net: do not call sock_put() on TIMEWAIT sockets
[pandora-kernel.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/secure_seq.h>
24 #include <net/ip.h>
25
26 int __inet6_hash(struct sock *sk, struct inet_timewait_sock *tw)
27 {
28         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
29         int twrefcnt = 0;
30
31         WARN_ON(!sk_unhashed(sk));
32
33         if (sk->sk_state == TCP_LISTEN) {
34                 struct inet_listen_hashbucket *ilb;
35
36                 ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
37                 spin_lock(&ilb->lock);
38                 __sk_nulls_add_node_rcu(sk, &ilb->head);
39                 spin_unlock(&ilb->lock);
40         } else {
41                 unsigned int hash;
42                 struct hlist_nulls_head *list;
43                 spinlock_t *lock;
44
45                 sk->sk_hash = hash = inet6_sk_ehashfn(sk);
46                 list = &inet_ehash_bucket(hashinfo, hash)->chain;
47                 lock = inet_ehash_lockp(hashinfo, hash);
48                 spin_lock(lock);
49                 __sk_nulls_add_node_rcu(sk, list);
50                 if (tw) {
51                         WARN_ON(sk->sk_hash != tw->tw_hash);
52                         twrefcnt = inet_twsk_unhash(tw);
53                 }
54                 spin_unlock(lock);
55         }
56
57         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
58         return twrefcnt;
59 }
60 EXPORT_SYMBOL(__inet6_hash);
61
62 /*
63  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
64  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
65  *
66  * The sockhash lock must be held as a reader here.
67  */
68 struct sock *__inet6_lookup_established(struct net *net,
69                                         struct inet_hashinfo *hashinfo,
70                                            const struct in6_addr *saddr,
71                                            const __be16 sport,
72                                            const struct in6_addr *daddr,
73                                            const u16 hnum,
74                                            const int dif)
75 {
76         struct sock *sk;
77         const struct hlist_nulls_node *node;
78         const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
79         /* Optimize here for direct hit, only listening connections can
80          * have wildcards anyways.
81          */
82         unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
83         unsigned int slot = hash & hashinfo->ehash_mask;
84         struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
85
86
87         rcu_read_lock();
88 begin:
89         sk_nulls_for_each_rcu(sk, node, &head->chain) {
90                 /* For IPV6 do the cheaper port and family tests first. */
91                 if (INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
92                         if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
93                                 goto begintw;
94                         if (!INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
95                                 sock_put(sk);
96                                 goto begin;
97                         }
98                 goto out;
99                 }
100         }
101         if (get_nulls_value(node) != slot)
102                 goto begin;
103
104 begintw:
105         /* Must check for a TIME_WAIT'er before going to listener hash. */
106         sk_nulls_for_each_rcu(sk, node, &head->twchain) {
107                 if (INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
108                         if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) {
109                                 sk = NULL;
110                                 goto out;
111                         }
112                         if (!INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
113                                 inet_twsk_put(inet_twsk(sk));
114                                 goto begintw;
115                         }
116                         goto out;
117                 }
118         }
119         if (get_nulls_value(node) != slot)
120                 goto begintw;
121         sk = NULL;
122 out:
123         rcu_read_unlock();
124         return sk;
125 }
126 EXPORT_SYMBOL(__inet6_lookup_established);
127
128 static inline int compute_score(struct sock *sk, struct net *net,
129                                 const unsigned short hnum,
130                                 const struct in6_addr *daddr,
131                                 const int dif)
132 {
133         int score = -1;
134
135         if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
136             sk->sk_family == PF_INET6) {
137                 const struct ipv6_pinfo *np = inet6_sk(sk);
138
139                 score = 1;
140                 if (!ipv6_addr_any(&np->rcv_saddr)) {
141                         if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
142                                 return -1;
143                         score++;
144                 }
145                 if (sk->sk_bound_dev_if) {
146                         if (sk->sk_bound_dev_if != dif)
147                                 return -1;
148                         score++;
149                 }
150         }
151         return score;
152 }
153
154 struct sock *inet6_lookup_listener(struct net *net,
155                 struct inet_hashinfo *hashinfo, const struct in6_addr *daddr,
156                 const unsigned short hnum, const int dif)
157 {
158         struct sock *sk;
159         const struct hlist_nulls_node *node;
160         struct sock *result;
161         int score, hiscore;
162         unsigned int hash = inet_lhashfn(net, hnum);
163         struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
164
165         rcu_read_lock();
166 begin:
167         result = NULL;
168         hiscore = -1;
169         sk_nulls_for_each(sk, node, &ilb->head) {
170                 score = compute_score(sk, net, hnum, daddr, dif);
171                 if (score > hiscore) {
172                         hiscore = score;
173                         result = sk;
174                 }
175         }
176         /*
177          * if the nulls value we got at the end of this lookup is
178          * not the expected one, we must restart lookup.
179          * We probably met an item that was moved to another chain.
180          */
181         if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
182                 goto begin;
183         if (result) {
184                 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
185                         result = NULL;
186                 else if (unlikely(compute_score(result, net, hnum, daddr,
187                                   dif) < hiscore)) {
188                         sock_put(result);
189                         goto begin;
190                 }
191         }
192         rcu_read_unlock();
193         return result;
194 }
195
196 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
197
198 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
199                           const struct in6_addr *saddr, const __be16 sport,
200                           const struct in6_addr *daddr, const __be16 dport,
201                           const int dif)
202 {
203         struct sock *sk;
204
205         local_bh_disable();
206         sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
207         local_bh_enable();
208
209         return sk;
210 }
211
212 EXPORT_SYMBOL_GPL(inet6_lookup);
213
214 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
215                                      struct sock *sk, const __u16 lport,
216                                      struct inet_timewait_sock **twp)
217 {
218         struct inet_hashinfo *hinfo = death_row->hashinfo;
219         struct inet_sock *inet = inet_sk(sk);
220         const struct ipv6_pinfo *np = inet6_sk(sk);
221         const struct in6_addr *daddr = &np->rcv_saddr;
222         const struct in6_addr *saddr = &np->daddr;
223         const int dif = sk->sk_bound_dev_if;
224         const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
225         struct net *net = sock_net(sk);
226         const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
227                                                 inet->inet_dport);
228         struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
229         spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
230         struct sock *sk2;
231         const struct hlist_nulls_node *node;
232         struct inet_timewait_sock *tw;
233         int twrefcnt = 0;
234
235         spin_lock(lock);
236
237         /* Check TIME-WAIT sockets first. */
238         sk_nulls_for_each(sk2, node, &head->twchain) {
239                 tw = inet_twsk(sk2);
240
241                 if (INET6_TW_MATCH(sk2, net, hash, saddr, daddr, ports, dif)) {
242                         if (twsk_unique(sk, sk2, twp))
243                                 goto unique;
244                         else
245                                 goto not_unique;
246                 }
247         }
248         tw = NULL;
249
250         /* And established part... */
251         sk_nulls_for_each(sk2, node, &head->chain) {
252                 if (INET6_MATCH(sk2, net, hash, saddr, daddr, ports, dif))
253                         goto not_unique;
254         }
255
256 unique:
257         /* Must record num and sport now. Otherwise we will see
258          * in hash table socket with a funny identity. */
259         inet->inet_num = lport;
260         inet->inet_sport = htons(lport);
261         sk->sk_hash = hash;
262         WARN_ON(!sk_unhashed(sk));
263         __sk_nulls_add_node_rcu(sk, &head->chain);
264         if (tw) {
265                 twrefcnt = inet_twsk_unhash(tw);
266                 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
267         }
268         spin_unlock(lock);
269         if (twrefcnt)
270                 inet_twsk_put(tw);
271         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
272
273         if (twp) {
274                 *twp = tw;
275         } else if (tw) {
276                 /* Silly. Should hash-dance instead... */
277                 inet_twsk_deschedule(tw, death_row);
278
279                 inet_twsk_put(tw);
280         }
281         return 0;
282
283 not_unique:
284         spin_unlock(lock);
285         return -EADDRNOTAVAIL;
286 }
287
288 static inline u32 inet6_sk_port_offset(const struct sock *sk)
289 {
290         const struct inet_sock *inet = inet_sk(sk);
291         const struct ipv6_pinfo *np = inet6_sk(sk);
292         return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
293                                           np->daddr.s6_addr32,
294                                           inet->inet_dport);
295 }
296
297 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
298                        struct sock *sk)
299 {
300         return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
301                         __inet6_check_established, __inet6_hash);
302 }
303
304 EXPORT_SYMBOL_GPL(inet6_hash_connect);