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 * Authors: Lotsa people, from code originally in tcp
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.
14 #ifndef _INET_HASHTABLES_H
15 #define _INET_HASHTABLES_H
18 #include <linux/interrupt.h>
19 #include <linux/ipv6.h>
20 #include <linux/list.h>
21 #include <linux/slab.h>
22 #include <linux/socket.h>
23 #include <linux/spinlock.h>
24 #include <linux/types.h>
25 #include <linux/wait.h>
26 #include <linux/vmalloc.h>
28 #include <net/inet_connection_sock.h>
29 #include <net/inet_sock.h>
31 #include <net/tcp_states.h>
33 #include <asm/atomic.h>
34 #include <asm/byteorder.h>
36 /* This is for all connections with a full identity, no wildcards.
37 * One chain is dedicated to TIME_WAIT sockets.
38 * I'll experiment with dynamic table growth later.
40 struct inet_ehash_bucket {
41 struct hlist_head chain;
42 struct hlist_head twchain;
45 /* There are a few simple rules, which allow for local port reuse by
46 * an application. In essence:
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
56 * Failing this, the port cannot be shared.
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.
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
76 struct inet_bind_bucket {
79 signed short fastreuse;
80 struct hlist_node node;
81 struct hlist_head owners;
84 #define inet_bind_bucket_for_each(tb, node, head) \
85 hlist_for_each_entry(tb, node, head, node)
87 struct inet_bind_hashbucket {
89 struct hlist_head chain;
92 /* This is for listening sockets, thus all sockets which possess wildcards. */
93 #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
95 struct inet_hashinfo {
96 /* This is for sockets with full identity only. Sockets here will
97 * always be without wildcards and will have the following invariant:
99 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
101 * TIME_WAIT sockets use a separate chain (twchain).
103 struct inet_ehash_bucket *ehash;
104 rwlock_t *ehash_locks;
105 unsigned int ehash_size;
106 unsigned int ehash_locks_mask;
108 /* Ok, let's try this, I give up, we do need a local binding
109 * TCP hash as well as the others for fast bind/connect.
111 struct inet_bind_hashbucket *bhash;
113 unsigned int bhash_size;
114 /* Note : 4 bytes padding on 64 bit arches */
116 /* All sockets in TCP_LISTEN state will be in here. This is the only
117 * table where wildcard'd TCP sockets can exist. Hash function here
118 * is just local port number.
120 struct hlist_head listening_hash[INET_LHTABLE_SIZE];
122 /* All the above members are written once at bootup and
123 * never written again _or_ are predominantly read-access.
125 * Now align to a new cache line as all the following members
128 rwlock_t lhash_lock ____cacheline_aligned;
129 atomic_t lhash_users;
130 wait_queue_head_t lhash_wait;
131 struct kmem_cache *bind_bucket_cachep;
134 static inline struct inet_ehash_bucket *inet_ehash_bucket(
135 struct inet_hashinfo *hashinfo,
138 return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)];
141 static inline rwlock_t *inet_ehash_lockp(
142 struct inet_hashinfo *hashinfo,
145 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
148 static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
150 unsigned int i, size = 256;
151 #if defined(CONFIG_PROVE_LOCKING)
152 unsigned int nr_pcpus = 2;
154 unsigned int nr_pcpus = num_possible_cpus();
164 if (sizeof(rwlock_t) != 0) {
166 if (size * sizeof(rwlock_t) > PAGE_SIZE)
167 hashinfo->ehash_locks = vmalloc(size * sizeof(rwlock_t));
170 hashinfo->ehash_locks = kmalloc(size * sizeof(rwlock_t),
172 if (!hashinfo->ehash_locks)
174 for (i = 0; i < size; i++)
175 rwlock_init(&hashinfo->ehash_locks[i]);
177 hashinfo->ehash_locks_mask = size - 1;
181 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
183 if (hashinfo->ehash_locks) {
185 unsigned int size = (hashinfo->ehash_locks_mask + 1) *
187 if (size > PAGE_SIZE)
188 vfree(hashinfo->ehash_locks);
191 kfree(hashinfo->ehash_locks);
192 hashinfo->ehash_locks = NULL;
196 extern struct inet_bind_bucket *
197 inet_bind_bucket_create(struct kmem_cache *cachep,
199 struct inet_bind_hashbucket *head,
200 const unsigned short snum);
201 extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
202 struct inet_bind_bucket *tb);
204 static inline int inet_bhashfn(const __u16 lport, const int bhash_size)
206 return lport & (bhash_size - 1);
209 extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
210 const unsigned short snum);
212 /* These can have wildcards, don't try too hard. */
213 static inline int inet_lhashfn(const unsigned short num)
215 return num & (INET_LHTABLE_SIZE - 1);
218 static inline int inet_sk_listen_hashfn(const struct sock *sk)
220 return inet_lhashfn(inet_sk(sk)->num);
223 /* Caller must disable local BH processing. */
224 static inline void __inet_inherit_port(struct sock *sk, struct sock *child)
226 struct inet_hashinfo *table = sk->sk_prot->hashinfo;
227 const int bhash = inet_bhashfn(inet_sk(child)->num, table->bhash_size);
228 struct inet_bind_hashbucket *head = &table->bhash[bhash];
229 struct inet_bind_bucket *tb;
231 spin_lock(&head->lock);
232 tb = inet_csk(sk)->icsk_bind_hash;
233 sk_add_bind_node(child, &tb->owners);
234 inet_csk(child)->icsk_bind_hash = tb;
235 spin_unlock(&head->lock);
238 static inline void inet_inherit_port(struct sock *sk, struct sock *child)
241 __inet_inherit_port(sk, child);
245 extern void inet_put_port(struct sock *sk);
247 extern void inet_listen_wlock(struct inet_hashinfo *hashinfo);
250 * - We may sleep inside this lock.
251 * - If sleeping is not required (or called from BH),
252 * use plain read_(un)lock(&inet_hashinfo.lhash_lock).
254 static inline void inet_listen_lock(struct inet_hashinfo *hashinfo)
256 /* read_lock synchronizes to candidates to writers */
257 read_lock(&hashinfo->lhash_lock);
258 atomic_inc(&hashinfo->lhash_users);
259 read_unlock(&hashinfo->lhash_lock);
262 static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo)
264 if (atomic_dec_and_test(&hashinfo->lhash_users))
265 wake_up(&hashinfo->lhash_wait);
268 extern void __inet_hash_nolisten(struct sock *sk);
269 extern void inet_hash(struct sock *sk);
270 extern void inet_unhash(struct sock *sk);
272 extern struct sock *__inet_lookup_listener(struct net *net,
273 struct inet_hashinfo *hashinfo,
275 const unsigned short hnum,
278 static inline struct sock *inet_lookup_listener(struct net *net,
279 struct inet_hashinfo *hashinfo,
280 __be32 daddr, __be16 dport, int dif)
282 return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif);
285 /* Socket demux engine toys. */
286 /* What happens here is ugly; there's a pair of adjacent fields in
287 struct inet_sock; __be16 dport followed by __u16 num. We want to
288 search by pair, so we combine the keys into a single 32bit value
289 and compare with 32bit value read from &...->dport. Let's at least
290 make sure that it's not mixed with anything else...
291 On 64bit targets we combine comparisons with pair of adjacent __be32
292 fields in the same way.
294 typedef __u32 __bitwise __portpair;
296 #define INET_COMBINED_PORTS(__sport, __dport) \
297 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
298 #else /* __LITTLE_ENDIAN */
299 #define INET_COMBINED_PORTS(__sport, __dport) \
300 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
303 #if (BITS_PER_LONG == 64)
304 typedef __u64 __bitwise __addrpair;
306 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
307 const __addrpair __name = (__force __addrpair) ( \
308 (((__force __u64)(__be32)(__saddr)) << 32) | \
309 ((__force __u64)(__be32)(__daddr)));
310 #else /* __LITTLE_ENDIAN */
311 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
312 const __addrpair __name = (__force __addrpair) ( \
313 (((__force __u64)(__be32)(__daddr)) << 32) | \
314 ((__force __u64)(__be32)(__saddr)));
315 #endif /* __BIG_ENDIAN */
316 #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
317 (((__sk)->sk_hash == (__hash)) && ((__sk)->sk_net == (__net)) && \
318 ((*((__addrpair *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \
319 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
320 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
321 #define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
322 (((__sk)->sk_hash == (__hash)) && ((__sk)->sk_net == (__net)) && \
323 ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \
324 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
325 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
326 #else /* 32-bit arch */
327 #define INET_ADDR_COOKIE(__name, __saddr, __daddr)
328 #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif) \
329 (((__sk)->sk_hash == (__hash)) && ((__sk)->sk_net == (__net)) && \
330 (inet_sk(__sk)->daddr == (__saddr)) && \
331 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
332 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
333 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
334 #define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif) \
335 (((__sk)->sk_hash == (__hash)) && ((__sk)->sk_net == (__net)) && \
336 (inet_twsk(__sk)->tw_daddr == (__saddr)) && \
337 (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
338 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
339 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
340 #endif /* 64-bit arch */
343 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
344 * not check it for lookups anymore, thanks Alexey. -DaveM
346 * Local BH must be disabled here.
348 extern struct sock * __inet_lookup_established(struct net *net,
349 struct inet_hashinfo *hashinfo,
350 const __be32 saddr, const __be16 sport,
351 const __be32 daddr, const u16 hnum, const int dif);
353 static inline struct sock *
354 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
355 const __be32 saddr, const __be16 sport,
356 const __be32 daddr, const __be16 dport,
359 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
363 static inline struct sock *__inet_lookup(struct net *net,
364 struct inet_hashinfo *hashinfo,
365 const __be32 saddr, const __be16 sport,
366 const __be32 daddr, const __be16 dport,
369 u16 hnum = ntohs(dport);
370 struct sock *sk = __inet_lookup_established(net, hashinfo,
371 saddr, sport, daddr, hnum, dif);
373 return sk ? : __inet_lookup_listener(net, hashinfo, daddr, hnum, dif);
376 static inline struct sock *inet_lookup(struct net *net,
377 struct inet_hashinfo *hashinfo,
378 const __be32 saddr, const __be16 sport,
379 const __be32 daddr, const __be16 dport,
385 sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
391 extern int __inet_hash_connect(struct inet_timewait_death_row *death_row,
392 struct sock *sk, u32 port_offset,
393 int (*check_established)(struct inet_timewait_death_row *,
394 struct sock *, __u16, struct inet_timewait_sock **),
395 void (*hash)(struct sock *sk));
396 extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
398 #endif /* _INET_HASHTABLES_H */