2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Generic INET transport hashtables
8 * Authors: Lotsa people, from code originally in tcp
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
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>
22 #include <net/inet_connection_sock.h>
23 #include <net/inet_hashtables.h>
27 * Allocate and initialize a new local port bind bucket.
28 * The bindhash mutex for snum's hash chain must be held here.
30 struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
31 struct inet_bind_hashbucket *head,
32 const unsigned short snum)
34 struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
39 INIT_HLIST_HEAD(&tb->owners);
40 hlist_add_head(&tb->node, &head->chain);
46 * Caller must hold hashbucket lock for this tb with local BH disabled
48 void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
50 if (hlist_empty(&tb->owners)) {
51 __hlist_del(&tb->node);
52 kmem_cache_free(cachep, tb);
56 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
57 const unsigned short snum)
59 inet_sk(sk)->num = snum;
60 sk_add_bind_node(sk, &tb->owners);
61 inet_csk(sk)->icsk_bind_hash = tb;
65 * Get rid of any references to a local port held by the given sock.
67 static void __inet_put_port(struct inet_hashinfo *hashinfo, struct sock *sk)
69 const int bhash = inet_bhashfn(inet_sk(sk)->num, hashinfo->bhash_size);
70 struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
71 struct inet_bind_bucket *tb;
73 spin_lock(&head->lock);
74 tb = inet_csk(sk)->icsk_bind_hash;
75 __sk_del_bind_node(sk);
76 inet_csk(sk)->icsk_bind_hash = NULL;
78 inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
79 spin_unlock(&head->lock);
82 void inet_put_port(struct inet_hashinfo *hashinfo, struct sock *sk)
85 __inet_put_port(hashinfo, sk);
89 EXPORT_SYMBOL(inet_put_port);
92 * This lock without WQ_FLAG_EXCLUSIVE is good on UP and it can be very bad on SMP.
93 * Look, when several writers sleep and reader wakes them up, all but one
94 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
95 * this, _but_ remember, it adds useless work on UP machines (wake up each
96 * exclusive lock release). It should be ifdefed really.
98 void inet_listen_wlock(struct inet_hashinfo *hashinfo)
99 __acquires(hashinfo->lhash_lock)
101 write_lock(&hashinfo->lhash_lock);
103 if (atomic_read(&hashinfo->lhash_users)) {
107 prepare_to_wait_exclusive(&hashinfo->lhash_wait,
108 &wait, TASK_UNINTERRUPTIBLE);
109 if (!atomic_read(&hashinfo->lhash_users))
111 write_unlock_bh(&hashinfo->lhash_lock);
113 write_lock_bh(&hashinfo->lhash_lock);
116 finish_wait(&hashinfo->lhash_wait, &wait);
120 EXPORT_SYMBOL(inet_listen_wlock);
123 * Don't inline this cruft. Here are some nice properties to exploit here. The
124 * BSD API does not allow a listening sock to specify the remote port nor the
125 * remote address for the connection. So always assume those are both
126 * wildcarded during the search since they can never be otherwise.
128 static struct sock *inet_lookup_listener_slow(const struct hlist_head *head,
130 const unsigned short hnum,
133 struct sock *result = NULL, *sk;
134 const struct hlist_node *node;
137 sk_for_each(sk, node, head) {
138 const struct inet_sock *inet = inet_sk(sk);
140 if (inet->num == hnum && !ipv6_only_sock(sk)) {
141 const __be32 rcv_saddr = inet->rcv_saddr;
142 int score = sk->sk_family == PF_INET ? 1 : 0;
145 if (rcv_saddr != daddr)
149 if (sk->sk_bound_dev_if) {
150 if (sk->sk_bound_dev_if != dif)
156 if (score > hiscore) {
165 /* Optimize the common listener case. */
166 struct sock *__inet_lookup_listener(struct inet_hashinfo *hashinfo,
167 const __be32 daddr, const unsigned short hnum,
170 struct sock *sk = NULL;
171 const struct hlist_head *head;
173 read_lock(&hashinfo->lhash_lock);
174 head = &hashinfo->listening_hash[inet_lhashfn(hnum)];
175 if (!hlist_empty(head)) {
176 const struct inet_sock *inet = inet_sk((sk = __sk_head(head)));
178 if (inet->num == hnum && !sk->sk_node.next &&
179 (!inet->rcv_saddr || inet->rcv_saddr == daddr) &&
180 (sk->sk_family == PF_INET || !ipv6_only_sock(sk)) &&
181 !sk->sk_bound_dev_if)
183 sk = inet_lookup_listener_slow(head, daddr, hnum, dif);
189 read_unlock(&hashinfo->lhash_lock);
192 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
194 struct sock * __inet_lookup_established(struct inet_hashinfo *hashinfo,
195 const __be32 saddr, const __be16 sport,
196 const __be32 daddr, const u16 hnum,
199 INET_ADDR_COOKIE(acookie, saddr, daddr)
200 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
202 const struct hlist_node *node;
203 /* Optimize here for direct hit, only listening connections can
204 * have wildcards anyways.
206 unsigned int hash = inet_ehashfn(daddr, hnum, saddr, sport);
207 struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash);
208 rwlock_t *lock = inet_ehash_lockp(hashinfo, hash);
210 prefetch(head->chain.first);
212 sk_for_each(sk, node, &head->chain) {
213 if (INET_MATCH(sk, hash, acookie, saddr, daddr, ports, dif))
214 goto hit; /* You sunk my battleship! */
217 /* Must check for a TIME_WAIT'er before going to listener hash. */
218 sk_for_each(sk, node, &head->twchain) {
219 if (INET_TW_MATCH(sk, hash, acookie, saddr, daddr, ports, dif))
230 EXPORT_SYMBOL_GPL(__inet_lookup_established);
232 /* called with local bh disabled */
233 static int __inet_check_established(struct inet_timewait_death_row *death_row,
234 struct sock *sk, __u16 lport,
235 struct inet_timewait_sock **twp)
237 struct inet_hashinfo *hinfo = death_row->hashinfo;
238 struct inet_sock *inet = inet_sk(sk);
239 __be32 daddr = inet->rcv_saddr;
240 __be32 saddr = inet->daddr;
241 int dif = sk->sk_bound_dev_if;
242 INET_ADDR_COOKIE(acookie, saddr, daddr)
243 const __portpair ports = INET_COMBINED_PORTS(inet->dport, lport);
244 unsigned int hash = inet_ehashfn(daddr, lport, saddr, inet->dport);
245 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
246 rwlock_t *lock = inet_ehash_lockp(hinfo, hash);
248 const struct hlist_node *node;
249 struct inet_timewait_sock *tw;
251 prefetch(head->chain.first);
254 /* Check TIME-WAIT sockets first. */
255 sk_for_each(sk2, node, &head->twchain) {
258 if (INET_TW_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif)) {
259 if (twsk_unique(sk, sk2, twp))
267 /* And established part... */
268 sk_for_each(sk2, node, &head->chain) {
269 if (INET_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif))
274 /* Must record num and sport now. Otherwise we will see
275 * in hash table socket with a funny identity. */
277 inet->sport = htons(lport);
279 BUG_TRAP(sk_unhashed(sk));
280 __sk_add_node(sk, &head->chain);
281 sock_prot_inuse_add(sk->sk_prot, 1);
286 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
288 /* Silly. Should hash-dance instead... */
289 inet_twsk_deschedule(tw, death_row);
290 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
299 return -EADDRNOTAVAIL;
302 static inline u32 inet_sk_port_offset(const struct sock *sk)
304 const struct inet_sock *inet = inet_sk(sk);
305 return secure_ipv4_port_ephemeral(inet->rcv_saddr, inet->daddr,
309 void __inet_hash_nolisten(struct inet_hashinfo *hashinfo, struct sock *sk)
311 struct hlist_head *list;
313 struct inet_ehash_bucket *head;
315 BUG_TRAP(sk_unhashed(sk));
317 sk->sk_hash = inet_sk_ehashfn(sk);
318 head = inet_ehash_bucket(hashinfo, sk->sk_hash);
320 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
323 __sk_add_node(sk, list);
324 sock_prot_inuse_add(sk->sk_prot, 1);
327 EXPORT_SYMBOL_GPL(__inet_hash_nolisten);
329 void __inet_hash(struct inet_hashinfo *hashinfo, struct sock *sk)
331 struct hlist_head *list;
334 if (sk->sk_state != TCP_LISTEN) {
335 __inet_hash_nolisten(hashinfo, sk);
339 BUG_TRAP(sk_unhashed(sk));
340 list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
341 lock = &hashinfo->lhash_lock;
343 inet_listen_wlock(hashinfo);
344 __sk_add_node(sk, list);
345 sock_prot_inuse_add(sk->sk_prot, 1);
347 wake_up(&hashinfo->lhash_wait);
349 EXPORT_SYMBOL_GPL(__inet_hash);
352 * Bind a port for a connect operation and hash it.
354 int inet_hash_connect(struct inet_timewait_death_row *death_row,
357 struct inet_hashinfo *hinfo = death_row->hashinfo;
358 const unsigned short snum = inet_sk(sk)->num;
359 struct inet_bind_hashbucket *head;
360 struct inet_bind_bucket *tb;
364 int i, remaining, low, high, port;
366 u32 offset = hint + inet_sk_port_offset(sk);
367 struct hlist_node *node;
368 struct inet_timewait_sock *tw = NULL;
370 inet_get_local_port_range(&low, &high);
371 remaining = (high - low) + 1;
374 for (i = 1; i <= remaining; i++) {
375 port = low + (i + offset) % remaining;
376 head = &hinfo->bhash[inet_bhashfn(port, hinfo->bhash_size)];
377 spin_lock(&head->lock);
379 /* Does not bother with rcv_saddr checks,
380 * because the established check is already
383 inet_bind_bucket_for_each(tb, node, &head->chain) {
384 if (tb->port == port) {
385 BUG_TRAP(!hlist_empty(&tb->owners));
386 if (tb->fastreuse >= 0)
388 if (!__inet_check_established(death_row,
396 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, head, port);
398 spin_unlock(&head->lock);
405 spin_unlock(&head->lock);
409 return -EADDRNOTAVAIL;
414 /* Head lock still held and bh's disabled */
415 inet_bind_hash(sk, tb, port);
416 if (sk_unhashed(sk)) {
417 inet_sk(sk)->sport = htons(port);
418 __inet_hash_nolisten(hinfo, sk);
420 spin_unlock(&head->lock);
423 inet_twsk_deschedule(tw, death_row);
431 head = &hinfo->bhash[inet_bhashfn(snum, hinfo->bhash_size)];
432 tb = inet_csk(sk)->icsk_bind_hash;
433 spin_lock_bh(&head->lock);
434 if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
435 __inet_hash_nolisten(hinfo, sk);
436 spin_unlock_bh(&head->lock);
439 spin_unlock(&head->lock);
440 /* No definite answer... Walk to established hash table */
441 ret = __inet_check_established(death_row, sk, snum, NULL);
448 EXPORT_SYMBOL_GPL(inet_hash_connect);