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/sock_reuseport.h>
  29
  30 static u32 inet_ehashfn(const struct net *net, const __be32 laddr,
  31                         const __u16 lport, const __be32 faddr,
  32                         const __be16 fport)
  33 {
  34         static u32 inet_ehash_secret __read_mostly;
  35
  36         net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret));
  37
  38         return __inet_ehashfn(laddr, lport, faddr, fport,
  39                               inet_ehash_secret + net_hash_mix(net));
  40 }
  41
  42 /* This function handles inet_sock, but also timewait and request sockets
  43  * for IPv4/IPv6.
  44  */
  45 u32 sk_ehashfn(const struct sock *sk)
  46 {
  47 #if IS_ENABLED(CONFIG_IPV6)
  48         if (sk->sk_family == AF_INET6 &&
  49             !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
  50                 return inet6_ehashfn(sock_net(sk),
  51                                      &sk->sk_v6_rcv_saddr, sk->sk_num,
  52                                      &sk->sk_v6_daddr, sk->sk_dport);
  53 #endif
  54         return inet_ehashfn(sock_net(sk),
  55                             sk->sk_rcv_saddr, sk->sk_num,
  56                             sk->sk_daddr, sk->sk_dport);
  57 }
  58
  59 /*
  60  * Allocate and initialize a new local port bind bucket.
  61  * The bindhash mutex for snum's hash chain must be held here.
  62  */
  63 struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
  64                                                  struct net *net,
  65                                                  struct inet_bind_hashbucket *head,
  66                                                  const unsigned short snum)
  67 {
  68         struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
  69
  70         if (tb) {
  71                 write_pnet(&tb->ib_net, net);
  72                 tb->port      = snum;
  73                 tb->fastreuse = 0;
  74                 tb->fastreuseport = 0;
  75                 tb->num_owners = 0;
  76                 INIT_HLIST_HEAD(&tb->owners);
  77                 hlist_add_head(&tb->node, &head->chain);
  78         }
  79         return tb;
  80 }
  81
  82 /*
  83  * Caller must hold hashbucket lock for this tb with local BH disabled
  84  */
  85 void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
  86 {
  87         if (hlist_empty(&tb->owners)) {
  88                 __hlist_del(&tb->node);
  89                 kmem_cache_free(cachep, tb);
  90         }
  91 }
  92
  93 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
  94                     const unsigned short snum)
  95 {
  96         inet_sk(sk)->inet_num = snum;
  97         sk_add_bind_node(sk, &tb->owners);
  98         tb->num_owners++;
  99         inet_csk(sk)->icsk_bind_hash = tb;
 100 }
 101
 102 /*
 103  * Get rid of any references to a local port held by the given sock.
 104  */
 105 static void __inet_put_port(struct sock *sk)
 106 {
 107         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 108         const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num,
 109                         hashinfo->bhash_size);
 110         struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
 111         struct inet_bind_bucket *tb;
 112
 113         spin_lock(&head->lock);
 114         tb = inet_csk(sk)->icsk_bind_hash;
 115         __sk_del_bind_node(sk);
 116         tb->num_owners--;
 117         inet_csk(sk)->icsk_bind_hash = NULL;
 118         inet_sk(sk)->inet_num = 0;
 119         inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
 120         spin_unlock(&head->lock);
 121 }
 122
 123 void inet_put_port(struct sock *sk)
 124 {
 125         local_bh_disable();
 126         __inet_put_port(sk);
 127         local_bh_enable();
 128 }
 129 EXPORT_SYMBOL(inet_put_port);
 130
 131 int __inet_inherit_port(const struct sock *sk, struct sock *child)
 132 {
 133         struct inet_hashinfo *table = sk->sk_prot->h.hashinfo;
 134         unsigned short port = inet_sk(child)->inet_num;
 135         const int bhash = inet_bhashfn(sock_net(sk), port,
 136                         table->bhash_size);
 137         struct inet_bind_hashbucket *head = &table->bhash[bhash];
 138         struct inet_bind_bucket *tb;
 139
 140         spin_lock(&head->lock);
 141         tb = inet_csk(sk)->icsk_bind_hash;
 142         if (unlikely(!tb)) {
 143                 spin_unlock(&head->lock);
 144                 return -ENOENT;
 145         }
 146         if (tb->port != port) {
 147                 /* NOTE: using tproxy and redirecting skbs to a proxy
 148                  * on a different listener port breaks the assumption
 149                  * that the listener socket's icsk_bind_hash is the same
 150                  * as that of the child socket. We have to look up or
 151                  * create a new bind bucket for the child here. */
 152                 inet_bind_bucket_for_each(tb, &head->chain) {
 153                         if (net_eq(ib_net(tb), sock_net(sk)) &&
 154                             tb->port == port)
 155                                 break;
 156                 }
 157                 if (!tb) {
 158                         tb = inet_bind_bucket_create(table->bind_bucket_cachep,
 159                                                      sock_net(sk), head, port);
 160                         if (!tb) {
 161                                 spin_unlock(&head->lock);
 162                                 return -ENOMEM;
 163                         }
 164                 }
 165         }
 166         inet_bind_hash(child, tb, port);
 167         spin_unlock(&head->lock);
 168
 169         return 0;
 170 }
 171 EXPORT_SYMBOL_GPL(__inet_inherit_port);
 172
 173 static inline int compute_score(struct sock *sk, struct net *net,
 174                                 const unsigned short hnum, const __be32 daddr,
 175                                 const int dif)
 176 {
 177         int score = -1;
 178         struct inet_sock *inet = inet_sk(sk);
 179
 180         if (net_eq(sock_net(sk), net) && inet->inet_num == hnum &&
 181                         !ipv6_only_sock(sk)) {
 182                 __be32 rcv_saddr = inet->inet_rcv_saddr;
 183                 score = sk->sk_family == PF_INET ? 2 : 1;
 184                 if (rcv_saddr) {
 185                         if (rcv_saddr != daddr)
 186                                 return -1;
 187                         score += 4;
 188                 }
 189                 if (sk->sk_bound_dev_if) {
 190                         if (sk->sk_bound_dev_if != dif)
 191                                 return -1;
 192                         score += 4;
 193                 }
 194                 if (sk->sk_incoming_cpu == raw_smp_processor_id())
 195                         score++;
 196         }
 197         return score;
 198 }
 199
 200 /*
 201  * Don't inline this cruft. Here are some nice properties to exploit here. The
 202  * BSD API does not allow a listening sock to specify the remote port nor the
 203  * remote address for the connection. So always assume those are both
 204  * wildcarded during the search since they can never be otherwise.
 205  */
 206
 207
 208 struct sock *__inet_lookup_listener(struct net *net,
 209                                     struct inet_hashinfo *hashinfo,
 210                                     struct sk_buff *skb, int doff,
 211                                     const __be32 saddr, __be16 sport,
 212                                     const __be32 daddr, const unsigned short hnum,
 213                                     const int dif)
 214 {
 215         struct sock *sk, *result;
 216         struct hlist_nulls_node *node;
 217         unsigned int hash = inet_lhashfn(net, hnum);
 218         struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
 219         int score, hiscore, matches = 0, reuseport = 0;
 220         bool select_ok = true;
 221         u32 phash = 0;
 222
 223         rcu_read_lock();
 224 begin:
 225         result = NULL;
 226         hiscore = 0;
 227         sk_nulls_for_each_rcu(sk, node, &ilb->head) {
 228                 score = compute_score(sk, net, hnum, daddr, dif);
 229                 if (score > hiscore) {
 230                         result = sk;
 231                         hiscore = score;
 232                         reuseport = sk->sk_reuseport;
 233                         if (reuseport) {
 234                                 phash = inet_ehashfn(net, daddr, hnum,
 235                                                      saddr, sport);
 236                                 if (select_ok) {
 237                                         struct sock *sk2;
 238                                         sk2 = reuseport_select_sock(sk, phash,
 239                                                                     skb, doff);
 240                                         if (sk2) {
 241                                                 result = sk2;
 242                                                 goto found;
 243                                         }
 244                                 }
 245                                 matches = 1;
 246                         }
 247                 } else if (score == hiscore && reuseport) {
 248                         matches++;
 249                         if (reciprocal_scale(phash, matches) == 0)
 250                                 result = sk;
 251                         phash = next_pseudo_random32(phash);
 252                 }
 253         }
 254         /*
 255          * if the nulls value we got at the end of this lookup is
 256          * not the expected one, we must restart lookup.
 257          * We probably met an item that was moved to another chain.
 258          */
 259         if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
 260                 goto begin;
 261         if (result) {
 262 found:
 263                 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
 264                         result = NULL;
 265                 else if (unlikely(compute_score(result, net, hnum, daddr,
 266                                   dif) < hiscore)) {
 267                         sock_put(result);
 268                         select_ok = false;
 269                         goto begin;
 270                 }
 271         }
 272         rcu_read_unlock();
 273         return result;
 274 }
 275 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
 276
 277 /* All sockets share common refcount, but have different destructors */
 278 void sock_gen_put(struct sock *sk)
 279 {
 280         if (!atomic_dec_and_test(&sk->sk_refcnt))
 281                 return;
 282
 283         if (sk->sk_state == TCP_TIME_WAIT)
 284                 inet_twsk_free(inet_twsk(sk));
 285         else if (sk->sk_state == TCP_NEW_SYN_RECV)
 286                 reqsk_free(inet_reqsk(sk));
 287         else
 288                 sk_free(sk);
 289 }
 290 EXPORT_SYMBOL_GPL(sock_gen_put);
 291
 292 void sock_edemux(struct sk_buff *skb)
 293 {
 294         sock_gen_put(skb->sk);
 295 }
 296 EXPORT_SYMBOL(sock_edemux);
 297
 298 struct sock *__inet_lookup_established(struct net *net,
 299                                   struct inet_hashinfo *hashinfo,
 300                                   const __be32 saddr, const __be16 sport,
 301                                   const __be32 daddr, const u16 hnum,
 302                                   const int dif)
 303 {
 304         INET_ADDR_COOKIE(acookie, saddr, daddr);
 305         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
 306         struct sock *sk;
 307         const struct hlist_nulls_node *node;
 308         /* Optimize here for direct hit, only listening connections can
 309          * have wildcards anyways.
 310          */
 311         unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
 312         unsigned int slot = hash & hashinfo->ehash_mask;
 313         struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
 314
 315         rcu_read_lock();
 316 begin:
 317         sk_nulls_for_each_rcu(sk, node, &head->chain) {
 318                 if (sk->sk_hash != hash)
 319                         continue;
 320                 if (likely(INET_MATCH(sk, net, acookie,
 321                                       saddr, daddr, ports, dif))) {
 322                         if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
 323                                 goto out;
 324                         if (unlikely(!INET_MATCH(sk, net, acookie,
 325                                                  saddr, daddr, ports, dif))) {
 326                                 sock_gen_put(sk);
 327                                 goto begin;
 328                         }
 329                         goto found;
 330                 }
 331         }
 332         /*
 333          * if the nulls value we got at the end of this lookup is
 334          * not the expected one, we must restart lookup.
 335          * We probably met an item that was moved to another chain.
 336          */
 337         if (get_nulls_value(node) != slot)
 338                 goto begin;
 339 out:
 340         sk = NULL;
 341 found:
 342         rcu_read_unlock();
 343         return sk;
 344 }
 345 EXPORT_SYMBOL_GPL(__inet_lookup_established);
 346
 347 /* called with local bh disabled */
 348 static int __inet_check_established(struct inet_timewait_death_row *death_row,
 349                                     struct sock *sk, __u16 lport,
 350                                     struct inet_timewait_sock **twp)
 351 {
 352         struct inet_hashinfo *hinfo = death_row->hashinfo;
 353         struct inet_sock *inet = inet_sk(sk);
 354         __be32 daddr = inet->inet_rcv_saddr;
 355         __be32 saddr = inet->inet_daddr;
 356         int dif = sk->sk_bound_dev_if;
 357         INET_ADDR_COOKIE(acookie, saddr, daddr);
 358         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
 359         struct net *net = sock_net(sk);
 360         unsigned int hash = inet_ehashfn(net, daddr, lport,
 361                                          saddr, inet->inet_dport);
 362         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 363         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
 364         struct sock *sk2;
 365         const struct hlist_nulls_node *node;
 366         struct inet_timewait_sock *tw = NULL;
 367
 368         spin_lock(lock);
 369
 370         sk_nulls_for_each(sk2, node, &head->chain) {
 371                 if (sk2->sk_hash != hash)
 372                         continue;
 373
 374                 if (likely(INET_MATCH(sk2, net, acookie,
 375                                          saddr, daddr, ports, dif))) {
 376                         if (sk2->sk_state == TCP_TIME_WAIT) {
 377                                 tw = inet_twsk(sk2);
 378                                 if (twsk_unique(sk, sk2, twp))
 379                                         break;
 380                         }
 381                         goto not_unique;
 382                 }
 383         }
 384
 385         /* Must record num and sport now. Otherwise we will see
 386          * in hash table socket with a funny identity.
 387          */
 388         inet->inet_num = lport;
 389         inet->inet_sport = htons(lport);
 390         sk->sk_hash = hash;
 391         WARN_ON(!sk_unhashed(sk));
 392         __sk_nulls_add_node_rcu(sk, &head->chain);
 393         if (tw) {
 394                 sk_nulls_del_node_init_rcu((struct sock *)tw);
 395                 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
 396         }
 397         spin_unlock(lock);
 398         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 399
 400         if (twp) {
 401                 *twp = tw;
 402         } else if (tw) {
 403                 /* Silly. Should hash-dance instead... */
 404                 inet_twsk_deschedule_put(tw);
 405         }
 406         return 0;
 407
 408 not_unique:
 409         spin_unlock(lock);
 410         return -EADDRNOTAVAIL;
 411 }
 412
 413 static u32 inet_sk_port_offset(const struct sock *sk)
 414 {
 415         const struct inet_sock *inet = inet_sk(sk);
 416
 417         return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
 418                                           inet->inet_daddr,
 419                                           inet->inet_dport);
 420 }
 421
 422 /* insert a socket into ehash, and eventually remove another one
 423  * (The another one can be a SYN_RECV or TIMEWAIT
 424  */
 425 bool inet_ehash_insert(struct sock *sk, struct sock *osk)
 426 {
 427         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 428         struct hlist_nulls_head *list;
 429         struct inet_ehash_bucket *head;
 430         spinlock_t *lock;
 431         bool ret = true;
 432
 433         WARN_ON_ONCE(!sk_unhashed(sk));
 434
 435         sk->sk_hash = sk_ehashfn(sk);
 436         head = inet_ehash_bucket(hashinfo, sk->sk_hash);
 437         list = &head->chain;
 438         lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 439
 440         spin_lock(lock);
 441         if (osk) {
 442                 WARN_ON_ONCE(sk->sk_hash != osk->sk_hash);
 443                 ret = sk_nulls_del_node_init_rcu(osk);
 444         }
 445         if (ret)
 446                 __sk_nulls_add_node_rcu(sk, list);
 447         spin_unlock(lock);
 448         return ret;
 449 }
 450
 451 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk)
 452 {
 453         bool ok = inet_ehash_insert(sk, osk);
 454
 455         if (ok) {
 456                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 457         } else {
 458                 percpu_counter_inc(sk->sk_prot->orphan_count);
 459                 sk->sk_state = TCP_CLOSE;
 460                 sock_set_flag(sk, SOCK_DEAD);
 461                 inet_csk_destroy_sock(sk);
 462         }
 463         return ok;
 464 }
 465 EXPORT_SYMBOL_GPL(inet_ehash_nolisten);
 466
 467 static int inet_reuseport_add_sock(struct sock *sk,
 468                                    struct inet_listen_hashbucket *ilb,
 469                                    int (*saddr_same)(const struct sock *sk1,
 470                                                      const struct sock *sk2,
 471                                                      bool match_wildcard))
 472 {
 473         struct sock *sk2;
 474         struct hlist_nulls_node *node;
 475         kuid_t uid = sock_i_uid(sk);
 476
 477         sk_nulls_for_each_rcu(sk2, node, &ilb->head) {
 478                 if (sk2 != sk &&
 479                     sk2->sk_family == sk->sk_family &&
 480                     ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 481                     sk2->sk_bound_dev_if == sk->sk_bound_dev_if &&
 482                     sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
 483                     saddr_same(sk, sk2, false))
 484                         return reuseport_add_sock(sk, sk2);
 485         }
 486
 487         /* Initial allocation may have already happened via setsockopt */
 488         if (!rcu_access_pointer(sk->sk_reuseport_cb))
 489                 return reuseport_alloc(sk);
 490         return 0;
 491 }
 492
 493 int __inet_hash(struct sock *sk, struct sock *osk,
 494                  int (*saddr_same)(const struct sock *sk1,
 495                                    const struct sock *sk2,
 496                                    bool match_wildcard))
 497 {
 498         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 499         struct inet_listen_hashbucket *ilb;
 500         int err = 0;
 501
 502         if (sk->sk_state != TCP_LISTEN) {
 503                 inet_ehash_nolisten(sk, osk);
 504                 return 0;
 505         }
 506         WARN_ON(!sk_unhashed(sk));
 507         ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
 508
 509         spin_lock(&ilb->lock);
 510         if (sk->sk_reuseport) {
 511                 err = inet_reuseport_add_sock(sk, ilb, saddr_same);
 512                 if (err)
 513                         goto unlock;
 514         }
 515         __sk_nulls_add_node_rcu(sk, &ilb->head);
 516         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 517 unlock:
 518         spin_unlock(&ilb->lock);
 519
 520         return err;
 521 }
 522 EXPORT_SYMBOL(__inet_hash);
 523
 524 int inet_hash(struct sock *sk)
 525 {
 526         int err = 0;
 527
 528         if (sk->sk_state != TCP_CLOSE) {
 529                 local_bh_disable();
 530                 err = __inet_hash(sk, NULL, ipv4_rcv_saddr_equal);
 531                 local_bh_enable();
 532         }
 533
 534         return err;
 535 }
 536 EXPORT_SYMBOL_GPL(inet_hash);
 537
 538 void inet_unhash(struct sock *sk)
 539 {
 540         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 541         spinlock_t *lock;
 542         int done;
 543
 544         if (sk_unhashed(sk))
 545                 return;
 546
 547         if (sk->sk_state == TCP_LISTEN)
 548                 lock = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)].lock;
 549         else
 550                 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 551
 552         spin_lock_bh(lock);
 553         if (rcu_access_pointer(sk->sk_reuseport_cb))
 554                 reuseport_detach_sock(sk);
 555         done = __sk_nulls_del_node_init_rcu(sk);
 556         if (done)
 557                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 558         spin_unlock_bh(lock);
 559 }
 560 EXPORT_SYMBOL_GPL(inet_unhash);
 561
 562 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 563                 struct sock *sk, u32 port_offset,
 564                 int (*check_established)(struct inet_timewait_death_row *,
 565                         struct sock *, __u16, struct inet_timewait_sock **))
 566 {
 567         struct inet_hashinfo *hinfo = death_row->hashinfo;
 568         struct inet_timewait_sock *tw = NULL;
 569         struct inet_bind_hashbucket *head;
 570         int port = inet_sk(sk)->inet_num;
 571         struct net *net = sock_net(sk);
 572         struct inet_bind_bucket *tb;
 573         u32 remaining, offset;
 574         int ret, i, low, high;
 575         static u32 hint;
 576
 577         if (port) {
 578                 head = &hinfo->bhash[inet_bhashfn(net, port,
 579                                                   hinfo->bhash_size)];
 580                 tb = inet_csk(sk)->icsk_bind_hash;
 581                 spin_lock_bh(&head->lock);
 582                 if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
 583                         inet_ehash_nolisten(sk, NULL);
 584                         spin_unlock_bh(&head->lock);
 585                         return 0;
 586                 }
 587                 spin_unlock(&head->lock);
 588                 /* No definite answer... Walk to established hash table */
 589                 ret = check_established(death_row, sk, port, NULL);
 590                 local_bh_enable();
 591                 return ret;
 592         }
 593
 594         inet_get_local_port_range(net, &low, &high);
 595         high++; /* [32768, 60999] -> [32768, 61000[ */
 596         remaining = high - low;
 597         if (likely(remaining > 1))
 598                 remaining &= ~1U;
 599
 600         offset = (hint + port_offset) % remaining;
 601         /* In first pass we try ports of @low parity.
 602          * inet_csk_get_port() does the opposite choice.
 603          */
 604         offset &= ~1U;
 605 other_parity_scan:
 606         port = low + offset;
 607         for (i = 0; i < remaining; i += 2, port += 2) {
 608                 if (unlikely(port >= high))
 609                         port -= remaining;
 610                 if (inet_is_local_reserved_port(net, port))
 611                         continue;
 612                 head = &hinfo->bhash[inet_bhashfn(net, port,
 613                                                   hinfo->bhash_size)];
 614                 spin_lock_bh(&head->lock);
 615
 616                 /* Does not bother with rcv_saddr checks, because
 617                  * the established check is already unique enough.
 618                  */
 619                 inet_bind_bucket_for_each(tb, &head->chain) {
 620                         if (net_eq(ib_net(tb), net) && tb->port == port) {
 621                                 if (tb->fastreuse >= 0 ||
 622                                     tb->fastreuseport >= 0)
 623                                         goto next_port;
 624                                 WARN_ON(hlist_empty(&tb->owners));
 625                                 if (!check_established(death_row, sk,
 626                                                        port, &tw))
 627                                         goto ok;
 628                                 goto next_port;
 629                         }
 630                 }
 631
 632                 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 633                                              net, head, port);
 634                 if (!tb) {
 635                         spin_unlock_bh(&head->lock);
 636                         return -ENOMEM;
 637                 }
 638                 tb->fastreuse = -1;
 639                 tb->fastreuseport = -1;
 640                 goto ok;
 641 next_port:
 642                 spin_unlock_bh(&head->lock);
 643                 cond_resched();
 644         }
 645
 646         offset++;
 647         if ((offset & 1) && remaining > 1)
 648                 goto other_parity_scan;
 649
 650         return -EADDRNOTAVAIL;
 651
 652 ok:
 653         hint += i + 2;
 654
 655         /* Head lock still held and bh's disabled */
 656         inet_bind_hash(sk, tb, port);
 657         if (sk_unhashed(sk)) {
 658                 inet_sk(sk)->inet_sport = htons(port);
 659                 inet_ehash_nolisten(sk, (struct sock *)tw);
 660         }
 661         if (tw)
 662                 inet_twsk_bind_unhash(tw, hinfo);
 663         spin_unlock(&head->lock);
 664         if (tw)
 665                 inet_twsk_deschedule_put(tw);
 666         local_bh_enable();
 667         return 0;
 668 }
 669
 670 /*
 671  * Bind a port for a connect operation and hash it.
 672  */
 673 int inet_hash_connect(struct inet_timewait_death_row *death_row,
 674                       struct sock *sk)
 675 {
 676         u32 port_offset = 0;
 677
 678         if (!inet_sk(sk)->inet_num)
 679                 port_offset = inet_sk_port_offset(sk);
 680         return __inet_hash_connect(death_row, sk, port_offset,
 681                                    __inet_check_established);
 682 }
 683 EXPORT_SYMBOL_GPL(inet_hash_connect);
 684
 685 void inet_hashinfo_init(struct inet_hashinfo *h)
 686 {
 687         int i;
 688
 689         for (i = 0; i < INET_LHTABLE_SIZE; i++) {
 690                 spin_lock_init(&h->listening_hash[i].lock);
 691                 INIT_HLIST_NULLS_HEAD(&h->listening_hash[i].head,
 692                                       i + LISTENING_NULLS_BASE);
 693                 }
 694 }
 695 EXPORT_SYMBOL_GPL(inet_hashinfo_init);
 696
 697 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
 698 {
 699         unsigned int locksz = sizeof(spinlock_t);
 700         unsigned int i, nblocks = 1;
 701
 702         if (locksz != 0) {
 703                 /* allocate 2 cache lines or at least one spinlock per cpu */
 704                 nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U);
 705                 nblocks = roundup_pow_of_two(nblocks * num_possible_cpus());
 706
 707                 /* no more locks than number of hash buckets */
 708                 nblocks = min(nblocks, hashinfo->ehash_mask + 1);
 709
 710                 hashinfo->ehash_locks = kmalloc_array(nblocks, locksz,
 711                                                       GFP_KERNEL | __GFP_NOWARN);
 712                 if (!hashinfo->ehash_locks)
 713                         hashinfo->ehash_locks = vmalloc(nblocks * locksz);
 714
 715                 if (!hashinfo->ehash_locks)
 716                         return -ENOMEM;
 717
 718                 for (i = 0; i < nblocks; i++)
 719                         spin_lock_init(&hashinfo->ehash_locks[i]);
 720         }
 721         hashinfo->ehash_locks_mask = nblocks - 1;
 722         return 0;
 723 }
 724 EXPORT_SYMBOL_GPL(inet_ehash_locks_alloc);