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 ACM |
17 | #include <linux/config.h> |
18 | ||
2d8c4ce5 | 19 | #include <linux/interrupt.h> |
33b62231 | 20 | #include <linux/ipv6.h> |
77d8bf9c ACM |
21 | #include <linux/list.h> |
22 | #include <linux/slab.h> | |
33b62231 | 23 | #include <linux/socket.h> |
77d8bf9c | 24 | #include <linux/spinlock.h> |
304a1618 | 25 | #include <linux/types.h> |
f3f05f70 | 26 | #include <linux/wait.h> |
304a1618 | 27 | |
463c84b9 | 28 | #include <net/inet_connection_sock.h> |
0a5578cf | 29 | #include <net/route.h> |
2d8c4ce5 | 30 | #include <net/sock.h> |
c752f073 | 31 | #include <net/tcp_states.h> |
2d8c4ce5 | 32 | |
f3f05f70 | 33 | #include <asm/atomic.h> |
e48c414e | 34 | #include <asm/byteorder.h> |
f3f05f70 | 35 | |
77d8bf9c ACM |
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 | */ | |
40 | struct inet_ehash_bucket { | |
41 | rwlock_t lock; | |
42 | struct hlist_head chain; | |
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 | * | |
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; | |
81c3d547 | 111 | unsigned int ehash_size; |
77d8bf9c ACM |
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; | |
2d8c4ce5 | 128 | kmem_cache_t *bind_bucket_cachep; |
77d8bf9c ACM |
129 | }; |
130 | ||
81c3d547 ED |
131 | static inline unsigned int inet_ehashfn(const __u32 laddr, const __u16 lport, |
132 | const __u32 faddr, const __u16 fport) | |
304a1618 | 133 | { |
81c3d547 | 134 | unsigned int h = (laddr ^ lport) ^ (faddr ^ fport); |
304a1618 ACM |
135 | h ^= h >> 16; |
136 | h ^= h >> 8; | |
81c3d547 | 137 | return h; |
304a1618 ACM |
138 | } |
139 | ||
81c3d547 | 140 | static inline int inet_sk_ehashfn(const struct sock *sk) |
304a1618 ACM |
141 | { |
142 | const struct inet_sock *inet = inet_sk(sk); | |
143 | const __u32 laddr = inet->rcv_saddr; | |
144 | const __u16 lport = inet->num; | |
145 | const __u32 faddr = inet->daddr; | |
146 | const __u16 fport = inet->dport; | |
147 | ||
81c3d547 ED |
148 | return inet_ehashfn(laddr, lport, faddr, fport); |
149 | } | |
150 | ||
151 | static inline struct inet_ehash_bucket *inet_ehash_bucket( | |
152 | struct inet_hashinfo *hashinfo, | |
153 | unsigned int hash) | |
154 | { | |
155 | return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)]; | |
304a1618 ACM |
156 | } |
157 | ||
77d8bf9c ACM |
158 | extern struct inet_bind_bucket * |
159 | inet_bind_bucket_create(kmem_cache_t *cachep, | |
160 | struct inet_bind_hashbucket *head, | |
161 | const unsigned short snum); | |
162 | extern void inet_bind_bucket_destroy(kmem_cache_t *cachep, | |
163 | struct inet_bind_bucket *tb); | |
164 | ||
165 | static inline int inet_bhashfn(const __u16 lport, const int bhash_size) | |
166 | { | |
167 | return lport & (bhash_size - 1); | |
168 | } | |
169 | ||
2d8c4ce5 ACM |
170 | extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, |
171 | const unsigned short snum); | |
172 | ||
77d8bf9c ACM |
173 | /* These can have wildcards, don't try too hard. */ |
174 | static inline int inet_lhashfn(const unsigned short num) | |
175 | { | |
176 | return num & (INET_LHTABLE_SIZE - 1); | |
177 | } | |
178 | ||
179 | static inline int inet_sk_listen_hashfn(const struct sock *sk) | |
180 | { | |
181 | return inet_lhashfn(inet_sk(sk)->num); | |
182 | } | |
183 | ||
2d8c4ce5 ACM |
184 | /* Caller must disable local BH processing. */ |
185 | static inline void __inet_inherit_port(struct inet_hashinfo *table, | |
186 | struct sock *sk, struct sock *child) | |
187 | { | |
188 | const int bhash = inet_bhashfn(inet_sk(child)->num, table->bhash_size); | |
189 | struct inet_bind_hashbucket *head = &table->bhash[bhash]; | |
190 | struct inet_bind_bucket *tb; | |
191 | ||
192 | spin_lock(&head->lock); | |
463c84b9 | 193 | tb = inet_csk(sk)->icsk_bind_hash; |
2d8c4ce5 | 194 | sk_add_bind_node(child, &tb->owners); |
463c84b9 | 195 | inet_csk(child)->icsk_bind_hash = tb; |
2d8c4ce5 ACM |
196 | spin_unlock(&head->lock); |
197 | } | |
198 | ||
199 | static inline void inet_inherit_port(struct inet_hashinfo *table, | |
200 | struct sock *sk, struct sock *child) | |
201 | { | |
202 | local_bh_disable(); | |
203 | __inet_inherit_port(table, sk, child); | |
204 | local_bh_enable(); | |
205 | } | |
206 | ||
207 | extern void inet_put_port(struct inet_hashinfo *table, struct sock *sk); | |
208 | ||
f3f05f70 ACM |
209 | extern void inet_listen_wlock(struct inet_hashinfo *hashinfo); |
210 | ||
211 | /* | |
212 | * - We may sleep inside this lock. | |
213 | * - If sleeping is not required (or called from BH), | |
214 | * use plain read_(un)lock(&inet_hashinfo.lhash_lock). | |
215 | */ | |
216 | static inline void inet_listen_lock(struct inet_hashinfo *hashinfo) | |
217 | { | |
218 | /* read_lock synchronizes to candidates to writers */ | |
219 | read_lock(&hashinfo->lhash_lock); | |
220 | atomic_inc(&hashinfo->lhash_users); | |
221 | read_unlock(&hashinfo->lhash_lock); | |
222 | } | |
223 | ||
224 | static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo) | |
225 | { | |
226 | if (atomic_dec_and_test(&hashinfo->lhash_users)) | |
227 | wake_up(&hashinfo->lhash_wait); | |
228 | } | |
229 | ||
230 | static inline void __inet_hash(struct inet_hashinfo *hashinfo, | |
231 | struct sock *sk, const int listen_possible) | |
232 | { | |
233 | struct hlist_head *list; | |
234 | rwlock_t *lock; | |
235 | ||
236 | BUG_TRAP(sk_unhashed(sk)); | |
237 | if (listen_possible && sk->sk_state == TCP_LISTEN) { | |
238 | list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)]; | |
239 | lock = &hashinfo->lhash_lock; | |
240 | inet_listen_wlock(hashinfo); | |
241 | } else { | |
81c3d547 ED |
242 | struct inet_ehash_bucket *head; |
243 | sk->sk_hash = inet_sk_ehashfn(sk); | |
244 | head = inet_ehash_bucket(hashinfo, sk->sk_hash); | |
245 | list = &head->chain; | |
246 | lock = &head->lock; | |
f3f05f70 ACM |
247 | write_lock(lock); |
248 | } | |
249 | __sk_add_node(sk, list); | |
250 | sock_prot_inc_use(sk->sk_prot); | |
251 | write_unlock(lock); | |
252 | if (listen_possible && sk->sk_state == TCP_LISTEN) | |
253 | wake_up(&hashinfo->lhash_wait); | |
254 | } | |
81849d10 ACM |
255 | |
256 | static inline void inet_hash(struct inet_hashinfo *hashinfo, struct sock *sk) | |
257 | { | |
258 | if (sk->sk_state != TCP_CLOSE) { | |
259 | local_bh_disable(); | |
260 | __inet_hash(hashinfo, sk, 1); | |
261 | local_bh_enable(); | |
262 | } | |
263 | } | |
264 | ||
265 | static inline void inet_unhash(struct inet_hashinfo *hashinfo, struct sock *sk) | |
266 | { | |
267 | rwlock_t *lock; | |
268 | ||
269 | if (sk_unhashed(sk)) | |
270 | goto out; | |
271 | ||
272 | if (sk->sk_state == TCP_LISTEN) { | |
273 | local_bh_disable(); | |
274 | inet_listen_wlock(hashinfo); | |
275 | lock = &hashinfo->lhash_lock; | |
276 | } else { | |
81c3d547 ED |
277 | lock = &inet_ehash_bucket(hashinfo, sk->sk_hash)->lock; |
278 | write_lock_bh(lock); | |
81849d10 ACM |
279 | } |
280 | ||
281 | if (__sk_del_node_init(sk)) | |
282 | sock_prot_dec_use(sk->sk_prot); | |
283 | write_unlock_bh(lock); | |
284 | out: | |
285 | if (sk->sk_state == TCP_LISTEN) | |
286 | wake_up(&hashinfo->lhash_wait); | |
287 | } | |
33b62231 | 288 | |
0a5578cf ACM |
289 | static inline int inet_iif(const struct sk_buff *skb) |
290 | { | |
291 | return ((struct rtable *)skb->dst)->rt_iif; | |
292 | } | |
293 | ||
33b62231 ACM |
294 | extern struct sock *__inet_lookup_listener(const struct hlist_head *head, |
295 | const u32 daddr, | |
296 | const unsigned short hnum, | |
297 | const int dif); | |
298 | ||
299 | /* Optimize the common listener case. */ | |
e48c414e ACM |
300 | static inline struct sock * |
301 | inet_lookup_listener(struct inet_hashinfo *hashinfo, | |
302 | const u32 daddr, | |
303 | const unsigned short hnum, const int dif) | |
33b62231 ACM |
304 | { |
305 | struct sock *sk = NULL; | |
e48c414e | 306 | const struct hlist_head *head; |
33b62231 ACM |
307 | |
308 | read_lock(&hashinfo->lhash_lock); | |
309 | head = &hashinfo->listening_hash[inet_lhashfn(hnum)]; | |
310 | if (!hlist_empty(head)) { | |
311 | const struct inet_sock *inet = inet_sk((sk = __sk_head(head))); | |
312 | ||
313 | if (inet->num == hnum && !sk->sk_node.next && | |
314 | (!inet->rcv_saddr || inet->rcv_saddr == daddr) && | |
315 | (sk->sk_family == PF_INET || !ipv6_only_sock(sk)) && | |
316 | !sk->sk_bound_dev_if) | |
317 | goto sherry_cache; | |
318 | sk = __inet_lookup_listener(head, daddr, hnum, dif); | |
319 | } | |
320 | if (sk) { | |
321 | sherry_cache: | |
322 | sock_hold(sk); | |
323 | } | |
324 | read_unlock(&hashinfo->lhash_lock); | |
325 | return sk; | |
326 | } | |
8feaf0c0 ACM |
327 | |
328 | /* Socket demux engine toys. */ | |
329 | #ifdef __BIG_ENDIAN | |
330 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
331 | (((__u32)(__sport) << 16) | (__u32)(__dport)) | |
332 | #else /* __LITTLE_ENDIAN */ | |
333 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
334 | (((__u32)(__dport) << 16) | (__u32)(__sport)) | |
335 | #endif | |
336 | ||
337 | #if (BITS_PER_LONG == 64) | |
338 | #ifdef __BIG_ENDIAN | |
339 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
340 | const __u64 __name = (((__u64)(__saddr)) << 32) | ((__u64)(__daddr)); | |
341 | #else /* __LITTLE_ENDIAN */ | |
342 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
343 | const __u64 __name = (((__u64)(__daddr)) << 32) | ((__u64)(__saddr)); | |
344 | #endif /* __BIG_ENDIAN */ | |
81c3d547 ED |
345 | #define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ |
346 | (((__sk)->sk_hash == (__hash)) && \ | |
347 | ((*((__u64 *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \ | |
8feaf0c0 ACM |
348 | ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \ |
349 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) | |
81c3d547 ED |
350 | #define INET_TW_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ |
351 | (((__sk)->sk_hash == (__hash)) && \ | |
352 | ((*((__u64 *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \ | |
8feaf0c0 ACM |
353 | ((*((__u32 *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \ |
354 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) | |
355 | #else /* 32-bit arch */ | |
356 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) | |
81c3d547 ED |
357 | #define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif) \ |
358 | (((__sk)->sk_hash == (__hash)) && \ | |
359 | (inet_sk(__sk)->daddr == (__saddr)) && \ | |
8feaf0c0 ACM |
360 | (inet_sk(__sk)->rcv_saddr == (__daddr)) && \ |
361 | ((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \ | |
362 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) | |
81c3d547 ED |
363 | #define INET_TW_MATCH(__sk, __hash,__cookie, __saddr, __daddr, __ports, __dif) \ |
364 | (((__sk)->sk_hash == (__hash)) && \ | |
365 | (inet_twsk(__sk)->tw_daddr == (__saddr)) && \ | |
8feaf0c0 ACM |
366 | (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \ |
367 | ((*((__u32 *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \ | |
368 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) | |
369 | #endif /* 64-bit arch */ | |
e48c414e ACM |
370 | |
371 | /* | |
372 | * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need | |
373 | * not check it for lookups anymore, thanks Alexey. -DaveM | |
374 | * | |
375 | * Local BH must be disabled here. | |
376 | */ | |
377 | static inline struct sock * | |
378 | __inet_lookup_established(struct inet_hashinfo *hashinfo, | |
379 | const u32 saddr, const u16 sport, | |
380 | const u32 daddr, const u16 hnum, | |
381 | const int dif) | |
382 | { | |
383 | INET_ADDR_COOKIE(acookie, saddr, daddr) | |
384 | const __u32 ports = INET_COMBINED_PORTS(sport, hnum); | |
385 | struct sock *sk; | |
386 | const struct hlist_node *node; | |
387 | /* Optimize here for direct hit, only listening connections can | |
388 | * have wildcards anyways. | |
389 | */ | |
81c3d547 ED |
390 | unsigned int hash = inet_ehashfn(daddr, hnum, saddr, sport); |
391 | struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash); | |
e48c414e | 392 | |
81c3d547 | 393 | prefetch(head->chain.first); |
e48c414e ACM |
394 | read_lock(&head->lock); |
395 | sk_for_each(sk, node, &head->chain) { | |
81c3d547 | 396 | if (INET_MATCH(sk, hash, acookie, saddr, daddr, ports, dif)) |
e48c414e ACM |
397 | goto hit; /* You sunk my battleship! */ |
398 | } | |
399 | ||
400 | /* Must check for a TIME_WAIT'er before going to listener hash. */ | |
401 | sk_for_each(sk, node, &(head + hashinfo->ehash_size)->chain) { | |
81c3d547 | 402 | if (INET_TW_MATCH(sk, hash, acookie, saddr, daddr, ports, dif)) |
e48c414e ACM |
403 | goto hit; |
404 | } | |
405 | sk = NULL; | |
406 | out: | |
407 | read_unlock(&head->lock); | |
408 | return sk; | |
409 | hit: | |
410 | sock_hold(sk); | |
411 | goto out; | |
412 | } | |
413 | ||
414 | static inline struct sock *__inet_lookup(struct inet_hashinfo *hashinfo, | |
415 | const u32 saddr, const u16 sport, | |
416 | const u32 daddr, const u16 hnum, | |
417 | const int dif) | |
418 | { | |
419 | struct sock *sk = __inet_lookup_established(hashinfo, saddr, sport, daddr, | |
420 | hnum, dif); | |
421 | return sk ? : inet_lookup_listener(hashinfo, daddr, hnum, dif); | |
422 | } | |
423 | ||
424 | static inline struct sock *inet_lookup(struct inet_hashinfo *hashinfo, | |
425 | const u32 saddr, const u16 sport, | |
426 | const u32 daddr, const u16 dport, | |
427 | const int dif) | |
428 | { | |
429 | struct sock *sk; | |
430 | ||
431 | local_bh_disable(); | |
432 | sk = __inet_lookup(hashinfo, saddr, sport, daddr, ntohs(dport), dif); | |
433 | local_bh_enable(); | |
434 | ||
435 | return sk; | |
436 | } | |
304a1618 | 437 | #endif /* _INET_HASHTABLES_H */ |