blob: 8fe267feb81e2af5e110f97e67916345df0493a1 [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
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -080025void __inet6_hash(struct sock *sk)
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;
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070028
Ilpo Järvinen547b7922008-07-25 21:43:18 -070029 WARN_ON(!sk_unhashed(sk));
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070030
31 if (sk->sk_state == TCP_LISTEN) {
Eric Dumazet5caea4e2008-11-20 00:40:07 -080032 struct inet_listen_hashbucket *ilb;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080033
Eric Dumazet5caea4e2008-11-20 00:40:07 -080034 ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
35 spin_lock(&ilb->lock);
Eric Dumazetc25eb3b2008-11-23 17:22:55 -080036 __sk_nulls_add_node_rcu(sk, &ilb->head);
Eric Dumazet5caea4e2008-11-20 00:40:07 -080037 spin_unlock(&ilb->lock);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070038 } else {
39 unsigned int hash;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080040 struct hlist_nulls_head *list;
Eric Dumazet9db66bd2008-11-20 20:39:09 -080041 spinlock_t *lock;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080042
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070043 sk->sk_hash = hash = inet6_sk_ehashfn(sk);
Eric Dumazet230140c2007-11-07 02:40:20 -080044 list = &inet_ehash_bucket(hashinfo, hash)->chain;
45 lock = inet_ehash_lockp(hashinfo, hash);
Eric Dumazet9db66bd2008-11-20 20:39:09 -080046 spin_lock(lock);
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080047 __sk_nulls_add_node_rcu(sk, list);
Eric Dumazet9db66bd2008-11-20 20:39:09 -080048 spin_unlock(lock);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070049 }
50
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -070051 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070052}
53EXPORT_SYMBOL(__inet6_hash);
54
55/*
56 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
57 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
58 *
59 * The sockhash lock must be held as a reader here.
60 */
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -080061struct sock *__inet6_lookup_established(struct net *net,
62 struct inet_hashinfo *hashinfo,
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070063 const struct in6_addr *saddr,
Al Virod2ecd9c2006-11-08 00:20:00 -080064 const __be16 sport,
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070065 const struct in6_addr *daddr,
66 const u16 hnum,
67 const int dif)
68{
69 struct sock *sk;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080070 const struct hlist_nulls_node *node;
Al Viro4f765d82006-09-27 18:43:07 -070071 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070072 /* Optimize here for direct hit, only listening connections can
73 * have wildcards anyways.
74 */
Pavel Emelyanov33de0142008-06-16 17:13:48 -070075 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080076 unsigned int slot = hash & (hashinfo->ehash_size - 1);
77 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070078
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080079
80 rcu_read_lock();
81begin:
82 sk_nulls_for_each_rcu(sk, node, &head->chain) {
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070083 /* For IPV6 do the cheaper port and family tests first. */
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080084 if (INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
85 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
86 goto begintw;
87 if (!INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
88 sock_put(sk);
89 goto begin;
90 }
91 goto out;
92 }
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070093 }
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080094 if (get_nulls_value(node) != slot)
95 goto begin;
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070096
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080097begintw:
98 /* Must check for a TIME_WAIT'er before going to listener hash. */
99 sk_nulls_for_each_rcu(sk, node, &head->twchain) {
100 if (INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
101 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) {
102 sk = NULL;
103 goto out;
104 }
105 if (!INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
106 sock_put(sk);
107 goto begintw;
108 }
109 goto out;
110 }
111 }
112 if (get_nulls_value(node) != slot)
113 goto begintw;
114 sk = NULL;
115out:
116 rcu_read_unlock();
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -0700117 return sk;
118}
119EXPORT_SYMBOL(__inet6_lookup_established);
120
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800121static int inline compute_score(struct sock *sk, struct net *net,
122 const unsigned short hnum,
123 const struct in6_addr *daddr,
124 const int dif)
125{
126 int score = -1;
127
128 if (net_eq(sock_net(sk), net) && inet_sk(sk)->num == hnum &&
129 sk->sk_family == PF_INET6) {
130 const struct ipv6_pinfo *np = inet6_sk(sk);
131
132 score = 1;
133 if (!ipv6_addr_any(&np->rcv_saddr)) {
134 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
135 return -1;
136 score++;
137 }
138 if (sk->sk_bound_dev_if) {
139 if (sk->sk_bound_dev_if != dif)
140 return -1;
141 score++;
142 }
143 }
144 return score;
145}
146
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800147struct sock *inet6_lookup_listener(struct net *net,
148 struct inet_hashinfo *hashinfo, const struct in6_addr *daddr,
149 const unsigned short hnum, const int dif)
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300150{
151 struct sock *sk;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800152 const struct hlist_nulls_node *node;
153 struct sock *result;
154 int score, hiscore;
155 unsigned int hash = inet_lhashfn(net, hnum);
156 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300157
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800158 rcu_read_lock();
159begin:
160 result = NULL;
161 hiscore = -1;
162 sk_nulls_for_each(sk, node, &ilb->head) {
163 score = compute_score(sk, net, hnum, daddr, dif);
164 if (score > hiscore) {
165 hiscore = score;
166 result = sk;
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300167 }
168 }
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800169 /*
170 * if the nulls value we got at the end of this lookup is
171 * not the expected one, we must restart lookup.
172 * We probably met an item that was moved to another chain.
173 */
174 if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
175 goto begin;
176 if (result) {
177 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
178 result = NULL;
179 else if (unlikely(compute_score(result, net, hnum, daddr,
180 dif) < hiscore)) {
181 sock_put(result);
182 goto begin;
183 }
184 }
185 rcu_read_unlock();
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300186 return result;
187}
188
189EXPORT_SYMBOL_GPL(inet6_lookup_listener);
190
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800191struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
Al Virod2ecd9c2006-11-08 00:20:00 -0800192 const struct in6_addr *saddr, const __be16 sport,
193 const struct in6_addr *daddr, const __be16 dport,
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300194 const int dif)
195{
196 struct sock *sk;
197
198 local_bh_disable();
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800199 sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300200 local_bh_enable();
201
202 return sk;
203}
204
205EXPORT_SYMBOL_GPL(inet6_lookup);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800206
207static int __inet6_check_established(struct inet_timewait_death_row *death_row,
208 struct sock *sk, const __u16 lport,
209 struct inet_timewait_sock **twp)
210{
211 struct inet_hashinfo *hinfo = death_row->hashinfo;
Herbert Xu3759fa92006-03-13 14:26:12 -0800212 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800213 const struct ipv6_pinfo *np = inet6_sk(sk);
214 const struct in6_addr *daddr = &np->rcv_saddr;
215 const struct in6_addr *saddr = &np->daddr;
216 const int dif = sk->sk_bound_dev_if;
Al Viro4f765d82006-09-27 18:43:07 -0700217 const __portpair ports = INET_COMBINED_PORTS(inet->dport, lport);
Pavel Emelyanov33de0142008-06-16 17:13:48 -0700218 struct net *net = sock_net(sk);
219 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800220 inet->dport);
221 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800222 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800223 struct sock *sk2;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800224 const struct hlist_nulls_node *node;
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800225 struct inet_timewait_sock *tw;
226
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800227 spin_lock(lock);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800228
229 /* Check TIME-WAIT sockets first. */
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800230 sk_nulls_for_each(sk2, node, &head->twchain) {
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800231 tw = inet_twsk(sk2);
232
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800233 if (INET6_TW_MATCH(sk2, net, hash, saddr, daddr, ports, dif)) {
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800234 if (twsk_unique(sk, sk2, twp))
235 goto unique;
236 else
237 goto not_unique;
238 }
239 }
240 tw = NULL;
241
242 /* And established part... */
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800243 sk_nulls_for_each(sk2, node, &head->chain) {
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800244 if (INET6_MATCH(sk2, net, hash, saddr, daddr, ports, dif))
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800245 goto not_unique;
246 }
247
248unique:
Herbert Xu3759fa92006-03-13 14:26:12 -0800249 /* Must record num and sport now. Otherwise we will see
250 * in hash table socket with a funny identity. */
251 inet->num = lport;
252 inet->sport = htons(lport);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700253 WARN_ON(!sk_unhashed(sk));
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800254 __sk_nulls_add_node_rcu(sk, &head->chain);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800255 sk->sk_hash = hash;
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800256 spin_unlock(lock);
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700257 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800258
259 if (twp != NULL) {
260 *twp = tw;
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700261 NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITRECYCLED);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800262 } else if (tw != NULL) {
263 /* Silly. Should hash-dance instead... */
264 inet_twsk_deschedule(tw, death_row);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700265 NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITRECYCLED);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800266
267 inet_twsk_put(tw);
268 }
269 return 0;
270
271not_unique:
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800272 spin_unlock(lock);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800273 return -EADDRNOTAVAIL;
274}
275
276static inline u32 inet6_sk_port_offset(const struct sock *sk)
277{
278 const struct inet_sock *inet = inet_sk(sk);
279 const struct ipv6_pinfo *np = inet6_sk(sk);
280 return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
281 np->daddr.s6_addr32,
282 inet->dport);
283}
284
285int inet6_hash_connect(struct inet_timewait_death_row *death_row,
286 struct sock *sk)
287{
Pavel Emelyanov5d8c0aa2008-02-05 03:14:44 -0800288 return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800289 __inet6_check_established, __inet6_hash);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800290}
291
292EXPORT_SYMBOL_GPL(inet6_hash_connect);