inet_hashtables.c 15.5 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
/*
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		Generic INET transport hashtables
 *
 * Authors:	Lotsa people, from code originally in tcp
 *
 *	This program is free software; you can redistribute it and/or
 *      modify it under the terms of the GNU General Public License
 *      as published by the Free Software Foundation; either version
 *      2 of the License, or (at your option) any later version.
 */

16
#include <linux/module.h>
17
#include <linux/random.h>
18
#include <linux/sched.h>
19
#include <linux/slab.h>
20
#include <linux/wait.h>
21

22
#include <net/inet_connection_sock.h>
23
#include <net/inet_hashtables.h>
24
#include <net/secure_seq.h>
25
#include <net/ip.h>
26

27 28 29
static u32 inet_ehashfn(const struct net *net, const __be32 laddr,
			const __u16 lport, const __be32 faddr,
			const __be16 fport)
30
{
31 32 33 34
	static u32 inet_ehash_secret __read_mostly;

	net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret));

35 36 37 38 39
	return __inet_ehashfn(laddr, lport, faddr, fport,
			      inet_ehash_secret + net_hash_mix(net));
}


40
u32 sk_ehashfn(const struct sock *sk)
41
{
42 43 44
	return inet_ehashfn(sock_net(sk),
			    sk->sk_rcv_saddr, sk->sk_num,
			    sk->sk_daddr, sk->sk_dport);
45 46
}

47 48 49 50
/*
 * Allocate and initialize a new local port bind bucket.
 * The bindhash mutex for snum's hash chain must be held here.
 */
51
struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
52
						 struct net *net,
53 54 55
						 struct inet_bind_hashbucket *head,
						 const unsigned short snum)
{
56
	struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
57 58

	if (tb != NULL) {
59
		write_pnet(&tb->ib_net, net);
60 61
		tb->port      = snum;
		tb->fastreuse = 0;
62
		tb->fastreuseport = 0;
63
		tb->num_owners = 0;
64 65 66 67 68 69 70 71 72
		INIT_HLIST_HEAD(&tb->owners);
		hlist_add_head(&tb->node, &head->chain);
	}
	return tb;
}

/*
 * Caller must hold hashbucket lock for this tb with local BH disabled
 */
73
void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
74 75 76 77 78 79
{
	if (hlist_empty(&tb->owners)) {
		__hlist_del(&tb->node);
		kmem_cache_free(cachep, tb);
	}
}
80 81 82 83

void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
		    const unsigned short snum)
{
84 85
	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;

86
	atomic_inc(&hashinfo->bsockets);
87

E
Eric Dumazet 已提交
88
	inet_sk(sk)->inet_num = snum;
89
	sk_add_bind_node(sk, &tb->owners);
90
	tb->num_owners++;
91
	inet_csk(sk)->icsk_bind_hash = tb;
92 93 94 95 96
}

/*
 * Get rid of any references to a local port held by the given sock.
 */
97
static void __inet_put_port(struct sock *sk)
98
{
99
	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
E
Eric Dumazet 已提交
100
	const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num,
101
			hashinfo->bhash_size);
102 103 104
	struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
	struct inet_bind_bucket *tb;

105
	atomic_dec(&hashinfo->bsockets);
106

107
	spin_lock(&head->lock);
108
	tb = inet_csk(sk)->icsk_bind_hash;
109
	__sk_del_bind_node(sk);
110
	tb->num_owners--;
111
	inet_csk(sk)->icsk_bind_hash = NULL;
E
Eric Dumazet 已提交
112
	inet_sk(sk)->inet_num = 0;
113 114 115 116
	inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
	spin_unlock(&head->lock);
}

117
void inet_put_port(struct sock *sk)
118 119
{
	local_bh_disable();
120
	__inet_put_port(sk);
121 122 123
	local_bh_enable();
}
EXPORT_SYMBOL(inet_put_port);
124

125
int __inet_inherit_port(struct sock *sk, struct sock *child)
126 127
{
	struct inet_hashinfo *table = sk->sk_prot->h.hashinfo;
128 129
	unsigned short port = inet_sk(child)->inet_num;
	const int bhash = inet_bhashfn(sock_net(sk), port,
130
			table->bhash_size);
131 132 133 134 135
	struct inet_bind_hashbucket *head = &table->bhash[bhash];
	struct inet_bind_bucket *tb;

	spin_lock(&head->lock);
	tb = inet_csk(sk)->icsk_bind_hash;
136 137 138 139 140 141
	if (tb->port != port) {
		/* NOTE: using tproxy and redirecting skbs to a proxy
		 * on a different listener port breaks the assumption
		 * that the listener socket's icsk_bind_hash is the same
		 * as that of the child socket. We have to look up or
		 * create a new bind bucket for the child here. */
142
		inet_bind_bucket_for_each(tb, &head->chain) {
143 144 145 146
			if (net_eq(ib_net(tb), sock_net(sk)) &&
			    tb->port == port)
				break;
		}
147
		if (!tb) {
148 149 150 151 152 153 154 155
			tb = inet_bind_bucket_create(table->bind_bucket_cachep,
						     sock_net(sk), head, port);
			if (!tb) {
				spin_unlock(&head->lock);
				return -ENOMEM;
			}
		}
	}
156
	inet_bind_hash(child, tb, port);
157
	spin_unlock(&head->lock);
158 159

	return 0;
160 161 162
}
EXPORT_SYMBOL_GPL(__inet_inherit_port);

163 164 165 166 167 168 169
static inline int compute_score(struct sock *sk, struct net *net,
				const unsigned short hnum, const __be32 daddr,
				const int dif)
{
	int score = -1;
	struct inet_sock *inet = inet_sk(sk);

E
Eric Dumazet 已提交
170
	if (net_eq(sock_net(sk), net) && inet->inet_num == hnum &&
171
			!ipv6_only_sock(sk)) {
E
Eric Dumazet 已提交
172
		__be32 rcv_saddr = inet->inet_rcv_saddr;
173
		score = sk->sk_family == PF_INET ? 2 : 1;
174 175 176
		if (rcv_saddr) {
			if (rcv_saddr != daddr)
				return -1;
177
			score += 4;
178 179 180 181
		}
		if (sk->sk_bound_dev_if) {
			if (sk->sk_bound_dev_if != dif)
				return -1;
182
			score += 4;
183 184 185 186 187
		}
	}
	return score;
}

188 189 190 191 192 193
/*
 * Don't inline this cruft. Here are some nice properties to exploit here. The
 * BSD API does not allow a listening sock to specify the remote port nor the
 * remote address for the connection. So always assume those are both
 * wildcarded during the search since they can never be otherwise.
 */
194

195

196 197
struct sock *__inet_lookup_listener(struct net *net,
				    struct inet_hashinfo *hashinfo,
198
				    const __be32 saddr, __be16 sport,
199
				    const __be32 daddr, const unsigned short hnum,
200
				    const int dif)
201
{
202 203 204 205
	struct sock *sk, *result;
	struct hlist_nulls_node *node;
	unsigned int hash = inet_lhashfn(net, hnum);
	struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
206 207
	int score, hiscore, matches = 0, reuseport = 0;
	u32 phash = 0;
208

209 210 211
	rcu_read_lock();
begin:
	result = NULL;
212
	hiscore = 0;
213 214 215 216 217
	sk_nulls_for_each_rcu(sk, node, &ilb->head) {
		score = compute_score(sk, net, hnum, daddr, dif);
		if (score > hiscore) {
			result = sk;
			hiscore = score;
218 219 220 221 222 223 224 225
			reuseport = sk->sk_reuseport;
			if (reuseport) {
				phash = inet_ehashfn(net, daddr, hnum,
						     saddr, sport);
				matches = 1;
			}
		} else if (score == hiscore && reuseport) {
			matches++;
226
			if (reciprocal_scale(phash, matches) == 0)
227 228
				result = sk;
			phash = next_pseudo_random32(phash);
229
		}
230
	}
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245
	/*
	 * if the nulls value we got at the end of this lookup is
	 * not the expected one, we must restart lookup.
	 * We probably met an item that was moved to another chain.
	 */
	if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
		goto begin;
	if (result) {
		if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
			result = NULL;
		else if (unlikely(compute_score(result, net, hnum, daddr,
				  dif) < hiscore)) {
			sock_put(result);
			goto begin;
		}
246
	}
247 248
	rcu_read_unlock();
	return result;
249
}
250
EXPORT_SYMBOL_GPL(__inet_lookup_listener);
251

E
Eric Dumazet 已提交
252 253 254 255 256 257 258 259
/* All sockets share common refcount, but have different destructors */
void sock_gen_put(struct sock *sk)
{
	if (!atomic_dec_and_test(&sk->sk_refcnt))
		return;

	if (sk->sk_state == TCP_TIME_WAIT)
		inet_twsk_free(inet_twsk(sk));
260 261
	else if (sk->sk_state == TCP_NEW_SYN_RECV)
		reqsk_free(inet_reqsk(sk));
E
Eric Dumazet 已提交
262 263 264 265 266
	else
		sk_free(sk);
}
EXPORT_SYMBOL_GPL(sock_gen_put);

267 268 269 270 271 272
void sock_edemux(struct sk_buff *skb)
{
	sock_gen_put(skb->sk);
}
EXPORT_SYMBOL(sock_edemux);

D
Daniel Baluta 已提交
273
struct sock *__inet_lookup_established(struct net *net,
274
				  struct inet_hashinfo *hashinfo,
275 276 277 278
				  const __be32 saddr, const __be16 sport,
				  const __be32 daddr, const u16 hnum,
				  const int dif)
{
279
	INET_ADDR_COOKIE(acookie, saddr, daddr);
280 281
	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
	struct sock *sk;
282
	const struct hlist_nulls_node *node;
283 284 285
	/* Optimize here for direct hit, only listening connections can
	 * have wildcards anyways.
	 */
286
	unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
287
	unsigned int slot = hash & hashinfo->ehash_mask;
288
	struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
289

290 291 292
	rcu_read_lock();
begin:
	sk_nulls_for_each_rcu(sk, node, &head->chain) {
293 294 295 296
		if (sk->sk_hash != hash)
			continue;
		if (likely(INET_MATCH(sk, net, acookie,
				      saddr, daddr, ports, dif))) {
297
			if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
E
Eric Dumazet 已提交
298
				goto out;
299 300
			if (unlikely(!INET_MATCH(sk, net, acookie,
						 saddr, daddr, ports, dif))) {
E
Eric Dumazet 已提交
301
				sock_gen_put(sk);
302 303
				goto begin;
			}
E
Eric Dumazet 已提交
304
			goto found;
305
		}
306
	}
307 308 309 310 311 312 313
	/*
	 * if the nulls value we got at the end of this lookup is
	 * not the expected one, we must restart lookup.
	 * We probably met an item that was moved to another chain.
	 */
	if (get_nulls_value(node) != slot)
		goto begin;
314
out:
E
Eric Dumazet 已提交
315 316
	sk = NULL;
found:
317
	rcu_read_unlock();
318 319 320 321
	return sk;
}
EXPORT_SYMBOL_GPL(__inet_lookup_established);

322 323 324 325 326 327 328
/* called with local bh disabled */
static int __inet_check_established(struct inet_timewait_death_row *death_row,
				    struct sock *sk, __u16 lport,
				    struct inet_timewait_sock **twp)
{
	struct inet_hashinfo *hinfo = death_row->hashinfo;
	struct inet_sock *inet = inet_sk(sk);
E
Eric Dumazet 已提交
329 330
	__be32 daddr = inet->inet_rcv_saddr;
	__be32 saddr = inet->inet_daddr;
331
	int dif = sk->sk_bound_dev_if;
332
	INET_ADDR_COOKIE(acookie, saddr, daddr);
E
Eric Dumazet 已提交
333
	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
334
	struct net *net = sock_net(sk);
E
Eric Dumazet 已提交
335 336
	unsigned int hash = inet_ehashfn(net, daddr, lport,
					 saddr, inet->inet_dport);
337
	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
338
	spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
339
	struct sock *sk2;
340
	const struct hlist_nulls_node *node;
E
Eric Dumazet 已提交
341
	struct inet_timewait_sock *tw = NULL;
342
	int twrefcnt = 0;
343

344
	spin_lock(lock);
345

346
	sk_nulls_for_each(sk2, node, &head->chain) {
347 348
		if (sk2->sk_hash != hash)
			continue;
E
Eric Dumazet 已提交
349

350
		if (likely(INET_MATCH(sk2, net, acookie,
E
Eric Dumazet 已提交
351 352 353 354 355 356
					 saddr, daddr, ports, dif))) {
			if (sk2->sk_state == TCP_TIME_WAIT) {
				tw = inet_twsk(sk2);
				if (twsk_unique(sk, sk2, twp))
					break;
			}
357
			goto not_unique;
E
Eric Dumazet 已提交
358
		}
359 360 361
	}

	/* Must record num and sport now. Otherwise we will see
E
Eric Dumazet 已提交
362 363
	 * in hash table socket with a funny identity.
	 */
E
Eric Dumazet 已提交
364 365
	inet->inet_num = lport;
	inet->inet_sport = htons(lport);
366
	sk->sk_hash = hash;
367
	WARN_ON(!sk_unhashed(sk));
368
	__sk_nulls_add_node_rcu(sk, &head->chain);
369 370 371 372
	if (tw) {
		twrefcnt = inet_twsk_unhash(tw);
		NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
	}
373
	spin_unlock(lock);
374 375
	if (twrefcnt)
		inet_twsk_put(tw);
376
	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
377 378 379 380 381 382 383 384 385 386 387 388

	if (twp) {
		*twp = tw;
	} else if (tw) {
		/* Silly. Should hash-dance instead... */
		inet_twsk_deschedule(tw, death_row);

		inet_twsk_put(tw);
	}
	return 0;

not_unique:
389
	spin_unlock(lock);
390 391 392 393 394 395
	return -EADDRNOTAVAIL;
}

static inline u32 inet_sk_port_offset(const struct sock *sk)
{
	const struct inet_sock *inet = inet_sk(sk);
E
Eric Dumazet 已提交
396 397 398
	return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
					  inet->inet_daddr,
					  inet->inet_dport);
399 400
}

401
int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw)
402
{
403
	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
404
	struct hlist_nulls_head *list;
405
	struct inet_ehash_bucket *head;
406
	spinlock_t *lock;
407
	int twrefcnt = 0;
408

409
	WARN_ON(!sk_unhashed(sk));
410

411
	sk->sk_hash = sk_ehashfn(sk);
412 413 414 415
	head = inet_ehash_bucket(hashinfo, sk->sk_hash);
	list = &head->chain;
	lock = inet_ehash_lockp(hashinfo, sk->sk_hash);

416
	spin_lock(lock);
417
	__sk_nulls_add_node_rcu(sk, list);
418 419 420 421
	if (tw) {
		WARN_ON(sk->sk_hash != tw->tw_hash);
		twrefcnt = inet_twsk_unhash(tw);
	}
422
	spin_unlock(lock);
423
	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
424
	return twrefcnt;
425 426 427
}
EXPORT_SYMBOL_GPL(__inet_hash_nolisten);

428
static void __inet_hash(struct sock *sk)
429
{
430
	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
431
	struct inet_listen_hashbucket *ilb;
432 433

	if (sk->sk_state != TCP_LISTEN) {
434
		__inet_hash_nolisten(sk, NULL);
435 436 437
		return;
	}

438
	WARN_ON(!sk_unhashed(sk));
439
	ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
440

441
	spin_lock(&ilb->lock);
442
	__sk_nulls_add_node_rcu(sk, &ilb->head);
443
	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
444
	spin_unlock(&ilb->lock);
445
}
446 447 448 449 450 451 452 453 454 455 456 457 458

void inet_hash(struct sock *sk)
{
	if (sk->sk_state != TCP_CLOSE) {
		local_bh_disable();
		__inet_hash(sk);
		local_bh_enable();
	}
}
EXPORT_SYMBOL_GPL(inet_hash);

void inet_unhash(struct sock *sk)
{
459
	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
460 461
	spinlock_t *lock;
	int done;
462 463

	if (sk_unhashed(sk))
464
		return;
465

466 467 468 469
	if (sk->sk_state == TCP_LISTEN)
		lock = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)].lock;
	else
		lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
470

471
	spin_lock_bh(lock);
C
Camelia Groza 已提交
472
	done = __sk_nulls_del_node_init_rcu(sk);
473 474
	if (done)
		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
475
	spin_unlock_bh(lock);
476 477
}
EXPORT_SYMBOL_GPL(inet_unhash);
478

479
int __inet_hash_connect(struct inet_timewait_death_row *death_row,
480
		struct sock *sk, u32 port_offset,
481 482
		int (*check_established)(struct inet_timewait_death_row *,
			struct sock *, __u16, struct inet_timewait_sock **),
483
		int (*hash)(struct sock *sk, struct inet_timewait_sock *twp))
484 485
{
	struct inet_hashinfo *hinfo = death_row->hashinfo;
E
Eric Dumazet 已提交
486
	const unsigned short snum = inet_sk(sk)->inet_num;
487 488
	struct inet_bind_hashbucket *head;
	struct inet_bind_bucket *tb;
489
	int ret;
490
	struct net *net = sock_net(sk);
491
	int twrefcnt = 1;
492

493
	if (!snum) {
494
		int i, remaining, low, high, port;
495
		static u32 hint;
496
		u32 offset = hint + port_offset;
497
		struct inet_timewait_sock *tw = NULL;
498

499
		inet_get_local_port_range(net, &low, &high);
500
		remaining = (high - low) + 1;
501

502
		local_bh_disable();
503 504
		for (i = 1; i <= remaining; i++) {
			port = low + (i + offset) % remaining;
505
			if (inet_is_local_reserved_port(net, port))
506
				continue;
507 508
			head = &hinfo->bhash[inet_bhashfn(net, port,
					hinfo->bhash_size)];
509
			spin_lock(&head->lock);
510

511 512 513 514
			/* Does not bother with rcv_saddr checks,
			 * because the established check is already
			 * unique enough.
			 */
515
			inet_bind_bucket_for_each(tb, &head->chain) {
O
Octavian Purdila 已提交
516 517
				if (net_eq(ib_net(tb), net) &&
				    tb->port == port) {
518 519
					if (tb->fastreuse >= 0 ||
					    tb->fastreuseport >= 0)
520
						goto next_port;
521
					WARN_ON(hlist_empty(&tb->owners));
522 523
					if (!check_established(death_row, sk,
								port, &tw))
524 525 526 527 528
						goto ok;
					goto next_port;
				}
			}

529 530
			tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
					net, head, port);
531 532 533 534 535
			if (!tb) {
				spin_unlock(&head->lock);
				break;
			}
			tb->fastreuse = -1;
536
			tb->fastreuseport = -1;
537 538 539 540 541 542 543 544
			goto ok;

		next_port:
			spin_unlock(&head->lock);
		}
		local_bh_enable();

		return -EADDRNOTAVAIL;
545 546 547 548

ok:
		hint += i;

549 550
		/* Head lock still held and bh's disabled */
		inet_bind_hash(sk, tb, port);
551
		if (sk_unhashed(sk)) {
E
Eric Dumazet 已提交
552
			inet_sk(sk)->inet_sport = htons(port);
553
			twrefcnt += hash(sk, tw);
554
		}
555 556
		if (tw)
			twrefcnt += inet_twsk_bind_unhash(tw, hinfo);
557
		spin_unlock(&head->lock);
558

559 560
		if (tw) {
			inet_twsk_deschedule(tw, death_row);
561 562 563 564
			while (twrefcnt) {
				twrefcnt--;
				inet_twsk_put(tw);
			}
565
		}
566 567 568

		ret = 0;
		goto out;
569
	}
570

571
	head = &hinfo->bhash[inet_bhashfn(net, snum, hinfo->bhash_size)];
572
	tb  = inet_csk(sk)->icsk_bind_hash;
573 574
	spin_lock_bh(&head->lock);
	if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
575
		hash(sk, NULL);
576 577 578 579 580
		spin_unlock_bh(&head->lock);
		return 0;
	} else {
		spin_unlock(&head->lock);
		/* No definite answer... Walk to established hash table */
581
		ret = check_established(death_row, sk, snum, NULL);
582 583 584 585 586
out:
		local_bh_enable();
		return ret;
	}
}
587 588 589 590 591 592 593

/*
 * Bind a port for a connect operation and hash it.
 */
int inet_hash_connect(struct inet_timewait_death_row *death_row,
		      struct sock *sk)
{
594
	return __inet_hash_connect(death_row, sk, inet_sk_port_offset(sk),
595 596
			__inet_check_established, __inet_hash_nolisten);
}
597
EXPORT_SYMBOL_GPL(inet_hash_connect);
598 599 600 601 602

void inet_hashinfo_init(struct inet_hashinfo *h)
{
	int i;

603
	atomic_set(&h->bsockets, 0);
604
	for (i = 0; i < INET_LHTABLE_SIZE; i++) {
605
		spin_lock_init(&h->listening_hash[i].lock);
606 607 608
		INIT_HLIST_NULLS_HEAD(&h->listening_hash[i].head,
				      i + LISTENING_NULLS_BASE);
		}
609 610
}
EXPORT_SYMBOL_GPL(inet_hashinfo_init);