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 | |
60063497 | 36 | #include <linux/atomic.h> |
e48c414e | 37 | #include <asm/byteorder.h> |
f3f05f70 | 38 | |
77d8bf9c | 39 | /* This is for all connections with a full identity, no wildcards. |
05dbc7b5 ED |
40 | * The 'e' prefix stands for Establish, but we really put all sockets |
41 | * but LISTEN ones. | |
77d8bf9c ACM |
42 | */ |
43 | struct inet_ehash_bucket { | |
3ab5aee7 | 44 | struct hlist_nulls_head chain; |
6d255361 | 45 | }; |
77d8bf9c ACM |
46 | |
47 | /* There are a few simple rules, which allow for local port reuse by | |
48 | * an application. In essence: | |
49 | * | |
50 | * 1) Sockets bound to different interfaces may share a local port. | |
51 | * Failing that, goto test 2. | |
52 | * 2) If all sockets have sk->sk_reuse set, and none of them are in | |
53 | * TCP_LISTEN state, the port may be shared. | |
54 | * Failing that, goto test 3. | |
55 | * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local | |
56 | * address, and none of them are the same, the port may be | |
57 | * shared. | |
58 | * Failing this, the port cannot be shared. | |
59 | * | |
60 | * The interesting point, is test #2. This is what an FTP server does | |
61 | * all day. To optimize this case we use a specific flag bit defined | |
62 | * below. As we add sockets to a bind bucket list, we perform a | |
63 | * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN)) | |
64 | * As long as all sockets added to a bind bucket pass this test, | |
65 | * the flag bit will be set. | |
66 | * The resulting situation is that tcp_v[46]_verify_bind() can just check | |
67 | * for this flag bit, if it is set and the socket trying to bind has | |
68 | * sk->sk_reuse set, we don't even have to walk the owners list at all, | |
69 | * we return that it is ok to bind this socket to the requested local port. | |
70 | * | |
71 | * Sounds like a lot of work, but it is worth it. In a more naive | |
72 | * implementation (ie. current FreeBSD etc.) the entire list of ports | |
73 | * must be walked for each data port opened by an ftp server. Needless | |
74 | * to say, this does not scale at all. With a couple thousand FTP | |
75 | * users logged onto your box, isn't it nice to know that new data | |
76 | * ports are created in O(1) time? I thought so. ;-) -DaveM | |
77 | */ | |
78 | struct inet_bind_bucket { | |
0c5c9fb5 | 79 | possible_net_t ib_net; |
77d8bf9c | 80 | unsigned short port; |
da5e3630 TH |
81 | signed char fastreuse; |
82 | signed char fastreuseport; | |
83 | kuid_t fastuid; | |
a9d8f911 | 84 | int num_owners; |
77d8bf9c ACM |
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 | ||
b67bfe0d SL |
94 | #define inet_bind_bucket_for_each(tb, head) \ |
95 | hlist_for_each_entry(tb, head, node) | |
77d8bf9c ACM |
96 | |
97 | struct inet_bind_hashbucket { | |
98 | spinlock_t lock; | |
99 | struct hlist_head chain; | |
100 | }; | |
101 | ||
c25eb3bf ED |
102 | /* |
103 | * Sockets can be hashed in established or listening table | |
104 | * We must use different 'nulls' end-of-chain value for listening | |
105 | * hash table, or we might find a socket that was closed and | |
106 | * reallocated/inserted into established hash table | |
107 | */ | |
108 | #define LISTENING_NULLS_BASE (1U << 29) | |
5caea4ea ED |
109 | struct inet_listen_hashbucket { |
110 | spinlock_t lock; | |
c25eb3bf | 111 | struct hlist_nulls_head head; |
5caea4ea ED |
112 | }; |
113 | ||
77d8bf9c ACM |
114 | /* This is for listening sockets, thus all sockets which possess wildcards. */ |
115 | #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */ | |
116 | ||
117 | struct inet_hashinfo { | |
118 | /* This is for sockets with full identity only. Sockets here will | |
119 | * always be without wildcards and will have the following invariant: | |
120 | * | |
121 | * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE | |
122 | * | |
77d8bf9c ACM |
123 | */ |
124 | struct inet_ehash_bucket *ehash; | |
9db66bdc | 125 | spinlock_t *ehash_locks; |
f373b53b | 126 | unsigned int ehash_mask; |
230140cf | 127 | unsigned int ehash_locks_mask; |
77d8bf9c ACM |
128 | |
129 | /* Ok, let's try this, I give up, we do need a local binding | |
130 | * TCP hash as well as the others for fast bind/connect. | |
131 | */ | |
132 | struct inet_bind_hashbucket *bhash; | |
133 | ||
cfcabdcc | 134 | unsigned int bhash_size; |
24dd1fa1 | 135 | /* 4 bytes hole on 64 bit */ |
77d8bf9c | 136 | |
5caea4ea | 137 | struct kmem_cache *bind_bucket_cachep; |
77d8bf9c ACM |
138 | |
139 | /* All the above members are written once at bootup and | |
140 | * never written again _or_ are predominantly read-access. | |
141 | * | |
142 | * Now align to a new cache line as all the following members | |
5caea4ea ED |
143 | * might be often dirty. |
144 | */ | |
145 | /* All sockets in TCP_LISTEN state will be in here. This is the only | |
146 | * table where wildcard'd TCP sockets can exist. Hash function here | |
147 | * is just local port number. | |
77d8bf9c | 148 | */ |
5caea4ea ED |
149 | struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE] |
150 | ____cacheline_aligned_in_smp; | |
151 | ||
24dd1fa1 | 152 | atomic_t bsockets; |
77d8bf9c ACM |
153 | }; |
154 | ||
81c3d547 ED |
155 | static inline struct inet_ehash_bucket *inet_ehash_bucket( |
156 | struct inet_hashinfo *hashinfo, | |
157 | unsigned int hash) | |
158 | { | |
f373b53b | 159 | return &hashinfo->ehash[hash & hashinfo->ehash_mask]; |
304a1618 ACM |
160 | } |
161 | ||
9db66bdc | 162 | static inline spinlock_t *inet_ehash_lockp( |
230140cf ED |
163 | struct inet_hashinfo *hashinfo, |
164 | unsigned int hash) | |
165 | { | |
166 | return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask]; | |
167 | } | |
168 | ||
169 | static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo) | |
170 | { | |
171 | unsigned int i, size = 256; | |
172 | #if defined(CONFIG_PROVE_LOCKING) | |
173 | unsigned int nr_pcpus = 2; | |
174 | #else | |
175 | unsigned int nr_pcpus = num_possible_cpus(); | |
176 | #endif | |
177 | if (nr_pcpus >= 4) | |
178 | size = 512; | |
179 | if (nr_pcpus >= 8) | |
180 | size = 1024; | |
181 | if (nr_pcpus >= 16) | |
182 | size = 2048; | |
183 | if (nr_pcpus >= 32) | |
184 | size = 4096; | |
94cd3e6c | 185 | if (sizeof(spinlock_t) != 0) { |
230140cf | 186 | #ifdef CONFIG_NUMA |
9db66bdc ED |
187 | if (size * sizeof(spinlock_t) > PAGE_SIZE) |
188 | hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t)); | |
230140cf ED |
189 | else |
190 | #endif | |
9db66bdc | 191 | hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t), |
230140cf ED |
192 | GFP_KERNEL); |
193 | if (!hashinfo->ehash_locks) | |
194 | return ENOMEM; | |
195 | for (i = 0; i < size; i++) | |
9db66bdc | 196 | spin_lock_init(&hashinfo->ehash_locks[i]); |
230140cf ED |
197 | } |
198 | hashinfo->ehash_locks_mask = size - 1; | |
199 | return 0; | |
200 | } | |
201 | ||
202 | static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo) | |
203 | { | |
204 | if (hashinfo->ehash_locks) { | |
205 | #ifdef CONFIG_NUMA | |
206 | unsigned int size = (hashinfo->ehash_locks_mask + 1) * | |
9db66bdc | 207 | sizeof(spinlock_t); |
230140cf ED |
208 | if (size > PAGE_SIZE) |
209 | vfree(hashinfo->ehash_locks); | |
210 | else | |
230140cf | 211 | #endif |
218ad12f | 212 | kfree(hashinfo->ehash_locks); |
230140cf ED |
213 | hashinfo->ehash_locks = NULL; |
214 | } | |
215 | } | |
216 | ||
1fd51155 JP |
217 | struct inet_bind_bucket * |
218 | inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net, | |
219 | struct inet_bind_hashbucket *head, | |
220 | const unsigned short snum); | |
221 | void inet_bind_bucket_destroy(struct kmem_cache *cachep, | |
222 | struct inet_bind_bucket *tb); | |
223 | ||
6eada011 ED |
224 | static inline u32 inet_bhashfn(const struct net *net, const __u16 lport, |
225 | const u32 bhash_size) | |
77d8bf9c | 226 | { |
0b441916 | 227 | return (lport + net_hash_mix(net)) & (bhash_size - 1); |
77d8bf9c ACM |
228 | } |
229 | ||
1fd51155 JP |
230 | void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, |
231 | const unsigned short snum); | |
2d8c4ce5 | 232 | |
77d8bf9c | 233 | /* These can have wildcards, don't try too hard. */ |
6eada011 | 234 | static inline u32 inet_lhashfn(const struct net *net, const unsigned short num) |
77d8bf9c | 235 | { |
0b441916 | 236 | return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1); |
77d8bf9c ACM |
237 | } |
238 | ||
239 | static inline int inet_sk_listen_hashfn(const struct sock *sk) | |
240 | { | |
c720c7e8 | 241 | return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num); |
77d8bf9c ACM |
242 | } |
243 | ||
2d8c4ce5 | 244 | /* Caller must disable local BH processing. */ |
1fd51155 | 245 | int __inet_inherit_port(struct sock *sk, struct sock *child); |
2d8c4ce5 | 246 | |
1fd51155 | 247 | void inet_put_port(struct sock *sk); |
2d8c4ce5 | 248 | |
5caea4ea | 249 | void inet_hashinfo_init(struct inet_hashinfo *h); |
f3f05f70 | 250 | |
1fd51155 | 251 | int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw); |
77a6a471 | 252 | int __inet_hash(struct sock *sk, struct inet_timewait_sock *tw); |
1fd51155 JP |
253 | void inet_hash(struct sock *sk); |
254 | void inet_unhash(struct sock *sk); | |
33b62231 | 255 | |
1fd51155 JP |
256 | struct sock *__inet_lookup_listener(struct net *net, |
257 | struct inet_hashinfo *hashinfo, | |
258 | const __be32 saddr, const __be16 sport, | |
259 | const __be32 daddr, | |
260 | const unsigned short hnum, | |
261 | const int dif); | |
8f491069 | 262 | |
c67499c0 PE |
263 | static inline struct sock *inet_lookup_listener(struct net *net, |
264 | struct inet_hashinfo *hashinfo, | |
da5e3630 | 265 | __be32 saddr, __be16 sport, |
c67499c0 | 266 | __be32 daddr, __be16 dport, int dif) |
8f491069 | 267 | { |
da5e3630 TH |
268 | return __inet_lookup_listener(net, hashinfo, saddr, sport, |
269 | daddr, ntohs(dport), dif); | |
8f491069 | 270 | } |
8feaf0c0 ACM |
271 | |
272 | /* Socket demux engine toys. */ | |
4f765d84 AV |
273 | /* What happens here is ugly; there's a pair of adjacent fields in |
274 | struct inet_sock; __be16 dport followed by __u16 num. We want to | |
275 | search by pair, so we combine the keys into a single 32bit value | |
276 | and compare with 32bit value read from &...->dport. Let's at least | |
277 | make sure that it's not mixed with anything else... | |
278 | On 64bit targets we combine comparisons with pair of adjacent __be32 | |
279 | fields in the same way. | |
280 | */ | |
8feaf0c0 ACM |
281 | #ifdef __BIG_ENDIAN |
282 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 283 | ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport))) |
8feaf0c0 ACM |
284 | #else /* __LITTLE_ENDIAN */ |
285 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 286 | ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport))) |
8feaf0c0 ACM |
287 | #endif |
288 | ||
289 | #if (BITS_PER_LONG == 64) | |
290 | #ifdef __BIG_ENDIAN | |
291 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
292 | const __addrpair __name = (__force __addrpair) ( \ |
293 | (((__force __u64)(__be32)(__saddr)) << 32) | \ | |
c7228317 | 294 | ((__force __u64)(__be32)(__daddr))) |
8feaf0c0 ACM |
295 | #else /* __LITTLE_ENDIAN */ |
296 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
297 | const __addrpair __name = (__force __addrpair) ( \ |
298 | (((__force __u64)(__be32)(__daddr)) << 32) | \ | |
c7228317 | 299 | ((__force __u64)(__be32)(__saddr))) |
8feaf0c0 | 300 | #endif /* __BIG_ENDIAN */ |
ce43b03e | 301 | #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \ |
50805466 ED |
302 | (((__sk)->sk_portpair == (__ports)) && \ |
303 | ((__sk)->sk_addrpair == (__cookie)) && \ | |
ce43b03e ED |
304 | (!(__sk)->sk_bound_dev_if || \ |
305 | ((__sk)->sk_bound_dev_if == (__dif))) && \ | |
306 | net_eq(sock_net(__sk), (__net))) | |
8feaf0c0 | 307 | #else /* 32-bit arch */ |
c7228317 JP |
308 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ |
309 | const int __name __deprecated __attribute__((unused)) | |
310 | ||
ce43b03e | 311 | #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \ |
50805466 ED |
312 | (((__sk)->sk_portpair == (__ports)) && \ |
313 | ((__sk)->sk_daddr == (__saddr)) && \ | |
314 | ((__sk)->sk_rcv_saddr == (__daddr)) && \ | |
ce43b03e ED |
315 | (!(__sk)->sk_bound_dev_if || \ |
316 | ((__sk)->sk_bound_dev_if == (__dif))) && \ | |
317 | net_eq(sock_net(__sk), (__net))) | |
8feaf0c0 | 318 | #endif /* 64-bit arch */ |
e48c414e ACM |
319 | |
320 | /* | |
321 | * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need | |
322 | * not check it for lookups anymore, thanks Alexey. -DaveM | |
323 | * | |
324 | * Local BH must be disabled here. | |
325 | */ | |
1fd51155 JP |
326 | struct sock *__inet_lookup_established(struct net *net, |
327 | struct inet_hashinfo *hashinfo, | |
328 | const __be32 saddr, const __be16 sport, | |
329 | const __be32 daddr, const u16 hnum, | |
330 | const int dif); | |
e48c414e | 331 | |
8f491069 | 332 | static inline struct sock * |
c67499c0 | 333 | inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo, |
fb99c848 AV |
334 | const __be32 saddr, const __be16 sport, |
335 | const __be32 daddr, const __be16 dport, | |
8f491069 HX |
336 | const int dif) |
337 | { | |
c67499c0 | 338 | return __inet_lookup_established(net, hashinfo, saddr, sport, daddr, |
8f491069 HX |
339 | ntohs(dport), dif); |
340 | } | |
341 | ||
c67499c0 PE |
342 | static inline struct sock *__inet_lookup(struct net *net, |
343 | struct inet_hashinfo *hashinfo, | |
fb99c848 AV |
344 | const __be32 saddr, const __be16 sport, |
345 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
346 | const int dif) |
347 | { | |
8f491069 | 348 | u16 hnum = ntohs(dport); |
c67499c0 PE |
349 | struct sock *sk = __inet_lookup_established(net, hashinfo, |
350 | saddr, sport, daddr, hnum, dif); | |
351 | ||
da5e3630 TH |
352 | return sk ? : __inet_lookup_listener(net, hashinfo, saddr, sport, |
353 | daddr, hnum, dif); | |
e48c414e ACM |
354 | } |
355 | ||
c67499c0 PE |
356 | static inline struct sock *inet_lookup(struct net *net, |
357 | struct inet_hashinfo *hashinfo, | |
fb99c848 AV |
358 | const __be32 saddr, const __be16 sport, |
359 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
360 | const int dif) |
361 | { | |
362 | struct sock *sk; | |
363 | ||
364 | local_bh_disable(); | |
c67499c0 | 365 | sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif); |
e48c414e ACM |
366 | local_bh_enable(); |
367 | ||
368 | return sk; | |
369 | } | |
a7f5e7f1 | 370 | |
9a1f27c4 ACM |
371 | static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo, |
372 | struct sk_buff *skb, | |
373 | const __be16 sport, | |
374 | const __be16 dport) | |
375 | { | |
41063e9d | 376 | struct sock *sk = skb_steal_sock(skb); |
9a1f27c4 ACM |
377 | const struct iphdr *iph = ip_hdr(skb); |
378 | ||
41063e9d | 379 | if (sk) |
23542618 KK |
380 | return sk; |
381 | else | |
adf30907 | 382 | return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, |
23542618 KK |
383 | iph->saddr, sport, |
384 | iph->daddr, dport, inet_iif(skb)); | |
9a1f27c4 ACM |
385 | } |
386 | ||
5b441f76 | 387 | u32 sk_ehashfn(const struct sock *sk); |
d1e559d0 ED |
388 | u32 inet6_ehashfn(const struct net *net, |
389 | const struct in6_addr *laddr, const u16 lport, | |
390 | const struct in6_addr *faddr, const __be16 fport); | |
391 | ||
392 | static inline void sk_daddr_set(struct sock *sk, __be32 addr) | |
393 | { | |
394 | sk->sk_daddr = addr; /* alias of inet_daddr */ | |
395 | #if IS_ENABLED(CONFIG_IPV6) | |
396 | ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr); | |
397 | #endif | |
398 | } | |
399 | ||
400 | static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr) | |
401 | { | |
402 | sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */ | |
403 | #if IS_ENABLED(CONFIG_IPV6) | |
404 | ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr); | |
405 | #endif | |
406 | } | |
5b441f76 | 407 | |
1fd51155 JP |
408 | int __inet_hash_connect(struct inet_timewait_death_row *death_row, |
409 | struct sock *sk, u32 port_offset, | |
410 | int (*check_established)(struct inet_timewait_death_row *, | |
411 | struct sock *, __u16, | |
b4d6444e | 412 | struct inet_timewait_sock **)); |
9327f705 | 413 | |
1fd51155 JP |
414 | int inet_hash_connect(struct inet_timewait_death_row *death_row, |
415 | struct sock *sk); | |
304a1618 | 416 | #endif /* _INET_HASHTABLES_H */ |