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
)
100 write_lock(&hashinfo
->lhash_lock
);
102 if (atomic_read(&hashinfo
->lhash_users
)) {
106 prepare_to_wait_exclusive(&hashinfo
->lhash_wait
,
107 &wait
, TASK_UNINTERRUPTIBLE
);
108 if (!atomic_read(&hashinfo
->lhash_users
))
110 write_unlock_bh(&hashinfo
->lhash_lock
);
112 write_lock_bh(&hashinfo
->lhash_lock
);
115 finish_wait(&hashinfo
->lhash_wait
, &wait
);
119 EXPORT_SYMBOL(inet_listen_wlock
);
122 * Don't inline this cruft. Here are some nice properties to exploit here. The
123 * BSD API does not allow a listening sock to specify the remote port nor the
124 * remote address for the connection. So always assume those are both
125 * wildcarded during the search since they can never be otherwise.
127 static struct sock
*inet_lookup_listener_slow(const struct hlist_head
*head
,
129 const unsigned short hnum
,
132 struct sock
*result
= NULL
, *sk
;
133 const struct hlist_node
*node
;
136 sk_for_each(sk
, node
, head
) {
137 const struct inet_sock
*inet
= inet_sk(sk
);
139 if (inet
->num
== hnum
&& !ipv6_only_sock(sk
)) {
140 const __be32 rcv_saddr
= inet
->rcv_saddr
;
141 int score
= sk
->sk_family
== PF_INET
? 1 : 0;
144 if (rcv_saddr
!= daddr
)
148 if (sk
->sk_bound_dev_if
) {
149 if (sk
->sk_bound_dev_if
!= dif
)
155 if (score
> hiscore
) {
164 /* Optimize the common listener case. */
165 struct sock
*__inet_lookup_listener(struct inet_hashinfo
*hashinfo
,
166 const __be32 daddr
, const unsigned short hnum
,
169 struct sock
*sk
= NULL
;
170 const struct hlist_head
*head
;
172 read_lock(&hashinfo
->lhash_lock
);
173 head
= &hashinfo
->listening_hash
[inet_lhashfn(hnum
)];
174 if (!hlist_empty(head
)) {
175 const struct inet_sock
*inet
= inet_sk((sk
= __sk_head(head
)));
177 if (inet
->num
== hnum
&& !sk
->sk_node
.next
&&
178 (!inet
->rcv_saddr
|| inet
->rcv_saddr
== daddr
) &&
179 (sk
->sk_family
== PF_INET
|| !ipv6_only_sock(sk
)) &&
180 !sk
->sk_bound_dev_if
)
182 sk
= inet_lookup_listener_slow(head
, daddr
, hnum
, dif
);
188 read_unlock(&hashinfo
->lhash_lock
);
191 EXPORT_SYMBOL_GPL(__inet_lookup_listener
);
193 struct sock
* __inet_lookup_established(struct inet_hashinfo
*hashinfo
,
194 const __be32 saddr
, const __be16 sport
,
195 const __be32 daddr
, const u16 hnum
,
198 INET_ADDR_COOKIE(acookie
, saddr
, daddr
)
199 const __portpair ports
= INET_COMBINED_PORTS(sport
, hnum
);
201 const struct hlist_node
*node
;
202 /* Optimize here for direct hit, only listening connections can
203 * have wildcards anyways.
205 unsigned int hash
= inet_ehashfn(daddr
, hnum
, saddr
, sport
);
206 struct inet_ehash_bucket
*head
= inet_ehash_bucket(hashinfo
, hash
);
208 prefetch(head
->chain
.first
);
209 read_lock(&head
->lock
);
210 sk_for_each(sk
, node
, &head
->chain
) {
211 if (INET_MATCH(sk
, hash
, acookie
, saddr
, daddr
, ports
, dif
))
212 goto hit
; /* You sunk my battleship! */
215 /* Must check for a TIME_WAIT'er before going to listener hash. */
216 sk_for_each(sk
, node
, &head
->twchain
) {
217 if (INET_TW_MATCH(sk
, hash
, acookie
, saddr
, daddr
, ports
, dif
))
222 read_unlock(&head
->lock
);
228 EXPORT_SYMBOL_GPL(__inet_lookup_established
);
230 /* called with local bh disabled */
231 static int __inet_check_established(struct inet_timewait_death_row
*death_row
,
232 struct sock
*sk
, __u16 lport
,
233 struct inet_timewait_sock
**twp
)
235 struct inet_hashinfo
*hinfo
= death_row
->hashinfo
;
236 struct inet_sock
*inet
= inet_sk(sk
);
237 __be32 daddr
= inet
->rcv_saddr
;
238 __be32 saddr
= inet
->daddr
;
239 int dif
= sk
->sk_bound_dev_if
;
240 INET_ADDR_COOKIE(acookie
, saddr
, daddr
)
241 const __portpair ports
= INET_COMBINED_PORTS(inet
->dport
, lport
);
242 unsigned int hash
= inet_ehashfn(daddr
, lport
, saddr
, inet
->dport
);
243 struct inet_ehash_bucket
*head
= inet_ehash_bucket(hinfo
, hash
);
245 const struct hlist_node
*node
;
246 struct inet_timewait_sock
*tw
;
248 prefetch(head
->chain
.first
);
249 write_lock(&head
->lock
);
251 /* Check TIME-WAIT sockets first. */
252 sk_for_each(sk2
, node
, &head
->twchain
) {
255 if (INET_TW_MATCH(sk2
, hash
, acookie
, saddr
, daddr
, ports
, dif
)) {
256 if (twsk_unique(sk
, sk2
, twp
))
264 /* And established part... */
265 sk_for_each(sk2
, node
, &head
->chain
) {
266 if (INET_MATCH(sk2
, hash
, acookie
, saddr
, daddr
, ports
, dif
))
271 /* Must record num and sport now. Otherwise we will see
272 * in hash table socket with a funny identity. */
274 inet
->sport
= htons(lport
);
276 BUG_TRAP(sk_unhashed(sk
));
277 __sk_add_node(sk
, &head
->chain
);
278 sock_prot_inc_use(sk
->sk_prot
);
279 write_unlock(&head
->lock
);
283 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED
);
285 /* Silly. Should hash-dance instead... */
286 inet_twsk_deschedule(tw
, death_row
);
287 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED
);
295 write_unlock(&head
->lock
);
296 return -EADDRNOTAVAIL
;
299 static inline u32
inet_sk_port_offset(const struct sock
*sk
)
301 const struct inet_sock
*inet
= inet_sk(sk
);
302 return secure_ipv4_port_ephemeral(inet
->rcv_saddr
, inet
->daddr
,
306 void __inet_hash_nolisten(struct inet_hashinfo
*hashinfo
, struct sock
*sk
)
308 struct hlist_head
*list
;
310 struct inet_ehash_bucket
*head
;
312 BUG_TRAP(sk_unhashed(sk
));
314 sk
->sk_hash
= inet_sk_ehashfn(sk
);
315 head
= inet_ehash_bucket(hashinfo
, sk
->sk_hash
);
320 __sk_add_node(sk
, list
);
321 sock_prot_inc_use(sk
->sk_prot
);
324 EXPORT_SYMBOL_GPL(__inet_hash_nolisten
);
326 void __inet_hash(struct inet_hashinfo
*hashinfo
, struct sock
*sk
)
328 struct hlist_head
*list
;
331 if (sk
->sk_state
!= TCP_LISTEN
) {
332 __inet_hash_nolisten(hashinfo
, sk
);
336 BUG_TRAP(sk_unhashed(sk
));
337 list
= &hashinfo
->listening_hash
[inet_sk_listen_hashfn(sk
)];
338 lock
= &hashinfo
->lhash_lock
;
340 inet_listen_wlock(hashinfo
);
341 __sk_add_node(sk
, list
);
342 sock_prot_inc_use(sk
->sk_prot
);
344 wake_up(&hashinfo
->lhash_wait
);
346 EXPORT_SYMBOL_GPL(__inet_hash
);
348 int __inet_hash_connect(struct inet_timewait_death_row
*death_row
,
349 struct sock
*sk
, u32 port_offset
,
350 int (*check_established
)(struct inet_timewait_death_row
*,
351 struct sock
*, __u16
, struct inet_timewait_sock
**),
352 void (*hash
)(struct inet_hashinfo
*, struct sock
*))
354 struct inet_hashinfo
*hinfo
= death_row
->hashinfo
;
355 const unsigned short snum
= inet_sk(sk
)->num
;
356 struct inet_bind_hashbucket
*head
;
357 struct inet_bind_bucket
*tb
;
361 int low
= sysctl_local_port_range
[0];
362 int high
= sysctl_local_port_range
[1];
363 int range
= high
- low
;
367 u32 offset
= hint
+ port_offset
;
368 struct hlist_node
*node
;
369 struct inet_timewait_sock
*tw
= NULL
;
372 for (i
= 1; i
<= range
; i
++) {
373 port
= low
+ (i
+ offset
) % range
;
374 head
= &hinfo
->bhash
[inet_bhashfn(port
, hinfo
->bhash_size
)];
375 spin_lock(&head
->lock
);
377 /* Does not bother with rcv_saddr checks,
378 * because the established check is already
381 inet_bind_bucket_for_each(tb
, node
, &head
->chain
) {
382 if (tb
->port
== port
) {
383 BUG_TRAP(!hlist_empty(&tb
->owners
));
384 if (tb
->fastreuse
>= 0)
386 if (!check_established(death_row
, sk
,
393 tb
= inet_bind_bucket_create(hinfo
->bind_bucket_cachep
, head
, port
);
395 spin_unlock(&head
->lock
);
402 spin_unlock(&head
->lock
);
406 return -EADDRNOTAVAIL
;
411 /* Head lock still held and bh's disabled */
412 inet_bind_hash(sk
, tb
, port
);
413 if (sk_unhashed(sk
)) {
414 inet_sk(sk
)->sport
= htons(port
);
417 spin_unlock(&head
->lock
);
420 inet_twsk_deschedule(tw
, death_row
);
428 head
= &hinfo
->bhash
[inet_bhashfn(snum
, hinfo
->bhash_size
)];
429 tb
= inet_csk(sk
)->icsk_bind_hash
;
430 spin_lock_bh(&head
->lock
);
431 if (sk_head(&tb
->owners
) == sk
&& !sk
->sk_bind_node
.next
) {
433 spin_unlock_bh(&head
->lock
);
436 spin_unlock(&head
->lock
);
437 /* No definite answer... Walk to established hash table */
438 ret
= check_established(death_row
, sk
, snum
, NULL
);
444 EXPORT_SYMBOL_GPL(__inet_hash_connect
);
447 * Bind a port for a connect operation and hash it.
449 int inet_hash_connect(struct inet_timewait_death_row
*death_row
,
452 return __inet_hash_connect(death_row
, sk
, inet_sk_port_offset(sk
),
453 __inet_check_established
, __inet_hash_nolisten
);
456 EXPORT_SYMBOL_GPL(inet_hash_connect
);