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> |
9e4505c4 | 26 | #include <linux/vmalloc.h> |
304a1618 | 27 | |
463c84b9 | 28 | #include <net/inet_connection_sock.h> |
14c85021 | 29 | #include <net/inet_sock.h> |
2d8c4ce5 | 30 | #include <net/sock.h> |
c752f073 | 31 | #include <net/tcp_states.h> |
0b441916 | 32 | #include <net/netns/hash.h> |
2d8c4ce5 | 33 | |
f3f05f70 | 34 | #include <asm/atomic.h> |
e48c414e | 35 | #include <asm/byteorder.h> |
f3f05f70 | 36 | |
77d8bf9c | 37 | /* This is for all connections with a full identity, no wildcards. |
dbca9b27 ED |
38 | * One chain is dedicated to TIME_WAIT sockets. |
39 | * I'll experiment with dynamic table growth later. | |
77d8bf9c ACM |
40 | */ |
41 | struct inet_ehash_bucket { | |
77d8bf9c | 42 | struct hlist_head chain; |
dbca9b27 | 43 | struct hlist_head twchain; |
6d255361 | 44 | }; |
77d8bf9c ACM |
45 | |
46 | /* There are a few simple rules, which allow for local port reuse by | |
47 | * an application. In essence: | |
48 | * | |
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 | |
56 | * shared. | |
57 | * Failing this, the port cannot be shared. | |
58 | * | |
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. | |
69 | * | |
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 | |
76 | */ | |
77 | struct inet_bind_bucket { | |
941b1d22 | 78 | struct net *ib_net; |
77d8bf9c ACM |
79 | unsigned short port; |
80 | signed short fastreuse; | |
81 | struct hlist_node node; | |
82 | struct hlist_head owners; | |
83 | }; | |
84 | ||
85 | #define inet_bind_bucket_for_each(tb, node, head) \ | |
86 | hlist_for_each_entry(tb, node, head, node) | |
87 | ||
88 | struct inet_bind_hashbucket { | |
89 | spinlock_t lock; | |
90 | struct hlist_head chain; | |
91 | }; | |
92 | ||
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. */ | |
95 | ||
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: | |
99 | * | |
100 | * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE | |
101 | * | |
dbca9b27 | 102 | * TIME_WAIT sockets use a separate chain (twchain). |
77d8bf9c ACM |
103 | */ |
104 | struct inet_ehash_bucket *ehash; | |
230140cf ED |
105 | rwlock_t *ehash_locks; |
106 | unsigned int ehash_size; | |
107 | unsigned int ehash_locks_mask; | |
77d8bf9c ACM |
108 | |
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. | |
111 | */ | |
112 | struct inet_bind_hashbucket *bhash; | |
113 | ||
cfcabdcc | 114 | unsigned int bhash_size; |
230140cf | 115 | /* Note : 4 bytes padding on 64 bit arches */ |
77d8bf9c ACM |
116 | |
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. | |
120 | */ | |
121 | struct hlist_head listening_hash[INET_LHTABLE_SIZE]; | |
122 | ||
123 | /* All the above members are written once at bootup and | |
124 | * never written again _or_ are predominantly read-access. | |
125 | * | |
126 | * Now align to a new cache line as all the following members | |
127 | * are often dirty. | |
128 | */ | |
129 | rwlock_t lhash_lock ____cacheline_aligned; | |
130 | atomic_t lhash_users; | |
131 | wait_queue_head_t lhash_wait; | |
e18b890b | 132 | struct kmem_cache *bind_bucket_cachep; |
77d8bf9c ACM |
133 | }; |
134 | ||
81c3d547 ED |
135 | static inline struct inet_ehash_bucket *inet_ehash_bucket( |
136 | struct inet_hashinfo *hashinfo, | |
137 | unsigned int hash) | |
138 | { | |
139 | return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)]; | |
304a1618 ACM |
140 | } |
141 | ||
230140cf ED |
142 | static inline rwlock_t *inet_ehash_lockp( |
143 | struct inet_hashinfo *hashinfo, | |
144 | unsigned int hash) | |
145 | { | |
146 | return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask]; | |
147 | } | |
148 | ||
149 | static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo) | |
150 | { | |
151 | unsigned int i, size = 256; | |
152 | #if defined(CONFIG_PROVE_LOCKING) | |
153 | unsigned int nr_pcpus = 2; | |
154 | #else | |
155 | unsigned int nr_pcpus = num_possible_cpus(); | |
156 | #endif | |
157 | if (nr_pcpus >= 4) | |
158 | size = 512; | |
159 | if (nr_pcpus >= 8) | |
160 | size = 1024; | |
161 | if (nr_pcpus >= 16) | |
162 | size = 2048; | |
163 | if (nr_pcpus >= 32) | |
164 | size = 4096; | |
165 | if (sizeof(rwlock_t) != 0) { | |
166 | #ifdef CONFIG_NUMA | |
167 | if (size * sizeof(rwlock_t) > PAGE_SIZE) | |
168 | hashinfo->ehash_locks = vmalloc(size * sizeof(rwlock_t)); | |
169 | else | |
170 | #endif | |
171 | hashinfo->ehash_locks = kmalloc(size * sizeof(rwlock_t), | |
172 | GFP_KERNEL); | |
173 | if (!hashinfo->ehash_locks) | |
174 | return ENOMEM; | |
175 | for (i = 0; i < size; i++) | |
176 | rwlock_init(&hashinfo->ehash_locks[i]); | |
177 | } | |
178 | hashinfo->ehash_locks_mask = size - 1; | |
179 | return 0; | |
180 | } | |
181 | ||
182 | static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo) | |
183 | { | |
184 | if (hashinfo->ehash_locks) { | |
185 | #ifdef CONFIG_NUMA | |
186 | unsigned int size = (hashinfo->ehash_locks_mask + 1) * | |
187 | sizeof(rwlock_t); | |
188 | if (size > PAGE_SIZE) | |
189 | vfree(hashinfo->ehash_locks); | |
190 | else | |
230140cf | 191 | #endif |
218ad12f | 192 | kfree(hashinfo->ehash_locks); |
230140cf ED |
193 | hashinfo->ehash_locks = NULL; |
194 | } | |
195 | } | |
196 | ||
77d8bf9c | 197 | extern struct inet_bind_bucket * |
e18b890b | 198 | inet_bind_bucket_create(struct kmem_cache *cachep, |
941b1d22 | 199 | struct net *net, |
77d8bf9c ACM |
200 | struct inet_bind_hashbucket *head, |
201 | const unsigned short snum); | |
e18b890b | 202 | extern void inet_bind_bucket_destroy(struct kmem_cache *cachep, |
77d8bf9c ACM |
203 | struct inet_bind_bucket *tb); |
204 | ||
7f635ab7 PE |
205 | static inline int inet_bhashfn(struct net *net, |
206 | const __u16 lport, const int bhash_size) | |
77d8bf9c | 207 | { |
0b441916 | 208 | return (lport + net_hash_mix(net)) & (bhash_size - 1); |
77d8bf9c ACM |
209 | } |
210 | ||
2d8c4ce5 ACM |
211 | extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, |
212 | const unsigned short snum); | |
213 | ||
77d8bf9c | 214 | /* These can have wildcards, don't try too hard. */ |
2086a650 | 215 | static inline int inet_lhashfn(struct net *net, const unsigned short num) |
77d8bf9c | 216 | { |
0b441916 | 217 | return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1); |
77d8bf9c ACM |
218 | } |
219 | ||
220 | static inline int inet_sk_listen_hashfn(const struct sock *sk) | |
221 | { | |
2086a650 | 222 | return inet_lhashfn(sock_net(sk), inet_sk(sk)->num); |
77d8bf9c ACM |
223 | } |
224 | ||
2d8c4ce5 | 225 | /* Caller must disable local BH processing. */ |
53083773 | 226 | extern void __inet_inherit_port(struct sock *sk, struct sock *child); |
2d8c4ce5 | 227 | |
ab1e0a13 | 228 | extern void inet_put_port(struct sock *sk); |
2d8c4ce5 | 229 | |
f3f05f70 ACM |
230 | extern void inet_listen_wlock(struct inet_hashinfo *hashinfo); |
231 | ||
232 | /* | |
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). | |
236 | */ | |
237 | static inline void inet_listen_lock(struct inet_hashinfo *hashinfo) | |
238 | { | |
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); | |
243 | } | |
244 | ||
245 | static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo) | |
246 | { | |
247 | if (atomic_dec_and_test(&hashinfo->lhash_users)) | |
248 | wake_up(&hashinfo->lhash_wait); | |
249 | } | |
250 | ||
ab1e0a13 ACM |
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); | |
33b62231 | 254 | |
c67499c0 PE |
255 | extern struct sock *__inet_lookup_listener(struct net *net, |
256 | struct inet_hashinfo *hashinfo, | |
fb99c848 | 257 | const __be32 daddr, |
8f491069 HX |
258 | const unsigned short hnum, |
259 | const int dif); | |
260 | ||
c67499c0 PE |
261 | static inline struct sock *inet_lookup_listener(struct net *net, |
262 | struct inet_hashinfo *hashinfo, | |
263 | __be32 daddr, __be16 dport, int dif) | |
8f491069 | 264 | { |
c67499c0 | 265 | return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif); |
8f491069 | 266 | } |
8feaf0c0 ACM |
267 | |
268 | /* Socket demux engine toys. */ | |
4f765d84 AV |
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. | |
276 | */ | |
277 | typedef __u32 __bitwise __portpair; | |
8feaf0c0 ACM |
278 | #ifdef __BIG_ENDIAN |
279 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 280 | ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport))) |
8feaf0c0 ACM |
281 | #else /* __LITTLE_ENDIAN */ |
282 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 283 | ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport))) |
8feaf0c0 ACM |
284 | #endif |
285 | ||
286 | #if (BITS_PER_LONG == 64) | |
4f765d84 | 287 | typedef __u64 __bitwise __addrpair; |
8feaf0c0 ACM |
288 | #ifdef __BIG_ENDIAN |
289 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
290 | const __addrpair __name = (__force __addrpair) ( \ |
291 | (((__force __u64)(__be32)(__saddr)) << 32) | \ | |
292 | ((__force __u64)(__be32)(__daddr))); | |
8feaf0c0 ACM |
293 | #else /* __LITTLE_ENDIAN */ |
294 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
295 | const __addrpair __name = (__force __addrpair) ( \ |
296 | (((__force __u64)(__be32)(__daddr)) << 32) | \ | |
297 | ((__force __u64)(__be32)(__saddr))); | |
8feaf0c0 | 298 | #endif /* __BIG_ENDIAN */ |
c67499c0 | 299 | #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ |
3b1e0a65 | 300 | (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \ |
4f765d84 AV |
301 | ((*((__addrpair *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \ |
302 | ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \ | |
8feaf0c0 | 303 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
c67499c0 | 304 | #define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ |
3b1e0a65 | 305 | (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \ |
4f765d84 AV |
306 | ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \ |
307 | ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \ | |
8feaf0c0 ACM |
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) | |
c67499c0 | 311 | #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif) \ |
3b1e0a65 | 312 | (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \ |
81c3d547 | 313 | (inet_sk(__sk)->daddr == (__saddr)) && \ |
8feaf0c0 | 314 | (inet_sk(__sk)->rcv_saddr == (__daddr)) && \ |
4f765d84 | 315 | ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \ |
8feaf0c0 | 316 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
c67499c0 | 317 | #define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif) \ |
3b1e0a65 | 318 | (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \ |
81c3d547 | 319 | (inet_twsk(__sk)->tw_daddr == (__saddr)) && \ |
8feaf0c0 | 320 | (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \ |
4f765d84 | 321 | ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \ |
8feaf0c0 ACM |
322 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
323 | #endif /* 64-bit arch */ | |
e48c414e ACM |
324 | |
325 | /* | |
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 | |
328 | * | |
329 | * Local BH must be disabled here. | |
330 | */ | |
c67499c0 PE |
331 | extern struct sock * __inet_lookup_established(struct net *net, |
332 | struct inet_hashinfo *hashinfo, | |
77a5ba55 PE |
333 | const __be32 saddr, const __be16 sport, |
334 | const __be32 daddr, const u16 hnum, const int dif); | |
e48c414e | 335 | |
8f491069 | 336 | static inline struct sock * |
c67499c0 | 337 | inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo, |
fb99c848 AV |
338 | const __be32 saddr, const __be16 sport, |
339 | const __be32 daddr, const __be16 dport, | |
8f491069 HX |
340 | const int dif) |
341 | { | |
c67499c0 | 342 | return __inet_lookup_established(net, hashinfo, saddr, sport, daddr, |
8f491069 HX |
343 | ntohs(dport), dif); |
344 | } | |
345 | ||
c67499c0 PE |
346 | static inline struct sock *__inet_lookup(struct net *net, |
347 | struct inet_hashinfo *hashinfo, | |
fb99c848 AV |
348 | const __be32 saddr, const __be16 sport, |
349 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
350 | const int dif) |
351 | { | |
8f491069 | 352 | u16 hnum = ntohs(dport); |
c67499c0 PE |
353 | struct sock *sk = __inet_lookup_established(net, hashinfo, |
354 | saddr, sport, daddr, hnum, dif); | |
355 | ||
356 | return sk ? : __inet_lookup_listener(net, hashinfo, daddr, hnum, dif); | |
e48c414e ACM |
357 | } |
358 | ||
c67499c0 PE |
359 | static inline struct sock *inet_lookup(struct net *net, |
360 | struct inet_hashinfo *hashinfo, | |
fb99c848 AV |
361 | const __be32 saddr, const __be16 sport, |
362 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
363 | const int dif) |
364 | { | |
365 | struct sock *sk; | |
366 | ||
367 | local_bh_disable(); | |
c67499c0 | 368 | sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif); |
e48c414e ACM |
369 | local_bh_enable(); |
370 | ||
371 | return sk; | |
372 | } | |
a7f5e7f1 | 373 | |
5ee31fc1 | 374 | extern int __inet_hash_connect(struct inet_timewait_death_row *death_row, |
5d8c0aa9 | 375 | struct sock *sk, u32 port_offset, |
5ee31fc1 PE |
376 | int (*check_established)(struct inet_timewait_death_row *, |
377 | struct sock *, __u16, struct inet_timewait_sock **), | |
ab1e0a13 | 378 | void (*hash)(struct sock *sk)); |
a7f5e7f1 ACM |
379 | extern int inet_hash_connect(struct inet_timewait_death_row *death_row, |
380 | struct sock *sk); | |
304a1618 | 381 | #endif /* _INET_HASHTABLES_H */ |