Commit | Line | Data |
---|---|---|
5324a040 ACM |
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 | * | |
d8313f5c | 8 | * Authors: Lotsa people, from code originally in tcp, generalised here |
67ba4152 | 9 | * by Arnaldo Carvalho de Melo <acme@mandriva.com> |
5324a040 ACM |
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 | ||
5324a040 | 17 | #include <linux/module.h> |
d8313f5c | 18 | #include <linux/random.h> |
5324a040 ACM |
19 | |
20 | #include <net/inet_connection_sock.h> | |
21 | #include <net/inet_hashtables.h> | |
22 | #include <net/inet6_hashtables.h> | |
6e5714ea | 23 | #include <net/secure_seq.h> |
d8313f5c | 24 | #include <net/ip.h> |
5324a040 | 25 | |
d1e559d0 ED |
26 | u32 inet6_ehashfn(const struct net *net, |
27 | const struct in6_addr *laddr, const u16 lport, | |
28 | const struct in6_addr *faddr, const __be16 fport) | |
b50026b5 | 29 | { |
1bbdceef HFS |
30 | 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 | ||
b50026b5 | 41 | return __inet6_ehashfn(lhash, lport, fhash, fport, |
1bbdceef | 42 | inet6_ehash_secret + net_hash_mix(net)); |
b50026b5 HFS |
43 | } |
44 | ||
b1a7ffcb DV |
45 | /* |
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 | */ | |
d86e0dac PE |
51 | struct sock *__inet6_lookup_established(struct net *net, |
52 | struct inet_hashinfo *hashinfo, | |
b1a7ffcb | 53 | const struct in6_addr *saddr, |
d2ecd9cc | 54 | const __be16 sport, |
b1a7ffcb DV |
55 | const struct in6_addr *daddr, |
56 | const u16 hnum, | |
57 | const int dif) | |
58 | { | |
59 | struct sock *sk; | |
3ab5aee7 | 60 | const struct hlist_nulls_node *node; |
4f765d84 | 61 | const __portpair ports = INET_COMBINED_PORTS(sport, hnum); |
b1a7ffcb DV |
62 | /* Optimize here for direct hit, only listening connections can |
63 | * have wildcards anyways. | |
64 | */ | |
33de014c | 65 | unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport); |
f373b53b | 66 | unsigned int slot = hash & hashinfo->ehash_mask; |
3ab5aee7 | 67 | struct inet_ehash_bucket *head = &hashinfo->ehash[slot]; |
b1a7ffcb | 68 | |
3ab5aee7 ED |
69 | |
70 | rcu_read_lock(); | |
71 | begin: | |
72 | sk_nulls_for_each_rcu(sk, node, &head->chain) { | |
ce43b03e ED |
73 | if (sk->sk_hash != hash) |
74 | continue; | |
efe4208f ED |
75 | if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif)) |
76 | continue; | |
05dbc7b5 ED |
77 | if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) |
78 | goto out; | |
79 | ||
efe4208f ED |
80 | if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) { |
81 | sock_gen_put(sk); | |
82 | goto begin; | |
3ab5aee7 | 83 | } |
efe4208f | 84 | goto found; |
b1a7ffcb | 85 | } |
3ab5aee7 ED |
86 | if (get_nulls_value(node) != slot) |
87 | goto begin; | |
3ab5aee7 | 88 | out: |
05dbc7b5 ED |
89 | sk = NULL; |
90 | found: | |
3ab5aee7 | 91 | rcu_read_unlock(); |
b1a7ffcb DV |
92 | return sk; |
93 | } | |
94 | EXPORT_SYMBOL(__inet6_lookup_established); | |
95 | ||
42b16b3f | 96 | static inline int compute_score(struct sock *sk, struct net *net, |
c25eb3bf ED |
97 | const unsigned short hnum, |
98 | const struct in6_addr *daddr, | |
99 | const int dif) | |
100 | { | |
101 | int score = -1; | |
102 | ||
c720c7e8 | 103 | if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum && |
c25eb3bf | 104 | sk->sk_family == PF_INET6) { |
c25eb3bf ED |
105 | |
106 | score = 1; | |
efe4208f ED |
107 | if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { |
108 | if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) | |
c25eb3bf ED |
109 | 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 | } | |
70da268b ED |
117 | if (sk->sk_incoming_cpu == raw_smp_processor_id()) |
118 | score++; | |
c25eb3bf ED |
119 | } |
120 | return score; | |
121 | } | |
122 | ||
d86e0dac | 123 | struct sock *inet6_lookup_listener(struct net *net, |
5ba24953 TH |
124 | struct inet_hashinfo *hashinfo, const struct in6_addr *saddr, |
125 | const __be16 sport, const struct in6_addr *daddr, | |
d86e0dac | 126 | const unsigned short hnum, const int dif) |
5324a040 ACM |
127 | { |
128 | struct sock *sk; | |
c25eb3bf ED |
129 | const struct hlist_nulls_node *node; |
130 | struct sock *result; | |
5ba24953 TH |
131 | int score, hiscore, matches = 0, reuseport = 0; |
132 | u32 phash = 0; | |
c25eb3bf ED |
133 | unsigned int hash = inet_lhashfn(net, hnum); |
134 | struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash]; | |
135 | ||
136 | rcu_read_lock(); | |
137 | begin: | |
138 | result = NULL; | |
5ba24953 | 139 | hiscore = 0; |
c25eb3bf ED |
140 | 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; | |
5ba24953 TH |
145 | 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++; | |
8fc54f68 | 153 | if (reciprocal_scale(phash, matches) == 0) |
5ba24953 TH |
154 | result = sk; |
155 | phash = next_pseudo_random32(phash); | |
5324a040 ACM |
156 | } |
157 | } | |
c25eb3bf ED |
158 | /* |
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(); | |
5324a040 ACM |
175 | return result; |
176 | } | |
5324a040 ACM |
177 | EXPORT_SYMBOL_GPL(inet6_lookup_listener); |
178 | ||
d86e0dac | 179 | struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo, |
d2ecd9cc AV |
180 | const struct in6_addr *saddr, const __be16 sport, |
181 | const struct in6_addr *daddr, const __be16 dport, | |
5324a040 ACM |
182 | const int dif) |
183 | { | |
184 | struct sock *sk; | |
185 | ||
186 | local_bh_disable(); | |
d86e0dac | 187 | sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif); |
5324a040 ACM |
188 | local_bh_enable(); |
189 | ||
190 | return sk; | |
191 | } | |
5324a040 | 192 | EXPORT_SYMBOL_GPL(inet6_lookup); |
d8313f5c ACM |
193 | |
194 | static 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; | |
3759fa9c | 199 | struct inet_sock *inet = inet_sk(sk); |
efe4208f ED |
200 | const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr; |
201 | const struct in6_addr *saddr = &sk->sk_v6_daddr; | |
d8313f5c | 202 | const int dif = sk->sk_bound_dev_if; |
c720c7e8 | 203 | const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport); |
33de014c PE |
204 | struct net *net = sock_net(sk); |
205 | const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr, | |
c720c7e8 | 206 | inet->inet_dport); |
d8313f5c | 207 | struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash); |
9db66bdc | 208 | spinlock_t *lock = inet_ehash_lockp(hinfo, hash); |
d8313f5c | 209 | struct sock *sk2; |
3ab5aee7 | 210 | const struct hlist_nulls_node *node; |
05dbc7b5 | 211 | struct inet_timewait_sock *tw = NULL; |
d8313f5c | 212 | |
9db66bdc | 213 | spin_lock(lock); |
d8313f5c | 214 | |
05dbc7b5 | 215 | sk_nulls_for_each(sk2, node, &head->chain) { |
ce43b03e ED |
216 | if (sk2->sk_hash != hash) |
217 | continue; | |
d8313f5c | 218 | |
efe4208f ED |
219 | if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) { |
220 | if (sk2->sk_state == TCP_TIME_WAIT) { | |
05dbc7b5 ED |
221 | tw = inet_twsk(sk2); |
222 | if (twsk_unique(sk, sk2, twp)) | |
efe4208f | 223 | break; |
05dbc7b5 | 224 | } |
d8313f5c | 225 | goto not_unique; |
efe4208f | 226 | } |
d8313f5c ACM |
227 | } |
228 | ||
3759fa9c | 229 | /* Must record num and sport now. Otherwise we will see |
efe4208f ED |
230 | * in hash table socket with a funny identity. |
231 | */ | |
c720c7e8 ED |
232 | inet->inet_num = lport; |
233 | inet->inet_sport = htons(lport); | |
13475a30 | 234 | sk->sk_hash = hash; |
547b792c | 235 | WARN_ON(!sk_unhashed(sk)); |
3ab5aee7 | 236 | __sk_nulls_add_node_rcu(sk, &head->chain); |
13475a30 | 237 | if (tw) { |
fc01538f | 238 | sk_nulls_del_node_init_rcu((struct sock *)tw); |
13475a30 ED |
239 | NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED); |
240 | } | |
9db66bdc | 241 | spin_unlock(lock); |
c29a0bc4 | 242 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
d8313f5c | 243 | |
13475a30 | 244 | if (twp) { |
d8313f5c | 245 | *twp = tw; |
13475a30 | 246 | } else if (tw) { |
d8313f5c | 247 | /* Silly. Should hash-dance instead... */ |
dbe7faa4 | 248 | inet_twsk_deschedule_put(tw); |
d8313f5c ACM |
249 | } |
250 | return 0; | |
251 | ||
252 | not_unique: | |
9db66bdc | 253 | spin_unlock(lock); |
d8313f5c ACM |
254 | return -EADDRNOTAVAIL; |
255 | } | |
256 | ||
e2baad9e | 257 | static u32 inet6_sk_port_offset(const struct sock *sk) |
d8313f5c ACM |
258 | { |
259 | const struct inet_sock *inet = inet_sk(sk); | |
efe4208f ED |
260 | |
261 | return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32, | |
262 | sk->sk_v6_daddr.s6_addr32, | |
c720c7e8 | 263 | inet->inet_dport); |
d8313f5c ACM |
264 | } |
265 | ||
266 | int inet6_hash_connect(struct inet_timewait_death_row *death_row, | |
267 | struct sock *sk) | |
268 | { | |
e2baad9e ED |
269 | 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, | |
b4d6444e | 274 | __inet6_check_established); |
d8313f5c | 275 | } |
d8313f5c | 276 | EXPORT_SYMBOL_GPL(inet6_hash_connect); |