net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License version 2 as
  11  * published by the Free Software Foundation.
  12  */
  13
  14 #include <linux/types.h>
  15 #include <linux/netfilter.h>
  16 #include <linux/module.h>
  17 #include <linux/sched.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/proc_fs.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/stddef.h>
  22 #include <linux/slab.h>
  23 #include <linux/random.h>
  24 #include <linux/jhash.h>
  25 #include <linux/err.h>
  26 #include <linux/percpu.h>
  27 #include <linux/moduleparam.h>
  28 #include <linux/notifier.h>
  29 #include <linux/kernel.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/socket.h>
  32 #include <linux/mm.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/rculist_nulls.h>
  35
  36 #include <net/netfilter/nf_conntrack.h>
  37 #include <net/netfilter/nf_conntrack_l3proto.h>
  38 #include <net/netfilter/nf_conntrack_l4proto.h>
  39 #include <net/netfilter/nf_conntrack_expect.h>
  40 #include <net/netfilter/nf_conntrack_helper.h>
  41 #include <net/netfilter/nf_conntrack_core.h>
  42 #include <net/netfilter/nf_conntrack_extend.h>
  43 #include <net/netfilter/nf_conntrack_acct.h>
  44 #include <net/netfilter/nf_conntrack_ecache.h>
  45 #include <net/netfilter/nf_conntrack_zones.h>
  46 #include <net/netfilter/nf_conntrack_timestamp.h>
  47 #include <net/netfilter/nf_nat.h>
  48 #include <net/netfilter/nf_nat_core.h>
  49
  50 #define NF_CONNTRACK_VERSION    "0.5.0"
  51
  52 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  53                                       enum nf_nat_manip_type manip,
  54                                       const struct nlattr *attr) __read_mostly;
  55 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  56
  57 DEFINE_SPINLOCK(nf_conntrack_lock);
  58 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  59
  60 unsigned int nf_conntrack_htable_size __read_mostly;
  61 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  62
  63 unsigned int nf_conntrack_max __read_mostly;
  64 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  65
  66 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  67 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  68
  69 unsigned int nf_conntrack_hash_rnd __read_mostly;
  70 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
  71
  72 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  73 {
  74         unsigned int n;
  75
  76         /* The direction must be ignored, so we hash everything up to the
  77          * destination ports (which is a multiple of 4) and treat the last
  78          * three bytes manually.
  79          */
  80         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  81         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  82                       (((__force __u16)tuple->dst.u.all << 16) |
  83                       tuple->dst.protonum));
  84 }
  85
  86 static u32 __hash_bucket(u32 hash, unsigned int size)
  87 {
  88         return ((u64)hash * size) >> 32;
  89 }
  90
  91 static u32 hash_bucket(u32 hash, const struct net *net)
  92 {
  93         return __hash_bucket(hash, net->ct.htable_size);
  94 }
  95
  96 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  97                                   u16 zone, unsigned int size)
  98 {
  99         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 100 }
 101
 102 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 103                                        const struct nf_conntrack_tuple *tuple)
 104 {
 105         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 106 }
 107
 108 bool
 109 nf_ct_get_tuple(const struct sk_buff *skb,
 110                 unsigned int nhoff,
 111                 unsigned int dataoff,
 112                 u_int16_t l3num,
 113                 u_int8_t protonum,
 114                 struct nf_conntrack_tuple *tuple,
 115                 const struct nf_conntrack_l3proto *l3proto,
 116                 const struct nf_conntrack_l4proto *l4proto)
 117 {
 118         memset(tuple, 0, sizeof(*tuple));
 119
 120         tuple->src.l3num = l3num;
 121         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 122                 return false;
 123
 124         tuple->dst.protonum = protonum;
 125         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 126
 127         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 128 }
 129 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 130
 131 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 132                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 133 {
 134         struct nf_conntrack_l3proto *l3proto;
 135         struct nf_conntrack_l4proto *l4proto;
 136         unsigned int protoff;
 137         u_int8_t protonum;
 138         int ret;
 139
 140         rcu_read_lock();
 141
 142         l3proto = __nf_ct_l3proto_find(l3num);
 143         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 144         if (ret != NF_ACCEPT) {
 145                 rcu_read_unlock();
 146                 return false;
 147         }
 148
 149         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 150
 151         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 152                               l3proto, l4proto);
 153
 154         rcu_read_unlock();
 155         return ret;
 156 }
 157 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 158
 159 bool
 160 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 161                    const struct nf_conntrack_tuple *orig,
 162                    const struct nf_conntrack_l3proto *l3proto,
 163                    const struct nf_conntrack_l4proto *l4proto)
 164 {
 165         memset(inverse, 0, sizeof(*inverse));
 166
 167         inverse->src.l3num = orig->src.l3num;
 168         if (l3proto->invert_tuple(inverse, orig) == 0)
 169                 return false;
 170
 171         inverse->dst.dir = !orig->dst.dir;
 172
 173         inverse->dst.protonum = orig->dst.protonum;
 174         return l4proto->invert_tuple(inverse, orig);
 175 }
 176 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 177
 178 static void
 179 clean_from_lists(struct nf_conn *ct)
 180 {
 181         pr_debug("clean_from_lists(%p)\n", ct);
 182         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 183         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 184
 185         /* Destroy all pending expectations */
 186         nf_ct_remove_expectations(ct);
 187 }
 188
 189 static void
 190 destroy_conntrack(struct nf_conntrack *nfct)
 191 {
 192         struct nf_conn *ct = (struct nf_conn *)nfct;
 193         struct net *net = nf_ct_net(ct);
 194         struct nf_conntrack_l4proto *l4proto;
 195
 196         pr_debug("destroy_conntrack(%p)\n", ct);
 197         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 198         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 199
 200         /* To make sure we don't get any weird locking issues here:
 201          * destroy_conntrack() MUST NOT be called with a write lock
 202          * to nf_conntrack_lock!!! -HW */
 203         rcu_read_lock();
 204         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 205         if (l4proto && l4proto->destroy)
 206                 l4proto->destroy(ct);
 207
 208         rcu_read_unlock();
 209
 210         spin_lock_bh(&nf_conntrack_lock);
 211         /* Expectations will have been removed in clean_from_lists,
 212          * except TFTP can create an expectation on the first packet,
 213          * before connection is in the list, so we need to clean here,
 214          * too. */
 215         nf_ct_remove_expectations(ct);
 216
 217         /* We overload first tuple to link into unconfirmed list. */
 218         if (!nf_ct_is_confirmed(ct)) {
 219                 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 220                 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 221         }
 222
 223         NF_CT_STAT_INC(net, delete);
 224         spin_unlock_bh(&nf_conntrack_lock);
 225
 226         if (ct->master)
 227                 nf_ct_put(ct->master);
 228
 229         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 230         nf_conntrack_free(ct);
 231 }
 232
 233 void nf_ct_delete_from_lists(struct nf_conn *ct)
 234 {
 235         struct net *net = nf_ct_net(ct);
 236
 237         nf_ct_helper_destroy(ct);
 238         spin_lock_bh(&nf_conntrack_lock);
 239         /* Inside lock so preempt is disabled on module removal path.
 240          * Otherwise we can get spurious warnings. */
 241         NF_CT_STAT_INC(net, delete_list);
 242         clean_from_lists(ct);
 243         spin_unlock_bh(&nf_conntrack_lock);
 244 }
 245 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
 246
 247 static void death_by_event(unsigned long ul_conntrack)
 248 {
 249         struct nf_conn *ct = (void *)ul_conntrack;
 250         struct net *net = nf_ct_net(ct);
 251
 252         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 253                 /* bad luck, let's retry again */
 254                 ct->timeout.expires = jiffies +
 255                         (random32() % net->ct.sysctl_events_retry_timeout);
 256                 add_timer(&ct->timeout);
 257                 return;
 258         }
 259         /* we've got the event delivered, now it's dying */
 260         set_bit(IPS_DYING_BIT, &ct->status);
 261         spin_lock(&nf_conntrack_lock);
 262         hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 263         spin_unlock(&nf_conntrack_lock);
 264         nf_ct_put(ct);
 265 }
 266
 267 void nf_ct_insert_dying_list(struct nf_conn *ct)
 268 {
 269         struct net *net = nf_ct_net(ct);
 270
 271         /* add this conntrack to the dying list */
 272         spin_lock_bh(&nf_conntrack_lock);
 273         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 274                              &net->ct.dying);
 275         spin_unlock_bh(&nf_conntrack_lock);
 276         /* set a new timer to retry event delivery */
 277         setup_timer(&ct->timeout, death_by_event, (unsigned long)ct);
 278         ct->timeout.expires = jiffies +
 279                 (random32() % net->ct.sysctl_events_retry_timeout);
 280         add_timer(&ct->timeout);
 281 }
 282 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list);
 283
 284 static void death_by_timeout(unsigned long ul_conntrack)
 285 {
 286         struct nf_conn *ct = (void *)ul_conntrack;
 287         struct nf_conn_tstamp *tstamp;
 288
 289         tstamp = nf_conn_tstamp_find(ct);
 290         if (tstamp && tstamp->stop == 0)
 291                 tstamp->stop = ktime_to_ns(ktime_get_real());
 292
 293         if (!test_bit(IPS_DYING_BIT, &ct->status) &&
 294             unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
 295                 /* destroy event was not delivered */
 296                 nf_ct_delete_from_lists(ct);
 297                 nf_ct_insert_dying_list(ct);
 298                 return;
 299         }
 300         set_bit(IPS_DYING_BIT, &ct->status);
 301         nf_ct_delete_from_lists(ct);
 302         nf_ct_put(ct);
 303 }
 304
 305 /*
 306  * Warning :
 307  * - Caller must take a reference on returned object
 308  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 309  * OR
 310  * - Caller must lock nf_conntrack_lock before calling this function
 311  */
 312 static struct nf_conntrack_tuple_hash *
 313 ____nf_conntrack_find(struct net *net, u16 zone,
 314                       const struct nf_conntrack_tuple *tuple, u32 hash)
 315 {
 316         struct nf_conntrack_tuple_hash *h;
 317         struct hlist_nulls_node *n;
 318         unsigned int bucket = hash_bucket(hash, net);
 319
 320         /* Disable BHs the entire time since we normally need to disable them
 321          * at least once for the stats anyway.
 322          */
 323         local_bh_disable();
 324 begin:
 325         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 326                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 327                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 328                         NF_CT_STAT_INC(net, found);
 329                         local_bh_enable();
 330                         return h;
 331                 }
 332                 NF_CT_STAT_INC(net, searched);
 333         }
 334         /*
 335          * if the nulls value we got at the end of this lookup is
 336          * not the expected one, we must restart lookup.
 337          * We probably met an item that was moved to another chain.
 338          */
 339         if (get_nulls_value(n) != bucket) {
 340                 NF_CT_STAT_INC(net, search_restart);
 341                 goto begin;
 342         }
 343         local_bh_enable();
 344
 345         return NULL;
 346 }
 347
 348 struct nf_conntrack_tuple_hash *
 349 __nf_conntrack_find(struct net *net, u16 zone,
 350                     const struct nf_conntrack_tuple *tuple)
 351 {
 352         return ____nf_conntrack_find(net, zone, tuple,
 353                                      hash_conntrack_raw(tuple, zone));
 354 }
 355 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 356
 357 /* Find a connection corresponding to a tuple. */
 358 static struct nf_conntrack_tuple_hash *
 359 __nf_conntrack_find_get(struct net *net, u16 zone,
 360                         const struct nf_conntrack_tuple *tuple, u32 hash)
 361 {
 362         struct nf_conntrack_tuple_hash *h;
 363         struct nf_conn *ct;
 364
 365         rcu_read_lock();
 366 begin:
 367         h = ____nf_conntrack_find(net, zone, tuple, hash);
 368         if (h) {
 369                 ct = nf_ct_tuplehash_to_ctrack(h);
 370                 if (unlikely(nf_ct_is_dying(ct) ||
 371                              !atomic_inc_not_zero(&ct->ct_general.use)))
 372                         h = NULL;
 373                 else {
 374                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 375                                      nf_ct_zone(ct) != zone)) {
 376                                 nf_ct_put(ct);
 377                                 goto begin;
 378                         }
 379                 }
 380         }
 381         rcu_read_unlock();
 382
 383         return h;
 384 }
 385
 386 struct nf_conntrack_tuple_hash *
 387 nf_conntrack_find_get(struct net *net, u16 zone,
 388                       const struct nf_conntrack_tuple *tuple)
 389 {
 390         return __nf_conntrack_find_get(net, zone, tuple,
 391                                        hash_conntrack_raw(tuple, zone));
 392 }
 393 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 394
 395 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 396                                        unsigned int hash,
 397                                        unsigned int repl_hash)
 398 {
 399         struct net *net = nf_ct_net(ct);
 400
 401         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 402                            &net->ct.hash[hash]);
 403         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 404                            &net->ct.hash[repl_hash]);
 405 }
 406
 407 void nf_conntrack_hash_insert(struct nf_conn *ct)
 408 {
 409         struct net *net = nf_ct_net(ct);
 410         unsigned int hash, repl_hash;
 411         u16 zone;
 412
 413         zone = nf_ct_zone(ct);
 414         hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 415         repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 416
 417         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 418 }
 419 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 420
 421 /* Confirm a connection given skb; places it in hash table */
 422 int
 423 __nf_conntrack_confirm(struct sk_buff *skb)
 424 {
 425         unsigned int hash, repl_hash;
 426         struct nf_conntrack_tuple_hash *h;
 427         struct nf_conn *ct;
 428         struct nf_conn_help *help;
 429         struct nf_conn_tstamp *tstamp;
 430         struct hlist_nulls_node *n;
 431         enum ip_conntrack_info ctinfo;
 432         struct net *net;
 433         u16 zone;
 434
 435         ct = nf_ct_get(skb, &ctinfo);
 436         net = nf_ct_net(ct);
 437
 438         /* ipt_REJECT uses nf_conntrack_attach to attach related
 439            ICMP/TCP RST packets in other direction.  Actual packet
 440            which created connection will be IP_CT_NEW or for an
 441            expected connection, IP_CT_RELATED. */
 442         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 443                 return NF_ACCEPT;
 444
 445         zone = nf_ct_zone(ct);
 446         /* reuse the hash saved before */
 447         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 448         hash = hash_bucket(hash, net);
 449         repl_hash = hash_conntrack(net, zone,
 450                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 451
 452         /* We're not in hash table, and we refuse to set up related
 453            connections for unconfirmed conns.  But packet copies and
 454            REJECT will give spurious warnings here. */
 455         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 456
 457         /* No external references means no one else could have
 458            confirmed us. */
 459         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 460         pr_debug("Confirming conntrack %p\n", ct);
 461
 462         spin_lock_bh(&nf_conntrack_lock);
 463
 464         /* We have to check the DYING flag inside the lock to prevent
 465            a race against nf_ct_get_next_corpse() possibly called from
 466            user context, else we insert an already 'dead' hash, blocking
 467            further use of that particular connection -JM */
 468
 469         if (unlikely(nf_ct_is_dying(ct))) {
 470                 spin_unlock_bh(&nf_conntrack_lock);
 471                 return NF_ACCEPT;
 472         }
 473
 474         /* See if there's one in the list already, including reverse:
 475            NAT could have grabbed it without realizing, since we're
 476            not in the hash.  If there is, we lost race. */
 477         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 478                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 479                                       &h->tuple) &&
 480                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 481                         goto out;
 482         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 483                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 484                                       &h->tuple) &&
 485                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 486                         goto out;
 487
 488         /* Remove from unconfirmed list */
 489         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 490
 491         /* Timer relative to confirmation time, not original
 492            setting time, otherwise we'd get timer wrap in
 493            weird delay cases. */
 494         ct->timeout.expires += jiffies;
 495         add_timer(&ct->timeout);
 496         atomic_inc(&ct->ct_general.use);
 497         ct->status |= IPS_CONFIRMED;
 498
 499         /* set conntrack timestamp, if enabled. */
 500         tstamp = nf_conn_tstamp_find(ct);
 501         if (tstamp) {
 502                 if (skb->tstamp.tv64 == 0)
 503                         __net_timestamp((struct sk_buff *)skb);
 504
 505                 tstamp->start = ktime_to_ns(skb->tstamp);
 506         }
 507         /* Since the lookup is lockless, hash insertion must be done after
 508          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 509          * guarantee that no other CPU can find the conntrack before the above
 510          * stores are visible.
 511          */
 512         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 513         NF_CT_STAT_INC(net, insert);
 514         spin_unlock_bh(&nf_conntrack_lock);
 515
 516         help = nfct_help(ct);
 517         if (help && help->helper)
 518                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 519
 520         nf_conntrack_event_cache(master_ct(ct) ?
 521                                  IPCT_RELATED : IPCT_NEW, ct);
 522         return NF_ACCEPT;
 523
 524 out:
 525         NF_CT_STAT_INC(net, insert_failed);
 526         spin_unlock_bh(&nf_conntrack_lock);
 527         return NF_DROP;
 528 }
 529 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 530
 531 /* Returns true if a connection correspondings to the tuple (required
 532    for NAT). */
 533 int
 534 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 535                          const struct nf_conn *ignored_conntrack)
 536 {
 537         struct net *net = nf_ct_net(ignored_conntrack);
 538         struct nf_conntrack_tuple_hash *h;
 539         struct hlist_nulls_node *n;
 540         struct nf_conn *ct;
 541         u16 zone = nf_ct_zone(ignored_conntrack);
 542         unsigned int hash = hash_conntrack(net, zone, tuple);
 543
 544         /* Disable BHs the entire time since we need to disable them at
 545          * least once for the stats anyway.
 546          */
 547         rcu_read_lock_bh();
 548         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 549                 ct = nf_ct_tuplehash_to_ctrack(h);
 550                 if (ct != ignored_conntrack &&
 551                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 552                     nf_ct_zone(ct) == zone) {
 553                         NF_CT_STAT_INC(net, found);
 554                         rcu_read_unlock_bh();
 555                         return 1;
 556                 }
 557                 NF_CT_STAT_INC(net, searched);
 558         }
 559         rcu_read_unlock_bh();
 560
 561         return 0;
 562 }
 563 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 564
 565 #define NF_CT_EVICTION_RANGE    8
 566
 567 /* There's a small race here where we may free a just-assured
 568    connection.  Too bad: we're in trouble anyway. */
 569 static noinline int early_drop(struct net *net, unsigned int hash)
 570 {
 571         /* Use oldest entry, which is roughly LRU */
 572         struct nf_conntrack_tuple_hash *h;
 573         struct nf_conn *ct = NULL, *tmp;
 574         struct hlist_nulls_node *n;
 575         unsigned int i, cnt = 0;
 576         int dropped = 0;
 577
 578         rcu_read_lock();
 579         for (i = 0; i < net->ct.htable_size; i++) {
 580                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 581                                          hnnode) {
 582                         tmp = nf_ct_tuplehash_to_ctrack(h);
 583                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 584                                 ct = tmp;
 585                         cnt++;
 586                 }
 587
 588                 if (ct != NULL) {
 589                         if (likely(!nf_ct_is_dying(ct) &&
 590                                    atomic_inc_not_zero(&ct->ct_general.use)))
 591                                 break;
 592                         else
 593                                 ct = NULL;
 594                 }
 595
 596                 if (cnt >= NF_CT_EVICTION_RANGE)
 597                         break;
 598
 599                 hash = (hash + 1) % net->ct.htable_size;
 600         }
 601         rcu_read_unlock();
 602
 603         if (!ct)
 604                 return dropped;
 605
 606         if (del_timer(&ct->timeout)) {
 607                 death_by_timeout((unsigned long)ct);
 608                 dropped = 1;
 609                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 610         }
 611         nf_ct_put(ct);
 612         return dropped;
 613 }
 614
 615 void init_nf_conntrack_hash_rnd(void)
 616 {
 617         unsigned int rand;
 618
 619         /*
 620          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 621          * Because there isn't enough entropy when system initializing,
 622          * and we initialize it as late as possible.
 623          */
 624         do {
 625                 get_random_bytes(&rand, sizeof(rand));
 626         } while (!rand);
 627         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 628 }
 629
 630 static struct nf_conn *
 631 __nf_conntrack_alloc(struct net *net, u16 zone,
 632                      const struct nf_conntrack_tuple *orig,
 633                      const struct nf_conntrack_tuple *repl,
 634                      gfp_t gfp, u32 hash)
 635 {
 636         struct nf_conn *ct;
 637
 638         if (unlikely(!nf_conntrack_hash_rnd)) {
 639                 init_nf_conntrack_hash_rnd();
 640                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 641                 hash = hash_conntrack_raw(orig, zone);
 642         }
 643
 644         /* We don't want any race condition at early drop stage */
 645         atomic_inc(&net->ct.count);
 646
 647         if (nf_conntrack_max &&
 648             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 649                 if (!early_drop(net, hash_bucket(hash, net))) {
 650                         atomic_dec(&net->ct.count);
 651                         if (net_ratelimit())
 652                                 printk(KERN_WARNING
 653                                        "nf_conntrack: table full, dropping"
 654                                        " packet.\n");
 655                         return ERR_PTR(-ENOMEM);
 656                 }
 657         }
 658
 659         /*
 660          * Do not use kmem_cache_zalloc(), as this cache uses
 661          * SLAB_DESTROY_BY_RCU.
 662          */
 663         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 664         if (ct == NULL) {
 665                 atomic_dec(&net->ct.count);
 666                 return ERR_PTR(-ENOMEM);
 667         }
 668         /*
 669          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 670          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 671          */
 672         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 673                offsetof(struct nf_conn, proto) -
 674                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 675         spin_lock_init(&ct->lock);
 676         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 677         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 678         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 679         /* save hash for reusing when confirming */
 680         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 681         /* Don't set timer yet: wait for confirmation */
 682         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 683         write_pnet(&ct->ct_net, net);
 684 #ifdef CONFIG_NF_CONNTRACK_ZONES
 685         if (zone) {
 686                 struct nf_conntrack_zone *nf_ct_zone;
 687
 688                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 689                 if (!nf_ct_zone)
 690                         goto out_free;
 691                 nf_ct_zone->id = zone;
 692         }
 693 #endif
 694         /*
 695          * changes to lookup keys must be done before setting refcnt to 1
 696          */
 697         smp_wmb();
 698         atomic_set(&ct->ct_general.use, 1);
 699         return ct;
 700
 701 #ifdef CONFIG_NF_CONNTRACK_ZONES
 702 out_free:
 703         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 704         return ERR_PTR(-ENOMEM);
 705 #endif
 706 }
 707
 708 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 709                                    const struct nf_conntrack_tuple *orig,
 710                                    const struct nf_conntrack_tuple *repl,
 711                                    gfp_t gfp)
 712 {
 713         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 714 }
 715 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 716
 717 void nf_conntrack_free(struct nf_conn *ct)
 718 {
 719         struct net *net = nf_ct_net(ct);
 720
 721         nf_ct_ext_destroy(ct);
 722         atomic_dec(&net->ct.count);
 723         nf_ct_ext_free(ct);
 724         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 725 }
 726 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 727
 728 /* Allocate a new conntrack: we return -ENOMEM if classification
 729    failed due to stress.  Otherwise it really is unclassifiable. */
 730 static struct nf_conntrack_tuple_hash *
 731 init_conntrack(struct net *net, struct nf_conn *tmpl,
 732                const struct nf_conntrack_tuple *tuple,
 733                struct nf_conntrack_l3proto *l3proto,
 734                struct nf_conntrack_l4proto *l4proto,
 735                struct sk_buff *skb,
 736                unsigned int dataoff, u32 hash)
 737 {
 738         struct nf_conn *ct;
 739         struct nf_conn_help *help;
 740         struct nf_conntrack_tuple repl_tuple;
 741         struct nf_conntrack_ecache *ecache;
 742         struct nf_conntrack_expect *exp;
 743         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 744
 745         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 746                 pr_debug("Can't invert tuple.\n");
 747                 return NULL;
 748         }
 749
 750         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 751                                   hash);
 752         if (IS_ERR(ct))
 753                 return (struct nf_conntrack_tuple_hash *)ct;
 754
 755         if (!l4proto->new(ct, skb, dataoff)) {
 756                 nf_conntrack_free(ct);
 757                 pr_debug("init conntrack: can't track with proto module\n");
 758                 return NULL;
 759         }
 760
 761         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 762         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 763
 764         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 765         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 766                                  ecache ? ecache->expmask : 0,
 767                              GFP_ATOMIC);
 768
 769         spin_lock_bh(&nf_conntrack_lock);
 770         exp = nf_ct_find_expectation(net, zone, tuple);
 771         if (exp) {
 772                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 773                          ct, exp);
 774                 /* Welcome, Mr. Bond.  We've been expecting you... */
 775                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 776                 ct->master = exp->master;
 777                 if (exp->helper) {
 778                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 779                         if (help)
 780                                 rcu_assign_pointer(help->helper, exp->helper);
 781                 }
 782
 783 #ifdef CONFIG_NF_CONNTRACK_MARK
 784                 ct->mark = exp->master->mark;
 785 #endif
 786 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 787                 ct->secmark = exp->master->secmark;
 788 #endif
 789                 nf_conntrack_get(&ct->master->ct_general);
 790                 NF_CT_STAT_INC(net, expect_new);
 791         } else {
 792                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 793                 NF_CT_STAT_INC(net, new);
 794         }
 795
 796         /* Overload tuple linked list to put us in unconfirmed list. */
 797         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 798                        &net->ct.unconfirmed);
 799
 800         spin_unlock_bh(&nf_conntrack_lock);
 801
 802         if (exp) {
 803                 if (exp->expectfn)
 804                         exp->expectfn(ct, exp);
 805                 nf_ct_expect_put(exp);
 806         }
 807
 808         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 809 }
 810
 811 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 812 static inline struct nf_conn *
 813 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 814                   struct sk_buff *skb,
 815                   unsigned int dataoff,
 816                   u_int16_t l3num,
 817                   u_int8_t protonum,
 818                   struct nf_conntrack_l3proto *l3proto,
 819                   struct nf_conntrack_l4proto *l4proto,
 820                   int *set_reply,
 821                   enum ip_conntrack_info *ctinfo)
 822 {
 823         struct nf_conntrack_tuple tuple;
 824         struct nf_conntrack_tuple_hash *h;
 825         struct nf_conn *ct;
 826         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 827         u32 hash;
 828
 829         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 830                              dataoff, l3num, protonum, &tuple, l3proto,
 831                              l4proto)) {
 832                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 833                 return NULL;
 834         }
 835
 836         /* look for tuple match */
 837         hash = hash_conntrack_raw(&tuple, zone);
 838         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 839         if (!h) {
 840                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 841                                    skb, dataoff, hash);
 842                 if (!h)
 843                         return NULL;
 844                 if (IS_ERR(h))
 845                         return (void *)h;
 846         }
 847         ct = nf_ct_tuplehash_to_ctrack(h);
 848
 849         /* It exists; we have (non-exclusive) reference. */
 850         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 851                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 852                 /* Please set reply bit if this packet OK */
 853                 *set_reply = 1;
 854         } else {
 855                 /* Once we've had two way comms, always ESTABLISHED. */
 856                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 857                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 858                         *ctinfo = IP_CT_ESTABLISHED;
 859                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 860                         pr_debug("nf_conntrack_in: related packet for %p\n",
 861                                  ct);
 862                         *ctinfo = IP_CT_RELATED;
 863                 } else {
 864                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 865                         *ctinfo = IP_CT_NEW;
 866                 }
 867                 *set_reply = 0;
 868         }
 869         skb->nfct = &ct->ct_general;
 870         skb->nfctinfo = *ctinfo;
 871         return ct;
 872 }
 873
 874 unsigned int
 875 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 876                 struct sk_buff *skb)
 877 {
 878         struct nf_conn *ct, *tmpl = NULL;
 879         enum ip_conntrack_info ctinfo;
 880         struct nf_conntrack_l3proto *l3proto;
 881         struct nf_conntrack_l4proto *l4proto;
 882         unsigned int dataoff;
 883         u_int8_t protonum;
 884         int set_reply = 0;
 885         int ret;
 886
 887         if (skb->nfct) {
 888                 /* Previously seen (loopback or untracked)?  Ignore. */
 889                 tmpl = (struct nf_conn *)skb->nfct;
 890                 if (!nf_ct_is_template(tmpl)) {
 891                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 892                         return NF_ACCEPT;
 893                 }
 894                 skb->nfct = NULL;
 895         }
 896
 897         /* rcu_read_lock()ed by nf_hook_slow */
 898         l3proto = __nf_ct_l3proto_find(pf);
 899         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 900                                    &dataoff, &protonum);
 901         if (ret <= 0) {
 902                 pr_debug("not prepared to track yet or error occurred\n");
 903                 NF_CT_STAT_INC_ATOMIC(net, error);
 904                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 905                 ret = -ret;
 906                 goto out;
 907         }
 908
 909         l4proto = __nf_ct_l4proto_find(pf, protonum);
 910
 911         /* It may be an special packet, error, unclean...
 912          * inverse of the return code tells to the netfilter
 913          * core what to do with the packet. */
 914         if (l4proto->error != NULL) {
 915                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 916                                      pf, hooknum);
 917                 if (ret <= 0) {
 918                         NF_CT_STAT_INC_ATOMIC(net, error);
 919                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 920                         ret = -ret;
 921                         goto out;
 922                 }
 923                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 924                 if (skb->nfct)
 925                         goto out;
 926         }
 927
 928         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 929                                l3proto, l4proto, &set_reply, &ctinfo);
 930         if (!ct) {
 931                 /* Not valid part of a connection */
 932                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 933                 ret = NF_ACCEPT;
 934                 goto out;
 935         }
 936
 937         if (IS_ERR(ct)) {
 938                 /* Too stressed to deal. */
 939                 NF_CT_STAT_INC_ATOMIC(net, drop);
 940                 ret = NF_DROP;
 941                 goto out;
 942         }
 943
 944         NF_CT_ASSERT(skb->nfct);
 945
 946         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 947         if (ret <= 0) {
 948                 /* Invalid: inverse of the return code tells
 949                  * the netfilter core what to do */
 950                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 951                 nf_conntrack_put(skb->nfct);
 952                 skb->nfct = NULL;
 953                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 954                 if (ret == -NF_DROP)
 955                         NF_CT_STAT_INC_ATOMIC(net, drop);
 956                 ret = -ret;
 957                 goto out;
 958         }
 959
 960         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 961                 nf_conntrack_event_cache(IPCT_REPLY, ct);
 962 out:
 963         if (tmpl) {
 964                 /* Special case: we have to repeat this hook, assign the
 965                  * template again to this packet. We assume that this packet
 966                  * has no conntrack assigned. This is used by nf_ct_tcp. */
 967                 if (ret == NF_REPEAT)
 968                         skb->nfct = (struct nf_conntrack *)tmpl;
 969                 else
 970                         nf_ct_put(tmpl);
 971         }
 972
 973         return ret;
 974 }
 975 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 976
 977 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 978                           const struct nf_conntrack_tuple *orig)
 979 {
 980         bool ret;
 981
 982         rcu_read_lock();
 983         ret = nf_ct_invert_tuple(inverse, orig,
 984                                  __nf_ct_l3proto_find(orig->src.l3num),
 985                                  __nf_ct_l4proto_find(orig->src.l3num,
 986                                                       orig->dst.protonum));
 987         rcu_read_unlock();
 988         return ret;
 989 }
 990 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 991
 992 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 993    implicitly racy: see __nf_conntrack_confirm */
 994 void nf_conntrack_alter_reply(struct nf_conn *ct,
 995                               const struct nf_conntrack_tuple *newreply)
 996 {
 997         struct nf_conn_help *help = nfct_help(ct);
 998
 999         /* Should be unconfirmed, so not in hash table yet */
1000         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1001
1002         pr_debug("Altering reply tuple of %p to ", ct);
1003         nf_ct_dump_tuple(newreply);
1004
1005         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1006         if (ct->master || (help && !hlist_empty(&help->expectations)))
1007                 return;
1008
1009         rcu_read_lock();
1010         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1011         rcu_read_unlock();
1012 }
1013 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1014
1015 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1016 void __nf_ct_refresh_acct(struct nf_conn *ct,
1017                           enum ip_conntrack_info ctinfo,
1018                           const struct sk_buff *skb,
1019                           unsigned long extra_jiffies,
1020                           int do_acct)
1021 {
1022         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1023         NF_CT_ASSERT(skb);
1024
1025         /* Only update if this is not a fixed timeout */
1026         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1027                 goto acct;
1028
1029         /* If not in hash table, timer will not be active yet */
1030         if (!nf_ct_is_confirmed(ct)) {
1031                 ct->timeout.expires = extra_jiffies;
1032         } else {
1033                 unsigned long newtime = jiffies + extra_jiffies;
1034
1035                 /* Only update the timeout if the new timeout is at least
1036                    HZ jiffies from the old timeout. Need del_timer for race
1037                    avoidance (may already be dying). */
1038                 if (newtime - ct->timeout.expires >= HZ)
1039                         mod_timer_pending(&ct->timeout, newtime);
1040         }
1041
1042 acct:
1043         if (do_acct) {
1044                 struct nf_conn_counter *acct;
1045
1046                 acct = nf_conn_acct_find(ct);
1047                 if (acct) {
1048                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1049                         atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes);
1050                 }
1051         }
1052 }
1053 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1054
1055 bool __nf_ct_kill_acct(struct nf_conn *ct,
1056                        enum ip_conntrack_info ctinfo,
1057                        const struct sk_buff *skb,
1058                        int do_acct)
1059 {
1060         if (do_acct) {
1061                 struct nf_conn_counter *acct;
1062
1063                 acct = nf_conn_acct_find(ct);
1064                 if (acct) {
1065                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1066                         atomic64_add(skb->len - skb_network_offset(skb),
1067                                      &acct[CTINFO2DIR(ctinfo)].bytes);
1068                 }
1069         }
1070
1071         if (del_timer(&ct->timeout)) {
1072                 ct->timeout.function((unsigned long)ct);
1073                 return true;
1074         }
1075         return false;
1076 }
1077 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1078
1079 #ifdef CONFIG_NF_CONNTRACK_ZONES
1080 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1081         .len    = sizeof(struct nf_conntrack_zone),
1082         .align  = __alignof__(struct nf_conntrack_zone),
1083         .id     = NF_CT_EXT_ZONE,
1084 };
1085 #endif
1086
1087 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1088
1089 #include <linux/netfilter/nfnetlink.h>
1090 #include <linux/netfilter/nfnetlink_conntrack.h>
1091 #include <linux/mutex.h>
1092
1093 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1094  * in ip_conntrack_core, since we don't want the protocols to autoload
1095  * or depend on ctnetlink */
1096 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1097                                const struct nf_conntrack_tuple *tuple)
1098 {
1099         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
1100         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
1101         return 0;
1102
1103 nla_put_failure:
1104         return -1;
1105 }
1106 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1107
1108 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1109         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1110         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1111 };
1112 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1113
1114 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1115                                struct nf_conntrack_tuple *t)
1116 {
1117         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1118                 return -EINVAL;
1119
1120         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1121         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1122
1123         return 0;
1124 }
1125 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1126
1127 int nf_ct_port_nlattr_tuple_size(void)
1128 {
1129         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1130 }
1131 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1132 #endif
1133
1134 /* Used by ipt_REJECT and ip6t_REJECT. */
1135 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1136 {
1137         struct nf_conn *ct;
1138         enum ip_conntrack_info ctinfo;
1139
1140         /* This ICMP is in reverse direction to the packet which caused it */
1141         ct = nf_ct_get(skb, &ctinfo);
1142         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1143                 ctinfo = IP_CT_RELATED_REPLY;
1144         else
1145                 ctinfo = IP_CT_RELATED;
1146
1147         /* Attach to new skbuff, and increment count */
1148         nskb->nfct = &ct->ct_general;
1149         nskb->nfctinfo = ctinfo;
1150         nf_conntrack_get(nskb->nfct);
1151 }
1152
1153 /* Bring out ya dead! */
1154 static struct nf_conn *
1155 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1156                 void *data, unsigned int *bucket)
1157 {
1158         struct nf_conntrack_tuple_hash *h;
1159         struct nf_conn *ct;
1160         struct hlist_nulls_node *n;
1161
1162         spin_lock_bh(&nf_conntrack_lock);
1163         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1164                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1165                         ct = nf_ct_tuplehash_to_ctrack(h);
1166                         if (iter(ct, data))
1167                                 goto found;
1168                 }
1169         }
1170         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1171                 ct = nf_ct_tuplehash_to_ctrack(h);
1172                 if (iter(ct, data))
1173                         set_bit(IPS_DYING_BIT, &ct->status);
1174         }
1175         spin_unlock_bh(&nf_conntrack_lock);
1176         return NULL;
1177 found:
1178         atomic_inc(&ct->ct_general.use);
1179         spin_unlock_bh(&nf_conntrack_lock);
1180         return ct;
1181 }
1182
1183 void nf_ct_iterate_cleanup(struct net *net,
1184                            int (*iter)(struct nf_conn *i, void *data),
1185                            void *data)
1186 {
1187         struct nf_conn *ct;
1188         unsigned int bucket = 0;
1189
1190         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1191                 /* Time to push up daises... */
1192                 if (del_timer(&ct->timeout))
1193                         death_by_timeout((unsigned long)ct);
1194                 /* ... else the timer will get him soon. */
1195
1196                 nf_ct_put(ct);
1197         }
1198 }
1199 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1200
1201 struct __nf_ct_flush_report {
1202         u32 pid;
1203         int report;
1204 };
1205
1206 static int kill_report(struct nf_conn *i, void *data)
1207 {
1208         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1209         struct nf_conn_tstamp *tstamp;
1210
1211         tstamp = nf_conn_tstamp_find(i);
1212         if (tstamp && tstamp->stop == 0)
1213                 tstamp->stop = ktime_to_ns(ktime_get_real());
1214
1215         /* If we fail to deliver the event, death_by_timeout() will retry */
1216         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1217                                       fr->pid, fr->report) < 0)
1218                 return 1;
1219
1220         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1221         set_bit(IPS_DYING_BIT, &i->status);
1222         return 1;
1223 }
1224
1225 static int kill_all(struct nf_conn *i, void *data)
1226 {
1227         return 1;
1228 }
1229
1230 void nf_ct_free_hashtable(void *hash, unsigned int size)
1231 {
1232         if (is_vmalloc_addr(hash))
1233                 vfree(hash);
1234         else
1235                 free_pages((unsigned long)hash,
1236                            get_order(sizeof(struct hlist_head) * size));
1237 }
1238 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1239
1240 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1241 {
1242         struct __nf_ct_flush_report fr = {
1243                 .pid    = pid,
1244                 .report = report,
1245         };
1246         nf_ct_iterate_cleanup(net, kill_report, &fr);
1247 }
1248 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1249
1250 static void nf_ct_release_dying_list(struct net *net)
1251 {
1252         struct nf_conntrack_tuple_hash *h;
1253         struct nf_conn *ct;
1254         struct hlist_nulls_node *n;
1255
1256         spin_lock_bh(&nf_conntrack_lock);
1257         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1258                 ct = nf_ct_tuplehash_to_ctrack(h);
1259                 /* never fails to remove them, no listeners at this point */
1260                 nf_ct_kill(ct);
1261         }
1262         spin_unlock_bh(&nf_conntrack_lock);
1263 }
1264
1265 static int untrack_refs(void)
1266 {
1267         int cnt = 0, cpu;
1268
1269         for_each_possible_cpu(cpu) {
1270                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1271
1272                 cnt += atomic_read(&ct->ct_general.use) - 1;
1273         }
1274         return cnt;
1275 }
1276
1277 static void nf_conntrack_cleanup_init_net(void)
1278 {
1279         while (untrack_refs() > 0)
1280                 schedule();
1281
1282         nf_conntrack_helper_fini();
1283         nf_conntrack_proto_fini();
1284 #ifdef CONFIG_NF_CONNTRACK_ZONES
1285         nf_ct_extend_unregister(&nf_ct_zone_extend);
1286 #endif
1287 }
1288
1289 static void nf_conntrack_cleanup_net(struct net *net)
1290 {
1291  i_see_dead_people:
1292         nf_ct_iterate_cleanup(net, kill_all, NULL);
1293         nf_ct_release_dying_list(net);
1294         if (atomic_read(&net->ct.count) != 0) {
1295                 schedule();
1296                 goto i_see_dead_people;
1297         }
1298
1299         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1300         nf_conntrack_ecache_fini(net);
1301         nf_conntrack_tstamp_fini(net);
1302         nf_conntrack_acct_fini(net);
1303         nf_conntrack_expect_fini(net);
1304         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1305         kfree(net->ct.slabname);
1306         free_percpu(net->ct.stat);
1307 }
1308
1309 /* Mishearing the voices in his head, our hero wonders how he's
1310    supposed to kill the mall. */
1311 void nf_conntrack_cleanup(struct net *net)
1312 {
1313         if (net_eq(net, &init_net))
1314                 RCU_INIT_POINTER(ip_ct_attach, NULL);
1315
1316         /* This makes sure all current packets have passed through
1317            netfilter framework.  Roll on, two-stage module
1318            delete... */
1319         synchronize_net();
1320
1321         nf_conntrack_cleanup_net(net);
1322
1323         if (net_eq(net, &init_net)) {
1324                 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1325                 nf_conntrack_cleanup_init_net();
1326         }
1327 }
1328
1329 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1330 {
1331         struct hlist_nulls_head *hash;
1332         unsigned int nr_slots, i;
1333         size_t sz;
1334
1335         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1336         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1337         sz = nr_slots * sizeof(struct hlist_nulls_head);
1338         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1339                                         get_order(sz));
1340         if (!hash) {
1341                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1342                 hash = vzalloc(sz);
1343         }
1344
1345         if (hash && nulls)
1346                 for (i = 0; i < nr_slots; i++)
1347                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1348
1349         return hash;
1350 }
1351 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1352
1353 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1354 {
1355         int i, bucket;
1356         unsigned int hashsize, old_size;
1357         struct hlist_nulls_head *hash, *old_hash;
1358         struct nf_conntrack_tuple_hash *h;
1359         struct nf_conn *ct;
1360
1361         if (current->nsproxy->net_ns != &init_net)
1362                 return -EOPNOTSUPP;
1363
1364         /* On boot, we can set this without any fancy locking. */
1365         if (!nf_conntrack_htable_size)
1366                 return param_set_uint(val, kp);
1367
1368         hashsize = simple_strtoul(val, NULL, 0);
1369         if (!hashsize)
1370                 return -EINVAL;
1371
1372         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1373         if (!hash)
1374                 return -ENOMEM;
1375
1376         /* Lookups in the old hash might happen in parallel, which means we
1377          * might get false negatives during connection lookup. New connections
1378          * created because of a false negative won't make it into the hash
1379          * though since that required taking the lock.
1380          */
1381         spin_lock_bh(&nf_conntrack_lock);
1382         for (i = 0; i < init_net.ct.htable_size; i++) {
1383                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1384                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1385                                         struct nf_conntrack_tuple_hash, hnnode);
1386                         ct = nf_ct_tuplehash_to_ctrack(h);
1387                         hlist_nulls_del_rcu(&h->hnnode);
1388                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1389                                                   hashsize);
1390                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1391                 }
1392         }
1393         old_size = init_net.ct.htable_size;
1394         old_hash = init_net.ct.hash;
1395
1396         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1397         init_net.ct.hash = hash;
1398         spin_unlock_bh(&nf_conntrack_lock);
1399
1400         nf_ct_free_hashtable(old_hash, old_size);
1401         return 0;
1402 }
1403 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1404
1405 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1406                   &nf_conntrack_htable_size, 0600);
1407
1408 void nf_ct_untracked_status_or(unsigned long bits)
1409 {
1410         int cpu;
1411
1412         for_each_possible_cpu(cpu)
1413                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1414 }
1415 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1416
1417 static int nf_conntrack_init_init_net(void)
1418 {
1419         int max_factor = 8;
1420         int ret, cpu;
1421
1422         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1423          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1424         if (!nf_conntrack_htable_size) {
1425                 nf_conntrack_htable_size
1426                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1427                            / sizeof(struct hlist_head));
1428                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1429                         nf_conntrack_htable_size = 16384;
1430                 if (nf_conntrack_htable_size < 32)
1431                         nf_conntrack_htable_size = 32;
1432
1433                 /* Use a max. factor of four by default to get the same max as
1434                  * with the old struct list_heads. When a table size is given
1435                  * we use the old value of 8 to avoid reducing the max.
1436                  * entries. */
1437                 max_factor = 4;
1438         }
1439         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1440
1441         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1442                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1443                nf_conntrack_max);
1444
1445         ret = nf_conntrack_proto_init();
1446         if (ret < 0)
1447                 goto err_proto;
1448
1449         ret = nf_conntrack_helper_init();
1450         if (ret < 0)
1451                 goto err_helper;
1452
1453 #ifdef CONFIG_NF_CONNTRACK_ZONES
1454         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1455         if (ret < 0)
1456                 goto err_extend;
1457 #endif
1458         /* Set up fake conntrack: to never be deleted, not in any hashes */
1459         for_each_possible_cpu(cpu) {
1460                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1461                 write_pnet(&ct->ct_net, &init_net);
1462                 atomic_set(&ct->ct_general.use, 1);
1463         }
1464         /*  - and look it like as a confirmed connection */
1465         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1466         return 0;
1467
1468 #ifdef CONFIG_NF_CONNTRACK_ZONES
1469 err_extend:
1470         nf_conntrack_helper_fini();
1471 #endif
1472 err_helper:
1473         nf_conntrack_proto_fini();
1474 err_proto:
1475         return ret;
1476 }
1477
1478 /*
1479  * We need to use special "null" values, not used in hash table
1480  */
1481 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1482 #define DYING_NULLS_VAL         ((1<<30)+1)
1483
1484 static int nf_conntrack_init_net(struct net *net)
1485 {
1486         int ret;
1487
1488         atomic_set(&net->ct.count, 0);
1489         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1490         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1491         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1492         if (!net->ct.stat) {
1493                 ret = -ENOMEM;
1494                 goto err_stat;
1495         }
1496
1497         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1498         if (!net->ct.slabname) {
1499                 ret = -ENOMEM;
1500                 goto err_slabname;
1501         }
1502
1503         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1504                                                         sizeof(struct nf_conn), 0,
1505                                                         SLAB_DESTROY_BY_RCU, NULL);
1506         if (!net->ct.nf_conntrack_cachep) {
1507                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1508                 ret = -ENOMEM;
1509                 goto err_cache;
1510         }
1511
1512         net->ct.htable_size = nf_conntrack_htable_size;
1513         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1514         if (!net->ct.hash) {
1515                 ret = -ENOMEM;
1516                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1517                 goto err_hash;
1518         }
1519         ret = nf_conntrack_expect_init(net);
1520         if (ret < 0)
1521                 goto err_expect;
1522         ret = nf_conntrack_acct_init(net);
1523         if (ret < 0)
1524                 goto err_acct;
1525         ret = nf_conntrack_tstamp_init(net);
1526         if (ret < 0)
1527                 goto err_tstamp;
1528         ret = nf_conntrack_ecache_init(net);
1529         if (ret < 0)
1530                 goto err_ecache;
1531
1532         return 0;
1533
1534 err_ecache:
1535         nf_conntrack_tstamp_fini(net);
1536 err_tstamp:
1537         nf_conntrack_acct_fini(net);
1538 err_acct:
1539         nf_conntrack_expect_fini(net);
1540 err_expect:
1541         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1542 err_hash:
1543         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1544 err_cache:
1545         kfree(net->ct.slabname);
1546 err_slabname:
1547         free_percpu(net->ct.stat);
1548 err_stat:
1549         return ret;
1550 }
1551
1552 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1553                         enum ip_conntrack_dir dir,
1554                         u32 seq);
1555 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);
1556
1557 int nf_conntrack_init(struct net *net)
1558 {
1559         int ret;
1560
1561         if (net_eq(net, &init_net)) {
1562                 ret = nf_conntrack_init_init_net();
1563                 if (ret < 0)
1564                         goto out_init_net;
1565         }
1566         ret = nf_conntrack_init_net(net);
1567         if (ret < 0)
1568                 goto out_net;
1569
1570         if (net_eq(net, &init_net)) {
1571                 /* For use by REJECT target */
1572                 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1573                 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1574
1575                 /* Howto get NAT offsets */
1576                 RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
1577         }
1578         return 0;
1579
1580 out_net:
1581         if (net_eq(net, &init_net))
1582                 nf_conntrack_cleanup_init_net();
1583 out_init_net:
1584         return ret;
1585 }