Merge git://git.infradead.org/mtd-2.6
[linux-2.6] / 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 inet_hashinfo *hashinfo,
26                                 struct sock *sk)
27 {
28         struct hlist_head *list;
29         rwlock_t *lock;
30
31         BUG_TRAP(sk_unhashed(sk));
32
33         if (sk->sk_state == TCP_LISTEN) {
34                 list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
35                 lock = &hashinfo->lhash_lock;
36                 inet_listen_wlock(hashinfo);
37         } else {
38                 unsigned int hash;
39                 sk->sk_hash = hash = inet6_sk_ehashfn(sk);
40                 hash &= (hashinfo->ehash_size - 1);
41                 list = &hashinfo->ehash[hash].chain;
42                 lock = &hashinfo->ehash[hash].lock;
43                 write_lock(lock);
44         }
45
46         __sk_add_node(sk, list);
47         sock_prot_inc_use(sk->sk_prot);
48         write_unlock(lock);
49 }
50 EXPORT_SYMBOL(__inet6_hash);
51
52 /*
53  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
54  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
55  *
56  * The sockhash lock must be held as a reader here.
57  */
58 struct sock *__inet6_lookup_established(struct inet_hashinfo *hashinfo,
59                                            const struct in6_addr *saddr,
60                                            const __be16 sport,
61                                            const struct in6_addr *daddr,
62                                            const u16 hnum,
63                                            const int dif)
64 {
65         struct sock *sk;
66         const struct hlist_node *node;
67         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
68         /* Optimize here for direct hit, only listening connections can
69          * have wildcards anyways.
70          */
71         unsigned int hash = inet6_ehashfn(daddr, hnum, saddr, sport);
72         struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash);
73
74         prefetch(head->chain.first);
75         read_lock(&head->lock);
76         sk_for_each(sk, node, &head->chain) {
77                 /* For IPV6 do the cheaper port and family tests first. */
78                 if (INET6_MATCH(sk, hash, saddr, daddr, ports, dif))
79                         goto hit; /* You sunk my battleship! */
80         }
81         /* Must check for a TIME_WAIT'er before going to listener hash. */
82         sk_for_each(sk, node, &head->twchain) {
83                 const struct inet_timewait_sock *tw = inet_twsk(sk);
84
85                 if(*((__portpair *)&(tw->tw_dport))     == ports        &&
86                    sk->sk_family                == PF_INET6) {
87                         const struct inet6_timewait_sock *tw6 = inet6_twsk(sk);
88
89                         if (ipv6_addr_equal(&tw6->tw_v6_daddr, saddr)   &&
90                             ipv6_addr_equal(&tw6->tw_v6_rcv_saddr, daddr)       &&
91                             (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif))
92                                 goto hit;
93                 }
94         }
95         read_unlock(&head->lock);
96         return NULL;
97
98 hit:
99         sock_hold(sk);
100         read_unlock(&head->lock);
101         return sk;
102 }
103 EXPORT_SYMBOL(__inet6_lookup_established);
104
105 struct sock *inet6_lookup_listener(struct inet_hashinfo *hashinfo,
106                                    const struct in6_addr *daddr,
107                                    const unsigned short hnum, const int dif)
108 {
109         struct sock *sk;
110         const struct hlist_node *node;
111         struct sock *result = NULL;
112         int score, hiscore = 0;
113
114         read_lock(&hashinfo->lhash_lock);
115         sk_for_each(sk, node, &hashinfo->listening_hash[inet_lhashfn(hnum)]) {
116                 if (inet_sk(sk)->num == hnum && sk->sk_family == PF_INET6) {
117                         const struct ipv6_pinfo *np = inet6_sk(sk);
118
119                         score = 1;
120                         if (!ipv6_addr_any(&np->rcv_saddr)) {
121                                 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
122                                         continue;
123                                 score++;
124                         }
125                         if (sk->sk_bound_dev_if) {
126                                 if (sk->sk_bound_dev_if != dif)
127                                         continue;
128                                 score++;
129                         }
130                         if (score == 3) {
131                                 result = sk;
132                                 break;
133                         }
134                         if (score > hiscore) {
135                                 hiscore = score;
136                                 result = sk;
137                         }
138                 }
139         }
140         if (result)
141                 sock_hold(result);
142         read_unlock(&hashinfo->lhash_lock);
143         return result;
144 }
145
146 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
147
148 struct sock *inet6_lookup(struct inet_hashinfo *hashinfo,
149                           const struct in6_addr *saddr, const __be16 sport,
150                           const struct in6_addr *daddr, const __be16 dport,
151                           const int dif)
152 {
153         struct sock *sk;
154
155         local_bh_disable();
156         sk = __inet6_lookup(hashinfo, saddr, sport, daddr, ntohs(dport), dif);
157         local_bh_enable();
158
159         return sk;
160 }
161
162 EXPORT_SYMBOL_GPL(inet6_lookup);
163
164 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
165                                      struct sock *sk, const __u16 lport,
166                                      struct inet_timewait_sock **twp)
167 {
168         struct inet_hashinfo *hinfo = death_row->hashinfo;
169         struct inet_sock *inet = inet_sk(sk);
170         const struct ipv6_pinfo *np = inet6_sk(sk);
171         const struct in6_addr *daddr = &np->rcv_saddr;
172         const struct in6_addr *saddr = &np->daddr;
173         const int dif = sk->sk_bound_dev_if;
174         const __portpair ports = INET_COMBINED_PORTS(inet->dport, lport);
175         const unsigned int hash = inet6_ehashfn(daddr, lport, saddr,
176                                                 inet->dport);
177         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
178         struct sock *sk2;
179         const struct hlist_node *node;
180         struct inet_timewait_sock *tw;
181
182         prefetch(head->chain.first);
183         write_lock(&head->lock);
184
185         /* Check TIME-WAIT sockets first. */
186         sk_for_each(sk2, node, &head->twchain) {
187                 const struct inet6_timewait_sock *tw6 = inet6_twsk(sk2);
188
189                 tw = inet_twsk(sk2);
190
191                 if(*((__portpair *)&(tw->tw_dport)) == ports             &&
192                    sk2->sk_family              == PF_INET6       &&
193                    ipv6_addr_equal(&tw6->tw_v6_daddr, saddr)     &&
194                    ipv6_addr_equal(&tw6->tw_v6_rcv_saddr, daddr) &&
195                    sk2->sk_bound_dev_if == sk->sk_bound_dev_if) {
196                         if (twsk_unique(sk, sk2, twp))
197                                 goto unique;
198                         else
199                                 goto not_unique;
200                 }
201         }
202         tw = NULL;
203
204         /* And established part... */
205         sk_for_each(sk2, node, &head->chain) {
206                 if (INET6_MATCH(sk2, hash, saddr, daddr, ports, dif))
207                         goto not_unique;
208         }
209
210 unique:
211         /* Must record num and sport now. Otherwise we will see
212          * in hash table socket with a funny identity. */
213         inet->num = lport;
214         inet->sport = htons(lport);
215         BUG_TRAP(sk_unhashed(sk));
216         __sk_add_node(sk, &head->chain);
217         sk->sk_hash = hash;
218         sock_prot_inc_use(sk->sk_prot);
219         write_unlock(&head->lock);
220
221         if (twp != NULL) {
222                 *twp = tw;
223                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
224         } else if (tw != NULL) {
225                 /* Silly. Should hash-dance instead... */
226                 inet_twsk_deschedule(tw, death_row);
227                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
228
229                 inet_twsk_put(tw);
230         }
231         return 0;
232
233 not_unique:
234         write_unlock(&head->lock);
235         return -EADDRNOTAVAIL;
236 }
237
238 static inline u32 inet6_sk_port_offset(const struct sock *sk)
239 {
240         const struct inet_sock *inet = inet_sk(sk);
241         const struct ipv6_pinfo *np = inet6_sk(sk);
242         return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
243                                           np->daddr.s6_addr32,
244                                           inet->dport);
245 }
246
247 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
248                        struct sock *sk)
249 {
250         struct inet_hashinfo *hinfo = death_row->hashinfo;
251         const unsigned short snum = inet_sk(sk)->num;
252         struct inet_bind_hashbucket *head;
253         struct inet_bind_bucket *tb;
254         int ret;
255
256         if (snum == 0) {
257                 const int low = sysctl_local_port_range[0];
258                 const int high = sysctl_local_port_range[1];
259                 const int range = high - low;
260                 int i, port;
261                 static u32 hint;
262                 const u32 offset = hint + inet6_sk_port_offset(sk);
263                 struct hlist_node *node;
264                 struct inet_timewait_sock *tw = NULL;
265
266                 local_bh_disable();
267                 for (i = 1; i <= range; i++) {
268                         port = low + (i + offset) % range;
269                         head = &hinfo->bhash[inet_bhashfn(port, hinfo->bhash_size)];
270                         spin_lock(&head->lock);
271
272                         /* Does not bother with rcv_saddr checks,
273                          * because the established check is already
274                          * unique enough.
275                          */
276                         inet_bind_bucket_for_each(tb, node, &head->chain) {
277                                 if (tb->port == port) {
278                                         BUG_TRAP(!hlist_empty(&tb->owners));
279                                         if (tb->fastreuse >= 0)
280                                                 goto next_port;
281                                         if (!__inet6_check_established(death_row,
282                                                                        sk, port,
283                                                                        &tw))
284                                                 goto ok;
285                                         goto next_port;
286                                 }
287                         }
288
289                         tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
290                                                      head, port);
291                         if (!tb) {
292                                 spin_unlock(&head->lock);
293                                 break;
294                         }
295                         tb->fastreuse = -1;
296                         goto ok;
297
298                 next_port:
299                         spin_unlock(&head->lock);
300                 }
301                 local_bh_enable();
302
303                 return -EADDRNOTAVAIL;
304
305 ok:
306                 hint += i;
307
308                 /* Head lock still held and bh's disabled */
309                 inet_bind_hash(sk, tb, port);
310                 if (sk_unhashed(sk)) {
311                         inet_sk(sk)->sport = htons(port);
312                         __inet6_hash(hinfo, sk);
313                 }
314                 spin_unlock(&head->lock);
315
316                 if (tw) {
317                         inet_twsk_deschedule(tw, death_row);
318                         inet_twsk_put(tw);
319                 }
320
321                 ret = 0;
322                 goto out;
323         }
324
325         head = &hinfo->bhash[inet_bhashfn(snum, hinfo->bhash_size)];
326         tb   = inet_csk(sk)->icsk_bind_hash;
327         spin_lock_bh(&head->lock);
328
329         if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
330                 __inet6_hash(hinfo, sk);
331                 spin_unlock_bh(&head->lock);
332                 return 0;
333         } else {
334                 spin_unlock(&head->lock);
335                 /* No definite answer... Walk to established hash table */
336                 ret = __inet6_check_established(death_row, sk, snum, NULL);
337 out:
338                 local_bh_enable();
339                 return ret;
340         }
341 }
342
343 EXPORT_SYMBOL_GPL(inet6_hash_connect);