]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - include/net/inet_hashtables.h
soreuseport: Prep for fast reuseport TCP socket selection
[mirror_ubuntu-bionic-kernel.git] / include / net / inet_hashtables.h
CommitLineData
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
60063497 35#include <linux/atomic.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 */
42struct 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 */
77struct inet_bind_bucket {
0c5c9fb5 78 possible_net_t ib_net;
77d8bf9c 79 unsigned short port;
da5e3630
TH
80 signed char fastreuse;
81 signed char fastreuseport;
82 kuid_t fastuid;
a9d8f911 83 int num_owners;
77d8bf9c
ACM
84 struct hlist_node node;
85 struct hlist_head owners;
86};
87
7a9546ee
ED
88static inline struct net *ib_net(struct inet_bind_bucket *ib)
89{
90 return read_pnet(&ib->ib_net);
91}
92
b67bfe0d
SL
93#define inet_bind_bucket_for_each(tb, head) \
94 hlist_for_each_entry(tb, head, node)
77d8bf9c
ACM
95
96struct inet_bind_hashbucket {
97 spinlock_t lock;
98 struct hlist_head chain;
99};
100
c25eb3bf
ED
101/*
102 * Sockets can be hashed in established or listening table
103 * We must use different 'nulls' end-of-chain value for listening
104 * hash table, or we might find a socket that was closed and
105 * reallocated/inserted into established hash table
106 */
107#define LISTENING_NULLS_BASE (1U << 29)
5caea4ea
ED
108struct inet_listen_hashbucket {
109 spinlock_t lock;
c25eb3bf 110 struct hlist_nulls_head head;
5caea4ea
ED
111};
112
77d8bf9c
ACM
113/* This is for listening sockets, thus all sockets which possess wildcards. */
114#define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
115
116struct inet_hashinfo {
117 /* This is for sockets with full identity only. Sockets here will
118 * always be without wildcards and will have the following invariant:
119 *
120 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
121 *
77d8bf9c
ACM
122 */
123 struct inet_ehash_bucket *ehash;
9db66bdc 124 spinlock_t *ehash_locks;
f373b53b 125 unsigned int ehash_mask;
230140cf 126 unsigned int ehash_locks_mask;
77d8bf9c
ACM
127
128 /* Ok, let's try this, I give up, we do need a local binding
129 * TCP hash as well as the others for fast bind/connect.
130 */
131 struct inet_bind_hashbucket *bhash;
132
cfcabdcc 133 unsigned int bhash_size;
24dd1fa1 134 /* 4 bytes hole on 64 bit */
77d8bf9c 135
5caea4ea 136 struct kmem_cache *bind_bucket_cachep;
77d8bf9c
ACM
137
138 /* All the above members are written once at bootup and
139 * never written again _or_ are predominantly read-access.
140 *
141 * Now align to a new cache line as all the following members
5caea4ea
ED
142 * might be often dirty.
143 */
144 /* All sockets in TCP_LISTEN state will be in here. This is the only
145 * table where wildcard'd TCP sockets can exist. Hash function here
146 * is just local port number.
77d8bf9c 147 */
5caea4ea
ED
148 struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE]
149 ____cacheline_aligned_in_smp;
77d8bf9c
ACM
150};
151
81c3d547
ED
152static inline struct inet_ehash_bucket *inet_ehash_bucket(
153 struct inet_hashinfo *hashinfo,
154 unsigned int hash)
155{
f373b53b 156 return &hashinfo->ehash[hash & hashinfo->ehash_mask];
304a1618
ACM
157}
158
9db66bdc 159static inline spinlock_t *inet_ehash_lockp(
230140cf
ED
160 struct inet_hashinfo *hashinfo,
161 unsigned int hash)
162{
163 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
164}
165
095dc8e0 166int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
230140cf
ED
167
168static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
169{
095dc8e0
ED
170 kvfree(hashinfo->ehash_locks);
171 hashinfo->ehash_locks = NULL;
230140cf
ED
172}
173
1fd51155
JP
174struct inet_bind_bucket *
175inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
176 struct inet_bind_hashbucket *head,
177 const unsigned short snum);
178void inet_bind_bucket_destroy(struct kmem_cache *cachep,
179 struct inet_bind_bucket *tb);
180
6eada011
ED
181static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
182 const u32 bhash_size)
77d8bf9c 183{
0b441916 184 return (lport + net_hash_mix(net)) & (bhash_size - 1);
77d8bf9c
ACM
185}
186
1fd51155
JP
187void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
188 const unsigned short snum);
2d8c4ce5 189
77d8bf9c 190/* These can have wildcards, don't try too hard. */
6eada011 191static inline u32 inet_lhashfn(const struct net *net, const unsigned short num)
77d8bf9c 192{
0b441916 193 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
77d8bf9c
ACM
194}
195
196static inline int inet_sk_listen_hashfn(const struct sock *sk)
197{
c720c7e8 198 return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
77d8bf9c
ACM
199}
200
2d8c4ce5 201/* Caller must disable local BH processing. */
1ce31c9e 202int __inet_inherit_port(const struct sock *sk, struct sock *child);
2d8c4ce5 203
1fd51155 204void inet_put_port(struct sock *sk);
2d8c4ce5 205
5caea4ea 206void inet_hashinfo_init(struct inet_hashinfo *h);
f3f05f70 207
5e0724d0
ED
208bool inet_ehash_insert(struct sock *sk, struct sock *osk);
209bool inet_ehash_nolisten(struct sock *sk, struct sock *osk);
fc01538f 210void __inet_hash(struct sock *sk, struct sock *osk);
086c653f 211int inet_hash(struct sock *sk);
1fd51155 212void inet_unhash(struct sock *sk);
33b62231 213
1fd51155
JP
214struct sock *__inet_lookup_listener(struct net *net,
215 struct inet_hashinfo *hashinfo,
a583636a 216 struct sk_buff *skb, int doff,
1fd51155
JP
217 const __be32 saddr, const __be16 sport,
218 const __be32 daddr,
219 const unsigned short hnum,
220 const int dif);
8f491069 221
c67499c0
PE
222static inline struct sock *inet_lookup_listener(struct net *net,
223 struct inet_hashinfo *hashinfo,
a583636a 224 struct sk_buff *skb, int doff,
da5e3630 225 __be32 saddr, __be16 sport,
c67499c0 226 __be32 daddr, __be16 dport, int dif)
8f491069 227{
a583636a 228 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
da5e3630 229 daddr, ntohs(dport), dif);
8f491069 230}
8feaf0c0
ACM
231
232/* Socket demux engine toys. */
4f765d84
AV
233/* What happens here is ugly; there's a pair of adjacent fields in
234 struct inet_sock; __be16 dport followed by __u16 num. We want to
235 search by pair, so we combine the keys into a single 32bit value
236 and compare with 32bit value read from &...->dport. Let's at least
237 make sure that it's not mixed with anything else...
238 On 64bit targets we combine comparisons with pair of adjacent __be32
239 fields in the same way.
240*/
8feaf0c0
ACM
241#ifdef __BIG_ENDIAN
242#define INET_COMBINED_PORTS(__sport, __dport) \
4f765d84 243 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
8feaf0c0
ACM
244#else /* __LITTLE_ENDIAN */
245#define INET_COMBINED_PORTS(__sport, __dport) \
4f765d84 246 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
8feaf0c0
ACM
247#endif
248
249#if (BITS_PER_LONG == 64)
250#ifdef __BIG_ENDIAN
251#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
4f765d84
AV
252 const __addrpair __name = (__force __addrpair) ( \
253 (((__force __u64)(__be32)(__saddr)) << 32) | \
c7228317 254 ((__force __u64)(__be32)(__daddr)))
8feaf0c0
ACM
255#else /* __LITTLE_ENDIAN */
256#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
4f765d84
AV
257 const __addrpair __name = (__force __addrpair) ( \
258 (((__force __u64)(__be32)(__daddr)) << 32) | \
c7228317 259 ((__force __u64)(__be32)(__saddr)))
8feaf0c0 260#endif /* __BIG_ENDIAN */
ce43b03e 261#define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
50805466
ED
262 (((__sk)->sk_portpair == (__ports)) && \
263 ((__sk)->sk_addrpair == (__cookie)) && \
ce43b03e
ED
264 (!(__sk)->sk_bound_dev_if || \
265 ((__sk)->sk_bound_dev_if == (__dif))) && \
266 net_eq(sock_net(__sk), (__net)))
8feaf0c0 267#else /* 32-bit arch */
c7228317
JP
268#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
269 const int __name __deprecated __attribute__((unused))
270
ce43b03e 271#define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif) \
50805466
ED
272 (((__sk)->sk_portpair == (__ports)) && \
273 ((__sk)->sk_daddr == (__saddr)) && \
274 ((__sk)->sk_rcv_saddr == (__daddr)) && \
ce43b03e
ED
275 (!(__sk)->sk_bound_dev_if || \
276 ((__sk)->sk_bound_dev_if == (__dif))) && \
277 net_eq(sock_net(__sk), (__net)))
8feaf0c0 278#endif /* 64-bit arch */
e48c414e
ACM
279
280/*
281 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
282 * not check it for lookups anymore, thanks Alexey. -DaveM
283 *
284 * Local BH must be disabled here.
285 */
1fd51155
JP
286struct sock *__inet_lookup_established(struct net *net,
287 struct inet_hashinfo *hashinfo,
288 const __be32 saddr, const __be16 sport,
289 const __be32 daddr, const u16 hnum,
290 const int dif);
e48c414e 291
8f491069 292static inline struct sock *
c67499c0 293 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
fb99c848
AV
294 const __be32 saddr, const __be16 sport,
295 const __be32 daddr, const __be16 dport,
8f491069
HX
296 const int dif)
297{
c67499c0 298 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
8f491069
HX
299 ntohs(dport), dif);
300}
301
c67499c0
PE
302static inline struct sock *__inet_lookup(struct net *net,
303 struct inet_hashinfo *hashinfo,
a583636a 304 struct sk_buff *skb, int doff,
fb99c848
AV
305 const __be32 saddr, const __be16 sport,
306 const __be32 daddr, const __be16 dport,
e48c414e
ACM
307 const int dif)
308{
8f491069 309 u16 hnum = ntohs(dport);
c67499c0
PE
310 struct sock *sk = __inet_lookup_established(net, hashinfo,
311 saddr, sport, daddr, hnum, dif);
312
a583636a
CG
313 return sk ? : __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
314 sport, daddr, hnum, dif);
e48c414e
ACM
315}
316
c67499c0
PE
317static inline struct sock *inet_lookup(struct net *net,
318 struct inet_hashinfo *hashinfo,
a583636a 319 struct sk_buff *skb, int doff,
fb99c848
AV
320 const __be32 saddr, const __be16 sport,
321 const __be32 daddr, const __be16 dport,
e48c414e
ACM
322 const int dif)
323{
324 struct sock *sk;
325
326 local_bh_disable();
a583636a
CG
327 sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
328 dport, dif);
e48c414e
ACM
329 local_bh_enable();
330
331 return sk;
332}
a7f5e7f1 333
9a1f27c4
ACM
334static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
335 struct sk_buff *skb,
a583636a 336 int doff,
9a1f27c4
ACM
337 const __be16 sport,
338 const __be16 dport)
339{
41063e9d 340 struct sock *sk = skb_steal_sock(skb);
9a1f27c4
ACM
341 const struct iphdr *iph = ip_hdr(skb);
342
41063e9d 343 if (sk)
23542618
KK
344 return sk;
345 else
a583636a
CG
346 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
347 doff, iph->saddr, sport,
23542618 348 iph->daddr, dport, inet_iif(skb));
9a1f27c4
ACM
349}
350
5b441f76 351u32 sk_ehashfn(const struct sock *sk);
d1e559d0
ED
352u32 inet6_ehashfn(const struct net *net,
353 const struct in6_addr *laddr, const u16 lport,
354 const struct in6_addr *faddr, const __be16 fport);
355
356static inline void sk_daddr_set(struct sock *sk, __be32 addr)
357{
358 sk->sk_daddr = addr; /* alias of inet_daddr */
359#if IS_ENABLED(CONFIG_IPV6)
360 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
361#endif
362}
363
364static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
365{
366 sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
367#if IS_ENABLED(CONFIG_IPV6)
368 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
369#endif
370}
5b441f76 371
1fd51155
JP
372int __inet_hash_connect(struct inet_timewait_death_row *death_row,
373 struct sock *sk, u32 port_offset,
374 int (*check_established)(struct inet_timewait_death_row *,
375 struct sock *, __u16,
b4d6444e 376 struct inet_timewait_sock **));
9327f705 377
1fd51155
JP
378int inet_hash_connect(struct inet_timewait_death_row *death_row,
379 struct sock *sk);
304a1618 380#endif /* _INET_HASHTABLES_H */