Commit | Line | Data |
---|---|---|
304a1618 ACM |
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 | ||
8feaf0c0 | 17 | |
2d8c4ce5 | 18 | #include <linux/interrupt.h> |
33b62231 | 19 | #include <linux/ipv6.h> |
77d8bf9c ACM |
20 | #include <linux/list.h> |
21 | #include <linux/slab.h> | |
33b62231 | 22 | #include <linux/socket.h> |
77d8bf9c | 23 | #include <linux/spinlock.h> |
304a1618 | 24 | #include <linux/types.h> |
f3f05f70 | 25 | #include <linux/wait.h> |
304a1618 | 26 | |
463c84b9 | 27 | #include <net/inet_connection_sock.h> |
14c85021 | 28 | #include <net/inet_sock.h> |
2d8c4ce5 | 29 | #include <net/sock.h> |
c752f073 | 30 | #include <net/tcp_states.h> |
2d8c4ce5 | 31 | |
f3f05f70 | 32 | #include <asm/atomic.h> |
e48c414e | 33 | #include <asm/byteorder.h> |
f3f05f70 | 34 | |
77d8bf9c | 35 | /* This is for all connections with a full identity, no wildcards. |
dbca9b27 ED |
36 | * One chain is dedicated to TIME_WAIT sockets. |
37 | * I'll experiment with dynamic table growth later. | |
77d8bf9c ACM |
38 | */ |
39 | struct inet_ehash_bucket { | |
40 | rwlock_t lock; | |
41 | struct hlist_head chain; | |
dbca9b27 | 42 | struct hlist_head twchain; |
6d255361 | 43 | }; |
77d8bf9c ACM |
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 | */ | |
76 | struct 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 | ||
86 | struct 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 | ||
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: | |
97 | * | |
98 | * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE | |
99 | * | |
dbca9b27 | 100 | * TIME_WAIT sockets use a separate chain (twchain). |
77d8bf9c ACM |
101 | */ |
102 | struct inet_ehash_bucket *ehash; | |
103 | ||
104 | /* Ok, let's try this, I give up, we do need a local binding | |
105 | * TCP hash as well as the others for fast bind/connect. | |
106 | */ | |
107 | struct inet_bind_hashbucket *bhash; | |
108 | ||
cfcabdcc | 109 | unsigned int bhash_size; |
81c3d547 | 110 | unsigned int ehash_size; |
77d8bf9c ACM |
111 | |
112 | /* All sockets in TCP_LISTEN state will be in here. This is the only | |
113 | * table where wildcard'd TCP sockets can exist. Hash function here | |
114 | * is just local port number. | |
115 | */ | |
116 | struct hlist_head listening_hash[INET_LHTABLE_SIZE]; | |
117 | ||
118 | /* All the above members are written once at bootup and | |
119 | * never written again _or_ are predominantly read-access. | |
120 | * | |
121 | * Now align to a new cache line as all the following members | |
122 | * are often dirty. | |
123 | */ | |
124 | rwlock_t lhash_lock ____cacheline_aligned; | |
125 | atomic_t lhash_users; | |
126 | wait_queue_head_t lhash_wait; | |
e18b890b | 127 | struct kmem_cache *bind_bucket_cachep; |
77d8bf9c ACM |
128 | }; |
129 | ||
81c3d547 ED |
130 | static inline struct inet_ehash_bucket *inet_ehash_bucket( |
131 | struct inet_hashinfo *hashinfo, | |
132 | unsigned int hash) | |
133 | { | |
134 | return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)]; | |
304a1618 ACM |
135 | } |
136 | ||
77d8bf9c | 137 | extern struct inet_bind_bucket * |
e18b890b | 138 | inet_bind_bucket_create(struct kmem_cache *cachep, |
77d8bf9c ACM |
139 | struct inet_bind_hashbucket *head, |
140 | const unsigned short snum); | |
e18b890b | 141 | extern void inet_bind_bucket_destroy(struct kmem_cache *cachep, |
77d8bf9c ACM |
142 | struct inet_bind_bucket *tb); |
143 | ||
144 | static inline int inet_bhashfn(const __u16 lport, const int bhash_size) | |
145 | { | |
146 | return lport & (bhash_size - 1); | |
147 | } | |
148 | ||
2d8c4ce5 ACM |
149 | extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, |
150 | const unsigned short snum); | |
151 | ||
77d8bf9c ACM |
152 | /* These can have wildcards, don't try too hard. */ |
153 | static inline int inet_lhashfn(const unsigned short num) | |
154 | { | |
155 | return num & (INET_LHTABLE_SIZE - 1); | |
156 | } | |
157 | ||
158 | static inline int inet_sk_listen_hashfn(const struct sock *sk) | |
159 | { | |
160 | return inet_lhashfn(inet_sk(sk)->num); | |
161 | } | |
162 | ||
2d8c4ce5 ACM |
163 | /* Caller must disable local BH processing. */ |
164 | static inline void __inet_inherit_port(struct inet_hashinfo *table, | |
165 | struct sock *sk, struct sock *child) | |
166 | { | |
167 | const int bhash = inet_bhashfn(inet_sk(child)->num, table->bhash_size); | |
168 | struct inet_bind_hashbucket *head = &table->bhash[bhash]; | |
169 | struct inet_bind_bucket *tb; | |
170 | ||
171 | spin_lock(&head->lock); | |
463c84b9 | 172 | tb = inet_csk(sk)->icsk_bind_hash; |
2d8c4ce5 | 173 | sk_add_bind_node(child, &tb->owners); |
463c84b9 | 174 | inet_csk(child)->icsk_bind_hash = tb; |
2d8c4ce5 ACM |
175 | spin_unlock(&head->lock); |
176 | } | |
177 | ||
178 | static inline void inet_inherit_port(struct inet_hashinfo *table, | |
179 | struct sock *sk, struct sock *child) | |
180 | { | |
181 | local_bh_disable(); | |
182 | __inet_inherit_port(table, sk, child); | |
183 | local_bh_enable(); | |
184 | } | |
185 | ||
186 | extern void inet_put_port(struct inet_hashinfo *table, struct sock *sk); | |
187 | ||
f3f05f70 ACM |
188 | extern void inet_listen_wlock(struct inet_hashinfo *hashinfo); |
189 | ||
190 | /* | |
191 | * - We may sleep inside this lock. | |
192 | * - If sleeping is not required (or called from BH), | |
193 | * use plain read_(un)lock(&inet_hashinfo.lhash_lock). | |
194 | */ | |
195 | static inline void inet_listen_lock(struct inet_hashinfo *hashinfo) | |
196 | { | |
197 | /* read_lock synchronizes to candidates to writers */ | |
198 | read_lock(&hashinfo->lhash_lock); | |
199 | atomic_inc(&hashinfo->lhash_users); | |
200 | read_unlock(&hashinfo->lhash_lock); | |
201 | } | |
202 | ||
203 | static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo) | |
204 | { | |
205 | if (atomic_dec_and_test(&hashinfo->lhash_users)) | |
206 | wake_up(&hashinfo->lhash_wait); | |
207 | } | |
208 | ||
209 | static inline void __inet_hash(struct inet_hashinfo *hashinfo, | |
210 | struct sock *sk, const int listen_possible) | |
211 | { | |
212 | struct hlist_head *list; | |
213 | rwlock_t *lock; | |
214 | ||
215 | BUG_TRAP(sk_unhashed(sk)); | |
216 | if (listen_possible && sk->sk_state == TCP_LISTEN) { | |
217 | list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)]; | |
218 | lock = &hashinfo->lhash_lock; | |
219 | inet_listen_wlock(hashinfo); | |
220 | } else { | |
81c3d547 ED |
221 | struct inet_ehash_bucket *head; |
222 | sk->sk_hash = inet_sk_ehashfn(sk); | |
223 | head = inet_ehash_bucket(hashinfo, sk->sk_hash); | |
224 | list = &head->chain; | |
225 | lock = &head->lock; | |
f3f05f70 ACM |
226 | write_lock(lock); |
227 | } | |
228 | __sk_add_node(sk, list); | |
229 | sock_prot_inc_use(sk->sk_prot); | |
230 | write_unlock(lock); | |
231 | if (listen_possible && sk->sk_state == TCP_LISTEN) | |
232 | wake_up(&hashinfo->lhash_wait); | |
233 | } | |
81849d10 ACM |
234 | |
235 | static inline void inet_hash(struct inet_hashinfo *hashinfo, struct sock *sk) | |
236 | { | |
237 | if (sk->sk_state != TCP_CLOSE) { | |
238 | local_bh_disable(); | |
239 | __inet_hash(hashinfo, sk, 1); | |
240 | local_bh_enable(); | |
241 | } | |
242 | } | |
243 | ||
244 | static inline void inet_unhash(struct inet_hashinfo *hashinfo, struct sock *sk) | |
245 | { | |
246 | rwlock_t *lock; | |
247 | ||
248 | if (sk_unhashed(sk)) | |
249 | goto out; | |
250 | ||
251 | if (sk->sk_state == TCP_LISTEN) { | |
252 | local_bh_disable(); | |
253 | inet_listen_wlock(hashinfo); | |
254 | lock = &hashinfo->lhash_lock; | |
255 | } else { | |
81c3d547 ED |
256 | lock = &inet_ehash_bucket(hashinfo, sk->sk_hash)->lock; |
257 | write_lock_bh(lock); | |
81849d10 ACM |
258 | } |
259 | ||
260 | if (__sk_del_node_init(sk)) | |
261 | sock_prot_dec_use(sk->sk_prot); | |
262 | write_unlock_bh(lock); | |
263 | out: | |
264 | if (sk->sk_state == TCP_LISTEN) | |
265 | wake_up(&hashinfo->lhash_wait); | |
266 | } | |
33b62231 | 267 | |
8f491069 | 268 | extern struct sock *__inet_lookup_listener(struct inet_hashinfo *hashinfo, |
fb99c848 | 269 | const __be32 daddr, |
8f491069 HX |
270 | const unsigned short hnum, |
271 | const int dif); | |
272 | ||
273 | static inline struct sock *inet_lookup_listener(struct inet_hashinfo *hashinfo, | |
fb99c848 | 274 | __be32 daddr, __be16 dport, int dif) |
8f491069 HX |
275 | { |
276 | return __inet_lookup_listener(hashinfo, daddr, ntohs(dport), dif); | |
277 | } | |
8feaf0c0 ACM |
278 | |
279 | /* Socket demux engine toys. */ | |
4f765d84 AV |
280 | /* What happens here is ugly; there's a pair of adjacent fields in |
281 | struct inet_sock; __be16 dport followed by __u16 num. We want to | |
282 | search by pair, so we combine the keys into a single 32bit value | |
283 | and compare with 32bit value read from &...->dport. Let's at least | |
284 | make sure that it's not mixed with anything else... | |
285 | On 64bit targets we combine comparisons with pair of adjacent __be32 | |
286 | fields in the same way. | |
287 | */ | |
288 | typedef __u32 __bitwise __portpair; | |
8feaf0c0 ACM |
289 | #ifdef __BIG_ENDIAN |
290 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 291 | ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport))) |
8feaf0c0 ACM |
292 | #else /* __LITTLE_ENDIAN */ |
293 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 294 | ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport))) |
8feaf0c0 ACM |
295 | #endif |
296 | ||
297 | #if (BITS_PER_LONG == 64) | |
4f765d84 | 298 | typedef __u64 __bitwise __addrpair; |
8feaf0c0 ACM |
299 | #ifdef __BIG_ENDIAN |
300 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
301 | const __addrpair __name = (__force __addrpair) ( \ |
302 | (((__force __u64)(__be32)(__saddr)) << 32) | \ | |
303 | ((__force __u64)(__be32)(__daddr))); | |
8feaf0c0 ACM |
304 | #else /* __LITTLE_ENDIAN */ |
305 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
306 | const __addrpair __name = (__force __addrpair) ( \ |
307 | (((__force __u64)(__be32)(__daddr)) << 32) | \ | |
308 | ((__force __u64)(__be32)(__saddr))); | |
8feaf0c0 | 309 | #endif /* __BIG_ENDIAN */ |
81c3d547 ED |
310 | #define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ |
311 | (((__sk)->sk_hash == (__hash)) && \ | |
4f765d84 AV |
312 | ((*((__addrpair *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \ |
313 | ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \ | |
8feaf0c0 | 314 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
81c3d547 ED |
315 | #define INET_TW_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ |
316 | (((__sk)->sk_hash == (__hash)) && \ | |
4f765d84 AV |
317 | ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \ |
318 | ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \ | |
8feaf0c0 ACM |
319 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
320 | #else /* 32-bit arch */ | |
321 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) | |
81c3d547 ED |
322 | #define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif) \ |
323 | (((__sk)->sk_hash == (__hash)) && \ | |
324 | (inet_sk(__sk)->daddr == (__saddr)) && \ | |
8feaf0c0 | 325 | (inet_sk(__sk)->rcv_saddr == (__daddr)) && \ |
4f765d84 | 326 | ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \ |
8feaf0c0 | 327 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
81c3d547 ED |
328 | #define INET_TW_MATCH(__sk, __hash,__cookie, __saddr, __daddr, __ports, __dif) \ |
329 | (((__sk)->sk_hash == (__hash)) && \ | |
330 | (inet_twsk(__sk)->tw_daddr == (__saddr)) && \ | |
8feaf0c0 | 331 | (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \ |
4f765d84 | 332 | ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \ |
8feaf0c0 ACM |
333 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
334 | #endif /* 64-bit arch */ | |
e48c414e ACM |
335 | |
336 | /* | |
337 | * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need | |
338 | * not check it for lookups anymore, thanks Alexey. -DaveM | |
339 | * | |
340 | * Local BH must be disabled here. | |
341 | */ | |
342 | static inline struct sock * | |
343 | __inet_lookup_established(struct inet_hashinfo *hashinfo, | |
fb99c848 AV |
344 | const __be32 saddr, const __be16 sport, |
345 | const __be32 daddr, const u16 hnum, | |
e48c414e ACM |
346 | const int dif) |
347 | { | |
348 | INET_ADDR_COOKIE(acookie, saddr, daddr) | |
4f765d84 | 349 | const __portpair ports = INET_COMBINED_PORTS(sport, hnum); |
e48c414e ACM |
350 | struct sock *sk; |
351 | const struct hlist_node *node; | |
352 | /* Optimize here for direct hit, only listening connections can | |
353 | * have wildcards anyways. | |
354 | */ | |
81c3d547 ED |
355 | unsigned int hash = inet_ehashfn(daddr, hnum, saddr, sport); |
356 | struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash); | |
e48c414e | 357 | |
81c3d547 | 358 | prefetch(head->chain.first); |
e48c414e ACM |
359 | read_lock(&head->lock); |
360 | sk_for_each(sk, node, &head->chain) { | |
81c3d547 | 361 | if (INET_MATCH(sk, hash, acookie, saddr, daddr, ports, dif)) |
e48c414e ACM |
362 | goto hit; /* You sunk my battleship! */ |
363 | } | |
364 | ||
365 | /* Must check for a TIME_WAIT'er before going to listener hash. */ | |
dbca9b27 | 366 | sk_for_each(sk, node, &head->twchain) { |
81c3d547 | 367 | if (INET_TW_MATCH(sk, hash, acookie, saddr, daddr, ports, dif)) |
e48c414e ACM |
368 | goto hit; |
369 | } | |
370 | sk = NULL; | |
371 | out: | |
372 | read_unlock(&head->lock); | |
373 | return sk; | |
374 | hit: | |
375 | sock_hold(sk); | |
376 | goto out; | |
377 | } | |
378 | ||
8f491069 HX |
379 | static inline struct sock * |
380 | inet_lookup_established(struct inet_hashinfo *hashinfo, | |
fb99c848 AV |
381 | const __be32 saddr, const __be16 sport, |
382 | const __be32 daddr, const __be16 dport, | |
8f491069 HX |
383 | const int dif) |
384 | { | |
385 | return __inet_lookup_established(hashinfo, saddr, sport, daddr, | |
386 | ntohs(dport), dif); | |
387 | } | |
388 | ||
e48c414e | 389 | static inline struct sock *__inet_lookup(struct inet_hashinfo *hashinfo, |
fb99c848 AV |
390 | const __be32 saddr, const __be16 sport, |
391 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
392 | const int dif) |
393 | { | |
8f491069 | 394 | u16 hnum = ntohs(dport); |
e48c414e ACM |
395 | struct sock *sk = __inet_lookup_established(hashinfo, saddr, sport, daddr, |
396 | hnum, dif); | |
8f491069 | 397 | return sk ? : __inet_lookup_listener(hashinfo, daddr, hnum, dif); |
e48c414e ACM |
398 | } |
399 | ||
400 | static inline struct sock *inet_lookup(struct inet_hashinfo *hashinfo, | |
fb99c848 AV |
401 | const __be32 saddr, const __be16 sport, |
402 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
403 | const int dif) |
404 | { | |
405 | struct sock *sk; | |
406 | ||
407 | local_bh_disable(); | |
8f491069 | 408 | sk = __inet_lookup(hashinfo, saddr, sport, daddr, dport, dif); |
e48c414e ACM |
409 | local_bh_enable(); |
410 | ||
411 | return sk; | |
412 | } | |
a7f5e7f1 ACM |
413 | |
414 | extern int inet_hash_connect(struct inet_timewait_death_row *death_row, | |
415 | struct sock *sk); | |
304a1618 | 416 | #endif /* _INET_HASHTABLES_H */ |