blob: d522dcf3031ab8ad7806af1d572e22751c93a6da [file] [log] [blame]
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -07001/*
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
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070017
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070018#include <linux/interrupt.h>
Arnaldo Carvalho de Melo9a1f27c2008-10-07 11:41:57 -070019#include <linux/ip.h>
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -070020#include <linux/ipv6.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070021#include <linux/list.h>
22#include <linux/slab.h>
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -070023#include <linux/socket.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070024#include <linux/spinlock.h>
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -070025#include <linux/types.h>
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -070026#include <linux/wait.h>
Eric Dumazet9e4505c2007-11-10 21:18:39 -080027#include <linux/vmalloc.h>
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -070028
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -070029#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020030#include <net/inet_sock.h>
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070031#include <net/sock.h>
Arnaldo Carvalho de Melo9a1f27c2008-10-07 11:41:57 -070032#include <net/route.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070033#include <net/tcp_states.h>
Pavel Emelyanov0b441912008-06-16 17:14:11 -070034#include <net/netns/hash.h>
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -070035
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -070036#include <asm/atomic.h>
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -070037#include <asm/byteorder.h>
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -070038
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070039/* This is for all connections with a full identity, no wildcards.
Eric Dumazetdbca9b2752007-02-08 14:16:46 -080040 * One chain is dedicated to TIME_WAIT sockets.
41 * I'll experiment with dynamic table growth later.
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070042 */
43struct inet_ehash_bucket {
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080044 struct hlist_nulls_head chain;
45 struct hlist_nulls_head twchain;
Eric Dumazet6d255362005-10-04 15:55:51 -070046};
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070047
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 {
Eric Dumazet7a9546e2008-11-12 00:54:20 -080080#ifdef CONFIG_NET_NS
Pavel Emelyanov941b1d22008-01-31 05:05:50 -080081 struct net *ib_net;
Eric Dumazet7a9546e2008-11-12 00:54:20 -080082#endif
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070083 unsigned short port;
84 signed short fastreuse;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -080085 int num_owners;
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070086 struct hlist_node node;
87 struct hlist_head owners;
88};
89
Eric Dumazet7a9546e2008-11-12 00:54:20 -080090static inline struct net *ib_net(struct inet_bind_bucket *ib)
91{
92 return read_pnet(&ib->ib_net);
93}
94
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070095#define inet_bind_bucket_for_each(tb, node, head) \
96 hlist_for_each_entry(tb, node, head, node)
97
98struct inet_bind_hashbucket {
99 spinlock_t lock;
100 struct hlist_head chain;
101};
102
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800103/*
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)
Eric Dumazet5caea4e2008-11-20 00:40:07 -0800110struct inet_listen_hashbucket {
111 spinlock_t lock;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800112 struct hlist_nulls_head head;
Eric Dumazet5caea4e2008-11-20 00:40:07 -0800113};
114
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700115/* 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 *
Eric Dumazetdbca9b2752007-02-08 14:16:46 -0800124 * TIME_WAIT sockets use a separate chain (twchain).
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700125 */
126 struct inet_ehash_bucket *ehash;
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800127 spinlock_t *ehash_locks;
Eric Dumazet230140c2007-11-07 02:40:20 -0800128 unsigned int ehash_size;
129 unsigned int ehash_locks_mask;
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700130
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
Stephen Hemmingercfcabdc2007-10-09 01:59:42 -0700136 unsigned int bhash_size;
Eric Dumazet24dd1fa2009-02-01 12:31:33 -0800137 /* 4 bytes hole on 64 bit */
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700138
Eric Dumazet5caea4e2008-11-20 00:40:07 -0800139 struct kmem_cache *bind_bucket_cachep;
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700140
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
Eric Dumazet5caea4e2008-11-20 00:40:07 -0800145 * might be often dirty.
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700146 */
Eric Dumazet5caea4e2008-11-20 00:40:07 -0800147 /* 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
Eric Dumazet24dd1fa2009-02-01 12:31:33 -0800154 atomic_t bsockets;
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700155};
156
Eric Dumazet81c3d542005-10-03 14:13:38 -0700157static 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_size - 1)];
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -0700162}
163
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800164static inline spinlock_t *inet_ehash_lockp(
Eric Dumazet230140c2007-11-07 02:40:20 -0800165 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;
Eric Dumazet94cd3e62009-01-27 17:45:10 -0800187 if (sizeof(spinlock_t) != 0) {
Eric Dumazet230140c2007-11-07 02:40:20 -0800188#ifdef CONFIG_NUMA
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800189 if (size * sizeof(spinlock_t) > PAGE_SIZE)
190 hashinfo->ehash_locks = vmalloc(size * sizeof(spinlock_t));
Eric Dumazet230140c2007-11-07 02:40:20 -0800191 else
192#endif
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800193 hashinfo->ehash_locks = kmalloc(size * sizeof(spinlock_t),
Eric Dumazet230140c2007-11-07 02:40:20 -0800194 GFP_KERNEL);
195 if (!hashinfo->ehash_locks)
196 return ENOMEM;
197 for (i = 0; i < size; i++)
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800198 spin_lock_init(&hashinfo->ehash_locks[i]);
Eric Dumazet230140c2007-11-07 02:40:20 -0800199 }
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) *
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800209 sizeof(spinlock_t);
Eric Dumazet230140c2007-11-07 02:40:20 -0800210 if (size > PAGE_SIZE)
211 vfree(hashinfo->ehash_locks);
212 else
Eric Dumazet230140c2007-11-07 02:40:20 -0800213#endif
Pavel Emelyanov218ad122007-11-26 20:23:31 +0800214 kfree(hashinfo->ehash_locks);
Eric Dumazet230140c2007-11-07 02:40:20 -0800215 hashinfo->ehash_locks = NULL;
216 }
217}
218
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700219extern struct inet_bind_bucket *
Christoph Lametere18b8902006-12-06 20:33:20 -0800220 inet_bind_bucket_create(struct kmem_cache *cachep,
Pavel Emelyanov941b1d22008-01-31 05:05:50 -0800221 struct net *net,
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700222 struct inet_bind_hashbucket *head,
223 const unsigned short snum);
Christoph Lametere18b8902006-12-06 20:33:20 -0800224extern void inet_bind_bucket_destroy(struct kmem_cache *cachep,
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700225 struct inet_bind_bucket *tb);
226
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700227static inline int inet_bhashfn(struct net *net,
228 const __u16 lport, const int bhash_size)
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700229{
Pavel Emelyanov0b441912008-06-16 17:14:11 -0700230 return (lport + net_hash_mix(net)) & (bhash_size - 1);
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700231}
232
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700233extern void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
234 const unsigned short snum);
235
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700236/* These can have wildcards, don't try too hard. */
Pavel Emelyanov2086a652008-06-16 17:13:08 -0700237static inline int inet_lhashfn(struct net *net, const unsigned short num)
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700238{
Pavel Emelyanov0b441912008-06-16 17:14:11 -0700239 return (num + net_hash_mix(net)) & (INET_LHTABLE_SIZE - 1);
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700240}
241
242static inline int inet_sk_listen_hashfn(const struct sock *sk)
243{
Pavel Emelyanov2086a652008-06-16 17:13:08 -0700244 return inet_lhashfn(sock_net(sk), inet_sk(sk)->num);
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -0700245}
246
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700247/* Caller must disable local BH processing. */
Pavel Emelyanov53083772008-04-17 23:18:15 -0700248extern void __inet_inherit_port(struct sock *sk, struct sock *child);
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700249
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800250extern void inet_put_port(struct sock *sk);
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700251
Eric Dumazet5caea4e2008-11-20 00:40:07 -0800252void inet_hashinfo_init(struct inet_hashinfo *h);
Arnaldo Carvalho de Melof3f05f72005-08-09 20:08:09 -0700253
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800254extern void __inet_hash_nolisten(struct sock *sk);
255extern void inet_hash(struct sock *sk);
256extern void inet_unhash(struct sock *sk);
Arnaldo Carvalho de Melo33b62232005-08-09 20:09:06 -0700257
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800258extern struct sock *__inet_lookup_listener(struct net *net,
259 struct inet_hashinfo *hashinfo,
Al Virofb99c842006-09-27 18:43:33 -0700260 const __be32 daddr,
Herbert Xu8f4910692006-08-09 15:47:12 -0700261 const unsigned short hnum,
262 const int dif);
263
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800264static inline struct sock *inet_lookup_listener(struct net *net,
265 struct inet_hashinfo *hashinfo,
266 __be32 daddr, __be16 dport, int dif)
Herbert Xu8f4910692006-08-09 15:47:12 -0700267{
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800268 return __inet_lookup_listener(net, hashinfo, daddr, ntohs(dport), dif);
Herbert Xu8f4910692006-08-09 15:47:12 -0700269}
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700270
271/* Socket demux engine toys. */
Al Viro4f765d82006-09-27 18:43:07 -0700272/* 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;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700281#ifdef __BIG_ENDIAN
282#define INET_COMBINED_PORTS(__sport, __dport) \
Al Viro4f765d82006-09-27 18:43:07 -0700283 ((__force __portpair)(((__force __u32)(__be16)(__sport) << 16) | (__u32)(__dport)))
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700284#else /* __LITTLE_ENDIAN */
285#define INET_COMBINED_PORTS(__sport, __dport) \
Al Viro4f765d82006-09-27 18:43:07 -0700286 ((__force __portpair)(((__u32)(__dport) << 16) | (__force __u32)(__be16)(__sport)))
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700287#endif
288
289#if (BITS_PER_LONG == 64)
Al Viro4f765d82006-09-27 18:43:07 -0700290typedef __u64 __bitwise __addrpair;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700291#ifdef __BIG_ENDIAN
292#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
Al Viro4f765d82006-09-27 18:43:07 -0700293 const __addrpair __name = (__force __addrpair) ( \
294 (((__force __u64)(__be32)(__saddr)) << 32) | \
295 ((__force __u64)(__be32)(__daddr)));
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700296#else /* __LITTLE_ENDIAN */
297#define INET_ADDR_COOKIE(__name, __saddr, __daddr) \
Al Viro4f765d82006-09-27 18:43:07 -0700298 const __addrpair __name = (__force __addrpair) ( \
299 (((__force __u64)(__be32)(__daddr)) << 32) | \
300 ((__force __u64)(__be32)(__saddr)));
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700301#endif /* __BIG_ENDIAN */
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800302#define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
Eric Dumazetf757fec2008-11-21 15:49:19 -0800303 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
Al Viro4f765d82006-09-27 18:43:07 -0700304 ((*((__addrpair *)&(inet_sk(__sk)->daddr))) == (__cookie)) && \
305 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700306 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800307#define INET_TW_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
Eric Dumazetf757fec2008-11-21 15:49:19 -0800308 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
Al Viro4f765d82006-09-27 18:43:07 -0700309 ((*((__addrpair *)&(inet_twsk(__sk)->tw_daddr))) == (__cookie)) && \
310 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700311 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
312#else /* 32-bit arch */
313#define INET_ADDR_COOKIE(__name, __saddr, __daddr)
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800314#define INET_MATCH(__sk, __net, __hash, __cookie, __saddr, __daddr, __ports, __dif) \
Eric Dumazetf757fec2008-11-21 15:49:19 -0800315 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
Eric Dumazet81c3d542005-10-03 14:13:38 -0700316 (inet_sk(__sk)->daddr == (__saddr)) && \
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700317 (inet_sk(__sk)->rcv_saddr == (__daddr)) && \
Al Viro4f765d82006-09-27 18:43:07 -0700318 ((*((__portpair *)&(inet_sk(__sk)->dport))) == (__ports)) && \
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700319 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800320#define INET_TW_MATCH(__sk, __net, __hash,__cookie, __saddr, __daddr, __ports, __dif) \
Eric Dumazetf757fec2008-11-21 15:49:19 -0800321 (((__sk)->sk_hash == (__hash)) && net_eq(sock_net(__sk), (__net)) && \
Eric Dumazet81c3d542005-10-03 14:13:38 -0700322 (inet_twsk(__sk)->tw_daddr == (__saddr)) && \
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700323 (inet_twsk(__sk)->tw_rcv_saddr == (__daddr)) && \
Al Viro4f765d82006-09-27 18:43:07 -0700324 ((*((__portpair *)&(inet_twsk(__sk)->tw_dport))) == (__ports)) && \
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700325 (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
326#endif /* 64-bit arch */
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700327
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 */
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800334extern struct sock * __inet_lookup_established(struct net *net,
335 struct inet_hashinfo *hashinfo,
Pavel Emelyanov77a5ba52007-12-20 15:32:17 -0800336 const __be32 saddr, const __be16 sport,
337 const __be32 daddr, const u16 hnum, const int dif);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700338
Herbert Xu8f4910692006-08-09 15:47:12 -0700339static inline struct sock *
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800340 inet_lookup_established(struct net *net, struct inet_hashinfo *hashinfo,
Al Virofb99c842006-09-27 18:43:33 -0700341 const __be32 saddr, const __be16 sport,
342 const __be32 daddr, const __be16 dport,
Herbert Xu8f4910692006-08-09 15:47:12 -0700343 const int dif)
344{
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800345 return __inet_lookup_established(net, hashinfo, saddr, sport, daddr,
Herbert Xu8f4910692006-08-09 15:47:12 -0700346 ntohs(dport), dif);
347}
348
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800349static inline struct sock *__inet_lookup(struct net *net,
350 struct inet_hashinfo *hashinfo,
Al Virofb99c842006-09-27 18:43:33 -0700351 const __be32 saddr, const __be16 sport,
352 const __be32 daddr, const __be16 dport,
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700353 const int dif)
354{
Herbert Xu8f4910692006-08-09 15:47:12 -0700355 u16 hnum = ntohs(dport);
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800356 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);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700360}
361
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800362static inline struct sock *inet_lookup(struct net *net,
363 struct inet_hashinfo *hashinfo,
Al Virofb99c842006-09-27 18:43:33 -0700364 const __be32 saddr, const __be16 sport,
365 const __be32 daddr, const __be16 dport,
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700366 const int dif)
367{
368 struct sock *sk;
369
370 local_bh_disable();
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800371 sk = __inet_lookup(net, hashinfo, saddr, sport, daddr, dport, dif);
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700372 local_bh_enable();
373
374 return sk;
375}
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800376
Arnaldo Carvalho de Melo9a1f27c2008-10-07 11:41:57 -0700377static inline struct sock *__inet_lookup_skb(struct inet_hashinfo *hashinfo,
378 struct sk_buff *skb,
379 const __be16 sport,
380 const __be16 dport)
381{
KOVACS Krisztian23542612008-10-07 12:41:01 -0700382 struct sock *sk;
Arnaldo Carvalho de Melo9a1f27c2008-10-07 11:41:57 -0700383 const struct iphdr *iph = ip_hdr(skb);
384
KOVACS Krisztian23542612008-10-07 12:41:01 -0700385 if (unlikely(sk = skb_steal_sock(skb)))
386 return sk;
387 else
Eric Dumazetadf30902009-06-02 05:19:30 +0000388 return __inet_lookup(dev_net(skb_dst(skb)->dev), hashinfo,
KOVACS Krisztian23542612008-10-07 12:41:01 -0700389 iph->saddr, sport,
390 iph->daddr, dport, inet_iif(skb));
Arnaldo Carvalho de Melo9a1f27c2008-10-07 11:41:57 -0700391}
392
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800393extern int __inet_hash_connect(struct inet_timewait_death_row *death_row,
Pavel Emelyanov5d8c0aa2008-02-05 03:14:44 -0800394 struct sock *sk, u32 port_offset,
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800395 int (*check_established)(struct inet_timewait_death_row *,
396 struct sock *, __u16, struct inet_timewait_sock **),
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800397 void (*hash)(struct sock *sk));
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800398extern int inet_hash_connect(struct inet_timewait_death_row *death_row,
399 struct sock *sk);
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -0700400#endif /* _INET_HASHTABLES_H */