net/ipv4/inet_hashtables.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              Generic INET transport hashtables
   8  *
   9  * Authors:     Lotsa people, from code originally in tcp
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/random.h>
  14 #include <linux/sched.h>
  15 #include <linux/slab.h>
  16 #include <linux/wait.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/memblock.h>
  19
  20 #include <net/addrconf.h>
  21 #include <net/inet_connection_sock.h>
  22 #include <net/inet_hashtables.h>
  23 #if IS_ENABLED(CONFIG_IPV6)
  24 #include <net/inet6_hashtables.h>
  25 #endif
  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 static 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                                                  int l3mdev)
  69 {
  70         struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
  71
  72         if (tb) {
  73                 write_pnet(&tb->ib_net, net);
  74                 tb->l3mdev    = l3mdev;
  75                 tb->port      = snum;
  76                 tb->fastreuse = 0;
  77                 tb->fastreuseport = 0;
  78                 INIT_HLIST_HEAD(&tb->owners);
  79                 hlist_add_head(&tb->node, &head->chain);
  80         }
  81         return tb;
  82 }
  83
  84 /*
  85  * Caller must hold hashbucket lock for this tb with local BH disabled
  86  */
  87 void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
  88 {
  89         if (hlist_empty(&tb->owners)) {
  90                 __hlist_del(&tb->node);
  91                 kmem_cache_free(cachep, tb);
  92         }
  93 }
  94
  95 bool inet_bind_bucket_match(const struct inet_bind_bucket *tb, const struct net *net,
  96                             unsigned short port, int l3mdev)
  97 {
  98         return net_eq(ib_net(tb), net) && tb->port == port &&
  99                 tb->l3mdev == l3mdev;
 100 }
 101
 102 static void inet_bind2_bucket_init(struct inet_bind2_bucket *tb,
 103                                    struct net *net,
 104                                    struct inet_bind_hashbucket *head,
 105                                    unsigned short port, int l3mdev,
 106                                    const struct sock *sk)
 107 {
 108         write_pnet(&tb->ib_net, net);
 109         tb->l3mdev    = l3mdev;
 110         tb->port      = port;
 111 #if IS_ENABLED(CONFIG_IPV6)
 112         tb->family    = sk->sk_family;
 113         if (sk->sk_family == AF_INET6)
 114                 tb->v6_rcv_saddr = sk->sk_v6_rcv_saddr;
 115         else
 116 #endif
 117                 tb->rcv_saddr = sk->sk_rcv_saddr;
 118         INIT_HLIST_HEAD(&tb->owners);
 119         hlist_add_head(&tb->node, &head->chain);
 120 }
 121
 122 struct inet_bind2_bucket *inet_bind2_bucket_create(struct kmem_cache *cachep,
 123                                                    struct net *net,
 124                                                    struct inet_bind_hashbucket *head,
 125                                                    unsigned short port,
 126                                                    int l3mdev,
 127                                                    const struct sock *sk)
 128 {
 129         struct inet_bind2_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
 130
 131         if (tb)
 132                 inet_bind2_bucket_init(tb, net, head, port, l3mdev, sk);
 133
 134         return tb;
 135 }
 136
 137 /* Caller must hold hashbucket lock for this tb with local BH disabled */
 138 void inet_bind2_bucket_destroy(struct kmem_cache *cachep, struct inet_bind2_bucket *tb)
 139 {
 140         if (hlist_empty(&tb->owners)) {
 141                 __hlist_del(&tb->node);
 142                 kmem_cache_free(cachep, tb);
 143         }
 144 }
 145
 146 static bool inet_bind2_bucket_addr_match(const struct inet_bind2_bucket *tb2,
 147                                          const struct sock *sk)
 148 {
 149 #if IS_ENABLED(CONFIG_IPV6)
 150         if (sk->sk_family != tb2->family)
 151                 return false;
 152
 153         if (sk->sk_family == AF_INET6)
 154                 return ipv6_addr_equal(&tb2->v6_rcv_saddr,
 155                                        &sk->sk_v6_rcv_saddr);
 156 #endif
 157         return tb2->rcv_saddr == sk->sk_rcv_saddr;
 158 }
 159
 160 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
 161                     struct inet_bind2_bucket *tb2, unsigned short port)
 162 {
 163         inet_sk(sk)->inet_num = port;
 164         sk_add_bind_node(sk, &tb->owners);
 165         inet_csk(sk)->icsk_bind_hash = tb;
 166         sk_add_bind2_node(sk, &tb2->owners);
 167         inet_csk(sk)->icsk_bind2_hash = tb2;
 168 }
 169
 170 /*
 171  * Get rid of any references to a local port held by the given sock.
 172  */
 173 static void __inet_put_port(struct sock *sk)
 174 {
 175         struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
 176         struct inet_bind_hashbucket *head, *head2;
 177         struct net *net = sock_net(sk);
 178         struct inet_bind_bucket *tb;
 179         int bhash;
 180
 181         bhash = inet_bhashfn(net, inet_sk(sk)->inet_num, hashinfo->bhash_size);
 182         head = &hashinfo->bhash[bhash];
 183         head2 = inet_bhashfn_portaddr(hashinfo, sk, net, inet_sk(sk)->inet_num);
 184
 185         spin_lock(&head->lock);
 186         tb = inet_csk(sk)->icsk_bind_hash;
 187         __sk_del_bind_node(sk);
 188         inet_csk(sk)->icsk_bind_hash = NULL;
 189         inet_sk(sk)->inet_num = 0;
 190         inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
 191
 192         spin_lock(&head2->lock);
 193         if (inet_csk(sk)->icsk_bind2_hash) {
 194                 struct inet_bind2_bucket *tb2 = inet_csk(sk)->icsk_bind2_hash;
 195
 196                 __sk_del_bind2_node(sk);
 197                 inet_csk(sk)->icsk_bind2_hash = NULL;
 198                 inet_bind2_bucket_destroy(hashinfo->bind2_bucket_cachep, tb2);
 199         }
 200         spin_unlock(&head2->lock);
 201
 202         spin_unlock(&head->lock);
 203 }
 204
 205 void inet_put_port(struct sock *sk)
 206 {
 207         local_bh_disable();
 208         __inet_put_port(sk);
 209         local_bh_enable();
 210 }
 211 EXPORT_SYMBOL(inet_put_port);
 212
 213 int __inet_inherit_port(const struct sock *sk, struct sock *child)
 214 {
 215         struct inet_hashinfo *table = tcp_or_dccp_get_hashinfo(sk);
 216         unsigned short port = inet_sk(child)->inet_num;
 217         struct inet_bind_hashbucket *head, *head2;
 218         bool created_inet_bind_bucket = false;
 219         struct net *net = sock_net(sk);
 220         bool update_fastreuse = false;
 221         struct inet_bind2_bucket *tb2;
 222         struct inet_bind_bucket *tb;
 223         int bhash, l3mdev;
 224
 225         bhash = inet_bhashfn(net, port, table->bhash_size);
 226         head = &table->bhash[bhash];
 227         head2 = inet_bhashfn_portaddr(table, child, net, port);
 228
 229         spin_lock(&head->lock);
 230         spin_lock(&head2->lock);
 231         tb = inet_csk(sk)->icsk_bind_hash;
 232         tb2 = inet_csk(sk)->icsk_bind2_hash;
 233         if (unlikely(!tb || !tb2)) {
 234                 spin_unlock(&head2->lock);
 235                 spin_unlock(&head->lock);
 236                 return -ENOENT;
 237         }
 238         if (tb->port != port) {
 239                 l3mdev = inet_sk_bound_l3mdev(sk);
 240
 241                 /* NOTE: using tproxy and redirecting skbs to a proxy
 242                  * on a different listener port breaks the assumption
 243                  * that the listener socket's icsk_bind_hash is the same
 244                  * as that of the child socket. We have to look up or
 245                  * create a new bind bucket for the child here. */
 246                 inet_bind_bucket_for_each(tb, &head->chain) {
 247                         if (inet_bind_bucket_match(tb, net, port, l3mdev))
 248                                 break;
 249                 }
 250                 if (!tb) {
 251                         tb = inet_bind_bucket_create(table->bind_bucket_cachep,
 252                                                      net, head, port, l3mdev);
 253                         if (!tb) {
 254                                 spin_unlock(&head2->lock);
 255                                 spin_unlock(&head->lock);
 256                                 return -ENOMEM;
 257                         }
 258                         created_inet_bind_bucket = true;
 259                 }
 260                 update_fastreuse = true;
 261
 262                 goto bhash2_find;
 263         } else if (!inet_bind2_bucket_addr_match(tb2, child)) {
 264                 l3mdev = inet_sk_bound_l3mdev(sk);
 265
 266 bhash2_find:
 267                 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, child);
 268                 if (!tb2) {
 269                         tb2 = inet_bind2_bucket_create(table->bind2_bucket_cachep,
 270                                                        net, head2, port,
 271                                                        l3mdev, child);
 272                         if (!tb2)
 273                                 goto error;
 274                 }
 275         }
 276         if (update_fastreuse)
 277                 inet_csk_update_fastreuse(tb, child);
 278         inet_bind_hash(child, tb, tb2, port);
 279         spin_unlock(&head2->lock);
 280         spin_unlock(&head->lock);
 281
 282         return 0;
 283
 284 error:
 285         if (created_inet_bind_bucket)
 286                 inet_bind_bucket_destroy(table->bind_bucket_cachep, tb);
 287         spin_unlock(&head2->lock);
 288         spin_unlock(&head->lock);
 289         return -ENOMEM;
 290 }
 291 EXPORT_SYMBOL_GPL(__inet_inherit_port);
 292
 293 static struct inet_listen_hashbucket *
 294 inet_lhash2_bucket_sk(struct inet_hashinfo *h, struct sock *sk)
 295 {
 296         u32 hash;
 297
 298 #if IS_ENABLED(CONFIG_IPV6)
 299         if (sk->sk_family == AF_INET6)
 300                 hash = ipv6_portaddr_hash(sock_net(sk),
 301                                           &sk->sk_v6_rcv_saddr,
 302                                           inet_sk(sk)->inet_num);
 303         else
 304 #endif
 305                 hash = ipv4_portaddr_hash(sock_net(sk),
 306                                           inet_sk(sk)->inet_rcv_saddr,
 307                                           inet_sk(sk)->inet_num);
 308         return inet_lhash2_bucket(h, hash);
 309 }
 310
 311 static inline int compute_score(struct sock *sk, struct net *net,
 312                                 const unsigned short hnum, const __be32 daddr,
 313                                 const int dif, const int sdif)
 314 {
 315         int score = -1;
 316
 317         if (net_eq(sock_net(sk), net) && sk->sk_num == hnum &&
 318                         !ipv6_only_sock(sk)) {
 319                 if (sk->sk_rcv_saddr != daddr)
 320                         return -1;
 321
 322                 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
 323                         return -1;
 324                 score =  sk->sk_bound_dev_if ? 2 : 1;
 325
 326                 if (sk->sk_family == PF_INET)
 327                         score++;
 328                 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
 329                         score++;
 330         }
 331         return score;
 332 }
 333
 334 static inline struct sock *lookup_reuseport(struct net *net, struct sock *sk,
 335                                             struct sk_buff *skb, int doff,
 336                                             __be32 saddr, __be16 sport,
 337                                             __be32 daddr, unsigned short hnum)
 338 {
 339         struct sock *reuse_sk = NULL;
 340         u32 phash;
 341
 342         if (sk->sk_reuseport) {
 343                 phash = inet_ehashfn(net, daddr, hnum, saddr, sport);
 344                 reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
 345         }
 346         return reuse_sk;
 347 }
 348
 349 /*
 350  * Here are some nice properties to exploit here. The BSD API
 351  * does not allow a listening sock to specify the remote port nor the
 352  * remote address for the connection. So always assume those are both
 353  * wildcarded during the search since they can never be otherwise.
 354  */
 355
 356 /* called with rcu_read_lock() : No refcount taken on the socket */
 357 static struct sock *inet_lhash2_lookup(struct net *net,
 358                                 struct inet_listen_hashbucket *ilb2,
 359                                 struct sk_buff *skb, int doff,
 360                                 const __be32 saddr, __be16 sport,
 361                                 const __be32 daddr, const unsigned short hnum,
 362                                 const int dif, const int sdif)
 363 {
 364         struct sock *sk, *result = NULL;
 365         struct hlist_nulls_node *node;
 366         int score, hiscore = 0;
 367
 368         sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) {
 369                 score = compute_score(sk, net, hnum, daddr, dif, sdif);
 370                 if (score > hiscore) {
 371                         result = lookup_reuseport(net, sk, skb, doff,
 372                                                   saddr, sport, daddr, hnum);
 373                         if (result)
 374                                 return result;
 375
 376                         result = sk;
 377                         hiscore = score;
 378                 }
 379         }
 380
 381         return result;
 382 }
 383
 384 static inline struct sock *inet_lookup_run_bpf(struct net *net,
 385                                                struct inet_hashinfo *hashinfo,
 386                                                struct sk_buff *skb, int doff,
 387                                                __be32 saddr, __be16 sport,
 388                                                __be32 daddr, u16 hnum, const int dif)
 389 {
 390         struct sock *sk, *reuse_sk;
 391         bool no_reuseport;
 392
 393         if (hashinfo != net->ipv4.tcp_death_row.hashinfo)
 394                 return NULL; /* only TCP is supported */
 395
 396         no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_TCP, saddr, sport,
 397                                             daddr, hnum, dif, &sk);
 398         if (no_reuseport || IS_ERR_OR_NULL(sk))
 399                 return sk;
 400
 401         reuse_sk = lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum);
 402         if (reuse_sk)
 403                 sk = reuse_sk;
 404         return sk;
 405 }
 406
 407 struct sock *__inet_lookup_listener(struct net *net,
 408                                     struct inet_hashinfo *hashinfo,
 409                                     struct sk_buff *skb, int doff,
 410                                     const __be32 saddr, __be16 sport,
 411                                     const __be32 daddr, const unsigned short hnum,
 412                                     const int dif, const int sdif)
 413 {
 414         struct inet_listen_hashbucket *ilb2;
 415         struct sock *result = NULL;
 416         unsigned int hash2;
 417
 418         /* Lookup redirect from BPF */
 419         if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
 420                 result = inet_lookup_run_bpf(net, hashinfo, skb, doff,
 421                                              saddr, sport, daddr, hnum, dif);
 422                 if (result)
 423                         goto done;
 424         }
 425
 426         hash2 = ipv4_portaddr_hash(net, daddr, hnum);
 427         ilb2 = inet_lhash2_bucket(hashinfo, hash2);
 428
 429         result = inet_lhash2_lookup(net, ilb2, skb, doff,
 430                                     saddr, sport, daddr, hnum,
 431                                     dif, sdif);
 432         if (result)
 433                 goto done;
 434
 435         /* Lookup lhash2 with INADDR_ANY */
 436         hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 437         ilb2 = inet_lhash2_bucket(hashinfo, hash2);
 438
 439         result = inet_lhash2_lookup(net, ilb2, skb, doff,
 440                                     saddr, sport, htonl(INADDR_ANY), hnum,
 441                                     dif, sdif);
 442 done:
 443         if (IS_ERR(result))
 444                 return NULL;
 445         return result;
 446 }
 447 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
 448
 449 /* All sockets share common refcount, but have different destructors */
 450 void sock_gen_put(struct sock *sk)
 451 {
 452         if (!refcount_dec_and_test(&sk->sk_refcnt))
 453                 return;
 454
 455         if (sk->sk_state == TCP_TIME_WAIT)
 456                 inet_twsk_free(inet_twsk(sk));
 457         else if (sk->sk_state == TCP_NEW_SYN_RECV)
 458                 reqsk_free(inet_reqsk(sk));
 459         else
 460                 sk_free(sk);
 461 }
 462 EXPORT_SYMBOL_GPL(sock_gen_put);
 463
 464 void sock_edemux(struct sk_buff *skb)
 465 {
 466         sock_gen_put(skb->sk);
 467 }
 468 EXPORT_SYMBOL(sock_edemux);
 469
 470 struct sock *__inet_lookup_established(struct net *net,
 471                                   struct inet_hashinfo *hashinfo,
 472                                   const __be32 saddr, const __be16 sport,
 473                                   const __be32 daddr, const u16 hnum,
 474                                   const int dif, const int sdif)
 475 {
 476         INET_ADDR_COOKIE(acookie, saddr, daddr);
 477         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
 478         struct sock *sk;
 479         const struct hlist_nulls_node *node;
 480         /* Optimize here for direct hit, only listening connections can
 481          * have wildcards anyways.
 482          */
 483         unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
 484         unsigned int slot = hash & hashinfo->ehash_mask;
 485         struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
 486
 487 begin:
 488         sk_nulls_for_each_rcu(sk, node, &head->chain) {
 489                 if (sk->sk_hash != hash)
 490                         continue;
 491                 if (likely(inet_match(net, sk, acookie, ports, dif, sdif))) {
 492                         if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
 493                                 goto out;
 494                         if (unlikely(!inet_match(net, sk, acookie,
 495                                                  ports, dif, sdif))) {
 496                                 sock_gen_put(sk);
 497                                 goto begin;
 498                         }
 499                         goto found;
 500                 }
 501         }
 502         /*
 503          * if the nulls value we got at the end of this lookup is
 504          * not the expected one, we must restart lookup.
 505          * We probably met an item that was moved to another chain.
 506          */
 507         if (get_nulls_value(node) != slot)
 508                 goto begin;
 509 out:
 510         sk = NULL;
 511 found:
 512         return sk;
 513 }
 514 EXPORT_SYMBOL_GPL(__inet_lookup_established);
 515
 516 /* called with local bh disabled */
 517 static int __inet_check_established(struct inet_timewait_death_row *death_row,
 518                                     struct sock *sk, __u16 lport,
 519                                     struct inet_timewait_sock **twp)
 520 {
 521         struct inet_hashinfo *hinfo = death_row->hashinfo;
 522         struct inet_sock *inet = inet_sk(sk);
 523         __be32 daddr = inet->inet_rcv_saddr;
 524         __be32 saddr = inet->inet_daddr;
 525         int dif = sk->sk_bound_dev_if;
 526         struct net *net = sock_net(sk);
 527         int sdif = l3mdev_master_ifindex_by_index(net, dif);
 528         INET_ADDR_COOKIE(acookie, saddr, daddr);
 529         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
 530         unsigned int hash = inet_ehashfn(net, daddr, lport,
 531                                          saddr, inet->inet_dport);
 532         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 533         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
 534         struct sock *sk2;
 535         const struct hlist_nulls_node *node;
 536         struct inet_timewait_sock *tw = NULL;
 537
 538         spin_lock(lock);
 539
 540         sk_nulls_for_each(sk2, node, &head->chain) {
 541                 if (sk2->sk_hash != hash)
 542                         continue;
 543
 544                 if (likely(inet_match(net, sk2, acookie, ports, dif, sdif))) {
 545                         if (sk2->sk_state == TCP_TIME_WAIT) {
 546                                 tw = inet_twsk(sk2);
 547                                 if (twsk_unique(sk, sk2, twp))
 548                                         break;
 549                         }
 550                         goto not_unique;
 551                 }
 552         }
 553
 554         /* Must record num and sport now. Otherwise we will see
 555          * in hash table socket with a funny identity.
 556          */
 557         inet->inet_num = lport;
 558         inet->inet_sport = htons(lport);
 559         sk->sk_hash = hash;
 560         WARN_ON(!sk_unhashed(sk));
 561         __sk_nulls_add_node_rcu(sk, &head->chain);
 562         if (tw) {
 563                 sk_nulls_del_node_init_rcu((struct sock *)tw);
 564                 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
 565         }
 566         spin_unlock(lock);
 567         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 568
 569         if (twp) {
 570                 *twp = tw;
 571         } else if (tw) {
 572                 /* Silly. Should hash-dance instead... */
 573                 inet_twsk_deschedule_put(tw);
 574         }
 575         return 0;
 576
 577 not_unique:
 578         spin_unlock(lock);
 579         return -EADDRNOTAVAIL;
 580 }
 581
 582 static u64 inet_sk_port_offset(const struct sock *sk)
 583 {
 584         const struct inet_sock *inet = inet_sk(sk);
 585
 586         return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
 587                                           inet->inet_daddr,
 588                                           inet->inet_dport);
 589 }
 590
 591 /* Searches for an exsiting socket in the ehash bucket list.
 592  * Returns true if found, false otherwise.
 593  */
 594 static bool inet_ehash_lookup_by_sk(struct sock *sk,
 595                                     struct hlist_nulls_head *list)
 596 {
 597         const __portpair ports = INET_COMBINED_PORTS(sk->sk_dport, sk->sk_num);
 598         const int sdif = sk->sk_bound_dev_if;
 599         const int dif = sk->sk_bound_dev_if;
 600         const struct hlist_nulls_node *node;
 601         struct net *net = sock_net(sk);
 602         struct sock *esk;
 603
 604         INET_ADDR_COOKIE(acookie, sk->sk_daddr, sk->sk_rcv_saddr);
 605
 606         sk_nulls_for_each_rcu(esk, node, list) {
 607                 if (esk->sk_hash != sk->sk_hash)
 608                         continue;
 609                 if (sk->sk_family == AF_INET) {
 610                         if (unlikely(inet_match(net, esk, acookie,
 611                                                 ports, dif, sdif))) {
 612                                 return true;
 613                         }
 614                 }
 615 #if IS_ENABLED(CONFIG_IPV6)
 616                 else if (sk->sk_family == AF_INET6) {
 617                         if (unlikely(inet6_match(net, esk,
 618                                                  &sk->sk_v6_daddr,
 619                                                  &sk->sk_v6_rcv_saddr,
 620                                                  ports, dif, sdif))) {
 621                                 return true;
 622                         }
 623                 }
 624 #endif
 625         }
 626         return false;
 627 }
 628
 629 /* Insert a socket into ehash, and eventually remove another one
 630  * (The another one can be a SYN_RECV or TIMEWAIT)
 631  * If an existing socket already exists, socket sk is not inserted,
 632  * and sets found_dup_sk parameter to true.
 633  */
 634 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk)
 635 {
 636         struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
 637         struct inet_ehash_bucket *head;
 638         struct hlist_nulls_head *list;
 639         spinlock_t *lock;
 640         bool ret = true;
 641
 642         WARN_ON_ONCE(!sk_unhashed(sk));
 643
 644         sk->sk_hash = sk_ehashfn(sk);
 645         head = inet_ehash_bucket(hashinfo, sk->sk_hash);
 646         list = &head->chain;
 647         lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 648
 649         spin_lock(lock);
 650         if (osk) {
 651                 WARN_ON_ONCE(sk->sk_hash != osk->sk_hash);
 652                 ret = sk_nulls_del_node_init_rcu(osk);
 653         } else if (found_dup_sk) {
 654                 *found_dup_sk = inet_ehash_lookup_by_sk(sk, list);
 655                 if (*found_dup_sk)
 656                         ret = false;
 657         }
 658
 659         if (ret)
 660                 __sk_nulls_add_node_rcu(sk, list);
 661
 662         spin_unlock(lock);
 663
 664         return ret;
 665 }
 666
 667 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk, bool *found_dup_sk)
 668 {
 669         bool ok = inet_ehash_insert(sk, osk, found_dup_sk);
 670
 671         if (ok) {
 672                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 673         } else {
 674                 this_cpu_inc(*sk->sk_prot->orphan_count);
 675                 inet_sk_set_state(sk, TCP_CLOSE);
 676                 sock_set_flag(sk, SOCK_DEAD);
 677                 inet_csk_destroy_sock(sk);
 678         }
 679         return ok;
 680 }
 681 EXPORT_SYMBOL_GPL(inet_ehash_nolisten);
 682
 683 static int inet_reuseport_add_sock(struct sock *sk,
 684                                    struct inet_listen_hashbucket *ilb)
 685 {
 686         struct inet_bind_bucket *tb = inet_csk(sk)->icsk_bind_hash;
 687         const struct hlist_nulls_node *node;
 688         struct sock *sk2;
 689         kuid_t uid = sock_i_uid(sk);
 690
 691         sk_nulls_for_each_rcu(sk2, node, &ilb->nulls_head) {
 692                 if (sk2 != sk &&
 693                     sk2->sk_family == sk->sk_family &&
 694                     ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 695                     sk2->sk_bound_dev_if == sk->sk_bound_dev_if &&
 696                     inet_csk(sk2)->icsk_bind_hash == tb &&
 697                     sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
 698                     inet_rcv_saddr_equal(sk, sk2, false))
 699                         return reuseport_add_sock(sk, sk2,
 700                                                   inet_rcv_saddr_any(sk));
 701         }
 702
 703         return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
 704 }
 705
 706 int __inet_hash(struct sock *sk, struct sock *osk)
 707 {
 708         struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
 709         struct inet_listen_hashbucket *ilb2;
 710         int err = 0;
 711
 712         if (sk->sk_state != TCP_LISTEN) {
 713                 local_bh_disable();
 714                 inet_ehash_nolisten(sk, osk, NULL);
 715                 local_bh_enable();
 716                 return 0;
 717         }
 718         WARN_ON(!sk_unhashed(sk));
 719         ilb2 = inet_lhash2_bucket_sk(hashinfo, sk);
 720
 721         spin_lock(&ilb2->lock);
 722         if (sk->sk_reuseport) {
 723                 err = inet_reuseport_add_sock(sk, ilb2);
 724                 if (err)
 725                         goto unlock;
 726         }
 727         if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 728                 sk->sk_family == AF_INET6)
 729                 __sk_nulls_add_node_tail_rcu(sk, &ilb2->nulls_head);
 730         else
 731                 __sk_nulls_add_node_rcu(sk, &ilb2->nulls_head);
 732         sock_set_flag(sk, SOCK_RCU_FREE);
 733         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 734 unlock:
 735         spin_unlock(&ilb2->lock);
 736
 737         return err;
 738 }
 739 EXPORT_SYMBOL(__inet_hash);
 740
 741 int inet_hash(struct sock *sk)
 742 {
 743         int err = 0;
 744
 745         if (sk->sk_state != TCP_CLOSE)
 746                 err = __inet_hash(sk, NULL);
 747
 748         return err;
 749 }
 750 EXPORT_SYMBOL_GPL(inet_hash);
 751
 752 void inet_unhash(struct sock *sk)
 753 {
 754         struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
 755
 756         if (sk_unhashed(sk))
 757                 return;
 758
 759         if (sk->sk_state == TCP_LISTEN) {
 760                 struct inet_listen_hashbucket *ilb2;
 761
 762                 ilb2 = inet_lhash2_bucket_sk(hashinfo, sk);
 763                 /* Don't disable bottom halves while acquiring the lock to
 764                  * avoid circular locking dependency on PREEMPT_RT.
 765                  */
 766                 spin_lock(&ilb2->lock);
 767                 if (sk_unhashed(sk)) {
 768                         spin_unlock(&ilb2->lock);
 769                         return;
 770                 }
 771
 772                 if (rcu_access_pointer(sk->sk_reuseport_cb))
 773                         reuseport_stop_listen_sock(sk);
 774
 775                 __sk_nulls_del_node_init_rcu(sk);
 776                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 777                 spin_unlock(&ilb2->lock);
 778         } else {
 779                 spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 780
 781                 spin_lock_bh(lock);
 782                 if (sk_unhashed(sk)) {
 783                         spin_unlock_bh(lock);
 784                         return;
 785                 }
 786                 __sk_nulls_del_node_init_rcu(sk);
 787                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 788                 spin_unlock_bh(lock);
 789         }
 790 }
 791 EXPORT_SYMBOL_GPL(inet_unhash);
 792
 793 static bool inet_bind2_bucket_match(const struct inet_bind2_bucket *tb,
 794                                     const struct net *net, unsigned short port,
 795                                     int l3mdev, const struct sock *sk)
 796 {
 797 #if IS_ENABLED(CONFIG_IPV6)
 798         if (sk->sk_family != tb->family)
 799                 return false;
 800
 801         if (sk->sk_family == AF_INET6)
 802                 return net_eq(ib2_net(tb), net) && tb->port == port &&
 803                         tb->l3mdev == l3mdev &&
 804                         ipv6_addr_equal(&tb->v6_rcv_saddr, &sk->sk_v6_rcv_saddr);
 805         else
 806 #endif
 807                 return net_eq(ib2_net(tb), net) && tb->port == port &&
 808                         tb->l3mdev == l3mdev && tb->rcv_saddr == sk->sk_rcv_saddr;
 809 }
 810
 811 bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb, const struct net *net,
 812                                       unsigned short port, int l3mdev, const struct sock *sk)
 813 {
 814 #if IS_ENABLED(CONFIG_IPV6)
 815         struct in6_addr addr_any = {};
 816
 817         if (sk->sk_family != tb->family)
 818                 return false;
 819
 820         if (sk->sk_family == AF_INET6)
 821                 return net_eq(ib2_net(tb), net) && tb->port == port &&
 822                         tb->l3mdev == l3mdev &&
 823                         ipv6_addr_equal(&tb->v6_rcv_saddr, &addr_any);
 824         else
 825 #endif
 826                 return net_eq(ib2_net(tb), net) && tb->port == port &&
 827                         tb->l3mdev == l3mdev && tb->rcv_saddr == 0;
 828 }
 829
 830 /* The socket's bhash2 hashbucket spinlock must be held when this is called */
 831 struct inet_bind2_bucket *
 832 inet_bind2_bucket_find(const struct inet_bind_hashbucket *head, const struct net *net,
 833                        unsigned short port, int l3mdev, const struct sock *sk)
 834 {
 835         struct inet_bind2_bucket *bhash2 = NULL;
 836
 837         inet_bind_bucket_for_each(bhash2, &head->chain)
 838                 if (inet_bind2_bucket_match(bhash2, net, port, l3mdev, sk))
 839                         break;
 840
 841         return bhash2;
 842 }
 843
 844 struct inet_bind_hashbucket *
 845 inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port)
 846 {
 847         struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
 848         u32 hash;
 849 #if IS_ENABLED(CONFIG_IPV6)
 850         struct in6_addr addr_any = {};
 851
 852         if (sk->sk_family == AF_INET6)
 853                 hash = ipv6_portaddr_hash(net, &addr_any, port);
 854         else
 855 #endif
 856                 hash = ipv4_portaddr_hash(net, 0, port);
 857
 858         return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
 859 }
 860
 861 static void inet_update_saddr(struct sock *sk, void *saddr, int family)
 862 {
 863         if (family == AF_INET) {
 864                 inet_sk(sk)->inet_saddr = *(__be32 *)saddr;
 865                 sk_rcv_saddr_set(sk, inet_sk(sk)->inet_saddr);
 866         }
 867 #if IS_ENABLED(CONFIG_IPV6)
 868         else {
 869                 sk->sk_v6_rcv_saddr = *(struct in6_addr *)saddr;
 870         }
 871 #endif
 872 }
 873
 874 static int __inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family, bool reset)
 875 {
 876         struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
 877         struct inet_bind_hashbucket *head, *head2;
 878         struct inet_bind2_bucket *tb2, *new_tb2;
 879         int l3mdev = inet_sk_bound_l3mdev(sk);
 880         int port = inet_sk(sk)->inet_num;
 881         struct net *net = sock_net(sk);
 882         int bhash;
 883
 884         if (!inet_csk(sk)->icsk_bind2_hash) {
 885                 /* Not bind()ed before. */
 886                 if (reset)
 887                         inet_reset_saddr(sk);
 888                 else
 889                         inet_update_saddr(sk, saddr, family);
 890
 891                 return 0;
 892         }
 893
 894         /* Allocate a bind2 bucket ahead of time to avoid permanently putting
 895          * the bhash2 table in an inconsistent state if a new tb2 bucket
 896          * allocation fails.
 897          */
 898         new_tb2 = kmem_cache_alloc(hinfo->bind2_bucket_cachep, GFP_ATOMIC);
 899         if (!new_tb2) {
 900                 if (reset) {
 901                         /* The (INADDR_ANY, port) bucket might have already
 902                          * been freed, then we cannot fixup icsk_bind2_hash,
 903                          * so we give up and unlink sk from bhash/bhash2 not
 904                          * to leave inconsistency in bhash2.
 905                          */
 906                         inet_put_port(sk);
 907                         inet_reset_saddr(sk);
 908                 }
 909
 910                 return -ENOMEM;
 911         }
 912
 913         bhash = inet_bhashfn(net, port, hinfo->bhash_size);
 914         head = &hinfo->bhash[bhash];
 915         head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
 916
 917         /* If we change saddr locklessly, another thread
 918          * iterating over bhash might see corrupted address.
 919          */
 920         spin_lock_bh(&head->lock);
 921
 922         spin_lock(&head2->lock);
 923         __sk_del_bind2_node(sk);
 924         inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep, inet_csk(sk)->icsk_bind2_hash);
 925         spin_unlock(&head2->lock);
 926
 927         if (reset)
 928                 inet_reset_saddr(sk);
 929         else
 930                 inet_update_saddr(sk, saddr, family);
 931
 932         head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
 933
 934         spin_lock(&head2->lock);
 935         tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
 936         if (!tb2) {
 937                 tb2 = new_tb2;
 938                 inet_bind2_bucket_init(tb2, net, head2, port, l3mdev, sk);
 939         }
 940         sk_add_bind2_node(sk, &tb2->owners);
 941         inet_csk(sk)->icsk_bind2_hash = tb2;
 942         spin_unlock(&head2->lock);
 943
 944         spin_unlock_bh(&head->lock);
 945
 946         if (tb2 != new_tb2)
 947                 kmem_cache_free(hinfo->bind2_bucket_cachep, new_tb2);
 948
 949         return 0;
 950 }
 951
 952 int inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family)
 953 {
 954         return __inet_bhash2_update_saddr(sk, saddr, family, false);
 955 }
 956 EXPORT_SYMBOL_GPL(inet_bhash2_update_saddr);
 957
 958 void inet_bhash2_reset_saddr(struct sock *sk)
 959 {
 960         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
 961                 __inet_bhash2_update_saddr(sk, NULL, 0, true);
 962 }
 963 EXPORT_SYMBOL_GPL(inet_bhash2_reset_saddr);
 964
 965 /* RFC 6056 3.3.4.  Algorithm 4: Double-Hash Port Selection Algorithm
 966  * Note that we use 32bit integers (vs RFC 'short integers')
 967  * because 2^16 is not a multiple of num_ephemeral and this
 968  * property might be used by clever attacker.
 969  *
 970  * RFC claims using TABLE_LENGTH=10 buckets gives an improvement, though
 971  * attacks were since demonstrated, thus we use 65536 by default instead
 972  * to really give more isolation and privacy, at the expense of 256kB
 973  * of kernel memory.
 974  */
 975 #define INET_TABLE_PERTURB_SIZE (1 << CONFIG_INET_TABLE_PERTURB_ORDER)
 976 static u32 *table_perturb;
 977
 978 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 979                 struct sock *sk, u64 port_offset,
 980                 int (*check_established)(struct inet_timewait_death_row *,
 981                         struct sock *, __u16, struct inet_timewait_sock **))
 982 {
 983         struct inet_hashinfo *hinfo = death_row->hashinfo;
 984         struct inet_bind_hashbucket *head, *head2;
 985         struct inet_timewait_sock *tw = NULL;
 986         int port = inet_sk(sk)->inet_num;
 987         struct net *net = sock_net(sk);
 988         struct inet_bind2_bucket *tb2;
 989         struct inet_bind_bucket *tb;
 990         bool tb_created = false;
 991         u32 remaining, offset;
 992         int ret, i, low, high;
 993         int l3mdev;
 994         u32 index;
 995
 996         if (port) {
 997                 head = &hinfo->bhash[inet_bhashfn(net, port,
 998                                                   hinfo->bhash_size)];
 999                 tb = inet_csk(sk)->icsk_bind_hash;
1000                 spin_lock_bh(&head->lock);
1001                 if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
1002                         inet_ehash_nolisten(sk, NULL, NULL);
1003                         spin_unlock_bh(&head->lock);
1004                         return 0;
1005                 }
1006                 spin_unlock(&head->lock);
1007                 /* No definite answer... Walk to established hash table */
1008                 ret = check_established(death_row, sk, port, NULL);
1009                 local_bh_enable();
1010                 return ret;
1011         }
1012
1013         l3mdev = inet_sk_bound_l3mdev(sk);
1014
1015         inet_get_local_port_range(net, &low, &high);
1016         high++; /* [32768, 60999] -> [32768, 61000[ */
1017         remaining = high - low;
1018         if (likely(remaining > 1))
1019                 remaining &= ~1U;
1020
1021         get_random_sleepable_once(table_perturb,
1022                                   INET_TABLE_PERTURB_SIZE * sizeof(*table_perturb));
1023         index = port_offset & (INET_TABLE_PERTURB_SIZE - 1);
1024
1025         offset = READ_ONCE(table_perturb[index]) + (port_offset >> 32);
1026         offset %= remaining;
1027
1028         /* In first pass we try ports of @low parity.
1029          * inet_csk_get_port() does the opposite choice.
1030          */
1031         offset &= ~1U;
1032 other_parity_scan:
1033         port = low + offset;
1034         for (i = 0; i < remaining; i += 2, port += 2) {
1035                 if (unlikely(port >= high))
1036                         port -= remaining;
1037                 if (inet_is_local_reserved_port(net, port))
1038                         continue;
1039                 head = &hinfo->bhash[inet_bhashfn(net, port,
1040                                                   hinfo->bhash_size)];
1041                 spin_lock_bh(&head->lock);
1042
1043                 /* Does not bother with rcv_saddr checks, because
1044                  * the established check is already unique enough.
1045                  */
1046                 inet_bind_bucket_for_each(tb, &head->chain) {
1047                         if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
1048                                 if (tb->fastreuse >= 0 ||
1049                                     tb->fastreuseport >= 0)
1050                                         goto next_port;
1051                                 WARN_ON(hlist_empty(&tb->owners));
1052                                 if (!check_established(death_row, sk,
1053                                                        port, &tw))
1054                                         goto ok;
1055                                 goto next_port;
1056                         }
1057                 }
1058
1059                 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
1060                                              net, head, port, l3mdev);
1061                 if (!tb) {
1062                         spin_unlock_bh(&head->lock);
1063                         return -ENOMEM;
1064                 }
1065                 tb_created = true;
1066                 tb->fastreuse = -1;
1067                 tb->fastreuseport = -1;
1068                 goto ok;
1069 next_port:
1070                 spin_unlock_bh(&head->lock);
1071                 cond_resched();
1072         }
1073
1074         offset++;
1075         if ((offset & 1) && remaining > 1)
1076                 goto other_parity_scan;
1077
1078         return -EADDRNOTAVAIL;
1079
1080 ok:
1081         /* Find the corresponding tb2 bucket since we need to
1082          * add the socket to the bhash2 table as well
1083          */
1084         head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
1085         spin_lock(&head2->lock);
1086
1087         tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
1088         if (!tb2) {
1089                 tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep, net,
1090                                                head2, port, l3mdev, sk);
1091                 if (!tb2)
1092                         goto error;
1093         }
1094
1095         /* Here we want to add a little bit of randomness to the next source
1096          * port that will be chosen. We use a max() with a random here so that
1097          * on low contention the randomness is maximal and on high contention
1098          * it may be inexistent.
1099          */
1100         i = max_t(int, i, get_random_u32_below(8) * 2);
1101         WRITE_ONCE(table_perturb[index], READ_ONCE(table_perturb[index]) + i + 2);
1102
1103         /* Head lock still held and bh's disabled */
1104         inet_bind_hash(sk, tb, tb2, port);
1105
1106         spin_unlock(&head2->lock);
1107
1108         if (sk_unhashed(sk)) {
1109                 inet_sk(sk)->inet_sport = htons(port);
1110                 inet_ehash_nolisten(sk, (struct sock *)tw, NULL);
1111         }
1112         if (tw)
1113                 inet_twsk_bind_unhash(tw, hinfo);
1114         spin_unlock(&head->lock);
1115         if (tw)
1116                 inet_twsk_deschedule_put(tw);
1117         local_bh_enable();
1118         return 0;
1119
1120 error:
1121         spin_unlock(&head2->lock);
1122         if (tb_created)
1123                 inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
1124         spin_unlock_bh(&head->lock);
1125         return -ENOMEM;
1126 }
1127
1128 /*
1129  * Bind a port for a connect operation and hash it.
1130  */
1131 int inet_hash_connect(struct inet_timewait_death_row *death_row,
1132                       struct sock *sk)
1133 {
1134         u64 port_offset = 0;
1135
1136         if (!inet_sk(sk)->inet_num)
1137                 port_offset = inet_sk_port_offset(sk);
1138         return __inet_hash_connect(death_row, sk, port_offset,
1139                                    __inet_check_established);
1140 }
1141 EXPORT_SYMBOL_GPL(inet_hash_connect);
1142
1143 static void init_hashinfo_lhash2(struct inet_hashinfo *h)
1144 {
1145         int i;
1146
1147         for (i = 0; i <= h->lhash2_mask; i++) {
1148                 spin_lock_init(&h->lhash2[i].lock);
1149                 INIT_HLIST_NULLS_HEAD(&h->lhash2[i].nulls_head,
1150                                       i + LISTENING_NULLS_BASE);
1151         }
1152 }
1153
1154 void __init inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
1155                                 unsigned long numentries, int scale,
1156                                 unsigned long low_limit,
1157                                 unsigned long high_limit)
1158 {
1159         h->lhash2 = alloc_large_system_hash(name,
1160                                             sizeof(*h->lhash2),
1161                                             numentries,
1162                                             scale,
1163                                             0,
1164                                             NULL,
1165                                             &h->lhash2_mask,
1166                                             low_limit,
1167                                             high_limit);
1168         init_hashinfo_lhash2(h);
1169
1170         /* this one is used for source ports of outgoing connections */
1171         table_perturb = alloc_large_system_hash("Table-perturb",
1172                                                 sizeof(*table_perturb),
1173                                                 INET_TABLE_PERTURB_SIZE,
1174                                                 0, 0, NULL, NULL,
1175                                                 INET_TABLE_PERTURB_SIZE,
1176                                                 INET_TABLE_PERTURB_SIZE);
1177 }
1178
1179 int inet_hashinfo2_init_mod(struct inet_hashinfo *h)
1180 {
1181         h->lhash2 = kmalloc_array(INET_LHTABLE_SIZE, sizeof(*h->lhash2), GFP_KERNEL);
1182         if (!h->lhash2)
1183                 return -ENOMEM;
1184
1185         h->lhash2_mask = INET_LHTABLE_SIZE - 1;
1186         /* INET_LHTABLE_SIZE must be a power of 2 */
1187         BUG_ON(INET_LHTABLE_SIZE & h->lhash2_mask);
1188
1189         init_hashinfo_lhash2(h);
1190         return 0;
1191 }
1192 EXPORT_SYMBOL_GPL(inet_hashinfo2_init_mod);
1193
1194 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
1195 {
1196         unsigned int locksz = sizeof(spinlock_t);
1197         unsigned int i, nblocks = 1;
1198
1199         if (locksz != 0) {
1200                 /* allocate 2 cache lines or at least one spinlock per cpu */
1201                 nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U);
1202                 nblocks = roundup_pow_of_two(nblocks * num_possible_cpus());
1203
1204                 /* no more locks than number of hash buckets */
1205                 nblocks = min(nblocks, hashinfo->ehash_mask + 1);
1206
1207                 hashinfo->ehash_locks = kvmalloc_array(nblocks, locksz, GFP_KERNEL);
1208                 if (!hashinfo->ehash_locks)
1209                         return -ENOMEM;
1210
1211                 for (i = 0; i < nblocks; i++)
1212                         spin_lock_init(&hashinfo->ehash_locks[i]);
1213         }
1214         hashinfo->ehash_locks_mask = nblocks - 1;
1215         return 0;
1216 }
1217 EXPORT_SYMBOL_GPL(inet_ehash_locks_alloc);
1218
1219 struct inet_hashinfo *inet_pernet_hashinfo_alloc(struct inet_hashinfo *hashinfo,
1220                                                  unsigned int ehash_entries)
1221 {
1222         struct inet_hashinfo *new_hashinfo;
1223         int i;
1224
1225         new_hashinfo = kmemdup(hashinfo, sizeof(*hashinfo), GFP_KERNEL);
1226         if (!new_hashinfo)
1227                 goto err;
1228
1229         new_hashinfo->ehash = vmalloc_huge(ehash_entries * sizeof(struct inet_ehash_bucket),
1230                                            GFP_KERNEL_ACCOUNT);
1231         if (!new_hashinfo->ehash)
1232                 goto free_hashinfo;
1233
1234         new_hashinfo->ehash_mask = ehash_entries - 1;
1235
1236         if (inet_ehash_locks_alloc(new_hashinfo))
1237                 goto free_ehash;
1238
1239         for (i = 0; i < ehash_entries; i++)
1240                 INIT_HLIST_NULLS_HEAD(&new_hashinfo->ehash[i].chain, i);
1241
1242         new_hashinfo->pernet = true;
1243
1244         return new_hashinfo;
1245
1246 free_ehash:
1247         vfree(new_hashinfo->ehash);
1248 free_hashinfo:
1249         kfree(new_hashinfo);
1250 err:
1251         return NULL;
1252 }
1253 EXPORT_SYMBOL_GPL(inet_pernet_hashinfo_alloc);
1254
1255 void inet_pernet_hashinfo_free(struct inet_hashinfo *hashinfo)
1256 {
1257         if (!hashinfo->pernet)
1258                 return;
1259
1260         inet_ehash_locks_free(hashinfo);
1261         vfree(hashinfo->ehash);
1262         kfree(hashinfo);
1263 }
1264 EXPORT_SYMBOL_GPL(inet_pernet_hashinfo_free);