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>
32 #include <net/netns/hash.h>
34 #include <asm/atomic.h>
35 #include <asm/byteorder.h>
37 /* This is for all connections with a full identity, no wildcards.
38 * One chain is dedicated to TIME_WAIT sockets.
39 * I'll experiment with dynamic table growth later.
41 struct inet_ehash_bucket {
42 struct hlist_head chain;
43 struct hlist_head twchain;
46 /* There are a few simple rules, which allow for local port reuse by
47 * an application. In essence:
49 * 1) Sockets bound to different interfaces may share a local port.
50 * Failing that, goto test 2.
51 * 2) If all sockets have sk->sk_reuse set, and none of them are in
52 * TCP_LISTEN state, the port may be shared.
53 * Failing that, goto test 3.
54 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
55 * address, and none of them are the same, the port may be
57 * Failing this, the port cannot be shared.
59 * The interesting point, is test #2. This is what an FTP server does
60 * all day. To optimize this case we use a specific flag bit defined
61 * below. As we add sockets to a bind bucket list, we perform a
62 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
63 * As long as all sockets added to a bind bucket pass this test,
64 * the flag bit will be set.
65 * The resulting situation is that tcp_v[46]_verify_bind() can just check
66 * for this flag bit, if it is set and the socket trying to bind has
67 * sk->sk_reuse set, we don't even have to walk the owners list at all,
68 * we return that it is ok to bind this socket to the requested local port.
70 * Sounds like a lot of work, but it is worth it. In a more naive
71 * implementation (ie. current FreeBSD etc.) the entire list of ports
72 * must be walked for each data port opened by an ftp server. Needless
73 * to say, this does not scale at all. With a couple thousand FTP
74 * users logged onto your box, isn't it nice to know that new data
75 * ports are created in O(1) time? I thought so. ;-) -DaveM
77 struct inet_bind_bucket {
80 signed short fastreuse;
81 struct hlist_node node;
82 struct hlist_head owners;
85 #define inet_bind_bucket_for_each(tb, node, head) \
86 hlist_for_each_entry(tb, node, head, node)
88 struct inet_bind_hashbucket {
90 struct hlist_head chain;
93 /* This is for listening sockets, thus all sockets which possess wildcards. */
94 #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
96 struct inet_hashinfo {
97 /* This is for sockets with full identity only. Sockets here will
98 * always be without wildcards and will have the following invariant:
100 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
102 * TIME_WAIT sockets use a separate chain (twchain).
104 struct inet_ehash_bucket *ehash;
105 rwlock_t *ehash_locks;
106 unsigned int ehash_size;
107 unsigned int ehash_locks_mask;
109 /* Ok, let's try this, I give up, we do need a local binding
110 * TCP hash as well as the others for fast bind/connect.
112 struct inet_bind_hashbucket *bhash;
114 unsigned int bhash_size;
115 /* Note : 4 bytes padding on 64 bit arches */
117 /* All sockets in TCP_LISTEN state will be in here. This is the only
118 * table where wildcard'd TCP sockets can exist. Hash function here
119 * is just local port number.
121 struct hlist_head listening_hash[INET_LHTABLE_SIZE];
123 /* All the above members are written once at bootup and
124 * never written again _or_ are predominantly read-access.
126 * Now align to a new cache line as all the following members
129 rwlock_t lhash_lock ____cacheline_aligned;
130 atomic_t lhash_users;
131 wait_queue_head_t lhash_wait;
132 struct kmem_cache *bind_bucket_cachep;
135 static inline struct inet_ehash_bucket *inet_ehash_bucket(
136 struct inet_hashinfo *hashinfo,
139 return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)];
142 static inline rwlock_t *inet_ehash_lockp(
143 struct inet_hashinfo *hashinfo,
146 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
149 static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
151 unsigned int i, size = 256;
152 #if defined(CONFIG_PROVE_LOCKING)
153 unsigned int nr_pcpus = 2;
155 unsigned int nr_pcpus = num_possible_cpus();
165 if (sizeof(rwlock_t) != 0) {
167 if (size * sizeof(rwlock_t) > PAGE_SIZE)
168 hashinfo->ehash_locks = vmalloc(size * sizeof(rwlock_t));
171 hashinfo->ehash_locks = kmalloc(size * sizeof(rwlock_t),
173 if (!hashinfo->ehash_locks)
175 for (i = 0; i < size; i++)
176 rwlock_init(&hashinfo->ehash_locks[i]);
178 hashinfo->ehash_locks_mask = size - 1;
182 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
184 if (hashinfo->ehash_locks) {
186 unsigned int size = (hashinfo->ehash_locks_mask + 1) *
188 if (size > PAGE_SIZE)
189 vfree(hashinfo->ehash_locks);
192 kfree(hashinfo->ehash_locks);
193 hashinfo->ehash_locks = NULL;
197 extern struct inet_bind_bucket *
198 inet_bind_bucket_create(struct kmem_cache *cachep,
200 struct inet_bind_hashbucket *head,
201 const unsigned short snum);
202 extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
203 struct inet_bind_bucket *tb);
205 static inline int inet_bhashfn(struct net *net,
206 const __u16 lport, const int bhash_size)
208 return (lport + net_hash_mix(net)) & (bhash_size - 1);
211 extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
212 const unsigned short snum);
214 /* These can have wildcards, don't try too hard. */
215 static inline int inet_lhashfn(struct net *net, const unsigned short num)
217 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
220 static inline int inet_sk_listen_hashfn(const struct sock *sk)
222 return inet_lhashfn(sock_net(sk), inet_sk(sk)->num);
225 /* Caller must disable local BH processing. */
226 extern void __inet_inherit_port(struct sock *sk, struct sock *child);
228 extern void inet_put_port(struct sock *sk);
230 extern void inet_listen_wlock(struct inet_hashinfo *hashinfo);
233 * - We may sleep inside this lock.
234 * - If sleeping is not required (or called from BH),
235 * use plain read_(un)lock(&inet_hashinfo.lhash_lock).
237 static inline void inet_listen_lock(struct inet_hashinfo *hashinfo)
239 /* read_lock synchronizes to candidates to writers */
240 read_lock(&hashinfo->lhash_lock);
241 atomic_inc(&hashinfo->lhash_users);
242 read_unlock(&hashinfo->lhash_lock);
245 static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo)
247 if (atomic_dec_and_test(&hashinfo->lhash_users))
248 wake_up(&hashinfo->lhash_wait);
251 extern void __inet_hash_nolisten(struct sock *sk);
252 extern void inet_hash(struct sock *sk);
253 extern void inet_unhash(struct sock *sk);
255 extern struct sock *__inet_lookup_listener(struct net *net,
256 struct inet_hashinfo *hashinfo,
258 const unsigned short hnum,
261 static inline struct sock *inet_lookup_listener(struct net *net,
262 struct inet_hashinfo *hashinfo,
263 __be32 daddr, __be16 dport, int dif)
265 return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif);
268 /* Socket demux engine toys. */
269 /* What happens here is ugly; there's a pair of adjacent fields in
270 struct inet_sock; __be16 dport followed by __u16 num. We want to
271 search by pair, so we combine the keys into a single 32bit value
272 and compare with 32bit value read from &...->dport. Let's at least
273 make sure that it's not mixed with anything else...
274 On 64bit targets we combine comparisons with pair of adjacent __be32
275 fields in the same way.
277 typedef __u32 __bitwise __portpair;
279 #define INET_COMBINED_PORTS(__sport, __dport) \
280 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
281 #else /* __LITTLE_ENDIAN */
282 #define INET_COMBINED_PORTS(__sport, __dport) \
283 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
286 #if (BITS_PER_LONG == 64)
287 typedef __u64 __bitwise __addrpair;
289 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
290 const __addrpair __name = (__force __addrpair) ( \
291 (((__force __u64)(__be32)(__saddr)) << 32) | \
292 ((__force __u64)(__be32)(__daddr)));
293 #else /* __LITTLE_ENDIAN */
294 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
295 const __addrpair __name = (__force __addrpair) ( \
296 (((__force __u64)(__be32)(__daddr)) << 32) | \
297 ((__force __u64)(__be32)(__saddr)));
298 #endif /* __BIG_ENDIAN */
299 #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
300 (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \
301 ((*((__addrpair *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \
302 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
303 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
304 #define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
305 (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \
306 ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \
307 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
308 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
309 #else /* 32-bit arch */
310 #define INET_ADDR_COOKIE(__name, __saddr, __daddr)
311 #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif) \
312 (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \
313 (inet_sk(__sk)->daddr == (__saddr)) && \
314 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
315 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
316 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
317 #define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif) \
318 (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \
319 (inet_twsk(__sk)->tw_daddr == (__saddr)) && \
320 (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
321 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
322 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
323 #endif /* 64-bit arch */
326 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
327 * not check it for lookups anymore, thanks Alexey. -DaveM
329 * Local BH must be disabled here.
331 extern struct sock * __inet_lookup_established(struct net *net,
332 struct inet_hashinfo *hashinfo,
333 const __be32 saddr, const __be16 sport,
334 const __be32 daddr, const u16 hnum, const int dif);
336 static inline struct sock *
337 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
338 const __be32 saddr, const __be16 sport,
339 const __be32 daddr, const __be16 dport,
342 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
346 static inline struct sock *__inet_lookup(struct net *net,
347 struct inet_hashinfo *hashinfo,
348 const __be32 saddr, const __be16 sport,
349 const __be32 daddr, const __be16 dport,
352 u16 hnum = ntohs(dport);
353 struct sock *sk = __inet_lookup_established(net, hashinfo,
354 saddr, sport, daddr, hnum, dif);
356 return sk ? : __inet_lookup_listener(net, hashinfo, daddr, hnum, dif);
359 static inline struct sock *inet_lookup(struct net *net,
360 struct inet_hashinfo *hashinfo,
361 const __be32 saddr, const __be16 sport,
362 const __be32 daddr, const __be16 dport,
368 sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
374 extern int __inet_hash_connect(struct inet_timewait_death_row *death_row,
375 struct sock *sk, u32 port_offset,
376 int (*check_established)(struct inet_timewait_death_row *,
377 struct sock *, __u16, struct inet_timewait_sock **),
378 void (*hash)(struct sock *sk));
379 extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
381 #endif /* _INET_HASHTABLES_H */