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/hotdata.h>
  27#include <net/ip.h>
  28#include <net/rps.h>
  29#include <net/secure_seq.h>
  30#include <net/sock_reuseport.h>
  31#include <net/tcp.h>
  32
  33u32 inet_ehashfn(const struct net *net, const __be32 laddr,
  34		 const __u16 lport, const __be32 faddr,
  35		 const __be16 fport)
  36{
  37	net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret));
  38
  39	return lport + __inet_ehashfn(laddr, 0, faddr, fport,
  40				      inet_ehash_secret + net_hash_mix(net));
  41}
  42EXPORT_SYMBOL_GPL(inet_ehashfn);
  43
  44/* This function handles inet_sock, but also timewait and request sockets
  45 * for IPv4/IPv6.
  46 */
  47static u32 sk_ehashfn(const struct sock *sk)
  48{
  49#if IS_ENABLED(CONFIG_IPV6)
  50	if (sk->sk_family == AF_INET6 &&
  51	    !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
  52		return inet6_ehashfn(sock_net(sk),
  53				     &sk->sk_v6_rcv_saddr, sk->sk_num,
  54				     &sk->sk_v6_daddr, sk->sk_dport);
  55#endif
  56	return inet_ehashfn(sock_net(sk),
  57			    sk->sk_rcv_saddr, sk->sk_num,
  58			    sk->sk_daddr, sk->sk_dport);
  59}
  60
  61static bool sk_is_connect_bind(const struct sock *sk)
  62{
  63	if (sk->sk_state == TCP_TIME_WAIT)
  64		return inet_twsk(sk)->tw_connect_bind;
  65	else
  66		return sk->sk_userlocks & SOCK_CONNECT_BIND;
  67}
  68
  69/*
  70 * Allocate and initialize a new local port bind bucket.
  71 * The bindhash mutex for snum's hash chain must be held here.
  72 */
  73struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
  74						 struct net *net,
  75						 struct inet_bind_hashbucket *head,
  76						 const unsigned short snum,
  77						 int l3mdev)
  78{
  79	struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
  80
  81	if (tb) {
  82		write_pnet(&tb->ib_net, net);
  83		tb->l3mdev    = l3mdev;
  84		tb->port      = snum;
  85		tb->fastreuse = 0;
  86		tb->fastreuseport = 0;
  87		INIT_HLIST_HEAD(&tb->bhash2);
  88		hlist_add_head_rcu(&tb->node, &head->chain);
  89	}
  90	return tb;
  91}
  92
  93/*
  94 * Caller must hold hashbucket lock for this tb with local BH disabled
  95 */
  96void inet_bind_bucket_destroy(struct inet_bind_bucket *tb)
  97{
  98	const struct inet_bind2_bucket *tb2;
  99
 100	if (hlist_empty(&tb->bhash2)) {
 101		hlist_del_rcu(&tb->node);
 102		kfree_rcu(tb, rcu);
 103		return;
 104	}
 105
 106	if (tb->fastreuse == -1 && tb->fastreuseport == -1)
 107		return;
 108	hlist_for_each_entry(tb2, &tb->bhash2, bhash_node) {
 109		if (tb2->fastreuse != -1 || tb2->fastreuseport != -1)
 110			return;
 111	}
 112	tb->fastreuse = -1;
 113	tb->fastreuseport = -1;
 114}
 115
 116bool inet_bind_bucket_match(const struct inet_bind_bucket *tb, const struct net *net,
 117			    unsigned short port, int l3mdev)
 118{
 119	return net_eq(ib_net(tb), net) && tb->port == port &&
 120		tb->l3mdev == l3mdev;
 121}
 122
 123static void inet_bind2_bucket_init(struct inet_bind2_bucket *tb2,
 124				   struct net *net,
 125				   struct inet_bind_hashbucket *head,
 126				   struct inet_bind_bucket *tb,
 127				   const struct sock *sk)
 128{
 129	write_pnet(&tb2->ib_net, net);
 130	tb2->l3mdev = tb->l3mdev;
 131	tb2->port = tb->port;
 132#if IS_ENABLED(CONFIG_IPV6)
 133	BUILD_BUG_ON(USHRT_MAX < (IPV6_ADDR_ANY | IPV6_ADDR_MAPPED));
 134	if (sk->sk_family == AF_INET6) {
 135		tb2->addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
 136		tb2->v6_rcv_saddr = sk->sk_v6_rcv_saddr;
 137	} else {
 138		tb2->addr_type = IPV6_ADDR_MAPPED;
 139		ipv6_addr_set_v4mapped(sk->sk_rcv_saddr, &tb2->v6_rcv_saddr);
 140	}
 141#else
 142	tb2->rcv_saddr = sk->sk_rcv_saddr;
 143#endif
 144	tb2->fastreuse = 0;
 145	tb2->fastreuseport = 0;
 146	INIT_HLIST_HEAD(&tb2->owners);
 147	hlist_add_head(&tb2->node, &head->chain);
 148	hlist_add_head(&tb2->bhash_node, &tb->bhash2);
 149}
 150
 151struct inet_bind2_bucket *inet_bind2_bucket_create(struct kmem_cache *cachep,
 152						   struct net *net,
 153						   struct inet_bind_hashbucket *head,
 154						   struct inet_bind_bucket *tb,
 155						   const struct sock *sk)
 156{
 157	struct inet_bind2_bucket *tb2 = kmem_cache_alloc(cachep, GFP_ATOMIC);
 158
 159	if (tb2)
 160		inet_bind2_bucket_init(tb2, net, head, tb, sk);
 161
 162	return tb2;
 163}
 164
 165/* Caller must hold hashbucket lock for this tb with local BH disabled */
 166void inet_bind2_bucket_destroy(struct kmem_cache *cachep, struct inet_bind2_bucket *tb)
 167{
 168	const struct sock *sk;
 169
 170	if (hlist_empty(&tb->owners)) {
 171		__hlist_del(&tb->node);
 172		__hlist_del(&tb->bhash_node);
 173		kmem_cache_free(cachep, tb);
 174		return;
 175	}
 176
 177	if (tb->fastreuse == -1 && tb->fastreuseport == -1)
 178		return;
 179	sk_for_each_bound(sk, &tb->owners) {
 180		if (!sk_is_connect_bind(sk))
 181			return;
 182	}
 183	tb->fastreuse = -1;
 184	tb->fastreuseport = -1;
 185}
 186
 187static bool inet_bind2_bucket_addr_match(const struct inet_bind2_bucket *tb2,
 188					 const struct sock *sk)
 189{
 190#if IS_ENABLED(CONFIG_IPV6)
 191	if (sk->sk_family == AF_INET6)
 192		return ipv6_addr_equal(&tb2->v6_rcv_saddr, &sk->sk_v6_rcv_saddr);
 193
 194	if (tb2->addr_type != IPV6_ADDR_MAPPED)
 195		return false;
 196#endif
 197	return tb2->rcv_saddr == sk->sk_rcv_saddr;
 198}
 199
 200void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
 201		    struct inet_bind2_bucket *tb2, unsigned short port)
 202{
 203	WRITE_ONCE(inet_sk(sk)->inet_num, port);
 204	inet_csk(sk)->icsk_bind_hash = tb;
 205	inet_csk(sk)->icsk_bind2_hash = tb2;
 206	sk_add_bind_node(sk, &tb2->owners);
 207}
 208
 209/*
 210 * Get rid of any references to a local port held by the given sock.
 211 */
 212static void __inet_put_port(struct sock *sk)
 213{
 214	struct inet_hashinfo *hashinfo = tcp_get_hashinfo(sk);
 215	struct inet_bind_hashbucket *head, *head2;
 216	struct net *net = sock_net(sk);
 217	struct inet_bind_bucket *tb;
 218	int bhash;
 219
 220	bhash = inet_bhashfn(net, inet_sk(sk)->inet_num, hashinfo->bhash_size);
 221	head = &hashinfo->bhash[bhash];
 222	head2 = inet_bhashfn_portaddr(hashinfo, sk, net, inet_sk(sk)->inet_num);
 223
 224	spin_lock(&head->lock);
 225	tb = inet_csk(sk)->icsk_bind_hash;
 226	inet_csk(sk)->icsk_bind_hash = NULL;
 227	WRITE_ONCE(inet_sk(sk)->inet_num, 0);
 228	sk->sk_userlocks &= ~SOCK_CONNECT_BIND;
 229
 230	spin_lock(&head2->lock);
 231	if (inet_csk(sk)->icsk_bind2_hash) {
 232		struct inet_bind2_bucket *tb2 = inet_csk(sk)->icsk_bind2_hash;
 233
 234		__sk_del_bind_node(sk);
 235		inet_csk(sk)->icsk_bind2_hash = NULL;
 236		inet_bind2_bucket_destroy(hashinfo->bind2_bucket_cachep, tb2);
 237	}
 238	spin_unlock(&head2->lock);
 239
 240	inet_bind_bucket_destroy(tb);
 241	spin_unlock(&head->lock);
 242}
 243
 244void inet_put_port(struct sock *sk)
 245{
 246	local_bh_disable();
 247	__inet_put_port(sk);
 248	local_bh_enable();
 249}
 250EXPORT_SYMBOL(inet_put_port);
 251
 252int __inet_inherit_port(const struct sock *sk, struct sock *child)
 253{
 254	struct inet_hashinfo *table = tcp_get_hashinfo(sk);
 255	unsigned short port = inet_sk(child)->inet_num;
 256	struct inet_bind_hashbucket *head, *head2;
 257	bool created_inet_bind_bucket = false;
 258	struct net *net = sock_net(sk);
 259	bool update_fastreuse = false;
 260	struct inet_bind2_bucket *tb2;
 261	struct inet_bind_bucket *tb;
 262	int bhash, l3mdev;
 263
 264	bhash = inet_bhashfn(net, port, table->bhash_size);
 265	head = &table->bhash[bhash];
 266	head2 = inet_bhashfn_portaddr(table, child, net, port);
 267
 268	spin_lock(&head->lock);
 269	spin_lock(&head2->lock);
 270	tb = inet_csk(sk)->icsk_bind_hash;
 271	tb2 = inet_csk(sk)->icsk_bind2_hash;
 272	if (unlikely(!tb || !tb2)) {
 273		spin_unlock(&head2->lock);
 274		spin_unlock(&head->lock);
 275		return -ENOENT;
 276	}
 277	if (tb->port != port) {
 278		l3mdev = inet_sk_bound_l3mdev(sk);
 279
 280		/* NOTE: using tproxy and redirecting skbs to a proxy
 281		 * on a different listener port breaks the assumption
 282		 * that the listener socket's icsk_bind_hash is the same
 283		 * as that of the child socket. We have to look up or
 284		 * create a new bind bucket for the child here. */
 285		inet_bind_bucket_for_each(tb, &head->chain) {
 286			if (inet_bind_bucket_match(tb, net, port, l3mdev))
 287				break;
 288		}
 289		if (!tb) {
 290			tb = inet_bind_bucket_create(table->bind_bucket_cachep,
 291						     net, head, port, l3mdev);
 292			if (!tb) {
 293				spin_unlock(&head2->lock);
 294				spin_unlock(&head->lock);
 295				return -ENOMEM;
 296			}
 297			created_inet_bind_bucket = true;
 298		}
 299		update_fastreuse = true;
 300
 301		goto bhash2_find;
 302	} else if (!inet_bind2_bucket_addr_match(tb2, child)) {
 303		l3mdev = inet_sk_bound_l3mdev(sk);
 304
 305bhash2_find:
 306		tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, child);
 307		if (!tb2) {
 308			tb2 = inet_bind2_bucket_create(table->bind2_bucket_cachep,
 309						       net, head2, tb, child);
 310			if (!tb2)
 311				goto error;
 312		}
 313	}
 314	if (update_fastreuse)
 315		inet_csk_update_fastreuse(child, tb, tb2);
 316	inet_bind_hash(child, tb, tb2, port);
 317	spin_unlock(&head2->lock);
 318	spin_unlock(&head->lock);
 319
 320	return 0;
 321
 322error:
 323	if (created_inet_bind_bucket)
 324		inet_bind_bucket_destroy(tb);
 325	spin_unlock(&head2->lock);
 326	spin_unlock(&head->lock);
 327	return -ENOMEM;
 328}
 329EXPORT_SYMBOL_GPL(__inet_inherit_port);
 330
 331static struct inet_listen_hashbucket *
 332inet_lhash2_bucket_sk(struct inet_hashinfo *h, struct sock *sk)
 333{
 334	u32 hash;
 335
 336#if IS_ENABLED(CONFIG_IPV6)
 337	if (sk->sk_family == AF_INET6)
 338		hash = ipv6_portaddr_hash(sock_net(sk),
 339					  &sk->sk_v6_rcv_saddr,
 340					  inet_sk(sk)->inet_num);
 341	else
 342#endif
 343		hash = ipv4_portaddr_hash(sock_net(sk),
 344					  inet_sk(sk)->inet_rcv_saddr,
 345					  inet_sk(sk)->inet_num);
 346	return inet_lhash2_bucket(h, hash);
 347}
 348
 349static inline int compute_score(struct sock *sk, const struct net *net,
 350				const unsigned short hnum, const __be32 daddr,
 351				const int dif, const int sdif)
 352{
 353	int score = -1;
 354
 355	if (net_eq(sock_net(sk), net) && READ_ONCE(sk->sk_num) == hnum &&
 356			!ipv6_only_sock(sk)) {
 357		if (sk->sk_rcv_saddr != daddr)
 358			return -1;
 359
 360		if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
 361			return -1;
 362		score =  sk->sk_bound_dev_if ? 2 : 1;
 363
 364		if (sk->sk_family == PF_INET)
 365			score++;
 366		if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
 367			score++;
 368	}
 369	return score;
 370}
 371
 372/**
 373 * inet_lookup_reuseport() - execute reuseport logic on AF_INET socket if necessary.
 374 * @net: network namespace.
 375 * @sk: AF_INET socket, must be in TCP_LISTEN state for TCP or TCP_CLOSE for UDP.
 376 * @skb: context for a potential SK_REUSEPORT program.
 377 * @doff: header offset.
 378 * @saddr: source address.
 379 * @sport: source port.
 380 * @daddr: destination address.
 381 * @hnum: destination port in host byte order.
 382 * @ehashfn: hash function used to generate the fallback hash.
 383 *
 384 * Return: NULL if sk doesn't have SO_REUSEPORT set, otherwise a pointer to
 385 *         the selected sock or an error.
 386 */
 387struct sock *inet_lookup_reuseport(const struct net *net, struct sock *sk,
 388				   struct sk_buff *skb, int doff,
 389				   __be32 saddr, __be16 sport,
 390				   __be32 daddr, unsigned short hnum,
 391				   inet_ehashfn_t *ehashfn)
 392{
 393	struct sock *reuse_sk = NULL;
 394	u32 phash;
 395
 396	if (sk->sk_reuseport) {
 397		phash = INDIRECT_CALL_2(ehashfn, udp_ehashfn, inet_ehashfn,
 398					net, daddr, hnum, saddr, sport);
 399		reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
 400	}
 401	return reuse_sk;
 402}
 403EXPORT_SYMBOL_GPL(inet_lookup_reuseport);
 404
 405/*
 406 * Here are some nice properties to exploit here. The BSD API
 407 * does not allow a listening sock to specify the remote port nor the
 408 * remote address for the connection. So always assume those are both
 409 * wildcarded during the search since they can never be otherwise.
 410 */
 411
 412/* called with rcu_read_lock() : No refcount taken on the socket */
 413static struct sock *inet_lhash2_lookup(const struct net *net,
 414				struct inet_listen_hashbucket *ilb2,
 415				struct sk_buff *skb, int doff,
 416				const __be32 saddr, __be16 sport,
 417				const __be32 daddr, const unsigned short hnum,
 418				const int dif, const int sdif)
 419{
 420	struct sock *sk, *result = NULL;
 421	struct hlist_nulls_node *node;
 422	int score, hiscore = 0;
 423
 424	sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) {
 425		score = compute_score(sk, net, hnum, daddr, dif, sdif);
 426		if (score > hiscore) {
 427			result = inet_lookup_reuseport(net, sk, skb, doff,
 428						       saddr, sport, daddr, hnum, inet_ehashfn);
 429			if (result)
 430				return result;
 431
 432			result = sk;
 433			hiscore = score;
 434		}
 435	}
 436
 437	return result;
 438}
 439
 440struct sock *inet_lookup_run_sk_lookup(const struct net *net,
 441				       int protocol,
 442				       struct sk_buff *skb, int doff,
 443				       __be32 saddr, __be16 sport,
 444				       __be32 daddr, u16 hnum, const int dif,
 445				       inet_ehashfn_t *ehashfn)
 446{
 447	struct sock *sk, *reuse_sk;
 448	bool no_reuseport;
 449
 450	no_reuseport = bpf_sk_lookup_run_v4(net, protocol, saddr, sport,
 451					    daddr, hnum, dif, &sk);
 452	if (no_reuseport || IS_ERR_OR_NULL(sk))
 453		return sk;
 454
 455	reuse_sk = inet_lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum,
 456					 ehashfn);
 457	if (reuse_sk)
 458		sk = reuse_sk;
 459	return sk;
 460}
 461
 462struct sock *__inet_lookup_listener(const struct net *net,
 463				    struct sk_buff *skb, int doff,
 464				    const __be32 saddr, __be16 sport,
 465				    const __be32 daddr, const unsigned short hnum,
 466				    const int dif, const int sdif)
 467{
 468	struct inet_listen_hashbucket *ilb2;
 469	struct inet_hashinfo *hashinfo;
 470	struct sock *result = NULL;
 471	unsigned int hash2;
 472
 473	/* Lookup redirect from BPF */
 474	if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
 475		result = inet_lookup_run_sk_lookup(net, IPPROTO_TCP, skb, doff,
 476						   saddr, sport, daddr, hnum, dif,
 477						   inet_ehashfn);
 478		if (result)
 479			goto done;
 480	}
 481
 482	hashinfo = net->ipv4.tcp_death_row.hashinfo;
 483	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
 484	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
 485
 486	result = inet_lhash2_lookup(net, ilb2, skb, doff,
 487				    saddr, sport, daddr, hnum,
 488				    dif, sdif);
 489	if (result)
 490		goto done;
 491
 492	/* Lookup lhash2 with INADDR_ANY */
 493	hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 494	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
 495
 496	result = inet_lhash2_lookup(net, ilb2, skb, doff,
 497				    saddr, sport, htonl(INADDR_ANY), hnum,
 498				    dif, sdif);
 499done:
 500	if (IS_ERR(result))
 501		return NULL;
 502	return result;
 503}
 504EXPORT_SYMBOL_GPL(__inet_lookup_listener);
 505
 506/* All sockets share common refcount, but have different destructors */
 507void sock_gen_put(struct sock *sk)
 508{
 509	if (!refcount_dec_and_test(&sk->sk_refcnt))
 510		return;
 511
 512	if (sk->sk_state == TCP_TIME_WAIT)
 513		inet_twsk_free(inet_twsk(sk));
 514	else if (sk->sk_state == TCP_NEW_SYN_RECV)
 515		reqsk_free(inet_reqsk(sk));
 516	else
 517		sk_free(sk);
 518}
 519EXPORT_SYMBOL_GPL(sock_gen_put);
 520
 521void sock_edemux(struct sk_buff *skb)
 522{
 523	sock_gen_put(skb->sk);
 524}
 525EXPORT_SYMBOL(sock_edemux);
 526
 527struct sock *__inet_lookup_established(const struct net *net,
 528				       const __be32 saddr, const __be16 sport,
 529				       const __be32 daddr, const u16 hnum,
 530				       const int dif, const int sdif)
 531{
 532	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
 533	INET_ADDR_COOKIE(acookie, saddr, daddr);
 534	const struct hlist_nulls_node *node;
 535	struct inet_ehash_bucket *head;
 536	struct inet_hashinfo *hashinfo;
 537	unsigned int hash, slot;
 538	struct sock *sk;
 539
 540	hashinfo = net->ipv4.tcp_death_row.hashinfo;
 541	hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
 542	slot = hash & hashinfo->ehash_mask;
 543	head = &hashinfo->ehash[slot];
 544
 545begin:
 546	sk_nulls_for_each_rcu(sk, node, &head->chain) {
 547		if (sk->sk_hash != hash)
 548			continue;
 549		if (likely(inet_match(net, sk, acookie, ports, dif, sdif))) {
 550			if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
 551				goto out;
 552			if (unlikely(!inet_match(net, sk, acookie,
 553						 ports, dif, sdif))) {
 554				sock_gen_put(sk);
 555				goto begin;
 556			}
 557			goto found;
 558		}
 559	}
 560	/*
 561	 * if the nulls value we got at the end of this lookup is
 562	 * not the expected one, we must restart lookup.
 563	 * We probably met an item that was moved to another chain.
 564	 */
 565	if (get_nulls_value(node) != slot)
 566		goto begin;
 567out:
 568	sk = NULL;
 569found:
 570	return sk;
 571}
 572EXPORT_SYMBOL_GPL(__inet_lookup_established);
 573
 574/* called with local bh disabled */
 575static int __inet_check_established(struct inet_timewait_death_row *death_row,
 576				    struct sock *sk, __u16 lport,
 577				    struct inet_timewait_sock **twp,
 578				    bool rcu_lookup,
 579				    u32 hash)
 580{
 581	struct inet_hashinfo *hinfo = death_row->hashinfo;
 582	struct inet_sock *inet = inet_sk(sk);
 583	__be32 daddr = inet->inet_rcv_saddr;
 584	__be32 saddr = inet->inet_daddr;
 585	int dif = sk->sk_bound_dev_if;
 586	struct net *net = sock_net(sk);
 587	int sdif = l3mdev_master_ifindex_by_index(net, dif);
 588	INET_ADDR_COOKIE(acookie, saddr, daddr);
 589	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
 590	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
 591	struct inet_timewait_sock *tw = NULL;
 592	const struct hlist_nulls_node *node;
 593	struct sock *sk2;
 594	spinlock_t *lock;
 595
 596	if (rcu_lookup) {
 597		sk_nulls_for_each(sk2, node, &head->chain) {
 598			if (sk2->sk_hash != hash ||
 599			    !inet_match(net, sk2, acookie, ports, dif, sdif))
 600				continue;
 601			if (sk2->sk_state == TCP_TIME_WAIT)
 602				break;
 603			return -EADDRNOTAVAIL;
 604		}
 605		return 0;
 606	}
 607
 608	lock = inet_ehash_lockp(hinfo, hash);
 609	spin_lock(lock);
 610
 611	sk_nulls_for_each(sk2, node, &head->chain) {
 612		if (sk2->sk_hash != hash)
 613			continue;
 614
 615		if (likely(inet_match(net, sk2, acookie, ports, dif, sdif))) {
 616			if (sk2->sk_state == TCP_TIME_WAIT) {
 617				tw = inet_twsk(sk2);
 618				if (tcp_twsk_unique(sk, sk2, twp))
 619					break;
 620			}
 621			goto not_unique;
 622		}
 623	}
 624
 625	/* Must record num and sport now. Otherwise we will see
 626	 * in hash table socket with a funny identity.
 627	 */
 628	inet->inet_num = lport;
 629	inet->inet_sport = htons(lport);
 630	sk->sk_hash = hash;
 631	WARN_ON(!sk_unhashed(sk));
 632	__sk_nulls_add_node_rcu(sk, &head->chain);
 633	if (tw) {
 634		sk_nulls_del_node_init_rcu((struct sock *)tw);
 635		__NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
 636	}
 637	spin_unlock(lock);
 638	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 639
 640	if (twp) {
 641		*twp = tw;
 642	} else if (tw) {
 643		/* Silly. Should hash-dance instead... */
 644		inet_twsk_deschedule_put(tw);
 645	}
 646	return 0;
 647
 648not_unique:
 649	spin_unlock(lock);
 650	return -EADDRNOTAVAIL;
 651}
 652
 653static u64 inet_sk_port_offset(const struct sock *sk)
 654{
 655	const struct inet_sock *inet = inet_sk(sk);
 656
 657	return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
 658					  inet->inet_daddr,
 659					  inet->inet_dport);
 660}
 661
 662/* Searches for an exsiting socket in the ehash bucket list.
 663 * Returns true if found, false otherwise.
 664 */
 665static bool inet_ehash_lookup_by_sk(struct sock *sk,
 666				    struct hlist_nulls_head *list)
 667{
 668	const __portpair ports = INET_COMBINED_PORTS(sk->sk_dport, sk->sk_num);
 669	const int sdif = sk->sk_bound_dev_if;
 670	const int dif = sk->sk_bound_dev_if;
 671	const struct hlist_nulls_node *node;
 672	struct net *net = sock_net(sk);
 673	struct sock *esk;
 674
 675	INET_ADDR_COOKIE(acookie, sk->sk_daddr, sk->sk_rcv_saddr);
 676
 677	sk_nulls_for_each_rcu(esk, node, list) {
 678		if (esk->sk_hash != sk->sk_hash)
 679			continue;
 680		if (sk->sk_family == AF_INET) {
 681			if (unlikely(inet_match(net, esk, acookie,
 682						ports, dif, sdif))) {
 683				return true;
 684			}
 685		}
 686#if IS_ENABLED(CONFIG_IPV6)
 687		else if (sk->sk_family == AF_INET6) {
 688			if (unlikely(inet6_match(net, esk,
 689						 &sk->sk_v6_daddr,
 690						 &sk->sk_v6_rcv_saddr,
 691						 ports, dif, sdif))) {
 692				return true;
 693			}
 694		}
 695#endif
 696	}
 697	return false;
 698}
 699
 700/* Insert a socket into ehash, and eventually remove another one
 701 * (The another one can be a SYN_RECV or TIMEWAIT)
 702 * If an existing socket already exists, socket sk is not inserted,
 703 * and sets found_dup_sk parameter to true.
 704 */
 705bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk)
 706{
 707	struct inet_hashinfo *hashinfo = tcp_get_hashinfo(sk);
 708	struct inet_ehash_bucket *head;
 709	struct hlist_nulls_head *list;
 710	spinlock_t *lock;
 711	bool ret = true;
 712
 713	WARN_ON_ONCE(!sk_unhashed(sk));
 714
 715	sk->sk_hash = sk_ehashfn(sk);
 716	head = inet_ehash_bucket(hashinfo, sk->sk_hash);
 717	list = &head->chain;
 718	lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 719
 720	spin_lock(lock);
 721	if (osk) {
 722		WARN_ON_ONCE(sk->sk_hash != osk->sk_hash);
 723		ret = sk_nulls_replace_node_init_rcu(osk, sk);
 724		goto unlock;
 725	}
 726
 727	if (found_dup_sk) {
 728		*found_dup_sk = inet_ehash_lookup_by_sk(sk, list);
 729		if (*found_dup_sk)
 730			ret = false;
 731	}
 732
 733	if (ret)
 734		__sk_nulls_add_node_rcu(sk, list);
 735
 736unlock:
 737	spin_unlock(lock);
 738
 739	return ret;
 740}
 741
 742bool inet_ehash_nolisten(struct sock *sk, struct sock *osk, bool *found_dup_sk)
 743{
 744	bool ok = inet_ehash_insert(sk, osk, found_dup_sk);
 745
 746	if (ok) {
 747		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 748	} else {
 749		tcp_orphan_count_inc();
 750		inet_sk_set_state(sk, TCP_CLOSE);
 751		sock_set_flag(sk, SOCK_DEAD);
 752		inet_csk_destroy_sock(sk);
 753	}
 754	return ok;
 755}
 756EXPORT_IPV6_MOD(inet_ehash_nolisten);
 757
 758static int inet_reuseport_add_sock(struct sock *sk,
 759				   struct inet_listen_hashbucket *ilb)
 760{
 761	struct inet_bind_bucket *tb = inet_csk(sk)->icsk_bind_hash;
 762	const struct hlist_nulls_node *node;
 763	kuid_t uid = sk_uid(sk);
 764	struct sock *sk2;
 765
 766	sk_nulls_for_each_rcu(sk2, node, &ilb->nulls_head) {
 767		if (sk2 != sk &&
 768		    sk2->sk_family == sk->sk_family &&
 769		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 770		    sk2->sk_bound_dev_if == sk->sk_bound_dev_if &&
 771		    inet_csk(sk2)->icsk_bind_hash == tb &&
 772		    sk2->sk_reuseport && uid_eq(uid, sk_uid(sk2)) &&
 773		    inet_rcv_saddr_equal(sk, sk2, false))
 774			return reuseport_add_sock(sk, sk2,
 775						  inet_rcv_saddr_any(sk));
 776	}
 777
 778	return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
 779}
 780
 781int inet_hash(struct sock *sk)
 782{
 783	struct inet_hashinfo *hashinfo = tcp_get_hashinfo(sk);
 784	struct inet_listen_hashbucket *ilb2;
 785	int err = 0;
 786
 787	if (sk->sk_state == TCP_CLOSE)
 788		return 0;
 789
 790	if (sk->sk_state != TCP_LISTEN) {
 791		local_bh_disable();
 792		inet_ehash_nolisten(sk, NULL, NULL);
 793		local_bh_enable();
 794		return 0;
 795	}
 796	WARN_ON(!sk_unhashed(sk));
 797	ilb2 = inet_lhash2_bucket_sk(hashinfo, sk);
 798
 799	spin_lock(&ilb2->lock);
 800	if (sk->sk_reuseport) {
 801		err = inet_reuseport_add_sock(sk, ilb2);
 802		if (err)
 803			goto unlock;
 804	}
 805	sock_set_flag(sk, SOCK_RCU_FREE);
 806	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 807		sk->sk_family == AF_INET6)
 808		__sk_nulls_add_node_tail_rcu(sk, &ilb2->nulls_head);
 809	else
 810		__sk_nulls_add_node_rcu(sk, &ilb2->nulls_head);
 811	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 812unlock:
 813	spin_unlock(&ilb2->lock);
 814
 815	return err;
 816}
 817EXPORT_IPV6_MOD(inet_hash);
 818
 819void inet_unhash(struct sock *sk)
 820{
 821	struct inet_hashinfo *hashinfo = tcp_get_hashinfo(sk);
 822
 823	if (sk_unhashed(sk))
 824		return;
 825
 826	sock_rps_delete_flow(sk);
 827	if (sk->sk_state == TCP_LISTEN) {
 828		struct inet_listen_hashbucket *ilb2;
 829
 830		ilb2 = inet_lhash2_bucket_sk(hashinfo, sk);
 831		/* Don't disable bottom halves while acquiring the lock to
 832		 * avoid circular locking dependency on PREEMPT_RT.
 833		 */
 834		spin_lock(&ilb2->lock);
 835		if (rcu_access_pointer(sk->sk_reuseport_cb))
 836			reuseport_stop_listen_sock(sk);
 837
 838		__sk_nulls_del_node_init_rcu(sk);
 839		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 840		spin_unlock(&ilb2->lock);
 841	} else {
 842		spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
 843
 844		spin_lock_bh(lock);
 845		__sk_nulls_del_node_init_rcu(sk);
 846		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 847		spin_unlock_bh(lock);
 848	}
 849}
 850EXPORT_IPV6_MOD(inet_unhash);
 851
 852static bool inet_bind2_bucket_match(const struct inet_bind2_bucket *tb,
 853				    const struct net *net, unsigned short port,
 854				    int l3mdev, const struct sock *sk)
 855{
 856	if (!net_eq(ib2_net(tb), net) || tb->port != port ||
 857	    tb->l3mdev != l3mdev)
 858		return false;
 859
 860	return inet_bind2_bucket_addr_match(tb, sk);
 861}
 862
 863bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb, const struct net *net,
 864				      unsigned short port, int l3mdev, const struct sock *sk)
 865{
 866	if (!net_eq(ib2_net(tb), net) || tb->port != port ||
 867	    tb->l3mdev != l3mdev)
 868		return false;
 869
 870#if IS_ENABLED(CONFIG_IPV6)
 871	if (tb->addr_type == IPV6_ADDR_ANY)
 872		return true;
 873
 874	if (tb->addr_type != IPV6_ADDR_MAPPED)
 875		return false;
 876
 877	if (sk->sk_family == AF_INET6 &&
 878	    !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
 879		return false;
 880#endif
 881	return tb->rcv_saddr == 0;
 882}
 883
 884/* The socket's bhash2 hashbucket spinlock must be held when this is called */
 885struct inet_bind2_bucket *
 886inet_bind2_bucket_find(const struct inet_bind_hashbucket *head, const struct net *net,
 887		       unsigned short port, int l3mdev, const struct sock *sk)
 888{
 889	struct inet_bind2_bucket *bhash2 = NULL;
 890
 891	inet_bind_bucket_for_each(bhash2, &head->chain)
 892		if (inet_bind2_bucket_match(bhash2, net, port, l3mdev, sk))
 893			break;
 894
 895	return bhash2;
 896}
 897
 898struct inet_bind_hashbucket *
 899inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port)
 900{
 901	struct inet_hashinfo *hinfo = tcp_get_hashinfo(sk);
 902	u32 hash;
 903
 904#if IS_ENABLED(CONFIG_IPV6)
 905	if (sk->sk_family == AF_INET6)
 906		hash = ipv6_portaddr_hash(net, &in6addr_any, port);
 907	else
 908#endif
 909		hash = ipv4_portaddr_hash(net, 0, port);
 910
 911	return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
 912}
 913
 914static void inet_update_saddr(struct sock *sk, void *saddr, int family)
 915{
 916	if (family == AF_INET) {
 917		inet_sk(sk)->inet_saddr = *(__be32 *)saddr;
 918		sk_rcv_saddr_set(sk, inet_sk(sk)->inet_saddr);
 919	}
 920#if IS_ENABLED(CONFIG_IPV6)
 921	else {
 922		sk->sk_v6_rcv_saddr = *(struct in6_addr *)saddr;
 923	}
 924#endif
 925}
 926
 927static int __inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family, bool reset)
 928{
 929	struct inet_hashinfo *hinfo = tcp_get_hashinfo(sk);
 930	struct inet_bind_hashbucket *head, *head2;
 931	struct inet_bind2_bucket *tb2, *new_tb2;
 932	int l3mdev = inet_sk_bound_l3mdev(sk);
 933	int port = inet_sk(sk)->inet_num;
 934	struct net *net = sock_net(sk);
 935	int bhash;
 936
 937	if (!inet_csk(sk)->icsk_bind2_hash) {
 938		/* Not bind()ed before. */
 939		if (reset)
 940			inet_reset_saddr(sk);
 941		else
 942			inet_update_saddr(sk, saddr, family);
 943
 944		return 0;
 945	}
 946
 947	/* Allocate a bind2 bucket ahead of time to avoid permanently putting
 948	 * the bhash2 table in an inconsistent state if a new tb2 bucket
 949	 * allocation fails.
 950	 */
 951	new_tb2 = kmem_cache_alloc(hinfo->bind2_bucket_cachep, GFP_ATOMIC);
 952	if (!new_tb2) {
 953		if (reset) {
 954			/* The (INADDR_ANY, port) bucket might have already
 955			 * been freed, then we cannot fixup icsk_bind2_hash,
 956			 * so we give up and unlink sk from bhash/bhash2 not
 957			 * to leave inconsistency in bhash2.
 958			 */
 959			inet_put_port(sk);
 960			inet_reset_saddr(sk);
 961		}
 962
 963		return -ENOMEM;
 964	}
 965
 966	bhash = inet_bhashfn(net, port, hinfo->bhash_size);
 967	head = &hinfo->bhash[bhash];
 968	head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
 969
 970	/* If we change saddr locklessly, another thread
 971	 * iterating over bhash might see corrupted address.
 972	 */
 973	spin_lock_bh(&head->lock);
 974
 975	spin_lock(&head2->lock);
 976	__sk_del_bind_node(sk);
 977	inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep, inet_csk(sk)->icsk_bind2_hash);
 978	spin_unlock(&head2->lock);
 979
 980	if (reset)
 981		inet_reset_saddr(sk);
 982	else
 983		inet_update_saddr(sk, saddr, family);
 984
 985	head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
 986
 987	spin_lock(&head2->lock);
 988	tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
 989	if (!tb2) {
 990		tb2 = new_tb2;
 991		inet_bind2_bucket_init(tb2, net, head2, inet_csk(sk)->icsk_bind_hash, sk);
 992		if (sk_is_connect_bind(sk)) {
 993			tb2->fastreuse = -1;
 994			tb2->fastreuseport = -1;
 995		}
 996	}
 997	inet_csk(sk)->icsk_bind2_hash = tb2;
 998	sk_add_bind_node(sk, &tb2->owners);
 999	spin_unlock(&head2->lock);
1000
1001	spin_unlock_bh(&head->lock);
1002
1003	if (tb2 != new_tb2)
1004		kmem_cache_free(hinfo->bind2_bucket_cachep, new_tb2);
1005
1006	return 0;
1007}
1008
1009int inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family)
1010{
1011	return __inet_bhash2_update_saddr(sk, saddr, family, false);
1012}
1013EXPORT_IPV6_MOD(inet_bhash2_update_saddr);
1014
1015void inet_bhash2_reset_saddr(struct sock *sk)
1016{
1017	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1018		__inet_bhash2_update_saddr(sk, NULL, 0, true);
1019}
1020EXPORT_IPV6_MOD(inet_bhash2_reset_saddr);
1021
1022/* RFC 6056 3.3.4.  Algorithm 4: Double-Hash Port Selection Algorithm
1023 * Note that we use 32bit integers (vs RFC 'short integers')
1024 * because 2^16 is not a multiple of num_ephemeral and this
1025 * property might be used by clever attacker.
1026 *
1027 * RFC claims using TABLE_LENGTH=10 buckets gives an improvement, though
1028 * attacks were since demonstrated, thus we use 65536 by default instead
1029 * to really give more isolation and privacy, at the expense of 256kB
1030 * of kernel memory.
1031 */
1032#define INET_TABLE_PERTURB_SIZE (1 << CONFIG_INET_TABLE_PERTURB_ORDER)
1033static u32 *table_perturb;
1034
1035int __inet_hash_connect(struct inet_timewait_death_row *death_row,
1036		struct sock *sk, u64 port_offset,
1037		u32 hash_port0,
1038		int (*check_established)(struct inet_timewait_death_row *,
1039			struct sock *, __u16, struct inet_timewait_sock **,
1040			bool rcu_lookup, u32 hash))
1041{
1042	struct inet_hashinfo *hinfo = death_row->hashinfo;
1043	struct inet_bind_hashbucket *head, *head2;
1044	struct inet_timewait_sock *tw = NULL;
1045	int port = inet_sk(sk)->inet_num;
1046	struct net *net = sock_net(sk);
1047	struct inet_bind2_bucket *tb2;
1048	struct inet_bind_bucket *tb;
1049	bool tb_created = false;
1050	u32 remaining, offset;
1051	int ret, i, low, high;
1052	bool local_ports;
1053	int step, l3mdev;
1054	u32 index;
1055
1056	if (port) {
1057		local_bh_disable();
1058		ret = check_established(death_row, sk, port, NULL, false,
1059					hash_port0 + port);
1060		local_bh_enable();
1061		return ret;
1062	}
1063
1064	l3mdev = inet_sk_bound_l3mdev(sk);
1065
1066	local_ports = inet_sk_get_local_port_range(sk, &low, &high);
1067	step = local_ports ? 1 : 2;
1068
1069	high++; /* [32768, 60999] -> [32768, 61000[ */
1070	remaining = high - low;
1071	if (!local_ports && remaining > 1)
1072		remaining &= ~1U;
1073
1074	get_random_sleepable_once(table_perturb,
1075				  INET_TABLE_PERTURB_SIZE * sizeof(*table_perturb));
1076	index = port_offset & (INET_TABLE_PERTURB_SIZE - 1);
1077
1078	offset = READ_ONCE(table_perturb[index]) + (port_offset >> 32);
1079	offset %= remaining;
1080
1081	/* In first pass we try ports of @low parity.
1082	 * inet_csk_get_port() does the opposite choice.
1083	 */
1084	if (!local_ports)
1085		offset &= ~1U;
1086other_parity_scan:
1087	port = low + offset;
1088	for (i = 0; i < remaining; i += step, port += step) {
1089		if (unlikely(port >= high))
1090			port -= remaining;
1091		if (inet_is_local_reserved_port(net, port))
1092			continue;
1093		head = &hinfo->bhash[inet_bhashfn(net, port,
1094						  hinfo->bhash_size)];
1095		rcu_read_lock();
1096		hlist_for_each_entry_rcu(tb, &head->chain, node) {
1097			if (!inet_bind_bucket_match(tb, net, port, l3mdev))
1098				continue;
1099			if (tb->fastreuse >= 0 || tb->fastreuseport >= 0) {
1100				rcu_read_unlock();
1101				goto next_port;
1102			}
1103			if (!check_established(death_row, sk, port, &tw, true,
1104					       hash_port0 + port))
1105				break;
1106			rcu_read_unlock();
1107			goto next_port;
1108		}
1109		rcu_read_unlock();
1110
1111		spin_lock_bh(&head->lock);
1112
1113		/* Does not bother with rcv_saddr checks, because
1114		 * the established check is already unique enough.
1115		 */
1116		inet_bind_bucket_for_each(tb, &head->chain) {
1117			if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
1118				if (tb->fastreuse >= 0 ||
1119				    tb->fastreuseport >= 0)
1120					goto next_port_unlock;
1121				WARN_ON(hlist_empty(&tb->bhash2));
1122				if (!check_established(death_row, sk,
1123						       port, &tw, false,
1124						       hash_port0 + port))
1125					goto ok;
1126				goto next_port_unlock;
1127			}
1128		}
1129
1130		tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
1131					     net, head, port, l3mdev);
1132		if (!tb) {
1133			spin_unlock_bh(&head->lock);
1134			return -ENOMEM;
1135		}
1136		tb_created = true;
1137		tb->fastreuse = -1;
1138		tb->fastreuseport = -1;
1139		goto ok;
1140next_port_unlock:
1141		spin_unlock_bh(&head->lock);
1142next_port:
1143		cond_resched();
1144	}
1145
1146	if (!local_ports) {
1147		offset++;
1148		if ((offset & 1) && remaining > 1)
1149			goto other_parity_scan;
1150	}
1151	return -EADDRNOTAVAIL;
1152
1153ok:
1154	/* Find the corresponding tb2 bucket since we need to
1155	 * add the socket to the bhash2 table as well
1156	 */
1157	head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
1158	spin_lock(&head2->lock);
1159
1160	tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
1161	if (!tb2) {
1162		tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep, net,
1163					       head2, tb, sk);
1164		if (!tb2)
1165			goto error;
1166		tb2->fastreuse = -1;
1167		tb2->fastreuseport = -1;
1168	}
1169
1170	/* Here we want to add a little bit of randomness to the next source
1171	 * port that will be chosen. We use a max() with a random here so that
1172	 * on low contention the randomness is maximal and on high contention
1173	 * it may be inexistent.
1174	 */
1175	i = max_t(int, i, get_random_u32_below(8) * step);
1176	WRITE_ONCE(table_perturb[index], READ_ONCE(table_perturb[index]) + i + step);
1177
1178	/* Head lock still held and bh's disabled */
1179	inet_bind_hash(sk, tb, tb2, port);
1180	sk->sk_userlocks |= SOCK_CONNECT_BIND;
1181
1182	if (sk_unhashed(sk)) {
1183		inet_sk(sk)->inet_sport = htons(port);
1184		inet_ehash_nolisten(sk, (struct sock *)tw, NULL);
1185	}
1186	if (tw)
1187		inet_twsk_bind_unhash(tw, hinfo);
1188
1189	spin_unlock(&head2->lock);
1190	spin_unlock(&head->lock);
1191
1192	if (tw)
1193		inet_twsk_deschedule_put(tw);
1194	local_bh_enable();
1195	return 0;
1196
1197error:
1198	if (sk_hashed(sk)) {
1199		spinlock_t *lock = inet_ehash_lockp(hinfo, sk->sk_hash);
1200
1201		sock_prot_inuse_add(net, sk->sk_prot, -1);
1202
1203		spin_lock(lock);
1204		__sk_nulls_del_node_init_rcu(sk);
1205		spin_unlock(lock);
1206
1207		sk->sk_hash = 0;
1208		inet_sk(sk)->inet_sport = 0;
1209		WRITE_ONCE(inet_sk(sk)->inet_num, 0);
1210
1211		if (tw)
1212			inet_twsk_bind_unhash(tw, hinfo);
1213	}
1214
1215	spin_unlock(&head2->lock);
1216	if (tb_created)
1217		inet_bind_bucket_destroy(tb);
1218	spin_unlock(&head->lock);
1219
1220	if (tw)
1221		inet_twsk_deschedule_put(tw);
1222
1223	local_bh_enable();
1224
1225	return -ENOMEM;
1226}
1227
1228/*
1229 * Bind a port for a connect operation and hash it.
1230 */
1231int inet_hash_connect(struct inet_timewait_death_row *death_row,
1232		      struct sock *sk)
1233{
1234	const struct inet_sock *inet = inet_sk(sk);
1235	const struct net *net = sock_net(sk);
1236	u64 port_offset = 0;
1237	u32 hash_port0;
1238
1239	if (!inet_sk(sk)->inet_num)
1240		port_offset = inet_sk_port_offset(sk);
1241
1242	hash_port0 = inet_ehashfn(net, inet->inet_rcv_saddr, 0,
1243				  inet->inet_daddr, inet->inet_dport);
1244
1245	return __inet_hash_connect(death_row, sk, port_offset, hash_port0,
1246				   __inet_check_established);
1247}
1248
1249static void init_hashinfo_lhash2(struct inet_hashinfo *h)
1250{
1251	int i;
1252
1253	for (i = 0; i <= h->lhash2_mask; i++) {
1254		spin_lock_init(&h->lhash2[i].lock);
1255		INIT_HLIST_NULLS_HEAD(&h->lhash2[i].nulls_head,
1256				      i + LISTENING_NULLS_BASE);
1257	}
1258}
1259
1260void __init inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
1261				unsigned long numentries, int scale,
1262				unsigned long low_limit,
1263				unsigned long high_limit)
1264{
1265	h->lhash2 = alloc_large_system_hash(name,
1266					    sizeof(*h->lhash2),
1267					    numentries,
1268					    scale,
1269					    0,
1270					    NULL,
1271					    &h->lhash2_mask,
1272					    low_limit,
1273					    high_limit);
1274	init_hashinfo_lhash2(h);
1275
1276	/* this one is used for source ports of outgoing connections */
1277	table_perturb = alloc_large_system_hash("Table-perturb",
1278						sizeof(*table_perturb),
1279						INET_TABLE_PERTURB_SIZE,
1280						0, 0, NULL, NULL,
1281						INET_TABLE_PERTURB_SIZE,
1282						INET_TABLE_PERTURB_SIZE);
1283}
1284
1285int inet_hashinfo2_init_mod(struct inet_hashinfo *h)
1286{
1287	h->lhash2 = kmalloc_array(INET_LHTABLE_SIZE, sizeof(*h->lhash2), GFP_KERNEL);
1288	if (!h->lhash2)
1289		return -ENOMEM;
1290
1291	h->lhash2_mask = INET_LHTABLE_SIZE - 1;
1292	/* INET_LHTABLE_SIZE must be a power of 2 */
1293	BUG_ON(INET_LHTABLE_SIZE & h->lhash2_mask);
1294
1295	init_hashinfo_lhash2(h);
1296	return 0;
1297}
1298
1299int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
1300{
1301	unsigned int locksz = sizeof(spinlock_t);
1302	unsigned int i, nblocks = 1;
1303	spinlock_t *ptr = NULL;
1304
1305	if (locksz == 0)
1306		goto set_mask;
1307
1308	/* Allocate 2 cache lines or at least one spinlock per cpu. */
1309	nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U) * num_possible_cpus();
1310
1311	/* At least one page per NUMA node. */
1312	nblocks = max(nblocks, num_online_nodes() * PAGE_SIZE / locksz);
1313
1314	nblocks = roundup_pow_of_two(nblocks);
1315
1316	/* No more locks than number of hash buckets. */
1317	nblocks = min(nblocks, hashinfo->ehash_mask + 1);
1318
1319	if (num_online_nodes() > 1) {
1320		/* Use vmalloc() to allow NUMA policy to spread pages
1321		 * on all available nodes if desired.
1322		 */
1323		ptr = vmalloc_array(nblocks, locksz);
1324	}
1325	if (!ptr) {
1326		ptr = kvmalloc_array(nblocks, locksz, GFP_KERNEL);
1327		if (!ptr)
1328			return -ENOMEM;
1329	}
1330	for (i = 0; i < nblocks; i++)
1331		spin_lock_init(&ptr[i]);
1332	hashinfo->ehash_locks = ptr;
1333set_mask:
1334	hashinfo->ehash_locks_mask = nblocks - 1;
1335	return 0;
1336}
1337
1338struct inet_hashinfo *inet_pernet_hashinfo_alloc(struct inet_hashinfo *hashinfo,
1339						 unsigned int ehash_entries)
1340{
1341	struct inet_hashinfo *new_hashinfo;
1342	int i;
1343
1344	new_hashinfo = kmemdup(hashinfo, sizeof(*hashinfo), GFP_KERNEL);
1345	if (!new_hashinfo)
1346		goto err;
1347
1348	new_hashinfo->ehash = vmalloc_huge(ehash_entries * sizeof(struct inet_ehash_bucket),
1349					   GFP_KERNEL_ACCOUNT);
1350	if (!new_hashinfo->ehash)
1351		goto free_hashinfo;
1352
1353	new_hashinfo->ehash_mask = ehash_entries - 1;
1354
1355	if (inet_ehash_locks_alloc(new_hashinfo))
1356		goto free_ehash;
1357
1358	for (i = 0; i < ehash_entries; i++)
1359		INIT_HLIST_NULLS_HEAD(&new_hashinfo->ehash[i].chain, i);
1360
1361	new_hashinfo->pernet = true;
1362
1363	return new_hashinfo;
1364
1365free_ehash:
1366	vfree(new_hashinfo->ehash);
1367free_hashinfo:
1368	kfree(new_hashinfo);
1369err:
1370	return NULL;
1371}
1372
1373void inet_pernet_hashinfo_free(struct inet_hashinfo *hashinfo)
1374{
1375	if (!hashinfo->pernet)
1376		return;
1377
1378	inet_ehash_locks_free(hashinfo);
1379	vfree(hashinfo->ehash);
1380	kfree(hashinfo);
1381}