blob: 262e13c02ec27dea15154d9b4af5fd413dd1c504 [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>
David S. Miller6e5714e2011-08-03 20:50:44 -070023#include <net/secure_seq.h>
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -080024#include <net/ip.h>
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -030025
Hannes Frederic Sowab50026b2013-10-19 21:48:52 +020026static unsigned int inet6_ehashfn(struct net *net,
27 const struct in6_addr *laddr,
28 const u16 lport,
29 const struct in6_addr *faddr,
30 const __be16 fport)
31{
Hannes Frederic Sowa1bbdcee2013-10-19 21:48:57 +020032 static u32 inet6_ehash_secret __read_mostly;
33 static u32 ipv6_hash_secret __read_mostly;
34
35 u32 lhash, fhash;
36
37 net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
38 net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
39
40 lhash = (__force u32)laddr->s6_addr32[3];
41 fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
42
Hannes Frederic Sowab50026b2013-10-19 21:48:52 +020043 return __inet6_ehashfn(lhash, lport, fhash, fport,
Hannes Frederic Sowa1bbdcee2013-10-19 21:48:57 +020044 inet6_ehash_secret + net_hash_mix(net));
Hannes Frederic Sowab50026b2013-10-19 21:48:52 +020045}
46
47static int inet6_sk_ehashfn(const struct sock *sk)
48{
49 const struct inet_sock *inet = inet_sk(sk);
50 const struct in6_addr *laddr = &sk->sk_v6_rcv_saddr;
51 const struct in6_addr *faddr = &sk->sk_v6_daddr;
52 const __u16 lport = inet->inet_num;
53 const __be16 fport = inet->inet_dport;
54 struct net *net = sock_net(sk);
55
56 return inet6_ehashfn(net, laddr, lport, faddr, fport);
57}
58
Eric Dumazet9327f702009-12-04 03:46:54 +000059int __inet6_hash(struct sock *sk, struct inet_timewait_sock *tw)
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070060{
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -070061 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
Eric Dumazet9327f702009-12-04 03:46:54 +000062 int twrefcnt = 0;
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070063
Ilpo Järvinen547b7922008-07-25 21:43:18 -070064 WARN_ON(!sk_unhashed(sk));
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070065
66 if (sk->sk_state == TCP_LISTEN) {
Eric Dumazet5caea4e2008-11-20 00:40:07 -080067 struct inet_listen_hashbucket *ilb;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080068
Eric Dumazet5caea4e2008-11-20 00:40:07 -080069 ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
70 spin_lock(&ilb->lock);
Eric Dumazetc25eb3b2008-11-23 17:22:55 -080071 __sk_nulls_add_node_rcu(sk, &ilb->head);
Eric Dumazet5caea4e2008-11-20 00:40:07 -080072 spin_unlock(&ilb->lock);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070073 } else {
74 unsigned int hash;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080075 struct hlist_nulls_head *list;
Eric Dumazet9db66bd2008-11-20 20:39:09 -080076 spinlock_t *lock;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080077
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070078 sk->sk_hash = hash = inet6_sk_ehashfn(sk);
Eric Dumazet230140c2007-11-07 02:40:20 -080079 list = &inet_ehash_bucket(hashinfo, hash)->chain;
80 lock = inet_ehash_lockp(hashinfo, hash);
Eric Dumazet9db66bd2008-11-20 20:39:09 -080081 spin_lock(lock);
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080082 __sk_nulls_add_node_rcu(sk, list);
Eric Dumazet9327f702009-12-04 03:46:54 +000083 if (tw) {
84 WARN_ON(sk->sk_hash != tw->tw_hash);
85 twrefcnt = inet_twsk_unhash(tw);
86 }
Eric Dumazet9db66bd2008-11-20 20:39:09 -080087 spin_unlock(lock);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070088 }
89
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -070090 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Eric Dumazet9327f702009-12-04 03:46:54 +000091 return twrefcnt;
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070092}
93EXPORT_SYMBOL(__inet6_hash);
94
95/*
96 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
97 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
98 *
99 * The sockhash lock must be held as a reader here.
100 */
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800101struct sock *__inet6_lookup_established(struct net *net,
102 struct inet_hashinfo *hashinfo,
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -0700103 const struct in6_addr *saddr,
Al Virod2ecd9c2006-11-08 00:20:00 -0800104 const __be16 sport,
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -0700105 const struct in6_addr *daddr,
106 const u16 hnum,
107 const int dif)
108{
109 struct sock *sk;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800110 const struct hlist_nulls_node *node;
Al Viro4f765d82006-09-27 18:43:07 -0700111 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -0700112 /* Optimize here for direct hit, only listening connections can
113 * have wildcards anyways.
114 */
Pavel Emelyanov33de0142008-06-16 17:13:48 -0700115 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
Eric Dumazetf373b532009-10-09 00:16:19 +0000116 unsigned int slot = hash & hashinfo->ehash_mask;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800117 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -0700118
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800119
120 rcu_read_lock();
121begin:
122 sk_nulls_for_each_rcu(sk, node, &head->chain) {
Eric Dumazetce43b032012-11-30 09:49:27 +0000123 if (sk->sk_hash != hash)
124 continue;
Eric Dumazetefe42082013-10-03 15:42:29 -0700125 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif))
126 continue;
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700127 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
128 goto out;
129
Eric Dumazetefe42082013-10-03 15:42:29 -0700130 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
131 sock_gen_put(sk);
132 goto begin;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800133 }
Eric Dumazetefe42082013-10-03 15:42:29 -0700134 goto found;
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -0700135 }
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800136 if (get_nulls_value(node) != slot)
137 goto begin;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800138out:
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700139 sk = NULL;
140found:
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800141 rcu_read_unlock();
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -0700142 return sk;
143}
144EXPORT_SYMBOL(__inet6_lookup_established);
145
Jesper Juhl42b16b32011-01-17 00:09:38 +0100146static inline int compute_score(struct sock *sk, struct net *net,
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800147 const unsigned short hnum,
148 const struct in6_addr *daddr,
149 const int dif)
150{
151 int score = -1;
152
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000153 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800154 sk->sk_family == PF_INET6) {
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800155
156 score = 1;
Eric Dumazetefe42082013-10-03 15:42:29 -0700157 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
158 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800159 return -1;
160 score++;
161 }
162 if (sk->sk_bound_dev_if) {
163 if (sk->sk_bound_dev_if != dif)
164 return -1;
165 score++;
166 }
167 }
168 return score;
169}
170
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800171struct sock *inet6_lookup_listener(struct net *net,
Tom Herbert5ba24952013-01-22 09:50:39 +0000172 struct inet_hashinfo *hashinfo, const struct in6_addr *saddr,
173 const __be16 sport, const struct in6_addr *daddr,
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800174 const unsigned short hnum, const int dif)
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300175{
176 struct sock *sk;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800177 const struct hlist_nulls_node *node;
178 struct sock *result;
Tom Herbert5ba24952013-01-22 09:50:39 +0000179 int score, hiscore, matches = 0, reuseport = 0;
180 u32 phash = 0;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800181 unsigned int hash = inet_lhashfn(net, hnum);
182 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300183
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800184 rcu_read_lock();
185begin:
186 result = NULL;
Tom Herbert5ba24952013-01-22 09:50:39 +0000187 hiscore = 0;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800188 sk_nulls_for_each(sk, node, &ilb->head) {
189 score = compute_score(sk, net, hnum, daddr, dif);
190 if (score > hiscore) {
191 hiscore = score;
192 result = sk;
Tom Herbert5ba24952013-01-22 09:50:39 +0000193 reuseport = sk->sk_reuseport;
194 if (reuseport) {
195 phash = inet6_ehashfn(net, daddr, hnum,
196 saddr, sport);
197 matches = 1;
198 }
199 } else if (score == hiscore && reuseport) {
200 matches++;
201 if (((u64)phash * matches) >> 32 == 0)
202 result = sk;
203 phash = next_pseudo_random32(phash);
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300204 }
205 }
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800206 /*
207 * if the nulls value we got at the end of this lookup is
208 * not the expected one, we must restart lookup.
209 * We probably met an item that was moved to another chain.
210 */
211 if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
212 goto begin;
213 if (result) {
214 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
215 result = NULL;
216 else if (unlikely(compute_score(result, net, hnum, daddr,
217 dif) < hiscore)) {
218 sock_put(result);
219 goto begin;
220 }
221 }
222 rcu_read_unlock();
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300223 return result;
224}
225
226EXPORT_SYMBOL_GPL(inet6_lookup_listener);
227
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800228struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
Al Virod2ecd9c2006-11-08 00:20:00 -0800229 const struct in6_addr *saddr, const __be16 sport,
230 const struct in6_addr *daddr, const __be16 dport,
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300231 const int dif)
232{
233 struct sock *sk;
234
235 local_bh_disable();
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800236 sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300237 local_bh_enable();
238
239 return sk;
240}
241
242EXPORT_SYMBOL_GPL(inet6_lookup);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800243
244static int __inet6_check_established(struct inet_timewait_death_row *death_row,
245 struct sock *sk, const __u16 lport,
246 struct inet_timewait_sock **twp)
247{
248 struct inet_hashinfo *hinfo = death_row->hashinfo;
Herbert Xu3759fa92006-03-13 14:26:12 -0800249 struct inet_sock *inet = inet_sk(sk);
Eric Dumazetefe42082013-10-03 15:42:29 -0700250 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
251 const struct in6_addr *saddr = &sk->sk_v6_daddr;
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800252 const int dif = sk->sk_bound_dev_if;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000253 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
Pavel Emelyanov33de0142008-06-16 17:13:48 -0700254 struct net *net = sock_net(sk);
255 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000256 inet->inet_dport);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800257 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800258 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800259 struct sock *sk2;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800260 const struct hlist_nulls_node *node;
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700261 struct inet_timewait_sock *tw = NULL;
Eric Dumazet13475a32009-12-02 22:31:19 +0000262 int twrefcnt = 0;
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800263
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800264 spin_lock(lock);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800265
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800266 sk_nulls_for_each(sk2, node, &head->chain) {
Eric Dumazetce43b032012-11-30 09:49:27 +0000267 if (sk2->sk_hash != hash)
268 continue;
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700269
Eric Dumazetefe42082013-10-03 15:42:29 -0700270 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) {
271 if (sk2->sk_state == TCP_TIME_WAIT) {
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700272 tw = inet_twsk(sk2);
273 if (twsk_unique(sk, sk2, twp))
Eric Dumazetefe42082013-10-03 15:42:29 -0700274 break;
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700275 }
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800276 goto not_unique;
Eric Dumazetefe42082013-10-03 15:42:29 -0700277 }
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800278 }
279
Herbert Xu3759fa92006-03-13 14:26:12 -0800280 /* Must record num and sport now. Otherwise we will see
Eric Dumazetefe42082013-10-03 15:42:29 -0700281 * in hash table socket with a funny identity.
282 */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000283 inet->inet_num = lport;
284 inet->inet_sport = htons(lport);
Eric Dumazet13475a32009-12-02 22:31:19 +0000285 sk->sk_hash = hash;
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700286 WARN_ON(!sk_unhashed(sk));
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800287 __sk_nulls_add_node_rcu(sk, &head->chain);
Eric Dumazet13475a32009-12-02 22:31:19 +0000288 if (tw) {
289 twrefcnt = inet_twsk_unhash(tw);
290 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
291 }
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800292 spin_unlock(lock);
Eric Dumazet13475a32009-12-02 22:31:19 +0000293 if (twrefcnt)
294 inet_twsk_put(tw);
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700295 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800296
Eric Dumazet13475a32009-12-02 22:31:19 +0000297 if (twp) {
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800298 *twp = tw;
Eric Dumazet13475a32009-12-02 22:31:19 +0000299 } else if (tw) {
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800300 /* Silly. Should hash-dance instead... */
301 inet_twsk_deschedule(tw, death_row);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800302
303 inet_twsk_put(tw);
304 }
305 return 0;
306
307not_unique:
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800308 spin_unlock(lock);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800309 return -EADDRNOTAVAIL;
310}
311
312static inline u32 inet6_sk_port_offset(const struct sock *sk)
313{
314 const struct inet_sock *inet = inet_sk(sk);
Eric Dumazetefe42082013-10-03 15:42:29 -0700315
316 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
317 sk->sk_v6_daddr.s6_addr32,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000318 inet->inet_dport);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800319}
320
321int inet6_hash_connect(struct inet_timewait_death_row *death_row,
322 struct sock *sk)
323{
Pavel Emelyanov5d8c0aa2008-02-05 03:14:44 -0800324 return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
Pavel Emelyanov5ee31fc2008-01-31 05:04:45 -0800325 __inet6_check_established, __inet6_hash);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800326}
327
328EXPORT_SYMBOL_GPL(inet6_hash_connect);