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 #include <linux/vmalloc.h>
  22
  23 #include <net/inet_connection_sock.h>
  24 #include <net/inet_hashtables.h>
  25 #include <net/secure_seq.h>
  26 #include <net/ip.h>
  27
  28 static u32 inet_ehashfn(const struct net *net, const __be32 laddr,
  29                         const __u16 lport, const __be32 faddr,
  30                         const __be16 fport)
  31 {
  32         static u32 inet_ehash_secret __read_mostly;
  33
  34         net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret));
  35
  36         return __inet_ehashfn(laddr, lport, faddr, fport,
  37                               inet_ehash_secret + net_hash_mix(net));
  38 }
  39
  40 /* This function handles inet_sock, but also timewait and request sockets
  41  * for IPv4/IPv6.
  42  */
  43 u32 sk_ehashfn(const struct sock *sk)
  44 {
  45 #if IS_ENABLED(CONFIG_IPV6)
  46         if (sk->sk_family == AF_INET6 &&
  47             !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
  48                 return inet6_ehashfn(sock_net(sk),
  49                                      &sk->sk_v6_rcv_saddr, sk->sk_num,
  50                                      &sk->sk_v6_daddr, sk->sk_dport);
  51 #endif
  52         return inet_ehashfn(sock_net(sk),
  53                             sk->sk_rcv_saddr, sk->sk_num,
  54                             sk->sk_daddr, sk->sk_dport);
  55 }
  56
  57 /*
  58  * Allocate and initialize a new local port bind bucket.
  59  * The bindhash mutex for snum's hash chain must be held here.
  60  */
  61 struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
  62                                                  struct net *net,
  63                                                  struct inet_bind_hashbucket *head,
  64                                                  const unsigned short snum)
  65 {
  66         struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
  67
  68         if (tb) {
  69                 write_pnet(&tb->ib_net, net);
  70                 tb->port      = snum;
  71                 tb->fastreuse = 0;
  72                 tb->fastreuseport = 0;
  73                 tb->num_owners = 0;
  74                 INIT_HLIST_HEAD(&tb->owners);
  75                 hlist_add_head(&tb->node, &head->chain);
  76         }
  77         return tb;
  78 }
  79
  80 /*
  81  * Caller must hold hashbucket lock for this tb with local BH disabled
  82  */
  83 void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
  84 {
  85         if (hlist_empty(&tb->owners)) {
  86                 __hlist_del(&tb->node);
  87                 kmem_cache_free(cachep, tb);
  88         }
  89 }
  90
  91 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
  92                     const unsigned short snum)
  93 {
  94         inet_sk(sk)->inet_num = snum;
  95         sk_add_bind_node(sk, &tb->owners);
  96         tb->num_owners++;
  97         inet_csk(sk)->icsk_bind_hash = tb;
  98 }
  99
 100 /*
 101  * Get rid of any references to a local port held by the given sock.
 102  */
 103 static void __inet_put_port(struct sock *sk)
 104 {
 105         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 106         const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num,
 107                         hashinfo->bhash_size);
 108         struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
 109         struct inet_bind_bucket *tb;
 110
 111         spin_lock(&head->lock);
 112         tb = inet_csk(sk)->icsk_bind_hash;
 113         __sk_del_bind_node(sk);
 114         tb->num_owners--;
 115         inet_csk(sk)->icsk_bind_hash = NULL;
 116         inet_sk(sk)->inet_num = 0;
 117         inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
 118         spin_unlock(&head->lock);
 119 }
 120
 121 void inet_put_port(struct sock *sk)
 122 {
 123         local_bh_disable();
 124         __inet_put_port(sk);
 125         local_bh_enable();
 126 }
 127 EXPORT_SYMBOL(inet_put_port);
 128
 129 int __inet_inherit_port(const struct sock *sk, struct sock *child)
 130 {
 131         struct inet_hashinfo *table = sk->sk_prot->h.hashinfo;
 132         unsigned short port = inet_sk(child)->inet_num;
 133         const int bhash = inet_bhashfn(sock_net(sk), port,
 134                         table->bhash_size);
 135         struct inet_bind_hashbucket *head = &table->bhash[bhash];
 136         struct inet_bind_bucket *tb;
 137
 138         spin_lock(&head->lock);
 139         tb = inet_csk(sk)->icsk_bind_hash;
 140         if (unlikely(!tb)) {
 141                 spin_unlock(&head->lock);
 142                 return -ENOENT;
 143         }
 144         if (tb->port != port) {
 145                 /* NOTE: using tproxy and redirecting skbs to a proxy
 146                  * on a different listener port breaks the assumption
 147                  * that the listener socket's icsk_bind_hash is the same
 148                  * as that of the child socket. We have to look up or
 149                  * create a new bind bucket for the child here. */
 150                 inet_bind_bucket_for_each(tb, &head->chain) {
 151                         if (net_eq(ib_net(tb), sock_net(sk)) &&
 152                             tb->port == port)
 153                                 break;
 154                 }
 155                 if (!tb) {
 156                         tb = inet_bind_bucket_create(table->bind_bucket_cachep,
 157                                                      sock_net(sk), head, port);
 158                         if (!tb) {
 159                                 spin_unlock(&head->lock);
 160                                 return -ENOMEM;
 161                         }
 162                 }
 163         }
 164         inet_bind_hash(child, tb, port);
 165         spin_unlock(&head->lock);
 166
 167         return 0;
 168 }
 169 EXPORT_SYMBOL_GPL(__inet_inherit_port);
 170
 171 static inline int compute_score(struct sock *sk, struct net *net,
 172                                 const unsigned short hnum, const __be32 daddr,
 173                                 const int dif)
 174 {
 175         int score = -1;
 176         struct inet_sock *inet = inet_sk(sk);
 177
 178         if (net_eq(sock_net(sk), net) && inet->inet_num == hnum &&
 179                         !ipv6_only_sock(sk)) {
 180                 __be32 rcv_saddr = inet->inet_rcv_saddr;
 181                 score = sk->sk_family == PF_INET ? 2 : 1;
 182                 if (rcv_saddr) {
 183                         if (rcv_saddr != daddr)
 184                                 return -1;
 185                         score += 4;
 186                 }
 187                 if (sk->sk_bound_dev_if) {
 188                         if (sk->sk_bound_dev_if != dif)
 189                                 return -1;
 190                         score += 4;
 191                 }
 192                 if (sk->sk_incoming_cpu == raw_smp_processor_id())
 193                         score++;
 194         }
 195         return score;
 196 }
 197
 198 /*
 199  * Don't inline this cruft. Here are some nice properties to exploit here. The
 200  * BSD API does not allow a listening sock to specify the remote port nor the
 201  * remote address for the connection. So always assume those are both
 202  * wildcarded during the search since they can never be otherwise.
 203  */
 204
 205
 206 struct sock *__inet_lookup_listener(struct net *net,
 207                                     struct inet_hashinfo *hashinfo,
 208                                     const __be32 saddr, __be16 sport,
 209                                     const __be32 daddr, const unsigned short hnum,
 210                                     const int dif)
 211 {
 212         struct sock *sk, *result;
 213         struct hlist_nulls_node *node;
 214         unsigned int hash = inet_lhashfn(net, hnum);
 215         struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
 216         int score, hiscore, matches = 0, reuseport = 0;
 217         u32 phash = 0;
 218
 219         rcu_read_lock();
 220 begin:
 221         result = NULL;
 222         hiscore = 0;
 223         sk_nulls_for_each_rcu(sk, node, &ilb->head) {
 224                 score = compute_score(sk, net, hnum, daddr, dif);
 225                 if (score > hiscore) {
 226                         result = sk;
 227                         hiscore = score;
 228                         reuseport = sk->sk_reuseport;
 229                         if (reuseport) {
 230                                 phash = inet_ehashfn(net, daddr, hnum,
 231                                                      saddr, sport);
 232                                 matches = 1;
 233                         }
 234                 } else if (score == hiscore && reuseport) {
 235                         matches++;
 236                         if (reciprocal_scale(phash, matches) == 0)
 237                                 result = sk;
 238                         phash = next_pseudo_random32(phash);
 239                 }
 240         }
 241         /*
 242          * if the nulls value we got at the end of this lookup is
 243          * not the expected one, we must restart lookup.
 244          * We probably met an item that was moved to another chain.
 245          */
 246         if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
 247                 goto begin;
 248         if (result) {
 249                 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
 250                         result = NULL;
 251                 else if (unlikely(compute_score(result, net, hnum, daddr,
 252                                   dif) < hiscore)) {
 253                         sock_put(result);
 254                         goto begin;
 255                 }
 256         }
 257         rcu_read_unlock();
 258         return result;
 259 }
 260 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
 261
 262 /* All sockets share common refcount, but have different destructors */
 263 void sock_gen_put(struct sock *sk)
 264 {
 265         if (!atomic_dec_and_test(&sk->sk_refcnt))
 266                 return;
 267
 268         if (sk->sk_state == TCP_TIME_WAIT)
 269                 inet_twsk_free(inet_twsk(sk));
 270         else if (sk->sk_state == TCP_NEW_SYN_RECV)
 271                 reqsk_free(inet_reqsk(sk));
 272         else
 273                 sk_free(sk);
 274 }
 275 EXPORT_SYMBOL_GPL(sock_gen_put);
 276
 277 void sock_edemux(struct sk_buff *skb)
 278 {
 279         sock_gen_put(skb->sk);
 280 }
 281 EXPORT_SYMBOL(sock_edemux);
 282
 283 struct sock *__inet_lookup_established(struct net *net,
 284                                   struct inet_hashinfo *hashinfo,
 285                                   const __be32 saddr, const __be16 sport,
 286                                   const __be32 daddr, const u16 hnum,
 287                                   const int dif)
 288 {
 289         INET_ADDR_COOKIE(acookie, saddr, daddr);
 290         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
 291         struct sock *sk;
 292         const struct hlist_nulls_node *node;
 293         /* Optimize here for direct hit, only listening connections can
 294          * have wildcards anyways.
 295          */
 296         unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
 297         unsigned int slot = hash & hashinfo->ehash_mask;
 298         struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
 299
 300         rcu_read_lock();
 301 begin:
 302         sk_nulls_for_each_rcu(sk, node, &head->chain) {
 303                 if (sk->sk_hash != hash)
 304                         continue;
 305                 if (likely(INET_MATCH(sk, net, acookie,
 306                                       saddr, daddr, ports, dif))) {
 307                         if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
 308                                 goto out;
 309                         if (unlikely(!INET_MATCH(sk, net, acookie,
 310                                                  saddr, daddr, ports, dif))) {
 311                                 sock_gen_put(sk);
 312                                 goto begin;
 313                         }
 314                         goto found;
 315                 }
 316         }
 317         /*
 318          * if the nulls value we got at the end of this lookup is
 319          * not the expected one, we must restart lookup.
 320          * We probably met an item that was moved to another chain.
 321          */
 322         if (get_nulls_value(node) != slot)
 323                 goto begin;
 324 out:
 325         sk = NULL;
 326 found:
 327         rcu_read_unlock();
 328         return sk;
 329 }
 330 EXPORT_SYMBOL_GPL(__inet_lookup_established);
 331
 332 /* called with local bh disabled */
 333 static int __inet_check_established(struct inet_timewait_death_row *death_row,
 334                                     struct sock *sk, __u16 lport,
 335                                     struct inet_timewait_sock **twp)
 336 {
 337         struct inet_hashinfo *hinfo = death_row->hashinfo;
 338         struct inet_sock *inet = inet_sk(sk);
 339         __be32 daddr = inet->inet_rcv_saddr;
 340         __be32 saddr = inet->inet_daddr;
 341         int dif = sk->sk_bound_dev_if;
 342         INET_ADDR_COOKIE(acookie, saddr, daddr);
 343         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
 344         struct net *net = sock_net(sk);
 345         unsigned int hash = inet_ehashfn(net, daddr, lport,
 346                                          saddr, inet->inet_dport);
 347         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 348         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
 349         struct sock *sk2;
 350         const struct hlist_nulls_node *node;
 351         struct inet_timewait_sock *tw = NULL;
 352
 353         spin_lock(lock);
 354
 355         sk_nulls_for_each(sk2, node, &head->chain) {
 356                 if (sk2->sk_hash != hash)
 357                         continue;
 358
 359                 if (likely(INET_MATCH(sk2, net, acookie,
 360                                          saddr, daddr, ports, dif))) {
 361                         if (sk2->sk_state == TCP_TIME_WAIT) {
 362                                 tw = inet_twsk(sk2);
 363                                 if (twsk_unique(sk, sk2, twp))
 364                                         break;
 365                         }
 366                         goto not_unique;
 367                 }
 368         }
 369
 370         /* Must record num and sport now. Otherwise we will see
 371          * in hash table socket with a funny identity.
 372          */
 373         inet->inet_num = lport;
 374         inet->inet_sport = htons(lport);
 375         sk->sk_hash = hash;
 376         WARN_ON(!sk_unhashed(sk));
 377         __sk_nulls_add_node_rcu(sk, &head->chain);
 378         if (tw) {
 379                 sk_nulls_del_node_init_rcu((struct sock *)tw);
 380                 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
 381         }
 382         spin_unlock(lock);
 383         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 384
 385         if (twp) {
 386                 *twp = tw;
 387         } else if (tw) {
 388                 /* Silly. Should hash-dance instead... */
 389                 inet_twsk_deschedule_put(tw);
 390         }
 391         return 0;
 392
 393 not_unique:
 394         spin_unlock(lock);
 395         return -EADDRNOTAVAIL;
 396 }
 397
 398 static u32 inet_sk_port_offset(const struct sock *sk)
 399 {
 400         const struct inet_sock *inet = inet_sk(sk);
 401
 402         return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
 403                                           inet->inet_daddr,
 404                                           inet->inet_dport);
 405 }
 406
 407 /* insert a socket into ehash, and eventually remove another one
 408  * (The another one can be a SYN_RECV or TIMEWAIT
 409  */
 410 bool inet_ehash_insert(struct sock *sk, struct sock *osk)
 411 {
 412         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 413         struct hlist_nulls_head *list;
 414         struct inet_ehash_bucket *head;
 415         spinlock_t *lock;
 416         bool ret = true;
 417
 418         WARN_ON_ONCE(!sk_unhashed(sk));
 419
 420         sk->sk_hash = sk_ehashfn(sk);
 421         head = inet_ehash_bucket(hashinfo, sk->sk_hash);
 422         list = &head->chain;
 423         lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 424
 425         spin_lock(lock);
 426         if (osk) {
 427                 WARN_ON_ONCE(sk->sk_hash != osk->sk_hash);
 428                 ret = sk_nulls_del_node_init_rcu(osk);
 429         }
 430         if (ret)
 431                 __sk_nulls_add_node_rcu(sk, list);
 432         spin_unlock(lock);
 433         return ret;
 434 }
 435
 436 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk)
 437 {
 438         bool ok = inet_ehash_insert(sk, osk);
 439
 440         if (ok) {
 441                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 442         } else {
 443                 percpu_counter_inc(sk->sk_prot->orphan_count);
 444                 sk->sk_state = TCP_CLOSE;
 445                 sock_set_flag(sk, SOCK_DEAD);
 446                 inet_csk_destroy_sock(sk);
 447         }
 448         return ok;
 449 }
 450 EXPORT_SYMBOL_GPL(inet_ehash_nolisten);
 451
 452 void __inet_hash(struct sock *sk, struct sock *osk)
 453 {
 454         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 455         struct inet_listen_hashbucket *ilb;
 456
 457         if (sk->sk_state != TCP_LISTEN) {
 458                 inet_ehash_nolisten(sk, osk);
 459                 return;
 460         }
 461         WARN_ON(!sk_unhashed(sk));
 462         ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
 463
 464         spin_lock(&ilb->lock);
 465         __sk_nulls_add_node_rcu(sk, &ilb->head);
 466         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 467         spin_unlock(&ilb->lock);
 468 }
 469 EXPORT_SYMBOL(__inet_hash);
 470
 471 void inet_hash(struct sock *sk)
 472 {
 473         if (sk->sk_state != TCP_CLOSE) {
 474                 local_bh_disable();
 475                 __inet_hash(sk, NULL);
 476                 local_bh_enable();
 477         }
 478 }
 479 EXPORT_SYMBOL_GPL(inet_hash);
 480
 481 void inet_unhash(struct sock *sk)
 482 {
 483         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 484         spinlock_t *lock;
 485         int done;
 486
 487         if (sk_unhashed(sk))
 488                 return;
 489
 490         if (sk->sk_state == TCP_LISTEN)
 491                 lock = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)].lock;
 492         else
 493                 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 494
 495         spin_lock_bh(lock);
 496         done = __sk_nulls_del_node_init_rcu(sk);
 497         if (done)
 498                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 499         spin_unlock_bh(lock);
 500 }
 501 EXPORT_SYMBOL_GPL(inet_unhash);
 502
 503 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 504                 struct sock *sk, u32 port_offset,
 505                 int (*check_established)(struct inet_timewait_death_row *,
 506                         struct sock *, __u16, struct inet_timewait_sock **))
 507 {
 508         struct inet_hashinfo *hinfo = death_row->hashinfo;
 509         const unsigned short snum = inet_sk(sk)->inet_num;
 510         struct inet_bind_hashbucket *head;
 511         struct inet_bind_bucket *tb;
 512         int ret;
 513         struct net *net = sock_net(sk);
 514
 515         if (!snum) {
 516                 int i, remaining, low, high, port;
 517                 static u32 hint;
 518                 u32 offset = hint + port_offset;
 519                 struct inet_timewait_sock *tw = NULL;
 520
 521                 inet_get_local_port_range(net, &low, &high);
 522                 remaining = (high - low) + 1;
 523
 524                 /* By starting with offset being an even number,
 525                  * we tend to leave about 50% of ports for other uses,
 526                  * like bind(0).
 527                  */
 528                 offset &= ~1;
 529
 530                 local_bh_disable();
 531                 for (i = 0; i < remaining; i++) {
 532                         port = low + (i + offset) % remaining;
 533                         if (inet_is_local_reserved_port(net, port))
 534                                 continue;
 535                         head = &hinfo->bhash[inet_bhashfn(net, port,
 536                                         hinfo->bhash_size)];
 537                         spin_lock(&head->lock);
 538
 539                         /* Does not bother with rcv_saddr checks,
 540                          * because the established check is already
 541                          * unique enough.
 542                          */
 543                         inet_bind_bucket_for_each(tb, &head->chain) {
 544                                 if (net_eq(ib_net(tb), net) &&
 545                                     tb->port == port) {
 546                                         if (tb->fastreuse >= 0 ||
 547                                             tb->fastreuseport >= 0)
 548                                                 goto next_port;
 549                                         WARN_ON(hlist_empty(&tb->owners));
 550                                         if (!check_established(death_row, sk,
 551                                                                 port, &tw))
 552                                                 goto ok;
 553                                         goto next_port;
 554                                 }
 555                         }
 556
 557                         tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 558                                         net, head, port);
 559                         if (!tb) {
 560                                 spin_unlock(&head->lock);
 561                                 break;
 562                         }
 563                         tb->fastreuse = -1;
 564                         tb->fastreuseport = -1;
 565                         goto ok;
 566
 567                 next_port:
 568                         spin_unlock(&head->lock);
 569                 }
 570                 local_bh_enable();
 571
 572                 return -EADDRNOTAVAIL;
 573
 574 ok:
 575                 hint += (i + 2) & ~1;
 576
 577                 /* Head lock still held and bh's disabled */
 578                 inet_bind_hash(sk, tb, port);
 579                 if (sk_unhashed(sk)) {
 580                         inet_sk(sk)->inet_sport = htons(port);
 581                         inet_ehash_nolisten(sk, (struct sock *)tw);
 582                 }
 583                 if (tw)
 584                         inet_twsk_bind_unhash(tw, hinfo);
 585                 spin_unlock(&head->lock);
 586
 587                 if (tw)
 588                         inet_twsk_deschedule_put(tw);
 589
 590                 ret = 0;
 591                 goto out;
 592         }
 593
 594         head = &hinfo->bhash[inet_bhashfn(net, snum, hinfo->bhash_size)];
 595         tb  = inet_csk(sk)->icsk_bind_hash;
 596         spin_lock_bh(&head->lock);
 597         if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
 598                 inet_ehash_nolisten(sk, NULL);
 599                 spin_unlock_bh(&head->lock);
 600                 return 0;
 601         } else {
 602                 spin_unlock(&head->lock);
 603                 /* No definite answer... Walk to established hash table */
 604                 ret = check_established(death_row, sk, snum, NULL);
 605 out:
 606                 local_bh_enable();
 607                 return ret;
 608         }
 609 }
 610
 611 /*
 612  * Bind a port for a connect operation and hash it.
 613  */
 614 int inet_hash_connect(struct inet_timewait_death_row *death_row,
 615                       struct sock *sk)
 616 {
 617         u32 port_offset = 0;
 618
 619         if (!inet_sk(sk)->inet_num)
 620                 port_offset = inet_sk_port_offset(sk);
 621         return __inet_hash_connect(death_row, sk, port_offset,
 622                                    __inet_check_established);
 623 }
 624 EXPORT_SYMBOL_GPL(inet_hash_connect);
 625
 626 void inet_hashinfo_init(struct inet_hashinfo *h)
 627 {
 628         int i;
 629
 630         for (i = 0; i < INET_LHTABLE_SIZE; i++) {
 631                 spin_lock_init(&h->listening_hash[i].lock);
 632                 INIT_HLIST_NULLS_HEAD(&h->listening_hash[i].head,
 633                                       i + LISTENING_NULLS_BASE);
 634                 }
 635 }
 636 EXPORT_SYMBOL_GPL(inet_hashinfo_init);
 637
 638 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
 639 {
 640         unsigned int locksz = sizeof(spinlock_t);
 641         unsigned int i, nblocks = 1;
 642
 643         if (locksz != 0) {
 644                 /* allocate 2 cache lines or at least one spinlock per cpu */
 645                 nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U);
 646                 nblocks = roundup_pow_of_two(nblocks * num_possible_cpus());
 647
 648                 /* no more locks than number of hash buckets */
 649                 nblocks = min(nblocks, hashinfo->ehash_mask + 1);
 650
 651                 hashinfo->ehash_locks = kmalloc_array(nblocks, locksz,
 652                                                       GFP_KERNEL | __GFP_NOWARN);
 653                 if (!hashinfo->ehash_locks)
 654                         hashinfo->ehash_locks = vmalloc(nblocks * locksz);
 655
 656                 if (!hashinfo->ehash_locks)
 657                         return -ENOMEM;
 658
 659                 for (i = 0; i < nblocks; i++)
 660                         spin_lock_init(&hashinfo->ehash_locks[i]);
 661         }
 662         hashinfo->ehash_locks_mask = nblocks - 1;
 663         return 0;
 664 }
 665 EXPORT_SYMBOL_GPL(inet_ehash_locks_alloc);