 /* There's a small race here where we may free a just-assured
 612    connection.  Too bad: we're in trouble anyway. */
 613 static noinline int early_drop(struct net *net, unsigned int hash)
 614 {
 615         /* Use oldest entry, which is roughly LRU */
 616         struct nf_conntrack_tuple_hash *h;
 617         struct nf_conn *ct = NULL, *tmp;
 618         struct hlist_nulls_node *n;
 619         unsigned int i, cnt = 0;
 620         int dropped = 0;
 621
 622         rcu_read_lock();
 623         for (i = 0; i < net->ct.htable_size; i++) {
 624                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 625                                          hnnode) {
 626                         tmp = nf_ct_tuplehash_to_ctrack(h);
 627                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 628                                 ct = tmp;
 629                         cnt++;
 630                 }
 631
 632                 if (ct != NULL) {
 633                         if (likely(!nf_ct_is_dying(ct) &&
 634                                    atomic_inc_not_zero(&ct->ct_general.use)))
 635                                 break;
 636                         else
 637                                 ct = NULL;
 638                 }
 639
 640                 if (cnt >= NF_CT_EVICTION_RANGE)
 641                         break;
 642
 643                 hash = (hash + 1) % net->ct.htable_size;
 644         }
 645         rcu_read_unlock();
 646
 647         if (!ct)
 648                 return dropped;
 649
 650         if (del_timer(&ct->timeout)) {
 651                 if (nf_ct_delete(ct, 0, 0)) {
 652                         dropped = 1;
 653                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 654                 }
 655         }
 656         nf_ct_put(ct);
 657         return dropped;
 658 }
 659
 660 void init_nf_conntrack_hash_rnd(void)
 661 {
 662         unsigned int rand;
 663
 664         /*
 665          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 666          * Because there isn't enough entropy when system initializing,
 667          * and we initialize it as late as possible.
 668          */
 669         do {
 670                 get_random_bytes(&rand, sizeof(rand));
 671         } while (!rand);
 672         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 673 }
 674
 675 static struct nf_conn *
 676 __nf_conntrack_alloc(struct net *net, u16 zone,
 677                      const struct nf_conntrack_tuple *orig,
 678                      const struct nf_conntrack_tuple *repl,
 679                      gfp_t gfp, u32 hash)
 680 {
 681         struct nf_conn *ct;
 682
 683         if (unlikely(!nf_conntrack_hash_rnd)) {
 684                 init_nf_conntrack_hash_rnd();
 685                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 686                 hash = hash_conntrack_raw(orig, zone);
 687         }
 688
 689         /* We don't want any race condition at early drop stage */
 690         atomic_inc(&net->ct.count);
 691
 692         if (nf_conntrack_max &&
 693             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 694                 if (!early_drop(net, hash_bucket(hash, net))) {
 695                         atomic_dec(&net->ct.count);
 696                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
 697                         return ERR_PTR(-ENOMEM);
 698                 }
 699         }
 700
 701         /*
 702          * Do not use kmem_cache_zalloc(), as this cache uses
 703          * SLAB_DESTROY_BY_RCU.
 704          */
 705         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 706         if (ct == NULL) {
 707                 atomic_dec(&net->ct.count);
 708                 return ERR_PTR(-ENOMEM);
 709         }
 710         /*
 711          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 712          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 713          */
 714         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 715                offsetof(struct nf_conn, proto) -
 716                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 717         spin_lock_init(&ct->lock);
 718         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 719         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 720         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 721         /* save hash for reusing when confirming */
 722         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 723         /* Don't set timer yet: wait for confirmation */
 724         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 725         write_pnet(&ct->ct_net, net);
 726 #ifdef CONFIG_NF_CONNTRACK_ZONES
 727         if (zone) {
 728                 struct nf_conntrack_zone *nf_ct_zone;
 729
 730                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 731                 if (!nf_ct_zone)
 732                         goto out_free;
 733                 nf_ct_zone->id = zone;
 734         }
 735 #endif
 736         /*
 737          * changes to lookup keys must be done before setting refcnt to 1
 738          */
 739         smp_wmb();
 740         atomic_set(&ct->ct_general.use, 1);
 741         return ct;
 742
 743 #ifdef CONFIG_NF_CONNTRACK_ZONES
 744 out_free:
 745         atomic_dec(&net->ct.count);
 746         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 747         return ERR_PTR(-ENOMEM);
 748 #endif
 749 }
 750
 751 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 752                                    const struct nf_conntrack_tuple *orig,
 753                                    const struct nf_conntrack_tuple *repl,
 754                                    gfp_t gfp)
 755 {
 756         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 757 }
 758 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 759
 760 void nf_conntrack_free(struct nf_conn *ct)
 761 {
 762         struct net *net = nf_ct_net(ct);
 763
 764         nf_ct_ext_destroy(ct);
 765         nf_ct_ext_free(ct);
 766         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 767         smp_mb__before_atomic_dec();
 768         atomic_dec(&net->ct.count);
 769 }
 770 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 771
 772
 773 /* Allocate a new conntrack: we return -ENOMEM if classification
 774    failed due to stress.  Otherwise it really is unclassifiable. */
 775 static struct nf_conntrack_tuple_hash *
 776 init_conntrack(struct net *net, struct nf_conn *tmpl,
 777                const struct nf_conntrack_tuple *tuple,
 778                struct nf_conntrack_l3proto *l3proto,
 779                struct nf_conntrack_l4proto *l4proto,
 780                struct sk_buff *skb,
 781                unsigned int dataoff, u32 hash)
 782 {
 783         struct nf_conn *ct;
 784         struct nf_conn_help *help;
 785         struct nf_conntrack_tuple repl_tuple;
 786         struct nf_conntrack_ecache *ecache;
 787         struct nf_conntrack_expect *exp;
 788         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 789         struct nf_conn_timeout *timeout_ext;
 790         unsigned int *timeouts;
 791
 792         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 793                 pr_debug("Can't invert tuple.\n");
 794                 return NULL;
 795         }
 796
 797         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 798                                   hash);
 799         if (IS_ERR(ct))
 800                 return (struct nf_conntrack_tuple_hash *)ct;
 801
 802         if (tmpl && nfct_synproxy(tmpl)) {
 803                 nfct_seqadj_ext_add(ct);
 804                 nfct_synproxy_ext_add(ct);
 805         }
 806
 807         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 808         if (timeout_ext)
 809                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 810         else
 811                 timeouts = l4proto->get_timeouts(net);
 812
 813         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 814                 nf_conntrack_free(ct);
 815                 pr_debug("init conntrack: can't track with proto module\n");
 816                 return NULL;
 817         }
 818
 819         if (timeout_ext)
 820                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 821
 822         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 823         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 824         nf_ct_labels_ext_add(ct);
 825
 826         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 827         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 828                                  ecache ? ecache->expmask : 0,
 829                              GFP_ATOMIC);
 830
 831         spin_lock_bh(&nf_conntrack_lock);
 832         exp = nf_ct_find_expectation(net, zone, tuple);
 833         if (exp) {
 834                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 835                          ct, exp);
 836                 /* Welcome, Mr. Bond.  We've been expecting you... */
 837                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 838                 ct->master = exp->master;
 839                 if (exp->helper) {
 840                         help = nf_ct_helper_ext_add(ct, exp->helper,
 841                                                     GFP_ATOMIC);
 842                         if (help)
 843                                 rcu_assign_pointer(help->helper, exp->helper);
 844                 }
 845
 846 #ifdef CONFIG_NF_CONNTRACK_MARK
 847                 ct->mark = exp->master->mark;
 848 #endif
 849 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 850                 ct->secmark = exp->master->secmark;
 851 #endif
 852                 nf_conntrack_get(&ct->master->ct_general);
 853                 NF_CT_STAT_INC(net, expect_new);
 854         } else {
 855                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 856                 NF_CT_STAT_INC(net, new);
 857         }
 858
 859         /* Overload tuple linked list to put us in unconfirmed list. */
 860         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 861                        &net->ct.unconfirmed);
 862
 863         spin_unlock_bh(&nf_conntrack_lock);
 864
 865         if (exp) {
 866                 if (exp->expectfn)
 867                         exp->expectfn(ct, exp);
 868                 nf_ct_expect_put(exp);
 869         }
 870
 871         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 872 }
 873
 874 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 875 static inline struct nf_conn *
 876 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 877                   struct sk_buff *skb,
 878                   unsigned int dataoff,
 879                   u_int16_t l3num,
 880                   u_int8_t protonum,
 881                   struct nf_conntrack_l3proto *l3proto,
 882                   struct nf_conntrack_l4proto *l4proto,
 883                   int *set_reply,
 884                   enum ip_conntrack_info *ctinfo)
 885 {
 886         struct nf_conntrack_tuple tuple;
 887         struct nf_conntrack_tuple_hash *h;
 888         struct nf_conn *ct;
 889         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 890         u32 hash;
 891
 892         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 893                              dataoff, l3num, protonum, &tuple, l3proto,
 894                              l4proto)) {
 895                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 896                 return NULL;
 897         }
 898
 899         /* look for tuple match */
 900         hash = hash_conntrack_raw(&tuple, zone);
 901         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 902         if (!h) {
 903                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 904                                    skb, dataoff, hash);
 905                 if (!h)
 906                         return NULL;
 907                 if (IS_ERR(h))
 908                         return (void *)h;
 909         }
 910         ct = nf_ct_tuplehash_to_ctrack(h);
 911
 912         /* It exists; we have (non-exclusive) reference. */
 913         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 914                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 915                 /* Please set reply bit if this packet OK */
 916                 *set_reply = 1;
 917         } else {
 918                 /* Once we've had two way comms, always ESTABLISHED. */
 919                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 920                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 921                         *ctinfo = IP_CT_ESTABLISHED;
 922                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 923                         pr_debug("nf_conntrack_in: related packet for %p\n",
 924                                  ct);
 925                         *ctinfo = IP_CT_RELATED;
 926                 } else {
 927                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 928                         *ctinfo = IP_CT_NEW;
 929                 }
 930                 *set_reply = 0;
 931         }
 932         skb->nfct = &ct->ct_general;
 933         skb->nfctinfo = *ctinfo;
 934         return ct;
 935 }
 936
 937 unsigned int
 938 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 939                 struct sk_buff *skb)
 940 {
 941         struct nf_conn *ct, *tmpl = NULL;
 942         enum ip_conntrack_info ctinfo;
 943         struct nf_conntrack_l3proto *l3proto;
 944         struct nf_conntrack_l4proto *l4proto;
 945         unsigned int *timeouts;
 946         unsigned int dataoff;
 947         u_int8_t protonum;
 948         int set_reply = 0;
 949         int ret;
 950
 951         if (skb->nfct) {
 952                 /* Previously seen (loopback or untracked)?  Ignore. */
 953                 tmpl = (struct nf_conn *)skb->nfct;
 954                 if (!nf_ct_is_template(tmpl)) {
 955                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 956                         return NF_ACCEPT;
 957                 }
 958                 skb->nfct = NULL;
 959         }
 960
 961         /* rcu_read_lock()ed by nf_hook_slow */
 962         l3proto = __nf_ct_l3proto_find(pf);
 963         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 964                                    &dataoff, &protonum);
 965         if (ret <= 0) {
 966                 pr_debug("not prepared to track yet or error occurred\n");
 967                 NF_CT_STAT_INC_ATOMIC(net, error);
 968                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 969                 ret = -ret;
 970                 goto out;
 971         }
 972
 973         l4proto = __nf_ct_l4proto_find(pf, protonum);
 974
 975         /* It may be an special packet, error, unclean...
 976          * inverse of the return code tells to the netfilter
 977          * core what to do with the packet. */
 978         if (l4proto->error != NULL) {
 979                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 980                                      pf, hooknum);
 981                 if (ret <= 0) {
 982                         NF_CT_STAT_INC_ATOMIC(net, error);
 983                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 984                         ret = -ret;
 985                         goto out;
 986                 }
 987                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 988                 if (skb->nfct)
 989                         goto out;
 990         }
 991
 992         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 993                                l3proto, l4proto, &set_reply, &ctinfo);
 994         if (!ct) {
 995                 /* Not valid part of a connection */
 996                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 997                 ret = NF_ACCEPT;
 998                 goto out;
 999         }
1000
1001         if (IS_ERR(ct)) {
1002                 /* Too stressed to deal. */
1003                 NF_CT_STAT_INC_ATOMIC(net, drop);
1004                 ret = NF_DROP;
1005                 goto out;
1006         }
1007
1008         NF_CT_ASSERT(skb->nfct);
1009
1010         /* Decide what timeout policy we want to apply to this flow. */
1011         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1012
1013         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1014         if (ret <= 0) {
1015                 /* Invalid: inverse of the return code tells
1016                  * the netfilter core what to do */
1017                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1018                 nf_conntrack_put(skb->nfct);
1019                 skb->nfct = NULL;
1020                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1021                 if (ret == -NF_DROP)
1022                         NF_CT_STAT_INC_ATOMIC(net, drop);
1023                 ret = -ret;
1024                 goto out;
1025         }
1026
1027         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1028                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1029 out:
1030         if (tmpl) {
1031                 /* Special case: we have to repeat this hook, assign the
1032                  * template again to this packet. We assume that this packet
1033                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1034                 if (ret == NF_REPEAT)
1035                         skb->nfct = (struct nf_conntrack *)tmpl;
1036                 else
1037                         nf_ct_put(tmpl);
1038         }
1039
1040         return ret;
1041 }
1042 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1043
1044 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1045                           const struct nf_conntrack_tuple *orig)
1046 {
1047         bool ret;
1048
1049         rcu_read_lock();
1050         ret = nf_ct_invert_tuple(inverse, orig,
1051                                  __nf_ct_l3proto_find(orig->src.l3num),
1052                                  __nf_ct_l4proto_find(orig->src.l3num,
1053                                                       orig->dst.protonum));
1054         rcu_read_unlock();
1055         return ret;
1056 }
1057 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1058
1059 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1060    implicitly racy: see __nf_conntrack_confirm */
1061 void nf_conntrack_alter_reply(struct nf_conn *ct,
1062                               const struct nf_conntrack_tuple *newreply)
1063 {
1064         struct nf_conn_help *help = nfct_help(ct);
1065
1066         /* Should be unconfirmed, so not in hash table yet */
1067         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1068
1069         pr_debug("Altering reply tuple of %p to ", ct);
1070         nf_ct_dump_tuple(newreply);
1071
1072         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1073         if (ct->master || (help && !hlist_empty(&help->expectations)))
1074                 return;
1075
1076         rcu_read_lock();
1077         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1078         rcu_read_unlock();
1079 }
1080 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1081
1082 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1083 void __nf_ct_refresh_acct(struct nf_conn *ct,
1084                           enum ip_conntrack_info ctinfo,
1085                           const struct sk_buff *skb,
1086                           unsigned long extra_jiffies,
1087                           int do_acct)
1088 {
1089         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1090         NF_CT_ASSERT(skb);
1091
1092         /* Only update if this is not a fixed timeout */
1093         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1094                 goto acct;
1095
1096         /* If not in hash table, timer will not be active yet */
1097         if (!nf_ct_is_confirmed(ct)) {
1098                 ct->timeout.expires = extra_jiffies;
1099         } else {
1100                 unsigned long newtime = jiffies + extra_jiffies;
1101
1102                 /* Only update the timeout if the new timeout is at least
1103                    HZ jiffies from the old timeout. Need del_timer for race
1104                    avoidance (may already be dying). */
1105                 if (newtime - ct->timeout.expires >= HZ)
1106                         mod_timer_pending(&ct->timeout, newtime);
1107         }
1108
1109 acct:
1110         if (do_acct) {
1111                 struct nf_conn_acct *acct;
1112
1113                 acct = nf_conn_acct_find(ct);
1114                 if (acct) {
1115                         struct nf_conn_counter *counter = acct->counter;
1116
1117                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1118                         atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1119                 }
1120         }
1121 }
1122 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1123
1124 bool __nf_ct_kill_acct(struct nf_conn *ct,
1125                        enum ip_conntrack_info ctinfo,
1126                        const struct sk_buff *skb,
1127                        int do_acct)
1128 {
1129         if (do_acct) {
1130                 struct nf_conn_acct *acct;
1131
1132                 acct = nf_conn_acct_find(ct);
1133                 if (acct) {
1134                         struct nf_conn_counter *counter = acct->counter;
1135
1136                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1137                         atomic64_add(skb->len - skb_network_offset(skb),
1138                                      &counter[CTINFO2DIR(ctinfo)].bytes);
1139                 }
1140         }
1141
1142         if (del_timer(&ct->timeout)) {
1143                 ct->timeout.function((unsigned long)ct);
1144                 return true;
1145         }
1146         return false;
1147 }
1148 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1149
1150 #ifdef CONFIG_NF_CONNTRACK_ZONES
1151 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1152         .len    = sizeof(struct nf_conntrack_zone),
1153         .align  = __alignof__(struct nf_conntrack_zone),
1154         .id     = NF_CT_EXT_ZONE,
1155 };
1156 #endif
1157
1158 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1159
1160 #include <linux/netfilter/nfnetlink.h>
1161 #include <linux/netfilter/nfnetlink_conntrack.h>
1162 #include <linux/mutex.h>
1163
1164 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1165  * in ip_conntrack_core, since we don't want the protocols to autoload
1166  * or depend on ctnetlink */
1167 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1168                                const struct nf_conntrack_tuple *tuple)
1169 {
1170         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1171             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1172                 goto nla_put_failure;
1173         return 0;
1174
1175 nla_put_failure:
1176         return -1;
1177 }
1178 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1179
1180 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1181         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1182         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1183 };
1184 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1185
1186 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1187                                struct nf_conntrack_tuple *t)
1188 {
1189         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1190                 return -EINVAL;
1191
1192         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1193         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1194
1195         return 0;
1196 }
1197 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1198
1199 int nf_ct_port_nlattr_tuple_size(void)
1200 {
1201         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1202 }
1203 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1204 #endif
1205
1206 /* Used by ipt_REJECT and ip6t_REJECT. */
1207 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1208 {
1209         struct nf_conn *ct;
1210         enum ip_conntrack_info ctinfo;
1211
1212         /* This ICMP is in reverse direction to the packet which caused it */
1213         ct = nf_ct_get(skb, &ctinfo);
1214         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1215                 ctinfo = IP_CT_RELATED_REPLY;
1216         else
1217                 ctinfo = IP_CT_RELATED;
1218
1219         /* Attach to new skbuff, and increment count */
1220         nskb->nfct = &ct->ct_general;
1221         nskb->nfctinfo = ctinfo;
1222         nf_conntrack_get(nskb->nfct);
1223 }
1224
1225 /* Bring out ya dead! */
1226 static struct nf_conn *
1227 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1228                 void *data, unsigned int *bucket)
1229 {
1230         struct nf_conntrack_tuple_hash *h;
1231         struct nf_conn *ct;
1232         struct hlist_nulls_node *n;
1233
1234         spin_lock_bh(&nf_conntrack_lock);
1235         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1236                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1237                         if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1238                                 continue;
1239                         ct = nf_ct_tuplehash_to_ctrack(h);
1240                         if (iter(ct, data))
1241                                 goto found;
1242                 }
1243         }
1244         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1245                 ct = nf_ct_tuplehash_to_ctrack(h);
1246                 if (iter(ct, data))
1247                         set_bit(IPS_DYING_BIT, &ct->status);
1248         }
1249         spin_unlock_bh(&nf_conntrack_lock);
1250         return NULL;
1251 found:
1252         atomic_inc(&ct->ct_general.use);
1253         spin_unlock_bh(&nf_conntrack_lock);
1254         return ct;
1255 }
1256
1257 void nf_ct_iterate_cleanup(struct net *net,
1258                            int (*iter)(struct nf_conn *i, void *data),
1259                            void *data, u32 portid, int report)
1260 {
1261         struct nf_conn *ct;
1262         unsigned int bucket = 0;
1263
1264         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1265                 /* Time to push up daises... */
1266                 if (del_timer(&ct->timeout))
1267                         nf_ct_delete(ct, portid, report);
1268
1269                 /* ... else the timer will get him soon. */
1270
1271                 nf_ct_put(ct);
1272         }
1273 }
1274 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1275
1276 static int kill_all(struct nf_conn *i, void *data)
1277 {
1278         return 1;
1279 }
1280
1281 void nf_ct_free_hashtable(void *hash, unsigned int size)
1282 {
1283         if (is_vmalloc_addr(hash))
1284                 vfree(hash);
1285         else
1286                 free_pages((unsigned long)hash,
1287                            get_order(sizeof(struct hlist_head) * size));
1288 }
1289 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1290
1291 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1292 {
1293         nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1294 }
1295 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1296
1297 static void nf_ct_release_dying_list(struct net *net)
1298 {
1299         struct nf_conntrack_tuple_hash *h;
1300         struct nf_conn *ct;
1301         struct hlist_nulls_node *n;
1302
1303         spin_lock_bh(&nf_conntrack_lock);
1304         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1305                 ct = nf_ct_tuplehash_to_ctrack(h);
1306                 /* never fails to remove them, no listeners at this point */
1307                 nf_ct_kill(ct);
1308         }
1309         spin_unlock_bh(&nf_conntrack_lock);
1310 }
1311
1312 static int untrack_refs(void)
1313 {
1314         int cnt = 0, cpu;
1315
1316         for_each_possible_cpu(cpu) {
1317                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1318
1319                 cnt += atomic_read(&ct->ct_general.use) - 1;
1320         }
1321         return cnt;
1322 }
1323
1324 void nf_conntrack_cleanup_start(void)
1325 {
1326         RCU_INIT_POINTER(ip_ct_attach, NULL);
1327 }
1328
1329 void nf_conntrack_cleanup_end(void)
1330 {
1331         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1332         while (untrack_refs() > 0)
1333                 schedule();
1334
1335 #ifdef CONFIG_NF_CONNTRACK_ZONES
1336         nf_ct_extend_unregister(&nf_ct_zone_extend);
1337 #endif
1338         nf_conntrack_proto_fini();
1339         nf_conntrack_seqadj_fini();
1340         nf_conntrack_labels_fini();
1341         nf_conntrack_helper_fini();
1342         nf_conntrack_timeout_fini();
1343         nf_conntrack_ecache_fini();
1344         nf_conntrack_tstamp_fini();
1345         nf_conntrack_acct_fini();
1346         nf_conntrack_expect_fini();
1347 }
1348
1349 /*
1350  * Mishearing the voices in his head, our hero wonders how he's
1351  * supposed to kill the mall.
1352  */
1353 void nf_conntrack_cleanup_net(struct net *net)
1354 {
1355         LIST_HEAD(single);
1356
1357         list_add(&net->exit_list, &single);
1358         nf_conntrack_cleanup_net_list(&single);
1359 }
1360
1361 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1362 {
1363         int busy;
1364         struct net *net;
1365
1366         /*
1367          * This makes sure all current packets have passed through
1368          *  netfilter framework.  Roll on, two-stage module
1369          *  delete...
1370          */
1371         synchronize_net();
1372 i_see_dead_people:
1373         busy = 0;
1374         list_for_each_entry(net, net_exit_list, exit_list) {
1375                 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1376                 nf_ct_release_dying_list(net);
1377                 if (atomic_read(&net->ct.count) != 0)
1378                         busy = 1;
1379         }
1380         if (busy) {
1381                 schedule();
1382                 goto i_see_dead_people;
1383         }
1384
1385         list_for_each_entry(net, net_exit_list, exit_list) {
1386                 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1387                 nf_conntrack_proto_pernet_fini(net);
1388                 nf_conntrack_helper_pernet_fini(net);
1389                 nf_conntrack_ecache_pernet_fini(net);
1390                 nf_conntrack_tstamp_pernet_fini(net);
1391                 nf_conntrack_acct_pernet_fini(net);
1392                 nf_conntrack_expect_pernet_fini(net);
1393                 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1394                 kfree(net->ct.slabname);
1395                 free_percpu(net->ct.stat);
1396         }
1397 }
1398
1399 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1400 {
1401         struct hlist_nulls_head *hash;
1402         unsigned int nr_slots, i;
1403         size_t sz;
1404
1405         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1406         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1407         sz = nr_slots * sizeof(struct hlist_nulls_head);
1408         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1409                                         get_order(sz));
1410         if (!hash) {
1411                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1412                 hash = vzalloc(sz);
1413         }
1414
1415         if (hash && nulls)
1416                 for (i = 0; i < nr_slots; i++)
1417                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1418
1419         return hash;
1420 }
1421 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1422
1423 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1424 {
1425         int i, bucket, rc;
1426         unsigned int hashsize, old_size;
1427         struct hlist_nulls_head *hash, *old_hash;
1428         struct nf_conntrack_tuple_hash *h;
1429         struct nf_conn *ct;
1430
1431         if (current->nsproxy->net_ns != &init_net)
1432                 return -EOPNOTSUPP;
1433
1434         /* On boot, we can set this without any fancy locking. */
1435         if (!nf_conntrack_htable_size)
1436                 return param_set_uint(val, kp);
1437
1438         rc = kstrtouint(val, 0, &hashsize);
1439         if (rc)
1440                 return rc;
1441         if (!hashsize)
1442                 return -EINVAL;
1443
1444         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1445         if (!hash)
1446                 return -ENOMEM;
1447
1448         /* Lookups in the old hash might happen in parallel, which means we
1449          * might get false negatives during connection lookup. New connections
1450          * created because of a false negative won't make it into the hash
1451          * though since that required taking the lock.
1452          */
1453         spin_lock_bh(&nf_conntrack_lock);
1454         for (i = 0; i < init_net.ct.htable_size; i++) {
1455                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1456                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1457                                         struct nf_conntrack_tuple_hash, hnnode);
1458                         ct = nf_ct_tuplehash_to_ctrack(h);
1459                         hlist_nulls_del_rcu(&h->hnnode);
1460                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1461                                                   hashsize);
1462                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1463                 }
1464         }
1465         old_size = init_net.ct.htable_size;
1466         old_hash = init_net.ct.hash;
1467
1468         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1469         init_net.ct.hash = hash;
1470         spin_unlock_bh(&nf_conntrack_lock);
1471
1472         nf_ct_free_hashtable(old_hash, old_size);
1473         return 0;
1474 }
1475 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1476
1477 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1478                   &nf_conntrack_htable_size, 0600);
1479
1480 void nf_ct_untracked_status_or(unsigned long bits)
1481 {
1482         int cpu;
1483
1484         for_each_possible_cpu(cpu)
1485                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1486 }
1487 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1488
1489 int nf_conntrack_init_start(void)
1490 {
1491         int max_factor = 8;
1492         int ret, cpu;
1493
1494         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1495          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1496         if (!nf_conntrack_htable_size) {
1497                 nf_conntrack_htable_size
1498                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1499                            / sizeof(struct hlist_head));
1500                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1501                         nf_conntrack_htable_size = 16384;
1502                 if (nf_conntrack_htable_size < 32)
1503                         nf_conntrack_htable_size = 32;
1504
1505                 /* Use a max. factor of four by default to get the same max as
1506                  * with the old struct list_heads. When a table size is given
1507                  * we use the old value of 8 to avoid reducing the max.
1508                  * entries. */
1509                 max_factor = 4;
1510         }
1511         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1512
1513         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1514                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1515                nf_conntrack_max);
1516
1517         ret = nf_conntrack_expect_init();
1518         if (ret < 0)
1519                 goto err_expect;
1520
1521         ret = nf_conntrack_acct_init();
1522         if (ret < 0)
1523                 goto err_acct;
1524
1525         ret = nf_conntrack_tstamp_init();
1526         if (ret < 0)
1527                 goto err_tstamp;
1528
1529         ret = nf_conntrack_ecache_init();
1530         if (ret < 0)
1531                 goto err_ecache;
1532
1533         ret = nf_conntrack_timeout_init();
1534         if (ret < 0)
1535                 goto err_timeout;
1536
1537         ret = nf_conntrack_helper_init();
1538         if (ret < 0)
1539                 goto err_helper;
1540
1541         ret = nf_conntrack_labels_init();
1542         if (ret < 0)
1543                 goto err_labels;
1544
1545         ret = nf_conntrack_seqadj_init();
1546         if (ret < 0)
1547                 goto err_seqadj;
1548
1549 #ifdef CONFIG_NF_CONNTRACK_ZONES
1550         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1551         if (ret < 0)
1552                 goto err_extend;
1553 #endif
1554         ret = nf_conntrack_proto_init();
1555         if (ret < 0)
1556                 goto err_proto;
1557
1558         /* Set up fake conntrack: to never be deleted, not in any hashes */
1559         for_each_possible_cpu(cpu) {
1560                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1561                 write_pnet(&ct->ct_net, &init_net);
1562                 atomic_set(&ct->ct_general.use, 1);
1563         }
1564         /*  - and look it like as a confirmed connection */
1565         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1566         return 0;
1567
1568 err_proto:
1569 #ifdef CONFIG_NF_CONNTRACK_ZONES
1570         nf_ct_extend_unregister(&nf_ct_zone_extend);
1571 err_extend:
1572 #endif
1573         nf_conntrack_seqadj_fini();
1574 err_seqadj:
1575         nf_conntrack_labels_fini();
1576 err_labels:
1577         nf_conntrack_helper_fini();
1578 err_helper:
1579         nf_conntrack_timeout_fini();
1580 err_timeout:
1581         nf_conntrack_ecache_fini();
1582 err_ecache:
1583         nf_conntrack_tstamp_fini();
1584 err_tstamp:
1585         nf_conntrack_acct_fini();
1586 err_acct:
1587         nf_conntrack_expect_fini();
1588 err_expect:
1589         return ret;
1590 }
1591
1592 void nf_conntrack_init_end(void)
1593 {
1594         /* For use by REJECT target */
1595         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1596         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1597 }
1598
1599 /*
1600  * We need to use special "null" values, not used in hash table
1601  */
1602 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1603 #define DYING_NULLS_VAL         ((1<<30)+1)
1604 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1605
1606 int nf_conntrack_init_net(struct net *net)
1607 {
1608         int ret;
1609
1610         atomic_set(&net->ct.count, 0);
1611         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1612         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1613         INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1614         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1615         if (!net->ct.stat) {
1616                 ret = -ENOMEM;
1617                 goto err_stat;
1618         }
1619
1620         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1621         if (!net->ct.slabname) {
1622                 ret = -ENOMEM;
1623                 goto err_slabname;
1624         }
1625
1626         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1627                                                         sizeof(struct nf_conn), 0,
1628                                                         SLAB_DESTROY_BY_RCU, NULL);
1629         if (!net->ct.nf_conntrack_cachep) {
1630                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1631                 ret = -ENOMEM;
1632                 goto err_cache;
1633         }
1634
1635         net->ct.htable_size = nf_conntrack_htable_size;
1636         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1637         if (!net->ct.hash) {
1638                 ret = -ENOMEM;
1639                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1640                 goto err_hash;
1641         }
1642         ret = nf_conntrack_expect_pernet_init(net);
1643         if (ret < 0)
1644                 goto err_expect;
1645         ret = nf_conntrack_acct_pernet_init(net);
1646         if (ret < 0)
1647                 goto err_acct;
1648         ret = nf_conntrack_tstamp_pernet_init(net);
1649         if (ret < 0)
1650                 goto err_tstamp;
1651         ret = nf_conntrack_ecache_pernet_init(net);
1652         if (ret < 0)
1653                 goto err_ecache;
1654         ret = nf_conntrack_helper_pernet_init(net);
1655         if (ret < 0)
1656                 goto err_helper;
1657         ret = nf_conntrack_proto_pernet_init(net);
1658         if (ret < 0)
1659                 goto err_proto;
1660         return 0;
1661
1662 err_proto:
1663         nf_conntrack_helper_pernet_fini(net);
1664 err_helper:
1665         nf_conntrack_ecache_pernet_fini(net);
1666 err_ecache:
1667         nf_conntrack_tstamp_pernet_fini(net);
1668 err_tstamp:
1669         nf_conntrack_acct_pernet_fini(net);
1670 err_acct:
1671         nf_conntrack_expect_pernet_fini(net);
1672 err_expect:
1673         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1674 err_hash:
1675         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1676 err_cache:
1677         kfree(net->ct.slabname);
1678 err_slabname:
1679         free_percpu(net->ct.stat);
1680 err_stat:
1681         return ret;
1682 }