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/addrconf.h>
  24 #include <net/inet_connection_sock.h>
  25 #include <net/inet_hashtables.h>
  26 #include <net/secure_seq.h>
  27 #include <net/ip.h>
  28 #include <net/tcp.h>
  29 #include <net/sock_reuseport.h>
  30
  31 static u32 inet_ehashfn(const struct net *net, const __be32 laddr,
  32                         const __u16 lport, const __be32 faddr,
  33                         const __be16 fport)
  34 {
  35         static u32 inet_ehash_secret __read_mostly;
  36
  37         net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret));
  38
  39         return __inet_ehashfn(laddr, lport, faddr, fport,
  40                               inet_ehash_secret + net_hash_mix(net));
  41 }
  42
  43 /* This function handles inet_sock, but also timewait and request sockets
  44  * for IPv4/IPv6.
  45  */
  46 u32 sk_ehashfn(const struct sock *sk)
  47 {
  48 #if IS_ENABLED(CONFIG_IPV6)
  49         if (sk->sk_family == AF_INET6 &&
  50             !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
  51                 return inet6_ehashfn(sock_net(sk),
  52                                      &sk->sk_v6_rcv_saddr, sk->sk_num,
  53                                      &sk->sk_v6_daddr, sk->sk_dport);
  54 #endif
  55         return inet_ehashfn(sock_net(sk),
  56                             sk->sk_rcv_saddr, sk->sk_num,
  57                             sk->sk_daddr, sk->sk_dport);
  58 }
  59
  60 /*
  61  * Allocate and initialize a new local port bind bucket.
  62  * The bindhash mutex for snum's hash chain must be held here.
  63  */
  64 struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
  65                                                  struct net *net,
  66                                                  struct inet_bind_hashbucket *head,
  67                                                  const unsigned short snum)
  68 {
  69         struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
  70
  71         if (tb) {
  72                 write_pnet(&tb->ib_net, net);
  73                 tb->port      = snum;
  74                 tb->fastreuse = 0;
  75                 tb->fastreuseport = 0;
  76                 tb->num_owners = 0;
  77                 INIT_HLIST_HEAD(&tb->owners);
  78                 hlist_add_head(&tb->node, &head->chain);
  79         }
  80         return tb;
  81 }
  82
  83 /*
  84  * Caller must hold hashbucket lock for this tb with local BH disabled
  85  */
  86 void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
  87 {
  88         if (hlist_empty(&tb->owners)) {
  89                 __hlist_del(&tb->node);
  90                 kmem_cache_free(cachep, tb);
  91         }
  92 }
  93
  94 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
  95                     const unsigned short snum)
  96 {
  97         inet_sk(sk)->inet_num = snum;
  98         sk_add_bind_node(sk, &tb->owners);
  99         tb->num_owners++;
 100         inet_csk(sk)->icsk_bind_hash = tb;
 101 }
 102
 103 /*
 104  * Get rid of any references to a local port held by the given sock.
 105  */
 106 static void __inet_put_port(struct sock *sk)
 107 {
 108         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 109         const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num,
 110                         hashinfo->bhash_size);
 111         struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
 112         struct inet_bind_bucket *tb;
 113
 114         spin_lock(&head->lock);
 115         tb = inet_csk(sk)->icsk_bind_hash;
 116         __sk_del_bind_node(sk);
 117         tb->num_owners--;
 118         inet_csk(sk)->icsk_bind_hash = NULL;
 119         inet_sk(sk)->inet_num = 0;
 120         inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
 121         spin_unlock(&head->lock);
 122 }
 123
 124 void inet_put_port(struct sock *sk)
 125 {
 126         local_bh_disable();
 127         __inet_put_port(sk);
 128         local_bh_enable();
 129 }
 130 EXPORT_SYMBOL(inet_put_port);
 131
 132 int __inet_inherit_port(const struct sock *sk, struct sock *child)
 133 {
 134         struct inet_hashinfo *table = sk->sk_prot->h.hashinfo;
 135         unsigned short port = inet_sk(child)->inet_num;
 136         const int bhash = inet_bhashfn(sock_net(sk), port,
 137                         table->bhash_size);
 138         struct inet_bind_hashbucket *head = &table->bhash[bhash];
 139         struct inet_bind_bucket *tb;
 140
 141         spin_lock(&head->lock);
 142         tb = inet_csk(sk)->icsk_bind_hash;
 143         if (unlikely(!tb)) {
 144                 spin_unlock(&head->lock);
 145                 return -ENOENT;
 146         }
 147         if (tb->port != port) {
 148                 /* NOTE: using tproxy and redirecting skbs to a proxy
 149                  * on a different listener port breaks the assumption
 150                  * that the listener socket's icsk_bind_hash is the same
 151                  * as that of the child socket. We have to look up or
 152                  * create a new bind bucket for the child here. */
 153                 inet_bind_bucket_for_each(tb, &head->chain) {
 154                         if (net_eq(ib_net(tb), sock_net(sk)) &&
 155                             tb->port == port)
 156                                 break;
 157                 }
 158                 if (!tb) {
 159                         tb = inet_bind_bucket_create(table->bind_bucket_cachep,
 160                                                      sock_net(sk), head, port);
 161                         if (!tb) {
 162                                 spin_unlock(&head->lock);
 163                                 return -ENOMEM;
 164                         }
 165                 }
 166         }
 167         inet_bind_hash(child, tb, port);
 168         spin_unlock(&head->lock);
 169
 170         return 0;
 171 }
 172 EXPORT_SYMBOL_GPL(__inet_inherit_port);
 173
 174 static inline int compute_score(struct sock *sk, struct net *net,
 175                                 const unsigned short hnum, const __be32 daddr,
 176                                 const int dif, bool exact_dif)
 177 {
 178         int score = -1;
 179         struct inet_sock *inet = inet_sk(sk);
 180
 181         if (net_eq(sock_net(sk), net) && inet->inet_num == hnum &&
 182                         !ipv6_only_sock(sk)) {
 183                 __be32 rcv_saddr = inet->inet_rcv_saddr;
 184                 score = sk->sk_family == PF_INET ? 2 : 1;
 185                 if (rcv_saddr) {
 186                         if (rcv_saddr != daddr)
 187                                 return -1;
 188                         score += 4;
 189                 }
 190                 if (sk->sk_bound_dev_if || exact_dif) {
 191                         if (sk->sk_bound_dev_if != dif)
 192                                 return -1;
 193                         score += 4;
 194                 }
 195                 if (sk->sk_incoming_cpu == raw_smp_processor_id())
 196                         score++;
 197         }
 198         return score;
 199 }
 200
 201 /*
 202  * Here are some nice properties to exploit here. The BSD API
 203  * does not allow a listening sock to specify the remote port nor the
 204  * remote address for the connection. So always assume those are both
 205  * wildcarded during the search since they can never be otherwise.
 206  */
 207
 208 /* called with rcu_read_lock() : No refcount taken on the socket */
 209 struct sock *__inet_lookup_listener(struct net *net,
 210                                     struct inet_hashinfo *hashinfo,
 211                                     struct sk_buff *skb, int doff,
 212                                     const __be32 saddr, __be16 sport,
 213                                     const __be32 daddr, const unsigned short hnum,
 214                                     const int dif)
 215 {
 216         unsigned int hash = inet_lhashfn(net, hnum);
 217         struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
 218         int score, hiscore = 0, matches = 0, reuseport = 0;
 219         bool exact_dif = inet_exact_dif_match(net, skb);
 220         struct sock *sk, *result = NULL;
 221         u32 phash = 0;
 222
 223         sk_for_each_rcu(sk, &ilb->head) {
 224                 score = compute_score(sk, net, hnum, daddr, dif, exact_dif);
 225                 if (score > hiscore) {
 226                         reuseport = sk->sk_reuseport;
 227                         if (reuseport) {
 228                                 phash = inet_ehashfn(net, daddr, hnum,
 229                                                      saddr, sport);
 230                                 result = reuseport_select_sock(sk, phash,
 231                                                                skb, doff);
 232                                 if (result)
 233                                         return result;
 234                                 matches = 1;
 235                         }
 236                         result = sk;
 237                         hiscore = score;
 238                 } else if (score == hiscore && reuseport) {
 239                         matches++;
 240                         if (reciprocal_scale(phash, matches) == 0)
 241                                 result = sk;
 242                         phash = next_pseudo_random32(phash);
 243                 }
 244         }
 245         return result;
 246 }
 247 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
 248
 249 /* All sockets share common refcount, but have different destructors */
 250 void sock_gen_put(struct sock *sk)
 251 {
 252         if (!atomic_dec_and_test(&sk->sk_refcnt))
 253                 return;
 254
 255         if (sk->sk_state == TCP_TIME_WAIT)
 256                 inet_twsk_free(inet_twsk(sk));
 257         else if (sk->sk_state == TCP_NEW_SYN_RECV)
 258                 reqsk_free(inet_reqsk(sk));
 259         else
 260                 sk_free(sk);
 261 }
 262 EXPORT_SYMBOL_GPL(sock_gen_put);
 263
 264 void sock_edemux(struct sk_buff *skb)
 265 {
 266         sock_gen_put(skb->sk);
 267 }
 268 EXPORT_SYMBOL(sock_edemux);
 269
 270 struct sock *__inet_lookup_established(struct net *net,
 271                                   struct inet_hashinfo *hashinfo,
 272                                   const __be32 saddr, const __be16 sport,
 273                                   const __be32 daddr, const u16 hnum,
 274                                   const int dif)
 275 {
 276         INET_ADDR_COOKIE(acookie, saddr, daddr);
 277         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
 278         struct sock *sk;
 279         const struct hlist_nulls_node *node;
 280         /* Optimize here for direct hit, only listening connections can
 281          * have wildcards anyways.
 282          */
 283         unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
 284         unsigned int slot = hash & hashinfo->ehash_mask;
 285         struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
 286
 287 begin:
 288         sk_nulls_for_each_rcu(sk, node, &head->chain) {
 289                 if (sk->sk_hash != hash)
 290                         continue;
 291                 if (likely(INET_MATCH(sk, net, acookie,
 292                                       saddr, daddr, ports, dif))) {
 293                         if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
 294                                 goto out;
 295                         if (unlikely(!INET_MATCH(sk, net, acookie,
 296                                                  saddr, daddr, ports, dif))) {
 297                                 sock_gen_put(sk);
 298                                 goto begin;
 299                         }
 300                         goto found;
 301                 }
 302         }
 303         /*
 304          * if the nulls value we got at the end of this lookup is
 305          * not the expected one, we must restart lookup.
 306          * We probably met an item that was moved to another chain.
 307          */
 308         if (get_nulls_value(node) != slot)
 309                 goto begin;
 310 out:
 311         sk = NULL;
 312 found:
 313         return sk;
 314 }
 315 EXPORT_SYMBOL_GPL(__inet_lookup_established);
 316
 317 /* called with local bh disabled */
 318 static int __inet_check_established(struct inet_timewait_death_row *death_row,
 319                                     struct sock *sk, __u16 lport,
 320                                     struct inet_timewait_sock **twp)
 321 {
 322         struct inet_hashinfo *hinfo = death_row->hashinfo;
 323         struct inet_sock *inet = inet_sk(sk);
 324         __be32 daddr = inet->inet_rcv_saddr;
 325         __be32 saddr = inet->inet_daddr;
 326         int dif = sk->sk_bound_dev_if;
 327         INET_ADDR_COOKIE(acookie, saddr, daddr);
 328         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
 329         struct net *net = sock_net(sk);
 330         unsigned int hash = inet_ehashfn(net, daddr, lport,
 331                                          saddr, inet->inet_dport);
 332         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 333         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
 334         struct sock *sk2;
 335         const struct hlist_nulls_node *node;
 336         struct inet_timewait_sock *tw = NULL;
 337
 338         spin_lock(lock);
 339
 340         sk_nulls_for_each(sk2, node, &head->chain) {
 341                 if (sk2->sk_hash != hash)
 342                         continue;
 343
 344                 if (likely(INET_MATCH(sk2, net, acookie,
 345                                          saddr, daddr, ports, dif))) {
 346                         if (sk2->sk_state == TCP_TIME_WAIT) {
 347                                 tw = inet_twsk(sk2);
 348                                 if (twsk_unique(sk, sk2, twp))
 349                                         break;
 350                         }
 351                         goto not_unique;
 352                 }
 353         }
 354
 355         /* Must record num and sport now. Otherwise we will see
 356          * in hash table socket with a funny identity.
 357          */
 358         inet->inet_num = lport;
 359         inet->inet_sport = htons(lport);
 360         sk->sk_hash = hash;
 361         WARN_ON(!sk_unhashed(sk));
 362         __sk_nulls_add_node_rcu(sk, &head->chain);
 363         if (tw) {
 364                 sk_nulls_del_node_init_rcu((struct sock *)tw);
 365                 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
 366         }
 367         spin_unlock(lock);
 368         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 369
 370         if (twp) {
 371                 *twp = tw;
 372         } else if (tw) {
 373                 /* Silly. Should hash-dance instead... */
 374                 inet_twsk_deschedule_put(tw);
 375         }
 376         return 0;
 377
 378 not_unique:
 379         spin_unlock(lock);
 380         return -EADDRNOTAVAIL;
 381 }
 382
 383 static u32 inet_sk_port_offset(const struct sock *sk)
 384 {
 385         const struct inet_sock *inet = inet_sk(sk);
 386
 387         return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
 388                                           inet->inet_daddr,
 389                                           inet->inet_dport);
 390 }
 391
 392 /* insert a socket into ehash, and eventually remove another one
 393  * (The another one can be a SYN_RECV or TIMEWAIT
 394  */
 395 bool inet_ehash_insert(struct sock *sk, struct sock *osk)
 396 {
 397         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 398         struct hlist_nulls_head *list;
 399         struct inet_ehash_bucket *head;
 400         spinlock_t *lock;
 401         bool ret = true;
 402
 403         WARN_ON_ONCE(!sk_unhashed(sk));
 404
 405         sk->sk_hash = sk_ehashfn(sk);
 406         head = inet_ehash_bucket(hashinfo, sk->sk_hash);
 407         list = &head->chain;
 408         lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 409
 410         spin_lock(lock);
 411         if (osk) {
 412                 WARN_ON_ONCE(sk->sk_hash != osk->sk_hash);
 413                 ret = sk_nulls_del_node_init_rcu(osk);
 414         }
 415         if (ret)
 416                 __sk_nulls_add_node_rcu(sk, list);
 417         spin_unlock(lock);
 418         return ret;
 419 }
 420
 421 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk)
 422 {
 423         bool ok = inet_ehash_insert(sk, osk);
 424
 425         if (ok) {
 426                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 427         } else {
 428                 percpu_counter_inc(sk->sk_prot->orphan_count);
 429                 sk->sk_state = TCP_CLOSE;
 430                 sock_set_flag(sk, SOCK_DEAD);
 431                 inet_csk_destroy_sock(sk);
 432         }
 433         return ok;
 434 }
 435 EXPORT_SYMBOL_GPL(inet_ehash_nolisten);
 436
 437 static int inet_reuseport_add_sock(struct sock *sk,
 438                                    struct inet_listen_hashbucket *ilb,
 439                                    int (*saddr_same)(const struct sock *sk1,
 440                                                      const struct sock *sk2,
 441                                                      bool match_wildcard))
 442 {
 443         struct inet_bind_bucket *tb = inet_csk(sk)->icsk_bind_hash;
 444         struct sock *sk2;
 445         kuid_t uid = sock_i_uid(sk);
 446
 447         sk_for_each_rcu(sk2, &ilb->head) {
 448                 if (sk2 != sk &&
 449                     sk2->sk_family == sk->sk_family &&
 450                     ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 451                     sk2->sk_bound_dev_if == sk->sk_bound_dev_if &&
 452                     inet_csk(sk2)->icsk_bind_hash == tb &&
 453                     sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
 454                     saddr_same(sk, sk2, false))
 455                         return reuseport_add_sock(sk, sk2);
 456         }
 457
 458         return reuseport_alloc(sk);
 459 }
 460
 461 int __inet_hash(struct sock *sk, struct sock *osk,
 462                  int (*saddr_same)(const struct sock *sk1,
 463                                    const struct sock *sk2,
 464                                    bool match_wildcard))
 465 {
 466         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 467         struct inet_listen_hashbucket *ilb;
 468         int err = 0;
 469
 470         if (sk->sk_state != TCP_LISTEN) {
 471                 inet_ehash_nolisten(sk, osk);
 472                 return 0;
 473         }
 474         WARN_ON(!sk_unhashed(sk));
 475         ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
 476
 477         spin_lock(&ilb->lock);
 478         if (sk->sk_reuseport) {
 479                 err = inet_reuseport_add_sock(sk, ilb, saddr_same);
 480                 if (err)
 481                         goto unlock;
 482         }
 483         if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 484                 sk->sk_family == AF_INET6)
 485                 hlist_add_tail_rcu(&sk->sk_node, &ilb->head);
 486         else
 487                 hlist_add_head_rcu(&sk->sk_node, &ilb->head);
 488         sock_set_flag(sk, SOCK_RCU_FREE);
 489         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 490 unlock:
 491         spin_unlock(&ilb->lock);
 492
 493         return err;
 494 }
 495 EXPORT_SYMBOL(__inet_hash);
 496
 497 int inet_hash(struct sock *sk)
 498 {
 499         int err = 0;
 500
 501         if (sk->sk_state != TCP_CLOSE) {
 502                 local_bh_disable();
 503                 err = __inet_hash(sk, NULL, ipv4_rcv_saddr_equal);
 504                 local_bh_enable();
 505         }
 506
 507         return err;
 508 }
 509 EXPORT_SYMBOL_GPL(inet_hash);
 510
 511 void inet_unhash(struct sock *sk)
 512 {
 513         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 514         spinlock_t *lock;
 515         bool listener = false;
 516         int done;
 517
 518         if (sk_unhashed(sk))
 519                 return;
 520
 521         if (sk->sk_state == TCP_LISTEN) {
 522                 lock = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)].lock;
 523                 listener = true;
 524         } else {
 525                 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 526         }
 527         spin_lock_bh(lock);
 528         if (rcu_access_pointer(sk->sk_reuseport_cb))
 529                 reuseport_detach_sock(sk);
 530         if (listener)
 531                 done = __sk_del_node_init(sk);
 532         else
 533                 done = __sk_nulls_del_node_init_rcu(sk);
 534         if (done)
 535                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 536         spin_unlock_bh(lock);
 537 }
 538 EXPORT_SYMBOL_GPL(inet_unhash);
 539
 540 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 541                 struct sock *sk, u32 port_offset,
 542                 int (*check_established)(struct inet_timewait_death_row *,
 543                         struct sock *, __u16, struct inet_timewait_sock **))
 544 {
 545         struct inet_hashinfo *hinfo = death_row->hashinfo;
 546         struct inet_timewait_sock *tw = NULL;
 547         struct inet_bind_hashbucket *head;
 548         int port = inet_sk(sk)->inet_num;
 549         struct net *net = sock_net(sk);
 550         struct inet_bind_bucket *tb;
 551         u32 remaining, offset;
 552         int ret, i, low, high;
 553         static u32 hint;
 554
 555         if (port) {
 556                 head = &hinfo->bhash[inet_bhashfn(net, port,
 557                                                   hinfo->bhash_size)];
 558                 tb = inet_csk(sk)->icsk_bind_hash;
 559                 spin_lock_bh(&head->lock);
 560                 if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
 561                         inet_ehash_nolisten(sk, NULL);
 562                         spin_unlock_bh(&head->lock);
 563                         return 0;
 564                 }
 565                 spin_unlock(&head->lock);
 566                 /* No definite answer... Walk to established hash table */
 567                 ret = check_established(death_row, sk, port, NULL);
 568                 local_bh_enable();
 569                 return ret;
 570         }
 571
 572         inet_get_local_port_range(net, &low, &high);
 573         high++; /* [32768, 60999] -> [32768, 61000[ */
 574         remaining = high - low;
 575         if (likely(remaining > 1))
 576                 remaining &= ~1U;
 577
 578         offset = (hint + port_offset) % remaining;
 579         /* In first pass we try ports of @low parity.
 580          * inet_csk_get_port() does the opposite choice.
 581          */
 582         offset &= ~1U;
 583 other_parity_scan:
 584         port = low + offset;
 585         for (i = 0; i < remaining; i += 2, port += 2) {
 586                 if (unlikely(port >= high))
 587                         port -= remaining;
 588                 if (inet_is_local_reserved_port(net, port))
 589                         continue;
 590                 head = &hinfo->bhash[inet_bhashfn(net, port,
 591                                                   hinfo->bhash_size)];
 592                 spin_lock_bh(&head->lock);
 593
 594                 /* Does not bother with rcv_saddr checks, because
 595                  * the established check is already unique enough.
 596                  */
 597                 inet_bind_bucket_for_each(tb, &head->chain) {
 598                         if (net_eq(ib_net(tb), net) && tb->port == port) {
 599                                 if (tb->fastreuse >= 0 ||
 600                                     tb->fastreuseport >= 0)
 601                                         goto next_port;
 602                                 WARN_ON(hlist_empty(&tb->owners));
 603                                 if (!check_established(death_row, sk,
 604                                                        port, &tw))
 605                                         goto ok;
 606                                 goto next_port;
 607                         }
 608                 }
 609
 610                 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 611                                              net, head, port);
 612                 if (!tb) {
 613                         spin_unlock_bh(&head->lock);
 614                         return -ENOMEM;
 615                 }
 616                 tb->fastreuse = -1;
 617                 tb->fastreuseport = -1;
 618                 goto ok;
 619 next_port:
 620                 spin_unlock_bh(&head->lock);
 621                 cond_resched();
 622         }
 623
 624         offset++;
 625         if ((offset & 1) && remaining > 1)
 626                 goto other_parity_scan;
 627
 628         return -EADDRNOTAVAIL;
 629
 630 ok:
 631         hint += i + 2;
 632
 633         /* Head lock still held and bh's disabled */
 634         inet_bind_hash(sk, tb, port);
 635         if (sk_unhashed(sk)) {
 636                 inet_sk(sk)->inet_sport = htons(port);
 637                 inet_ehash_nolisten(sk, (struct sock *)tw);
 638         }
 639         if (tw)
 640                 inet_twsk_bind_unhash(tw, hinfo);
 641         spin_unlock(&head->lock);
 642         if (tw)
 643                 inet_twsk_deschedule_put(tw);
 644         local_bh_enable();
 645         return 0;
 646 }
 647
 648 /*
 649  * Bind a port for a connect operation and hash it.
 650  */
 651 int inet_hash_connect(struct inet_timewait_death_row *death_row,
 652                       struct sock *sk)
 653 {
 654         u32 port_offset = 0;
 655
 656         if (!inet_sk(sk)->inet_num)
 657                 port_offset = inet_sk_port_offset(sk);
 658         return __inet_hash_connect(death_row, sk, port_offset,
 659                                    __inet_check_established);
 660 }
 661 EXPORT_SYMBOL_GPL(inet_hash_connect);
 662
 663 void inet_hashinfo_init(struct inet_hashinfo *h)
 664 {
 665         int i;
 666
 667         for (i = 0; i < INET_LHTABLE_SIZE; i++) {
 668                 spin_lock_init(&h->listening_hash[i].lock);
 669                 INIT_HLIST_HEAD(&h->listening_hash[i].head);
 670         }
 671 }
 672 EXPORT_SYMBOL_GPL(inet_hashinfo_init);
 673
 674 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
 675 {
 676         unsigned int locksz = sizeof(spinlock_t);
 677         unsigned int i, nblocks = 1;
 678
 679         if (locksz != 0) {
 680                 /* allocate 2 cache lines or at least one spinlock per cpu */
 681                 nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U);
 682                 nblocks = roundup_pow_of_two(nblocks * num_possible_cpus());
 683
 684                 /* no more locks than number of hash buckets */
 685                 nblocks = min(nblocks, hashinfo->ehash_mask + 1);
 686
 687                 hashinfo->ehash_locks = kmalloc_array(nblocks, locksz,
 688                                                       GFP_KERNEL | __GFP_NOWARN);
 689                 if (!hashinfo->ehash_locks)
 690                         hashinfo->ehash_locks = vmalloc(nblocks * locksz);
 691
 692                 if (!hashinfo->ehash_locks)
 693                         return -ENOMEM;
 694
 695                 for (i = 0; i < nblocks; i++)
 696                         spin_lock_init(&hashinfo->ehash_locks[i]);
 697         }
 698         hashinfo->ehash_locks_mask = nblocks - 1;
 699         return 0;
 700 }
 701 EXPORT_SYMBOL_GPL(inet_ehash_locks_alloc);