blob: b53197233709c71af7630581ee6dab590e15f5a3 [file] [log] [blame]
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -03001/*
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 * Generic INET6 transport hashtables
7 *
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -08008 * Authors: Lotsa people, from code originally in tcp, generalised here
9 * by Arnaldo Carvalho de Melo <acme@mandriva.com>
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -030010 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
15 */
16
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -030017#include <linux/module.h>
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -080018#include <linux/random.h>
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -030019
20#include <net/inet_connection_sock.h>
21#include <net/inet_hashtables.h>
22#include <net/inet6_hashtables.h>
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -080023#include <net/ip.h>
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -030024
Eric Dumazet9327f702009-12-04 03:46:54 +000025int __inet6_hash(struct sock *sk, struct inet_timewait_sock *tw)
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070026{
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -070027 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
Eric Dumazet9327f702009-12-04 03:46:54 +000028 int twrefcnt = 0;
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070029
Ilpo Järvinen547b7922008-07-25 21:43:18 -070030 WARN_ON(!sk_unhashed(sk));
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070031
32 if (sk->sk_state == TCP_LISTEN) {
Eric Dumazet5caea4e2008-11-20 00:40:07 -080033 struct inet_listen_hashbucket *ilb;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080034
Eric Dumazet5caea4e2008-11-20 00:40:07 -080035 ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
36 spin_lock(&ilb->lock);
Eric Dumazetc25eb3b2008-11-23 17:22:55 -080037 __sk_nulls_add_node_rcu(sk, &ilb->head);
Eric Dumazet5caea4e2008-11-20 00:40:07 -080038 spin_unlock(&ilb->lock);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070039 } else {
40 unsigned int hash;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080041 struct hlist_nulls_head *list;
Eric Dumazet9db66bd2008-11-20 20:39:09 -080042 spinlock_t *lock;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080043
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070044 sk->sk_hash = hash = inet6_sk_ehashfn(sk);
Eric Dumazet230140c2007-11-07 02:40:20 -080045 list = &inet_ehash_bucket(hashinfo, hash)->chain;
46 lock = inet_ehash_lockp(hashinfo, hash);
Eric Dumazet9db66bd2008-11-20 20:39:09 -080047 spin_lock(lock);
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080048 __sk_nulls_add_node_rcu(sk, list);
Eric Dumazet9327f702009-12-04 03:46:54 +000049 if (tw) {
50 WARN_ON(sk->sk_hash != tw->tw_hash);
51 twrefcnt = inet_twsk_unhash(tw);
52 }
Eric Dumazet9db66bd2008-11-20 20:39:09 -080053 spin_unlock(lock);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070054 }
55
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -070056 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Eric Dumazet9327f702009-12-04 03:46:54 +000057 return twrefcnt;
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070058}
59EXPORT_SYMBOL(__inet6_hash);
60
61/*
62 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
63 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
64 *
65 * The sockhash lock must be held as a reader here.
66 */
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -080067struct sock *__inet6_lookup_established(struct net *net,
68 struct inet_hashinfo *hashinfo,
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070069 const struct in6_addr *saddr,
Al Virod2ecd9c2006-11-08 00:20:00 -080070 const __be16 sport,
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070071 const struct in6_addr *daddr,
72 const u16 hnum,
73 const int dif)
74{
75 struct sock *sk;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080076 const struct hlist_nulls_node *node;
Al Viro4f765d82006-09-27 18:43:07 -070077 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070078 /* Optimize here for direct hit, only listening connections can
79 * have wildcards anyways.
80 */
Pavel Emelyanov33de0142008-06-16 17:13:48 -070081 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
Eric Dumazetf373b532009-10-09 00:16:19 +000082 unsigned int slot = hash & hashinfo->ehash_mask;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080083 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070084
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080085
86 rcu_read_lock();
87begin:
88 sk_nulls_for_each_rcu(sk, node, &head->chain) {
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070089 /* For IPV6 do the cheaper port and family tests first. */
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080090 if (INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
91 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
92 goto begintw;
93 if (!INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
94 sock_put(sk);
95 goto begin;
96 }
97 goto out;
98 }
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070099 }
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800100 if (get_nulls_value(node) != slot)
101 goto begin;
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -0700102
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800103begintw:
104 /* Must check for a TIME_WAIT'er before going to listener hash. */
105 sk_nulls_for_each_rcu(sk, node, &head->twchain) {
106 if (INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
107 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) {
108 sk = NULL;
109 goto out;
110 }
111 if (!INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
112 sock_put(sk);
113 goto begintw;
114 }
115 goto out;
116 }
117 }
118 if (get_nulls_value(node) != slot)
119 goto begintw;
120 sk = NULL;
121out:
122 rcu_read_unlock();
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -0700123 return sk;
124}
125EXPORT_SYMBOL(__inet6_lookup_established);
126
Jesper Juhl42b16b32011-01-17 00:09:38 +0100127static inline int compute_score(struct sock *sk, struct net *net,
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800128 const unsigned short hnum,
129 const struct in6_addr *daddr,
130 const int dif)
131{
132 int score = -1;
133
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000134 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800135 sk->sk_family == PF_INET6) {
136 const struct ipv6_pinfo *np = inet6_sk(sk);
137
138 score = 1;
139 if (!ipv6_addr_any(&np->rcv_saddr)) {
140 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
141 return -1;
142 score++;
143 }
144 if (sk->sk_bound_dev_if) {
145 if (sk->sk_bound_dev_if != dif)
146 return -1;
147 score++;
148 }
149 }
150 return score;
151}
152
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800153struct sock *inet6_lookup_listener(struct net *net,
154 struct inet_hashinfo *hashinfo, const struct in6_addr *daddr,
155 const unsigned short hnum, const int dif)
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300156{
157 struct sock *sk;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800158 const struct hlist_nulls_node *node;
159 struct sock *result;
160 int score, hiscore;
161 unsigned int hash = inet_lhashfn(net, hnum);
162 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300163
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800164 rcu_read_lock();
165begin:
166 result = NULL;
167 hiscore = -1;
168 sk_nulls_for_each(sk, node, &ilb->head) {
169 score = compute_score(sk, net, hnum, daddr, dif);
170 if (score > hiscore) {
171 hiscore = score;
172 result = sk;
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300173 }
174 }
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800175 /*
176 * if the nulls value we got at the end of this lookup is
177 * not the expected one, we must restart lookup.
178 * We probably met an item that was moved to another chain.
179 */
180 if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
181 goto begin;
182 if (result) {
183 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
184 result = NULL;
185 else if (unlikely(compute_score(result, net, hnum, daddr,
186 dif) < hiscore)) {
187 sock_put(result);
188 goto begin;
189 }
190 }
191 rcu_read_unlock();
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300192 return result;
193}
194
195EXPORT_SYMBOL_GPL(inet6_lookup_listener);
196
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800197struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
Al Virod2ecd9c2006-11-08 00:20:00 -0800198 const struct in6_addr *saddr, const __be16 sport,
199 const struct in6_addr *daddr, const __be16 dport,
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300200 const int dif)
201{
202 struct sock *sk;
203
204 local_bh_disable();
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800205 sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300206 local_bh_enable();
207
208 return sk;
209}
210
211EXPORT_SYMBOL_GPL(inet6_lookup);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800212
213static int __inet6_check_established(struct inet_timewait_death_row *death_row,
214 struct sock *sk, const __u16 lport,
215 struct inet_timewait_sock **twp)
216{
217 struct inet_hashinfo *hinfo = death_row->hashinfo;
Herbert Xu3759fa92006-03-13 14:26:12 -0800218 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800219 const struct ipv6_pinfo *np = inet6_sk(sk);
220 const struct in6_addr *daddr = &np->rcv_saddr;
221 const struct in6_addr *saddr = &np->daddr;
222 const int dif = sk->sk_bound_dev_if;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000223 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
Pavel Emelyanov33de0142008-06-16 17:13:48 -0700224 struct net *net = sock_net(sk);
225 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000226 inet->inet_dport);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800227 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800228 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800229 struct sock *sk2;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800230 const struct hlist_nulls_node *node;
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800231 struct inet_timewait_sock *tw;
Eric Dumazet13475a32009-12-02 22:31:19 +0000232 int twrefcnt = 0;
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800233
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800234 spin_lock(lock);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800235
236 /* Check TIME-WAIT sockets first. */
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800237 sk_nulls_for_each(sk2, node, &head->twchain) {
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800238 tw = inet_twsk(sk2);
239
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800240 if (INET6_TW_MATCH(sk2, net, hash, saddr, daddr, ports, dif)) {
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800241 if (twsk_unique(sk, sk2, twp))
242 goto unique;
243 else
244 goto not_unique;
245 }
246 }
247 tw = NULL;
248
249 /* And established part... */
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800250 sk_nulls_for_each(sk2, node, &head->chain) {
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800251 if (INET6_MATCH(sk2, net, hash, saddr, daddr, ports, dif))
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800252 goto not_unique;
253 }
254
255unique:
Herbert Xu3759fa92006-03-13 14:26:12 -0800256 /* Must record num and sport now. Otherwise we will see
257 * in hash table socket with a funny identity. */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000258 inet->inet_num = lport;
259 inet->inet_sport = htons(lport);
Eric Dumazet13475a32009-12-02 22:31:19 +0000260 sk->sk_hash = hash;
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700261 WARN_ON(!sk_unhashed(sk));
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800262 __sk_nulls_add_node_rcu(sk, &head->chain);
Eric Dumazet13475a32009-12-02 22:31:19 +0000263 if (tw) {
264 twrefcnt = inet_twsk_unhash(tw);
265 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
266 }
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800267 spin_unlock(lock);
Eric Dumazet13475a32009-12-02 22:31:19 +0000268 if (twrefcnt)
269 inet_twsk_put(tw);
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700270 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800271
Eric Dumazet13475a32009-12-02 22:31:19 +0000272 if (twp) {
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800273 *twp = tw;
Eric Dumazet13475a32009-12-02 22:31:19 +0000274 } else if (tw) {
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800275 /* Silly. Should hash-dance instead... */
276 inet_twsk_deschedule(tw, death_row);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800277
278 inet_twsk_put(tw);
279 }
280 return 0;
281
282not_unique:
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800283 spin_unlock(lock);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800284 return -EADDRNOTAVAIL;
285}
286
287static inline u32 inet6_sk_port_offset(const struct sock *sk)
288{
289 const struct inet_sock *inet = inet_sk(sk);
290 const struct ipv6_pinfo *np = inet6_sk(sk);
291 return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
292 np->daddr.s6_addr32,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000293 inet->inet_dport);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800294}
295
296int inet6_hash_connect(struct inet_timewait_death_row *death_row,
297 struct sock *sk)
298{
Pavel Emelyanov5d8c0aa2008-02-05 03:14:44 -0800299 return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800300 __inet6_check_established, __inet6_hash);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800301}
302
303EXPORT_SYMBOL_GPL(inet6_hash_connect);