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 {
78 signed short fastreuse;
79 struct hlist_node node;
80 struct hlist_head owners;
83 #define inet_bind_bucket_for_each(tb, node, head) \
84 hlist_for_each_entry(tb, node, head, node)
86 struct inet_bind_hashbucket {
88 struct hlist_head chain;
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. */
94 struct 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:
98 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
100 * TIME_WAIT sockets use a separate chain (twchain).
102 struct inet_ehash_bucket *ehash;
103 rwlock_t *ehash_locks;
104 unsigned int ehash_size;
105 unsigned int ehash_locks_mask;
107 /* Ok, let's try this, I give up, we do need a local binding
108 * TCP hash as well as the others for fast bind/connect.
110 struct inet_bind_hashbucket *bhash;
112 unsigned int bhash_size;
113 /* Note : 4 bytes padding on 64 bit arches */
115 /* All sockets in TCP_LISTEN state will be in here. This is the only
116 * table where wildcard'd TCP sockets can exist. Hash function here
117 * is just local port number.
119 struct hlist_head listening_hash[INET_LHTABLE_SIZE];
121 /* All the above members are written once at bootup and
122 * never written again _or_ are predominantly read-access.
124 * Now align to a new cache line as all the following members
127 rwlock_t lhash_lock ____cacheline_aligned;
128 atomic_t lhash_users;
129 wait_queue_head_t lhash_wait;
130 struct kmem_cache *bind_bucket_cachep;
133 static inline struct inet_ehash_bucket *inet_ehash_bucket(
134 struct inet_hashinfo *hashinfo,
137 return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)];
140 static inline rwlock_t *inet_ehash_lockp(
141 struct inet_hashinfo *hashinfo,
144 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
147 static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
149 unsigned int i, size = 256;
150 #if defined(CONFIG_PROVE_LOCKING)
151 unsigned int nr_pcpus = 2;
153 unsigned int nr_pcpus = num_possible_cpus();
163 if (sizeof(rwlock_t) != 0) {
165 if (size * sizeof(rwlock_t) > PAGE_SIZE)
166 hashinfo->ehash_locks = vmalloc(size * sizeof(rwlock_t));
169 hashinfo->ehash_locks = kmalloc(size * sizeof(rwlock_t),
171 if (!hashinfo->ehash_locks)
173 for (i = 0; i < size; i++)
174 rwlock_init(&hashinfo->ehash_locks[i]);
176 hashinfo->ehash_locks_mask = size - 1;
180 static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
182 if (hashinfo->ehash_locks) {
184 unsigned int size = (hashinfo->ehash_locks_mask + 1) *
186 if (size > PAGE_SIZE)
187 vfree(hashinfo->ehash_locks);
190 kfree(hashinfo->ehash_locks);
191 hashinfo->ehash_locks = NULL;
195 extern struct inet_bind_bucket *
196 inet_bind_bucket_create(struct kmem_cache *cachep,
197 struct inet_bind_hashbucket *head,
198 const unsigned short snum);
199 extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
200 struct inet_bind_bucket *tb);
202 static inline int inet_bhashfn(const __u16 lport, const int bhash_size)
204 return lport & (bhash_size - 1);
207 extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
208 const unsigned short snum);
210 /* These can have wildcards, don't try too hard. */
211 static inline int inet_lhashfn(const unsigned short num)
213 return num & (INET_LHTABLE_SIZE - 1);
216 static inline int inet_sk_listen_hashfn(const struct sock *sk)
218 return inet_lhashfn(inet_sk(sk)->num);
221 /* Caller must disable local BH processing. */
222 static inline void __inet_inherit_port(struct inet_hashinfo *table,
223 struct sock *sk, struct sock *child)
225 const int bhash = inet_bhashfn(inet_sk(child)->num, table->bhash_size);
226 struct inet_bind_hashbucket *head = &table->bhash[bhash];
227 struct inet_bind_bucket *tb;
229 spin_lock(&head->lock);
230 tb = inet_csk(sk)->icsk_bind_hash;
231 sk_add_bind_node(child, &tb->owners);
232 inet_csk(child)->icsk_bind_hash = tb;
233 spin_unlock(&head->lock);
236 static inline void inet_inherit_port(struct inet_hashinfo *table,
237 struct sock *sk, struct sock *child)
240 __inet_inherit_port(table, sk, child);
244 extern void inet_put_port(struct inet_hashinfo *table, struct sock *sk);
246 extern void inet_listen_wlock(struct inet_hashinfo *hashinfo);
249 * - We may sleep inside this lock.
250 * - If sleeping is not required (or called from BH),
251 * use plain read_(un)lock(&inet_hashinfo.lhash_lock).
253 static inline void inet_listen_lock(struct inet_hashinfo *hashinfo)
255 /* read_lock synchronizes to candidates to writers */
256 read_lock(&hashinfo->lhash_lock);
257 atomic_inc(&hashinfo->lhash_users);
258 read_unlock(&hashinfo->lhash_lock);
261 static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo)
263 if (atomic_dec_and_test(&hashinfo->lhash_users))
264 wake_up(&hashinfo->lhash_wait);
267 static inline void __inet_hash(struct inet_hashinfo *hashinfo,
268 struct sock *sk, const int listen_possible)
270 struct hlist_head *list;
273 BUG_TRAP(sk_unhashed(sk));
274 if (listen_possible && sk->sk_state == TCP_LISTEN) {
275 list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
276 lock = &hashinfo->lhash_lock;
277 inet_listen_wlock(hashinfo);
279 struct inet_ehash_bucket *head;
280 sk->sk_hash = inet_sk_ehashfn(sk);
281 head = inet_ehash_bucket(hashinfo, sk->sk_hash);
283 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
286 __sk_add_node(sk, list);
287 sock_prot_inc_use(sk->sk_prot);
289 if (listen_possible && sk->sk_state == TCP_LISTEN)
290 wake_up(&hashinfo->lhash_wait);
293 static inline void inet_hash(struct inet_hashinfo *hashinfo, struct sock *sk)
295 if (sk->sk_state != TCP_CLOSE) {
297 __inet_hash(hashinfo, sk, 1);
302 static inline void inet_unhash(struct inet_hashinfo *hashinfo, struct sock *sk)
309 if (sk->sk_state == TCP_LISTEN) {
311 inet_listen_wlock(hashinfo);
312 lock = &hashinfo->lhash_lock;
314 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
318 if (__sk_del_node_init(sk))
319 sock_prot_dec_use(sk->sk_prot);
320 write_unlock_bh(lock);
322 if (sk->sk_state == TCP_LISTEN)
323 wake_up(&hashinfo->lhash_wait);
326 extern struct sock *__inet_lookup_listener(struct inet_hashinfo *hashinfo,
328 const unsigned short hnum,
331 static inline struct sock *inet_lookup_listener(struct inet_hashinfo *hashinfo,
332 __be32 daddr, __be16 dport, int dif)
334 return __inet_lookup_listener(hashinfo, daddr, ntohs(dport), dif);
337 /* Socket demux engine toys. */
338 /* What happens here is ugly; there's a pair of adjacent fields in
339 struct inet_sock; __be16 dport followed by __u16 num. We want to
340 search by pair, so we combine the keys into a single 32bit value
341 and compare with 32bit value read from &...->dport. Let's at least
342 make sure that it's not mixed with anything else...
343 On 64bit targets we combine comparisons with pair of adjacent __be32
344 fields in the same way.
346 typedef __u32 __bitwise __portpair;
348 #define INET_COMBINED_PORTS(__sport, __dport) \
349 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
350 #else /* __LITTLE_ENDIAN */
351 #define INET_COMBINED_PORTS(__sport, __dport) \
352 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
355 #if (BITS_PER_LONG == 64)
356 typedef __u64 __bitwise __addrpair;
358 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
359 const __addrpair __name = (__force __addrpair) ( \
360 (((__force __u64)(__be32)(__saddr)) << 32) | \
361 ((__force __u64)(__be32)(__daddr)));
362 #else /* __LITTLE_ENDIAN */
363 #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
364 const __addrpair __name = (__force __addrpair) ( \
365 (((__force __u64)(__be32)(__daddr)) << 32) | \
366 ((__force __u64)(__be32)(__saddr)));
367 #endif /* __BIG_ENDIAN */
368 #define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
369 (((__sk)->sk_hash == (__hash)) && \
370 ((*((__addrpair *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \
371 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
372 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
373 #define INET_TW_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
374 (((__sk)->sk_hash == (__hash)) && \
375 ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \
376 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
377 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
378 #else /* 32-bit arch */
379 #define INET_ADDR_COOKIE(__name, __saddr, __daddr)
380 #define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif) \
381 (((__sk)->sk_hash == (__hash)) && \
382 (inet_sk(__sk)->daddr == (__saddr)) && \
383 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
384 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
385 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
386 #define INET_TW_MATCH(__sk, __hash,__cookie, __saddr, __daddr, __ports, __dif) \
387 (((__sk)->sk_hash == (__hash)) && \
388 (inet_twsk(__sk)->tw_daddr == (__saddr)) && \
389 (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
390 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
391 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
392 #endif /* 64-bit arch */
395 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
396 * not check it for lookups anymore, thanks Alexey. -DaveM
398 * Local BH must be disabled here.
400 static inline struct sock *
401 __inet_lookup_established(struct inet_hashinfo *hashinfo,
402 const __be32 saddr, const __be16 sport,
403 const __be32 daddr, const u16 hnum,
406 INET_ADDR_COOKIE(acookie, saddr, daddr)
407 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
409 const struct hlist_node *node;
410 /* Optimize here for direct hit, only listening connections can
411 * have wildcards anyways.
413 unsigned int hash = inet_ehashfn(daddr, hnum, saddr, sport);
414 struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash);
415 rwlock_t *lock = inet_ehash_lockp(hashinfo, hash);
417 prefetch(head->chain.first);
419 sk_for_each(sk, node, &head->chain) {
420 if (INET_MATCH(sk, hash, acookie, saddr, daddr, ports, dif))
421 goto hit; /* You sunk my battleship! */
424 /* Must check for a TIME_WAIT'er before going to listener hash. */
425 sk_for_each(sk, node, &head->twchain) {
426 if (INET_TW_MATCH(sk, hash, acookie, saddr, daddr, ports, dif))
438 static inline struct sock *
439 inet_lookup_established(struct inet_hashinfo *hashinfo,
440 const __be32 saddr, const __be16 sport,
441 const __be32 daddr, const __be16 dport,
444 return __inet_lookup_established(hashinfo, saddr, sport, daddr,
448 static inline struct sock *__inet_lookup(struct inet_hashinfo *hashinfo,
449 const __be32 saddr, const __be16 sport,
450 const __be32 daddr, const __be16 dport,
453 u16 hnum = ntohs(dport);
454 struct sock *sk = __inet_lookup_established(hashinfo, saddr, sport, daddr,
456 return sk ? : __inet_lookup_listener(hashinfo, daddr, hnum, dif);
459 static inline struct sock *inet_lookup(struct inet_hashinfo *hashinfo,
460 const __be32 saddr, const __be16 sport,
461 const __be32 daddr, const __be16 dport,
467 sk = __inet_lookup(hashinfo, saddr, sport, daddr, dport, dif);
473 extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
475 #endif /* _INET_HASHTABLES_H */