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
,
59 const int dif
, const int sdif
)
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
];
73 sk_nulls_for_each_rcu(sk
, node
, &head
->chain
) {
74 if (sk
->sk_hash
!= hash
)
76 if (!INET6_MATCH(sk
, net
, saddr
, daddr
, ports
, dif
, sdif
))
78 if (unlikely(!refcount_inc_not_zero(&sk
->sk_refcnt
)))
81 if (unlikely(!INET6_MATCH(sk
, net
, saddr
, daddr
, ports
, dif
, sdif
))) {
87 if (get_nulls_value(node
) != slot
)
94 EXPORT_SYMBOL(__inet6_lookup_established
);
96 static inline int compute_score(struct sock
*sk
, struct net
*net
,
97 const unsigned short hnum
,
98 const struct in6_addr
*daddr
,
99 const int dif
, const int sdif
, bool exact_dif
)
103 if (net_eq(sock_net(sk
), net
) && inet_sk(sk
)->inet_num
== hnum
&&
104 sk
->sk_family
== PF_INET6
) {
107 if (!ipv6_addr_any(&sk
->sk_v6_rcv_saddr
)) {
108 if (!ipv6_addr_equal(&sk
->sk_v6_rcv_saddr
, daddr
))
112 if (sk
->sk_bound_dev_if
|| exact_dif
) {
113 bool dev_match
= (sk
->sk_bound_dev_if
== dif
||
114 sk
->sk_bound_dev_if
== sdif
);
118 if (sk
->sk_bound_dev_if
)
121 if (READ_ONCE(sk
->sk_incoming_cpu
) == raw_smp_processor_id())
127 /* called with rcu_read_lock() */
128 struct sock
*inet6_lookup_listener(struct net
*net
,
129 struct inet_hashinfo
*hashinfo
,
130 struct sk_buff
*skb
, int doff
,
131 const struct in6_addr
*saddr
,
132 const __be16 sport
, const struct in6_addr
*daddr
,
133 const unsigned short hnum
, const int dif
, const int sdif
)
135 unsigned int hash
= inet_lhashfn(net
, hnum
);
136 struct inet_listen_hashbucket
*ilb
= &hashinfo
->listening_hash
[hash
];
137 int score
, hiscore
= 0, matches
= 0, reuseport
= 0;
138 bool exact_dif
= inet6_exact_dif_match(net
, skb
);
139 struct sock
*sk
, *result
= NULL
;
140 struct hlist_nulls_node
*node
;
143 sk_nulls_for_each(sk
, node
, &ilb
->nulls_head
) {
144 score
= compute_score(sk
, net
, hnum
, daddr
, dif
, sdif
, exact_dif
);
145 if (score
> hiscore
) {
146 reuseport
= sk
->sk_reuseport
;
148 phash
= inet6_ehashfn(net
, daddr
, hnum
,
150 result
= reuseport_select_sock(sk
, phash
,
158 } else if (score
== hiscore
&& reuseport
) {
160 if (reciprocal_scale(phash
, matches
) == 0)
162 phash
= next_pseudo_random32(phash
);
167 EXPORT_SYMBOL_GPL(inet6_lookup_listener
);
169 struct sock
*inet6_lookup(struct net
*net
, struct inet_hashinfo
*hashinfo
,
170 struct sk_buff
*skb
, int doff
,
171 const struct in6_addr
*saddr
, const __be16 sport
,
172 const struct in6_addr
*daddr
, const __be16 dport
,
178 sk
= __inet6_lookup(net
, hashinfo
, skb
, doff
, saddr
, sport
, daddr
,
179 ntohs(dport
), dif
, 0, &refcounted
);
180 if (sk
&& !refcounted
&& !refcount_inc_not_zero(&sk
->sk_refcnt
))
184 EXPORT_SYMBOL_GPL(inet6_lookup
);
186 static int __inet6_check_established(struct inet_timewait_death_row
*death_row
,
187 struct sock
*sk
, const __u16 lport
,
188 struct inet_timewait_sock
**twp
)
190 struct inet_hashinfo
*hinfo
= death_row
->hashinfo
;
191 struct inet_sock
*inet
= inet_sk(sk
);
192 const struct in6_addr
*daddr
= &sk
->sk_v6_rcv_saddr
;
193 const struct in6_addr
*saddr
= &sk
->sk_v6_daddr
;
194 const int dif
= sk
->sk_bound_dev_if
;
195 struct net
*net
= sock_net(sk
);
196 const int sdif
= l3mdev_master_ifindex_by_index(net
, dif
);
197 const __portpair ports
= INET_COMBINED_PORTS(inet
->inet_dport
, lport
);
198 const unsigned int hash
= inet6_ehashfn(net
, daddr
, lport
, saddr
,
200 struct inet_ehash_bucket
*head
= inet_ehash_bucket(hinfo
, hash
);
201 spinlock_t
*lock
= inet_ehash_lockp(hinfo
, hash
);
203 const struct hlist_nulls_node
*node
;
204 struct inet_timewait_sock
*tw
= NULL
;
208 sk_nulls_for_each(sk2
, node
, &head
->chain
) {
209 if (sk2
->sk_hash
!= hash
)
212 if (likely(INET6_MATCH(sk2
, net
, saddr
, daddr
, ports
,
214 if (sk2
->sk_state
== TCP_TIME_WAIT
) {
216 if (twsk_unique(sk
, sk2
, twp
))
223 /* Must record num and sport now. Otherwise we will see
224 * in hash table socket with a funny identity.
226 inet
->inet_num
= lport
;
227 inet
->inet_sport
= htons(lport
);
229 WARN_ON(!sk_unhashed(sk
));
230 __sk_nulls_add_node_rcu(sk
, &head
->chain
);
232 sk_nulls_del_node_init_rcu((struct sock
*)tw
);
233 __NET_INC_STATS(net
, LINUX_MIB_TIMEWAITRECYCLED
);
236 sock_prot_inuse_add(sock_net(sk
), sk
->sk_prot
, 1);
241 /* Silly. Should hash-dance instead... */
242 inet_twsk_deschedule_put(tw
);
248 return -EADDRNOTAVAIL
;
251 static u32
inet6_sk_port_offset(const struct sock
*sk
)
253 const struct inet_sock
*inet
= inet_sk(sk
);
255 return secure_ipv6_port_ephemeral(sk
->sk_v6_rcv_saddr
.s6_addr32
,
256 sk
->sk_v6_daddr
.s6_addr32
,
260 int inet6_hash_connect(struct inet_timewait_death_row
*death_row
,
265 if (!inet_sk(sk
)->inet_num
)
266 port_offset
= inet6_sk_port_offset(sk
);
267 return __inet_hash_connect(death_row
, sk
, port_offset
,
268 __inet6_check_established
);
270 EXPORT_SYMBOL_GPL(inet6_hash_connect
);
272 int inet6_hash(struct sock
*sk
)
276 if (sk
->sk_state
!= TCP_CLOSE
) {
278 err
= __inet_hash(sk
, NULL
);
284 EXPORT_SYMBOL_GPL(inet6_hash
);