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.
6 * Generic INET6 transport hashtables
8 * Authors: Lotsa people, from code originally in tcp, generalised here
9 * by Arnaldo Carvalho de Melo <acme@mandriva.com>
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.
17 #include <linux/module.h>
18 #include <linux/random.h>
20 #include <net/addrconf.h>
21 #include <net/inet_connection_sock.h>
22 #include <net/inet_hashtables.h>
23 #include <net/inet6_hashtables.h>
24 #include <net/secure_seq.h>
26 #include <net/sock_reuseport.h>
28 u32
inet6_ehashfn(const struct net
*net
,
29 const struct in6_addr
*laddr
, const u16 lport
,
30 const struct in6_addr
*faddr
, const __be16 fport
)
32 static u32 inet6_ehash_secret __read_mostly
;
33 static u32 ipv6_hash_secret __read_mostly
;
37 net_get_random_once(&inet6_ehash_secret
, sizeof(inet6_ehash_secret
));
38 net_get_random_once(&ipv6_hash_secret
, sizeof(ipv6_hash_secret
));
40 lhash
= (__force u32
)laddr
->s6_addr32
[3];
41 fhash
= __ipv6_addr_jhash(faddr
, ipv6_hash_secret
);
43 return __inet6_ehashfn(lhash
, lport
, fhash
, fport
,
44 inet6_ehash_secret
+ net_hash_mix(net
));
48 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
49 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
51 * The sockhash lock must be held as a reader here.
53 struct sock
*__inet6_lookup_established(struct net
*net
,
54 struct inet_hashinfo
*hashinfo
,
55 const struct in6_addr
*saddr
,
57 const struct in6_addr
*daddr
,
62 const struct hlist_nulls_node
*node
;
63 const __portpair ports
= INET_COMBINED_PORTS(sport
, hnum
);
64 /* Optimize here for direct hit, only listening connections can
65 * have wildcards anyways.
67 unsigned int hash
= inet6_ehashfn(net
, daddr
, hnum
, saddr
, sport
);
68 unsigned int slot
= hash
& hashinfo
->ehash_mask
;
69 struct inet_ehash_bucket
*head
= &hashinfo
->ehash
[slot
];
74 sk_nulls_for_each_rcu(sk
, node
, &head
->chain
) {
75 if (sk
->sk_hash
!= hash
)
77 if (!INET6_MATCH(sk
, net
, saddr
, daddr
, ports
, dif
))
79 if (unlikely(!atomic_inc_not_zero(&sk
->sk_refcnt
)))
82 if (unlikely(!INET6_MATCH(sk
, net
, saddr
, daddr
, ports
, dif
))) {
88 if (get_nulls_value(node
) != slot
)
96 EXPORT_SYMBOL(__inet6_lookup_established
);
98 static inline int compute_score(struct sock
*sk
, struct net
*net
,
99 const unsigned short hnum
,
100 const struct in6_addr
*daddr
,
105 if (net_eq(sock_net(sk
), net
) && inet_sk(sk
)->inet_num
== hnum
&&
106 sk
->sk_family
== PF_INET6
) {
109 if (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
)) {
110 if (!ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, daddr
))
114 if (sk
->sk_bound_dev_if
) {
115 if (sk
->sk_bound_dev_if
!= dif
)
119 if (sk
->sk_incoming_cpu
== raw_smp_processor_id())
125 struct sock
*inet6_lookup_listener(struct net
*net
,
126 struct inet_hashinfo
*hashinfo
,
127 struct sk_buff
*skb
, int doff
,
128 const struct in6_addr
*saddr
,
129 const __be16 sport
, const struct in6_addr
*daddr
,
130 const unsigned short hnum
, const int dif
)
133 const struct hlist_nulls_node
*node
;
135 int score
, hiscore
, matches
= 0, reuseport
= 0;
136 bool select_ok
= true;
138 unsigned int hash
= inet_lhashfn(net
, hnum
);
139 struct inet_listen_hashbucket
*ilb
= &hashinfo
->listening_hash
[hash
];
145 sk_nulls_for_each(sk
, node
, &ilb
->head
) {
146 score
= compute_score(sk
, net
, hnum
, daddr
, dif
);
147 if (score
> hiscore
) {
150 reuseport
= sk
->sk_reuseport
;
152 phash
= inet6_ehashfn(net
, daddr
, hnum
,
156 sk2
= reuseport_select_sock(sk
, phash
,
165 } else if (score
== hiscore
&& reuseport
) {
167 if (reciprocal_scale(phash
, matches
) == 0)
169 phash
= next_pseudo_random32(phash
);
173 * if the nulls value we got at the end of this lookup is
174 * not the expected one, we must restart lookup.
175 * We probably met an item that was moved to another chain.
177 if (get_nulls_value(node
) != hash
+ LISTENING_NULLS_BASE
)
181 if (unlikely(!atomic_inc_not_zero(&result
->sk_refcnt
)))
183 else if (unlikely(compute_score(result
, net
, hnum
, daddr
,
193 EXPORT_SYMBOL_GPL(inet6_lookup_listener
);
195 struct sock
*inet6_lookup(struct net
*net
, struct inet_hashinfo
*hashinfo
,
196 struct sk_buff
*skb
, int doff
,
197 const struct in6_addr
*saddr
, const __be16 sport
,
198 const struct in6_addr
*daddr
, const __be16 dport
,
204 sk
= __inet6_lookup(net
, hashinfo
, skb
, doff
, saddr
, sport
, daddr
,
210 EXPORT_SYMBOL_GPL(inet6_lookup
);
212 static int __inet6_check_established(struct inet_timewait_death_row
*death_row
,
213 struct sock
*sk
, const __u16 lport
,
214 struct inet_timewait_sock
**twp
)
216 struct inet_hashinfo
*hinfo
= death_row
->hashinfo
;
217 struct inet_sock
*inet
= inet_sk(sk
);
218 const struct in6_addr
*daddr
= &sk
->sk_v6_rcv_saddr
;
219 const struct in6_addr
*saddr
= &sk
->sk_v6_daddr
;
220 const int dif
= sk
->sk_bound_dev_if
;
221 const __portpair ports
= INET_COMBINED_PORTS(inet
->inet_dport
, lport
);
222 struct net
*net
= sock_net(sk
);
223 const unsigned int hash
= inet6_ehashfn(net
, daddr
, lport
, saddr
,
225 struct inet_ehash_bucket
*head
= inet_ehash_bucket(hinfo
, hash
);
226 spinlock_t
*lock
= inet_ehash_lockp(hinfo
, hash
);
228 const struct hlist_nulls_node
*node
;
229 struct inet_timewait_sock
*tw
= NULL
;
233 sk_nulls_for_each(sk2
, node
, &head
->chain
) {
234 if (sk2
->sk_hash
!= hash
)
237 if (likely(INET6_MATCH(sk2
, net
, saddr
, daddr
, ports
, dif
))) {
238 if (sk2
->sk_state
== TCP_TIME_WAIT
) {
240 if (twsk_unique(sk
, sk2
, twp
))
247 /* Must record num and sport now. Otherwise we will see
248 * in hash table socket with a funny identity.
250 inet
->inet_num
= lport
;
251 inet
->inet_sport
= htons(lport
);
253 WARN_ON(!sk_unhashed(sk
));
254 __sk_nulls_add_node_rcu(sk
, &head
->chain
);
256 sk_nulls_del_node_init_rcu((struct sock
*)tw
);
257 NET_INC_STATS_BH(net
, LINUX_MIB_TIMEWAITRECYCLED
);
260 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
265 /* Silly. Should hash-dance instead... */
266 inet_twsk_deschedule_put(tw
);
272 return -EADDRNOTAVAIL
;
275 static u32
inet6_sk_port_offset(const struct sock
*sk
)
277 const struct inet_sock
*inet
= inet_sk(sk
);
279 return secure_ipv6_port_ephemeral(sk
->sk_v6_rcv_saddr
.s6_addr32
,
280 sk
->sk_v6_daddr
.s6_addr32
,
284 int inet6_hash_connect(struct inet_timewait_death_row
*death_row
,
289 if (!inet_sk(sk
)->inet_num
)
290 port_offset
= inet6_sk_port_offset(sk
);
291 return __inet_hash_connect(death_row
, sk
, port_offset
,
292 __inet6_check_established
);
294 EXPORT_SYMBOL_GPL(inet6_hash_connect
);
296 int inet6_hash(struct sock
*sk
)
298 if (sk
->sk_state
!= TCP_CLOSE
) {
300 __inet_hash(sk
, NULL
, ipv6_rcv_saddr_equal
);
306 EXPORT_SYMBOL_GPL(inet6_hash
);
308 /* match_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
309 * only, and any IPv4 addresses if not IPv6 only
310 * match_wildcard == false: addresses must be exactly the same, i.e.
311 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
312 * and 0.0.0.0 equals to 0.0.0.0 only
314 int ipv6_rcv_saddr_equal(const struct sock
*sk
, const struct sock
*sk2
,
317 const struct in6_addr
*sk2_rcv_saddr6
= inet6_rcv_saddr(sk2
);
318 int sk2_ipv6only
= inet_v6_ipv6only(sk2
);
319 int addr_type
= ipv6_addr_type(&sk
->sk_v6_rcv_saddr
);
320 int addr_type2
= sk2_rcv_saddr6
? ipv6_addr_type(sk2_rcv_saddr6
) : IPV6_ADDR_MAPPED
;
322 /* if both are mapped, treat as IPv4 */
323 if (addr_type
== IPV6_ADDR_MAPPED
&& addr_type2
== IPV6_ADDR_MAPPED
) {
325 if (sk
->sk_rcv_saddr
== sk2
->sk_rcv_saddr
)
327 if (!sk
->sk_rcv_saddr
|| !sk2
->sk_rcv_saddr
)
328 return match_wildcard
;
333 if (addr_type
== IPV6_ADDR_ANY
&& addr_type2
== IPV6_ADDR_ANY
)
336 if (addr_type2
== IPV6_ADDR_ANY
&& match_wildcard
&&
337 !(sk2_ipv6only
&& addr_type
== IPV6_ADDR_MAPPED
))
340 if (addr_type
== IPV6_ADDR_ANY
&& match_wildcard
&&
341 !(ipv6_only_sock(sk
) && addr_type2
== IPV6_ADDR_MAPPED
))
344 if (sk2_rcv_saddr6
&&
345 ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, sk2_rcv_saddr6
))
350 EXPORT_SYMBOL_GPL(ipv6_rcv_saddr_equal
);