Pileus Git - ~andy/linux/blob - net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  */
  14
  15 #include <linux/types.h>
  16 #include <linux/netfilter.h>
  17 #include <linux/module.h>
  18 #include <linux/sched.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/proc_fs.h>
  21 #include <linux/vmalloc.h>
  22 #include <linux/stddef.h>
  23 #include <linux/slab.h>
  24 #include <linux/random.h>
  25 #include <linux/jhash.h>
  26 #include <linux/err.h>
  27 #include <linux/percpu.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/notifier.h>
  30 #include <linux/kernel.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/socket.h>
  33 #include <linux/mm.h>
  34 #include <linux/nsproxy.h>
  35 #include <linux/rculist_nulls.h>
  36
  37 #include <net/netfilter/nf_conntrack.h>
  38 #include <net/netfilter/nf_conntrack_l3proto.h>
  39 #include <net/netfilter/nf_conntrack_l4proto.h>
  40 #include <net/netfilter/nf_conntrack_expect.h>
  41 #include <net/netfilter/nf_conntrack_helper.h>
  42 #include <net/netfilter/nf_conntrack_seqadj.h>
  43 #include <net/netfilter/nf_conntrack_core.h>
  44 #include <net/netfilter/nf_conntrack_extend.h>
  45 #include <net/netfilter/nf_conntrack_acct.h>
  46 #include <net/netfilter/nf_conntrack_ecache.h>
  47 #include <net/netfilter/nf_conntrack_zones.h>
  48 #include <net/netfilter/nf_conntrack_timestamp.h>
  49 #include <net/netfilter/nf_conntrack_timeout.h>
  50 #include <net/netfilter/nf_conntrack_labels.h>
  51 #include <net/netfilter/nf_conntrack_synproxy.h>
  52 #include <net/netfilter/nf_nat.h>
  53 #include <net/netfilter/nf_nat_core.h>
  54 #include <net/netfilter/nf_nat_helper.h>
  55
  56 #define NF_CONNTRACK_VERSION    "0.5.0"
  57
  58 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  59                                       enum nf_nat_manip_type manip,
  60                                       const struct nlattr *attr) __read_mostly;
  61 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  62
  63 DEFINE_SPINLOCK(nf_conntrack_lock);
  64 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  65
  66 unsigned int nf_conntrack_htable_size __read_mostly;
  67 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  68
  69 unsigned int nf_conntrack_max __read_mostly;
  70 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  71
  72 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
  73 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
  74
  75 unsigned int nf_conntrack_hash_rnd __read_mostly;
  76 EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd);
  77
  78 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone)
  79 {
  80         unsigned int n;
  81
  82         /* The direction must be ignored, so we hash everything up to the
  83          * destination ports (which is a multiple of 4) and treat the last
  84          * three bytes manually.
  85          */
  86         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  87         return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^
  88                       (((__force __u16)tuple->dst.u.all << 16) |
  89                       tuple->dst.protonum));
  90 }
  91
  92 static u32 __hash_bucket(u32 hash, unsigned int size)
  93 {
  94         return ((u64)hash * size) >> 32;
  95 }
  96
  97 static u32 hash_bucket(u32 hash, const struct net *net)
  98 {
  99         return __hash_bucket(hash, net->ct.htable_size);
 100 }
 101
 102 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
 103                                   u16 zone, unsigned int size)
 104 {
 105         return __hash_bucket(hash_conntrack_raw(tuple, zone), size);
 106 }
 107
 108 static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
 109                                        const struct nf_conntrack_tuple *tuple)
 110 {
 111         return __hash_conntrack(tuple, zone, net->ct.htable_size);
 112 }
 113
 114 bool
 115 nf_ct_get_tuple(const struct sk_buff *skb,
 116                 unsigned int nhoff,
 117                 unsigned int dataoff,
 118                 u_int16_t l3num,
 119                 u_int8_t protonum,
 120                 struct nf_conntrack_tuple *tuple,
 121                 const struct nf_conntrack_l3proto *l3proto,
 122                 const struct nf_conntrack_l4proto *l4proto)
 123 {
 124         memset(tuple, 0, sizeof(*tuple));
 125
 126         tuple->src.l3num = l3num;
 127         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 128                 return false;
 129
 130         tuple->dst.protonum = protonum;
 131         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 132
 133         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 134 }
 135 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 136
 137 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 138                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 139 {
 140         struct nf_conntrack_l3proto *l3proto;
 141         struct nf_conntrack_l4proto *l4proto;
 142         unsigned int protoff;
 143         u_int8_t protonum;
 144         int ret;
 145
 146         rcu_read_lock();
 147
 148         l3proto = __nf_ct_l3proto_find(l3num);
 149         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 150         if (ret != NF_ACCEPT) {
 151                 rcu_read_unlock();
 152                 return false;
 153         }
 154
 155         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 156
 157         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 158                               l3proto, l4proto);
 159
 160         rcu_read_unlock();
 161         return ret;
 162 }
 163 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 164
 165 bool
 166 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 167                    const struct nf_conntrack_tuple *orig,
 168                    const struct nf_conntrack_l3proto *l3proto,
 169                    const struct nf_conntrack_l4proto *l4proto)
 170 {
 171         memset(inverse, 0, sizeof(*inverse));
 172
 173         inverse->src.l3num = orig->src.l3num;
 174         if (l3proto->invert_tuple(inverse, orig) == 0)
 175                 return false;
 176
 177         inverse->dst.dir = !orig->dst.dir;
 178
 179         inverse->dst.protonum = orig->dst.protonum;
 180         return l4proto->invert_tuple(inverse, orig);
 181 }
 182 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 183
 184 static void
 185 clean_from_lists(struct nf_conn *ct)
 186 {
 187         pr_debug("clean_from_lists(%p)\n", ct);
 188         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 189         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 190
 191         /* Destroy all pending expectations */
 192         nf_ct_remove_expectations(ct);
 193 }
 194
 195 static void
 196 destroy_conntrack(struct nf_conntrack *nfct)
 197 {
 198         struct nf_conn *ct = (struct nf_conn *)nfct;
 199         struct net *net = nf_ct_net(ct);
 200         struct nf_conntrack_l4proto *l4proto;
 201
 202         pr_debug("destroy_conntrack(%p)\n", ct);
 203         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 204         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 205
 206         /* To make sure we don't get any weird locking issues here:
 207          * destroy_conntrack() MUST NOT be called with a write lock
 208          * to nf_conntrack_lock!!! -HW */
 209         rcu_read_lock();
 210         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 211         if (l4proto && l4proto->destroy)
 212                 l4proto->destroy(ct);
 213
 214         rcu_read_unlock();
 215
 216         spin_lock_bh(&nf_conntrack_lock);
 217         /* Expectations will have been removed in clean_from_lists,
 218          * except TFTP can create an expectation on the first packet,
 219          * before connection is in the list, so we need to clean here,
 220          * too. */
 221         nf_ct_remove_expectations(ct);
 222
 223         /* We overload first tuple to link into unconfirmed or dying list.*/
 224         BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 225         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 226
 227         NF_CT_STAT_INC(net, delete);
 228         spin_unlock_bh(&nf_conntrack_lock);
 229
 230         if (ct->master)
 231                 nf_ct_put(ct->master);
 232
 233         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 234         nf_conntrack_free(ct);
 235 }
 236
 237 static void nf_ct_delete_from_lists(struct nf_conn *ct)
 238 {
 239         struct net *net = nf_ct_net(ct);
 240
 241         nf_ct_helper_destroy(ct);
 242         spin_lock_bh(&nf_conntrack_lock);
 243         /* Inside lock so preempt is disabled on module removal path.
 244          * Otherwise we can get spurious warnings. */
 245         NF_CT_STAT_INC(net, delete_list);
 246         clean_from_lists(ct);
 247         /* add this conntrack to the dying list */
 248         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 249                              &net->ct.dying);
 250         spin_unlock_bh(&nf_conntrack_lock);
 251 }
 252
 253 static void death_by_event(unsigned long ul_conntrack)
 254 {
 255         struct nf_conn *ct = (void *)ul_conntrack;
 256         struct net *net = nf_ct_net(ct);
 257         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 258
 259         BUG_ON(ecache == NULL);
 260
 261         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
 262                 /* bad luck, let's retry again */
 263                 ecache->timeout.expires = jiffies +
 264                         (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 265                 add_timer(&ecache->timeout);
 266                 return;
 267         }
 268         /* we've got the event delivered, now it's dying */
 269         set_bit(IPS_DYING_BIT, &ct->status);
 270         nf_ct_put(ct);
 271 }
 272
 273 static void nf_ct_dying_timeout(struct nf_conn *ct)
 274 {
 275         struct net *net = nf_ct_net(ct);
 276         struct nf_conntrack_ecache *ecache = nf_ct_ecache_find(ct);
 277
 278         BUG_ON(ecache == NULL);
 279
 280         /* set a new timer to retry event delivery */
 281         setup_timer(&ecache->timeout, death_by_event, (unsigned long)ct);
 282         ecache->timeout.expires = jiffies +
 283                 (prandom_u32() % net->ct.sysctl_events_retry_timeout);
 284         add_timer(&ecache->timeout);
 285 }
 286
 287 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
 288 {
 289         struct nf_conn_tstamp *tstamp;
 290
 291         tstamp = nf_conn_tstamp_find(ct);
 292         if (tstamp && tstamp->stop == 0)
 293                 tstamp->stop = ktime_to_ns(ktime_get_real());
 294
 295         if (!nf_ct_is_dying(ct) &&
 296             unlikely(nf_conntrack_event_report(IPCT_DESTROY, ct,
 297             portid, report) < 0)) {
 298                 /* destroy event was not delivered */
 299                 nf_ct_delete_from_lists(ct);
 300                 nf_ct_dying_timeout(ct);
 301                 return false;
 302         }
 303         set_bit(IPS_DYING_BIT, &ct->status);
 304         nf_ct_delete_from_lists(ct);
 305         nf_ct_put(ct);
 306         return true;
 307 }
 308 EXPORT_SYMBOL_GPL(nf_ct_delete);
 309
 310 static void death_by_timeout(unsigned long ul_conntrack)
 311 {
 312         nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
 313 }
 314
 315 static inline bool
 316 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
 317                         const struct nf_conntrack_tuple *tuple,
 318                         u16 zone)
 319 {
 320         struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 321
 322         /* A conntrack can be recreated with the equal tuple,
 323          * so we need to check that the conntrack is confirmed
 324          */
 325         return nf_ct_tuple_equal(tuple, &h->tuple) &&
 326                 nf_ct_zone(ct) == zone &&
 327                 nf_ct_is_confirmed(ct);
 328 }
 329
 330 /*
 331  * Warning :
 332  * - Caller must take a reference on returned object
 333  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 334  * OR
 335  * - Caller must lock nf_conntrack_lock before calling this function
 336  */
 337 static struct nf_conntrack_tuple_hash *
 338 ____nf_conntrack_find(struct net *net, u16 zone,
 339                       const struct nf_conntrack_tuple *tuple, u32 hash)
 340 {
 341         struct nf_conntrack_tuple_hash *h;
 342         struct hlist_nulls_node *n;
 343         unsigned int bucket = hash_bucket(hash, net);
 344
 345         /* Disable BHs the entire time since we normally need to disable them
 346          * at least once for the stats anyway.
 347          */
 348         local_bh_disable();
 349 begin:
 350         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
 351                 if (nf_ct_key_equal(h, tuple, zone)) {
 352                         NF_CT_STAT_INC(net, found);
 353                         local_bh_enable();
 354                         return h;
 355                 }
 356                 NF_CT_STAT_INC(net, searched);
 357         }
 358         /*
 359          * if the nulls value we got at the end of this lookup is
 360          * not the expected one, we must restart lookup.
 361          * We probably met an item that was moved to another chain.
 362          */
 363         if (get_nulls_value(n) != bucket) {
 364                 NF_CT_STAT_INC(net, search_restart);
 365                 goto begin;
 366         }
 367         local_bh_enable();
 368
 369         return NULL;
 370 }
 371
 372 /* Find a connection corresponding to a tuple. */
 373 static struct nf_conntrack_tuple_hash *
 374 __nf_conntrack_find_get(struct net *net, u16 zone,
 375                         const struct nf_conntrack_tuple *tuple, u32 hash)
 376 {
 377         struct nf_conntrack_tuple_hash *h;
 378         struct nf_conn *ct;
 379
 380         rcu_read_lock();
 381 begin:
 382         h = ____nf_conntrack_find(net, zone, tuple, hash);
 383         if (h) {
 384                 ct = nf_ct_tuplehash_to_ctrack(h);
 385                 if (unlikely(nf_ct_is_dying(ct) ||
 386                              !atomic_inc_not_zero(&ct->ct_general.use)))
 387                         h = NULL;
 388                 else {
 389                         if (unlikely(!nf_ct_key_equal(h, tuple, zone))) {
 390                                 nf_ct_put(ct);
 391                                 goto begin;
 392                         }
 393                 }
 394         }
 395         rcu_read_unlock();
 396
 397         return h;
 398 }
 399
 400 struct nf_conntrack_tuple_hash *
 401 nf_conntrack_find_get(struct net *net, u16 zone,
 402                       const struct nf_conntrack_tuple *tuple)
 403 {
 404         return __nf_conntrack_find_get(net, zone, tuple,
 405                                        hash_conntrack_raw(tuple, zone));
 406 }
 407 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 408
 409 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 410                                        unsigned int hash,
 411                                        unsigned int repl_hash)
 412 {
 413         struct net *net = nf_ct_net(ct);
 414
 415         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 416                            &net->ct.hash[hash]);
 417         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 418                            &net->ct.hash[repl_hash]);
 419 }
 420
 421 int
 422 nf_conntrack_hash_check_insert(struct nf_conn *ct)
 423 {
 424         struct net *net = nf_ct_net(ct);
 425         unsigned int hash, repl_hash;
 426         struct nf_conntrack_tuple_hash *h;
 427         struct hlist_nulls_node *n;
 428         u16 zone;
 429
 430         zone = nf_ct_zone(ct);
 431         hash = hash_conntrack(net, zone,
 432                               &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 433         repl_hash = hash_conntrack(net, zone,
 434                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 435
 436         spin_lock_bh(&nf_conntrack_lock);
 437
 438         /* See if there's one in the list already, including reverse */
 439         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 440                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 441                                       &h->tuple) &&
 442                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 443                         goto out;
 444         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 445                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 446                                       &h->tuple) &&
 447                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 448                         goto out;
 449
 450         add_timer(&ct->timeout);
 451         smp_wmb();
 452         /* The caller holds a reference to this object */
 453         atomic_set(&ct->ct_general.use, 2);
 454         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 455         NF_CT_STAT_INC(net, insert);
 456         spin_unlock_bh(&nf_conntrack_lock);
 457
 458         return 0;
 459
 460 out:
 461         NF_CT_STAT_INC(net, insert_failed);
 462         spin_unlock_bh(&nf_conntrack_lock);
 463         return -EEXIST;
 464 }
 465 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
 466
 467 /* deletion from this larval template list happens via nf_ct_put() */
 468 void nf_conntrack_tmpl_insert(struct net *net, struct nf_conn *tmpl)
 469 {
 470         __set_bit(IPS_TEMPLATE_BIT, &tmpl->status);
 471         __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
 472         nf_conntrack_get(&tmpl->ct_general);
 473
 474         spin_lock_bh(&nf_conntrack_lock);
 475         /* Overload tuple linked list to put us in template list. */
 476         hlist_nulls_add_head_rcu(&tmpl->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 477                                  &net->ct.tmpl);
 478         spin_unlock_bh(&nf_conntrack_lock);
 479 }
 480 EXPORT_SYMBOL_GPL(nf_conntrack_tmpl_insert);
 481
 482 /* Confirm a connection given skb; places it in hash table */
 483 int
 484 __nf_conntrack_confirm(struct sk_buff *skb)
 485 {
 486         unsigned int hash, repl_hash;
 487         struct nf_conntrack_tuple_hash *h;
 488         struct nf_conn *ct;
 489         struct nf_conn_help *help;
 490         struct nf_conn_tstamp *tstamp;
 491         struct hlist_nulls_node *n;
 492         enum ip_conntrack_info ctinfo;
 493         struct net *net;
 494         u16 zone;
 495
 496         ct = nf_ct_get(skb, &ctinfo);
 497         net = nf_ct_net(ct);
 498
 499         /* ipt_REJECT uses nf_conntrack_attach to attach related
 500            ICMP/TCP RST packets in other direction.  Actual packet
 501            which created connection will be IP_CT_NEW or for an
 502            expected connection, IP_CT_RELATED. */
 503         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 504                 return NF_ACCEPT;
 505
 506         zone = nf_ct_zone(ct);
 507         /* reuse the hash saved before */
 508         hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
 509         hash = hash_bucket(hash, net);
 510         repl_hash = hash_conntrack(net, zone,
 511                                    &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 512
 513         /* We're not in hash table, and we refuse to set up related
 514            connections for unconfirmed conns.  But packet copies and
 515            REJECT will give spurious warnings here. */
 516         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 517
 518         /* No external references means no one else could have
 519            confirmed us. */
 520         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 521         pr_debug("Confirming conntrack %p\n", ct);
 522
 523         spin_lock_bh(&nf_conntrack_lock);
 524
 525         /* We have to check the DYING flag inside the lock to prevent
 526            a race against nf_ct_get_next_corpse() possibly called from
 527            user context, else we insert an already 'dead' hash, blocking
 528            further use of that particular connection -JM */
 529
 530         if (unlikely(nf_ct_is_dying(ct))) {
 531                 spin_unlock_bh(&nf_conntrack_lock);
 532                 return NF_ACCEPT;
 533         }
 534
 535         /* See if there's one in the list already, including reverse:
 536            NAT could have grabbed it without realizing, since we're
 537            not in the hash.  If there is, we lost race. */
 538         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 539                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 540                                       &h->tuple) &&
 541                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 542                         goto out;
 543         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 544                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 545                                       &h->tuple) &&
 546                     zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
 547                         goto out;
 548
 549         /* Remove from unconfirmed list */
 550         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 551
 552         /* Timer relative to confirmation time, not original
 553            setting time, otherwise we'd get timer wrap in
 554            weird delay cases. */
 555         ct->timeout.expires += jiffies;
 556         add_timer(&ct->timeout);
 557         atomic_inc(&ct->ct_general.use);
 558         ct->status |= IPS_CONFIRMED;
 559
 560         /* set conntrack timestamp, if enabled. */
 561         tstamp = nf_conn_tstamp_find(ct);
 562         if (tstamp) {
 563                 if (skb->tstamp.tv64 == 0)
 564                         __net_timestamp(skb);
 565
 566                 tstamp->start = ktime_to_ns(skb->tstamp);
 567         }
 568         /* Since the lookup is lockless, hash insertion must be done after
 569          * starting the timer and setting the CONFIRMED bit. The RCU barriers
 570          * guarantee that no other CPU can find the conntrack before the above
 571          * stores are visible.
 572          */
 573         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 574         NF_CT_STAT_INC(net, insert);
 575         spin_unlock_bh(&nf_conntrack_lock);
 576
 577         help = nfct_help(ct);
 578         if (help && help->helper)
 579                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 580
 581         nf_conntrack_event_cache(master_ct(ct) ?
 582                                  IPCT_RELATED : IPCT_NEW, ct);
 583         return NF_ACCEPT;
 584
 585 out:
 586         NF_CT_STAT_INC(net, insert_failed);
 587         spin_unlock_bh(&nf_conntrack_lock);
 588         return NF_DROP;
 589 }
 590 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 591
 592 /* Returns true if a connection correspondings to the tuple (required
 593    for NAT). */
 594 int
 595 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 596                          const struct nf_conn *ignored_conntrack)
 597 {
 598         struct net *net = nf_ct_net(ignored_conntrack);
 599         struct nf_conntrack_tuple_hash *h;
 600         struct hlist_nulls_node *n;
 601         struct nf_conn *ct;
 602         u16 zone = nf_ct_zone(ignored_conntrack);
 603         unsigned int hash = hash_conntrack(net, zone, tuple);
 604
 605         /* Disable BHs the entire time since we need to disable them at
 606          * least once for the stats anyway.
 607          */
 608         rcu_read_lock_bh();
 609         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 610                 ct = nf_ct_tuplehash_to_ctrack(h);
 611                 if (ct != ignored_conntrack &&
 612                     nf_ct_tuple_equal(tuple, &h->tuple) &&
 613                     nf_ct_zone(ct) == zone) {
 614                         NF_CT_STAT_INC(net, found);
 615                         rcu_read_unlock_bh();
 616                         return 1;
 617                 }
 618                 NF_CT_STAT_INC(net, searched);
 619         }
 620         rcu_read_unlock_bh();
 621
 622         return 0;
 623 }
 624 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 625
 626 #define NF_CT_EVICTION_RANGE    8
 627
 628 /* There's a small race here where we may free a just-assured
 629    connection.  Too bad: we're in trouble anyway. */
 630 static noinline int early_drop(struct net *net, unsigned int hash)
 631 {
 632         /* Use oldest entry, which is roughly LRU */
 633         struct nf_conntrack_tuple_hash *h;
 634         struct nf_conn *ct = NULL, *tmp;
 635         struct hlist_nulls_node *n;
 636         unsigned int i, cnt = 0;
 637         int dropped = 0;
 638
 639         rcu_read_lock();
 640         for (i = 0; i < net->ct.htable_size; i++) {
 641                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 642                                          hnnode) {
 643                         tmp = nf_ct_tuplehash_to_ctrack(h);
 644                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 645                                 ct = tmp;
 646                         cnt++;
 647                 }
 648
 649                 if (ct != NULL) {
 650                         if (likely(!nf_ct_is_dying(ct) &&
 651                                    atomic_inc_not_zero(&ct->ct_general.use)))
 652                                 break;
 653                         else
 654                                 ct = NULL;
 655                 }
 656
 657                 if (cnt >= NF_CT_EVICTION_RANGE)
 658                         break;
 659
 660                 hash = (hash + 1) % net->ct.htable_size;
 661         }
 662         rcu_read_unlock();
 663
 664         if (!ct)
 665                 return dropped;
 666
 667         if (del_timer(&ct->timeout)) {
 668                 if (nf_ct_delete(ct, 0, 0)) {
 669                         dropped = 1;
 670                         NF_CT_STAT_INC_ATOMIC(net, early_drop);
 671                 }
 672         }
 673         nf_ct_put(ct);
 674         return dropped;
 675 }
 676
 677 void init_nf_conntrack_hash_rnd(void)
 678 {
 679         unsigned int rand;
 680
 681         /*
 682          * Why not initialize nf_conntrack_rnd in a "init()" function ?
 683          * Because there isn't enough entropy when system initializing,
 684          * and we initialize it as late as possible.
 685          */
 686         do {
 687                 get_random_bytes(&rand, sizeof(rand));
 688         } while (!rand);
 689         cmpxchg(&nf_conntrack_hash_rnd, 0, rand);
 690 }
 691
 692 static struct nf_conn *
 693 __nf_conntrack_alloc(struct net *net, u16 zone,
 694                      const struct nf_conntrack_tuple *orig,
 695                      const struct nf_conntrack_tuple *repl,
 696                      gfp_t gfp, u32 hash)
 697 {
 698         struct nf_conn *ct;
 699
 700         if (unlikely(!nf_conntrack_hash_rnd)) {
 701                 init_nf_conntrack_hash_rnd();
 702                 /* recompute the hash as nf_conntrack_hash_rnd is initialized */
 703                 hash = hash_conntrack_raw(orig, zone);
 704         }
 705
 706         /* We don't want any race condition at early drop stage */
 707         atomic_inc(&net->ct.count);
 708
 709         if (nf_conntrack_max &&
 710             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 711                 if (!early_drop(net, hash_bucket(hash, net))) {
 712                         atomic_dec(&net->ct.count);
 713                         net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
 714                         return ERR_PTR(-ENOMEM);
 715                 }
 716         }
 717
 718         /*
 719          * Do not use kmem_cache_zalloc(), as this cache uses
 720          * SLAB_DESTROY_BY_RCU.
 721          */
 722         ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp);
 723         if (ct == NULL) {
 724                 atomic_dec(&net->ct.count);
 725                 return ERR_PTR(-ENOMEM);
 726         }
 727         /*
 728          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
 729          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
 730          */
 731         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
 732                offsetof(struct nf_conn, proto) -
 733                offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
 734         spin_lock_init(&ct->lock);
 735         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 736         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
 737         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 738         /* save hash for reusing when confirming */
 739         *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
 740         /* Don't set timer yet: wait for confirmation */
 741         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 742         write_pnet(&ct->ct_net, net);
 743 #ifdef CONFIG_NF_CONNTRACK_ZONES
 744         if (zone) {
 745                 struct nf_conntrack_zone *nf_ct_zone;
 746
 747                 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
 748                 if (!nf_ct_zone)
 749                         goto out_free;
 750                 nf_ct_zone->id = zone;
 751         }
 752 #endif
 753         /* Because we use RCU lookups, we set ct_general.use to zero before
 754          * this is inserted in any list.
 755          */
 756         atomic_set(&ct->ct_general.use, 0);
 757         return ct;
 758
 759 #ifdef CONFIG_NF_CONNTRACK_ZONES
 760 out_free:
 761         atomic_dec(&net->ct.count);
 762         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 763         return ERR_PTR(-ENOMEM);
 764 #endif
 765 }
 766
 767 struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
 768                                    const struct nf_conntrack_tuple *orig,
 769                                    const struct nf_conntrack_tuple *repl,
 770                                    gfp_t gfp)
 771 {
 772         return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
 773 }
 774 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 775
 776 void nf_conntrack_free(struct nf_conn *ct)
 777 {
 778         struct net *net = nf_ct_net(ct);
 779
 780         /* A freed object has refcnt == 0, that's
 781          * the golden rule for SLAB_DESTROY_BY_RCU
 782          */
 783         NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
 784
 785         nf_ct_ext_destroy(ct);
 786         nf_ct_ext_free(ct);
 787         kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
 788         smp_mb__before_atomic_dec();
 789         atomic_dec(&net->ct.count);
 790 }
 791 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 792
 793
 794 /* Allocate a new conntrack: we return -ENOMEM if classification
 795    failed due to stress.  Otherwise it really is unclassifiable. */
 796 static struct nf_conntrack_tuple_hash *
 797 init_conntrack(struct net *net, struct nf_conn *tmpl,
 798                const struct nf_conntrack_tuple *tuple,
 799                struct nf_conntrack_l3proto *l3proto,
 800                struct nf_conntrack_l4proto *l4proto,
 801                struct sk_buff *skb,
 802                unsigned int dataoff, u32 hash)
 803 {
 804         struct nf_conn *ct;
 805         struct nf_conn_help *help;
 806         struct nf_conntrack_tuple repl_tuple;
 807         struct nf_conntrack_ecache *ecache;
 808         struct nf_conntrack_expect *exp;
 809         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 810         struct nf_conn_timeout *timeout_ext;
 811         unsigned int *timeouts;
 812
 813         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 814                 pr_debug("Can't invert tuple.\n");
 815                 return NULL;
 816         }
 817
 818         ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
 819                                   hash);
 820         if (IS_ERR(ct))
 821                 return (struct nf_conntrack_tuple_hash *)ct;
 822
 823         if (tmpl && nfct_synproxy(tmpl)) {
 824                 nfct_seqadj_ext_add(ct);
 825                 nfct_synproxy_ext_add(ct);
 826         }
 827
 828         timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
 829         if (timeout_ext)
 830                 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext);
 831         else
 832                 timeouts = l4proto->get_timeouts(net);
 833
 834         if (!l4proto->new(ct, skb, dataoff, timeouts)) {
 835                 nf_conntrack_free(ct);
 836                 pr_debug("init conntrack: can't track with proto module\n");
 837                 return NULL;
 838         }
 839
 840         if (timeout_ext)
 841                 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC);
 842
 843         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 844         nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
 845         nf_ct_labels_ext_add(ct);
 846
 847         ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
 848         nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
 849                                  ecache ? ecache->expmask : 0,
 850                              GFP_ATOMIC);
 851
 852         spin_lock_bh(&nf_conntrack_lock);
 853         exp = nf_ct_find_expectation(net, zone, tuple);
 854         if (exp) {
 855                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 856                          ct, exp);
 857                 /* Welcome, Mr. Bond.  We've been expecting you... */
 858                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 859                 ct->master = exp->master;
 860                 if (exp->helper) {
 861                         help = nf_ct_helper_ext_add(ct, exp->helper,
 862                                                     GFP_ATOMIC);
 863                         if (help)
 864                                 rcu_assign_pointer(help->helper, exp->helper);
 865                 }
 866
 867 #ifdef CONFIG_NF_CONNTRACK_MARK
 868                 ct->mark = exp->master->mark;
 869 #endif
 870 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 871                 ct->secmark = exp->master->secmark;
 872 #endif
 873                 nf_conntrack_get(&ct->master->ct_general);
 874                 NF_CT_STAT_INC(net, expect_new);
 875         } else {
 876                 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
 877                 NF_CT_STAT_INC(net, new);
 878         }
 879
 880         /* Now it is inserted into the unconfirmed list, bump refcount */
 881         nf_conntrack_get(&ct->ct_general);
 882
 883         /* Overload tuple linked list to put us in unconfirmed list. */
 884         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 885                        &net->ct.unconfirmed);
 886
 887         spin_unlock_bh(&nf_conntrack_lock);
 888
 889         if (exp) {
 890                 if (exp->expectfn)
 891                         exp->expectfn(ct, exp);
 892                 nf_ct_expect_put(exp);
 893         }
 894
 895         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 896 }
 897
 898 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 899 static inline struct nf_conn *
 900 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
 901                   struct sk_buff *skb,
 902                   unsigned int dataoff,
 903                   u_int16_t l3num,
 904                   u_int8_t protonum,
 905                   struct nf_conntrack_l3proto *l3proto,
 906                   struct nf_conntrack_l4proto *l4proto,
 907                   int *set_reply,
 908                   enum ip_conntrack_info *ctinfo)
 909 {
 910         struct nf_conntrack_tuple tuple;
 911         struct nf_conntrack_tuple_hash *h;
 912         struct nf_conn *ct;
 913         u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 914         u32 hash;
 915
 916         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 917                              dataoff, l3num, protonum, &tuple, l3proto,
 918                              l4proto)) {
 919                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 920                 return NULL;
 921         }
 922
 923         /* look for tuple match */
 924         hash = hash_conntrack_raw(&tuple, zone);
 925         h = __nf_conntrack_find_get(net, zone, &tuple, hash);
 926         if (!h) {
 927                 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
 928                                    skb, dataoff, hash);
 929                 if (!h)
 930                         return NULL;
 931                 if (IS_ERR(h))
 932                         return (void *)h;
 933         }
 934         ct = nf_ct_tuplehash_to_ctrack(h);
 935
 936         /* It exists; we have (non-exclusive) reference. */
 937         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 938                 *ctinfo = IP_CT_ESTABLISHED_REPLY;
 939                 /* Please set reply bit if this packet OK */
 940                 *set_reply = 1;
 941         } else {
 942                 /* Once we've had two way comms, always ESTABLISHED. */
 943                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 944                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 945                         *ctinfo = IP_CT_ESTABLISHED;
 946                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 947                         pr_debug("nf_conntrack_in: related packet for %p\n",
 948                                  ct);
 949                         *ctinfo = IP_CT_RELATED;
 950                 } else {
 951                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 952                         *ctinfo = IP_CT_NEW;
 953                 }
 954                 *set_reply = 0;
 955         }
 956         skb->nfct = &ct->ct_general;
 957         skb->nfctinfo = *ctinfo;
 958         return ct;
 959 }
 960
 961 unsigned int
 962 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 963                 struct sk_buff *skb)
 964 {
 965         struct nf_conn *ct, *tmpl = NULL;
 966         enum ip_conntrack_info ctinfo;
 967         struct nf_conntrack_l3proto *l3proto;
 968         struct nf_conntrack_l4proto *l4proto;
 969         unsigned int *timeouts;
 970         unsigned int dataoff;
 971         u_int8_t protonum;
 972         int set_reply = 0;
 973         int ret;
 974
 975         if (skb->nfct) {
 976                 /* Previously seen (loopback or untracked)?  Ignore. */
 977                 tmpl = (struct nf_conn *)skb->nfct;
 978                 if (!nf_ct_is_template(tmpl)) {
 979                         NF_CT_STAT_INC_ATOMIC(net, ignore);
 980                         return NF_ACCEPT;
 981                 }
 982                 skb->nfct = NULL;
 983         }
 984
 985         /* rcu_read_lock()ed by nf_hook_slow */
 986         l3proto = __nf_ct_l3proto_find(pf);
 987         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 988                                    &dataoff, &protonum);
 989         if (ret <= 0) {
 990                 pr_debug("not prepared to track yet or error occurred\n");
 991                 NF_CT_STAT_INC_ATOMIC(net, error);
 992                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 993                 ret = -ret;
 994                 goto out;
 995         }
 996
 997         l4proto = __nf_ct_l4proto_find(pf, protonum);
 998
 999         /* It may be an special packet, error, unclean...
1000          * inverse of the return code tells to the netfilter
1001          * core what to do with the packet. */
1002         if (l4proto->error != NULL) {
1003                 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1004                                      pf, hooknum);
1005                 if (ret <= 0) {
1006                         NF_CT_STAT_INC_ATOMIC(net, error);
1007                         NF_CT_STAT_INC_ATOMIC(net, invalid);
1008                         ret = -ret;
1009                         goto out;
1010                 }
1011                 /* ICMP[v6] protocol trackers may assign one conntrack. */
1012                 if (skb->nfct)
1013                         goto out;
1014         }
1015
1016         ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1017                                l3proto, l4proto, &set_reply, &ctinfo);
1018         if (!ct) {
1019                 /* Not valid part of a connection */
1020                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1021                 ret = NF_ACCEPT;
1022                 goto out;
1023         }
1024
1025         if (IS_ERR(ct)) {
1026                 /* Too stressed to deal. */
1027                 NF_CT_STAT_INC_ATOMIC(net, drop);
1028                 ret = NF_DROP;
1029                 goto out;
1030         }
1031
1032         NF_CT_ASSERT(skb->nfct);
1033
1034         /* Decide what timeout policy we want to apply to this flow. */
1035         timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1036
1037         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1038         if (ret <= 0) {
1039                 /* Invalid: inverse of the return code tells
1040                  * the netfilter core what to do */
1041                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1042                 nf_conntrack_put(skb->nfct);
1043                 skb->nfct = NULL;
1044                 NF_CT_STAT_INC_ATOMIC(net, invalid);
1045                 if (ret == -NF_DROP)
1046                         NF_CT_STAT_INC_ATOMIC(net, drop);
1047                 ret = -ret;
1048                 goto out;
1049         }
1050
1051         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1052                 nf_conntrack_event_cache(IPCT_REPLY, ct);
1053 out:
1054         if (tmpl) {
1055                 /* Special case: we have to repeat this hook, assign the
1056                  * template again to this packet. We assume that this packet
1057                  * has no conntrack assigned. This is used by nf_ct_tcp. */
1058                 if (ret == NF_REPEAT)
1059                         skb->nfct = (struct nf_conntrack *)tmpl;
1060                 else
1061                         nf_ct_put(tmpl);
1062         }
1063
1064         return ret;
1065 }
1066 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1067
1068 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1069                           const struct nf_conntrack_tuple *orig)
1070 {
1071         bool ret;
1072
1073         rcu_read_lock();
1074         ret = nf_ct_invert_tuple(inverse, orig,
1075                                  __nf_ct_l3proto_find(orig->src.l3num),
1076                                  __nf_ct_l4proto_find(orig->src.l3num,
1077                                                       orig->dst.protonum));
1078         rcu_read_unlock();
1079         return ret;
1080 }
1081 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1082
1083 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
1084    implicitly racy: see __nf_conntrack_confirm */
1085 void nf_conntrack_alter_reply(struct nf_conn *ct,
1086                               const struct nf_conntrack_tuple *newreply)
1087 {
1088         struct nf_conn_help *help = nfct_help(ct);
1089
1090         /* Should be unconfirmed, so not in hash table yet */
1091         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1092
1093         pr_debug("Altering reply tuple of %p to ", ct);
1094         nf_ct_dump_tuple(newreply);
1095
1096         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1097         if (ct->master || (help && !hlist_empty(&help->expectations)))
1098                 return;
1099
1100         rcu_read_lock();
1101         __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1102         rcu_read_unlock();
1103 }
1104 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1105
1106 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1107 void __nf_ct_refresh_acct(struct nf_conn *ct,
1108                           enum ip_conntrack_info ctinfo,
1109                           const struct sk_buff *skb,
1110                           unsigned long extra_jiffies,
1111                           int do_acct)
1112 {
1113         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1114         NF_CT_ASSERT(skb);
1115
1116         /* Only update if this is not a fixed timeout */
1117         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1118                 goto acct;
1119
1120         /* If not in hash table, timer will not be active yet */
1121         if (!nf_ct_is_confirmed(ct)) {
1122                 ct->timeout.expires = extra_jiffies;
1123         } else {
1124                 unsigned long newtime = jiffies + extra_jiffies;
1125
1126                 /* Only update the timeout if the new timeout is at least
1127                    HZ jiffies from the old timeout. Need del_timer for race
1128                    avoidance (may already be dying). */
1129                 if (newtime - ct->timeout.expires >= HZ)
1130                         mod_timer_pending(&ct->timeout, newtime);
1131         }
1132
1133 acct:
1134         if (do_acct) {
1135                 struct nf_conn_acct *acct;
1136
1137                 acct = nf_conn_acct_find(ct);
1138                 if (acct) {
1139                         struct nf_conn_counter *counter = acct->counter;
1140
1141                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1142                         atomic64_add(skb->len, &counter[CTINFO2DIR(ctinfo)].bytes);
1143                 }
1144         }
1145 }
1146 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1147
1148 bool __nf_ct_kill_acct(struct nf_conn *ct,
1149                        enum ip_conntrack_info ctinfo,
1150                        const struct sk_buff *skb,
1151                        int do_acct)
1152 {
1153         if (do_acct) {
1154                 struct nf_conn_acct *acct;
1155
1156                 acct = nf_conn_acct_find(ct);
1157                 if (acct) {
1158                         struct nf_conn_counter *counter = acct->counter;
1159
1160                         atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
1161                         atomic64_add(skb->len - skb_network_offset(skb),
1162                                      &counter[CTINFO2DIR(ctinfo)].bytes);
1163                 }
1164         }
1165
1166         if (del_timer(&ct->timeout)) {
1167                 ct->timeout.function((unsigned long)ct);
1168                 return true;
1169         }
1170         return false;
1171 }
1172 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1173
1174 #ifdef CONFIG_NF_CONNTRACK_ZONES
1175 static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
1176         .len    = sizeof(struct nf_conntrack_zone),
1177         .align  = __alignof__(struct nf_conntrack_zone),
1178         .id     = NF_CT_EXT_ZONE,
1179 };
1180 #endif
1181
1182 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1183
1184 #include <linux/netfilter/nfnetlink.h>
1185 #include <linux/netfilter/nfnetlink_conntrack.h>
1186 #include <linux/mutex.h>
1187
1188 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1189  * in ip_conntrack_core, since we don't want the protocols to autoload
1190  * or depend on ctnetlink */
1191 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1192                                const struct nf_conntrack_tuple *tuple)
1193 {
1194         if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1195             nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1196                 goto nla_put_failure;
1197         return 0;
1198
1199 nla_put_failure:
1200         return -1;
1201 }
1202 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1203
1204 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1205         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
1206         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
1207 };
1208 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1209
1210 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1211                                struct nf_conntrack_tuple *t)
1212 {
1213         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1214                 return -EINVAL;
1215
1216         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1217         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1218
1219         return 0;
1220 }
1221 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1222
1223 int nf_ct_port_nlattr_tuple_size(void)
1224 {
1225         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1226 }
1227 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1228 #endif
1229
1230 /* Used by ipt_REJECT and ip6t_REJECT. */
1231 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1232 {
1233         struct nf_conn *ct;
1234         enum ip_conntrack_info ctinfo;
1235
1236         /* This ICMP is in reverse direction to the packet which caused it */
1237         ct = nf_ct_get(skb, &ctinfo);
1238         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1239                 ctinfo = IP_CT_RELATED_REPLY;
1240         else
1241                 ctinfo = IP_CT_RELATED;
1242
1243         /* Attach to new skbuff, and increment count */
1244         nskb->nfct = &ct->ct_general;
1245         nskb->nfctinfo = ctinfo;
1246         nf_conntrack_get(nskb->nfct);
1247 }
1248
1249 /* Bring out ya dead! */
1250 static struct nf_conn *
1251 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1252                 void *data, unsigned int *bucket)
1253 {
1254         struct nf_conntrack_tuple_hash *h;
1255         struct nf_conn *ct;
1256         struct hlist_nulls_node *n;
1257
1258         spin_lock_bh(&nf_conntrack_lock);
1259         for (; *bucket < net->ct.htable_size; (*bucket)++) {
1260                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1261                         if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1262                                 continue;
1263                         ct = nf_ct_tuplehash_to_ctrack(h);
1264                         if (iter(ct, data))
1265                                 goto found;
1266                 }
1267         }
1268         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1269                 ct = nf_ct_tuplehash_to_ctrack(h);
1270                 if (iter(ct, data))
1271                         set_bit(IPS_DYING_BIT, &ct->status);
1272         }
1273         spin_unlock_bh(&nf_conntrack_lock);
1274         return NULL;
1275 found:
1276         atomic_inc(&ct->ct_general.use);
1277         spin_unlock_bh(&nf_conntrack_lock);
1278         return ct;
1279 }
1280
1281 void nf_ct_iterate_cleanup(struct net *net,
1282                            int (*iter)(struct nf_conn *i, void *data),
1283                            void *data, u32 portid, int report)
1284 {
1285         struct nf_conn *ct;
1286         unsigned int bucket = 0;
1287
1288         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1289                 /* Time to push up daises... */
1290                 if (del_timer(&ct->timeout))
1291                         nf_ct_delete(ct, portid, report);
1292
1293                 /* ... else the timer will get him soon. */
1294
1295                 nf_ct_put(ct);
1296         }
1297 }
1298 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1299
1300 static int kill_all(struct nf_conn *i, void *data)
1301 {
1302         return 1;
1303 }
1304
1305 void nf_ct_free_hashtable(void *hash, unsigned int size)
1306 {
1307         if (is_vmalloc_addr(hash))
1308                 vfree(hash);
1309         else
1310                 free_pages((unsigned long)hash,
1311                            get_order(sizeof(struct hlist_head) * size));
1312 }
1313 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1314
1315 void nf_conntrack_flush_report(struct net *net, u32 portid, int report)
1316 {
1317         nf_ct_iterate_cleanup(net, kill_all, NULL, portid, report);
1318 }
1319 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1320
1321 static void nf_ct_release_dying_list(struct net *net)
1322 {
1323         struct nf_conntrack_tuple_hash *h;
1324         struct nf_conn *ct;
1325         struct hlist_nulls_node *n;
1326
1327         spin_lock_bh(&nf_conntrack_lock);
1328         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1329                 ct = nf_ct_tuplehash_to_ctrack(h);
1330                 /* never fails to remove them, no listeners at this point */
1331                 nf_ct_kill(ct);
1332         }
1333         spin_unlock_bh(&nf_conntrack_lock);
1334 }
1335
1336 static int untrack_refs(void)
1337 {
1338         int cnt = 0, cpu;
1339
1340         for_each_possible_cpu(cpu) {
1341                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1342
1343                 cnt += atomic_read(&ct->ct_general.use) - 1;
1344         }
1345         return cnt;
1346 }
1347
1348 void nf_conntrack_cleanup_start(void)
1349 {
1350         RCU_INIT_POINTER(ip_ct_attach, NULL);
1351 }
1352
1353 void nf_conntrack_cleanup_end(void)
1354 {
1355         RCU_INIT_POINTER(nf_ct_destroy, NULL);
1356         while (untrack_refs() > 0)
1357                 schedule();
1358
1359 #ifdef CONFIG_NF_CONNTRACK_ZONES
1360         nf_ct_extend_unregister(&nf_ct_zone_extend);
1361 #endif
1362         nf_conntrack_proto_fini();
1363         nf_conntrack_seqadj_fini();
1364         nf_conntrack_labels_fini();
1365         nf_conntrack_helper_fini();
1366         nf_conntrack_timeout_fini();
1367         nf_conntrack_ecache_fini();
1368         nf_conntrack_tstamp_fini();
1369         nf_conntrack_acct_fini();
1370         nf_conntrack_expect_fini();
1371 }
1372
1373 /*
1374  * Mishearing the voices in his head, our hero wonders how he's
1375  * supposed to kill the mall.
1376  */
1377 void nf_conntrack_cleanup_net(struct net *net)
1378 {
1379         LIST_HEAD(single);
1380
1381         list_add(&net->exit_list, &single);
1382         nf_conntrack_cleanup_net_list(&single);
1383 }
1384
1385 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1386 {
1387         int busy;
1388         struct net *net;
1389
1390         /*
1391          * This makes sure all current packets have passed through
1392          *  netfilter framework.  Roll on, two-stage module
1393          *  delete...
1394          */
1395         synchronize_net();
1396 i_see_dead_people:
1397         busy = 0;
1398         list_for_each_entry(net, net_exit_list, exit_list) {
1399                 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1400                 nf_ct_release_dying_list(net);
1401                 if (atomic_read(&net->ct.count) != 0)
1402                         busy = 1;
1403         }
1404         if (busy) {
1405                 schedule();
1406                 goto i_see_dead_people;
1407         }
1408
1409         list_for_each_entry(net, net_exit_list, exit_list) {
1410                 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1411                 nf_conntrack_proto_pernet_fini(net);
1412                 nf_conntrack_helper_pernet_fini(net);
1413                 nf_conntrack_ecache_pernet_fini(net);
1414                 nf_conntrack_tstamp_pernet_fini(net);
1415                 nf_conntrack_acct_pernet_fini(net);
1416                 nf_conntrack_expect_pernet_fini(net);
1417                 kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1418                 kfree(net->ct.slabname);
1419                 free_percpu(net->ct.stat);
1420         }
1421 }
1422
1423 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1424 {
1425         struct hlist_nulls_head *hash;
1426         unsigned int nr_slots, i;
1427         size_t sz;
1428
1429         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1430         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1431         sz = nr_slots * sizeof(struct hlist_nulls_head);
1432         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1433                                         get_order(sz));
1434         if (!hash) {
1435                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1436                 hash = vzalloc(sz);
1437         }
1438
1439         if (hash && nulls)
1440                 for (i = 0; i < nr_slots; i++)
1441                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1442
1443         return hash;
1444 }
1445 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1446
1447 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1448 {
1449         int i, bucket, rc;
1450         unsigned int hashsize, old_size;
1451         struct hlist_nulls_head *hash, *old_hash;
1452         struct nf_conntrack_tuple_hash *h;
1453         struct nf_conn *ct;
1454
1455         if (current->nsproxy->net_ns != &init_net)
1456                 return -EOPNOTSUPP;
1457
1458         /* On boot, we can set this without any fancy locking. */
1459         if (!nf_conntrack_htable_size)
1460                 return param_set_uint(val, kp);
1461
1462         rc = kstrtouint(val, 0, &hashsize);
1463         if (rc)
1464                 return rc;
1465         if (!hashsize)
1466                 return -EINVAL;
1467
1468         hash = nf_ct_alloc_hashtable(&hashsize, 1);
1469         if (!hash)
1470                 return -ENOMEM;
1471
1472         /* Lookups in the old hash might happen in parallel, which means we
1473          * might get false negatives during connection lookup. New connections
1474          * created because of a false negative won't make it into the hash
1475          * though since that required taking the lock.
1476          */
1477         spin_lock_bh(&nf_conntrack_lock);
1478         for (i = 0; i < init_net.ct.htable_size; i++) {
1479                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1480                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1481                                         struct nf_conntrack_tuple_hash, hnnode);
1482                         ct = nf_ct_tuplehash_to_ctrack(h);
1483                         hlist_nulls_del_rcu(&h->hnnode);
1484                         bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
1485                                                   hashsize);
1486                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1487                 }
1488         }
1489         old_size = init_net.ct.htable_size;
1490         old_hash = init_net.ct.hash;
1491
1492         init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
1493         init_net.ct.hash = hash;
1494         spin_unlock_bh(&nf_conntrack_lock);
1495
1496         nf_ct_free_hashtable(old_hash, old_size);
1497         return 0;
1498 }
1499 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1500
1501 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1502                   &nf_conntrack_htable_size, 0600);
1503
1504 void nf_ct_untracked_status_or(unsigned long bits)
1505 {
1506         int cpu;
1507
1508         for_each_possible_cpu(cpu)
1509                 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1510 }
1511 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1512
1513 int nf_conntrack_init_start(void)
1514 {
1515         int max_factor = 8;
1516         int ret, cpu;
1517
1518         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1519          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1520         if (!nf_conntrack_htable_size) {
1521                 nf_conntrack_htable_size
1522                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1523                            / sizeof(struct hlist_head));
1524                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1525                         nf_conntrack_htable_size = 16384;
1526                 if (nf_conntrack_htable_size < 32)
1527                         nf_conntrack_htable_size = 32;
1528
1529                 /* Use a max. factor of four by default to get the same max as
1530                  * with the old struct list_heads. When a table size is given
1531                  * we use the old value of 8 to avoid reducing the max.
1532                  * entries. */
1533                 max_factor = 4;
1534         }
1535         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1536
1537         printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1538                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1539                nf_conntrack_max);
1540
1541         ret = nf_conntrack_expect_init();
1542         if (ret < 0)
1543                 goto err_expect;
1544
1545         ret = nf_conntrack_acct_init();
1546         if (ret < 0)
1547                 goto err_acct;
1548
1549         ret = nf_conntrack_tstamp_init();
1550         if (ret < 0)
1551                 goto err_tstamp;
1552
1553         ret = nf_conntrack_ecache_init();
1554         if (ret < 0)
1555                 goto err_ecache;
1556
1557         ret = nf_conntrack_timeout_init();
1558         if (ret < 0)
1559                 goto err_timeout;
1560
1561         ret = nf_conntrack_helper_init();
1562         if (ret < 0)
1563                 goto err_helper;
1564
1565         ret = nf_conntrack_labels_init();
1566         if (ret < 0)
1567                 goto err_labels;
1568
1569         ret = nf_conntrack_seqadj_init();
1570         if (ret < 0)
1571                 goto err_seqadj;
1572
1573 #ifdef CONFIG_NF_CONNTRACK_ZONES
1574         ret = nf_ct_extend_register(&nf_ct_zone_extend);
1575         if (ret < 0)
1576                 goto err_extend;
1577 #endif
1578         ret = nf_conntrack_proto_init();
1579         if (ret < 0)
1580                 goto err_proto;
1581
1582         /* Set up fake conntrack: to never be deleted, not in any hashes */
1583         for_each_possible_cpu(cpu) {
1584                 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1585                 write_pnet(&ct->ct_net, &init_net);
1586                 atomic_set(&ct->ct_general.use, 1);
1587         }
1588         /*  - and look it like as a confirmed connection */
1589         nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1590         return 0;
1591
1592 err_proto:
1593 #ifdef CONFIG_NF_CONNTRACK_ZONES
1594         nf_ct_extend_unregister(&nf_ct_zone_extend);
1595 err_extend:
1596 #endif
1597         nf_conntrack_seqadj_fini();
1598 err_seqadj:
1599         nf_conntrack_labels_fini();
1600 err_labels:
1601         nf_conntrack_helper_fini();
1602 err_helper:
1603         nf_conntrack_timeout_fini();
1604 err_timeout:
1605         nf_conntrack_ecache_fini();
1606 err_ecache:
1607         nf_conntrack_tstamp_fini();
1608 err_tstamp:
1609         nf_conntrack_acct_fini();
1610 err_acct:
1611         nf_conntrack_expect_fini();
1612 err_expect:
1613         return ret;
1614 }
1615
1616 void nf_conntrack_init_end(void)
1617 {
1618         /* For use by REJECT target */
1619         RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1620         RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1621 }
1622
1623 /*
1624  * We need to use special "null" values, not used in hash table
1625  */
1626 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1627 #define DYING_NULLS_VAL         ((1<<30)+1)
1628 #define TEMPLATE_NULLS_VAL      ((1<<30)+2)
1629
1630 int nf_conntrack_init_net(struct net *net)
1631 {
1632         int ret;
1633
1634         atomic_set(&net->ct.count, 0);
1635         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1636         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1637         INIT_HLIST_NULLS_HEAD(&net->ct.tmpl, TEMPLATE_NULLS_VAL);
1638         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1639         if (!net->ct.stat) {
1640                 ret = -ENOMEM;
1641                 goto err_stat;
1642         }
1643
1644         net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net);
1645         if (!net->ct.slabname) {
1646                 ret = -ENOMEM;
1647                 goto err_slabname;
1648         }
1649
1650         net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname,
1651                                                         sizeof(struct nf_conn), 0,
1652                                                         SLAB_DESTROY_BY_RCU, NULL);
1653         if (!net->ct.nf_conntrack_cachep) {
1654                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1655                 ret = -ENOMEM;
1656                 goto err_cache;
1657         }
1658
1659         net->ct.htable_size = nf_conntrack_htable_size;
1660         net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
1661         if (!net->ct.hash) {
1662                 ret = -ENOMEM;
1663                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1664                 goto err_hash;
1665         }
1666         ret = nf_conntrack_expect_pernet_init(net);
1667         if (ret < 0)
1668                 goto err_expect;
1669         ret = nf_conntrack_acct_pernet_init(net);
1670         if (ret < 0)
1671                 goto err_acct;
1672         ret = nf_conntrack_tstamp_pernet_init(net);
1673         if (ret < 0)
1674                 goto err_tstamp;
1675         ret = nf_conntrack_ecache_pernet_init(net);
1676         if (ret < 0)
1677                 goto err_ecache;
1678         ret = nf_conntrack_helper_pernet_init(net);
1679         if (ret < 0)
1680                 goto err_helper;
1681         ret = nf_conntrack_proto_pernet_init(net);
1682         if (ret < 0)
1683                 goto err_proto;
1684         return 0;
1685
1686 err_proto:
1687         nf_conntrack_helper_pernet_fini(net);
1688 err_helper:
1689         nf_conntrack_ecache_pernet_fini(net);
1690 err_ecache:
1691         nf_conntrack_tstamp_pernet_fini(net);
1692 err_tstamp:
1693         nf_conntrack_acct_pernet_fini(net);
1694 err_acct:
1695         nf_conntrack_expect_pernet_fini(net);
1696 err_expect:
1697         nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
1698 err_hash:
1699         kmem_cache_destroy(net->ct.nf_conntrack_cachep);
1700 err_cache:
1701         kfree(net->ct.slabname);
1702 err_slabname:
1703         free_percpu(net->ct.stat);
1704 err_stat:
1705         return ret;
1706 }