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; |
3c82a21f | 82 | int l3mdev; |
77d8bf9c | 83 | unsigned short port; |
da5e3630 TH |
84 | signed char fastreuse; |
85 | signed char fastreuseport; | |
86 | kuid_t fastuid; | |
637bc8bb JB |
87 | #if IS_ENABLED(CONFIG_IPV6) |
88 | struct in6_addr fast_v6_rcv_saddr; | |
89 | #endif | |
90 | __be32 fast_rcv_saddr; | |
91 | unsigned short fast_sk_family; | |
92 | bool fast_ipv6_only; | |
77d8bf9c ACM |
93 | struct hlist_node node; |
94 | struct hlist_head owners; | |
95 | }; | |
96 | ||
7a9546ee ED |
97 | static inline struct net *ib_net(struct inet_bind_bucket *ib) |
98 | { | |
99 | return read_pnet(&ib->ib_net); | |
100 | } | |
101 | ||
b67bfe0d SL |
102 | #define inet_bind_bucket_for_each(tb, head) \ |
103 | hlist_for_each_entry(tb, head, node) | |
77d8bf9c ACM |
104 | |
105 | struct inet_bind_hashbucket { | |
106 | spinlock_t lock; | |
107 | struct hlist_head chain; | |
108 | }; | |
109 | ||
c25eb3bf ED |
110 | /* |
111 | * Sockets can be hashed in established or listening table | |
c25eb3bf | 112 | */ |
5caea4ea ED |
113 | struct inet_listen_hashbucket { |
114 | spinlock_t lock; | |
76d013b2 | 115 | unsigned int count; |
3b24d854 | 116 | struct hlist_head head; |
5caea4ea ED |
117 | }; |
118 | ||
77d8bf9c ACM |
119 | /* This is for listening sockets, thus all sockets which possess wildcards. */ |
120 | #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */ | |
121 | ||
122 | struct inet_hashinfo { | |
123 | /* This is for sockets with full identity only. Sockets here will | |
124 | * always be without wildcards and will have the following invariant: | |
125 | * | |
126 | * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE | |
127 | * | |
77d8bf9c ACM |
128 | */ |
129 | struct inet_ehash_bucket *ehash; | |
9db66bdc | 130 | spinlock_t *ehash_locks; |
f373b53b | 131 | unsigned int ehash_mask; |
230140cf | 132 | unsigned int ehash_locks_mask; |
77d8bf9c ACM |
133 | |
134 | /* Ok, let's try this, I give up, we do need a local binding | |
135 | * TCP hash as well as the others for fast bind/connect. | |
136 | */ | |
61b7c691 | 137 | struct kmem_cache *bind_bucket_cachep; |
77d8bf9c | 138 | struct inet_bind_hashbucket *bhash; |
cfcabdcc | 139 | unsigned int bhash_size; |
77d8bf9c | 140 | |
61b7c691 MKL |
141 | /* The 2nd listener table hashed by local port and address */ |
142 | unsigned int lhash2_mask; | |
143 | struct inet_listen_hashbucket *lhash2; | |
77d8bf9c ACM |
144 | |
145 | /* All the above members are written once at bootup and | |
146 | * never written again _or_ are predominantly read-access. | |
147 | * | |
148 | * Now align to a new cache line as all the following members | |
5caea4ea ED |
149 | * might be often dirty. |
150 | */ | |
61b7c691 MKL |
151 | /* All sockets in TCP_LISTEN state will be in listening_hash. |
152 | * This is the only table where wildcard'd TCP sockets can | |
153 | * exist. listening_hash is only hashed by local port number. | |
154 | * If lhash2 is initialized, the same socket will also be hashed | |
155 | * to lhash2 by port and address. | |
77d8bf9c | 156 | */ |
5caea4ea ED |
157 | struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE] |
158 | ____cacheline_aligned_in_smp; | |
77d8bf9c ACM |
159 | }; |
160 | ||
61b7c691 MKL |
161 | #define inet_lhash2_for_each_icsk_rcu(__icsk, list) \ |
162 | hlist_for_each_entry_rcu(__icsk, list, icsk_listen_portaddr_node) | |
163 | ||
164 | static inline struct inet_listen_hashbucket * | |
165 | inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash) | |
166 | { | |
167 | return &h->lhash2[hash & h->lhash2_mask]; | |
168 | } | |
169 | ||
81c3d547 ED |
170 | static inline struct inet_ehash_bucket *inet_ehash_bucket( |
171 | struct inet_hashinfo *hashinfo, | |
172 | unsigned int hash) | |
173 | { | |
f373b53b | 174 | return &hashinfo->ehash[hash & hashinfo->ehash_mask]; |
304a1618 ACM |
175 | } |
176 | ||
9db66bdc | 177 | static inline spinlock_t *inet_ehash_lockp( |
230140cf ED |
178 | struct inet_hashinfo *hashinfo, |
179 | unsigned int hash) | |
180 | { | |
181 | return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask]; | |
182 | } | |
183 | ||
095dc8e0 | 184 | int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo); |
230140cf ED |
185 | |
186 | static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo) | |
187 | { | |
095dc8e0 ED |
188 | kvfree(hashinfo->ehash_locks); |
189 | hashinfo->ehash_locks = NULL; | |
230140cf ED |
190 | } |
191 | ||
e7819058 MM |
192 | static inline bool inet_sk_bound_dev_eq(struct net *net, int bound_dev_if, |
193 | int dif, int sdif) | |
194 | { | |
195 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) | |
196 | return inet_bound_dev_eq(!!net->ipv4.sysctl_tcp_l3mdev_accept, | |
197 | bound_dev_if, dif, sdif); | |
198 | #else | |
199 | return inet_bound_dev_eq(true, bound_dev_if, dif, sdif); | |
200 | #endif | |
201 | } | |
202 | ||
1fd51155 JP |
203 | struct inet_bind_bucket * |
204 | inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net, | |
205 | struct inet_bind_hashbucket *head, | |
3c82a21f | 206 | const unsigned short snum, int l3mdev); |
1fd51155 JP |
207 | void inet_bind_bucket_destroy(struct kmem_cache *cachep, |
208 | struct inet_bind_bucket *tb); | |
209 | ||
6eada011 ED |
210 | static inline u32 inet_bhashfn(const struct net *net, const __u16 lport, |
211 | const u32 bhash_size) | |
77d8bf9c | 212 | { |
0b441916 | 213 | return (lport + net_hash_mix(net)) & (bhash_size - 1); |
77d8bf9c ACM |
214 | } |
215 | ||
1fd51155 JP |
216 | void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, |
217 | const unsigned short snum); | |
2d8c4ce5 | 218 | |
77d8bf9c | 219 | /* These can have wildcards, don't try too hard. */ |
6eada011 | 220 | static inline u32 inet_lhashfn(const struct net *net, const unsigned short num) |
77d8bf9c | 221 | { |
0b441916 | 222 | return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1); |
77d8bf9c ACM |
223 | } |
224 | ||
225 | static inline int inet_sk_listen_hashfn(const struct sock *sk) | |
226 | { | |
c720c7e8 | 227 | return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num); |
77d8bf9c ACM |
228 | } |
229 | ||
2d8c4ce5 | 230 | /* Caller must disable local BH processing. */ |
1ce31c9e | 231 | int __inet_inherit_port(const struct sock *sk, struct sock *child); |
2d8c4ce5 | 232 | |
1fd51155 | 233 | void inet_put_port(struct sock *sk); |
2d8c4ce5 | 234 | |
5caea4ea | 235 | void inet_hashinfo_init(struct inet_hashinfo *h); |
61b7c691 MKL |
236 | void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name, |
237 | unsigned long numentries, int scale, | |
238 | unsigned long low_limit, | |
239 | unsigned long high_limit); | |
c92c81df | 240 | int inet_hashinfo2_init_mod(struct inet_hashinfo *h); |
f3f05f70 | 241 | |
5e0724d0 ED |
242 | bool inet_ehash_insert(struct sock *sk, struct sock *osk); |
243 | bool inet_ehash_nolisten(struct sock *sk, struct sock *osk); | |
fe38d2a1 | 244 | int __inet_hash(struct sock *sk, struct sock *osk); |
086c653f | 245 | int inet_hash(struct sock *sk); |
1fd51155 | 246 | void inet_unhash(struct sock *sk); |
33b62231 | 247 | |
1fd51155 JP |
248 | struct sock *__inet_lookup_listener(struct net *net, |
249 | struct inet_hashinfo *hashinfo, | |
a583636a | 250 | struct sk_buff *skb, int doff, |
1fd51155 JP |
251 | const __be32 saddr, const __be16 sport, |
252 | const __be32 daddr, | |
253 | const unsigned short hnum, | |
3fa6f616 | 254 | const int dif, const int sdif); |
8f491069 | 255 | |
c67499c0 PE |
256 | static inline struct sock *inet_lookup_listener(struct net *net, |
257 | struct inet_hashinfo *hashinfo, | |
a583636a | 258 | struct sk_buff *skb, int doff, |
da5e3630 | 259 | __be32 saddr, __be16 sport, |
3fa6f616 | 260 | __be32 daddr, __be16 dport, int dif, int sdif) |
8f491069 | 261 | { |
a583636a | 262 | return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport, |
3fa6f616 | 263 | daddr, ntohs(dport), dif, sdif); |
8f491069 | 264 | } |
8feaf0c0 ACM |
265 | |
266 | /* Socket demux engine toys. */ | |
4f765d84 AV |
267 | /* What happens here is ugly; there's a pair of adjacent fields in |
268 | struct inet_sock; __be16 dport followed by __u16 num. We want to | |
269 | search by pair, so we combine the keys into a single 32bit value | |
270 | and compare with 32bit value read from &...->dport. Let's at least | |
271 | make sure that it's not mixed with anything else... | |
272 | On 64bit targets we combine comparisons with pair of adjacent __be32 | |
273 | fields in the same way. | |
274 | */ | |
8feaf0c0 ACM |
275 | #ifdef __BIG_ENDIAN |
276 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 277 | ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport))) |
8feaf0c0 ACM |
278 | #else /* __LITTLE_ENDIAN */ |
279 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 280 | ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport))) |
8feaf0c0 ACM |
281 | #endif |
282 | ||
283 | #if (BITS_PER_LONG == 64) | |
284 | #ifdef __BIG_ENDIAN | |
285 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
286 | const __addrpair __name = (__force __addrpair) ( \ |
287 | (((__force __u64)(__be32)(__saddr)) << 32) | \ | |
c7228317 | 288 | ((__force __u64)(__be32)(__daddr))) |
8feaf0c0 ACM |
289 | #else /* __LITTLE_ENDIAN */ |
290 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
291 | const __addrpair __name = (__force __addrpair) ( \ |
292 | (((__force __u64)(__be32)(__daddr)) << 32) | \ | |
c7228317 | 293 | ((__force __u64)(__be32)(__saddr))) |
8feaf0c0 | 294 | #endif /* __BIG_ENDIAN */ |
3fa6f616 | 295 | #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \ |
50805466 ED |
296 | (((__sk)->sk_portpair == (__ports)) && \ |
297 | ((__sk)->sk_addrpair == (__cookie)) && \ | |
3c82a21f RS |
298 | (((__sk)->sk_bound_dev_if == (__dif)) || \ |
299 | ((__sk)->sk_bound_dev_if == (__sdif))) && \ | |
ce43b03e | 300 | net_eq(sock_net(__sk), (__net))) |
8feaf0c0 | 301 | #else /* 32-bit arch */ |
c7228317 JP |
302 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ |
303 | const int __name __deprecated __attribute__((unused)) | |
304 | ||
3fa6f616 | 305 | #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \ |
50805466 ED |
306 | (((__sk)->sk_portpair == (__ports)) && \ |
307 | ((__sk)->sk_daddr == (__saddr)) && \ | |
308 | ((__sk)->sk_rcv_saddr == (__daddr)) && \ | |
3c82a21f RS |
309 | (((__sk)->sk_bound_dev_if == (__dif)) || \ |
310 | ((__sk)->sk_bound_dev_if == (__sdif))) && \ | |
ce43b03e | 311 | net_eq(sock_net(__sk), (__net))) |
8feaf0c0 | 312 | #endif /* 64-bit arch */ |
e48c414e | 313 | |
ee3cf32a | 314 | /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need |
e48c414e | 315 | * not check it for lookups anymore, thanks Alexey. -DaveM |
e48c414e | 316 | */ |
1fd51155 JP |
317 | struct sock *__inet_lookup_established(struct net *net, |
318 | struct inet_hashinfo *hashinfo, | |
319 | const __be32 saddr, const __be16 sport, | |
320 | const __be32 daddr, const u16 hnum, | |
3fa6f616 | 321 | const int dif, const int sdif); |
e48c414e | 322 | |
8f491069 | 323 | static inline struct sock * |
c67499c0 | 324 | inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo, |
fb99c848 AV |
325 | const __be32 saddr, const __be16 sport, |
326 | const __be32 daddr, const __be16 dport, | |
8f491069 HX |
327 | const int dif) |
328 | { | |
c67499c0 | 329 | return __inet_lookup_established(net, hashinfo, saddr, sport, daddr, |
3fa6f616 | 330 | ntohs(dport), dif, 0); |
8f491069 HX |
331 | } |
332 | ||
c67499c0 PE |
333 | static inline struct sock *__inet_lookup(struct net *net, |
334 | struct inet_hashinfo *hashinfo, | |
a583636a | 335 | struct sk_buff *skb, int doff, |
fb99c848 AV |
336 | const __be32 saddr, const __be16 sport, |
337 | const __be32 daddr, const __be16 dport, | |
3fa6f616 | 338 | const int dif, const int sdif, |
3b24d854 | 339 | bool *refcounted) |
e48c414e | 340 | { |
8f491069 | 341 | u16 hnum = ntohs(dport); |
3b24d854 | 342 | struct sock *sk; |
c67499c0 | 343 | |
3b24d854 | 344 | sk = __inet_lookup_established(net, hashinfo, saddr, sport, |
3fa6f616 | 345 | daddr, hnum, dif, sdif); |
3b24d854 ED |
346 | *refcounted = true; |
347 | if (sk) | |
348 | return sk; | |
349 | *refcounted = false; | |
350 | return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, | |
3fa6f616 | 351 | sport, daddr, hnum, dif, sdif); |
e48c414e ACM |
352 | } |
353 | ||
c67499c0 PE |
354 | static inline struct sock *inet_lookup(struct net *net, |
355 | struct inet_hashinfo *hashinfo, | |
a583636a | 356 | struct sk_buff *skb, int doff, |
fb99c848 AV |
357 | const __be32 saddr, const __be16 sport, |
358 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
359 | const int dif) |
360 | { | |
361 | struct sock *sk; | |
3b24d854 | 362 | bool refcounted; |
e48c414e | 363 | |
a583636a | 364 | sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr, |
3fa6f616 | 365 | dport, dif, 0, &refcounted); |
e48c414e | 366 | |
41c6d650 | 367 | if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt)) |
3b24d854 | 368 | sk = NULL; |
e48c414e ACM |
369 | return sk; |
370 | } | |
a7f5e7f1 | 371 | |
9a1f27c4 ACM |
372 | static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo, |
373 | struct sk_buff *skb, | |
a583636a | 374 | int doff, |
9a1f27c4 | 375 | const __be16 sport, |
3b24d854 | 376 | const __be16 dport, |
3fa6f616 | 377 | const int sdif, |
3b24d854 | 378 | bool *refcounted) |
9a1f27c4 | 379 | { |
41063e9d | 380 | struct sock *sk = skb_steal_sock(skb); |
9a1f27c4 ACM |
381 | const struct iphdr *iph = ip_hdr(skb); |
382 | ||
3b24d854 | 383 | *refcounted = true; |
41063e9d | 384 | if (sk) |
23542618 | 385 | return sk; |
3b24d854 ED |
386 | |
387 | return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb, | |
388 | doff, iph->saddr, sport, | |
3fa6f616 | 389 | iph->daddr, dport, inet_iif(skb), sdif, |
3b24d854 | 390 | refcounted); |
9a1f27c4 ACM |
391 | } |
392 | ||
d1e559d0 ED |
393 | u32 inet6_ehashfn(const struct net *net, |
394 | const struct in6_addr *laddr, const u16 lport, | |
395 | const struct in6_addr *faddr, const __be16 fport); | |
396 | ||
397 | static inline void sk_daddr_set(struct sock *sk, __be32 addr) | |
398 | { | |
399 | sk->sk_daddr = addr; /* alias of inet_daddr */ | |
400 | #if IS_ENABLED(CONFIG_IPV6) | |
401 | ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr); | |
402 | #endif | |
403 | } | |
404 | ||
405 | static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr) | |
406 | { | |
407 | sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */ | |
408 | #if IS_ENABLED(CONFIG_IPV6) | |
409 | ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr); | |
410 | #endif | |
411 | } | |
5b441f76 | 412 | |
1fd51155 JP |
413 | int __inet_hash_connect(struct inet_timewait_death_row *death_row, |
414 | struct sock *sk, u32 port_offset, | |
415 | int (*check_established)(struct inet_timewait_death_row *, | |
416 | struct sock *, __u16, | |
b4d6444e | 417 | struct inet_timewait_sock **)); |
9327f705 | 418 | |
1fd51155 JP |
419 | int inet_hash_connect(struct inet_timewait_death_row *death_row, |
420 | struct sock *sk); | |
304a1618 | 421 | #endif /* _INET_HASHTABLES_H */ |