xref: /linux/net/ipv6/inet6_hashtables.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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/secure_seq.h>
24 #include <net/ip.h>
25 
26 int __inet6_hash(struct sock *sk, struct inet_timewait_sock *tw)
27 {
28 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
29 	int twrefcnt = 0;
30 
31 	WARN_ON(!sk_unhashed(sk));
32 
33 	if (sk->sk_state == TCP_LISTEN) {
34 		struct inet_listen_hashbucket *ilb;
35 
36 		ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
37 		spin_lock(&ilb->lock);
38 		__sk_nulls_add_node_rcu(sk, &ilb->head);
39 		spin_unlock(&ilb->lock);
40 	} else {
41 		unsigned int hash;
42 		struct hlist_nulls_head *list;
43 		spinlock_t *lock;
44 
45 		sk->sk_hash = hash = inet6_sk_ehashfn(sk);
46 		list = &inet_ehash_bucket(hashinfo, hash)->chain;
47 		lock = inet_ehash_lockp(hashinfo, hash);
48 		spin_lock(lock);
49 		__sk_nulls_add_node_rcu(sk, list);
50 		if (tw) {
51 			WARN_ON(sk->sk_hash != tw->tw_hash);
52 			twrefcnt = inet_twsk_unhash(tw);
53 		}
54 		spin_unlock(lock);
55 	}
56 
57 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
58 	return twrefcnt;
59 }
60 EXPORT_SYMBOL(__inet6_hash);
61 
62 /*
63  * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
64  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
65  *
66  * The sockhash lock must be held as a reader here.
67  */
68 struct sock *__inet6_lookup_established(struct net *net,
69 					struct inet_hashinfo *hashinfo,
70 					   const struct in6_addr *saddr,
71 					   const __be16 sport,
72 					   const struct in6_addr *daddr,
73 					   const u16 hnum,
74 					   const int dif)
75 {
76 	struct sock *sk;
77 	const struct hlist_nulls_node *node;
78 	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
79 	/* Optimize here for direct hit, only listening connections can
80 	 * have wildcards anyways.
81 	 */
82 	unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
83 	unsigned int slot = hash & hashinfo->ehash_mask;
84 	struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
85 
86 
87 	rcu_read_lock();
88 begin:
89 	sk_nulls_for_each_rcu(sk, node, &head->chain) {
90 		if (sk->sk_hash != hash)
91 			continue;
92 		if (likely(INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
93 			if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
94 				goto begintw;
95 			if (unlikely(!INET6_MATCH(sk, net, saddr, daddr,
96 						  ports, dif))) {
97 				sock_put(sk);
98 				goto begin;
99 			}
100 		goto out;
101 		}
102 	}
103 	if (get_nulls_value(node) != slot)
104 		goto begin;
105 
106 begintw:
107 	/* Must check for a TIME_WAIT'er before going to listener hash. */
108 	sk_nulls_for_each_rcu(sk, node, &head->twchain) {
109 		if (sk->sk_hash != hash)
110 			continue;
111 		if (likely(INET6_TW_MATCH(sk, net, saddr, daddr,
112 					  ports, dif))) {
113 			if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) {
114 				sk = NULL;
115 				goto out;
116 			}
117 			if (unlikely(!INET6_TW_MATCH(sk, net, saddr, daddr,
118 						     ports, dif))) {
119 				sock_put(sk);
120 				goto begintw;
121 			}
122 			goto out;
123 		}
124 	}
125 	if (get_nulls_value(node) != slot)
126 		goto begintw;
127 	sk = NULL;
128 out:
129 	rcu_read_unlock();
130 	return sk;
131 }
132 EXPORT_SYMBOL(__inet6_lookup_established);
133 
134 static inline int compute_score(struct sock *sk, struct net *net,
135 				const unsigned short hnum,
136 				const struct in6_addr *daddr,
137 				const int dif)
138 {
139 	int score = -1;
140 
141 	if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
142 	    sk->sk_family == PF_INET6) {
143 		const struct ipv6_pinfo *np = inet6_sk(sk);
144 
145 		score = 1;
146 		if (!ipv6_addr_any(&np->rcv_saddr)) {
147 			if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
148 				return -1;
149 			score++;
150 		}
151 		if (sk->sk_bound_dev_if) {
152 			if (sk->sk_bound_dev_if != dif)
153 				return -1;
154 			score++;
155 		}
156 	}
157 	return score;
158 }
159 
160 struct sock *inet6_lookup_listener(struct net *net,
161 		struct inet_hashinfo *hashinfo, const struct in6_addr *daddr,
162 		const unsigned short hnum, const int dif)
163 {
164 	struct sock *sk;
165 	const struct hlist_nulls_node *node;
166 	struct sock *result;
167 	int score, hiscore;
168 	unsigned int hash = inet_lhashfn(net, hnum);
169 	struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
170 
171 	rcu_read_lock();
172 begin:
173 	result = NULL;
174 	hiscore = -1;
175 	sk_nulls_for_each(sk, node, &ilb->head) {
176 		score = compute_score(sk, net, hnum, daddr, dif);
177 		if (score > hiscore) {
178 			hiscore = score;
179 			result = sk;
180 		}
181 	}
182 	/*
183 	 * if the nulls value we got at the end of this lookup is
184 	 * not the expected one, we must restart lookup.
185 	 * We probably met an item that was moved to another chain.
186 	 */
187 	if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
188 		goto begin;
189 	if (result) {
190 		if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
191 			result = NULL;
192 		else if (unlikely(compute_score(result, net, hnum, daddr,
193 				  dif) < hiscore)) {
194 			sock_put(result);
195 			goto begin;
196 		}
197 	}
198 	rcu_read_unlock();
199 	return result;
200 }
201 
202 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
203 
204 struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
205 			  const struct in6_addr *saddr, const __be16 sport,
206 			  const struct in6_addr *daddr, const __be16 dport,
207 			  const int dif)
208 {
209 	struct sock *sk;
210 
211 	local_bh_disable();
212 	sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
213 	local_bh_enable();
214 
215 	return sk;
216 }
217 
218 EXPORT_SYMBOL_GPL(inet6_lookup);
219 
220 static int __inet6_check_established(struct inet_timewait_death_row *death_row,
221 				     struct sock *sk, const __u16 lport,
222 				     struct inet_timewait_sock **twp)
223 {
224 	struct inet_hashinfo *hinfo = death_row->hashinfo;
225 	struct inet_sock *inet = inet_sk(sk);
226 	const struct ipv6_pinfo *np = inet6_sk(sk);
227 	const struct in6_addr *daddr = &np->rcv_saddr;
228 	const struct in6_addr *saddr = &np->daddr;
229 	const int dif = sk->sk_bound_dev_if;
230 	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
231 	struct net *net = sock_net(sk);
232 	const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
233 						inet->inet_dport);
234 	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
235 	spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
236 	struct sock *sk2;
237 	const struct hlist_nulls_node *node;
238 	struct inet_timewait_sock *tw;
239 	int twrefcnt = 0;
240 
241 	spin_lock(lock);
242 
243 	/* Check TIME-WAIT sockets first. */
244 	sk_nulls_for_each(sk2, node, &head->twchain) {
245 		if (sk2->sk_hash != hash)
246 			continue;
247 
248 		if (likely(INET6_TW_MATCH(sk2, net, saddr, daddr,
249 					  ports, dif))) {
250 			tw = inet_twsk(sk2);
251 			if (twsk_unique(sk, sk2, twp))
252 				goto unique;
253 			else
254 				goto not_unique;
255 		}
256 	}
257 	tw = NULL;
258 
259 	/* And established part... */
260 	sk_nulls_for_each(sk2, node, &head->chain) {
261 		if (sk2->sk_hash != hash)
262 			continue;
263 		if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif)))
264 			goto not_unique;
265 	}
266 
267 unique:
268 	/* Must record num and sport now. Otherwise we will see
269 	 * in hash table socket with a funny identity. */
270 	inet->inet_num = lport;
271 	inet->inet_sport = htons(lport);
272 	sk->sk_hash = hash;
273 	WARN_ON(!sk_unhashed(sk));
274 	__sk_nulls_add_node_rcu(sk, &head->chain);
275 	if (tw) {
276 		twrefcnt = inet_twsk_unhash(tw);
277 		NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
278 	}
279 	spin_unlock(lock);
280 	if (twrefcnt)
281 		inet_twsk_put(tw);
282 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
283 
284 	if (twp) {
285 		*twp = tw;
286 	} else if (tw) {
287 		/* Silly. Should hash-dance instead... */
288 		inet_twsk_deschedule(tw, death_row);
289 
290 		inet_twsk_put(tw);
291 	}
292 	return 0;
293 
294 not_unique:
295 	spin_unlock(lock);
296 	return -EADDRNOTAVAIL;
297 }
298 
299 static inline u32 inet6_sk_port_offset(const struct sock *sk)
300 {
301 	const struct inet_sock *inet = inet_sk(sk);
302 	const struct ipv6_pinfo *np = inet6_sk(sk);
303 	return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
304 					  np->daddr.s6_addr32,
305 					  inet->inet_dport);
306 }
307 
308 int inet6_hash_connect(struct inet_timewait_death_row *death_row,
309 		       struct sock *sk)
310 {
311 	return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
312 			__inet6_check_established, __inet6_hash);
313 }
314 
315 EXPORT_SYMBOL_GPL(inet6_hash_connect);
316