blob: 44981906fb913e7afbb4f7e03cd0192b9fdcda12 [file] [log] [blame]
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -07001/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Generic INET transport hashtables
7 *
8 * Authors: Lotsa people, from code originally in tcp
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 */
15
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070016#include <linux/module.h>
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -080017#include <linux/random.h>
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -070018#include <linux/sched.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070019#include <linux/slab.h>
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -070020#include <linux/wait.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070021
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -070022#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070023#include <net/inet_hashtables.h>
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -080024#include <net/ip.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070025
26/*
27 * Allocate and initialize a new local port bind bucket.
28 * The bindhash mutex for snum's hash chain must be held here.
29 */
Christoph Lametere18b8902006-12-06 20:33:20 -080030struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
Pavel Emelyanov941b1d22008-01-31 05:05:50 -080031 struct net *net,
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070032 struct inet_bind_hashbucket *head,
33 const unsigned short snum)
34{
Christoph Lameter54e6ecb2006-12-06 20:33:16 -080035 struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070036
37 if (tb != NULL) {
Denis V. Lunev8c5da492008-04-16 02:01:11 -070038 tb->ib_net = hold_net(net);
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070039 tb->port = snum;
40 tb->fastreuse = 0;
41 INIT_HLIST_HEAD(&tb->owners);
42 hlist_add_head(&tb->node, &head->chain);
43 }
44 return tb;
45}
46
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070047/*
48 * Caller must hold hashbucket lock for this tb with local BH disabled
49 */
Christoph Lametere18b8902006-12-06 20:33:20 -080050void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070051{
52 if (hlist_empty(&tb->owners)) {
53 __hlist_del(&tb->node);
Denis V. Lunev8c5da492008-04-16 02:01:11 -070054 release_net(tb->ib_net);
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070055 kmem_cache_free(cachep, tb);
56 }
57}
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070058
59void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
60 const unsigned short snum)
61{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -070062 inet_sk(sk)->num = snum;
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070063 sk_add_bind_node(sk, &tb->owners);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -070064 inet_csk(sk)->icsk_bind_hash = tb;
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070065}
66
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070067/*
68 * Get rid of any references to a local port held by the given sock.
69 */
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -080070static void __inet_put_port(struct sock *sk)
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070071{
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -070072 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -070073 const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->num,
74 hashinfo->bhash_size);
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070075 struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
76 struct inet_bind_bucket *tb;
77
78 spin_lock(&head->lock);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -070079 tb = inet_csk(sk)->icsk_bind_hash;
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070080 __sk_del_bind_node(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -070081 inet_csk(sk)->icsk_bind_hash = NULL;
82 inet_sk(sk)->num = 0;
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070083 inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
84 spin_unlock(&head->lock);
85}
86
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -080087void inet_put_port(struct sock *sk)
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070088{
89 local_bh_disable();
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -080090 __inet_put_port(sk);
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070091 local_bh_enable();
92}
93
94EXPORT_SYMBOL(inet_put_port);
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -070095
Pavel Emelyanov53083772008-04-17 23:18:15 -070096void __inet_inherit_port(struct sock *sk, struct sock *child)
97{
98 struct inet_hashinfo *table = sk->sk_prot->h.hashinfo;
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -070099 const int bhash = inet_bhashfn(sock_net(sk), inet_sk(child)->num,
100 table->bhash_size);
Pavel Emelyanov53083772008-04-17 23:18:15 -0700101 struct inet_bind_hashbucket *head = &table->bhash[bhash];
102 struct inet_bind_bucket *tb;
103
104 spin_lock(&head->lock);
105 tb = inet_csk(sk)->icsk_bind_hash;
106 sk_add_bind_node(child, &tb->owners);
107 inet_csk(child)->icsk_bind_hash = tb;
108 spin_unlock(&head->lock);
109}
110
111EXPORT_SYMBOL_GPL(__inet_inherit_port);
112
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -0700113/*
114 * This lock without WQ_FLAG_EXCLUSIVE is good on UP and it can be very bad on SMP.
115 * Look, when several writers sleep and reader wakes them up, all but one
116 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
117 * this, _but_ remember, it adds useless work on UP machines (wake up each
118 * exclusive lock release). It should be ifdefed really.
119 */
120void inet_listen_wlock(struct inet_hashinfo *hashinfo)
Eric Dumazet9a429c42008-01-01 21:58:02 -0800121 __acquires(hashinfo->lhash_lock)
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -0700122{
123 write_lock(&hashinfo->lhash_lock);
124
125 if (atomic_read(&hashinfo->lhash_users)) {
126 DEFINE_WAIT(wait);
127
128 for (;;) {
129 prepare_to_wait_exclusive(&hashinfo->lhash_wait,
130 &wait, TASK_UNINTERRUPTIBLE);
131 if (!atomic_read(&hashinfo->lhash_users))
132 break;
133 write_unlock_bh(&hashinfo->lhash_lock);
134 schedule();
135 write_lock_bh(&hashinfo->lhash_lock);
136 }
137
138 finish_wait(&hashinfo->lhash_wait, &wait);
139 }
140}
141
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -0700142/*
143 * Don't inline this cruft. Here are some nice properties to exploit here. The
144 * BSD API does not allow a listening sock to specify the remote port nor the
145 * remote address for the connection. So always assume those are both
146 * wildcarded during the search since they can never be otherwise.
147 */
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800148static struct sock *inet_lookup_listener_slow(struct net *net,
149 const struct hlist_head *head,
Al Virofb99c842006-09-27 18:43:33 -0700150 const __be32 daddr,
Herbert Xu8f4910692006-08-09 15:47:12 -0700151 const unsigned short hnum,
152 const int dif)
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -0700153{
154 struct sock *result = NULL, *sk;
155 const struct hlist_node *node;
156 int hiscore = -1;
157
158 sk_for_each(sk, node, head) {
159 const struct inet_sock *inet = inet_sk(sk);
160
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +0900161 if (net_eq(sock_net(sk), net) && inet->num == hnum &&
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800162 !ipv6_only_sock(sk)) {
Al Virofb99c842006-09-27 18:43:33 -0700163 const __be32 rcv_saddr = inet->rcv_saddr;
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -0700164 int score = sk->sk_family == PF_INET ? 1 : 0;
165
166 if (rcv_saddr) {
167 if (rcv_saddr != daddr)
168 continue;
169 score += 2;
170 }
171 if (sk->sk_bound_dev_if) {
172 if (sk->sk_bound_dev_if != dif)
173 continue;
174 score += 2;
175 }
176 if (score == 5)
177 return sk;
178 if (score > hiscore) {
179 hiscore = score;
180 result = sk;
181 }
182 }
183 }
184 return result;
185}
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700186
Herbert Xu99a92ff2006-08-08 02:18:10 -0700187/* Optimize the common listener case. */
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800188struct sock *__inet_lookup_listener(struct net *net,
189 struct inet_hashinfo *hashinfo,
Al Virofb99c842006-09-27 18:43:33 -0700190 const __be32 daddr, const unsigned short hnum,
Herbert Xu8f4910692006-08-09 15:47:12 -0700191 const int dif)
Herbert Xu99a92ff2006-08-08 02:18:10 -0700192{
193 struct sock *sk = NULL;
194 const struct hlist_head *head;
195
196 read_lock(&hashinfo->lhash_lock);
Pavel Emelyanov2086a652008-06-16 17:13:08 -0700197 head = &hashinfo->listening_hash[inet_lhashfn(net, hnum)];
Herbert Xu99a92ff2006-08-08 02:18:10 -0700198 if (!hlist_empty(head)) {
199 const struct inet_sock *inet = inet_sk((sk = __sk_head(head)));
200
201 if (inet->num == hnum && !sk->sk_node.next &&
202 (!inet->rcv_saddr || inet->rcv_saddr == daddr) &&
203 (sk->sk_family == PF_INET || !ipv6_only_sock(sk)) &&
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +0900204 !sk->sk_bound_dev_if && net_eq(sock_net(sk), net))
Herbert Xu99a92ff2006-08-08 02:18:10 -0700205 goto sherry_cache;
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800206 sk = inet_lookup_listener_slow(net, head, daddr, hnum, dif);
Herbert Xu99a92ff2006-08-08 02:18:10 -0700207 }
208 if (sk) {
209sherry_cache:
210 sock_hold(sk);
211 }
212 read_unlock(&hashinfo->lhash_lock);
213 return sk;
214}
Herbert Xu8f4910692006-08-09 15:47:12 -0700215EXPORT_SYMBOL_GPL(__inet_lookup_listener);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800216
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800217struct sock * __inet_lookup_established(struct net *net,
218 struct inet_hashinfo *hashinfo,
Pavel Emelyanov77a5ba52007-12-20 15:32:17 -0800219 const __be32 saddr, const __be16 sport,
220 const __be32 daddr, const u16 hnum,
221 const int dif)
222{
223 INET_ADDR_COOKIE(acookie, saddr, daddr)
224 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
225 struct sock *sk;
226 const struct hlist_node *node;
227 /* Optimize here for direct hit, only listening connections can
228 * have wildcards anyways.
229 */
Pavel Emelyanov9f26b3a2008-06-16 17:13:27 -0700230 unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
Pavel Emelyanov77a5ba52007-12-20 15:32:17 -0800231 struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash);
232 rwlock_t *lock = inet_ehash_lockp(hashinfo, hash);
233
234 prefetch(head->chain.first);
235 read_lock(lock);
236 sk_for_each(sk, node, &head->chain) {
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800237 if (INET_MATCH(sk, net, hash, acookie,
238 saddr, daddr, ports, dif))
Pavel Emelyanov77a5ba52007-12-20 15:32:17 -0800239 goto hit; /* You sunk my battleship! */
240 }
241
242 /* Must check for a TIME_WAIT'er before going to listener hash. */
243 sk_for_each(sk, node, &head->twchain) {
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800244 if (INET_TW_MATCH(sk, net, hash, acookie,
245 saddr, daddr, ports, dif))
Pavel Emelyanov77a5ba52007-12-20 15:32:17 -0800246 goto hit;
247 }
248 sk = NULL;
249out:
250 read_unlock(lock);
251 return sk;
252hit:
253 sock_hold(sk);
254 goto out;
255}
256EXPORT_SYMBOL_GPL(__inet_lookup_established);
257
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800258/* called with local bh disabled */
259static int __inet_check_established(struct inet_timewait_death_row *death_row,
260 struct sock *sk, __u16 lport,
261 struct inet_timewait_sock **twp)
262{
263 struct inet_hashinfo *hinfo = death_row->hashinfo;
264 struct inet_sock *inet = inet_sk(sk);
Al Virofb99c842006-09-27 18:43:33 -0700265 __be32 daddr = inet->rcv_saddr;
266 __be32 saddr = inet->daddr;
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800267 int dif = sk->sk_bound_dev_if;
268 INET_ADDR_COOKIE(acookie, saddr, daddr)
Al Viro4f765d82006-09-27 18:43:07 -0700269 const __portpair ports = INET_COMBINED_PORTS(inet->dport, lport);
Pavel Emelyanov9f26b3a2008-06-16 17:13:27 -0700270 struct net *net = sock_net(sk);
271 unsigned int hash = inet_ehashfn(net, daddr, lport, saddr, inet->dport);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800272 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
Eric Dumazet230140c2007-11-07 02:40:20 -0800273 rwlock_t *lock = inet_ehash_lockp(hinfo, hash);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800274 struct sock *sk2;
275 const struct hlist_node *node;
276 struct inet_timewait_sock *tw;
277
278 prefetch(head->chain.first);
Eric Dumazet230140c2007-11-07 02:40:20 -0800279 write_lock(lock);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800280
281 /* Check TIME-WAIT sockets first. */
Eric Dumazetdbca9b2752007-02-08 14:16:46 -0800282 sk_for_each(sk2, node, &head->twchain) {
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800283 tw = inet_twsk(sk2);
284
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800285 if (INET_TW_MATCH(sk2, net, hash, acookie,
286 saddr, daddr, ports, dif)) {
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800287 if (twsk_unique(sk, sk2, twp))
288 goto unique;
289 else
290 goto not_unique;
291 }
292 }
293 tw = NULL;
294
295 /* And established part... */
296 sk_for_each(sk2, node, &head->chain) {
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800297 if (INET_MATCH(sk2, net, hash, acookie,
298 saddr, daddr, ports, dif))
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800299 goto not_unique;
300 }
301
302unique:
303 /* Must record num and sport now. Otherwise we will see
304 * in hash table socket with a funny identity. */
305 inet->num = lport;
306 inet->sport = htons(lport);
307 sk->sk_hash = hash;
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700308 WARN_ON(!sk_unhashed(sk));
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800309 __sk_add_node(sk, &head->chain);
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700310 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Eric Dumazet230140c2007-11-07 02:40:20 -0800311 write_unlock(lock);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800312
313 if (twp) {
314 *twp = tw;
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700315 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800316 } else if (tw) {
317 /* Silly. Should hash-dance instead... */
318 inet_twsk_deschedule(tw, death_row);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700319 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800320
321 inet_twsk_put(tw);
322 }
323
324 return 0;
325
326not_unique:
Eric Dumazet230140c2007-11-07 02:40:20 -0800327 write_unlock(lock);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800328 return -EADDRNOTAVAIL;
329}
330
331static inline u32 inet_sk_port_offset(const struct sock *sk)
332{
333 const struct inet_sock *inet = inet_sk(sk);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900334 return secure_ipv4_port_ephemeral(inet->rcv_saddr, inet->daddr,
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800335 inet->dport);
336}
337
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800338void __inet_hash_nolisten(struct sock *sk)
Pavel Emelyanov152da812007-12-20 15:31:33 -0800339{
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -0700340 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
Pavel Emelyanov152da812007-12-20 15:31:33 -0800341 struct hlist_head *list;
342 rwlock_t *lock;
343 struct inet_ehash_bucket *head;
344
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700345 WARN_ON(!sk_unhashed(sk));
Pavel Emelyanov152da812007-12-20 15:31:33 -0800346
347 sk->sk_hash = inet_sk_ehashfn(sk);
348 head = inet_ehash_bucket(hashinfo, sk->sk_hash);
349 list = &head->chain;
350 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
351
352 write_lock(lock);
353 __sk_add_node(sk, list);
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700354 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Pavel Emelyanov152da812007-12-20 15:31:33 -0800355 write_unlock(lock);
356}
357EXPORT_SYMBOL_GPL(__inet_hash_nolisten);
358
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800359static void __inet_hash(struct sock *sk)
Pavel Emelyanov152da812007-12-20 15:31:33 -0800360{
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -0700361 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
Pavel Emelyanov152da812007-12-20 15:31:33 -0800362 struct hlist_head *list;
363 rwlock_t *lock;
364
365 if (sk->sk_state != TCP_LISTEN) {
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800366 __inet_hash_nolisten(sk);
Pavel Emelyanov152da812007-12-20 15:31:33 -0800367 return;
368 }
369
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700370 WARN_ON(!sk_unhashed(sk));
Pavel Emelyanov152da812007-12-20 15:31:33 -0800371 list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
372 lock = &hashinfo->lhash_lock;
373
374 inet_listen_wlock(hashinfo);
375 __sk_add_node(sk, list);
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700376 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Pavel Emelyanov152da812007-12-20 15:31:33 -0800377 write_unlock(lock);
378 wake_up(&hashinfo->lhash_wait);
379}
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800380
381void inet_hash(struct sock *sk)
382{
383 if (sk->sk_state != TCP_CLOSE) {
384 local_bh_disable();
385 __inet_hash(sk);
386 local_bh_enable();
387 }
388}
389EXPORT_SYMBOL_GPL(inet_hash);
390
391void inet_unhash(struct sock *sk)
392{
393 rwlock_t *lock;
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -0700394 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800395
396 if (sk_unhashed(sk))
397 goto out;
398
399 if (sk->sk_state == TCP_LISTEN) {
400 local_bh_disable();
401 inet_listen_wlock(hashinfo);
402 lock = &hashinfo->lhash_lock;
403 } else {
404 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
405 write_lock_bh(lock);
406 }
407
408 if (__sk_del_node_init(sk))
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700409 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800410 write_unlock_bh(lock);
411out:
412 if (sk->sk_state == TCP_LISTEN)
413 wake_up(&hashinfo->lhash_wait);
414}
415EXPORT_SYMBOL_GPL(inet_unhash);
Pavel Emelyanov152da812007-12-20 15:31:33 -0800416
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800417int __inet_hash_connect(struct inet_timewait_death_row *death_row,
Pavel Emelyanov5d8c0aa2008-02-05 03:14:44 -0800418 struct sock *sk, u32 port_offset,
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800419 int (*check_established)(struct inet_timewait_death_row *,
420 struct sock *, __u16, struct inet_timewait_sock **),
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800421 void (*hash)(struct sock *sk))
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800422{
423 struct inet_hashinfo *hinfo = death_row->hashinfo;
424 const unsigned short snum = inet_sk(sk)->num;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900425 struct inet_bind_hashbucket *head;
426 struct inet_bind_bucket *tb;
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800427 int ret;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900428 struct net *net = sock_net(sk);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800429
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900430 if (!snum) {
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700431 int i, remaining, low, high, port;
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800432 static u32 hint;
Pavel Emelyanov5d8c0aa2008-02-05 03:14:44 -0800433 u32 offset = hint + port_offset;
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800434 struct hlist_node *node;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900435 struct inet_timewait_sock *tw = NULL;
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800436
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700437 inet_get_local_port_range(&low, &high);
Anton Arapova25de532007-10-18 22:00:17 -0700438 remaining = (high - low) + 1;
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700439
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900440 local_bh_disable();
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700441 for (i = 1; i <= remaining; i++) {
442 port = low + (i + offset) % remaining;
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700443 head = &hinfo->bhash[inet_bhashfn(net, port,
444 hinfo->bhash_size)];
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900445 spin_lock(&head->lock);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800446
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900447 /* Does not bother with rcv_saddr checks,
448 * because the established check is already
449 * unique enough.
450 */
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800451 inet_bind_bucket_for_each(tb, node, &head->chain) {
Pavel Emelyanov941b1d22008-01-31 05:05:50 -0800452 if (tb->ib_net == net && tb->port == port) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700453 WARN_ON(hlist_empty(&tb->owners));
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900454 if (tb->fastreuse >= 0)
455 goto next_port;
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800456 if (!check_established(death_row, sk,
457 port, &tw))
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900458 goto ok;
459 goto next_port;
460 }
461 }
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800462
Pavel Emelyanov941b1d22008-01-31 05:05:50 -0800463 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
464 net, head, port);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900465 if (!tb) {
466 spin_unlock(&head->lock);
467 break;
468 }
469 tb->fastreuse = -1;
470 goto ok;
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800471
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900472 next_port:
473 spin_unlock(&head->lock);
474 }
475 local_bh_enable();
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800476
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900477 return -EADDRNOTAVAIL;
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800478
479ok:
480 hint += i;
481
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900482 /* Head lock still held and bh's disabled */
483 inet_bind_hash(sk, tb, port);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800484 if (sk_unhashed(sk)) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900485 inet_sk(sk)->sport = htons(port);
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800486 hash(sk);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900487 }
488 spin_unlock(&head->lock);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800489
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900490 if (tw) {
491 inet_twsk_deschedule(tw, death_row);
492 inet_twsk_put(tw);
493 }
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800494
495 ret = 0;
496 goto out;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900497 }
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800498
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700499 head = &hinfo->bhash[inet_bhashfn(net, snum, hinfo->bhash_size)];
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900500 tb = inet_csk(sk)->icsk_bind_hash;
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800501 spin_lock_bh(&head->lock);
502 if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800503 hash(sk);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800504 spin_unlock_bh(&head->lock);
505 return 0;
506 } else {
507 spin_unlock(&head->lock);
508 /* No definite answer... Walk to established hash table */
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800509 ret = check_established(death_row, sk, snum, NULL);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800510out:
511 local_bh_enable();
512 return ret;
513 }
514}
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800515
516/*
517 * Bind a port for a connect operation and hash it.
518 */
519int inet_hash_connect(struct inet_timewait_death_row *death_row,
520 struct sock *sk)
521{
Pavel Emelyanov5d8c0aa2008-02-05 03:14:44 -0800522 return __inet_hash_connect(death_row, sk, inet_sk_port_offset(sk),
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800523 __inet_check_established, __inet_hash_nolisten);
524}
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800525
526EXPORT_SYMBOL_GPL(inet_hash_connect);