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> |
9a1f27c4 | 19 | #include <linux/ip.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> |
9e4505c4 | 27 | #include <linux/vmalloc.h> |
304a1618 | 28 | |
463c84b9 | 29 | #include <net/inet_connection_sock.h> |
14c85021 | 30 | #include <net/inet_sock.h> |
2d8c4ce5 | 31 | #include <net/sock.h> |
9a1f27c4 | 32 | #include <net/route.h> |
c752f073 | 33 | #include <net/tcp_states.h> |
0b441916 | 34 | #include <net/netns/hash.h> |
2d8c4ce5 | 35 | |
f3f05f70 | 36 | #include <asm/atomic.h> |
e48c414e | 37 | #include <asm/byteorder.h> |
f3f05f70 | 38 | |
77d8bf9c | 39 | /* This is for all connections with a full identity, no wildcards. |
dbca9b27 ED |
40 | * One chain is dedicated to TIME_WAIT sockets. |
41 | * I'll experiment with dynamic table growth later. | |
77d8bf9c ACM |
42 | */ |
43 | struct inet_ehash_bucket { | |
3ab5aee7 ED |
44 | struct hlist_nulls_head chain; |
45 | struct hlist_nulls_head twchain; | |
6d255361 | 46 | }; |
77d8bf9c ACM |
47 | |
48 | /* There are a few simple rules, which allow for local port reuse by | |
49 | * an application. In essence: | |
50 | * | |
51 | * 1) Sockets bound to different interfaces may share a local port. | |
52 | * Failing that, goto test 2. | |
53 | * 2) If all sockets have sk->sk_reuse set, and none of them are in | |
54 | * TCP_LISTEN state, the port may be shared. | |
55 | * Failing that, goto test 3. | |
56 | * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local | |
57 | * address, and none of them are the same, the port may be | |
58 | * shared. | |
59 | * Failing this, the port cannot be shared. | |
60 | * | |
61 | * The interesting point, is test #2. This is what an FTP server does | |
62 | * all day. To optimize this case we use a specific flag bit defined | |
63 | * below. As we add sockets to a bind bucket list, we perform a | |
64 | * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN)) | |
65 | * As long as all sockets added to a bind bucket pass this test, | |
66 | * the flag bit will be set. | |
67 | * The resulting situation is that tcp_v[46]_verify_bind() can just check | |
68 | * for this flag bit, if it is set and the socket trying to bind has | |
69 | * sk->sk_reuse set, we don't even have to walk the owners list at all, | |
70 | * we return that it is ok to bind this socket to the requested local port. | |
71 | * | |
72 | * Sounds like a lot of work, but it is worth it. In a more naive | |
73 | * implementation (ie. current FreeBSD etc.) the entire list of ports | |
74 | * must be walked for each data port opened by an ftp server. Needless | |
75 | * to say, this does not scale at all. With a couple thousand FTP | |
76 | * users logged onto your box, isn't it nice to know that new data | |
77 | * ports are created in O(1) time? I thought so. ;-) -DaveM | |
78 | */ | |
79 | struct inet_bind_bucket { | |
7a9546ee | 80 | #ifdef CONFIG_NET_NS |
941b1d22 | 81 | struct net *ib_net; |
7a9546ee | 82 | #endif |
77d8bf9c ACM |
83 | unsigned short port; |
84 | signed short fastreuse; | |
85 | struct hlist_node node; | |
86 | struct hlist_head owners; | |
87 | }; | |
88 | ||
7a9546ee ED |
89 | static inline struct net *ib_net(struct inet_bind_bucket *ib) |
90 | { | |
91 | return read_pnet(&ib->ib_net); | |
92 | } | |
93 | ||
77d8bf9c ACM |
94 | #define inet_bind_bucket_for_each(tb, node, head) \ |
95 | hlist_for_each_entry(tb, node, head, node) | |
96 | ||
97 | struct inet_bind_hashbucket { | |
98 | spinlock_t lock; | |
99 | struct hlist_head chain; | |
100 | }; | |
101 | ||
5caea4ea ED |
102 | struct inet_listen_hashbucket { |
103 | spinlock_t lock; | |
104 | struct hlist_head head; | |
105 | }; | |
106 | ||
77d8bf9c ACM |
107 | /* This is for listening sockets, thus all sockets which possess wildcards. */ |
108 | #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */ | |
109 | ||
110 | struct inet_hashinfo { | |
111 | /* This is for sockets with full identity only. Sockets here will | |
112 | * always be without wildcards and will have the following invariant: | |
113 | * | |
114 | * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE | |
115 | * | |
dbca9b27 | 116 | * TIME_WAIT sockets use a separate chain (twchain). |
77d8bf9c ACM |
117 | */ |
118 | struct inet_ehash_bucket *ehash; | |
9db66bdc | 119 | spinlock_t *ehash_locks; |
230140cf ED |
120 | unsigned int ehash_size; |
121 | unsigned int ehash_locks_mask; | |
77d8bf9c ACM |
122 | |
123 | /* Ok, let's try this, I give up, we do need a local binding | |
124 | * TCP hash as well as the others for fast bind/connect. | |
125 | */ | |
126 | struct inet_bind_hashbucket *bhash; | |
127 | ||
cfcabdcc | 128 | unsigned int bhash_size; |
230140cf | 129 | /* Note : 4 bytes padding on 64 bit arches */ |
77d8bf9c | 130 | |
5caea4ea | 131 | struct kmem_cache *bind_bucket_cachep; |
77d8bf9c ACM |
132 | |
133 | /* All the above members are written once at bootup and | |
134 | * never written again _or_ are predominantly read-access. | |
135 | * | |
136 | * Now align to a new cache line as all the following members | |
5caea4ea ED |
137 | * might be often dirty. |
138 | */ | |
139 | /* All sockets in TCP_LISTEN state will be in here. This is the only | |
140 | * table where wildcard'd TCP sockets can exist. Hash function here | |
141 | * is just local port number. | |
77d8bf9c | 142 | */ |
5caea4ea ED |
143 | struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE] |
144 | ____cacheline_aligned_in_smp; | |
145 | ||
77d8bf9c ACM |
146 | }; |
147 | ||
81c3d547 ED |
148 | static inline struct inet_ehash_bucket *inet_ehash_bucket( |
149 | struct inet_hashinfo *hashinfo, | |
150 | unsigned int hash) | |
151 | { | |
152 | return &hashinfo->ehash[hash & (hashinfo->ehash_size - 1)]; | |
304a1618 ACM |
153 | } |
154 | ||
9db66bdc | 155 | static inline spinlock_t *inet_ehash_lockp( |
230140cf ED |
156 | struct inet_hashinfo *hashinfo, |
157 | unsigned int hash) | |
158 | { | |
159 | return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask]; | |
160 | } | |
161 | ||
162 | static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo) | |
163 | { | |
164 | unsigned int i, size = 256; | |
165 | #if defined(CONFIG_PROVE_LOCKING) | |
166 | unsigned int nr_pcpus = 2; | |
167 | #else | |
168 | unsigned int nr_pcpus = num_possible_cpus(); | |
169 | #endif | |
170 | if (nr_pcpus >= 4) | |
171 | size = 512; | |
172 | if (nr_pcpus >= 8) | |
173 | size = 1024; | |
174 | if (nr_pcpus >= 16) | |
175 | size = 2048; | |
176 | if (nr_pcpus >= 32) | |
177 | size = 4096; | |
178 | if (sizeof(rwlock_t) != 0) { | |
179 | #ifdef CONFIG_NUMA | |
9db66bdc ED |
180 | if (size * sizeof(spinlock_t) > PAGE_SIZE) |
181 | hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t)); | |
230140cf ED |
182 | else |
183 | #endif | |
9db66bdc | 184 | hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t), |
230140cf ED |
185 | GFP_KERNEL); |
186 | if (!hashinfo->ehash_locks) | |
187 | return ENOMEM; | |
188 | for (i = 0; i < size; i++) | |
9db66bdc | 189 | spin_lock_init(&hashinfo->ehash_locks[i]); |
230140cf ED |
190 | } |
191 | hashinfo->ehash_locks_mask = size - 1; | |
192 | return 0; | |
193 | } | |
194 | ||
195 | static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo) | |
196 | { | |
197 | if (hashinfo->ehash_locks) { | |
198 | #ifdef CONFIG_NUMA | |
199 | unsigned int size = (hashinfo->ehash_locks_mask + 1) * | |
9db66bdc | 200 | sizeof(spinlock_t); |
230140cf ED |
201 | if (size > PAGE_SIZE) |
202 | vfree(hashinfo->ehash_locks); | |
203 | else | |
230140cf | 204 | #endif |
218ad12f | 205 | kfree(hashinfo->ehash_locks); |
230140cf ED |
206 | hashinfo->ehash_locks = NULL; |
207 | } | |
208 | } | |
209 | ||
77d8bf9c | 210 | extern struct inet_bind_bucket * |
e18b890b | 211 | inet_bind_bucket_create(struct kmem_cache *cachep, |
941b1d22 | 212 | struct net *net, |
77d8bf9c ACM |
213 | struct inet_bind_hashbucket *head, |
214 | const unsigned short snum); | |
e18b890b | 215 | extern void inet_bind_bucket_destroy(struct kmem_cache *cachep, |
77d8bf9c ACM |
216 | struct inet_bind_bucket *tb); |
217 | ||
7f635ab7 PE |
218 | static inline int inet_bhashfn(struct net *net, |
219 | const __u16 lport, const int bhash_size) | |
77d8bf9c | 220 | { |
0b441916 | 221 | return (lport + net_hash_mix(net)) & (bhash_size - 1); |
77d8bf9c ACM |
222 | } |
223 | ||
2d8c4ce5 ACM |
224 | extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, |
225 | const unsigned short snum); | |
226 | ||
77d8bf9c | 227 | /* These can have wildcards, don't try too hard. */ |
2086a650 | 228 | static inline int inet_lhashfn(struct net *net, const unsigned short num) |
77d8bf9c | 229 | { |
0b441916 | 230 | return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1); |
77d8bf9c ACM |
231 | } |
232 | ||
233 | static inline int inet_sk_listen_hashfn(const struct sock *sk) | |
234 | { | |
2086a650 | 235 | return inet_lhashfn(sock_net(sk), inet_sk(sk)->num); |
77d8bf9c ACM |
236 | } |
237 | ||
2d8c4ce5 | 238 | /* Caller must disable local BH processing. */ |
53083773 | 239 | extern void __inet_inherit_port(struct sock *sk, struct sock *child); |
2d8c4ce5 | 240 | |
ab1e0a13 | 241 | extern void inet_put_port(struct sock *sk); |
2d8c4ce5 | 242 | |
5caea4ea | 243 | void inet_hashinfo_init(struct inet_hashinfo *h); |
f3f05f70 | 244 | |
ab1e0a13 ACM |
245 | extern void __inet_hash_nolisten(struct sock *sk); |
246 | extern void inet_hash(struct sock *sk); | |
247 | extern void inet_unhash(struct sock *sk); | |
33b62231 | 248 | |
c67499c0 PE |
249 | extern struct sock *__inet_lookup_listener(struct net *net, |
250 | struct inet_hashinfo *hashinfo, | |
fb99c848 | 251 | const __be32 daddr, |
8f491069 HX |
252 | const unsigned short hnum, |
253 | const int dif); | |
254 | ||
c67499c0 PE |
255 | static inline struct sock *inet_lookup_listener(struct net *net, |
256 | struct inet_hashinfo *hashinfo, | |
257 | __be32 daddr, __be16 dport, int dif) | |
8f491069 | 258 | { |
c67499c0 | 259 | return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif); |
8f491069 | 260 | } |
8feaf0c0 ACM |
261 | |
262 | /* Socket demux engine toys. */ | |
4f765d84 AV |
263 | /* What happens here is ugly; there's a pair of adjacent fields in |
264 | struct inet_sock; __be16 dport followed by __u16 num. We want to | |
265 | search by pair, so we combine the keys into a single 32bit value | |
266 | and compare with 32bit value read from &...->dport. Let's at least | |
267 | make sure that it's not mixed with anything else... | |
268 | On 64bit targets we combine comparisons with pair of adjacent __be32 | |
269 | fields in the same way. | |
270 | */ | |
271 | typedef __u32 __bitwise __portpair; | |
8feaf0c0 ACM |
272 | #ifdef __BIG_ENDIAN |
273 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 274 | ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport))) |
8feaf0c0 ACM |
275 | #else /* __LITTLE_ENDIAN */ |
276 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 277 | ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport))) |
8feaf0c0 ACM |
278 | #endif |
279 | ||
280 | #if (BITS_PER_LONG == 64) | |
4f765d84 | 281 | typedef __u64 __bitwise __addrpair; |
8feaf0c0 ACM |
282 | #ifdef __BIG_ENDIAN |
283 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
284 | const __addrpair __name = (__force __addrpair) ( \ |
285 | (((__force __u64)(__be32)(__saddr)) << 32) | \ | |
286 | ((__force __u64)(__be32)(__daddr))); | |
8feaf0c0 ACM |
287 | #else /* __LITTLE_ENDIAN */ |
288 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
289 | const __addrpair __name = (__force __addrpair) ( \ |
290 | (((__force __u64)(__be32)(__daddr)) << 32) | \ | |
291 | ((__force __u64)(__be32)(__saddr))); | |
8feaf0c0 | 292 | #endif /* __BIG_ENDIAN */ |
c67499c0 | 293 | #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ |
3b1e0a65 | 294 | (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \ |
4f765d84 AV |
295 | ((*((__addrpair *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \ |
296 | ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \ | |
8feaf0c0 | 297 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
c67499c0 | 298 | #define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\ |
3b1e0a65 | 299 | (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \ |
4f765d84 AV |
300 | ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \ |
301 | ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \ | |
8feaf0c0 ACM |
302 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
303 | #else /* 32-bit arch */ | |
304 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) | |
c67499c0 | 305 | #define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif) \ |
3b1e0a65 | 306 | (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \ |
81c3d547 | 307 | (inet_sk(__sk)->daddr == (__saddr)) && \ |
8feaf0c0 | 308 | (inet_sk(__sk)->rcv_saddr == (__daddr)) && \ |
4f765d84 | 309 | ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \ |
8feaf0c0 | 310 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
c67499c0 | 311 | #define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif) \ |
3b1e0a65 | 312 | (((__sk)->sk_hash == (__hash)) && sock_net((__sk)) == (__net) && \ |
81c3d547 | 313 | (inet_twsk(__sk)->tw_daddr == (__saddr)) && \ |
8feaf0c0 | 314 | (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \ |
4f765d84 | 315 | ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \ |
8feaf0c0 ACM |
316 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
317 | #endif /* 64-bit arch */ | |
e48c414e ACM |
318 | |
319 | /* | |
320 | * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need | |
321 | * not check it for lookups anymore, thanks Alexey. -DaveM | |
322 | * | |
323 | * Local BH must be disabled here. | |
324 | */ | |
c67499c0 PE |
325 | extern struct sock * __inet_lookup_established(struct net *net, |
326 | struct inet_hashinfo *hashinfo, | |
77a5ba55 PE |
327 | const __be32 saddr, const __be16 sport, |
328 | const __be32 daddr, const u16 hnum, const int dif); | |
e48c414e | 329 | |
8f491069 | 330 | static inline struct sock * |
c67499c0 | 331 | inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo, |
fb99c848 AV |
332 | const __be32 saddr, const __be16 sport, |
333 | const __be32 daddr, const __be16 dport, | |
8f491069 HX |
334 | const int dif) |
335 | { | |
c67499c0 | 336 | return __inet_lookup_established(net, hashinfo, saddr, sport, daddr, |
8f491069 HX |
337 | ntohs(dport), dif); |
338 | } | |
339 | ||
c67499c0 PE |
340 | static inline struct sock *__inet_lookup(struct net *net, |
341 | struct inet_hashinfo *hashinfo, | |
fb99c848 AV |
342 | const __be32 saddr, const __be16 sport, |
343 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
344 | const int dif) |
345 | { | |
8f491069 | 346 | u16 hnum = ntohs(dport); |
c67499c0 PE |
347 | struct sock *sk = __inet_lookup_established(net, hashinfo, |
348 | saddr, sport, daddr, hnum, dif); | |
349 | ||
350 | return sk ? : __inet_lookup_listener(net, hashinfo, daddr, hnum, dif); | |
e48c414e ACM |
351 | } |
352 | ||
c67499c0 PE |
353 | static inline struct sock *inet_lookup(struct net *net, |
354 | struct inet_hashinfo *hashinfo, | |
fb99c848 AV |
355 | const __be32 saddr, const __be16 sport, |
356 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
357 | const int dif) |
358 | { | |
359 | struct sock *sk; | |
360 | ||
361 | local_bh_disable(); | |
c67499c0 | 362 | sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif); |
e48c414e ACM |
363 | local_bh_enable(); |
364 | ||
365 | return sk; | |
366 | } | |
a7f5e7f1 | 367 | |
9a1f27c4 ACM |
368 | static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo, |
369 | struct sk_buff *skb, | |
370 | const __be16 sport, | |
371 | const __be16 dport) | |
372 | { | |
23542618 | 373 | struct sock *sk; |
9a1f27c4 ACM |
374 | const struct iphdr *iph = ip_hdr(skb); |
375 | ||
23542618 KK |
376 | if (unlikely(sk = skb_steal_sock(skb))) |
377 | return sk; | |
378 | else | |
379 | return __inet_lookup(dev_net(skb->dst->dev), hashinfo, | |
380 | iph->saddr, sport, | |
381 | iph->daddr, dport, inet_iif(skb)); | |
9a1f27c4 ACM |
382 | } |
383 | ||
5ee31fc1 | 384 | extern int __inet_hash_connect(struct inet_timewait_death_row *death_row, |
5d8c0aa9 | 385 | struct sock *sk, u32 port_offset, |
5ee31fc1 PE |
386 | int (*check_established)(struct inet_timewait_death_row *, |
387 | struct sock *, __u16, struct inet_timewait_sock **), | |
ab1e0a13 | 388 | void (*hash)(struct sock *sk)); |
a7f5e7f1 ACM |
389 | extern int inet_hash_connect(struct inet_timewait_death_row *death_row, |
390 | struct sock *sk); | |
304a1618 | 391 | #endif /* _INET_HASHTABLES_H */ |