Commit | Line | Data |
---|---|---|
2874c5fd | 1 | /* SPDX-License-Identifier: GPL-2.0-or-later */ |
304a1618 ACM |
2 | /* |
3 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
4 | * operating system. INET is implemented using the BSD Socket | |
5 | * interface as the means of communication with the user level. | |
6 | * | |
7 | * Authors: Lotsa people, from code originally in tcp | |
304a1618 ACM |
8 | */ |
9 | ||
10 | #ifndef _INET_HASHTABLES_H | |
11 | #define _INET_HASHTABLES_H | |
12 | ||
8feaf0c0 | 13 | |
2d8c4ce5 | 14 | #include <linux/interrupt.h> |
9a1f27c4 | 15 | #include <linux/ip.h> |
33b62231 | 16 | #include <linux/ipv6.h> |
77d8bf9c ACM |
17 | #include <linux/list.h> |
18 | #include <linux/slab.h> | |
33b62231 | 19 | #include <linux/socket.h> |
77d8bf9c | 20 | #include <linux/spinlock.h> |
304a1618 | 21 | #include <linux/types.h> |
f3f05f70 | 22 | #include <linux/wait.h> |
304a1618 | 23 | |
463c84b9 | 24 | #include <net/inet_connection_sock.h> |
14c85021 | 25 | #include <net/inet_sock.h> |
2d8c4ce5 | 26 | #include <net/sock.h> |
9a1f27c4 | 27 | #include <net/route.h> |
c752f073 | 28 | #include <net/tcp_states.h> |
0b441916 | 29 | #include <net/netns/hash.h> |
2d8c4ce5 | 30 | |
41c6d650 | 31 | #include <linux/refcount.h> |
e48c414e | 32 | #include <asm/byteorder.h> |
f3f05f70 | 33 | |
77d8bf9c | 34 | /* This is for all connections with a full identity, no wildcards. |
05dbc7b5 ED |
35 | * The 'e' prefix stands for Establish, but we really put all sockets |
36 | * but LISTEN ones. | |
77d8bf9c ACM |
37 | */ |
38 | struct inet_ehash_bucket { | |
3ab5aee7 | 39 | struct hlist_nulls_head chain; |
6d255361 | 40 | }; |
77d8bf9c ACM |
41 | |
42 | /* There are a few simple rules, which allow for local port reuse by | |
43 | * an application. In essence: | |
44 | * | |
45 | * 1) Sockets bound to different interfaces may share a local port. | |
46 | * Failing that, goto test 2. | |
47 | * 2) If all sockets have sk->sk_reuse set, and none of them are in | |
48 | * TCP_LISTEN state, the port may be shared. | |
49 | * Failing that, goto test 3. | |
50 | * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local | |
51 | * address, and none of them are the same, the port may be | |
52 | * shared. | |
53 | * Failing this, the port cannot be shared. | |
54 | * | |
55 | * The interesting point, is test #2. This is what an FTP server does | |
56 | * all day. To optimize this case we use a specific flag bit defined | |
57 | * below. As we add sockets to a bind bucket list, we perform a | |
58 | * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN)) | |
59 | * As long as all sockets added to a bind bucket pass this test, | |
60 | * the flag bit will be set. | |
61 | * The resulting situation is that tcp_v[46]_verify_bind() can just check | |
62 | * for this flag bit, if it is set and the socket trying to bind has | |
63 | * sk->sk_reuse set, we don't even have to walk the owners list at all, | |
64 | * we return that it is ok to bind this socket to the requested local port. | |
65 | * | |
66 | * Sounds like a lot of work, but it is worth it. In a more naive | |
67 | * implementation (ie. current FreeBSD etc.) the entire list of ports | |
68 | * must be walked for each data port opened by an ftp server. Needless | |
69 | * to say, this does not scale at all. With a couple thousand FTP | |
70 | * users logged onto your box, isn't it nice to know that new data | |
71 | * ports are created in O(1) time? I thought so. ;-) -DaveM | |
72 | */ | |
637bc8bb JB |
73 | #define FASTREUSEPORT_ANY 1 |
74 | #define FASTREUSEPORT_STRICT 2 | |
75 | ||
77d8bf9c | 76 | struct inet_bind_bucket { |
0c5c9fb5 | 77 | possible_net_t ib_net; |
3c82a21f | 78 | int l3mdev; |
77d8bf9c | 79 | unsigned short port; |
da5e3630 TH |
80 | signed char fastreuse; |
81 | signed char fastreuseport; | |
82 | kuid_t fastuid; | |
637bc8bb JB |
83 | #if IS_ENABLED(CONFIG_IPV6) |
84 | struct in6_addr fast_v6_rcv_saddr; | |
85 | #endif | |
86 | __be32 fast_rcv_saddr; | |
87 | unsigned short fast_sk_family; | |
88 | bool fast_ipv6_only; | |
77d8bf9c ACM |
89 | struct hlist_node node; |
90 | struct hlist_head owners; | |
91 | }; | |
92 | ||
7a9546ee ED |
93 | static inline struct net *ib_net(struct inet_bind_bucket *ib) |
94 | { | |
95 | return read_pnet(&ib->ib_net); | |
96 | } | |
97 | ||
b67bfe0d SL |
98 | #define inet_bind_bucket_for_each(tb, head) \ |
99 | hlist_for_each_entry(tb, head, node) | |
77d8bf9c ACM |
100 | |
101 | struct inet_bind_hashbucket { | |
102 | spinlock_t lock; | |
103 | struct hlist_head chain; | |
104 | }; | |
105 | ||
c25eb3bf ED |
106 | /* |
107 | * Sockets can be hashed in established or listening table | |
c25eb3bf | 108 | */ |
5caea4ea ED |
109 | struct inet_listen_hashbucket { |
110 | spinlock_t lock; | |
76d013b2 | 111 | unsigned int count; |
3b24d854 | 112 | struct hlist_head head; |
5caea4ea ED |
113 | }; |
114 | ||
77d8bf9c ACM |
115 | /* This is for listening sockets, thus all sockets which possess wildcards. */ |
116 | #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */ | |
117 | ||
118 | struct inet_hashinfo { | |
119 | /* This is for sockets with full identity only. Sockets here will | |
120 | * always be without wildcards and will have the following invariant: | |
121 | * | |
122 | * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE | |
123 | * | |
77d8bf9c ACM |
124 | */ |
125 | struct inet_ehash_bucket *ehash; | |
9db66bdc | 126 | spinlock_t *ehash_locks; |
f373b53b | 127 | unsigned int ehash_mask; |
230140cf | 128 | unsigned int ehash_locks_mask; |
77d8bf9c ACM |
129 | |
130 | /* Ok, let's try this, I give up, we do need a local binding | |
131 | * TCP hash as well as the others for fast bind/connect. | |
132 | */ | |
61b7c691 | 133 | struct kmem_cache *bind_bucket_cachep; |
77d8bf9c | 134 | struct inet_bind_hashbucket *bhash; |
cfcabdcc | 135 | unsigned int bhash_size; |
77d8bf9c | 136 | |
61b7c691 MKL |
137 | /* The 2nd listener table hashed by local port and address */ |
138 | unsigned int lhash2_mask; | |
139 | struct inet_listen_hashbucket *lhash2; | |
77d8bf9c ACM |
140 | |
141 | /* All the above members are written once at bootup and | |
142 | * never written again _or_ are predominantly read-access. | |
143 | * | |
144 | * Now align to a new cache line as all the following members | |
5caea4ea ED |
145 | * might be often dirty. |
146 | */ | |
61b7c691 MKL |
147 | /* All sockets in TCP_LISTEN state will be in listening_hash. |
148 | * This is the only table where wildcard'd TCP sockets can | |
149 | * exist. listening_hash is only hashed by local port number. | |
150 | * If lhash2 is initialized, the same socket will also be hashed | |
151 | * to lhash2 by port and address. | |
77d8bf9c | 152 | */ |
5caea4ea ED |
153 | struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE] |
154 | ____cacheline_aligned_in_smp; | |
77d8bf9c ACM |
155 | }; |
156 | ||
61b7c691 MKL |
157 | #define inet_lhash2_for_each_icsk_rcu(__icsk, list) \ |
158 | hlist_for_each_entry_rcu(__icsk, list, icsk_listen_portaddr_node) | |
159 | ||
160 | static inline struct inet_listen_hashbucket * | |
161 | inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash) | |
162 | { | |
163 | return &h->lhash2[hash & h->lhash2_mask]; | |
164 | } | |
165 | ||
81c3d547 ED |
166 | static inline struct inet_ehash_bucket *inet_ehash_bucket( |
167 | struct inet_hashinfo *hashinfo, | |
168 | unsigned int hash) | |
169 | { | |
f373b53b | 170 | return &hashinfo->ehash[hash & hashinfo->ehash_mask]; |
304a1618 ACM |
171 | } |
172 | ||
9db66bdc | 173 | static inline spinlock_t *inet_ehash_lockp( |
230140cf ED |
174 | struct inet_hashinfo *hashinfo, |
175 | unsigned int hash) | |
176 | { | |
177 | return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask]; | |
178 | } | |
179 | ||
095dc8e0 | 180 | int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo); |
230140cf ED |
181 | |
182 | static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo) | |
183 | { | |
095dc8e0 ED |
184 | kvfree(hashinfo->ehash_locks); |
185 | hashinfo->ehash_locks = NULL; | |
230140cf ED |
186 | } |
187 | ||
e7819058 MM |
188 | static inline bool inet_sk_bound_dev_eq(struct net *net, int bound_dev_if, |
189 | int dif, int sdif) | |
190 | { | |
191 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) | |
192 | return inet_bound_dev_eq(!!net->ipv4.sysctl_tcp_l3mdev_accept, | |
193 | bound_dev_if, dif, sdif); | |
194 | #else | |
195 | return inet_bound_dev_eq(true, bound_dev_if, dif, sdif); | |
196 | #endif | |
197 | } | |
198 | ||
1fd51155 JP |
199 | struct inet_bind_bucket * |
200 | inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net, | |
201 | struct inet_bind_hashbucket *head, | |
3c82a21f | 202 | const unsigned short snum, int l3mdev); |
1fd51155 JP |
203 | void inet_bind_bucket_destroy(struct kmem_cache *cachep, |
204 | struct inet_bind_bucket *tb); | |
205 | ||
6eada011 ED |
206 | static inline u32 inet_bhashfn(const struct net *net, const __u16 lport, |
207 | const u32 bhash_size) | |
77d8bf9c | 208 | { |
0b441916 | 209 | return (lport + net_hash_mix(net)) & (bhash_size - 1); |
77d8bf9c ACM |
210 | } |
211 | ||
1fd51155 JP |
212 | void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, |
213 | const unsigned short snum); | |
2d8c4ce5 | 214 | |
77d8bf9c | 215 | /* These can have wildcards, don't try too hard. */ |
6eada011 | 216 | static inline u32 inet_lhashfn(const struct net *net, const unsigned short num) |
77d8bf9c | 217 | { |
0b441916 | 218 | return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1); |
77d8bf9c ACM |
219 | } |
220 | ||
221 | static inline int inet_sk_listen_hashfn(const struct sock *sk) | |
222 | { | |
c720c7e8 | 223 | return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num); |
77d8bf9c ACM |
224 | } |
225 | ||
2d8c4ce5 | 226 | /* Caller must disable local BH processing. */ |
1ce31c9e | 227 | int __inet_inherit_port(const struct sock *sk, struct sock *child); |
2d8c4ce5 | 228 | |
1fd51155 | 229 | void inet_put_port(struct sock *sk); |
2d8c4ce5 | 230 | |
5caea4ea | 231 | void inet_hashinfo_init(struct inet_hashinfo *h); |
61b7c691 MKL |
232 | void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name, |
233 | unsigned long numentries, int scale, | |
234 | unsigned long low_limit, | |
235 | unsigned long high_limit); | |
c92c81df | 236 | int inet_hashinfo2_init_mod(struct inet_hashinfo *h); |
f3f05f70 | 237 | |
5e0724d0 ED |
238 | bool inet_ehash_insert(struct sock *sk, struct sock *osk); |
239 | bool inet_ehash_nolisten(struct sock *sk, struct sock *osk); | |
fe38d2a1 | 240 | int __inet_hash(struct sock *sk, struct sock *osk); |
086c653f | 241 | int inet_hash(struct sock *sk); |
1fd51155 | 242 | void inet_unhash(struct sock *sk); |
33b62231 | 243 | |
1fd51155 JP |
244 | struct sock *__inet_lookup_listener(struct net *net, |
245 | struct inet_hashinfo *hashinfo, | |
a583636a | 246 | struct sk_buff *skb, int doff, |
1fd51155 JP |
247 | const __be32 saddr, const __be16 sport, |
248 | const __be32 daddr, | |
249 | const unsigned short hnum, | |
3fa6f616 | 250 | const int dif, const int sdif); |
8f491069 | 251 | |
c67499c0 PE |
252 | static inline struct sock *inet_lookup_listener(struct net *net, |
253 | struct inet_hashinfo *hashinfo, | |
a583636a | 254 | struct sk_buff *skb, int doff, |
da5e3630 | 255 | __be32 saddr, __be16 sport, |
3fa6f616 | 256 | __be32 daddr, __be16 dport, int dif, int sdif) |
8f491069 | 257 | { |
a583636a | 258 | return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport, |
3fa6f616 | 259 | daddr, ntohs(dport), dif, sdif); |
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 | */ | |
8feaf0c0 ACM |
271 | #ifdef __BIG_ENDIAN |
272 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 273 | ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport))) |
8feaf0c0 ACM |
274 | #else /* __LITTLE_ENDIAN */ |
275 | #define INET_COMBINED_PORTS(__sport, __dport) \ | |
4f765d84 | 276 | ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport))) |
8feaf0c0 ACM |
277 | #endif |
278 | ||
279 | #if (BITS_PER_LONG == 64) | |
280 | #ifdef __BIG_ENDIAN | |
281 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
282 | const __addrpair __name = (__force __addrpair) ( \ |
283 | (((__force __u64)(__be32)(__saddr)) << 32) | \ | |
c7228317 | 284 | ((__force __u64)(__be32)(__daddr))) |
8feaf0c0 ACM |
285 | #else /* __LITTLE_ENDIAN */ |
286 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ | |
4f765d84 AV |
287 | const __addrpair __name = (__force __addrpair) ( \ |
288 | (((__force __u64)(__be32)(__daddr)) << 32) | \ | |
c7228317 | 289 | ((__force __u64)(__be32)(__saddr))) |
8feaf0c0 | 290 | #endif /* __BIG_ENDIAN */ |
3fa6f616 | 291 | #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \ |
50805466 ED |
292 | (((__sk)->sk_portpair == (__ports)) && \ |
293 | ((__sk)->sk_addrpair == (__cookie)) && \ | |
3c82a21f RS |
294 | (((__sk)->sk_bound_dev_if == (__dif)) || \ |
295 | ((__sk)->sk_bound_dev_if == (__sdif))) && \ | |
ce43b03e | 296 | net_eq(sock_net(__sk), (__net))) |
8feaf0c0 | 297 | #else /* 32-bit arch */ |
c7228317 JP |
298 | #define INET_ADDR_COOKIE(__name, __saddr, __daddr) \ |
299 | const int __name __deprecated __attribute__((unused)) | |
300 | ||
3fa6f616 | 301 | #define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \ |
50805466 ED |
302 | (((__sk)->sk_portpair == (__ports)) && \ |
303 | ((__sk)->sk_daddr == (__saddr)) && \ | |
304 | ((__sk)->sk_rcv_saddr == (__daddr)) && \ | |
3c82a21f RS |
305 | (((__sk)->sk_bound_dev_if == (__dif)) || \ |
306 | ((__sk)->sk_bound_dev_if == (__sdif))) && \ | |
ce43b03e | 307 | net_eq(sock_net(__sk), (__net))) |
8feaf0c0 | 308 | #endif /* 64-bit arch */ |
e48c414e | 309 | |
ee3cf32a | 310 | /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need |
e48c414e | 311 | * not check it for lookups anymore, thanks Alexey. -DaveM |
e48c414e | 312 | */ |
1fd51155 JP |
313 | struct sock *__inet_lookup_established(struct net *net, |
314 | struct inet_hashinfo *hashinfo, | |
315 | const __be32 saddr, const __be16 sport, | |
316 | const __be32 daddr, const u16 hnum, | |
3fa6f616 | 317 | const int dif, const int sdif); |
e48c414e | 318 | |
8f491069 | 319 | static inline struct sock * |
c67499c0 | 320 | inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo, |
fb99c848 AV |
321 | const __be32 saddr, const __be16 sport, |
322 | const __be32 daddr, const __be16 dport, | |
8f491069 HX |
323 | const int dif) |
324 | { | |
c67499c0 | 325 | return __inet_lookup_established(net, hashinfo, saddr, sport, daddr, |
3fa6f616 | 326 | ntohs(dport), dif, 0); |
8f491069 HX |
327 | } |
328 | ||
c67499c0 PE |
329 | static inline struct sock *__inet_lookup(struct net *net, |
330 | struct inet_hashinfo *hashinfo, | |
a583636a | 331 | struct sk_buff *skb, int doff, |
fb99c848 AV |
332 | const __be32 saddr, const __be16 sport, |
333 | const __be32 daddr, const __be16 dport, | |
3fa6f616 | 334 | const int dif, const int sdif, |
3b24d854 | 335 | bool *refcounted) |
e48c414e | 336 | { |
8f491069 | 337 | u16 hnum = ntohs(dport); |
3b24d854 | 338 | struct sock *sk; |
c67499c0 | 339 | |
3b24d854 | 340 | sk = __inet_lookup_established(net, hashinfo, saddr, sport, |
3fa6f616 | 341 | daddr, hnum, dif, sdif); |
3b24d854 ED |
342 | *refcounted = true; |
343 | if (sk) | |
344 | return sk; | |
345 | *refcounted = false; | |
346 | return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, | |
3fa6f616 | 347 | sport, daddr, hnum, dif, sdif); |
e48c414e ACM |
348 | } |
349 | ||
c67499c0 PE |
350 | static inline struct sock *inet_lookup(struct net *net, |
351 | struct inet_hashinfo *hashinfo, | |
a583636a | 352 | struct sk_buff *skb, int doff, |
fb99c848 AV |
353 | const __be32 saddr, const __be16 sport, |
354 | const __be32 daddr, const __be16 dport, | |
e48c414e ACM |
355 | const int dif) |
356 | { | |
357 | struct sock *sk; | |
3b24d854 | 358 | bool refcounted; |
e48c414e | 359 | |
a583636a | 360 | sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr, |
3fa6f616 | 361 | dport, dif, 0, &refcounted); |
e48c414e | 362 | |
41c6d650 | 363 | if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt)) |
3b24d854 | 364 | sk = NULL; |
e48c414e ACM |
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, | |
a583636a | 370 | int doff, |
9a1f27c4 | 371 | const __be16 sport, |
3b24d854 | 372 | const __be16 dport, |
3fa6f616 | 373 | const int sdif, |
3b24d854 | 374 | bool *refcounted) |
9a1f27c4 | 375 | { |
41063e9d | 376 | struct sock *sk = skb_steal_sock(skb); |
9a1f27c4 ACM |
377 | const struct iphdr *iph = ip_hdr(skb); |
378 | ||
3b24d854 | 379 | *refcounted = true; |
41063e9d | 380 | if (sk) |
23542618 | 381 | return sk; |
3b24d854 ED |
382 | |
383 | return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb, | |
384 | doff, iph->saddr, sport, | |
3fa6f616 | 385 | iph->daddr, dport, inet_iif(skb), sdif, |
3b24d854 | 386 | refcounted); |
9a1f27c4 ACM |
387 | } |
388 | ||
d1e559d0 ED |
389 | u32 inet6_ehashfn(const struct net *net, |
390 | const struct in6_addr *laddr, const u16 lport, | |
391 | const struct in6_addr *faddr, const __be16 fport); | |
392 | ||
393 | static inline void sk_daddr_set(struct sock *sk, __be32 addr) | |
394 | { | |
395 | sk->sk_daddr = addr; /* alias of inet_daddr */ | |
396 | #if IS_ENABLED(CONFIG_IPV6) | |
397 | ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr); | |
398 | #endif | |
399 | } | |
400 | ||
401 | static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr) | |
402 | { | |
403 | sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */ | |
404 | #if IS_ENABLED(CONFIG_IPV6) | |
405 | ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr); | |
406 | #endif | |
407 | } | |
5b441f76 | 408 | |
1fd51155 JP |
409 | int __inet_hash_connect(struct inet_timewait_death_row *death_row, |
410 | struct sock *sk, u32 port_offset, | |
411 | int (*check_established)(struct inet_timewait_death_row *, | |
412 | struct sock *, __u16, | |
b4d6444e | 413 | struct inet_timewait_sock **)); |
9327f705 | 414 | |
1fd51155 JP |
415 | int inet_hash_connect(struct inet_timewait_death_row *death_row, |
416 | struct sock *sk); | |
304a1618 | 417 | #endif /* _INET_HASHTABLES_H */ |