]>
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> |
304a1618 | 27 | |
463c84b9 | 28 | #include <net/inet_connection_sock.h> |
14c85021 | 29 | #include <net/inet_sock.h> |
2d8c4ce5 | 30 | #include <net/sock.h> |
9a1f27c4 | 31 | #include <net/route.h> |
c752f073 | 32 | #include <net/tcp_states.h> |
0b441916 | 33 | #include <net/netns/hash.h> |
2d8c4ce5 | 34 | |
41c6d650 | 35 | #include <linux/refcount.h> |
e48c414e | 36 | #include <asm/byteorder.h> |
f3f05f70 | 37 | |
77d8bf9c | 38 | /* This is for all connections with a full identity, no wildcards. |
05dbc7b5 ED |
39 | * The 'e' prefix stands for Establish, but we really put all sockets |
40 | * but LISTEN ones. | |
77d8bf9c ACM |
41 | */ |
42 | struct inet_ehash_bucket { | |
3ab5aee7 | 43 | struct hlist_nulls_head chain; |
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 | */ | |
637bc8bb JB |
77 | #define FASTREUSEPORT_ANY 1 |
78 | #define FASTREUSEPORT_STRICT 2 | |
79 | ||
77d8bf9c | 80 | struct inet_bind_bucket { |
0c5c9fb5 | 81 | possible_net_t ib_net; |
77d8bf9c | 82 | unsigned short port; |
da5e3630 TH |
83 | signed char fastreuse; |
84 | signed char fastreuseport; | |
85 | kuid_t fastuid; | |
637bc8bb JB |
86 | #if IS_ENABLED(CONFIG_IPV6) |
87 | struct in6_addr fast_v6_rcv_saddr; | |
88 | #endif | |
89 | __be32 fast_rcv_saddr; | |
90 | unsigned short fast_sk_family; | |
91 | bool fast_ipv6_only; | |
77d8bf9c ACM |
92 | struct hlist_node node; |
93 | struct hlist_head owners; | |
94 | }; | |
95 | ||
7a9546ee ED |
96 | static inline struct net *ib_net(struct inet_bind_bucket *ib) |
97 | { | |
98 | return read_pnet(&ib->ib_net); | |
99 | } | |
100 | ||
b67bfe0d SL |
101 | #define inet_bind_bucket_for_each(tb, head) \ |
102 | hlist_for_each_entry(tb, head, node) | |
77d8bf9c ACM |
103 | |
104 | struct inet_bind_hashbucket { | |
105 | spinlock_t lock; | |
106 | struct hlist_head chain; | |
107 | }; | |
108 | ||
6cb8e266 ED |
109 | /* Sockets can be hashed in established or listening table. |
110 | * We must use different 'nulls' end-of-chain value for all hash buckets : | |
111 | * A socket might transition from ESTABLISH to LISTEN state without | |
112 | * RCU grace period. A lookup in ehash table needs to handle this case. | |
c25eb3bf | 113 | */ |
6cb8e266 | 114 | #define LISTENING_NULLS_BASE (1U << 29) |
5caea4ea ED |
115 | struct inet_listen_hashbucket { |
116 | spinlock_t lock; | |
6cb8e266 ED |
117 | union { |
118 | struct hlist_head head; | |
119 | struct hlist_nulls_head nulls_head; | |
120 | }; | |
5caea4ea ED |
121 | }; |
122 | ||
77d8bf9c ACM |
123 | /* This is for listening sockets, thus all sockets which possess wildcards. */ |
124 | #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */ | |
125 | ||
126 | struct inet_hashinfo { | |
127 | /* This is for sockets with full identity only. Sockets here will | |
128 | * always be without wildcards and will have the following invariant: | |
129 | * | |
130 | * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE | |
131 | * | |
77d8bf9c ACM |
132 | */ |
133 | struct inet_ehash_bucket *ehash; | |
9db66bdc | 134 | spinlock_t *ehash_locks; |
f373b53b | 135 | unsigned int ehash_mask; |
230140cf | 136 | unsigned int ehash_locks_mask; |
77d8bf9c ACM |
137 | |
138 | /* Ok, let's try this, I give up, we do need a local binding | |
139 | * TCP hash as well as the others for fast bind/connect. | |
140 | */ | |
141 | struct inet_bind_hashbucket *bhash; | |
142 | ||
cfcabdcc | 143 | unsigned int bhash_size; |
24dd1fa1 | 144 | /* 4 bytes hole on 64 bit */ |
77d8bf9c | 145 | |
5caea4ea | 146 | struct kmem_cache *bind_bucket_cachep; |
77d8bf9c ACM |
147 | |
148 | /* All the above members are written once at bootup and | |
149 | * never written again _or_ are predominantly read-access. | |
150 | * | |
151 | * Now align to a new cache line as all the following members | |
5caea4ea ED |
152 | * might be often dirty. |
153 | */ | |
154 | /* All sockets in TCP_LISTEN state will be in here. This is the only | |
155 | * table where wildcard'd TCP sockets can exist. Hash function here | |
156 | * is just local port number. | |
77d8bf9c | 157 | */ |
5caea4ea ED |
158 | struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE] |
159 | ____cacheline_aligned_in_smp; | |
77d8bf9c ACM |
160 | }; |
161 | ||
81c3d547 ED |
162 | static inline struct inet_ehash_bucket *inet_ehash_bucket( |
163 | struct inet_hashinfo *hashinfo, | |
164 | unsigned int hash) | |
165 | { | |
f373b53b | 166 | return &hashinfo->ehash[hash & hashinfo->ehash_mask]; |
304a1618 ACM |
167 | } |
168 | ||
9db66bdc | 169 | static inline spinlock_t *inet_ehash_lockp( |
230140cf ED |
170 | struct inet_hashinfo *hashinfo, |
171 | unsigned int hash) | |
172 | { | |
173 | return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask]; | |
174 | } | |
175 | ||
095dc8e0 | 176 | int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo); |
230140cf ED |
177 | |
178 | static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo) | |
179 | { | |
095dc8e0 ED |
180 | kvfree(hashinfo->ehash_locks); |
181 | hashinfo->ehash_locks = NULL; | |
230140cf ED |
182 | } |
183 | ||
1fd51155 JP |
184 | struct inet_bind_bucket * |
185 | inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net, | |
186 | struct inet_bind_hashbucket *head, | |
187 | const unsigned short snum); | |
188 | void inet_bind_bucket_destroy(struct kmem_cache *cachep, | |
189 | struct inet_bind_bucket *tb); | |
190 | ||
6eada011 ED |
191 | static inline u32 inet_bhashfn(const struct net *net, const __u16 lport, |
192 | const u32 bhash_size) | |
77d8bf9c | 193 | { |
0b441916 | 194 | return (lport + net_hash_mix(net)) & (bhash_size - 1); |
77d8bf9c ACM |
195 | } |
196 | ||
1fd51155 JP |
197 | void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, |
198 | const unsigned short snum); | |
2d8c4ce5 | 199 | |
77d8bf9c | 200 | /* These can have wildcards, don't try too hard. */ |
6eada011 | 201 | static inline u32 inet_lhashfn(const struct net *net, const unsigned short num) |
77d8bf9c | 202 | { |
0b441916 | 203 | return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1); |
77d8bf9c ACM |
204 | } |
205 | ||
206 | static inline int inet_sk_listen_hashfn(const struct sock *sk) | |
207 | { | |
c720c7e8 | 208 | return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num); |
77d8bf9c ACM |
209 | } |
210 | ||
2d8c4ce5 | 211 | /* Caller must disable local BH processing. */ |
1ce31c9e | 212 | int __inet_inherit_port(const struct sock *sk, struct sock *child); |
2d8c4ce5 | 213 | |
1fd51155 | 214 | void inet_put_port(struct sock *sk); |
2d8c4ce5 | 215 | |
5caea4ea | 216 | void inet_hashinfo_init(struct inet_hashinfo *h); |
f3f05f70 | 217 | |
5e0724d0 ED |
218 | bool inet_ehash_insert(struct sock *sk, struct sock *osk); |
219 | bool inet_ehash_nolisten(struct sock *sk, struct sock *osk); | |
fe38d2a1 | 220 | int __inet_hash(struct sock *sk, struct sock *osk); |
086c653f | 221 | int inet_hash(struct sock *sk); |
1fd51155 | 222 | void inet_unhash(struct sock *sk); |
33b62231 | 223 | |
1fd51155 JP |
224 | struct sock *__inet_lookup_listener(struct net *net, |
225 | struct inet_hashinfo *hashinfo, | |
a583636a | 226 | struct sk_buff *skb, int doff, |
1fd51155 JP |
227 | const __be32 saddr, const __be16 sport, |
228 | const __be32 daddr, | |
229 | const unsigned short hnum, | |
3fa6f616 | 230 | const int dif, const int sdif); |
8f491069 | 231 | |
c67499c0 PE |
232 | static inline struct sock *inet_lookup_listener(struct net *net, |
233 | struct inet_hashinfo *hashinfo, | |
a583636a | 234 | struct sk_buff *skb, int doff, |
da5e3630 | 235 | __be32 saddr, __be16 sport, |
3fa6f616 | 236 | __be32 daddr, __be16 dport, int dif, int sdif) |
8f491069 | 237 | { |
a583636a | 238 | return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport, |
3fa6f616 | 239 | daddr, ntohs(dport), dif, sdif); |
8f491069 | 240 | } |
8feaf0c0 ACM |
241 | |
242 | /* Socket demux engine toys. */ | |
4f765d84 AV |
243 | /* What happens here is ugly; there's a pair of adjacent fields in |
244 | struct inet_sock; __be16 dport followed by __u16 num. We want to | |
245 | search by pair, so we combine the keys into a single 32bit value | |
246 | and compare with 32bit value read from &...->dport. Let's at least | |
247 | make sure that it's not mixed with anything else... | |
248 | On 64bit targets we combine comparisons with pair of adjacent __be32 | |
249 | fields in the same way. | |
250 | */ | |
8feaf0c0 ACM |
251 | #ifdef __BIG_ENDIAN |
252 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 253 | ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport))) |
8feaf0c0 ACM |
254 | #else /* __LITTLE_ENDIAN */ |
255 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 256 | ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport))) |
8feaf0c0 ACM |
257 | #endif |
258 | ||
259 | #if (BITS_PER_LONG == 64) | |
260 | #ifdef __BIG_ENDIAN | |
261 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
262 | const __addrpair __name = (__force __addrpair) ( \ |
263 | (((__force __u64)(__be32)(__saddr)) << 32) | \ | |
c7228317 | 264 | ((__force __u64)(__be32)(__daddr))) |
8feaf0c0 ACM |
265 | #else /* __LITTLE_ENDIAN */ |
266 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
267 | const __addrpair __name = (__force __addrpair) ( \ |
268 | (((__force __u64)(__be32)(__daddr)) << 32) | \ | |
c7228317 | 269 | ((__force __u64)(__be32)(__saddr))) |
8feaf0c0 | 270 | #endif /* __BIG_ENDIAN */ |
3fa6f616 | 271 | #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \ |
50805466 ED |
272 | (((__sk)->sk_portpair == (__ports)) && \ |
273 | ((__sk)->sk_addrpair == (__cookie)) && \ | |
ce43b03e | 274 | (!(__sk)->sk_bound_dev_if || \ |
3fa6f616 DA |
275 | ((__sk)->sk_bound_dev_if == (__dif)) || \ |
276 | ((__sk)->sk_bound_dev_if == (__sdif))) && \ | |
ce43b03e | 277 | net_eq(sock_net(__sk), (__net))) |
8feaf0c0 | 278 | #else /* 32-bit arch */ |
c7228317 JP |
279 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ |
280 | const int __name __deprecated __attribute__((unused)) | |
281 | ||
3fa6f616 | 282 | #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \ |
50805466 ED |
283 | (((__sk)->sk_portpair == (__ports)) && \ |
284 | ((__sk)->sk_daddr == (__saddr)) && \ | |
285 | ((__sk)->sk_rcv_saddr == (__daddr)) && \ | |
ce43b03e | 286 | (!(__sk)->sk_bound_dev_if || \ |
3fa6f616 DA |
287 | ((__sk)->sk_bound_dev_if == (__dif)) || \ |
288 | ((__sk)->sk_bound_dev_if == (__sdif))) && \ | |
ce43b03e | 289 | net_eq(sock_net(__sk), (__net))) |
8feaf0c0 | 290 | #endif /* 64-bit arch */ |
e48c414e | 291 | |
ee3cf32a | 292 | /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need |
e48c414e | 293 | * not check it for lookups anymore, thanks Alexey. -DaveM |
e48c414e | 294 | */ |
1fd51155 JP |
295 | struct sock *__inet_lookup_established(struct net *net, |
296 | struct inet_hashinfo *hashinfo, | |
297 | const __be32 saddr, const __be16 sport, | |
298 | const __be32 daddr, const u16 hnum, | |
3fa6f616 | 299 | const int dif, const int sdif); |
e48c414e | 300 | |
8f491069 | 301 | static inline struct sock * |
c67499c0 | 302 | inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo, |
fb99c848 AV |
303 | const __be32 saddr, const __be16 sport, |
304 | const __be32 daddr, const __be16 dport, | |
8f491069 HX |
305 | const int dif) |
306 | { | |
c67499c0 | 307 | return __inet_lookup_established(net, hashinfo, saddr, sport, daddr, |
3fa6f616 | 308 | ntohs(dport), dif, 0); |
8f491069 HX |
309 | } |
310 | ||
c67499c0 PE |
311 | static inline struct sock *__inet_lookup(struct net *net, |
312 | struct inet_hashinfo *hashinfo, | |
a583636a | 313 | struct sk_buff *skb, int doff, |
fb99c848 AV |
314 | const __be32 saddr, const __be16 sport, |
315 | const __be32 daddr, const __be16 dport, | |
3fa6f616 | 316 | const int dif, const int sdif, |
3b24d854 | 317 | bool *refcounted) |
e48c414e | 318 | { |
8f491069 | 319 | u16 hnum = ntohs(dport); |
3b24d854 | 320 | struct sock *sk; |
c67499c0 | 321 | |
3b24d854 | 322 | sk = __inet_lookup_established(net, hashinfo, saddr, sport, |
3fa6f616 | 323 | daddr, hnum, dif, sdif); |
3b24d854 ED |
324 | *refcounted = true; |
325 | if (sk) | |
326 | return sk; | |
327 | *refcounted = false; | |
328 | return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, | |
3fa6f616 | 329 | sport, daddr, hnum, dif, sdif); |
e48c414e ACM |
330 | } |
331 | ||
c67499c0 PE |
332 | static inline struct sock *inet_lookup(struct net *net, |
333 | struct inet_hashinfo *hashinfo, | |
a583636a | 334 | struct sk_buff *skb, int doff, |
fb99c848 AV |
335 | const __be32 saddr, const __be16 sport, |
336 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
337 | const int dif) |
338 | { | |
339 | struct sock *sk; | |
3b24d854 | 340 | bool refcounted; |
e48c414e | 341 | |
a583636a | 342 | sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr, |
3fa6f616 | 343 | dport, dif, 0, &refcounted); |
e48c414e | 344 | |
41c6d650 | 345 | if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt)) |
3b24d854 | 346 | sk = NULL; |
e48c414e ACM |
347 | return sk; |
348 | } | |
a7f5e7f1 | 349 | |
9a1f27c4 ACM |
350 | static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo, |
351 | struct sk_buff *skb, | |
a583636a | 352 | int doff, |
9a1f27c4 | 353 | const __be16 sport, |
3b24d854 | 354 | const __be16 dport, |
3fa6f616 | 355 | const int sdif, |
3b24d854 | 356 | bool *refcounted) |
9a1f27c4 | 357 | { |
41063e9d | 358 | struct sock *sk = skb_steal_sock(skb); |
9a1f27c4 ACM |
359 | const struct iphdr *iph = ip_hdr(skb); |
360 | ||
3b24d854 | 361 | *refcounted = true; |
41063e9d | 362 | if (sk) |
23542618 | 363 | return sk; |
3b24d854 ED |
364 | |
365 | return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb, | |
366 | doff, iph->saddr, sport, | |
3fa6f616 | 367 | iph->daddr, dport, inet_iif(skb), sdif, |
3b24d854 | 368 | refcounted); |
9a1f27c4 ACM |
369 | } |
370 | ||
d1e559d0 ED |
371 | u32 inet6_ehashfn(const struct net *net, |
372 | const struct in6_addr *laddr, const u16 lport, | |
373 | const struct in6_addr *faddr, const __be16 fport); | |
374 | ||
375 | static inline void sk_daddr_set(struct sock *sk, __be32 addr) | |
376 | { | |
377 | sk->sk_daddr = addr; /* alias of inet_daddr */ | |
378 | #if IS_ENABLED(CONFIG_IPV6) | |
379 | ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr); | |
380 | #endif | |
381 | } | |
382 | ||
383 | static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr) | |
384 | { | |
385 | sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */ | |
386 | #if IS_ENABLED(CONFIG_IPV6) | |
387 | ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr); | |
388 | #endif | |
389 | } | |
5b441f76 | 390 | |
1fd51155 JP |
391 | int __inet_hash_connect(struct inet_timewait_death_row *death_row, |
392 | struct sock *sk, u32 port_offset, | |
393 | int (*check_established)(struct inet_timewait_death_row *, | |
394 | struct sock *, __u16, | |
b4d6444e | 395 | struct inet_timewait_sock **)); |
9327f705 | 396 | |
1fd51155 JP |
397 | int inet_hash_connect(struct inet_timewait_death_row *death_row, |
398 | struct sock *sk); | |
304a1618 | 399 | #endif /* _INET_HASHTABLES_H */ |