net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  */
  14
  15 #include <linux/types.h>
  16 #include <linux/netfilter.h>
  17 #include <linux/module.h>
  18 #include <linux/sched.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/proc_fs.h>
  21 #include <linux/vmalloc.h>
  22 #include <linux/stddef.h>
  23 #include <linux/slab.h>
  24 #include <linux/random.h>
  25 #include <linux/jhash.h>
  26 #include <linux/err.h>
  27 #include <linux/percpu.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/notifier.h>
  30 #include <linux/kernel.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/socket.h>
  33 #include <linux/mm.h>
  34 #include <linux/nsproxy.h>
  35 #include <linux/rculist_nulls.h>
  36
  37 #include <net/netfilter/nf_conntrack.h>
  38 #include <net/netfilter/nf_conntrack_l3proto.h>
  39 #include <net/netfilter/nf_conntrack_l4proto.h>
  40 #include <net/netfilter/nf_conntrack_expect.h>
  41 #include <net/netfilter/nf_conntrack_helper.h>
  42 #include <net/netfilter/nf_conntrack_seqadj.h>
  43 #include <net/netfilter/nf_conntrack_core.h>
  44 #include <net/netfilter/nf_conntrack_extend.h>
  45 #include <net/netfilter/nf_conntrack_acct.h>
  46 #include <net/netfilter/nf_conntrack_ecache.h>
  47 #include <net/netfilter/nf_conntrack_zones.h>
  48 #include <net/netfilter/nf_conntrack_timestamp.h>
  49 #include <net/netfilter/nf_conntrack_timeout.h>
  50 #include <net/netfilter/nf_conntrack_labels.h>
  51 #include <net/netfilter/nf_nat.h>
  52 #include <net/netfilter/nf_nat_core.h>
  53 #include <net/netfilter/nf_nat_helper.h>
  54
  55 #define NF_CONNTRACK_VERSION    "0.5.0"
  56
  57 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  58                                       enum nf_nat_manip_type manip,
  59                                       const struct nlattr *attr) __read_mostly;
  60 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  61
  62 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
  63                               struct nf_conn *ct,
  64                               enum ip_conntrack_info ctinfo,
  65                               unsigned int protoff);
  66 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
  67
  68 DEFINE_SPINLOCK(nf_conntrack_lock);
  69 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  70
  71 unsigned int nf_conntrack_htable_size __read_mostly;
  72 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  73
  74 unsigned int nf_conntrack_max __read_mostly;
  75 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  76
  77 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  78 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  79
  80 unsigned int nf_conntrack_hash_rnd __read_mostly;
  81 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
  82
  83 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  84 {
  85         unsigned int n;
  86
  87         /* The direction must be ignored, so we hash everything up to the
  88          * destination ports (which is a multiple of 4) and treat the last
  89          * three bytes manually.
  90          */
  91         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  92         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  93                       (((__force __u16)tuple->dst.u.all << 16) |
  94                       tuple->dst.protonum));
  95 }
  96
  97 static u32 __hash_bucket(u32 hash, unsigned int size)
  98 {
  99         return ((u64)hash * size) >> 32;
 100 }
 101
 102 static u32 hash_bucket(u32 hash, const struct net *net)
 103 {
 104         return __hash_bucket(hash, net->ct.htable_size);
 105 }
 106
 107 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
 108                                   u16 zone, unsigned int size)
 109 {
 110         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 111 }
 112
 113 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 114                                        const struct nf_conntrack_tuple *tuple)
 115 {
 116         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 117 }
 118
 119 bool
 120 nf_ct_get_tuple(const struct sk_buff *skb,
 121                 unsigned int nhoff,
 122                 unsigned int dataoff,
 123                 u_int16_t l3num,
 124                 u_int8_t protonum,
 125                 struct nf_conntrack_tuple *tuple,
 126                 const struct nf_conntrack_l3proto *l3proto,
 127                 const struct nf_conntrack_l4proto *l4proto)
 128 {
 129         memset(tuple, 0, sizeof(*tuple));
 130
 131         tuple->src.l3num = l3num;
 132         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 133                 return false;
 134
 135         tuple->dst.protonum = protonum;
 136         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 137
 138         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 139 }
 140 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 141
 142 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 143                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 144 {
 145         struct nf_conntrack_l3proto *l3proto;
 146         struct nf_conntrack_l4proto *l4proto;
 147         unsigned int protoff;
 148         u_int8_t protonum;
 149         int ret;
 150
 151         rcu_read_lock();
 152
 153         l3proto = __nf_ct_l3proto_find(l3num);
 154         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 155         if (ret != NF_ACCEPT) {
 156                 rcu_read_unlock();
 157                 return false;
 158         }
 159
 160         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 161
 162         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 163                               l3proto, l4proto);
 164
 165         rcu_read_unlock();
 166         return ret;
 167 }
 168 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 169
 170 bool
 171 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 172                    const struct nf_conntrack_tuple *orig,
 173                    const struct nf_conntrack_l3proto *l3proto,
 174                    const struct nf_conntrack_l4proto *l4proto)
 175 {
 176         memset(inverse, 0, sizeof(*inverse));
 177
 178         inverse->src.l3num = orig->src.l3num;
 179         if (l3proto->invert_tuple(inverse, orig) == 0)
 180                 return false;
 181
 182         inverse->dst.dir = !orig->dst.dir;
 183
 184         inverse->dst.protonum = orig->dst.protonum;
 185         return l4proto->invert_tuple(inverse, orig);
 186 }
 187 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 188
 189 static void
 190 clean_from_lists(struct nf_conn *ct)
 191 {
 192         pr_debug("clean_from_lists(%p)\n", ct);
 193         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 194         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 195
 196         /* Destroy all pending expectations */
 197         nf_ct_remove_expectations(ct);
 198 }
 199
 200 static void
 201 destroy_conntrack(struct nf_conntrack *nfct)
 202 {
 203         struct nf_conn *ct = (struct nf_conn *)nfct;
 204         struct net *net = nf_ct_net(ct);
 205         struct nf_conntrack_l4proto *l4proto;
 206
 207         pr_debug("destroy_conntrack(%p)\n", ct);
 208         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 209         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 210
 211         /* To make sure we don't get any weird locking issues here:
 212          * destroy_conntrack() MUST NOT be called with a write lock
 213          * to nf_conntrack_lock!!! -HW */
 214         rcu_read_lock();
 215         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 216         if (l4proto && l4proto->destroy)
 217                 l4proto->destroy(ct);
 218
 219         rcu_read_unlock();
 220
 221         spin_lock_bh(&nf_conntrack_lock);
 222         /* Expectations will have been removed in clean_from_lists,
 223          * except TFTP can create an expectation on the first packet,
 224          * before connection is in the list, so we need to clean here,
 225          * too. */
 226         nf_ct_remove_expectations(ct);
 227
 228         /* We overload first tuple to link into unconfirmed or dying list.*/
 229         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 230         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 231
 232         NF_CT_STAT_INC(net, delete);
 233         spin_unlock_bh(&nf_conntrack_lock);
 234
 235         if (ct->master)
 236                 nf_ct_put(ct->master);
 237
 238         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 239         nf_conntrack_free(ct);
 240 }
 241
 242 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 243 {
 244         struct net *net = nf_ct_net(ct);
 245
 246         nf_ct_helper_destroy(ct);
 247         spin_lock_bh(&nf_conntrack_lock);
 248         /* Inside lock so preempt is disabled on module removal path.
 249          * Otherwise we can get spurious warnings. */
 250         NF_CT_STAT_INC(net, delete_list);
 251         clean_from_lists(ct);
 252         /* add this conntrack to the dying list */
 253         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 254                              &net->ct.dying);
 255         spin_unlock_bh(&nf_conntrack_lock);
 256 }
 257
 258 static void death_by_event(unsigned long ul_conntrack)
 259 {
 260         struct nf_conn *ct = (void *)ul_conntrack;
 261         struct net *net = nf_ct_net(ct);
 262         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 263
 264         BUG_ON(ecache == NULL);
 265
 266         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 267                 /* bad luck, let's retry again */
 268                 ecache->timeout.expires = jiffies +
 269                         (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 270                 add_timer(&ecache->timeout);
 271                 return;
 272         }
 273         /* we've got the event delivered, now it's dying */
 274         set_bit(IPS_DYING_BIT, &ct->status);
 275         nf_ct_put(ct);
 276 }
 277
 278 static void nf_ct_dying_timeout(struct nf_conn *ct)
 279 {
 280         struct net *net = nf_ct_net(ct);
 281         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 282
 283         BUG_ON(ecache == NULL);
 284
 285         /* set a new timer to retry event delivery */
 286         setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
 287         ecache->timeout.expires = jiffies +
 288                 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 289         add_timer(&ecache->timeout);
 290 }
 291
 292 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 293 {
 294         struct nf_conn_tstamp *tstamp;
 295
 296         tstamp = nf_conn_tstamp_find(ct);
 297         if (tstamp && tstamp->stop == 0)
 298                 tstamp->stop = ktime_to_ns(ktime_get_real());
 299
 300         if (!nf_ct_is_dying(ct) &&
 301             unlikely(nf_conntrack_event_report(IPCT_DESTROY, ct,
 302             portid, report) < 0)) {
 303                 /* destroy event was not delivered */
 304                 nf_ct_delete_from_lists(ct);
 305                 nf_ct_dying_timeout(ct);
 306                 return false;
 307         }
 308         set_bit(IPS_DYING_BIT, &ct->status);
 309         nf_ct_delete_from_lists(ct);
 310         nf_ct_put(ct);
 311         return true;
 312 }
 313 EXPORT_SYMBOL_GPL(nf_ct_delete);
 314
 315 static void death_by_timeout(unsigned long ul_conntrack)
 316 {
 317         nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
 318 }
 319
 320 /*
 321  * Warning :
 322  * - Caller must take a reference on returned object
 323  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 324  * OR
 325  * - Caller must lock nf_conntrack_lock before calling this function
 326  */
 327 static struct nf_conntrack_tuple_hash *
 328 ____nf_conntrack_find(struct net *net, u16 zone,
 329                       const struct nf_conntrack_tuple *tuple, u32 hash)
 330 {
 331         struct nf_conntrack_tuple_hash *h;
 332         struct hlist_nulls_node *n;
 333         unsigned int bucket = hash_bucket(hash, net);
 334
 335         /* Disable BHs the entire time since we normally need to disable them
 336          * at least once for the stats anyway.
 337          */
 338         local_bh_disable();
 339 begin:
 340         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 341                 if (nf_ct_tuple_equal(tuple, &h->tuple) &&
 342                     nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
 343                         NF_CT_STAT_INC(net, found);
 344                         local_bh_enable();
 345                         return h;
 346                 }
 347                 NF_CT_STAT_INC(net, searched);
 348         }
 349         /*
 350          * if the nulls value we got at the end of this lookup is
 351          * not the expected one, we must restart lookup.
 352          * We probably met an item that was moved to another chain.
 353          */
 354         if (get_nulls_value(n) != bucket) {
 355                 NF_CT_STAT_INC(net, search_restart);
 356                 goto begin;
 357         }
 358         local_bh_enable();
 359
 360         return NULL;
 361 }
 362
 363 struct nf_conntrack_tuple_hash *
 364 __nf_conntrack_find(struct net *net, u16 zone,
 365                     const struct nf_conntrack_tuple *tuple)
 366 {
 367         return ____nf_conntrack_find(net, zone, tuple,
 368                                      hash_conntrack_raw(tuple, zone));
 369 }
 370 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 371
 372 /* Find a connection corresponding to a tuple. */
 373 static struct nf_conntrack_tuple_hash *
 374 __nf_conntrack_find_get(struct net *net, u16 zone,
 375                         const struct nf_conntrack_tuple *tuple, u32 hash)
 376 {
 377         struct nf_conntrack_tuple_hash *h;
 378         struct nf_conn *ct;
 379
 380         rcu_read_lock();
 381 begin:
 382         h = ____nf_conntrack_find(net, zone, tuple, hash);
 383         if (h) {
 384                 ct = nf_ct_tuplehash_to_ctrack(h);
 385                 if (unlikely(nf_ct_is_dying(ct) ||
 386                              !atomic_inc_not_zero(&ct->ct_general.use)))
 387                         h = NULL;
 388                 else {
 389                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
 390                                      nf_ct_zone(ct) != zone)) {
 391                                 nf_ct_put(ct);
 392                                 goto begin;
 393                         }
 394                 }
 395         }
 396         rcu_read_unlock();
 397
 398         return h;
 399 }
 400
 401 struct nf_conntrack_tuple_hash *
 402 nf_conntrack_find_get(struct net *net, u16 zone,
 403                       const struct nf_conntrack_tuple *tuple)
 404 {
 405         return __nf_conntrack_find_get(net, zone, tuple,
 406                                        hash_conntrack_raw(tuple, zone));
 407 }
 408 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 409
 410 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 411                                        unsigned int hash,
 412                                        unsigned int repl_hash)
 413 {
 414         struct net *net = nf_ct_net(ct);
 415
 416         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 417                            &net->ct.hash[hash]);
 418         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 419                            &net->ct.hash[repl_hash]);
 420 }
 421
 422 int
 423 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 424 {
 425         struct net *net = nf_ct_net(ct);
 426         unsigned int hash, repl_hash;
 427         struct nf_conntrack_tuple_hash *h;
 428         struct hlist_nulls_node *n;
 429         u16 zone;
 430
 431         zone = nf_ct_zone(ct);
 432         hash = hash_conntrack(net, zone,
 433                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 434         repl_hash = hash_conntrack(net, zone,
 435                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 436
 437         spin_lock_bh(&nf_conntrack_lock);
 438
 439         /* See if there's one in the list already, including reverse */
 440         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 441                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 442                                       &h->tuple) &&
 443                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 444                         goto out;
 445         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 446                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 447                                       &h->tuple) &&
 448                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 449                         goto out;
 450
 451         add_timer(&ct->timeout);
 452         nf_conntrack_get(&ct->ct_general);
 453         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 454         NF_CT_STAT_INC(net, insert);
 455         spin_unlock_bh(&nf_conntrack_lock);
 456
 457         return 0;
 458
 459 out:
 460         NF_CT_STAT_INC(net, insert_failed);
 461         spin_unlock_bh(&nf_conntrack_lock);
 462         return -EEXIST;
 463 }
 464 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 465
 466 /* Confirm a connection given skb; places it in hash table */
 467 int
 468 __nf_conntrack_confirm(struct sk_buff *skb)
 469 {
 470         unsigned int hash, repl_hash;
 471         struct nf_conntrack_tuple_hash *h;
 472         struct nf_conn *ct;
 473         struct nf_conn_help *help;
 474         struct nf_conn_tstamp *tstamp;
 475         struct hlist_nulls_node *n;
 476         enum ip_conntrack_info ctinfo;
 477         struct net *net;
 478         u16 zone;
 479
 480         ct = nf_ct_get(skb, &ctinfo);
 481         net = nf_ct_net(ct);
 482
 483         /* ipt_REJECT uses nf_conntrack_attach to attach related
 484            ICMP/TCP RST packets in other direction.  Actual packet
 485            which created connection will be IP_CT_NEW or for an
 486            expected connection, IP_CT_RELATED. */
 487         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 488                 return NF_ACCEPT;
 489
 490         zone = nf_ct_zone(ct);
 491         /* reuse the hash saved before */
 492         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 493         hash = hash_bucket(hash, net);
 494         repl_hash = hash_conntrack(net, zone,
 495                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 496
 497         /* We're not in hash table, and we refuse to set up related
 498            connections for unconfirmed conns.  But packet copies and
 499            REJECT will give spurious warnings here. */
 500         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 501
 502         /* No external references means no one else could have
 503            confirmed us. */
 504         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 505         pr_debug("Confirming conntrack %p\n", ct);
 506
 507         spin_lock_bh(&nf_conntrack_lock);
 508
 509         /* We have to check the DYING flag inside the lock to prevent
 510            a race against nf_ct_get_next_corpse() possibly called from
 511            user context, else we insert an already 'dead' hash, blocking
 512            further use of that particular connection -JM */
 513
 514         if (unlikely(nf_ct_is_dying(ct))) {
 515                 spin_unlock_bh(&nf_conntrack_lock);
 516                 return NF_ACCEPT;
 517         }
 518
 519         /* See if there's one in the list already, including reverse:
 520            NAT could have grabbed it without realizing, since we're
 521            not in the hash.  If there is, we lost race. */
 522         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 523                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 524                                       &h->tuple) &&
 525                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 526                         goto out;
 527         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 528                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 529                                       &h->tuple) &&
 530                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 531                         goto out;
 532
 533         /* Remove from unconfirmed list */
 534         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 535
 536         /* Timer relative to confirmation time, not original
 537            setting time, otherwise we'd get timer wrap in
 538            weird delay cases. */
 539         ct->timeout.expires += jiffies;
 540         add_timer(&ct->timeout);
 541         atomic_inc(&ct->ct_general.use);
 542         ct->status |= IPS_CONFIRMED;
 543
 544         /* set conntrack timestamp, if enabled. */
 545         tstamp = nf_conn_tstamp_find(ct);
 546         if (tstamp) {
 547                 if (skb->tstamp.tv64 == 0)
 548                         __net_timestamp(skb);
 549
 550                 tstamp->start = ktime_to_ns(skb->tstamp);
 551         }
 552         /* Since the lookup is lockless, hash insertion must be done after
 553          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 554          * guarantee that no other CPU can find the conntrack before the above
 555          * stores are visible.
 556          */
 557         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 558         NF_CT_STAT_INC(net, insert);
 559         spin_unlock_bh(&nf_conntrack_lock);
 560
 561         help = nfct_help(ct);
 562         if (help && help->helper)
 563                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 564
 565         nf_conntrack_event_cache(master_ct(ct) ?
 566                                  IPCT_RELATED : IPCT_NEW, ct);
 567         return NF_ACCEPT;
 568
 569 out:
 570         NF_CT_STAT_INC(net, insert_failed);
 571         spin_unlock_bh(&nf_conntrack_lock);
 572         return NF_DROP;
 573 }
 574 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 575
 576 /* Returns true if a connection correspondings to the tuple (required
 577    for NAT). */
 578 int
 579 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 580                          const struct nf_conn *ignored_conntrack)
 581 {
 582         struct net *net = nf_ct_net(ignored_conntrack);
 583         struct nf_conntrack_tuple_hash *h;
 584         struct hlist_nulls_node *n;
 585         struct nf_conn *ct;
 586         u16 zone = nf_ct_zone(ignored_conntrack);
 587         unsigned int hash = hash_conntrack(net, zone, tuple);
 588
 589         /* Disable BHs the entire time since we need to disable them at
 590          * least once for the stats anyway.
 591          */
 592         rcu_read_lock_bh();
 593         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 594                 ct = nf_ct_tuplehash_to_ctrack(h);
 595                 if (ct != ignored_conntrack &&
 596                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 597                     nf_ct_zone(ct) == zone) {
 598                         NF_CT_STAT_INC(net, found);
 599                         rcu_read_unlock_bh();
 600                         return 1;
 601                 }
 602                 NF_CT_STAT_INC(net, searched);
 603         }
 604         rcu_read_unlock_bh();
 605
 606         return 0;
 607 }
 608 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 609
 610 #define NF_CT_EVICTION_RANGE    8
 611
 612 /* There's a small race here where we may free a just-assured
 613    connection.  Too bad: we're in trouble anyway. */
 614 static noinline int early_drop(struct net *net, unsigned int hash)
 615 {
 616         /* Use oldest entry, which is roughly LRU */
 617         struct nf_conntrack_tuple_hash *h;
 618         struct nf_conn *ct = NULL, *tmp;
 619         struct hlist_nulls_node *n;
 620         unsigned int i, cnt = 0;
 621         int dropped = 0;
 622
 623         rcu_read_lock();
 624         for (i = 0; i < net->ct.htable_size; i++) {
 625                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 626                                          hnnode) {
 627                         tmp = nf_ct_tuplehash_to_ctrack(h);
 628                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 629                                 ct = tmp;
 630                         cnt++;
 631                 }
 632
 633                 if (ct != NULL) {
 634                         if (likely(!nf_ct_is_dying(ct) &&
 635                                    atomic_inc_not_zero(&ct->ct_general.use)))
 636                                 break;
 637                         else
 638                                 ct = NULL;
 639                 }
 640
 641                 if (cnt >= NF_CT_EVICTION_RANGE)
 642                         break;
 643
 644                 hash = (hash + 1) % net->ct.htable_size;
 645         }
 646         rcu_read_unlock();
 647
 648         if (!ct)
 649                 return dropped;
 650
 651         if (del_timer(&ct->timeout)) {
 652                 if (nf_ct_delete(ct, 0, 0)) {
 653                         dropped = 1;
 654                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 655                 }
 656         }
 657         nf_ct_put(ct);
 658         return dropped;
 659 }
 660
 661 void init_nf_conntrack_hash_rnd(void)
 662 {
 663         unsigned int rand;
 664
 665         /*
 666          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 667          * Because there isn't enough entropy when system initializing,
 668          * and we initialize it as late as possible.
 669          */
 670         do {
 671                 get_random_bytes(&rand, sizeof(rand));
 672         } while (!rand);
 673         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 674 }
 675
 676 static struct nf_conn *
 677 __nf_conntrack_alloc(struct net *net, u16 zone,
 678                      const struct nf_conntrack_tuple *orig,
 679                      const struct nf_conntrack_tuple *repl,
 680                      gfp_t gfp, u32 hash)
 681 {
 682         struct nf_conn *ct;
 683
 684         if (unlikely(!nf_conntrack_hash_rnd)) {
 685                 init_nf_conntrack_hash_rnd();
 686                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 687                 hash = hash_conntrack_raw(orig, zone);
 688         }
 689
 690         /* We don't want any race condition at early drop stage */
 691         atomic_inc(&net->ct.count);
 692
 693         if (nf_conntrack_max &&
 694             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 695                 if (!early_drop(net, hash_bucket(hash, net))) {
 696                         atomic_dec(&net->ct.count);
 697                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
 698                         return ERR_PTR(-ENOMEM);
 699                 }
 700         }
 701
 702         /*
 703          * Do not use kmem_cache_zalloc(), as this cache uses
 704          * SLAB_DESTROY_BY_RCU.
 705          */
 706         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 707         if (ct == NULL) {
 708                 atomic_dec(&net->ct.count);
 709                 return ERR_PTR(-ENOMEM);
 710         }
 711         /*
 712          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 713          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 714          */
 715         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 716                offsetof(struct nf_conn, proto) -
 717                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 718         spin_lock_init(&ct->lock);
 719         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 720         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 721         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 722         /* save hash for reusing when confirming */
 723         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 724         /* Don't set timer yet: wait for confirmation */
 725         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 726         write_pnet(&ct->ct_net, net);
 727 #ifdef CONFIG_NF_CONNTRACK_ZONES
 728         if (zone) {
 729                 struct nf_conntrack_zone *nf_ct_zone;
 730
 731                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 732                 if (!nf_ct_zone)
 733                         goto out_free;
 734                 nf_ct_zone->id = zone;
 735         }
 736 #endif
 737         /*
 738          * changes to lookup keys must be done before setting refcnt to 1
 739          */
 740         smp_wmb();
 741         atomic_set(&ct->ct_general.use, 1);
 742         return ct;
 743
 744 #ifdef CONFIG_NF_CONNTRACK_ZONES
 745 out_free:
 746         atomic_dec(&net->ct.count);
 747         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 748         return ERR_PTR(-ENOMEM);
 749 #endif
 750 }
 751
 752 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 753                                    const struct nf_conntrack_tuple *orig,
 754                                    const struct nf_conntrack_tuple *repl,
 755                                    gfp_t gfp)
 756 {
 757         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 758 }
 759 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 760
 761 void nf_conntrack_free(struct nf_conn *ct)
 762 {
 763         struct net *net = nf_ct_net(ct);
 764
 765         nf_ct_ext_destroy(ct);
 766         atomic_dec(&net->ct.count);
 767         nf_ct_ext_free(ct);
 768         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 769 }
 770 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 771
 772
 773 /* Allocate a new conntrack: we return -ENOMEM if classification
 774    failed due to stress.  Otherwise it really is unclassifiable. */
 775 static struct nf_conntrack_tuple_hash *
 776 init_conntrack(struct net *net, struct nf_conn *tmpl,
 777                const struct nf_conntrack_tuple *tuple,
 778                struct nf_conntrack_l3proto *l3proto,
 779                struct nf_conntrack_l4proto *l4proto,
 780                struct sk_buff *skb,
 781                unsigned int dataoff, u32 hash)
 782 {
 783         struct nf_conn *ct;
 784         struct nf_conn_help *help;
 785         struct nf_conntrack_tuple repl_tuple;
 786         struct nf_conntrack_ecache *ecache;
 787         struct nf_conntrack_expect *exp;
 788         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 789         struct nf_conn_timeout *timeout_ext;
 790         unsigned int *timeouts;
 791
 792         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 793                 pr_debug("Can't invert tuple.\n");
 794                 return NULL;
 795         }
 796
 797         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 798                                   hash);
 799         if (IS_ERR(ct))
 800                 return (struct nf_conntrack_tuple_hash *)ct;
 801
 802         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 803         if (timeout_ext)
 804                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 805         else
 806                 timeouts = l4proto->get_timeouts(net);
 807
 808         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 809                 nf_conntrack_free(ct);
 810                 pr_debug("init conntrack: can't track with proto module\n");
 811                 return NULL;
 812         }
 813
 814         if (timeout_ext)
 815                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 816
 817         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 818         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 819         nf_ct_labels_ext_add(ct);
 820
 821         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 822         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 823                                  ecache ? ecache->expmask : 0,
 824                              GFP_ATOMIC);
 825
 826         spin_lock_bh(&nf_conntrack_lock);
 827         exp = nf_ct_find_expectation(net, zone, tuple);
 828         if (exp) {
 829                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 830                          ct, exp);
 831                 /* Welcome, Mr. Bond.  We've been expecting you... */
 832                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 833                 ct->master = exp->master;
 834                 if (exp->helper) {
 835                         help = nf_ct_helper_ext_add(ct, exp->helper,
 836                                                     GFP_ATOMIC);
 837                         if (help)
 838                                 rcu_assign_pointer(help->helper, exp->helper);
 839                 }
 840
 841 #ifdef CONFIG_NF_CONNTRACK_MARK
 842                 ct->mark = exp->master->mark;
 843 #endif
 844 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 845                 ct->secmark = exp->master->secmark;
 846 #endif
 847                 nf_conntrack_get(&ct->master->ct_general);
 848                 NF_CT_STAT_INC(net, expect_new);
 849         } else {
 850                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 851                 NF_CT_STAT_INC(net, new);
 852         }
 853
 854         /* Overload tuple linked list to put us in unconfirmed list. */
 855         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 856                        &net->ct.unconfirmed);
 857
 858         spin_unlock_bh(&nf_conntrack_lock);
 859
 860         if (exp) {
 861                 if (exp->expectfn)
 862                         exp->expectfn(ct, exp);
 863                 nf_ct_expect_put(exp);
 864         }
 865
 866         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 867 }
 868
 869 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 870 static inline struct nf_conn *
 871 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 872                   struct sk_buff *skb,
 873                   unsigned int dataoff,
 874                   u_int16_t l3num,
 875                   u_int8_t protonum,
 876                   struct nf_conntrack_l3proto *l3proto,
 877                   struct nf_conntrack_l4proto *l4proto,
 878                   int *set_reply,
 879                   enum ip_conntrack_info *ctinfo)
 880 {
 881         struct nf_conntrack_tuple tuple;
 882         struct nf_conntrack_tuple_hash *h;
 883         struct nf_conn *ct;
 884         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 885         u32 hash;
 886
 887         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 888                              dataoff, l3num, protonum, &tuple, l3proto,
 889                              l4proto)) {
 890                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 891                 return NULL;
 892         }
 893
 894         /* look for tuple match */
 895         hash = hash_conntrack_raw(&tuple, zone);
 896         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 897         if (!h) {
 898                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 899                                    skb, dataoff, hash);
 900                 if (!h)
 901                         return NULL;
 902                 if (IS_ERR(h))
 903                         return (void *)h;
 904         }
 905         ct = nf_ct_tuplehash_to_ctrack(h);
 906
 907         /* It exists; we have (non-exclusive) reference. */
 908         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 909                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 910                 /* Please set reply bit if this packet OK */
 911                 *set_reply = 1;
 912         } else {
 913                 /* Once we've had two way comms, always ESTABLISHED. */
 914                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 915                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 916                         *ctinfo = IP_CT_ESTABLISHED;
 917                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 918                         pr_debug("nf_conntrack_in: related packet for %p\n",
 919                                  ct);
 920                         *ctinfo = IP_CT_RELATED;
 921                 } else {
 922                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 923                         *ctinfo = IP_CT_NEW;
 924                 }
 925                 *set_reply = 0;
 926         }
 927         skb->nfct = &ct->ct_general;
 928         skb->nfctinfo = *ctinfo;
 929         return ct;
 930 }
 931
 932 unsigned int
 933 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 934                 struct sk_buff *skb)
 935 {
 936         struct nf_conn *ct, *tmpl = NULL;
 937         enum ip_conntrack_info ctinfo;
 938         struct nf_conntrack_l3proto *l3proto;
 939         struct nf_conntrack_l4proto *l4proto;
 940         unsigned int *timeouts;
 941         unsigned int dataoff;
 942         u_int8_t protonum;
 943         int set_reply = 0;
 944         int ret;
 945
 946         if (skb->nfct) {
 947                 /* Previously seen (loopback or untracked)?  Ignore. */
 948                 tmpl = (struct nf_conn *)skb->nfct;
 949                 if (!nf_ct_is_template(tmpl)) {
 950                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 951                         return NF_ACCEPT;
 952                 }
 953                 skb->nfct = NULL;
 954         }
 955
 956         /* rcu_read_lock()ed by nf_hook_slow */
 957         l3proto = __nf_ct_l3proto_find(pf);
 958         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 959                                    &dataoff, &protonum);
 960         if (ret <= 0) {
 961                 pr_debug("not prepared to track yet or error occurred\n");
 962                 NF_CT_STAT_INC_ATOMIC(net, error);
 963                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 964                 ret = -ret;
 965                 goto out;
 966         }
 967
 968         l4proto = __nf_ct_l4proto_find(pf, protonum);
 969
 970         /* It may be an special packet, error, unclean...
 971          * inverse of the return code tells to the netfilter
 972          * core what to do with the packet. */
 973         if (l4proto->error != NULL) {
 974                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
 975                                      pf, hooknum);
 976                 if (ret <= 0) {
 977                         NF_CT_STAT_INC_ATOMIC(net, error);
 978                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 979                         ret = -ret;
 980                         goto out;
 981                 }
 982                 /* ICMP[v6] protocol trackers may assign one conntrack. */
 983                 if (skb->nfct)
 984                         goto out;
 985         }
 986
 987         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
 988                                l3proto, l4proto, &set_reply, &ctinfo);
 989         if (!ct) {
 990                 /* Not valid part of a connection */
 991                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 992                 ret = NF_ACCEPT;
 993                 goto out;
 994         }
 995
 996         if (IS_ERR(ct)) {
 997                 /* Too stressed to deal. */
 998                 NF_CT_STAT_INC_ATOMIC(net, drop);
 999                 ret = NF_DROP;
1000                 goto out;
1001         }
1002
1003         NF_CT_ASSERT(skb->nfct);
1004
1005         /* Decide what timeout policy we want to apply to this flow. */
1006         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1007
1008         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1009         if (ret <= 0) {
1010                 /* Invalid: inverse of the return code tells
1011                  * the netfilter core what to do */
1012                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1013                 nf_conntrack_put(skb->nfct);
1014                 skb->nfct = NULL;
1015                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1016                 if (ret == -NF_DROP)
1017                         NF_CT_STAT_INC_ATOMIC(net, drop);
1018                 ret = -ret;
1019                 goto out;
1020         }
1021
1022         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1023                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1024 out:
1025         if (tmpl) {
1026                 /* Special case: we have to repeat this hook, assign the
1027                  * template again to this packet. We assume that this packet
1028                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1029                 if (ret == NF_REPEAT)
1030                         skb->nfct = (struct nf_conntrack *)tmpl;
1031                 else
1032                         nf_ct_put(tmpl);
1033         }
1034
1035         return ret;
1036 }
1037 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1038
1039 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1040                           const struct nf_conntrack_tuple *orig)
1041 {
1042         bool ret;
1043
1044         rcu_read_lock();
1045         ret = nf_ct_invert_tuple(inverse, orig,
1046                                  __nf_ct_l3proto_find(orig->src.l3num),
1047                                  __nf_ct_l4proto_find(orig->src.l3num,
1048                                                       orig->dst.protonum));
1049         rcu_read_unlock();
1050         return ret;
1051 }
1052 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1053
1054 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1055    implicitly racy: see __nf_conntrack_confirm */
1056 void nf_conntrack_alter_reply(struct nf_conn *ct,
1057                               const struct nf_conntrack_tuple *newreply)
1058 {
1059         struct nf_conn_help *help = nfct_help(ct);
1060
1061         /* Should be unconfirmed, so not in hash table yet */
1062         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1063
1064         pr_debug("Altering reply tuple of %p to ", ct);
1065         nf_ct_dump_tuple(newreply);
1066
1067         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1068         if (ct->master || (help && !hlist_empty(&help->expectations)))
1069                 return;
1070
1071         rcu_read_lock();
1072         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1073         rcu_read_unlock();
1074 }
1075 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1076
1077 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1078 void __nf_ct_refresh_acct(struct nf_conn *ct,
1079                           enum ip_conntrack_info ctinfo,
1080                           const struct sk_buff *skb,
1081                           unsigned long extra_jiffies,
1082                           int do_acct)
1083 {
1084         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1085         NF_CT_ASSERT(skb);
1086
1087         /* Only update if this is not a fixed timeout */
1088         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1089                 goto acct;
1090
1091         /* If not in hash table, timer will not be active yet */
1092         if (!nf_ct_is_confirmed(ct)) {
1093                 ct->timeout.expires = extra_jiffies;
1094         } else {
1095                 unsigned long newtime = jiffies + extra_jiffies;
1096
1097                 /* Only update the timeout if the new timeout is at least
1098                    HZ jiffies from the old timeout. Need del_timer for race
1099                    avoidance (may already be dying). */
1100                 if (newtime - ct->timeout.expires >= HZ)
1101                         mod_timer_pending(&ct->timeout, newtime);
1102         }
1103
1104 acct:
1105         if (do_acct) {
1106                 struct nf_conn_counter *acct;
1107
1108                 acct = nf_conn_acct_find(ct);
1109                 if (acct) {
1110                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1111                         atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes);
1112                 }
1113         }
1114 }
1115 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1116
1117 bool __nf_ct_kill_acct(struct nf_conn *ct,
1118                        enum ip_conntrack_info ctinfo,
1119                        const struct sk_buff *skb,
1120                        int do_acct)
1121 {
1122         if (do_acct) {
1123                 struct nf_conn_counter *acct;
1124
1125                 acct = nf_conn_acct_find(ct);
1126                 if (acct) {
1127                         atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets);
1128                         atomic64_add(skb->len - skb_network_offset(skb),
1129                                      &acct[CTINFO2DIR(ctinfo)].bytes);
1130                 }
1131         }
1132
1133         if (del_timer(&ct->timeout)) {
1134                 ct->timeout.function((unsigned long)ct);
1135                 return true;
1136         }
1137         return false;
1138 }
1139 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1140
1141 #ifdef CONFIG_NF_CONNTRACK_ZONES
1142 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1143         .len    = sizeof(struct nf_conntrack_zone),
1144         .align  = __alignof__(struct nf_conntrack_zone),
1145         .id     = NF_CT_EXT_ZONE,
1146 };
1147 #endif
1148
1149 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1150
1151 #include <linux/netfilter/nfnetlink.h>
1152 #include <linux/netfilter/nfnetlink_conntrack.h>
1153 #include <linux/mutex.h>
1154
1155 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1156  * in ip_conntrack_core, since we don't want the protocols to autoload
1157  * or depend on ctnetlink */
1158 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1159                                const struct nf_conntrack_tuple *tuple)
1160 {
1161         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1162             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1163                 goto nla_put_failure;
1164         return 0;
1165
1166 nla_put_failure:
1167         return -1;
1168 }
1169 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1170
1171 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1172         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1173         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1174 };
1175 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1176
1177 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1178                                struct nf_conntrack_tuple *t)
1179 {
1180         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1181                 return -EINVAL;
1182
1183         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1184         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1185
1186         return 0;
1187 }
1188 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1189
1190 int nf_ct_port_nlattr_tuple_size(void)
1191 {
1192         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1193 }
1194 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1195 #endif
1196
1197 /* Used by ipt_REJECT and ip6t_REJECT. */
1198 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1199 {
1200         struct nf_conn *ct;
1201         enum ip_conntrack_info ctinfo;
1202
1203         /* This ICMP is in reverse direction to the packet which caused it */
1204         ct = nf_ct_get(skb, &ctinfo);
1205         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1206                 ctinfo = IP_CT_RELATED_REPLY;
1207         else
1208                 ctinfo = IP_CT_RELATED;
1209
1210         /* Attach to new skbuff, and increment count */
1211         nskb->nfct = &ct->ct_general;
1212         nskb->nfctinfo = ctinfo;
1213         nf_conntrack_get(nskb->nfct);
1214 }
1215
1216 /* Bring out ya dead! */
1217 static struct nf_conn *
1218 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1219                 void *data, unsigned int *bucket)
1220 {
1221         struct nf_conntrack_tuple_hash *h;
1222         struct nf_conn *ct;
1223         struct hlist_nulls_node *n;
1224
1225         spin_lock_bh(&nf_conntrack_lock);
1226         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1227                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1228                         if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1229                                 continue;
1230                         ct = nf_ct_tuplehash_to_ctrack(h);
1231                         if (iter(ct, data))
1232                                 goto found;
1233                 }
1234         }
1235         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1236                 ct = nf_ct_tuplehash_to_ctrack(h);
1237                 if (iter(ct, data))
1238                         set_bit(IPS_DYING_BIT, &ct->status);
1239         }
1240         spin_unlock_bh(&nf_conntrack_lock);
1241         return NULL;
1242 found:
1243         atomic_inc(&ct->ct_general.use);
1244         spin_unlock_bh(&nf_conntrack_lock);
1245         return ct;
1246 }
1247
1248 void nf_ct_iterate_cleanup(struct net *net,
1249                            int (*iter)(struct nf_conn *i, void *data),
1250                            void *data, u32 portid, int report)
1251 {
1252         struct nf_conn *ct;
1253         unsigned int bucket = 0;
1254
1255         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1256                 /* Time to push up daises... */
1257                 if (del_timer(&ct->timeout))
1258                         nf_ct_delete(ct, portid, report);
1259
1260                 /* ... else the timer will get him soon. */
1261
1262                 nf_ct_put(ct);
1263         }
1264 }
1265 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1266
1267 static int kill_all(struct nf_conn *i, void *data)
1268 {
1269         return 1;
1270 }
1271
1272 void nf_ct_free_hashtable(void *hash, unsigned int size)
1273 {
1274         if (is_vmalloc_addr(hash))
1275                 vfree(hash);
1276         else
1277                 free_pages((unsigned long)hash,
1278                            get_order(sizeof(struct hlist_head) * size));
1279 }
1280 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1281
1282 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1283 {
1284         nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1285 }
1286 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1287
1288 static void nf_ct_release_dying_list(struct net *net)
1289 {
1290         struct nf_conntrack_tuple_hash *h;
1291         struct nf_conn *ct;
1292         struct hlist_nulls_node *n;
1293
1294         spin_lock_bh(&nf_conntrack_lock);
1295         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1296                 ct = nf_ct_tuplehash_to_ctrack(h);
1297                 /* never fails to remove them, no listeners at this point */
1298                 nf_ct_kill(ct);
1299         }
1300         spin_unlock_bh(&nf_conntrack_lock);
1301 }
1302
1303 static int untrack_refs(void)
1304 {
1305         int cnt = 0, cpu;
1306
1307         for_each_possible_cpu(cpu) {
1308                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1309
1310                 cnt += atomic_read(&ct->ct_general.use) - 1;
1311         }
1312         return cnt;
1313 }
1314
1315 void nf_conntrack_cleanup_start(void)
1316 {
1317         RCU_INIT_POINTER(ip_ct_attach, NULL);
1318 }
1319
1320 void nf_conntrack_cleanup_end(void)
1321 {
1322         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1323         while (untrack_refs() > 0)
1324                 schedule();
1325
1326 #ifdef CONFIG_NF_CONNTRACK_ZONES
1327         nf_ct_extend_unregister(&nf_ct_zone_extend);
1328 #endif
1329         nf_conntrack_proto_fini();
1330         nf_conntrack_seqadj_fini();
1331         nf_conntrack_labels_fini();
1332         nf_conntrack_helper_fini();
1333         nf_conntrack_timeout_fini();
1334         nf_conntrack_ecache_fini();
1335         nf_conntrack_tstamp_fini();
1336         nf_conntrack_acct_fini();
1337         nf_conntrack_expect_fini();
1338 }
1339
1340 /*
1341  * Mishearing the voices in his head, our hero wonders how he's
1342  * supposed to kill the mall.
1343  */
1344 void nf_conntrack_cleanup_net(struct net *net)
1345 {
1346         LIST_HEAD(single);
1347
1348         list_add(&net->exit_list, &single);
1349         nf_conntrack_cleanup_net_list(&single);
1350 }
1351
1352 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1353 {
1354         int busy;
1355         struct net *net;
1356
1357         /*
1358          * This makes sure all current packets have passed through
1359          *  netfilter framework.  Roll on, two-stage module
1360          *  delete...
1361          */
1362         synchronize_net();
1363 i_see_dead_people:
1364         busy = 0;
1365         list_for_each_entry(net, net_exit_list, exit_list) {
1366                 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1367                 nf_ct_release_dying_list(net);
1368                 if (atomic_read(&net->ct.count) != 0)
1369                         busy = 1;
1370         }
1371         if (busy) {
1372                 schedule();
1373                 goto i_see_dead_people;
1374         }
1375
1376         list_for_each_entry(net, net_exit_list, exit_list) {
1377                 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1378                 nf_conntrack_proto_pernet_fini(net);
1379                 nf_conntrack_helper_pernet_fini(net);
1380                 nf_conntrack_ecache_pernet_fini(net);
1381                 nf_conntrack_tstamp_pernet_fini(net);
1382                 nf_conntrack_acct_pernet_fini(net);
1383                 nf_conntrack_expect_pernet_fini(net);
1384                 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1385                 kfree(net->ct.slabname);
1386                 free_percpu(net->ct.stat);
1387         }
1388 }
1389
1390 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1391 {
1392         struct hlist_nulls_head *hash;
1393         unsigned int nr_slots, i;
1394         size_t sz;
1395
1396         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1397         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1398         sz = nr_slots * sizeof(struct hlist_nulls_head);
1399         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1400                                         get_order(sz));
1401         if (!hash) {
1402                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1403                 hash = vzalloc(sz);
1404         }
1405
1406         if (hash && nulls)
1407                 for (i = 0; i < nr_slots; i++)
1408                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1409
1410         return hash;
1411 }
1412 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1413
1414 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1415 {
1416         int i, bucket, rc;
1417         unsigned int hashsize, old_size;
1418         struct hlist_nulls_head *hash, *old_hash;
1419         struct nf_conntrack_tuple_hash *h;
1420         struct nf_conn *ct;
1421
1422         if (current->nsproxy->net_ns != &init_net)
1423                 return -EOPNOTSUPP;
1424
1425         /* On boot, we can set this without any fancy locking. */
1426         if (!nf_conntrack_htable_size)
1427                 return param_set_uint(val, kp);
1428
1429         rc = kstrtouint(val, 0, &hashsize);
1430         if (rc)
1431                 return rc;
1432         if (!hashsize)
1433                 return -EINVAL;
1434
1435         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1436         if (!hash)
1437                 return -ENOMEM;
1438
1439         /* Lookups in the old hash might happen in parallel, which means we
1440          * might get false negatives during connection lookup. New connections
1441          * created because of a false negative won't make it into the hash
1442          * though since that required taking the lock.
1443          */
1444         spin_lock_bh(&nf_conntrack_lock);
1445         for (i = 0; i < init_net.ct.htable_size; i++) {
1446                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1447                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1448                                         struct nf_conntrack_tuple_hash, hnnode);
1449                         ct = nf_ct_tuplehash_to_ctrack(h);
1450                         hlist_nulls_del_rcu(&h->hnnode);
1451                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1452                                                   hashsize);
1453                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1454                 }
1455         }
1456         old_size = init_net.ct.htable_size;
1457         old_hash = init_net.ct.hash;
1458
1459         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1460         init_net.ct.hash = hash;
1461         spin_unlock_bh(&nf_conntrack_lock);
1462
1463         nf_ct_free_hashtable(old_hash, old_size);
1464         return 0;
1465 }
1466 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1467
1468 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1469                   &nf_conntrack_htable_size, 0600);
1470
1471 void nf_ct_untracked_status_or(unsigned long bits)
1472 {
1473         int cpu;
1474
1475         for_each_possible_cpu(cpu)
1476                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1477 }
1478 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1479
1480 int nf_conntrack_init_start(void)
1481 {
1482         int max_factor = 8;
1483         int ret, cpu;
1484
1485         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1486          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1487         if (!nf_conntrack_htable_size) {
1488                 nf_conntrack_htable_size
1489                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1490                            / sizeof(struct hlist_head));
1491                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1492                         nf_conntrack_htable_size = 16384;
1493                 if (nf_conntrack_htable_size < 32)
1494                         nf_conntrack_htable_size = 32;
1495
1496                 /* Use a max. factor of four by default to get the same max as
1497                  * with the old struct list_heads. When a table size is given
1498                  * we use the old value of 8 to avoid reducing the max.
1499                  * entries. */
1500                 max_factor = 4;
1501         }
1502         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1503
1504         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1505                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1506                nf_conntrack_max);
1507
1508         ret = nf_conntrack_expect_init();
1509         if (ret < 0)
1510                 goto err_expect;
1511
1512         ret = nf_conntrack_acct_init();
1513         if (ret < 0)
1514                 goto err_acct;
1515
1516         ret = nf_conntrack_tstamp_init();
1517         if (ret < 0)
1518                 goto err_tstamp;
1519
1520         ret = nf_conntrack_ecache_init();
1521         if (ret < 0)
1522                 goto err_ecache;
1523
1524         ret = nf_conntrack_timeout_init();
1525         if (ret < 0)
1526                 goto err_timeout;
1527
1528         ret = nf_conntrack_helper_init();
1529         if (ret < 0)
1530                 goto err_helper;
1531
1532         ret = nf_conntrack_labels_init();
1533         if (ret < 0)
1534                 goto err_labels;
1535
1536         ret = nf_conntrack_seqadj_init();
1537         if (ret < 0)
1538                 goto err_seqadj;
1539
1540 #ifdef CONFIG_NF_CONNTRACK_ZONES
1541         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1542         if (ret < 0)
1543                 goto err_extend;
1544 #endif
1545         ret = nf_conntrack_proto_init();
1546         if (ret < 0)
1547                 goto err_proto;
1548
1549         /* Set up fake conntrack: to never be deleted, not in any hashes */
1550         for_each_possible_cpu(cpu) {
1551                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1552                 write_pnet(&ct->ct_net, &init_net);
1553                 atomic_set(&ct->ct_general.use, 1);
1554         }
1555         /*  - and look it like as a confirmed connection */
1556         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1557         return 0;
1558
1559 err_proto:
1560 #ifdef CONFIG_NF_CONNTRACK_ZONES
1561         nf_ct_extend_unregister(&nf_ct_zone_extend);
1562 err_extend:
1563 #endif
1564         nf_conntrack_seqadj_fini();
1565 err_seqadj:
1566         nf_conntrack_labels_fini();
1567 err_labels:
1568         nf_conntrack_helper_fini();
1569 err_helper:
1570         nf_conntrack_timeout_fini();
1571 err_timeout:
1572         nf_conntrack_ecache_fini();
1573 err_ecache:
1574         nf_conntrack_tstamp_fini();
1575 err_tstamp:
1576         nf_conntrack_acct_fini();
1577 err_acct:
1578         nf_conntrack_expect_fini();
1579 err_expect:
1580         return ret;
1581 }
1582
1583 void nf_conntrack_init_end(void)
1584 {
1585         /* For use by REJECT target */
1586         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1587         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1588 }
1589
1590 /*
1591  * We need to use special "null" values, not used in hash table
1592  */
1593 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1594 #define DYING_NULLS_VAL         ((1<<30)+1)
1595 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1596
1597 int nf_conntrack_init_net(struct net *net)
1598 {
1599         int ret;
1600
1601         atomic_set(&net->ct.count, 0);
1602         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1603         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1604         INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1605         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1606         if (!net->ct.stat) {
1607                 ret = -ENOMEM;
1608                 goto err_stat;
1609         }
1610
1611         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1612         if (!net->ct.slabname) {
1613                 ret = -ENOMEM;
1614                 goto err_slabname;
1615         }
1616
1617         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1618                                                         sizeof(struct nf_conn), 0,
1619                                                         SLAB_DESTROY_BY_RCU, NULL);
1620         if (!net->ct.nf_conntrack_cachep) {
1621                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1622                 ret = -ENOMEM;
1623                 goto err_cache;
1624         }
1625
1626         net->ct.htable_size = nf_conntrack_htable_size;
1627         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1628         if (!net->ct.hash) {
1629                 ret = -ENOMEM;
1630                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1631                 goto err_hash;
1632         }
1633         ret = nf_conntrack_expect_pernet_init(net);
1634         if (ret < 0)
1635                 goto err_expect;
1636         ret = nf_conntrack_acct_pernet_init(net);
1637         if (ret < 0)
1638                 goto err_acct;
1639         ret = nf_conntrack_tstamp_pernet_init(net);
1640         if (ret < 0)
1641                 goto err_tstamp;
1642         ret = nf_conntrack_ecache_pernet_init(net);
1643         if (ret < 0)
1644                 goto err_ecache;
1645         ret = nf_conntrack_helper_pernet_init(net);
1646         if (ret < 0)
1647                 goto err_helper;
1648         ret = nf_conntrack_proto_pernet_init(net);
1649         if (ret < 0)
1650                 goto err_proto;
1651         return 0;
1652
1653 err_proto:
1654         nf_conntrack_helper_pernet_fini(net);
1655 err_helper:
1656         nf_conntrack_ecache_pernet_fini(net);
1657 err_ecache:
1658         nf_conntrack_tstamp_pernet_fini(net);
1659 err_tstamp:
1660         nf_conntrack_acct_pernet_fini(net);
1661 err_acct:
1662         nf_conntrack_expect_pernet_fini(net);
1663 err_expect:
1664         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1665 err_hash:
1666         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1667 err_cache:
1668         kfree(net->ct.slabname);
1669 err_slabname:
1670         free_percpu(net->ct.stat);
1671 err_stat:
1672         return ret;
1673 }