xref: /linux/net/ipv6/inet6_hashtables.c (revision e26207a3819684e9b4450a2d30bdd065fa92d9c7)
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 int __inet6_hash(struct sock *sk, struct inet_timewait_sock *tw)
26 {
27 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
28 	int twrefcnt = 0;
29 
30 	WARN_ON(!sk_unhashed(sk));
31 
32 	if (sk->sk_state == TCP_LISTEN) {
33 		struct inet_listen_hashbucket *ilb;
34 
35 		ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
36 		spin_lock(&ilb->lock);
37 		__sk_nulls_add_node_rcu(sk, &ilb->head);
38 		spin_unlock(&ilb->lock);
39 	} else {
40 		unsigned int hash;
41 		struct hlist_nulls_head *list;
42 		spinlock_t *lock;
43 
44 		sk->sk_hash = hash = inet6_sk_ehashfn(sk);
45 		list = &inet_ehash_bucket(hashinfo, hash)->chain;
46 		lock = inet_ehash_lockp(hashinfo, hash);
47 		spin_lock(lock);
48 		__sk_nulls_add_node_rcu(sk, list);
49 		if (tw) {
50 			WARN_ON(sk->sk_hash != tw->tw_hash);
51 			twrefcnt = inet_twsk_unhash(tw);
52 		}
53 		spin_unlock(lock);
54 	}
55 
56 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
57 	return twrefcnt;
58 }
59 EXPORT_SYMBOL(__inet6_hash);
60 
61 /*
62  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
63  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
64  *
65  * The sockhash lock must be held as a reader here.
66  */
67 struct sock *__inet6_lookup_established(struct net *net,
68 					struct inet_hashinfo *hashinfo,
69 					   const struct in6_addr *saddr,
70 					   const __be16 sport,
71 					   const struct in6_addr *daddr,
72 					   const u16 hnum,
73 					   const int dif)
74 {
75 	struct sock *sk;
76 	const struct hlist_nulls_node *node;
77 	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
78 	/* Optimize here for direct hit, only listening connections can
79 	 * have wildcards anyways.
80 	 */
81 	unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
82 	unsigned int slot = hash & hashinfo->ehash_mask;
83 	struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
84 
85 
86 	rcu_read_lock();
87 begin:
88 	sk_nulls_for_each_rcu(sk, node, &head->chain) {
89 		/* For IPV6 do the cheaper port and family tests first. */
90 		if (INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
91 			if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
92 				goto begintw;
93 			if (!INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
94 				sock_put(sk);
95 				goto begin;
96 			}
97 		goto out;
98 		}
99 	}
100 	if (get_nulls_value(node) != slot)
101 		goto begin;
102 
103 begintw:
104 	/* Must check for a TIME_WAIT'er before going to listener hash. */
105 	sk_nulls_for_each_rcu(sk, node, &head->twchain) {
106 		if (INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
107 			if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) {
108 				sk = NULL;
109 				goto out;
110 			}
111 			if (!INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
112 				sock_put(sk);
113 				goto begintw;
114 			}
115 			goto out;
116 		}
117 	}
118 	if (get_nulls_value(node) != slot)
119 		goto begintw;
120 	sk = NULL;
121 out:
122 	rcu_read_unlock();
123 	return sk;
124 }
125 EXPORT_SYMBOL(__inet6_lookup_established);
126 
127 static int inline compute_score(struct sock *sk, struct net *net,
128 				const unsigned short hnum,
129 				const struct in6_addr *daddr,
130 				const int dif)
131 {
132 	int score = -1;
133 
134 	if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
135 	    sk->sk_family == PF_INET6) {
136 		const struct ipv6_pinfo *np = inet6_sk(sk);
137 
138 		score = 1;
139 		if (!ipv6_addr_any(&np->rcv_saddr)) {
140 			if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
141 				return -1;
142 			score++;
143 		}
144 		if (sk->sk_bound_dev_if) {
145 			if (sk->sk_bound_dev_if != dif)
146 				return -1;
147 			score++;
148 		}
149 	}
150 	return score;
151 }
152 
153 struct sock *inet6_lookup_listener(struct net *net,
154 		struct inet_hashinfo *hashinfo, const struct in6_addr *daddr,
155 		const unsigned short hnum, const int dif)
156 {
157 	struct sock *sk;
158 	const struct hlist_nulls_node *node;
159 	struct sock *result;
160 	int score, hiscore;
161 	unsigned int hash = inet_lhashfn(net, hnum);
162 	struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
163 
164 	rcu_read_lock();
165 begin:
166 	result = NULL;
167 	hiscore = -1;
168 	sk_nulls_for_each(sk, node, &ilb->head) {
169 		score = compute_score(sk, net, hnum, daddr, dif);
170 		if (score > hiscore) {
171 			hiscore = score;
172 			result = sk;
173 		}
174 	}
175 	/*
176 	 * if the nulls value we got at the end of this lookup is
177 	 * not the expected one, we must restart lookup.
178 	 * We probably met an item that was moved to another chain.
179 	 */
180 	if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
181 		goto begin;
182 	if (result) {
183 		if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
184 			result = NULL;
185 		else if (unlikely(compute_score(result, net, hnum, daddr,
186 				  dif) < hiscore)) {
187 			sock_put(result);
188 			goto begin;
189 		}
190 	}
191 	rcu_read_unlock();
192 	return result;
193 }
194 
195 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
196 
197 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
198 			  const struct in6_addr *saddr, const __be16 sport,
199 			  const struct in6_addr *daddr, const __be16 dport,
200 			  const int dif)
201 {
202 	struct sock *sk;
203 
204 	local_bh_disable();
205 	sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
206 	local_bh_enable();
207 
208 	return sk;
209 }
210 
211 EXPORT_SYMBOL_GPL(inet6_lookup);
212 
213 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
214 				     struct sock *sk, const __u16 lport,
215 				     struct inet_timewait_sock **twp)
216 {
217 	struct inet_hashinfo *hinfo = death_row->hashinfo;
218 	struct inet_sock *inet = inet_sk(sk);
219 	const struct ipv6_pinfo *np = inet6_sk(sk);
220 	const struct in6_addr *daddr = &np->rcv_saddr;
221 	const struct in6_addr *saddr = &np->daddr;
222 	const int dif = sk->sk_bound_dev_if;
223 	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
224 	struct net *net = sock_net(sk);
225 	const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
226 						inet->inet_dport);
227 	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
228 	spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
229 	struct sock *sk2;
230 	const struct hlist_nulls_node *node;
231 	struct inet_timewait_sock *tw;
232 	int twrefcnt = 0;
233 
234 	spin_lock(lock);
235 
236 	/* Check TIME-WAIT sockets first. */
237 	sk_nulls_for_each(sk2, node, &head->twchain) {
238 		tw = inet_twsk(sk2);
239 
240 		if (INET6_TW_MATCH(sk2, net, hash, saddr, daddr, ports, dif)) {
241 			if (twsk_unique(sk, sk2, twp))
242 				goto unique;
243 			else
244 				goto not_unique;
245 		}
246 	}
247 	tw = NULL;
248 
249 	/* And established part... */
250 	sk_nulls_for_each(sk2, node, &head->chain) {
251 		if (INET6_MATCH(sk2, net, hash, saddr, daddr, ports, dif))
252 			goto not_unique;
253 	}
254 
255 unique:
256 	/* Must record num and sport now. Otherwise we will see
257 	 * in hash table socket with a funny identity. */
258 	inet->inet_num = lport;
259 	inet->inet_sport = htons(lport);
260 	sk->sk_hash = hash;
261 	WARN_ON(!sk_unhashed(sk));
262 	__sk_nulls_add_node_rcu(sk, &head->chain);
263 	if (tw) {
264 		twrefcnt = inet_twsk_unhash(tw);
265 		NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
266 	}
267 	spin_unlock(lock);
268 	if (twrefcnt)
269 		inet_twsk_put(tw);
270 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
271 
272 	if (twp) {
273 		*twp = tw;
274 	} else if (tw) {
275 		/* Silly. Should hash-dance instead... */
276 		inet_twsk_deschedule(tw, death_row);
277 
278 		inet_twsk_put(tw);
279 	}
280 	return 0;
281 
282 not_unique:
283 	spin_unlock(lock);
284 	return -EADDRNOTAVAIL;
285 }
286 
287 static inline u32 inet6_sk_port_offset(const struct sock *sk)
288 {
289 	const struct inet_sock *inet = inet_sk(sk);
290 	const struct ipv6_pinfo *np = inet6_sk(sk);
291 	return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
292 					  np->daddr.s6_addr32,
293 					  inet->inet_dport);
294 }
295 
296 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
297 		       struct sock *sk)
298 {
299 	return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
300 			__inet6_check_established, __inet6_hash);
301 }
302 
303 EXPORT_SYMBOL_GPL(inet6_hash_connect);
304