c813e294ec0c94d2aac13149f528683a85987d91
[linux-2.6.git] / net / ipv6 / inet6_hashtables.c
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  *              Generic INET6 transport hashtables
7  *
8  * Authors:     Lotsa people, from code originally in tcp, generalised here
9  *              by Arnaldo Carvalho de Melo <acme@mandriva.com>
10  *
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
17 #include <linux/module.h>
18 #include <linux/random.h>
19
20 #include <net/inet_connection_sock.h>
21 #include <net/inet_hashtables.h>
22 #include <net/inet6_hashtables.h>
23 #include <net/ip.h>
24
25 void __inet6_hash(struct sock *sk)
26 {
27         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
28
29         WARN_ON(!sk_unhashed(sk));
30
31         if (sk->sk_state == TCP_LISTEN) {
32                 struct inet_listen_hashbucket *ilb;
33
34                 ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
35                 spin_lock(&ilb->lock);
36                 __sk_nulls_add_node_rcu(sk, &ilb->head);
37                 spin_unlock(&ilb->lock);
38         } else {
39                 unsigned int hash;
40                 struct hlist_nulls_head *list;
41                 spinlock_t *lock;
42
43                 sk->sk_hash = hash = inet6_sk_ehashfn(sk);
44                 list = &inet_ehash_bucket(hashinfo, hash)->chain;
45                 lock = inet_ehash_lockp(hashinfo, hash);
46                 spin_lock(lock);
47                 __sk_nulls_add_node_rcu(sk, list);
48                 spin_unlock(lock);
49         }
50
51         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
52 }
53 EXPORT_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  */
61 struct sock *__inet6_lookup_established(struct net *net,
62                                         struct inet_hashinfo *hashinfo,
63                                            const struct in6_addr *saddr,
64                                            const __be16 sport,
65                                            const struct in6_addr *daddr,
66                                            const u16 hnum,
67                                            const int dif)
68 {
69         struct sock *sk;
70         const struct hlist_nulls_node *node;
71         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
72         /* Optimize here for direct hit, only listening connections can
73          * have wildcards anyways.
74          */
75         unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
76         unsigned int slot = hash & hashinfo->ehash_mask;
77         struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
78
79
80         rcu_read_lock();
81 begin:
82         sk_nulls_for_each_rcu(sk, node, &head->chain) {
83                 /* For IPV6 do the cheaper port and family tests first. */
84                 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                 }
93         }
94         if (get_nulls_value(node) != slot)
95                 goto begin;
96
97 begintw:
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;
115 out:
116         rcu_read_unlock();
117         return sk;
118 }
119 EXPORT_SYMBOL(__inet6_lookup_established);
120
121 static 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)->inet_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
147 struct 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)
150 {
151         struct sock *sk;
152         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];
157
158         rcu_read_lock();
159 begin:
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;
167                 }
168         }
169         /*
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();
186         return result;
187 }
188
189 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
190
191 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
192                           const struct in6_addr *saddr, const __be16 sport,
193                           const struct in6_addr *daddr, const __be16 dport,
194                           const int dif)
195 {
196         struct sock *sk;
197
198         local_bh_disable();
199         sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
200         local_bh_enable();
201
202         return sk;
203 }
204
205 EXPORT_SYMBOL_GPL(inet6_lookup);
206
207 static 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;
212         struct inet_sock *inet = inet_sk(sk);
213         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;
217         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
218         struct net *net = sock_net(sk);
219         const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
220                                                 inet->inet_dport);
221         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
222         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
223         struct sock *sk2;
224         const struct hlist_nulls_node *node;
225         struct inet_timewait_sock *tw;
226         int twrefcnt = 0;
227
228         spin_lock(lock);
229
230         /* Check TIME-WAIT sockets first. */
231         sk_nulls_for_each(sk2, node, &head->twchain) {
232                 tw = inet_twsk(sk2);
233
234                 if (INET6_TW_MATCH(sk2, net, hash, saddr, daddr, ports, dif)) {
235                         if (twsk_unique(sk, sk2, twp))
236                                 goto unique;
237                         else
238                                 goto not_unique;
239                 }
240         }
241         tw = NULL;
242
243         /* And established part... */
244         sk_nulls_for_each(sk2, node, &head->chain) {
245                 if (INET6_MATCH(sk2, net, hash, saddr, daddr, ports, dif))
246                         goto not_unique;
247         }
248
249 unique:
250         /* Must record num and sport now. Otherwise we will see
251          * in hash table socket with a funny identity. */
252         inet->inet_num = lport;
253         inet->inet_sport = htons(lport);
254         sk->sk_hash = hash;
255         WARN_ON(!sk_unhashed(sk));
256         __sk_nulls_add_node_rcu(sk, &head->chain);
257         if (tw) {
258                 twrefcnt = inet_twsk_unhash(tw);
259                 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
260         }
261         spin_unlock(lock);
262         if (twrefcnt)
263                 inet_twsk_put(tw);
264         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
265
266         if (twp) {
267                 *twp = tw;
268         } else if (tw) {
269                 /* Silly. Should hash-dance instead... */
270                 inet_twsk_deschedule(tw, death_row);
271
272                 inet_twsk_put(tw);
273         }
274         return 0;
275
276 not_unique:
277         spin_unlock(lock);
278         return -EADDRNOTAVAIL;
279 }
280
281 static inline u32 inet6_sk_port_offset(const struct sock *sk)
282 {
283         const struct inet_sock *inet = inet_sk(sk);
284         const struct ipv6_pinfo *np = inet6_sk(sk);
285         return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
286                                           np->daddr.s6_addr32,
287                                           inet->inet_dport);
288 }
289
290 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
291                        struct sock *sk)
292 {
293         return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
294                         __inet6_check_established, __inet6_hash);
295 }
296
297 EXPORT_SYMBOL_GPL(inet6_hash_connect);