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 int
 408 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 409 {
 410         struct net *net = nf_ct_net(ct);
 411         unsigned int hash, repl_hash;
 412         struct nf_conntrack_tuple_hash *h;
 413         struct hlist_nulls_node *n;
 414         u16 zone;
 415
 416         zone = nf_ct_zone(ct);
 417         hash = hash_conntrack(net, zone,
 418                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 419         repl_hash = hash_conntrack(net, zone,
 420                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 421
 422         spin_lock_bh(&nf_conntrack_lock);
 423
 424         /* See if there's one in the list already, including reverse */
 425         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 426                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 427                                       &h->tuple) &&
 428                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 429                         goto out;
 430         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 431                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 432                                       &h->tuple) &&
 433                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 434                         goto out;
 435
 436         add_timer(&ct->timeout);
 437         nf_conntrack_get(&ct->ct_general);
 438         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 439         NF_CT_STAT_INC(net, insert);
 440         spin_unlock_bh(&nf_conntrack_lock);
 441
 442         return 0;
 443
 444 out:
 445         NF_CT_STAT_INC(net, insert_failed);
 446         spin_unlock_bh(&nf_conntrack_lock);
 447         return -EEXIST;
 448 }
 449 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 450
 451 /* Confirm a connection given skb; places it in hash table */
 452 int
 453 __nf_conntrack_confirm(struct sk_buff *skb)
 454 {
 455         unsigned int hash, repl_hash;
 456         struct nf_conntrack_tuple_hash *h;
 457         struct nf_conn *ct;
 458         struct nf_conn_help *help;
 459         struct nf_conn_tstamp *tstamp;
 460         struct hlist_nulls_node *n;
 461         enum ip_conntrack_info ctinfo;
 462         struct net *net;
 463         u16 zone;
 464
 465         ct = nf_ct_get(skb, &ctinfo);
 466         net = nf_ct_net(ct);
 467
 468         /* ipt_REJECT uses nf_conntrack_attach to attach related
 469            ICMP/TCP RST packets in other direction.  Actual packet
 470            which created connection will be IP_CT_NEW or for an
 471            expected connection, IP_CT_RELATED. */
 472         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 473                 return NF_ACCEPT;
 474
 475         zone = nf_ct_zone(ct);
 476         /* reuse the hash saved before */
 477         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 478         hash = hash_bucket(hash, net);
 479         repl_hash = hash_conntrack(net, zone,
 480                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 481
 482         /* We're not in hash table, and we refuse to set up related
 483            connections for unconfirmed conns.  But packet copies and
 484            REJECT will give spurious warnings here. */
 485         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 486
 487         /* No external references means no one else could have
 488            confirmed us. */
 489         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 490         pr_debug("Confirming conntrack %p\n", ct);
 491
 492         spin_lock_bh(&nf_conntrack_lock);
 493
 494         /* We have to check the DYING flag inside the lock to prevent
 495            a race against nf_ct_get_next_corpse() possibly called from
 496            user context, else we insert an already 'dead' hash, blocking
 497            further use of that particular connection -JM */
 498
 499         if (unlikely(nf_ct_is_dying(ct))) {
 500                 spin_unlock_bh(&nf_conntrack_lock);
 501                 return NF_ACCEPT;
 502         }
 503
 504         /* See if there's one in the list already, including reverse:
 505            NAT could have grabbed it without realizing, since we're
 506            not in the hash.  If there is, we lost race. */
 507         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 508                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 509                                       &h->tuple) &&
 510                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 511                         goto out;
 512         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 513                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 514                                       &h->tuple) &&
 515                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 516                         goto out;
 517
 518         /* Remove from unconfirmed list */
 519         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 520
 521         /* Timer relative to confirmation time, not original
 522            setting time, otherwise we'd get timer wrap in
 523            weird delay cases. */
 524         ct->timeout.expires += jiffies;
 525         add_timer(&ct->timeout);
 526         atomic_inc(&ct->ct_general.use);
 527         ct->status |= IPS_CONFIRMED;
 528
 529         /* set conntrack timestamp, if enabled. */
 530         tstamp = nf_conn_tstamp_find(ct);
 531         if (tstamp) {
 532                 if (skb->tstamp.tv64 == 0)
 533                         __net_timestamp((struct sk_buff *)skb);
 534
 535                 tstamp->start = ktime_to_ns(skb->tstamp);
 536         }
 537         /* Since the lookup is lockless, hash insertion must be done after
 538          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 539          * guarantee that no other CPU can find the conntrack before the above
 540          * stores are visible.
 541          */
 542         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 543         NF_CT_STAT_INC(net, insert);
 544         spin_unlock_bh(&nf_conntrack_lock);
 545
 546         help = nfct_help(ct);
 547         if (help && help->helper)
 548                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 549
 550         nf_conntrack_event_cache(master_ct(ct) ?
 551                                  IPCT_RELATED : IPCT_NEW, ct);
 552         return NF_ACCEPT;
 553
 554 out:
 555         NF_CT_STAT_INC(net, insert_failed);
 556         spin_unlock_bh(&nf_conntrack_lock);
 557         return NF_DROP;
 558 }
 559 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 560
 561 /* Returns true if a connection correspondings to the tuple (required
 562    for NAT). */
 563 int
 564 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 565                          const struct nf_conn *ignored_conntrack)
 566 {
 567         struct net *net = nf_ct_net(ignored_conntrack);
 568         struct nf_conntrack_tuple_hash *h;
 569         struct hlist_nulls_node *n;
 570         struct nf_conn *ct;
 571         u16 zone = nf_ct_zone(ignored_conntrack);
 572         unsigned int hash = hash_conntrack(net, zone, tuple);
 573
 574         /* Disable BHs the entire time since we need to disable them at
 575          * least once for the stats anyway.
 576          */
 577         rcu_read_lock_bh();
 578         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 579                 ct = nf_ct_tuplehash_to_ctrack(h);
 580                 if (ct != ignored_conntrack &&
 581                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 582                     nf_ct_zone(ct) == zone) {
 583                         NF_CT_STAT_INC(net, found);
 584                         rcu_read_unlock_bh();
 585                         return 1;
 586                 }
 587                 NF_CT_STAT_INC(net, searched);
 588         }
 589         rcu_read_unlock_bh();
 590
 591         return 0;
 592 }
 593 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 594
 595 #define NF_CT_EVICTION_RANGE    8
 596
 597 /* There's a small race here where we may free a just-assured
 598    connection.  Too bad: we're in trouble anyway. */
 599 static noinline int early_drop(struct net *net, unsigned int hash)
 600 {
 601         /* Use oldest entry, which is roughly LRU */
 602         struct nf_conntrack_tuple_hash *h;
 603         struct nf_conn *ct = NULL, *tmp;
 604         struct hlist_nulls_node *n;
 605         unsigned int i, cnt = 0;
 606         int dropped = 0;
 607
 608         rcu_read_lock();
 609         for (i = 0; i < net->ct.htable_size; i++) {
 610                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 611                                          hnnode) {
 612                         tmp = nf_ct_tuplehash_to_ctrack(h);
 613                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 614                                 ct = tmp;
 615                         cnt++;
 616                 }
 617
 618                 if (ct != NULL) {
 619                         if (likely(!nf_ct_is_dying(ct) &&
 620                                    atomic_inc_not_zero(&ct->ct_general.use)))
 621                                 break;
 622                         else
 623                                 ct = NULL;
 624                 }
 625
 626                 if (cnt >= NF_CT_EVICTION_RANGE)
 627                         break;
 628
 629                 hash = (hash + 1) % net->ct.htable_size;
 630         }
 631         rcu_read_unlock();
 632
 633         if (!ct)
 634                 return dropped;
 635
 636         if (del_timer(&ct->timeout)) {
 637                 death_by_timeout((unsigned long)ct);
 638                 dropped = 1;
 639                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 640         }
 641         nf_ct_put(ct);
 642         return dropped;
 643 }
 644
 645 void init_nf_conntrack_hash_rnd(void)
 646 {
 647         unsigned int rand;
 648
 649         /*
 650          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 651          * Because there isn't enough entropy when system initializing,
 652          * and we initialize it as late as possible.
 653          */
 654         do {
 655                 get_random_bytes(&rand, sizeof(rand));
 656         } while (!rand);
 657         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 658 }
 659
 660 static struct nf_conn *
 661 __nf_conntrack_alloc(struct net *net, u16 zone,
 662                      const struct nf_conntrack_tuple *orig,
 663                      const struct nf_conntrack_tuple *repl,
 664                      gfp_t gfp, u32 hash)
 665 {
 666         struct nf_conn *ct;
 667
 668         if (unlikely(!nf_conntrack_hash_rnd)) {
 669                 init_nf_conntrack_hash_rnd();
 670                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 671                 hash = hash_conntrack_raw(orig, zone);
 672         }
 673
 674         /* We don't want any race condition at early drop stage */
 675         atomic_inc(&net->ct.count);
 676
 677         if (nf_conntrack_max &&
 678             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 679                 if (!early_drop(net, hash_bucket(hash, net))) {
 680                         atomic_dec(&net->ct.count);
 681                         if (net_ratelimit())
 682                                 printk(KERN_WARNING
 683                                        "nf_conntrack: table full, dropping"
 684                                        " packet.\n");
 685                         return ERR_PTR(-ENOMEM);
 686                 }
 687         }
 688
 689         /*
 690          * Do not use kmem_cache_zalloc(), as this cache uses
 691          * SLAB_DESTROY_BY_RCU.
 692          */
 693         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 694         if (ct == NULL) {
 695                 atomic_dec(&net->ct.count);
 696                 return ERR_PTR(-ENOMEM);
 697         }
 698         /*
 699          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 700          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 701          */
 702         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 703                offsetof(struct nf_conn, proto) -
 704                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 705         spin_lock_init(&ct->lock);
 706         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 707         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 708         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 709         /* save hash for reusing when confirming */
 710         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 711         /* Don't set timer yet: wait for confirmation */
 712         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 713         write_pnet(&ct->ct_net, net);
 714 #ifdef CONFIG_NF_CONNTRACK_ZONES
 715         if (zone) {
 716                 struct nf_conntrack_zone *nf_ct_zone;
 717
 718                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 719                 if (!nf_ct_zone)
 720                         goto out_free;
 721                 nf_ct_zone->id = zone;
 722         }
 723 #endif
 724         /*
 725          * changes to lookup keys must be done before setting refcnt to 1
 726          */
 727         smp_wmb();
 728         atomic_set(&ct->ct_general.use, 1);
 729         return ct;
 730
 731 #ifdef CONFIG_NF_CONNTRACK_ZONES
 732 out_free:
 733         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 734         return ERR_PTR(-ENOMEM);
 735 #endif
 736 }
 737
 738 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 739                                    const struct nf_conntrack_tuple *orig,
 740                                    const struct nf_conntrack_tuple *repl,
 741                                    gfp_t gfp)
 742 {
 743         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 744 }
 745 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 746
 747 void nf_conntrack_free(struct nf_conn *ct)
 748 {
 749         struct net *net = nf_ct_net(ct);
 750
 751         nf_ct_ext_destroy(ct);
 752         atomic_dec(&net->ct.count);
 753         nf_ct_ext_free(ct);
 754         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 755 }
 756 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 757
 758 /* Allocate a new conntrack: we return -ENOMEM if classification
 759    failed due to stress.  Otherwise it really is unclassifiable. */
 760 static struct nf_conntrack_tuple_hash *
 761 init_conntrack(struct net *net, struct nf_conn *tmpl,
 762                const struct nf_conntrack_tuple *tuple,
 763                struct nf_conntrack_l3proto *l3proto,
 764                struct nf_conntrack_l4proto *l4proto,
 765                struct sk_buff *skb,
 766                unsigned int dataoff, u32 hash)
 767 {
 768         struct nf_conn *ct;
 769         struct nf_conn_help *help;
 770         struct nf_conntrack_tuple repl_tuple;
 771         struct nf_conntrack_ecache *ecache;
 772         struct nf_conntrack_expect *exp;
 773         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 774
 775         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 776                 pr_debug("Can't invert tuple.\n");
 777                 return NULL;
 778         }
 779
 780         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 781                                   hash);
 782         if (IS_ERR(ct))
 783                 return (struct nf_conntrack_tuple_hash *)ct;
 784
 785         if (!l4proto->new(ct, skb, dataoff)) {
 786                 nf_conntrack_free(ct);
 787                 pr_debug("init conntrack: can't track with proto module\n");
 788                 return NULL;
 789         }
 790
 791         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 792         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 793
 794         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 795         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 796                                  ecache ? ecache->expmask : 0,
 797                              GFP_ATOMIC);
 798
 799         spin_lock_bh(&nf_conntrack_lock);
 800         exp = nf_ct_find_expectation(net, zone, tuple);
 801         if (exp) {
 802                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 803                          ct, exp);
 804                 /* Welcome, Mr. Bond.  We've been expecting you... */
 805                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 806                 ct->master = exp->master;
 807                 if (exp->helper) {
 808                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 809                         if (help)
 810                                 rcu_assign_pointer(help->helper, exp->helper);
 811                 }
 812
 813 #ifdef CONFIG_NF_CONNTRACK_MARK
 814                 ct->mark = exp->master->mark;
 815 #endif
 816 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 817                 ct->secmark = exp->master->secmark;
 818 #endif
 819                 nf_conntrack_get(&ct->master->ct_general);
 820                 NF_CT_STAT_INC(net, expect_new);
 821         } else {
 822                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 823                 NF_CT_STAT_INC(net, new);
 824         }
 825
 826         /* Overload tuple linked list to put us in unconfirmed list. */
 827         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 828                        &net->ct.unconfirmed);
 829
 830         spin_unlock_bh(&nf_conntrack_lock);
 831
 832         if (exp) {
 833                 if (exp->expectfn)
 834                         exp->expectfn(ct, exp);
 835                 nf_ct_expect_put(exp);
 836         }
 837
 838         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 839 }
 840
 841 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 842 static inline struct nf_conn *
 843 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 844                   struct sk_buff *skb,
 845                   unsigned int dataoff,
 846                   u_int16_t l3num,
 847                   u_int8_t protonum,
 848                   struct nf_conntrack_l3proto *l3proto,
 849                   struct nf_conntrack_l4proto *l4proto,
 850                   int *set_reply,
 851                   enum ip_conntrack_info *ctinfo)
 852 {
 853         struct nf_conntrack_tuple tuple;
 854         struct nf_conntrack_tuple_hash *h;
 855         struct nf_conn *ct;
 856         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 857         u32 hash;
 858
 859         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 860                              dataoff, l3num, protonum, &tuple, l3proto,
 861                              l4proto)) {
 862                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 863                 return NULL;
 864         }
 865
 866         /* look for tuple match */
 867         hash = hash_conntrack_raw(&tuple, zone);
 868         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 869         if (!h) {
 870                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 871                                    skb, dataoff, hash);
 872                 if (!h)
 873                         return NULL;
 874                 if (IS_ERR(h))
 875                         return (void *)h;
 876         }
 877         ct = nf_ct_tuplehash_to_ctrack(h);
 878
 879         /* It exists; we have (non-exclusive) reference. */
 880         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 881                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 882                 /* Please set reply bit if this packet OK */
 883                 *set_reply = 1;
 884         } else {
 885                 /* Once we've had two way comms, always ESTABLISHED. */
 886                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 887                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 888                         *ctinfo = IP_CT_ESTABLISHED;
 889                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 890                         pr_debug("nf_conntrack_in: related packet for %p\n",
 891                                  ct);
 892                         *ctinfo = IP_CT_RELATED;
 893                 } else {
 894                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 895                         *ctinfo = IP_CT_NEW;
 896                 }
 897                 *set_reply = 0;
 898         }
 899         skb->nfct = &ct->ct_general;
 900         skb->nfctinfo = *ctinfo;
 901         return ct;
 902 }
 903
 904 unsigned int
 905 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 906                 struct sk_buff *skb)
 907 {
 908         struct nf_conn *ct, *tmpl = NULL;
 909         enum ip_conntrack_info ctinfo;
 910         struct nf_conntrack_l3proto *l3proto;
 911         struct nf_conntrack_l4proto *l4proto;
 912         unsigned int dataoff;
 913         u_int8_t protonum;
 914         int set_reply = 0;
 915         int ret;
 916
 917         if (skb->nfct) {
 918                 /* Previously seen (loopback or untracked)?  Ignore. */
 919                 tmpl = (struct nf_conn *)skb->nfct;
 920                 if (!nf_ct_is_template(tmpl)) {
 921                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 922                         return NF_ACCEPT;
 923                 }
 924                 skb->nfct = NULL;
 925         }
 926
 927         /* rcu_read_lock()ed by nf_hook_slow */
 928         l3proto = __nf_ct_l3proto_find(pf);
 929         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 930                                    &dataoff, &protonum);
 931         if (ret <= 0) {
 932                 pr_debug("not prepared to track yet or error occurred\n");
 933                 NF_CT_STAT_INC_ATOMIC(net, error);
 934                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 935                 ret = -ret;
 936                 goto out;
 937         }
 938
 939         l4proto = __nf_ct_l4proto_find(pf, protonum);
 940
 941         /* It may be an special packet, error, unclean...
 942          * inverse of the return code tells to the netfilter
 943          * core what to do with the packet. */
 944         if (l4proto->error != NULL) {
 945                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 946                                      pf, hooknum);
 947                 if (ret <= 0) {
 948                         NF_CT_STAT_INC_ATOMIC(net, error);
 949                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 950                         ret = -ret;
 951                         goto out;
 952                 }
 953                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 954                 if (skb->nfct)
 955                         goto out;
 956         }
 957
 958         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 959                                l3proto, l4proto, &set_reply, &ctinfo);
 960         if (!ct) {
 961                 /* Not valid part of a connection */
 962                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 963                 ret = NF_ACCEPT;
 964                 goto out;
 965         }
 966
 967         if (IS_ERR(ct)) {
 968                 /* Too stressed to deal. */
 969                 NF_CT_STAT_INC_ATOMIC(net, drop);
 970                 ret = NF_DROP;
 971                 goto out;
 972         }
 973
 974         NF_CT_ASSERT(skb->nfct);
 975
 976         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 977         if (ret <= 0) {
 978                 /* Invalid: inverse of the return code tells
 979                  * the netfilter core what to do */
 980                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 981                 nf_conntrack_put(skb->nfct);
 982                 skb->nfct = NULL;
 983                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 984                 if (ret == -NF_DROP)
 985                         NF_CT_STAT_INC_ATOMIC(net, drop);
 986                 ret = -ret;
 987                 goto out;
 988         }
 989
 990         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 991                 nf_conntrack_event_cache(IPCT_REPLY, ct);
 992 out:
 993         if (tmpl) {
 994                 /* Special case: we have to repeat this hook, assign the
 995                  * template again to this packet. We assume that this packet
 996                  * has no conntrack assigned. This is used by nf_ct_tcp. */
 997                 if (ret == NF_REPEAT)
 998                         skb->nfct = (struct nf_conntrack *)tmpl;
 999                 else
1000                         nf_ct_put(tmpl);
1001         }
1002
1003         return ret;
1004 }
1005 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1006
1007 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1008                           const struct nf_conntrack_tuple *orig)
1009 {
1010         bool ret;
1011
1012         rcu_read_lock();
1013         ret = nf_ct_invert_tuple(inverse, orig,
1014                                  __nf_ct_l3proto_find(orig->src.l3num),
1015                                  __nf_ct_l4proto_find(orig->src.l3num,
1016                                                       orig->dst.protonum));
1017         rcu_read_unlock();
1018         return ret;
1019 }
1020 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1021
1022 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1023    implicitly racy: see __nf_conntrack_confirm */
1024 void nf_conntrack_alter_reply(struct nf_conn *ct,
1025                               const struct nf_conntrack_tuple *newreply)
1026 {
1027         struct nf_conn_help *help = nfct_help(ct);
1028
1029         /* Should be unconfirmed, so not in hash table yet */
1030         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1031
1032         pr_debug("Altering reply tuple of %p to ", ct);
1033         nf_ct_dump_tuple(newreply);
1034
1035         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1036         if (ct->master || (help && !hlist_empty(&help->expectations)))
1037                 return;
1038
1039         rcu_read_lock();
1040         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1041         rcu_read_unlock();
1042 }
1043 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1044
1045 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1046 void __nf_ct_refresh_acct(struct nf_conn *ct,
1047                           enum ip_conntrack_info ctinfo,
1048                           const struct sk_buff *skb,
1049                           unsigned long extra_jiffies,
1050                           int do_acct)
1051 {
1052         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1053         NF_CT_ASSERT(skb);
1054
1055         /* Only update if this is not a fixed timeout */
1056         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1057                 goto acct;
1058
1059         /* If not in hash table, timer will not be active yet */
1060         if (!nf_ct_is_confirmed(ct)) {
1061                 ct->timeout.expires = extra_jiffies;
1062         } else {
1063                 unsigned long newtime = jiffies + extra_jiffies;
1064
1065                 /* Only update the timeout if the new timeout is at least
1066                    HZ jiffies from the old timeout. Need del_timer for race
1067                    avoidance (may already be dying). */
1068                 if (newtime - ct->timeout.expires >= HZ)
1069                         mod_timer_pending(&ct->timeout, newtime);
1070         }
1071
1072 acct:
1073         if (do_acct) {
1074                 struct nf_conn_counter *acct;
1075
1076                 acct = nf_conn_acct_find(ct);
1077                 if (acct) {
1078                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1079                         atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes);
1080                 }
1081         }
1082 }
1083 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1084
1085 bool __nf_ct_kill_acct(struct nf_conn *ct,
1086                        enum ip_conntrack_info ctinfo,
1087                        const struct sk_buff *skb,
1088                        int do_acct)
1089 {
1090         if (do_acct) {
1091                 struct nf_conn_counter *acct;
1092
1093                 acct = nf_conn_acct_find(ct);
1094                 if (acct) {
1095                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1096                         atomic64_add(skb->len - skb_network_offset(skb),
1097                                      &acct[CTINFO2DIR(ctinfo)].bytes);
1098                 }
1099         }
1100
1101         if (del_timer(&ct->timeout)) {
1102                 ct->timeout.function((unsigned long)ct);
1103                 return true;
1104         }
1105         return false;
1106 }
1107 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1108
1109 #ifdef CONFIG_NF_CONNTRACK_ZONES
1110 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1111         .len    = sizeof(struct nf_conntrack_zone),
1112         .align  = __alignof__(struct nf_conntrack_zone),
1113         .id     = NF_CT_EXT_ZONE,
1114 };
1115 #endif
1116
1117 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1118
1119 #include <linux/netfilter/nfnetlink.h>
1120 #include <linux/netfilter/nfnetlink_conntrack.h>
1121 #include <linux/mutex.h>
1122
1123 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1124  * in ip_conntrack_core, since we don't want the protocols to autoload
1125  * or depend on ctnetlink */
1126 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1127                                const struct nf_conntrack_tuple *tuple)
1128 {
1129         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
1130         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
1131         return 0;
1132
1133 nla_put_failure:
1134         return -1;
1135 }
1136 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1137
1138 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1139         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1140         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1141 };
1142 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1143
1144 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1145                                struct nf_conntrack_tuple *t)
1146 {
1147         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1148                 return -EINVAL;
1149
1150         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1151         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1152
1153         return 0;
1154 }
1155 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1156
1157 int nf_ct_port_nlattr_tuple_size(void)
1158 {
1159         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1160 }
1161 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1162 #endif
1163
1164 /* Used by ipt_REJECT and ip6t_REJECT. */
1165 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
1166 {
1167         struct nf_conn *ct;
1168         enum ip_conntrack_info ctinfo;
1169
1170         /* This ICMP is in reverse direction to the packet which caused it */
1171         ct = nf_ct_get(skb, &ctinfo);
1172         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1173                 ctinfo = IP_CT_RELATED_REPLY;
1174         else
1175                 ctinfo = IP_CT_RELATED;
1176
1177         /* Attach to new skbuff, and increment count */
1178         nskb->nfct = &ct->ct_general;
1179         nskb->nfctinfo = ctinfo;
1180         nf_conntrack_get(nskb->nfct);
1181 }
1182
1183 /* Bring out ya dead! */
1184 static struct nf_conn *
1185 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1186                 void *data, unsigned int *bucket)
1187 {
1188         struct nf_conntrack_tuple_hash *h;
1189         struct nf_conn *ct;
1190         struct hlist_nulls_node *n;
1191
1192         spin_lock_bh(&nf_conntrack_lock);
1193         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1194                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1195                         ct = nf_ct_tuplehash_to_ctrack(h);
1196                         if (iter(ct, data))
1197                                 goto found;
1198                 }
1199         }
1200         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1201                 ct = nf_ct_tuplehash_to_ctrack(h);
1202                 if (iter(ct, data))
1203                         set_bit(IPS_DYING_BIT, &ct->status);
1204         }
1205         spin_unlock_bh(&nf_conntrack_lock);
1206         return NULL;
1207 found:
1208         atomic_inc(&ct->ct_general.use);
1209         spin_unlock_bh(&nf_conntrack_lock);
1210         return ct;
1211 }
1212
1213 void nf_ct_iterate_cleanup(struct net *net,
1214                            int (*iter)(struct nf_conn *i, void *data),
1215                            void *data)
1216 {
1217         struct nf_conn *ct;
1218         unsigned int bucket = 0;
1219
1220         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1221                 /* Time to push up daises... */
1222                 if (del_timer(&ct->timeout))
1223                         death_by_timeout((unsigned long)ct);
1224                 /* ... else the timer will get him soon. */
1225
1226                 nf_ct_put(ct);
1227         }
1228 }
1229 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1230
1231 struct __nf_ct_flush_report {
1232         u32 pid;
1233         int report;
1234 };
1235
1236 static int kill_report(struct nf_conn *i, void *data)
1237 {
1238         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1239         struct nf_conn_tstamp *tstamp;
1240
1241         tstamp = nf_conn_tstamp_find(i);
1242         if (tstamp && tstamp->stop == 0)
1243                 tstamp->stop = ktime_to_ns(ktime_get_real());
1244
1245         /* If we fail to deliver the event, death_by_timeout() will retry */
1246         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1247                                       fr->pid, fr->report) < 0)
1248                 return 1;
1249
1250         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1251         set_bit(IPS_DYING_BIT, &i->status);
1252         return 1;
1253 }
1254
1255 static int kill_all(struct nf_conn *i, void *data)
1256 {
1257         return 1;
1258 }
1259
1260 void nf_ct_free_hashtable(void *hash, unsigned int size)
1261 {
1262         if (is_vmalloc_addr(hash))
1263                 vfree(hash);
1264         else
1265                 free_pages((unsigned long)hash,
1266                            get_order(sizeof(struct hlist_head) * size));
1267 }
1268 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1269
1270 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1271 {
1272         struct __nf_ct_flush_report fr = {
1273                 .pid    = pid,
1274                 .report = report,
1275         };
1276         nf_ct_iterate_cleanup(net, kill_report, &fr);
1277 }
1278 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1279
1280 static void nf_ct_release_dying_list(struct net *net)
1281 {
1282         struct nf_conntrack_tuple_hash *h;
1283         struct nf_conn *ct;
1284         struct hlist_nulls_node *n;
1285
1286         spin_lock_bh(&nf_conntrack_lock);
1287         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1288                 ct = nf_ct_tuplehash_to_ctrack(h);
1289                 /* never fails to remove them, no listeners at this point */
1290                 nf_ct_kill(ct);
1291         }
1292         spin_unlock_bh(&nf_conntrack_lock);
1293 }
1294
1295 static int untrack_refs(void)
1296 {
1297         int cnt = 0, cpu;
1298
1299         for_each_possible_cpu(cpu) {
1300                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1301
1302                 cnt += atomic_read(&ct->ct_general.use) - 1;
1303         }
1304         return cnt;
1305 }
1306
1307 static void nf_conntrack_cleanup_init_net(void)
1308 {
1309         while (untrack_refs() > 0)
1310                 schedule();
1311
1312         nf_conntrack_helper_fini();
1313         nf_conntrack_proto_fini();
1314 #ifdef CONFIG_NF_CONNTRACK_ZONES
1315         nf_ct_extend_unregister(&nf_ct_zone_extend);
1316 #endif
1317 }
1318
1319 static void nf_conntrack_cleanup_net(struct net *net)
1320 {
1321  i_see_dead_people:
1322         nf_ct_iterate_cleanup(net, kill_all, NULL);
1323         nf_ct_release_dying_list(net);
1324         if (atomic_read(&net->ct.count) != 0) {
1325                 schedule();
1326                 goto i_see_dead_people;
1327         }
1328
1329         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1330         nf_conntrack_ecache_fini(net);
1331         nf_conntrack_tstamp_fini(net);
1332         nf_conntrack_acct_fini(net);
1333         nf_conntrack_expect_fini(net);
1334         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1335         kfree(net->ct.slabname);
1336         free_percpu(net->ct.stat);
1337 }
1338
1339 /* Mishearing the voices in his head, our hero wonders how he's
1340    supposed to kill the mall. */
1341 void nf_conntrack_cleanup(struct net *net)
1342 {
1343         if (net_eq(net, &init_net))
1344                 RCU_INIT_POINTER(ip_ct_attach, NULL);
1345
1346         /* This makes sure all current packets have passed through
1347            netfilter framework.  Roll on, two-stage module
1348            delete... */
1349         synchronize_net();
1350
1351         nf_conntrack_cleanup_net(net);
1352
1353         if (net_eq(net, &init_net)) {
1354                 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1355                 nf_conntrack_cleanup_init_net();
1356         }
1357 }
1358
1359 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1360 {
1361         struct hlist_nulls_head *hash;
1362         unsigned int nr_slots, i;
1363         size_t sz;
1364
1365         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1366         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1367         sz = nr_slots * sizeof(struct hlist_nulls_head);
1368         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1369                                         get_order(sz));
1370         if (!hash) {
1371                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1372                 hash = vzalloc(sz);
1373         }
1374
1375         if (hash && nulls)
1376                 for (i = 0; i < nr_slots; i++)
1377                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1378
1379         return hash;
1380 }
1381 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1382
1383 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1384 {
1385         int i, bucket;
1386         unsigned int hashsize, old_size;
1387         struct hlist_nulls_head *hash, *old_hash;
1388         struct nf_conntrack_tuple_hash *h;
1389         struct nf_conn *ct;
1390
1391         if (current->nsproxy->net_ns != &init_net)
1392                 return -EOPNOTSUPP;
1393
1394         /* On boot, we can set this without any fancy locking. */
1395         if (!nf_conntrack_htable_size)
1396                 return param_set_uint(val, kp);
1397
1398         hashsize = simple_strtoul(val, NULL, 0);
1399         if (!hashsize)
1400                 return -EINVAL;
1401
1402         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1403         if (!hash)
1404                 return -ENOMEM;
1405
1406         /* Lookups in the old hash might happen in parallel, which means we
1407          * might get false negatives during connection lookup. New connections
1408          * created because of a false negative won't make it into the hash
1409          * though since that required taking the lock.
1410          */
1411         spin_lock_bh(&nf_conntrack_lock);
1412         for (i = 0; i < init_net.ct.htable_size; i++) {
1413                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1414                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1415                                         struct nf_conntrack_tuple_hash, hnnode);
1416                         ct = nf_ct_tuplehash_to_ctrack(h);
1417                         hlist_nulls_del_rcu(&h->hnnode);
1418                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1419                                                   hashsize);
1420                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1421                 }
1422         }
1423         old_size = init_net.ct.htable_size;
1424         old_hash = init_net.ct.hash;
1425
1426         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1427         init_net.ct.hash = hash;
1428         spin_unlock_bh(&nf_conntrack_lock);
1429
1430         nf_ct_free_hashtable(old_hash, old_size);
1431         return 0;
1432 }
1433 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1434
1435 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1436                   &nf_conntrack_htable_size, 0600);
1437
1438 void nf_ct_untracked_status_or(unsigned long bits)
1439 {
1440         int cpu;
1441
1442         for_each_possible_cpu(cpu)
1443                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1444 }
1445 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1446
1447 static int nf_conntrack_init_init_net(void)
1448 {
1449         int max_factor = 8;
1450         int ret, cpu;
1451
1452         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1453          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1454         if (!nf_conntrack_htable_size) {
1455                 nf_conntrack_htable_size
1456                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1457                            / sizeof(struct hlist_head));
1458                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1459                         nf_conntrack_htable_size = 16384;
1460                 if (nf_conntrack_htable_size < 32)
1461                         nf_conntrack_htable_size = 32;
1462
1463                 /* Use a max. factor of four by default to get the same max as
1464                  * with the old struct list_heads. When a table size is given
1465                  * we use the old value of 8 to avoid reducing the max.
1466                  * entries. */
1467                 max_factor = 4;
1468         }
1469         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1470
1471         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1472                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1473                nf_conntrack_max);
1474
1475         ret = nf_conntrack_proto_init();
1476         if (ret < 0)
1477                 goto err_proto;
1478
1479         ret = nf_conntrack_helper_init();
1480         if (ret < 0)
1481                 goto err_helper;
1482
1483 #ifdef CONFIG_NF_CONNTRACK_ZONES
1484         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1485         if (ret < 0)
1486                 goto err_extend;
1487 #endif
1488         /* Set up fake conntrack: to never be deleted, not in any hashes */
1489         for_each_possible_cpu(cpu) {
1490                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1491                 write_pnet(&ct->ct_net, &init_net);
1492                 atomic_set(&ct->ct_general.use, 1);
1493         }
1494         /*  - and look it like as a confirmed connection */
1495         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1496         return 0;
1497
1498 #ifdef CONFIG_NF_CONNTRACK_ZONES
1499 err_extend:
1500         nf_conntrack_helper_fini();
1501 #endif
1502 err_helper:
1503         nf_conntrack_proto_fini();
1504 err_proto:
1505         return ret;
1506 }
1507
1508 /*
1509  * We need to use special "null" values, not used in hash table
1510  */
1511 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1512 #define DYING_NULLS_VAL         ((1<<30)+1)
1513
1514 static int nf_conntrack_init_net(struct net *net)
1515 {
1516         int ret;
1517
1518         atomic_set(&net->ct.count, 0);
1519         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1520         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1521         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1522         if (!net->ct.stat) {
1523                 ret = -ENOMEM;
1524                 goto err_stat;
1525         }
1526
1527         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1528         if (!net->ct.slabname) {
1529                 ret = -ENOMEM;
1530                 goto err_slabname;
1531         }
1532
1533         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1534                                                         sizeof(struct nf_conn), 0,
1535                                                         SLAB_DESTROY_BY_RCU, NULL);
1536         if (!net->ct.nf_conntrack_cachep) {
1537                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1538                 ret = -ENOMEM;
1539                 goto err_cache;
1540         }
1541
1542         net->ct.htable_size = nf_conntrack_htable_size;
1543         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1544         if (!net->ct.hash) {
1545                 ret = -ENOMEM;
1546                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1547                 goto err_hash;
1548         }
1549         ret = nf_conntrack_expect_init(net);
1550         if (ret < 0)
1551                 goto err_expect;
1552         ret = nf_conntrack_acct_init(net);
1553         if (ret < 0)
1554                 goto err_acct;
1555         ret = nf_conntrack_tstamp_init(net);
1556         if (ret < 0)
1557                 goto err_tstamp;
1558         ret = nf_conntrack_ecache_init(net);
1559         if (ret < 0)
1560                 goto err_ecache;
1561
1562         return 0;
1563
1564 err_ecache:
1565         nf_conntrack_tstamp_fini(net);
1566 err_tstamp:
1567         nf_conntrack_acct_fini(net);
1568 err_acct:
1569         nf_conntrack_expect_fini(net);
1570 err_expect:
1571         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1572 err_hash:
1573         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1574 err_cache:
1575         kfree(net->ct.slabname);
1576 err_slabname:
1577         free_percpu(net->ct.stat);
1578 err_stat:
1579         return ret;
1580 }
1581
1582 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1583                         enum ip_conntrack_dir dir,
1584                         u32 seq);
1585 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);
1586
1587 int nf_conntrack_init(struct net *net)
1588 {
1589         int ret;
1590
1591         if (net_eq(net, &init_net)) {
1592                 ret = nf_conntrack_init_init_net();
1593                 if (ret < 0)
1594                         goto out_init_net;
1595         }
1596         ret = nf_conntrack_init_net(net);
1597         if (ret < 0)
1598                 goto out_net;
1599
1600         if (net_eq(net, &init_net)) {
1601                 /* For use by REJECT target */
1602                 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1603                 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1604
1605                 /* Howto get NAT offsets */
1606                 RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
1607         }
1608         return 0;
1609
1610 out_net:
1611         if (net_eq(net, &init_net))
1612                 nf_conntrack_cleanup_init_net();
1613 out_init_net:
1614         return ret;
1615 }