inet6_hashtables.c 7.5 KB
Newer Older
1 2 3 4 5 6 7
/*
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		Generic INET6 transport hashtables
 *
8
 * Authors:	Lotsa people, from code originally in tcp, generalised here
9
 *		by Arnaldo Carvalho de Melo <acme@mandriva.com>
10 11 12 13 14 15 16 17
 *
 *	This program is free software; you can redistribute it and/or
 *      modify it under the terms of the GNU General Public License
 *      as published by the Free Software Foundation; either version
 *      2 of the License, or (at your option) any later version.
 */

#include <linux/module.h>
18
#include <linux/random.h>
19 20 21 22

#include <net/inet_connection_sock.h>
#include <net/inet_hashtables.h>
#include <net/inet6_hashtables.h>
23
#include <net/secure_seq.h>
24
#include <net/ip.h>
25

26 27 28
u32 inet6_ehashfn(const struct net *net,
		  const struct in6_addr *laddr, const u16 lport,
		  const struct in6_addr *faddr, const __be16 fport)
29
{
30 31 32 33 34 35 36 37 38 39 40
	static u32 inet6_ehash_secret __read_mostly;
	static u32 ipv6_hash_secret __read_mostly;

	u32 lhash, fhash;

	net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
	net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));

	lhash = (__force u32)laddr->s6_addr32[3];
	fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);

41
	return __inet6_ehashfn(lhash, lport, fhash, fport,
42
			       inet6_ehash_secret + net_hash_mix(net));
43 44
}

45 46 47 48 49 50
/*
 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
 *
 * The sockhash lock must be held as a reader here.
 */
51 52
struct sock *__inet6_lookup_established(struct net *net,
					struct inet_hashinfo *hashinfo,
53
					   const struct in6_addr *saddr,
A
Al Viro 已提交
54
					   const __be16 sport,
55 56 57 58 59
					   const struct in6_addr *daddr,
					   const u16 hnum,
					   const int dif)
{
	struct sock *sk;
60
	const struct hlist_nulls_node *node;
A
Al Viro 已提交
61
	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
62 63 64
	/* Optimize here for direct hit, only listening connections can
	 * have wildcards anyways.
	 */
65
	unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
66
	unsigned int slot = hash & hashinfo->ehash_mask;
67
	struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
68

69 70 71 72

	rcu_read_lock();
begin:
	sk_nulls_for_each_rcu(sk, node, &head->chain) {
73 74
		if (sk->sk_hash != hash)
			continue;
75 76
		if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif))
			continue;
E
Eric Dumazet 已提交
77 78 79
		if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
			goto out;

80 81 82
		if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
			sock_gen_put(sk);
			goto begin;
83
		}
84
		goto found;
85
	}
86 87 88
	if (get_nulls_value(node) != slot)
		goto begin;
out:
E
Eric Dumazet 已提交
89 90
	sk = NULL;
found:
91
	rcu_read_unlock();
92 93 94 95
	return sk;
}
EXPORT_SYMBOL(__inet6_lookup_established);

96
static inline int compute_score(struct sock *sk, struct net *net,
97 98 99 100 101 102
				const unsigned short hnum,
				const struct in6_addr *daddr,
				const int dif)
{
	int score = -1;

E
Eric Dumazet 已提交
103
	if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
104 105 106
	    sk->sk_family == PF_INET6) {

		score = 1;
107 108
		if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
			if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
109 110 111 112 113 114 115 116 117 118 119 120
				return -1;
			score++;
		}
		if (sk->sk_bound_dev_if) {
			if (sk->sk_bound_dev_if != dif)
				return -1;
			score++;
		}
	}
	return score;
}

121
struct sock *inet6_lookup_listener(struct net *net,
122 123
		struct inet_hashinfo *hashinfo, const struct in6_addr *saddr,
		const __be16 sport, const struct in6_addr *daddr,
124
		const unsigned short hnum, const int dif)
125 126
{
	struct sock *sk;
127 128
	const struct hlist_nulls_node *node;
	struct sock *result;
129 130
	int score, hiscore, matches = 0, reuseport = 0;
	u32 phash = 0;
131 132 133 134 135 136
	unsigned int hash = inet_lhashfn(net, hnum);
	struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];

	rcu_read_lock();
begin:
	result = NULL;
137
	hiscore = 0;
138 139 140 141 142
	sk_nulls_for_each(sk, node, &ilb->head) {
		score = compute_score(sk, net, hnum, daddr, dif);
		if (score > hiscore) {
			hiscore = score;
			result = sk;
143 144 145 146 147 148 149 150
			reuseport = sk->sk_reuseport;
			if (reuseport) {
				phash = inet6_ehashfn(net, daddr, hnum,
						      saddr, sport);
				matches = 1;
			}
		} else if (score == hiscore && reuseport) {
			matches++;
151
			if (reciprocal_scale(phash, matches) == 0)
152 153
				result = sk;
			phash = next_pseudo_random32(phash);
154 155
		}
	}
156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172
	/*
	 * if the nulls value we got at the end of this lookup is
	 * not the expected one, we must restart lookup.
	 * We probably met an item that was moved to another chain.
	 */
	if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
		goto begin;
	if (result) {
		if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
			result = NULL;
		else if (unlikely(compute_score(result, net, hnum, daddr,
				  dif) < hiscore)) {
			sock_put(result);
			goto begin;
		}
	}
	rcu_read_unlock();
173 174 175 176
	return result;
}
EXPORT_SYMBOL_GPL(inet6_lookup_listener);

177
struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
A
Al Viro 已提交
178 179
			  const struct in6_addr *saddr, const __be16 sport,
			  const struct in6_addr *daddr, const __be16 dport,
180 181 182 183 184
			  const int dif)
{
	struct sock *sk;

	local_bh_disable();
185
	sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
186 187 188 189 190
	local_bh_enable();

	return sk;
}
EXPORT_SYMBOL_GPL(inet6_lookup);
191 192 193 194 195 196

static int __inet6_check_established(struct inet_timewait_death_row *death_row,
				     struct sock *sk, const __u16 lport,
				     struct inet_timewait_sock **twp)
{
	struct inet_hashinfo *hinfo = death_row->hashinfo;
197
	struct inet_sock *inet = inet_sk(sk);
198 199
	const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
	const struct in6_addr *saddr = &sk->sk_v6_daddr;
200
	const int dif = sk->sk_bound_dev_if;
E
Eric Dumazet 已提交
201
	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
202 203
	struct net *net = sock_net(sk);
	const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
E
Eric Dumazet 已提交
204
						inet->inet_dport);
205
	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
206
	spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
207
	struct sock *sk2;
208
	const struct hlist_nulls_node *node;
E
Eric Dumazet 已提交
209
	struct inet_timewait_sock *tw = NULL;
210
	int twrefcnt = 0;
211

212
	spin_lock(lock);
213

E
Eric Dumazet 已提交
214
	sk_nulls_for_each(sk2, node, &head->chain) {
215 216
		if (sk2->sk_hash != hash)
			continue;
217

218 219
		if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) {
			if (sk2->sk_state == TCP_TIME_WAIT) {
E
Eric Dumazet 已提交
220 221
				tw = inet_twsk(sk2);
				if (twsk_unique(sk, sk2, twp))
222
					break;
E
Eric Dumazet 已提交
223
			}
224
			goto not_unique;
225
		}
226 227
	}

228
	/* Must record num and sport now. Otherwise we will see
229 230
	 * in hash table socket with a funny identity.
	 */
E
Eric Dumazet 已提交
231 232
	inet->inet_num = lport;
	inet->inet_sport = htons(lport);
233
	sk->sk_hash = hash;
234
	WARN_ON(!sk_unhashed(sk));
235
	__sk_nulls_add_node_rcu(sk, &head->chain);
236 237 238 239
	if (tw) {
		twrefcnt = inet_twsk_unhash(tw);
		NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
	}
240
	spin_unlock(lock);
241 242
	if (twrefcnt)
		inet_twsk_put(tw);
243
	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
244

245
	if (twp) {
246
		*twp = tw;
247
	} else if (tw) {
248 249 250 251 252 253 254 255
		/* Silly. Should hash-dance instead... */
		inet_twsk_deschedule(tw, death_row);

		inet_twsk_put(tw);
	}
	return 0;

not_unique:
256
	spin_unlock(lock);
257 258 259 260 261 262
	return -EADDRNOTAVAIL;
}

static inline u32 inet6_sk_port_offset(const struct sock *sk)
{
	const struct inet_sock *inet = inet_sk(sk);
263 264 265

	return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
					  sk->sk_v6_daddr.s6_addr32,
E
Eric Dumazet 已提交
266
					  inet->inet_dport);
267 268 269 270 271
}

int inet6_hash_connect(struct inet_timewait_death_row *death_row,
		       struct sock *sk)
{
272
	return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
E
Eric Dumazet 已提交
273
			__inet6_check_established, __inet_hash_nolisten);
274 275
}
EXPORT_SYMBOL_GPL(inet6_hash_connect);