release/src-rt-6.x.4708/linux/linux-2.6.36/net/ipv4/inet_hashtables.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Generic INET transport hashtables
   7  *
   8  * Authors:     Lotsa people, from code originally in tcp
   9  *
  10  *      This program is free software; you can redistribute it and/or
  11  *      modify it under the terms of the GNU General Public License
  12  *      as published by the Free Software Foundation; either version
  13  *      2 of the License, or (at your option) any later version.
  14  */
  15
  16 #include <linux/module.h>
  17 #include <linux/random.h>
  18 #include <linux/sched.h>
  19 #include <linux/slab.h>
  20 #include <linux/wait.h>
  21
  22 #include <net/inet_connection_sock.h>
  23 #include <net/inet_hashtables.h>
  24 #include <net/ip.h>
  25
  26 #include <typedefs.h>
  27 #include <bcmdefs.h>
  28
  29 /*
  30  * Allocate and initialize a new local port bind bucket.
  31  * The bindhash mutex for snum's hash chain must be held here.
  32  */
  33 struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
  34                                                  struct net *net,
  35                                                  struct inet_bind_hashbucket *head,
  36                                                  const unsigned short snum)
  37 {
  38         struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
  39
  40         if (tb != NULL) {
  41                 write_pnet(&tb->ib_net, hold_net(net));
  42                 tb->port      = snum;
  43                 tb->fastreuse = 0;
  44                 tb->num_owners = 0;
  45                 INIT_HLIST_HEAD(&tb->owners);
  46                 hlist_add_head(&tb->node, &head->chain);
  47         }
  48         return tb;
  49 }
  50
  51 /*
  52  * Caller must hold hashbucket lock for this tb with local BH disabled
  53  */
  54 void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
  55 {
  56         if (hlist_empty(&tb->owners)) {
  57                 __hlist_del(&tb->node);
  58                 release_net(ib_net(tb));
  59                 kmem_cache_free(cachep, tb);
  60         }
  61 }
  62
  63 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
  64                     const unsigned short snum)
  65 {
  66         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
  67
  68         atomic_inc(&hashinfo->bsockets);
  69
  70         inet_sk(sk)->inet_num = snum;
  71         sk_add_bind_node(sk, &tb->owners);
  72         tb->num_owners++;
  73         inet_csk(sk)->icsk_bind_hash = tb;
  74 }
  75
  76 /*
  77  * Get rid of any references to a local port held by the given sock.
  78  */
  79 static void __inet_put_port(struct sock *sk)
  80 {
  81         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
  82         const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num,
  83                         hashinfo->bhash_size);
  84         struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
  85         struct inet_bind_bucket *tb;
  86
  87         atomic_dec(&hashinfo->bsockets);
  88
  89         spin_lock(&head->lock);
  90         tb = inet_csk(sk)->icsk_bind_hash;
  91         __sk_del_bind_node(sk);
  92         tb->num_owners--;
  93         inet_csk(sk)->icsk_bind_hash = NULL;
  94         inet_sk(sk)->inet_num = 0;
  95         inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
  96         spin_unlock(&head->lock);
  97 }
  98
  99 void inet_put_port(struct sock *sk)
 100 {
 101         local_bh_disable();
 102         __inet_put_port(sk);
 103         local_bh_enable();
 104 }
 105 EXPORT_SYMBOL(inet_put_port);
 106
 107 void __inet_inherit_port(struct sock *sk, struct sock *child)
 108 {
 109         struct inet_hashinfo *table = sk->sk_prot->h.hashinfo;
 110         const int bhash = inet_bhashfn(sock_net(sk), inet_sk(child)->inet_num,
 111                         table->bhash_size);
 112         struct inet_bind_hashbucket *head = &table->bhash[bhash];
 113         struct inet_bind_bucket *tb;
 114
 115         spin_lock(&head->lock);
 116         tb = inet_csk(sk)->icsk_bind_hash;
 117         sk_add_bind_node(child, &tb->owners);
 118         inet_csk(child)->icsk_bind_hash = tb;
 119         spin_unlock(&head->lock);
 120 }
 121 EXPORT_SYMBOL_GPL(__inet_inherit_port);
 122
 123 static inline int compute_score(struct sock *sk, struct net *net,
 124                                 const unsigned short hnum, const __be32 daddr,
 125                                 const int dif)
 126 {
 127         int score = -1;
 128         struct inet_sock *inet = inet_sk(sk);
 129
 130         if (net_eq(sock_net(sk), net) && inet->inet_num == hnum &&
 131                         !ipv6_only_sock(sk)) {
 132                 __be32 rcv_saddr = inet->inet_rcv_saddr;
 133                 score = sk->sk_family == PF_INET ? 1 : 0;
 134                 if (rcv_saddr) {
 135                         if (rcv_saddr != daddr)
 136                                 return -1;
 137                         score += 2;
 138                 }
 139                 if (sk->sk_bound_dev_if) {
 140                         if (sk->sk_bound_dev_if != dif)
 141                                 return -1;
 142                         score += 2;
 143                 }
 144         }
 145         return score;
 146 }
 147
 148 /*
 149  * Don't inline this cruft. Here are some nice properties to exploit here. The
 150  * BSD API does not allow a listening sock to specify the remote port nor the
 151  * remote address for the connection. So always assume those are both
 152  * wildcarded during the search since they can never be otherwise.
 153  */
 154
 155
 156 struct sock *__inet_lookup_listener(struct net *net,
 157                                     struct inet_hashinfo *hashinfo,
 158                                     const __be32 daddr, const unsigned short hnum,
 159                                     const int dif)
 160 {
 161         struct sock *sk, *result;
 162         struct hlist_nulls_node *node;
 163         unsigned int hash = inet_lhashfn(net, hnum);
 164         struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
 165         int score, hiscore;
 166
 167         rcu_read_lock();
 168 begin:
 169         result = NULL;
 170         hiscore = -1;
 171         sk_nulls_for_each_rcu(sk, node, &ilb->head) {
 172                 score = compute_score(sk, net, hnum, daddr, dif);
 173                 if (score > hiscore) {
 174                         result = sk;
 175                         hiscore = score;
 176                 }
 177         }
 178         /*
 179          * if the nulls value we got at the end of this lookup is
 180          * not the expected one, we must restart lookup.
 181          * We probably met an item that was moved to another chain.
 182          */
 183         if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
 184                 goto begin;
 185         if (result) {
 186                 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
 187                         result = NULL;
 188                 else if (unlikely(compute_score(result, net, hnum, daddr,
 189                                   dif) < hiscore)) {
 190                         sock_put(result);
 191                         goto begin;
 192                 }
 193         }
 194         rcu_read_unlock();
 195         return result;
 196 }
 197 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
 198
 199 struct sock * BCMFASTPATH_HOST __inet_lookup_established(struct net *net,
 200                                   struct inet_hashinfo *hashinfo,
 201                                   const __be32 saddr, const __be16 sport,
 202                                   const __be32 daddr, const u16 hnum,
 203                                   const int dif)
 204 {
 205         INET_ADDR_COOKIE(acookie, saddr, daddr)
 206         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
 207         struct sock *sk;
 208         const struct hlist_nulls_node *node;
 209         /* Optimize here for direct hit, only listening connections can
 210          * have wildcards anyways.
 211          */
 212         unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
 213         unsigned int slot = hash & hashinfo->ehash_mask;
 214         struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
 215
 216         rcu_read_lock();
 217 begin:
 218         sk_nulls_for_each_rcu(sk, node, &head->chain) {
 219                 if (INET_MATCH(sk, net, hash, acookie,
 220                                         saddr, daddr, ports, dif)) {
 221                         if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
 222                                 goto begintw;
 223                         if (unlikely(!INET_MATCH(sk, net, hash, acookie,
 224                                 saddr, daddr, ports, dif))) {
 225                                 sock_put(sk);
 226                                 goto begin;
 227                         }
 228                         goto out;
 229                 }
 230         }
 231         /*
 232          * if the nulls value we got at the end of this lookup is
 233          * not the expected one, we must restart lookup.
 234          * We probably met an item that was moved to another chain.
 235          */
 236         if (get_nulls_value(node) != slot)
 237                 goto begin;
 238
 239 begintw:
 240         /* Must check for a TIME_WAIT'er before going to listener hash. */
 241         sk_nulls_for_each_rcu(sk, node, &head->twchain) {
 242                 if (INET_TW_MATCH(sk, net, hash, acookie,
 243                                         saddr, daddr, ports, dif)) {
 244                         if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) {
 245                                 sk = NULL;
 246                                 goto out;
 247                         }
 248                         if (unlikely(!INET_TW_MATCH(sk, net, hash, acookie,
 249                                  saddr, daddr, ports, dif))) {
 250                                 sock_put(sk);
 251                                 goto begintw;
 252                         }
 253                         goto out;
 254                 }
 255         }
 256         /*
 257          * if the nulls value we got at the end of this lookup is
 258          * not the expected one, we must restart lookup.
 259          * We probably met an item that was moved to another chain.
 260          */
 261         if (get_nulls_value(node) != slot)
 262                 goto begintw;
 263         sk = NULL;
 264 out:
 265         rcu_read_unlock();
 266         return sk;
 267 }
 268 EXPORT_SYMBOL_GPL(__inet_lookup_established);
 269
 270 /* called with local bh disabled */
 271 static int __inet_check_established(struct inet_timewait_death_row *death_row,
 272                                     struct sock *sk, __u16 lport,
 273                                     struct inet_timewait_sock **twp)
 274 {
 275         struct inet_hashinfo *hinfo = death_row->hashinfo;
 276         struct inet_sock *inet = inet_sk(sk);
 277         __be32 daddr = inet->inet_rcv_saddr;
 278         __be32 saddr = inet->inet_daddr;
 279         int dif = sk->sk_bound_dev_if;
 280         INET_ADDR_COOKIE(acookie, saddr, daddr)
 281         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
 282         struct net *net = sock_net(sk);
 283         unsigned int hash = inet_ehashfn(net, daddr, lport,
 284                                          saddr, inet->inet_dport);
 285         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 286         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
 287         struct sock *sk2;
 288         const struct hlist_nulls_node *node;
 289         struct inet_timewait_sock *tw;
 290         int twrefcnt = 0;
 291
 292         spin_lock(lock);
 293
 294         /* Check TIME-WAIT sockets first. */
 295         sk_nulls_for_each(sk2, node, &head->twchain) {
 296                 tw = inet_twsk(sk2);
 297
 298                 if (INET_TW_MATCH(sk2, net, hash, acookie,
 299                                         saddr, daddr, ports, dif)) {
 300                         if (twsk_unique(sk, sk2, twp))
 301                                 goto unique;
 302                         else
 303                                 goto not_unique;
 304                 }
 305         }
 306         tw = NULL;
 307
 308         /* And established part... */
 309         sk_nulls_for_each(sk2, node, &head->chain) {
 310                 if (INET_MATCH(sk2, net, hash, acookie,
 311                                         saddr, daddr, ports, dif))
 312                         goto not_unique;
 313         }
 314
 315 unique:
 316         /* Must record num and sport now. Otherwise we will see
 317          * in hash table socket with a funny identity. */
 318         inet->inet_num = lport;
 319         inet->inet_sport = htons(lport);
 320         sk->sk_hash = hash;
 321         WARN_ON(!sk_unhashed(sk));
 322         __sk_nulls_add_node_rcu(sk, &head->chain);
 323         if (tw) {
 324                 twrefcnt = inet_twsk_unhash(tw);
 325                 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
 326         }
 327         spin_unlock(lock);
 328         if (twrefcnt)
 329                 inet_twsk_put(tw);
 330         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 331
 332         if (twp) {
 333                 *twp = tw;
 334         } else if (tw) {
 335                 /* Silly. Should hash-dance instead... */
 336                 inet_twsk_deschedule(tw, death_row);
 337
 338                 inet_twsk_put(tw);
 339         }
 340         return 0;
 341
 342 not_unique:
 343         spin_unlock(lock);
 344         return -EADDRNOTAVAIL;
 345 }
 346
 347 static inline u32 inet_sk_port_offset(const struct sock *sk)
 348 {
 349         const struct inet_sock *inet = inet_sk(sk);
 350         return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
 351                                           inet->inet_daddr,
 352                                           inet->inet_dport);
 353 }
 354
 355 int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw)
 356 {
 357         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 358         struct hlist_nulls_head *list;
 359         spinlock_t *lock;
 360         struct inet_ehash_bucket *head;
 361         int twrefcnt = 0;
 362
 363         WARN_ON(!sk_unhashed(sk));
 364
 365         sk->sk_hash = inet_sk_ehashfn(sk);
 366         head = inet_ehash_bucket(hashinfo, sk->sk_hash);
 367         list = &head->chain;
 368         lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 369
 370         spin_lock(lock);
 371         __sk_nulls_add_node_rcu(sk, list);
 372         if (tw) {
 373                 WARN_ON(sk->sk_hash != tw->tw_hash);
 374                 twrefcnt = inet_twsk_unhash(tw);
 375         }
 376         spin_unlock(lock);
 377         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 378         return twrefcnt;
 379 }
 380 EXPORT_SYMBOL_GPL(__inet_hash_nolisten);
 381
 382 static void __inet_hash(struct sock *sk)
 383 {
 384         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 385         struct inet_listen_hashbucket *ilb;
 386
 387         if (sk->sk_state != TCP_LISTEN) {
 388                 __inet_hash_nolisten(sk, NULL);
 389                 return;
 390         }
 391
 392         WARN_ON(!sk_unhashed(sk));
 393         ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
 394
 395         spin_lock(&ilb->lock);
 396         __sk_nulls_add_node_rcu(sk, &ilb->head);
 397         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 398         spin_unlock(&ilb->lock);
 399 }
 400
 401 void inet_hash(struct sock *sk)
 402 {
 403         if (sk->sk_state != TCP_CLOSE) {
 404                 local_bh_disable();
 405                 __inet_hash(sk);
 406                 local_bh_enable();
 407         }
 408 }
 409 EXPORT_SYMBOL_GPL(inet_hash);
 410
 411 void inet_unhash(struct sock *sk)
 412 {
 413         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 414         spinlock_t *lock;
 415         int done;
 416
 417         if (sk_unhashed(sk))
 418                 return;
 419
 420         if (sk->sk_state == TCP_LISTEN)
 421                 lock = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)].lock;
 422         else
 423                 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 424
 425         spin_lock_bh(lock);
 426         done =__sk_nulls_del_node_init_rcu(sk);
 427         if (done)
 428                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 429         spin_unlock_bh(lock);
 430 }
 431 EXPORT_SYMBOL_GPL(inet_unhash);
 432
 433 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 434                 struct sock *sk, u32 port_offset,
 435                 int (*check_established)(struct inet_timewait_death_row *,
 436                         struct sock *, __u16, struct inet_timewait_sock **),
 437                 int (*hash)(struct sock *sk, struct inet_timewait_sock *twp))
 438 {
 439         struct inet_hashinfo *hinfo = death_row->hashinfo;
 440         const unsigned short snum = inet_sk(sk)->inet_num;
 441         struct inet_bind_hashbucket *head;
 442         struct inet_bind_bucket *tb;
 443         int ret;
 444         struct net *net = sock_net(sk);
 445         int twrefcnt = 1;
 446
 447         if (!snum) {
 448                 int i, remaining, low, high, port;
 449                 static u32 hint;
 450                 u32 offset = hint + port_offset;
 451                 struct hlist_node *node;
 452                 struct inet_timewait_sock *tw = NULL;
 453
 454                 inet_get_local_port_range(&low, &high);
 455                 remaining = (high - low) + 1;
 456
 457                 local_bh_disable();
 458                 for (i = 1; i <= remaining; i++) {
 459                         port = low + (i + offset) % remaining;
 460                         if (inet_is_reserved_local_port(port))
 461                                 continue;
 462                         head = &hinfo->bhash[inet_bhashfn(net, port,
 463                                         hinfo->bhash_size)];
 464                         spin_lock(&head->lock);
 465
 466                         /* Does not bother with rcv_saddr checks,
 467                          * because the established check is already
 468                          * unique enough.
 469                          */
 470                         inet_bind_bucket_for_each(tb, node, &head->chain) {
 471                                 if (net_eq(ib_net(tb), net) &&
 472                                     tb->port == port) {
 473                                         if (tb->fastreuse >= 0)
 474                                                 goto next_port;
 475                                         WARN_ON(hlist_empty(&tb->owners));
 476                                         if (!check_established(death_row, sk,
 477                                                                 port, &tw))
 478                                                 goto ok;
 479                                         goto next_port;
 480                                 }
 481                         }
 482
 483                         tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 484                                         net, head, port);
 485                         if (!tb) {
 486                                 spin_unlock(&head->lock);
 487                                 break;
 488                         }
 489                         tb->fastreuse = -1;
 490                         goto ok;
 491
 492                 next_port:
 493                         spin_unlock(&head->lock);
 494                 }
 495                 local_bh_enable();
 496
 497                 return -EADDRNOTAVAIL;
 498
 499 ok:
 500                 hint += i;
 501
 502                 /* Head lock still held and bh's disabled */
 503                 inet_bind_hash(sk, tb, port);
 504                 if (sk_unhashed(sk)) {
 505                         inet_sk(sk)->inet_sport = htons(port);
 506                         twrefcnt += hash(sk, tw);
 507                 }
 508                 if (tw)
 509                         twrefcnt += inet_twsk_bind_unhash(tw, hinfo);
 510                 spin_unlock(&head->lock);
 511
 512                 if (tw) {
 513                         inet_twsk_deschedule(tw, death_row);
 514                         while (twrefcnt) {
 515                                 twrefcnt--;
 516                                 inet_twsk_put(tw);
 517                         }
 518                 }
 519
 520                 ret = 0;
 521                 goto out;
 522         }
 523
 524         head = &hinfo->bhash[inet_bhashfn(net, snum, hinfo->bhash_size)];
 525         tb  = inet_csk(sk)->icsk_bind_hash;
 526         spin_lock_bh(&head->lock);
 527         if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
 528                 hash(sk, NULL);
 529                 spin_unlock_bh(&head->lock);
 530                 return 0;
 531         } else {
 532                 spin_unlock(&head->lock);
 533                 /* No definite answer... Walk to established hash table */
 534                 ret = check_established(death_row, sk, snum, NULL);
 535 out:
 536                 local_bh_enable();
 537                 return ret;
 538         }
 539 }
 540
 541 /*
 542  * Bind a port for a connect operation and hash it.
 543  */
 544 int inet_hash_connect(struct inet_timewait_death_row *death_row,
 545                       struct sock *sk)
 546 {
 547         return __inet_hash_connect(death_row, sk, inet_sk_port_offset(sk),
 548                         __inet_check_established, __inet_hash_nolisten);
 549 }
 550 EXPORT_SYMBOL_GPL(inet_hash_connect);
 551
 552 void inet_hashinfo_init(struct inet_hashinfo *h)
 553 {
 554         int i;
 555
 556         atomic_set(&h->bsockets, 0);
 557         for (i = 0; i < INET_LHTABLE_SIZE; i++) {
 558                 spin_lock_init(&h->listening_hash[i].lock);
 559                 INIT_HLIST_NULLS_HEAD(&h->listening_hash[i].head,
 560                                       i + LISTENING_NULLS_BASE);
 561                 }
 562 }
 563 EXPORT_SYMBOL_GPL(inet_hashinfo_init);