xref: /openbmc/linux/net/ipv6/inet6_hashtables.c (revision 545e4006)
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 sock *sk)
26 {
27 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
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 		list = &inet_ehash_bucket(hashinfo, hash)->chain;
41 		lock = inet_ehash_lockp(hashinfo, hash);
42 		write_lock(lock);
43 	}
44 
45 	__sk_add_node(sk, list);
46 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
47 	write_unlock(lock);
48 }
49 EXPORT_SYMBOL(__inet6_hash);
50 
51 /*
52  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
53  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
54  *
55  * The sockhash lock must be held as a reader here.
56  */
57 struct sock *__inet6_lookup_established(struct net *net,
58 					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(net, daddr, hnum, saddr, sport);
72 	struct inet_ehash_bucket *head = inet_ehash_bucket(hashinfo, hash);
73 	rwlock_t *lock = inet_ehash_lockp(hashinfo, hash);
74 
75 	prefetch(head->chain.first);
76 	read_lock(lock);
77 	sk_for_each(sk, node, &head->chain) {
78 		/* For IPV6 do the cheaper port and family tests first. */
79 		if (INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif))
80 			goto hit; /* You sunk my battleship! */
81 	}
82 	/* Must check for a TIME_WAIT'er before going to listener hash. */
83 	sk_for_each(sk, node, &head->twchain) {
84 		if (INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif))
85 			goto hit;
86 	}
87 	read_unlock(lock);
88 	return NULL;
89 
90 hit:
91 	sock_hold(sk);
92 	read_unlock(lock);
93 	return sk;
94 }
95 EXPORT_SYMBOL(__inet6_lookup_established);
96 
97 struct sock *inet6_lookup_listener(struct net *net,
98 		struct inet_hashinfo *hashinfo, const struct in6_addr *daddr,
99 		const unsigned short hnum, const int dif)
100 {
101 	struct sock *sk;
102 	const struct hlist_node *node;
103 	struct sock *result = NULL;
104 	int score, hiscore = 0;
105 
106 	read_lock(&hashinfo->lhash_lock);
107 	sk_for_each(sk, node,
108 			&hashinfo->listening_hash[inet_lhashfn(net, hnum)]) {
109 		if (net_eq(sock_net(sk), net) && inet_sk(sk)->num == hnum &&
110 				sk->sk_family == PF_INET6) {
111 			const struct ipv6_pinfo *np = inet6_sk(sk);
112 
113 			score = 1;
114 			if (!ipv6_addr_any(&np->rcv_saddr)) {
115 				if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
116 					continue;
117 				score++;
118 			}
119 			if (sk->sk_bound_dev_if) {
120 				if (sk->sk_bound_dev_if != dif)
121 					continue;
122 				score++;
123 			}
124 			if (score == 3) {
125 				result = sk;
126 				break;
127 			}
128 			if (score > hiscore) {
129 				hiscore = score;
130 				result = sk;
131 			}
132 		}
133 	}
134 	if (result)
135 		sock_hold(result);
136 	read_unlock(&hashinfo->lhash_lock);
137 	return result;
138 }
139 
140 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
141 
142 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
143 			  const struct in6_addr *saddr, const __be16 sport,
144 			  const struct in6_addr *daddr, const __be16 dport,
145 			  const int dif)
146 {
147 	struct sock *sk;
148 
149 	local_bh_disable();
150 	sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
151 	local_bh_enable();
152 
153 	return sk;
154 }
155 
156 EXPORT_SYMBOL_GPL(inet6_lookup);
157 
158 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
159 				     struct sock *sk, const __u16 lport,
160 				     struct inet_timewait_sock **twp)
161 {
162 	struct inet_hashinfo *hinfo = death_row->hashinfo;
163 	struct inet_sock *inet = inet_sk(sk);
164 	const struct ipv6_pinfo *np = inet6_sk(sk);
165 	const struct in6_addr *daddr = &np->rcv_saddr;
166 	const struct in6_addr *saddr = &np->daddr;
167 	const int dif = sk->sk_bound_dev_if;
168 	const __portpair ports = INET_COMBINED_PORTS(inet->dport, lport);
169 	struct net *net = sock_net(sk);
170 	const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
171 						inet->dport);
172 	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
173 	rwlock_t *lock = inet_ehash_lockp(hinfo, hash);
174 	struct sock *sk2;
175 	const struct hlist_node *node;
176 	struct inet_timewait_sock *tw;
177 
178 	prefetch(head->chain.first);
179 	write_lock(lock);
180 
181 	/* Check TIME-WAIT sockets first. */
182 	sk_for_each(sk2, node, &head->twchain) {
183 		tw = inet_twsk(sk2);
184 
185 		if (INET6_TW_MATCH(sk2, net, hash, saddr, daddr, ports, dif)) {
186 			if (twsk_unique(sk, sk2, twp))
187 				goto unique;
188 			else
189 				goto not_unique;
190 		}
191 	}
192 	tw = NULL;
193 
194 	/* And established part... */
195 	sk_for_each(sk2, node, &head->chain) {
196 		if (INET6_MATCH(sk2, net, hash, saddr, daddr, ports, dif))
197 			goto not_unique;
198 	}
199 
200 unique:
201 	/* Must record num and sport now. Otherwise we will see
202 	 * in hash table socket with a funny identity. */
203 	inet->num = lport;
204 	inet->sport = htons(lport);
205 	BUG_TRAP(sk_unhashed(sk));
206 	__sk_add_node(sk, &head->chain);
207 	sk->sk_hash = hash;
208 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
209 	write_unlock(lock);
210 
211 	if (twp != NULL) {
212 		*twp = tw;
213 		NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITRECYCLED);
214 	} else if (tw != NULL) {
215 		/* Silly. Should hash-dance instead... */
216 		inet_twsk_deschedule(tw, death_row);
217 		NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITRECYCLED);
218 
219 		inet_twsk_put(tw);
220 	}
221 	return 0;
222 
223 not_unique:
224 	write_unlock(lock);
225 	return -EADDRNOTAVAIL;
226 }
227 
228 static inline u32 inet6_sk_port_offset(const struct sock *sk)
229 {
230 	const struct inet_sock *inet = inet_sk(sk);
231 	const struct ipv6_pinfo *np = inet6_sk(sk);
232 	return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
233 					  np->daddr.s6_addr32,
234 					  inet->dport);
235 }
236 
237 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
238 		       struct sock *sk)
239 {
240 	return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
241 			__inet6_check_established, __inet6_hash);
242 }
243 
244 EXPORT_SYMBOL_GPL(inet6_hash_connect);
245