Loading...
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
17
18#include <linux/interrupt.h>
19#include <linux/ip.h>
20#include <linux/ipv6.h>
21#include <linux/list.h>
22#include <linux/slab.h>
23#include <linux/socket.h>
24#include <linux/spinlock.h>
25#include <linux/types.h>
26#include <linux/wait.h>
27#include <linux/vmalloc.h>
28
29#include <net/inet_connection_sock.h>
30#include <net/inet_sock.h>
31#include <net/sock.h>
32#include <net/route.h>
33#include <net/tcp_states.h>
34#include <net/netns/hash.h>
35
36#include <linux/atomic.h>
37#include <asm/byteorder.h>
38
39/* This is for all connections with a full identity, no wildcards.
40 * One chain is dedicated to TIME_WAIT sockets.
41 * I'll experiment with dynamic table growth later.
42 */
43struct inet_ehash_bucket {
44 struct hlist_nulls_head chain;
45 struct hlist_nulls_head twchain;
46};
47
48/* There are a few simple rules, which allow for local port reuse by
49 * an application. In essence:
50 *
51 * 1) Sockets bound to different interfaces may share a local port.
52 * Failing that, goto test 2.
53 * 2) If all sockets have sk->sk_reuse set, and none of them are in
54 * TCP_LISTEN state, the port may be shared.
55 * Failing that, goto test 3.
56 * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local
57 * address, and none of them are the same, the port may be
58 * shared.
59 * Failing this, the port cannot be shared.
60 *
61 * The interesting point, is test #2. This is what an FTP server does
62 * all day. To optimize this case we use a specific flag bit defined
63 * below. As we add sockets to a bind bucket list, we perform a
64 * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN))
65 * As long as all sockets added to a bind bucket pass this test,
66 * the flag bit will be set.
67 * The resulting situation is that tcp_v[46]_verify_bind() can just check
68 * for this flag bit, if it is set and the socket trying to bind has
69 * sk->sk_reuse set, we don't even have to walk the owners list at all,
70 * we return that it is ok to bind this socket to the requested local port.
71 *
72 * Sounds like a lot of work, but it is worth it. In a more naive
73 * implementation (ie. current FreeBSD etc.) the entire list of ports
74 * must be walked for each data port opened by an ftp server. Needless
75 * to say, this does not scale at all. With a couple thousand FTP
76 * users logged onto your box, isn't it nice to know that new data
77 * ports are created in O(1) time? I thought so. ;-) -DaveM
78 */
79struct inet_bind_bucket {
80#ifdef CONFIG_NET_NS
81 struct net *ib_net;
82#endif
83 unsigned short port;
84 signed short fastreuse;
85 int num_owners;
86 struct hlist_node node;
87 struct hlist_head owners;
88};
89
90static inline struct net *ib_net(struct inet_bind_bucket *ib)
91{
92 return read_pnet(&ib->ib_net);
93}
94
95#define inet_bind_bucket_for_each(tb, pos, head) \
96 hlist_for_each_entry(tb, pos, head, node)
97
98struct inet_bind_hashbucket {
99 spinlock_t lock;
100 struct hlist_head chain;
101};
102
103/*
104 * Sockets can be hashed in established or listening table
105 * We must use different 'nulls' end-of-chain value for listening
106 * hash table, or we might find a socket that was closed and
107 * reallocated/inserted into established hash table
108 */
109#define LISTENING_NULLS_BASE (1U << 29)
110struct inet_listen_hashbucket {
111 spinlock_t lock;
112 struct hlist_nulls_head head;
113};
114
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
118struct 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 *
124 * TIME_WAIT sockets use a separate chain (twchain).
125 */
126 struct inet_ehash_bucket *ehash;
127 spinlock_t *ehash_locks;
128 unsigned int ehash_mask;
129 unsigned int ehash_locks_mask;
130
131 /* Ok, let's try this, I give up, we do need a local binding
132 * TCP hash as well as the others for fast bind/connect.
133 */
134 struct inet_bind_hashbucket *bhash;
135
136 unsigned int bhash_size;
137 /* 4 bytes hole on 64 bit */
138
139 struct kmem_cache *bind_bucket_cachep;
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
145 * might be often dirty.
146 */
147 /* All sockets in TCP_LISTEN state will be in here. This is the only
148 * table where wildcard'd TCP sockets can exist. Hash function here
149 * is just local port number.
150 */
151 struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE]
152 ____cacheline_aligned_in_smp;
153
154 atomic_t bsockets;
155};
156
157static inline struct inet_ehash_bucket *inet_ehash_bucket(
158 struct inet_hashinfo *hashinfo,
159 unsigned int hash)
160{
161 return &hashinfo->ehash[hash & hashinfo->ehash_mask];
162}
163
164static inline spinlock_t *inet_ehash_lockp(
165 struct inet_hashinfo *hashinfo,
166 unsigned int hash)
167{
168 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
169}
170
171static inline int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
172{
173 unsigned int i, size = 256;
174#if defined(CONFIG_PROVE_LOCKING)
175 unsigned int nr_pcpus = 2;
176#else
177 unsigned int nr_pcpus = num_possible_cpus();
178#endif
179 if (nr_pcpus >= 4)
180 size = 512;
181 if (nr_pcpus >= 8)
182 size = 1024;
183 if (nr_pcpus >= 16)
184 size = 2048;
185 if (nr_pcpus >= 32)
186 size = 4096;
187 if (sizeof(spinlock_t) != 0) {
188#ifdef CONFIG_NUMA
189 if (size * sizeof(spinlock_t) > PAGE_SIZE)
190 hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t));
191 else
192#endif
193 hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t),
194 GFP_KERNEL);
195 if (!hashinfo->ehash_locks)
196 return ENOMEM;
197 for (i = 0; i < size; i++)
198 spin_lock_init(&hashinfo->ehash_locks[i]);
199 }
200 hashinfo->ehash_locks_mask = size - 1;
201 return 0;
202}
203
204static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
205{
206 if (hashinfo->ehash_locks) {
207#ifdef CONFIG_NUMA
208 unsigned int size = (hashinfo->ehash_locks_mask + 1) *
209 sizeof(spinlock_t);
210 if (size > PAGE_SIZE)
211 vfree(hashinfo->ehash_locks);
212 else
213#endif
214 kfree(hashinfo->ehash_locks);
215 hashinfo->ehash_locks = NULL;
216 }
217}
218
219extern struct inet_bind_bucket *
220 inet_bind_bucket_create(struct kmem_cache *cachep,
221 struct net *net,
222 struct inet_bind_hashbucket *head,
223 const unsigned short snum);
224extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
225 struct inet_bind_bucket *tb);
226
227static inline int inet_bhashfn(struct net *net,
228 const __u16 lport, const int bhash_size)
229{
230 return (lport + net_hash_mix(net)) & (bhash_size - 1);
231}
232
233extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
234 const unsigned short snum);
235
236/* These can have wildcards, don't try too hard. */
237static inline int inet_lhashfn(struct net *net, const unsigned short num)
238{
239 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
240}
241
242static inline int inet_sk_listen_hashfn(const struct sock *sk)
243{
244 return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
245}
246
247/* Caller must disable local BH processing. */
248extern int __inet_inherit_port(struct sock *sk, struct sock *child);
249
250extern void inet_put_port(struct sock *sk);
251
252void inet_hashinfo_init(struct inet_hashinfo *h);
253
254extern int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw);
255extern void inet_hash(struct sock *sk);
256extern void inet_unhash(struct sock *sk);
257
258extern struct sock *__inet_lookup_listener(struct net *net,
259 struct inet_hashinfo *hashinfo,
260 const __be32 daddr,
261 const unsigned short hnum,
262 const int dif);
263
264static inline struct sock *inet_lookup_listener(struct net *net,
265 struct inet_hashinfo *hashinfo,
266 __be32 daddr, __be16 dport, int dif)
267{
268 return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif);
269}
270
271/* Socket demux engine toys. */
272/* What happens here is ugly; there's a pair of adjacent fields in
273 struct inet_sock; __be16 dport followed by __u16 num. We want to
274 search by pair, so we combine the keys into a single 32bit value
275 and compare with 32bit value read from &...->dport. Let's at least
276 make sure that it's not mixed with anything else...
277 On 64bit targets we combine comparisons with pair of adjacent __be32
278 fields in the same way.
279*/
280typedef __u32 __bitwise __portpair;
281#ifdef __BIG_ENDIAN
282#define INET_COMBINED_PORTS(__sport, __dport) \
283 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
284#else /* __LITTLE_ENDIAN */
285#define INET_COMBINED_PORTS(__sport, __dport) \
286 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
287#endif
288
289#if (BITS_PER_LONG == 64)
290typedef __u64 __bitwise __addrpair;
291#ifdef __BIG_ENDIAN
292#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
293 const __addrpair __name = (__force __addrpair) ( \
294 (((__force __u64)(__be32)(__saddr)) << 32) | \
295 ((__force __u64)(__be32)(__daddr)));
296#else /* __LITTLE_ENDIAN */
297#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
298 const __addrpair __name = (__force __addrpair) ( \
299 (((__force __u64)(__be32)(__daddr)) << 32) | \
300 ((__force __u64)(__be32)(__saddr)));
301#endif /* __BIG_ENDIAN */
302#define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
303 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
304 ((*((__addrpair *)&(inet_sk(__sk)->inet_daddr))) == (__cookie)) && \
305 ((*((__portpair *)&(inet_sk(__sk)->inet_dport))) == (__ports)) && \
306 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
307#define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
308 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
309 ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \
310 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
311 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
312#else /* 32-bit arch */
313#define INET_ADDR_COOKIE(__name, __saddr, __daddr)
314#define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif) \
315 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
316 (inet_sk(__sk)->inet_daddr == (__saddr)) && \
317 (inet_sk(__sk)->inet_rcv_saddr == (__daddr)) && \
318 ((*((__portpair *)&(inet_sk(__sk)->inet_dport))) == (__ports)) && \
319 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
320#define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif) \
321 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
322 (inet_twsk(__sk)->tw_daddr == (__saddr)) && \
323 (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
324 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
325 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
326#endif /* 64-bit arch */
327
328/*
329 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
330 * not check it for lookups anymore, thanks Alexey. -DaveM
331 *
332 * Local BH must be disabled here.
333 */
334extern struct sock * __inet_lookup_established(struct net *net,
335 struct inet_hashinfo *hashinfo,
336 const __be32 saddr, const __be16 sport,
337 const __be32 daddr, const u16 hnum, const int dif);
338
339static inline struct sock *
340 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
341 const __be32 saddr, const __be16 sport,
342 const __be32 daddr, const __be16 dport,
343 const int dif)
344{
345 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
346 ntohs(dport), dif);
347}
348
349static inline struct sock *__inet_lookup(struct net *net,
350 struct inet_hashinfo *hashinfo,
351 const __be32 saddr, const __be16 sport,
352 const __be32 daddr, const __be16 dport,
353 const int dif)
354{
355 u16 hnum = ntohs(dport);
356 struct sock *sk = __inet_lookup_established(net, hashinfo,
357 saddr, sport, daddr, hnum, dif);
358
359 return sk ? : __inet_lookup_listener(net, hashinfo, daddr, hnum, dif);
360}
361
362static inline struct sock *inet_lookup(struct net *net,
363 struct inet_hashinfo *hashinfo,
364 const __be32 saddr, const __be16 sport,
365 const __be32 daddr, const __be16 dport,
366 const int dif)
367{
368 struct sock *sk;
369
370 local_bh_disable();
371 sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
372 local_bh_enable();
373
374 return sk;
375}
376
377static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
378 struct sk_buff *skb,
379 const __be16 sport,
380 const __be16 dport)
381{
382 struct sock *sk;
383 const struct iphdr *iph = ip_hdr(skb);
384
385 if (unlikely(sk = skb_steal_sock(skb)))
386 return sk;
387 else
388 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo,
389 iph->saddr, sport,
390 iph->daddr, dport, inet_iif(skb));
391}
392
393extern int __inet_hash_connect(struct inet_timewait_death_row *death_row,
394 struct sock *sk,
395 u32 port_offset,
396 int (*check_established)(struct inet_timewait_death_row *,
397 struct sock *, __u16, struct inet_timewait_sock **),
398 int (*hash)(struct sock *sk, struct inet_timewait_sock *twp));
399
400extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
401 struct sock *sk);
402#endif /* _INET_HASHTABLES_H */
1/* SPDX-License-Identifier: GPL-2.0-or-later */
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
8 */
9
10#ifndef _INET_HASHTABLES_H
11#define _INET_HASHTABLES_H
12
13
14#include <linux/interrupt.h>
15#include <linux/ip.h>
16#include <linux/ipv6.h>
17#include <linux/list.h>
18#include <linux/slab.h>
19#include <linux/socket.h>
20#include <linux/spinlock.h>
21#include <linux/types.h>
22#include <linux/wait.h>
23
24#include <net/inet_connection_sock.h>
25#include <net/inet_sock.h>
26#include <net/sock.h>
27#include <net/route.h>
28#include <net/tcp_states.h>
29#include <net/netns/hash.h>
30
31#include <linux/refcount.h>
32#include <asm/byteorder.h>
33
34/* This is for all connections with a full identity, no wildcards.
35 * The 'e' prefix stands for Establish, but we really put all sockets
36 * but LISTEN ones.
37 */
38struct inet_ehash_bucket {
39 struct hlist_nulls_head chain;
40};
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 */
73#define FASTREUSEPORT_ANY 1
74#define FASTREUSEPORT_STRICT 2
75
76struct inet_bind_bucket {
77 possible_net_t ib_net;
78 int l3mdev;
79 unsigned short port;
80 signed char fastreuse;
81 signed char fastreuseport;
82 kuid_t fastuid;
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;
89 struct hlist_node node;
90 struct hlist_head owners;
91};
92
93static inline struct net *ib_net(struct inet_bind_bucket *ib)
94{
95 return read_pnet(&ib->ib_net);
96}
97
98#define inet_bind_bucket_for_each(tb, head) \
99 hlist_for_each_entry(tb, head, node)
100
101struct inet_bind_hashbucket {
102 spinlock_t lock;
103 struct hlist_head chain;
104};
105
106/* Sockets can be hashed in established or listening table.
107 * We must use different 'nulls' end-of-chain value for all hash buckets :
108 * A socket might transition from ESTABLISH to LISTEN state without
109 * RCU grace period. A lookup in ehash table needs to handle this case.
110 */
111#define LISTENING_NULLS_BASE (1U << 29)
112struct inet_listen_hashbucket {
113 spinlock_t lock;
114 unsigned int count;
115 union {
116 struct hlist_head head;
117 struct hlist_nulls_head nulls_head;
118 };
119};
120
121/* This is for listening sockets, thus all sockets which possess wildcards. */
122#define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */
123
124struct inet_hashinfo {
125 /* This is for sockets with full identity only. Sockets here will
126 * always be without wildcards and will have the following invariant:
127 *
128 * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE
129 *
130 */
131 struct inet_ehash_bucket *ehash;
132 spinlock_t *ehash_locks;
133 unsigned int ehash_mask;
134 unsigned int ehash_locks_mask;
135
136 /* Ok, let's try this, I give up, we do need a local binding
137 * TCP hash as well as the others for fast bind/connect.
138 */
139 struct kmem_cache *bind_bucket_cachep;
140 struct inet_bind_hashbucket *bhash;
141 unsigned int bhash_size;
142
143 /* The 2nd listener table hashed by local port and address */
144 unsigned int lhash2_mask;
145 struct inet_listen_hashbucket *lhash2;
146
147 /* All the above members are written once at bootup and
148 * never written again _or_ are predominantly read-access.
149 *
150 * Now align to a new cache line as all the following members
151 * might be often dirty.
152 */
153 /* All sockets in TCP_LISTEN state will be in listening_hash.
154 * This is the only table where wildcard'd TCP sockets can
155 * exist. listening_hash is only hashed by local port number.
156 * If lhash2 is initialized, the same socket will also be hashed
157 * to lhash2 by port and address.
158 */
159 struct inet_listen_hashbucket listening_hash[INET_LHTABLE_SIZE]
160 ____cacheline_aligned_in_smp;
161};
162
163#define inet_lhash2_for_each_icsk_rcu(__icsk, list) \
164 hlist_for_each_entry_rcu(__icsk, list, icsk_listen_portaddr_node)
165
166static inline struct inet_listen_hashbucket *
167inet_lhash2_bucket(struct inet_hashinfo *h, u32 hash)
168{
169 return &h->lhash2[hash & h->lhash2_mask];
170}
171
172static inline struct inet_ehash_bucket *inet_ehash_bucket(
173 struct inet_hashinfo *hashinfo,
174 unsigned int hash)
175{
176 return &hashinfo->ehash[hash & hashinfo->ehash_mask];
177}
178
179static inline spinlock_t *inet_ehash_lockp(
180 struct inet_hashinfo *hashinfo,
181 unsigned int hash)
182{
183 return &hashinfo->ehash_locks[hash & hashinfo->ehash_locks_mask];
184}
185
186int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo);
187
188static inline void inet_hashinfo2_free_mod(struct inet_hashinfo *h)
189{
190 kfree(h->lhash2);
191 h->lhash2 = NULL;
192}
193
194static inline void inet_ehash_locks_free(struct inet_hashinfo *hashinfo)
195{
196 kvfree(hashinfo->ehash_locks);
197 hashinfo->ehash_locks = NULL;
198}
199
200static inline bool inet_sk_bound_dev_eq(struct net *net, int bound_dev_if,
201 int dif, int sdif)
202{
203#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
204 return inet_bound_dev_eq(!!net->ipv4.sysctl_tcp_l3mdev_accept,
205 bound_dev_if, dif, sdif);
206#else
207 return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
208#endif
209}
210
211struct inet_bind_bucket *
212inet_bind_bucket_create(struct kmem_cache *cachep, struct net *net,
213 struct inet_bind_hashbucket *head,
214 const unsigned short snum, int l3mdev);
215void inet_bind_bucket_destroy(struct kmem_cache *cachep,
216 struct inet_bind_bucket *tb);
217
218static inline u32 inet_bhashfn(const struct net *net, const __u16 lport,
219 const u32 bhash_size)
220{
221 return (lport + net_hash_mix(net)) & (bhash_size - 1);
222}
223
224void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
225 const unsigned short snum);
226
227/* These can have wildcards, don't try too hard. */
228static inline u32 inet_lhashfn(const struct net *net, const unsigned short num)
229{
230 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
231}
232
233static inline int inet_sk_listen_hashfn(const struct sock *sk)
234{
235 return inet_lhashfn(sock_net(sk), inet_sk(sk)->inet_num);
236}
237
238/* Caller must disable local BH processing. */
239int __inet_inherit_port(const struct sock *sk, struct sock *child);
240
241void inet_put_port(struct sock *sk);
242
243void inet_hashinfo_init(struct inet_hashinfo *h);
244void inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
245 unsigned long numentries, int scale,
246 unsigned long low_limit,
247 unsigned long high_limit);
248int inet_hashinfo2_init_mod(struct inet_hashinfo *h);
249
250bool inet_ehash_insert(struct sock *sk, struct sock *osk);
251bool inet_ehash_nolisten(struct sock *sk, struct sock *osk);
252int __inet_hash(struct sock *sk, struct sock *osk);
253int inet_hash(struct sock *sk);
254void inet_unhash(struct sock *sk);
255
256struct sock *__inet_lookup_listener(struct net *net,
257 struct inet_hashinfo *hashinfo,
258 struct sk_buff *skb, int doff,
259 const __be32 saddr, const __be16 sport,
260 const __be32 daddr,
261 const unsigned short hnum,
262 const int dif, const int sdif);
263
264static inline struct sock *inet_lookup_listener(struct net *net,
265 struct inet_hashinfo *hashinfo,
266 struct sk_buff *skb, int doff,
267 __be32 saddr, __be16 sport,
268 __be32 daddr, __be16 dport, int dif, int sdif)
269{
270 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr, sport,
271 daddr, ntohs(dport), dif, sdif);
272}
273
274/* Socket demux engine toys. */
275/* What happens here is ugly; there's a pair of adjacent fields in
276 struct inet_sock; __be16 dport followed by __u16 num. We want to
277 search by pair, so we combine the keys into a single 32bit value
278 and compare with 32bit value read from &...->dport. Let's at least
279 make sure that it's not mixed with anything else...
280 On 64bit targets we combine comparisons with pair of adjacent __be32
281 fields in the same way.
282*/
283#ifdef __BIG_ENDIAN
284#define INET_COMBINED_PORTS(__sport, __dport) \
285 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
286#else /* __LITTLE_ENDIAN */
287#define INET_COMBINED_PORTS(__sport, __dport) \
288 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
289#endif
290
291#if (BITS_PER_LONG == 64)
292#ifdef __BIG_ENDIAN
293#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
294 const __addrpair __name = (__force __addrpair) ( \
295 (((__force __u64)(__be32)(__saddr)) << 32) | \
296 ((__force __u64)(__be32)(__daddr)))
297#else /* __LITTLE_ENDIAN */
298#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
299 const __addrpair __name = (__force __addrpair) ( \
300 (((__force __u64)(__be32)(__daddr)) << 32) | \
301 ((__force __u64)(__be32)(__saddr)))
302#endif /* __BIG_ENDIAN */
303#define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
304 (((__sk)->sk_portpair == (__ports)) && \
305 ((__sk)->sk_addrpair == (__cookie)) && \
306 (((__sk)->sk_bound_dev_if == (__dif)) || \
307 ((__sk)->sk_bound_dev_if == (__sdif))) && \
308 net_eq(sock_net(__sk), (__net)))
309#else /* 32-bit arch */
310#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
311 const int __name __deprecated __attribute__((unused))
312
313#define INET_MATCH(__sk, __net, __cookie, __saddr, __daddr, __ports, __dif, __sdif) \
314 (((__sk)->sk_portpair == (__ports)) && \
315 ((__sk)->sk_daddr == (__saddr)) && \
316 ((__sk)->sk_rcv_saddr == (__daddr)) && \
317 (((__sk)->sk_bound_dev_if == (__dif)) || \
318 ((__sk)->sk_bound_dev_if == (__sdif))) && \
319 net_eq(sock_net(__sk), (__net)))
320#endif /* 64-bit arch */
321
322/* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so we need
323 * not check it for lookups anymore, thanks Alexey. -DaveM
324 */
325struct sock *__inet_lookup_established(struct net *net,
326 struct inet_hashinfo *hashinfo,
327 const __be32 saddr, const __be16 sport,
328 const __be32 daddr, const u16 hnum,
329 const int dif, const int sdif);
330
331static inline struct sock *
332 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
333 const __be32 saddr, const __be16 sport,
334 const __be32 daddr, const __be16 dport,
335 const int dif)
336{
337 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
338 ntohs(dport), dif, 0);
339}
340
341static inline struct sock *__inet_lookup(struct net *net,
342 struct inet_hashinfo *hashinfo,
343 struct sk_buff *skb, int doff,
344 const __be32 saddr, const __be16 sport,
345 const __be32 daddr, const __be16 dport,
346 const int dif, const int sdif,
347 bool *refcounted)
348{
349 u16 hnum = ntohs(dport);
350 struct sock *sk;
351
352 sk = __inet_lookup_established(net, hashinfo, saddr, sport,
353 daddr, hnum, dif, sdif);
354 *refcounted = true;
355 if (sk)
356 return sk;
357 *refcounted = false;
358 return __inet_lookup_listener(net, hashinfo, skb, doff, saddr,
359 sport, daddr, hnum, dif, sdif);
360}
361
362static inline struct sock *inet_lookup(struct net *net,
363 struct inet_hashinfo *hashinfo,
364 struct sk_buff *skb, int doff,
365 const __be32 saddr, const __be16 sport,
366 const __be32 daddr, const __be16 dport,
367 const int dif)
368{
369 struct sock *sk;
370 bool refcounted;
371
372 sk = __inet_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
373 dport, dif, 0, &refcounted);
374
375 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
376 sk = NULL;
377 return sk;
378}
379
380static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
381 struct sk_buff *skb,
382 int doff,
383 const __be16 sport,
384 const __be16 dport,
385 const int sdif,
386 bool *refcounted)
387{
388 struct sock *sk = skb_steal_sock(skb, refcounted);
389 const struct iphdr *iph = ip_hdr(skb);
390
391 if (sk)
392 return sk;
393
394 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo, skb,
395 doff, iph->saddr, sport,
396 iph->daddr, dport, inet_iif(skb), sdif,
397 refcounted);
398}
399
400u32 inet6_ehashfn(const struct net *net,
401 const struct in6_addr *laddr, const u16 lport,
402 const struct in6_addr *faddr, const __be16 fport);
403
404static inline void sk_daddr_set(struct sock *sk, __be32 addr)
405{
406 sk->sk_daddr = addr; /* alias of inet_daddr */
407#if IS_ENABLED(CONFIG_IPV6)
408 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_daddr);
409#endif
410}
411
412static inline void sk_rcv_saddr_set(struct sock *sk, __be32 addr)
413{
414 sk->sk_rcv_saddr = addr; /* alias of inet_rcv_saddr */
415#if IS_ENABLED(CONFIG_IPV6)
416 ipv6_addr_set_v4mapped(addr, &sk->sk_v6_rcv_saddr);
417#endif
418}
419
420int __inet_hash_connect(struct inet_timewait_death_row *death_row,
421 struct sock *sk, u32 port_offset,
422 int (*check_established)(struct inet_timewait_death_row *,
423 struct sock *, __u16,
424 struct inet_timewait_sock **));
425
426int inet_hash_connect(struct inet_timewait_death_row *death_row,
427 struct sock *sk);
428#endif /* _INET_HASHTABLES_H */