blob: 98e0bb3014feac529f6e2fb4d1e34c41ef8ecb49 [file] [log] [blame]
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -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 * Authors: Lotsa people, from code originally in tcp
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14#ifndef _INET_HASHTABLES_H
15#define _INET_HASHTABLES_H
16
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -070017
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070018#include <linux/interrupt.h>
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -070019#include <linux/ipv6.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070020#include <linux/list.h>
21#include <linux/slab.h>
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -070022#include <linux/socket.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070023#include <linux/spinlock.h>
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -070024#include <linux/types.h>
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -070025#include <linux/wait.h>
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -070026
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -070027#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020028#include <net/inet_sock.h>
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -070029#include <net/route.h>
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070030#include <net/sock.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070031#include <net/tcp_states.h>
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070032
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -070033#include <asm/atomic.h>
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070034#include <asm/byteorder.h>
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -070035
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070036/* This is for all connections with a full identity, no wildcards.
37 * New scheme, half the table is for TIME_WAIT, the other half is
38 * for the rest. I'll experiment with dynamic table growth later.
39 */
40struct inet_ehash_bucket {
41 rwlock_t lock;
42 struct hlist_head chain;
Eric Dumazet6d255362005-10-04 15:55:51 -070043};
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070044
45/* There are a few simple rules, which allow for local port reuse by
46 * an application. In essence:
47 *
48 * 1) Sockets bound to different interfaces may share a local port.
49 * Failing that, goto test 2.
50 * 2) If all sockets have sk->sk_reuse set, and none of them are in
51 * TCP_LISTEN state, the port may be shared.
52 * Failing that, goto test 3.
53 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
54 * address, and none of them are the same, the port may be
55 * shared.
56 * Failing this, the port cannot be shared.
57 *
58 * The interesting point, is test #2. This is what an FTP server does
59 * all day. To optimize this case we use a specific flag bit defined
60 * below. As we add sockets to a bind bucket list, we perform a
61 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
62 * As long as all sockets added to a bind bucket pass this test,
63 * the flag bit will be set.
64 * The resulting situation is that tcp_v[46]_verify_bind() can just check
65 * for this flag bit, if it is set and the socket trying to bind has
66 * sk->sk_reuse set, we don't even have to walk the owners list at all,
67 * we return that it is ok to bind this socket to the requested local port.
68 *
69 * Sounds like a lot of work, but it is worth it. In a more naive
70 * implementation (ie. current FreeBSD etc.) the entire list of ports
71 * must be walked for each data port opened by an ftp server. Needless
72 * to say, this does not scale at all. With a couple thousand FTP
73 * users logged onto your box, isn't it nice to know that new data
74 * ports are created in O(1) time? I thought so. ;-) -DaveM
75 */
76struct inet_bind_bucket {
77 unsigned short port;
78 signed short fastreuse;
79 struct hlist_node node;
80 struct hlist_head owners;
81};
82
83#define inet_bind_bucket_for_each(tb, node, head) \
84 hlist_for_each_entry(tb, node, head, node)
85
86struct inet_bind_hashbucket {
87 spinlock_t lock;
88 struct hlist_head chain;
89};
90
91/* This is for listening sockets, thus all sockets which possess wildcards. */
92#define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
93
94struct inet_hashinfo {
95 /* This is for sockets with full identity only. Sockets here will
96 * always be without wildcards and will have the following invariant:
97 *
98 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
99 *
100 * First half of the table is for sockets not in TIME_WAIT, second half
101 * is for TIME_WAIT sockets only.
102 */
103 struct inet_ehash_bucket *ehash;
104
105 /* Ok, let's try this, I give up, we do need a local binding
106 * TCP hash as well as the others for fast bind/connect.
107 */
108 struct inet_bind_hashbucket *bhash;
109
110 int bhash_size;
Eric Dumazet81c3d542005-10-03 14:13:38 -0700111 unsigned int ehash_size;
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700112
113 /* All sockets in TCP_LISTEN state will be in here. This is the only
114 * table where wildcard'd TCP sockets can exist. Hash function here
115 * is just local port number.
116 */
117 struct hlist_head listening_hash[INET_LHTABLE_SIZE];
118
119 /* All the above members are written once at bootup and
120 * never written again _or_ are predominantly read-access.
121 *
122 * Now align to a new cache line as all the following members
123 * are often dirty.
124 */
125 rwlock_t lhash_lock ____cacheline_aligned;
126 atomic_t lhash_users;
127 wait_queue_head_t lhash_wait;
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700128 kmem_cache_t *bind_bucket_cachep;
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700129};
130
Eric Dumazet81c3d542005-10-03 14:13:38 -0700131static inline struct inet_ehash_bucket *inet_ehash_bucket(
132 struct inet_hashinfo *hashinfo,
133 unsigned int hash)
134{
135 return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)];
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -0700136}
137
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700138extern struct inet_bind_bucket *
139 inet_bind_bucket_create(kmem_cache_t *cachep,
140 struct inet_bind_hashbucket *head,
141 const unsigned short snum);
142extern void inet_bind_bucket_destroy(kmem_cache_t *cachep,
143 struct inet_bind_bucket *tb);
144
145static inline int inet_bhashfn(const __u16 lport, const int bhash_size)
146{
147 return lport & (bhash_size - 1);
148}
149
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700150extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
151 const unsigned short snum);
152
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700153/* These can have wildcards, don't try too hard. */
154static inline int inet_lhashfn(const unsigned short num)
155{
156 return num & (INET_LHTABLE_SIZE - 1);
157}
158
159static inline int inet_sk_listen_hashfn(const struct sock *sk)
160{
161 return inet_lhashfn(inet_sk(sk)->num);
162}
163
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700164/* Caller must disable local BH processing. */
165static inline void __inet_inherit_port(struct inet_hashinfo *table,
166 struct sock *sk, struct sock *child)
167{
168 const int bhash = inet_bhashfn(inet_sk(child)->num, table->bhash_size);
169 struct inet_bind_hashbucket *head = &table->bhash[bhash];
170 struct inet_bind_bucket *tb;
171
172 spin_lock(&head->lock);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700173 tb = inet_csk(sk)->icsk_bind_hash;
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700174 sk_add_bind_node(child, &tb->owners);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700175 inet_csk(child)->icsk_bind_hash = tb;
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700176 spin_unlock(&head->lock);
177}
178
179static inline void inet_inherit_port(struct inet_hashinfo *table,
180 struct sock *sk, struct sock *child)
181{
182 local_bh_disable();
183 __inet_inherit_port(table, sk, child);
184 local_bh_enable();
185}
186
187extern void inet_put_port(struct inet_hashinfo *table, struct sock *sk);
188
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -0700189extern void inet_listen_wlock(struct inet_hashinfo *hashinfo);
190
191/*
192 * - We may sleep inside this lock.
193 * - If sleeping is not required (or called from BH),
194 * use plain read_(un)lock(&inet_hashinfo.lhash_lock).
195 */
196static inline void inet_listen_lock(struct inet_hashinfo *hashinfo)
197{
198 /* read_lock synchronizes to candidates to writers */
199 read_lock(&hashinfo->lhash_lock);
200 atomic_inc(&hashinfo->lhash_users);
201 read_unlock(&hashinfo->lhash_lock);
202}
203
204static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo)
205{
206 if (atomic_dec_and_test(&hashinfo->lhash_users))
207 wake_up(&hashinfo->lhash_wait);
208}
209
210static inline void __inet_hash(struct inet_hashinfo *hashinfo,
211 struct sock *sk, const int listen_possible)
212{
213 struct hlist_head *list;
214 rwlock_t *lock;
215
216 BUG_TRAP(sk_unhashed(sk));
217 if (listen_possible && sk->sk_state == TCP_LISTEN) {
218 list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
219 lock = &hashinfo->lhash_lock;
220 inet_listen_wlock(hashinfo);
221 } else {
Eric Dumazet81c3d542005-10-03 14:13:38 -0700222 struct inet_ehash_bucket *head;
223 sk->sk_hash = inet_sk_ehashfn(sk);
224 head = inet_ehash_bucket(hashinfo, sk->sk_hash);
225 list = &head->chain;
226 lock = &head->lock;
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -0700227 write_lock(lock);
228 }
229 __sk_add_node(sk, list);
230 sock_prot_inc_use(sk->sk_prot);
231 write_unlock(lock);
232 if (listen_possible && sk->sk_state == TCP_LISTEN)
233 wake_up(&hashinfo->lhash_wait);
234}
Arnaldo Carvalho de Melo81849d12005-08-09 20:08:50 -0700235
236static inline void inet_hash(struct inet_hashinfo *hashinfo, struct sock *sk)
237{
238 if (sk->sk_state != TCP_CLOSE) {
239 local_bh_disable();
240 __inet_hash(hashinfo, sk, 1);
241 local_bh_enable();
242 }
243}
244
245static inline void inet_unhash(struct inet_hashinfo *hashinfo, struct sock *sk)
246{
247 rwlock_t *lock;
248
249 if (sk_unhashed(sk))
250 goto out;
251
252 if (sk->sk_state == TCP_LISTEN) {
253 local_bh_disable();
254 inet_listen_wlock(hashinfo);
255 lock = &hashinfo->lhash_lock;
256 } else {
Eric Dumazet81c3d542005-10-03 14:13:38 -0700257 lock = &inet_ehash_bucket(hashinfo, sk->sk_hash)->lock;
258 write_lock_bh(lock);
Arnaldo Carvalho de Melo81849d12005-08-09 20:08:50 -0700259 }
260
261 if (__sk_del_node_init(sk))
262 sock_prot_dec_use(sk->sk_prot);
263 write_unlock_bh(lock);
264out:
265 if (sk->sk_state == TCP_LISTEN)
266 wake_up(&hashinfo->lhash_wait);
267}
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -0700268
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700269static inline int inet_iif(const struct sk_buff *skb)
270{
271 return ((struct rtable *)skb->dst)->rt_iif;
272}
273
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -0700274extern struct sock *__inet_lookup_listener(const struct hlist_head *head,
275 const u32 daddr,
276 const unsigned short hnum,
277 const int dif);
278
279/* Optimize the common listener case. */
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700280static inline struct sock *
281 inet_lookup_listener(struct inet_hashinfo *hashinfo,
282 const u32 daddr,
283 const unsigned short hnum, const int dif)
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -0700284{
285 struct sock *sk = NULL;
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700286 const struct hlist_head *head;
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -0700287
288 read_lock(&hashinfo->lhash_lock);
289 head = &hashinfo->listening_hash[inet_lhashfn(hnum)];
290 if (!hlist_empty(head)) {
291 const struct inet_sock *inet = inet_sk((sk = __sk_head(head)));
292
293 if (inet->num == hnum && !sk->sk_node.next &&
294 (!inet->rcv_saddr || inet->rcv_saddr == daddr) &&
295 (sk->sk_family == PF_INET || !ipv6_only_sock(sk)) &&
296 !sk->sk_bound_dev_if)
297 goto sherry_cache;
298 sk = __inet_lookup_listener(head, daddr, hnum, dif);
299 }
300 if (sk) {
301sherry_cache:
302 sock_hold(sk);
303 }
304 read_unlock(&hashinfo->lhash_lock);
305 return sk;
306}
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700307
308/* Socket demux engine toys. */
309#ifdef __BIG_ENDIAN
310#define INET_COMBINED_PORTS(__sport, __dport) \
311 (((__u32)(__sport) << 16) | (__u32)(__dport))
312#else /* __LITTLE_ENDIAN */
313#define INET_COMBINED_PORTS(__sport, __dport) \
314 (((__u32)(__dport) << 16) | (__u32)(__sport))
315#endif
316
317#if (BITS_PER_LONG == 64)
318#ifdef __BIG_ENDIAN
319#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
320 const __u64 __name = (((__u64)(__saddr)) << 32) | ((__u64)(__daddr));
321#else /* __LITTLE_ENDIAN */
322#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
323 const __u64 __name = (((__u64)(__daddr)) << 32) | ((__u64)(__saddr));
324#endif /* __BIG_ENDIAN */
Eric Dumazet81c3d542005-10-03 14:13:38 -0700325#define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
326 (((__sk)->sk_hash == (__hash)) && \
327 ((*((__u64 *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700328 ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \
329 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
Eric Dumazet81c3d542005-10-03 14:13:38 -0700330#define INET_TW_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
331 (((__sk)->sk_hash == (__hash)) && \
332 ((*((__u64 *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700333 ((*((__u32 *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
334 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
335#else /* 32-bit arch */
336#define INET_ADDR_COOKIE(__name, __saddr, __daddr)
Eric Dumazet81c3d542005-10-03 14:13:38 -0700337#define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif) \
338 (((__sk)->sk_hash == (__hash)) && \
339 (inet_sk(__sk)->daddr == (__saddr)) && \
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700340 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
341 ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \
342 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
Eric Dumazet81c3d542005-10-03 14:13:38 -0700343#define INET_TW_MATCH(__sk, __hash,__cookie, __saddr, __daddr, __ports, __dif) \
344 (((__sk)->sk_hash == (__hash)) && \
345 (inet_twsk(__sk)->tw_daddr == (__saddr)) && \
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700346 (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
347 ((*((__u32 *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
348 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
349#endif /* 64-bit arch */
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700350
351/*
352 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
353 * not check it for lookups anymore, thanks Alexey. -DaveM
354 *
355 * Local BH must be disabled here.
356 */
357static inline struct sock *
358 __inet_lookup_established(struct inet_hashinfo *hashinfo,
359 const u32 saddr, const u16 sport,
360 const u32 daddr, const u16 hnum,
361 const int dif)
362{
363 INET_ADDR_COOKIE(acookie, saddr, daddr)
364 const __u32 ports = INET_COMBINED_PORTS(sport, hnum);
365 struct sock *sk;
366 const struct hlist_node *node;
367 /* Optimize here for direct hit, only listening connections can
368 * have wildcards anyways.
369 */
Eric Dumazet81c3d542005-10-03 14:13:38 -0700370 unsigned int hash = inet_ehashfn(daddr, hnum, saddr, sport);
371 struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700372
Eric Dumazet81c3d542005-10-03 14:13:38 -0700373 prefetch(head->chain.first);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700374 read_lock(&head->lock);
375 sk_for_each(sk, node, &head->chain) {
Eric Dumazet81c3d542005-10-03 14:13:38 -0700376 if (INET_MATCH(sk, hash, acookie, saddr, daddr, ports, dif))
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700377 goto hit; /* You sunk my battleship! */
378 }
379
380 /* Must check for a TIME_WAIT'er before going to listener hash. */
381 sk_for_each(sk, node, &(head + hashinfo->ehash_size)->chain) {
Eric Dumazet81c3d542005-10-03 14:13:38 -0700382 if (INET_TW_MATCH(sk, hash, acookie, saddr, daddr, ports, dif))
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700383 goto hit;
384 }
385 sk = NULL;
386out:
387 read_unlock(&head->lock);
388 return sk;
389hit:
390 sock_hold(sk);
391 goto out;
392}
393
394static inline struct sock *__inet_lookup(struct inet_hashinfo *hashinfo,
395 const u32 saddr, const u16 sport,
396 const u32 daddr, const u16 hnum,
397 const int dif)
398{
399 struct sock *sk = __inet_lookup_established(hashinfo, saddr, sport, daddr,
400 hnum, dif);
401 return sk ? : inet_lookup_listener(hashinfo, daddr, hnum, dif);
402}
403
404static inline struct sock *inet_lookup(struct inet_hashinfo *hashinfo,
405 const u32 saddr, const u16 sport,
406 const u32 daddr, const u16 dport,
407 const int dif)
408{
409 struct sock *sk;
410
411 local_bh_disable();
412 sk = __inet_lookup(hashinfo, saddr, sport, daddr, ntohs(dport), dif);
413 local_bh_enable();
414
415 return sk;
416}
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800417
418extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
419 struct sock *sk);
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -0700420#endif /* _INET_HASHTABLES_H */