aboutsummaryrefslogtreecommitdiffstats
path: root/include/net/inet_hashtables.h
diff options
context:
space:
mode:
authorJeff Garzik <jgarzik@pobox.com>2005-09-01 18:02:01 -0400
committerJeff Garzik <jgarzik@pobox.com>2005-09-01 18:02:01 -0400
commite3ee3b78f83688a0ae4315e8be71b2eac559904a (patch)
treedeb03bcdd020262af450ed23382d7c921263f5cf /include/net/inet_hashtables.h
parent91cb70c1769d9b72dd1efe40c31f01005820b09e (diff)
parent6b39374a27eb4be7e9d82145ae270ba02ea90dc8 (diff)
/spare/repo/netdev-2.6 branch 'master'
Diffstat (limited to 'include/net/inet_hashtables.h')
-rw-r--r--include/net/inet_hashtables.h427
1 files changed, 427 insertions, 0 deletions
diff --git a/include/net/inet_hashtables.h b/include/net/inet_hashtables.h
new file mode 100644
index 000000000000..646b6ea7fe26
--- /dev/null
+++ b/include/net/inet_hashtables.h
@@ -0,0 +1,427 @@
1/*
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
17#include <linux/config.h>
18
19#include <linux/interrupt.h>
20#include <linux/ipv6.h>
21#include <linux/list.h>
22#include <linux/slab.h>
23#include <linux/socket.h>
24#include <linux/spinlock.h>
25#include <linux/types.h>
26#include <linux/wait.h>
27
28#include <net/inet_connection_sock.h>
29#include <net/route.h>
30#include <net/sock.h>
31#include <net/tcp_states.h>
32
33#include <asm/atomic.h>
34#include <asm/byteorder.h>
35
36/* 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;
43} __attribute__((__aligned__(8)));
44
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;
111 int ehash_size;
112
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;
128 spinlock_t portalloc_lock;
129 kmem_cache_t *bind_bucket_cachep;
130 int port_rover;
131};
132
133static inline int inet_ehashfn(const __u32 laddr, const __u16 lport,
134 const __u32 faddr, const __u16 fport,
135 const int ehash_size)
136{
137 int h = (laddr ^ lport) ^ (faddr ^ fport);
138 h ^= h >> 16;
139 h ^= h >> 8;
140 return h & (ehash_size - 1);
141}
142
143static inline int inet_sk_ehashfn(const struct sock *sk, const int ehash_size)
144{
145 const struct inet_sock *inet = inet_sk(sk);
146 const __u32 laddr = inet->rcv_saddr;
147 const __u16 lport = inet->num;
148 const __u32 faddr = inet->daddr;
149 const __u16 fport = inet->dport;
150
151 return inet_ehashfn(laddr, lport, faddr, fport, ehash_size);
152}
153
154extern struct inet_bind_bucket *
155 inet_bind_bucket_create(kmem_cache_t *cachep,
156 struct inet_bind_hashbucket *head,
157 const unsigned short snum);
158extern void inet_bind_bucket_destroy(kmem_cache_t *cachep,
159 struct inet_bind_bucket *tb);
160
161static inline int inet_bhashfn(const __u16 lport, const int bhash_size)
162{
163 return lport & (bhash_size - 1);
164}
165
166extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
167 const unsigned short snum);
168
169/* These can have wildcards, don't try too hard. */
170static inline int inet_lhashfn(const unsigned short num)
171{
172 return num & (INET_LHTABLE_SIZE - 1);
173}
174
175static inline int inet_sk_listen_hashfn(const struct sock *sk)
176{
177 return inet_lhashfn(inet_sk(sk)->num);
178}
179
180/* Caller must disable local BH processing. */
181static inline void __inet_inherit_port(struct inet_hashinfo *table,
182 struct sock *sk, struct sock *child)
183{
184 const int bhash = inet_bhashfn(inet_sk(child)->num, table->bhash_size);
185 struct inet_bind_hashbucket *head = &table->bhash[bhash];
186 struct inet_bind_bucket *tb;
187
188 spin_lock(&head->lock);
189 tb = inet_csk(sk)->icsk_bind_hash;
190 sk_add_bind_node(child, &tb->owners);
191 inet_csk(child)->icsk_bind_hash = tb;
192 spin_unlock(&head->lock);
193}
194
195static inline void inet_inherit_port(struct inet_hashinfo *table,
196 struct sock *sk, struct sock *child)
197{
198 local_bh_disable();
199 __inet_inherit_port(table, sk, child);
200 local_bh_enable();
201}
202
203extern void inet_put_port(struct inet_hashinfo *table, struct sock *sk);
204
205extern void inet_listen_wlock(struct inet_hashinfo *hashinfo);
206
207/*
208 * - We may sleep inside this lock.
209 * - If sleeping is not required (or called from BH),
210 * use plain read_(un)lock(&inet_hashinfo.lhash_lock).
211 */
212static inline void inet_listen_lock(struct inet_hashinfo *hashinfo)
213{
214 /* read_lock synchronizes to candidates to writers */
215 read_lock(&hashinfo->lhash_lock);
216 atomic_inc(&hashinfo->lhash_users);
217 read_unlock(&hashinfo->lhash_lock);
218}
219
220static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo)
221{
222 if (atomic_dec_and_test(&hashinfo->lhash_users))
223 wake_up(&hashinfo->lhash_wait);
224}
225
226static inline void __inet_hash(struct inet_hashinfo *hashinfo,
227 struct sock *sk, const int listen_possible)
228{
229 struct hlist_head *list;
230 rwlock_t *lock;
231
232 BUG_TRAP(sk_unhashed(sk));
233 if (listen_possible && sk->sk_state == TCP_LISTEN) {
234 list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
235 lock = &hashinfo->lhash_lock;
236 inet_listen_wlock(hashinfo);
237 } else {
238 sk->sk_hashent = inet_sk_ehashfn(sk, hashinfo->ehash_size);
239 list = &hashinfo->ehash[sk->sk_hashent].chain;
240 lock = &hashinfo->ehash[sk->sk_hashent].lock;
241 write_lock(lock);
242 }
243 __sk_add_node(sk, list);
244 sock_prot_inc_use(sk->sk_prot);
245 write_unlock(lock);
246 if (listen_possible && sk->sk_state == TCP_LISTEN)
247 wake_up(&hashinfo->lhash_wait);
248}
249
250static inline void inet_hash(struct inet_hashinfo *hashinfo, struct sock *sk)
251{
252 if (sk->sk_state != TCP_CLOSE) {
253 local_bh_disable();
254 __inet_hash(hashinfo, sk, 1);
255 local_bh_enable();
256 }
257}
258
259static inline void inet_unhash(struct inet_hashinfo *hashinfo, struct sock *sk)
260{
261 rwlock_t *lock;
262
263 if (sk_unhashed(sk))
264 goto out;
265
266 if (sk->sk_state == TCP_LISTEN) {
267 local_bh_disable();
268 inet_listen_wlock(hashinfo);
269 lock = &hashinfo->lhash_lock;
270 } else {
271 struct inet_ehash_bucket *head = &hashinfo->ehash[sk->sk_hashent];
272 lock = &head->lock;
273 write_lock_bh(&head->lock);
274 }
275
276 if (__sk_del_node_init(sk))
277 sock_prot_dec_use(sk->sk_prot);
278 write_unlock_bh(lock);
279out:
280 if (sk->sk_state == TCP_LISTEN)
281 wake_up(&hashinfo->lhash_wait);
282}
283
284static inline int inet_iif(const struct sk_buff *skb)
285{
286 return ((struct rtable *)skb->dst)->rt_iif;
287}
288
289extern struct sock *__inet_lookup_listener(const struct hlist_head *head,
290 const u32 daddr,
291 const unsigned short hnum,
292 const int dif);
293
294/* Optimize the common listener case. */
295static inline struct sock *
296 inet_lookup_listener(struct inet_hashinfo *hashinfo,
297 const u32 daddr,
298 const unsigned short hnum, const int dif)
299{
300 struct sock *sk = NULL;
301 const struct hlist_head *head;
302
303 read_lock(&hashinfo->lhash_lock);
304 head = &hashinfo->listening_hash[inet_lhashfn(hnum)];
305 if (!hlist_empty(head)) {
306 const struct inet_sock *inet = inet_sk((sk = __sk_head(head)));
307
308 if (inet->num == hnum && !sk->sk_node.next &&
309 (!inet->rcv_saddr || inet->rcv_saddr == daddr) &&
310 (sk->sk_family == PF_INET || !ipv6_only_sock(sk)) &&
311 !sk->sk_bound_dev_if)
312 goto sherry_cache;
313 sk = __inet_lookup_listener(head, daddr, hnum, dif);
314 }
315 if (sk) {
316sherry_cache:
317 sock_hold(sk);
318 }
319 read_unlock(&hashinfo->lhash_lock);
320 return sk;
321}
322
323/* Socket demux engine toys. */
324#ifdef __BIG_ENDIAN
325#define INET_COMBINED_PORTS(__sport, __dport) \
326 (((__u32)(__sport) << 16) | (__u32)(__dport))
327#else /* __LITTLE_ENDIAN */
328#define INET_COMBINED_PORTS(__sport, __dport) \
329 (((__u32)(__dport) << 16) | (__u32)(__sport))
330#endif
331
332#if (BITS_PER_LONG == 64)
333#ifdef __BIG_ENDIAN
334#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
335 const __u64 __name = (((__u64)(__saddr)) << 32) | ((__u64)(__daddr));
336#else /* __LITTLE_ENDIAN */
337#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
338 const __u64 __name = (((__u64)(__daddr)) << 32) | ((__u64)(__saddr));
339#endif /* __BIG_ENDIAN */
340#define INET_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
341 (((*((__u64 *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \
342 ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \
343 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
344#define INET_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif)\
345 (((*((__u64 *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \
346 ((*((__u32 *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
347 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
348#else /* 32-bit arch */
349#define INET_ADDR_COOKIE(__name, __saddr, __daddr)
350#define INET_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif) \
351 ((inet_sk(__sk)->daddr == (__saddr)) && \
352 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
353 ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \
354 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
355#define INET_TW_MATCH(__sk, __cookie, __saddr, __daddr, __ports, __dif) \
356 ((inet_twsk(__sk)->tw_daddr == (__saddr)) && \
357 (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
358 ((*((__u32 *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
359 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
360#endif /* 64-bit arch */
361
362/*
363 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
364 * not check it for lookups anymore, thanks Alexey. -DaveM
365 *
366 * Local BH must be disabled here.
367 */
368static inline struct sock *
369 __inet_lookup_established(struct inet_hashinfo *hashinfo,
370 const u32 saddr, const u16 sport,
371 const u32 daddr, const u16 hnum,
372 const int dif)
373{
374 INET_ADDR_COOKIE(acookie, saddr, daddr)
375 const __u32 ports = INET_COMBINED_PORTS(sport, hnum);
376 struct sock *sk;
377 const struct hlist_node *node;
378 /* Optimize here for direct hit, only listening connections can
379 * have wildcards anyways.
380 */
381 const int hash = inet_ehashfn(daddr, hnum, saddr, sport, hashinfo->ehash_size);
382 struct inet_ehash_bucket *head = &hashinfo->ehash[hash];
383
384 read_lock(&head->lock);
385 sk_for_each(sk, node, &head->chain) {
386 if (INET_MATCH(sk, acookie, saddr, daddr, ports, dif))
387 goto hit; /* You sunk my battleship! */
388 }
389
390 /* Must check for a TIME_WAIT'er before going to listener hash. */
391 sk_for_each(sk, node, &(head + hashinfo->ehash_size)->chain) {
392 if (INET_TW_MATCH(sk, acookie, saddr, daddr, ports, dif))
393 goto hit;
394 }
395 sk = NULL;
396out:
397 read_unlock(&head->lock);
398 return sk;
399hit:
400 sock_hold(sk);
401 goto out;
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 hnum,
407 const int dif)
408{
409 struct sock *sk = __inet_lookup_established(hashinfo, saddr, sport, daddr,
410 hnum, dif);
411 return sk ? : inet_lookup_listener(hashinfo, daddr, hnum, dif);
412}
413
414static inline struct sock *inet_lookup(struct inet_hashinfo *hashinfo,
415 const u32 saddr, const u16 sport,
416 const u32 daddr, const u16 dport,
417 const int dif)
418{
419 struct sock *sk;
420
421 local_bh_disable();
422 sk = __inet_lookup(hashinfo, saddr, sport, daddr, ntohs(dport), dif);
423 local_bh_enable();
424
425 return sk;
426}
427#endif /* _INET_HASHTABLES_H */