blob: 21ace5a2bf7c27f7e70a212cf0d94432d029112d [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
Ian Morris67ba4152014-08-24 21:53:10 +01009 * 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
Eric Dumazetd1e559d2015-03-18 14:05:35 -070026u32 inet6_ehashfn(const struct net *net,
27 const struct in6_addr *laddr, const u16 lport,
28 const struct in6_addr *faddr, const __be16 fport)
Hannes Frederic Sowab50026b2013-10-19 21:48:52 +020029{
Hannes Frederic Sowa1bbdcee2013-10-19 21:48:57 +020030 static u32 inet6_ehash_secret __read_mostly;
31 static u32 ipv6_hash_secret __read_mostly;
32
33 u32 lhash, fhash;
34
35 net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
36 net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
37
38 lhash = (__force u32)laddr->s6_addr32[3];
39 fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
40
Hannes Frederic Sowab50026b2013-10-19 21:48:52 +020041 return __inet6_ehashfn(lhash, lport, fhash, fport,
Hannes Frederic Sowa1bbdcee2013-10-19 21:48:57 +020042 inet6_ehash_secret + net_hash_mix(net));
Hannes Frederic Sowab50026b2013-10-19 21:48:52 +020043}
44
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070045/*
46 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
47 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
48 *
49 * The sockhash lock must be held as a reader here.
50 */
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -080051struct sock *__inet6_lookup_established(struct net *net,
52 struct inet_hashinfo *hashinfo,
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070053 const struct in6_addr *saddr,
Al Virod2ecd9c2006-11-08 00:20:00 -080054 const __be16 sport,
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070055 const struct in6_addr *daddr,
56 const u16 hnum,
57 const int dif)
58{
59 struct sock *sk;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080060 const struct hlist_nulls_node *node;
Al Viro4f765d82006-09-27 18:43:07 -070061 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070062 /* Optimize here for direct hit, only listening connections can
63 * have wildcards anyways.
64 */
Pavel Emelyanov33de0142008-06-16 17:13:48 -070065 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
Eric Dumazetf373b532009-10-09 00:16:19 +000066 unsigned int slot = hash & hashinfo->ehash_mask;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080067 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070068
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080069
70 rcu_read_lock();
71begin:
72 sk_nulls_for_each_rcu(sk, node, &head->chain) {
Eric Dumazetce43b032012-11-30 09:49:27 +000073 if (sk->sk_hash != hash)
74 continue;
Eric Dumazetefe42082013-10-03 15:42:29 -070075 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif))
76 continue;
Eric Dumazet05dbc7b2013-10-03 00:22:02 -070077 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
78 goto out;
79
Eric Dumazetefe42082013-10-03 15:42:29 -070080 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
81 sock_gen_put(sk);
82 goto begin;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080083 }
Eric Dumazetefe42082013-10-03 15:42:29 -070084 goto found;
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070085 }
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080086 if (get_nulls_value(node) != slot)
87 goto begin;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080088out:
Eric Dumazet05dbc7b2013-10-03 00:22:02 -070089 sk = NULL;
90found:
Eric Dumazet3ab5aee2008-11-16 19:40:17 -080091 rcu_read_unlock();
Denis Vlasenkob1a7ffc2006-04-09 22:48:59 -070092 return sk;
93}
94EXPORT_SYMBOL(__inet6_lookup_established);
95
Jesper Juhl42b16b32011-01-17 00:09:38 +010096static inline int compute_score(struct sock *sk, struct net *net,
Eric Dumazetc25eb3b2008-11-23 17:22:55 -080097 const unsigned short hnum,
98 const struct in6_addr *daddr,
99 const int dif)
100{
101 int score = -1;
102
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000103 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800104 sk->sk_family == PF_INET6) {
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800105
106 score = 1;
Eric Dumazetefe42082013-10-03 15:42:29 -0700107 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
108 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800109 return -1;
110 score++;
111 }
112 if (sk->sk_bound_dev_if) {
113 if (sk->sk_bound_dev_if != dif)
114 return -1;
115 score++;
116 }
Eric Dumazet70da2682015-10-08 19:33:21 -0700117 if (sk->sk_incoming_cpu == raw_smp_processor_id())
118 score++;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800119 }
120 return score;
121}
122
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800123struct sock *inet6_lookup_listener(struct net *net,
Tom Herbert5ba24952013-01-22 09:50:39 +0000124 struct inet_hashinfo *hashinfo, const struct in6_addr *saddr,
125 const __be16 sport, const struct in6_addr *daddr,
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800126 const unsigned short hnum, const int dif)
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300127{
128 struct sock *sk;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800129 const struct hlist_nulls_node *node;
130 struct sock *result;
Tom Herbert5ba24952013-01-22 09:50:39 +0000131 int score, hiscore, matches = 0, reuseport = 0;
132 u32 phash = 0;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800133 unsigned int hash = inet_lhashfn(net, hnum);
134 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300135
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800136 rcu_read_lock();
137begin:
138 result = NULL;
Tom Herbert5ba24952013-01-22 09:50:39 +0000139 hiscore = 0;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800140 sk_nulls_for_each(sk, node, &ilb->head) {
141 score = compute_score(sk, net, hnum, daddr, dif);
142 if (score > hiscore) {
143 hiscore = score;
144 result = sk;
Tom Herbert5ba24952013-01-22 09:50:39 +0000145 reuseport = sk->sk_reuseport;
146 if (reuseport) {
147 phash = inet6_ehashfn(net, daddr, hnum,
148 saddr, sport);
149 matches = 1;
150 }
151 } else if (score == hiscore && reuseport) {
152 matches++;
Daniel Borkmann8fc54f62014-08-23 20:58:54 +0200153 if (reciprocal_scale(phash, matches) == 0)
Tom Herbert5ba24952013-01-22 09:50:39 +0000154 result = sk;
155 phash = next_pseudo_random32(phash);
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300156 }
157 }
Eric Dumazetc25eb3b2008-11-23 17:22:55 -0800158 /*
159 * if the nulls value we got at the end of this lookup is
160 * not the expected one, we must restart lookup.
161 * We probably met an item that was moved to another chain.
162 */
163 if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
164 goto begin;
165 if (result) {
166 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
167 result = NULL;
168 else if (unlikely(compute_score(result, net, hnum, daddr,
169 dif) < hiscore)) {
170 sock_put(result);
171 goto begin;
172 }
173 }
174 rcu_read_unlock();
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300175 return result;
176}
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300177EXPORT_SYMBOL_GPL(inet6_lookup_listener);
178
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800179struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
Al Virod2ecd9c2006-11-08 00:20:00 -0800180 const struct in6_addr *saddr, const __be16 sport,
181 const struct in6_addr *daddr, const __be16 dport,
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300182 const int dif)
183{
184 struct sock *sk;
185
186 local_bh_disable();
Pavel Emelyanovd86e0da2008-01-31 05:07:21 -0800187 sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300188 local_bh_enable();
189
190 return sk;
191}
Arnaldo Carvalho de Melo5324a042005-08-12 09:26:18 -0300192EXPORT_SYMBOL_GPL(inet6_lookup);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800193
194static int __inet6_check_established(struct inet_timewait_death_row *death_row,
195 struct sock *sk, const __u16 lport,
196 struct inet_timewait_sock **twp)
197{
198 struct inet_hashinfo *hinfo = death_row->hashinfo;
Herbert Xu3759fa92006-03-13 14:26:12 -0800199 struct inet_sock *inet = inet_sk(sk);
Eric Dumazetefe42082013-10-03 15:42:29 -0700200 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
201 const struct in6_addr *saddr = &sk->sk_v6_daddr;
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800202 const int dif = sk->sk_bound_dev_if;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000203 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
Pavel Emelyanov33de0142008-06-16 17:13:48 -0700204 struct net *net = sock_net(sk);
205 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000206 inet->inet_dport);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800207 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800208 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800209 struct sock *sk2;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800210 const struct hlist_nulls_node *node;
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700211 struct inet_timewait_sock *tw = NULL;
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800212
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800213 spin_lock(lock);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800214
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800215 sk_nulls_for_each(sk2, node, &head->chain) {
Eric Dumazetce43b032012-11-30 09:49:27 +0000216 if (sk2->sk_hash != hash)
217 continue;
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700218
Eric Dumazetefe42082013-10-03 15:42:29 -0700219 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) {
220 if (sk2->sk_state == TCP_TIME_WAIT) {
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700221 tw = inet_twsk(sk2);
222 if (twsk_unique(sk, sk2, twp))
Eric Dumazetefe42082013-10-03 15:42:29 -0700223 break;
Eric Dumazet05dbc7b2013-10-03 00:22:02 -0700224 }
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800225 goto not_unique;
Eric Dumazetefe42082013-10-03 15:42:29 -0700226 }
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800227 }
228
Herbert Xu3759fa92006-03-13 14:26:12 -0800229 /* Must record num and sport now. Otherwise we will see
Eric Dumazetefe42082013-10-03 15:42:29 -0700230 * in hash table socket with a funny identity.
231 */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000232 inet->inet_num = lport;
233 inet->inet_sport = htons(lport);
Eric Dumazet13475a32009-12-02 22:31:19 +0000234 sk->sk_hash = hash;
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700235 WARN_ON(!sk_unhashed(sk));
Eric Dumazet3ab5aee2008-11-16 19:40:17 -0800236 __sk_nulls_add_node_rcu(sk, &head->chain);
Eric Dumazet13475a32009-12-02 22:31:19 +0000237 if (tw) {
Eric Dumazetfc01538f2015-07-08 14:28:29 -0700238 sk_nulls_del_node_init_rcu((struct sock *)tw);
Eric Dumazet13475a32009-12-02 22:31:19 +0000239 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
240 }
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800241 spin_unlock(lock);
Pavel Emelyanovc29a0bc2008-03-31 19:41:46 -0700242 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800243
Eric Dumazet13475a32009-12-02 22:31:19 +0000244 if (twp) {
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800245 *twp = tw;
Eric Dumazet13475a32009-12-02 22:31:19 +0000246 } else if (tw) {
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800247 /* Silly. Should hash-dance instead... */
Eric Dumazetdbe7faa2015-07-08 14:28:30 -0700248 inet_twsk_deschedule_put(tw);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800249 }
250 return 0;
251
252not_unique:
Eric Dumazet9db66bd2008-11-20 20:39:09 -0800253 spin_unlock(lock);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800254 return -EADDRNOTAVAIL;
255}
256
Eric Dumazete2baad92015-05-27 10:46:02 -0700257static u32 inet6_sk_port_offset(const struct sock *sk)
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800258{
259 const struct inet_sock *inet = inet_sk(sk);
Eric Dumazetefe42082013-10-03 15:42:29 -0700260
261 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
262 sk->sk_v6_daddr.s6_addr32,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000263 inet->inet_dport);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800264}
265
266int inet6_hash_connect(struct inet_timewait_death_row *death_row,
267 struct sock *sk)
268{
Eric Dumazete2baad92015-05-27 10:46:02 -0700269 u32 port_offset = 0;
270
271 if (!inet_sk(sk)->inet_num)
272 port_offset = inet6_sk_port_offset(sk);
273 return __inet_hash_connect(death_row, sk, port_offset,
Eric Dumazetb4d64442015-03-18 14:05:37 -0700274 __inet6_check_established);
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800275}
Arnaldo Carvalho de Melod8313f52005-12-13 23:25:44 -0800276EXPORT_SYMBOL_GPL(inet6_hash_connect);