bpf: Introduce BPF_PROG_TYPE_SK_REUSEPORT
[linux-block.git] / net / ipv6 / inet6_hashtables.c
CommitLineData
5324a040
ACM
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 *
d8313f5c 8 * Authors: Lotsa people, from code originally in tcp, generalised here
67ba4152 9 * by Arnaldo Carvalho de Melo <acme@mandriva.com>
5324a040
ACM
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
5324a040 17#include <linux/module.h>
d8313f5c 18#include <linux/random.h>
5324a040 19
c125e80b 20#include <net/addrconf.h>
5324a040
ACM
21#include <net/inet_connection_sock.h>
22#include <net/inet_hashtables.h>
23#include <net/inet6_hashtables.h>
6e5714ea 24#include <net/secure_seq.h>
d8313f5c 25#include <net/ip.h>
c125e80b 26#include <net/sock_reuseport.h>
5324a040 27
d1e559d0
ED
28u32 inet6_ehashfn(const struct net *net,
29 const struct in6_addr *laddr, const u16 lport,
30 const struct in6_addr *faddr, const __be16 fport)
b50026b5 31{
1bbdceef
HFS
32 static u32 inet6_ehash_secret __read_mostly;
33 static u32 ipv6_hash_secret __read_mostly;
34
35 u32 lhash, fhash;
36
37 net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
38 net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
39
40 lhash = (__force u32)laddr->s6_addr32[3];
41 fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
42
b50026b5 43 return __inet6_ehashfn(lhash, lport, fhash, fport,
1bbdceef 44 inet6_ehash_secret + net_hash_mix(net));
b50026b5
HFS
45}
46
b1a7ffcb
DV
47/*
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
50 *
51 * The sockhash lock must be held as a reader here.
52 */
d86e0dac
PE
53struct sock *__inet6_lookup_established(struct net *net,
54 struct inet_hashinfo *hashinfo,
b1a7ffcb 55 const struct in6_addr *saddr,
d2ecd9cc 56 const __be16 sport,
b1a7ffcb
DV
57 const struct in6_addr *daddr,
58 const u16 hnum,
4297a0ef 59 const int dif, const int sdif)
b1a7ffcb
DV
60{
61 struct sock *sk;
3ab5aee7 62 const struct hlist_nulls_node *node;
4f765d84 63 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
b1a7ffcb
DV
64 /* Optimize here for direct hit, only listening connections can
65 * have wildcards anyways.
66 */
33de014c 67 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
f373b53b 68 unsigned int slot = hash & hashinfo->ehash_mask;
3ab5aee7 69 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
b1a7ffcb 70
3ab5aee7 71
3ab5aee7
ED
72begin:
73 sk_nulls_for_each_rcu(sk, node, &head->chain) {
ce43b03e
ED
74 if (sk->sk_hash != hash)
75 continue;
4297a0ef 76 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif, sdif))
efe4208f 77 continue;
41c6d650 78 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
05dbc7b5
ED
79 goto out;
80
4297a0ef 81 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif, sdif))) {
efe4208f
ED
82 sock_gen_put(sk);
83 goto begin;
3ab5aee7 84 }
efe4208f 85 goto found;
b1a7ffcb 86 }
3ab5aee7
ED
87 if (get_nulls_value(node) != slot)
88 goto begin;
3ab5aee7 89out:
05dbc7b5
ED
90 sk = NULL;
91found:
b1a7ffcb
DV
92 return sk;
93}
94EXPORT_SYMBOL(__inet6_lookup_established);
95
42b16b3f 96static inline int compute_score(struct sock *sk, struct net *net,
c25eb3bf
ED
97 const unsigned short hnum,
98 const struct in6_addr *daddr,
4297a0ef 99 const int dif, const int sdif, bool exact_dif)
c25eb3bf
ED
100{
101 int score = -1;
102
c720c7e8 103 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
c25eb3bf 104 sk->sk_family == PF_INET6) {
c25eb3bf
ED
105
106 score = 1;
efe4208f
ED
107 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
108 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
c25eb3bf
ED
109 return -1;
110 score++;
111 }
a04a480d 112 if (sk->sk_bound_dev_if || exact_dif) {
4297a0ef
DA
113 bool dev_match = (sk->sk_bound_dev_if == dif ||
114 sk->sk_bound_dev_if == sdif);
115
8c43bd17 116 if (!dev_match)
c25eb3bf 117 return -1;
8c43bd17 118 if (sk->sk_bound_dev_if)
4297a0ef 119 score++;
c25eb3bf 120 }
70da268b
ED
121 if (sk->sk_incoming_cpu == raw_smp_processor_id())
122 score++;
c25eb3bf
ED
123 }
124 return score;
125}
126
3b24d854 127/* called with rcu_read_lock() */
61b7c691
MKL
128static struct sock *inet6_lhash2_lookup(struct net *net,
129 struct inet_listen_hashbucket *ilb2,
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)
134{
135 bool exact_dif = inet6_exact_dif_match(net, skb);
136 struct inet_connection_sock *icsk;
137 struct sock *sk, *result = NULL;
138 int score, hiscore = 0;
139 u32 phash = 0;
140
141 inet_lhash2_for_each_icsk_rcu(icsk, &ilb2->head) {
142 sk = (struct sock *)icsk;
143 score = compute_score(sk, net, hnum, daddr, dif, sdif,
144 exact_dif);
145 if (score > hiscore) {
146 if (sk->sk_reuseport) {
147 phash = inet6_ehashfn(net, daddr, hnum,
148 saddr, sport);
149 result = reuseport_select_sock(sk, phash,
150 skb, doff);
151 if (result)
152 return result;
153 }
154 result = sk;
155 hiscore = score;
156 }
157 }
158
159 return result;
160}
161
d86e0dac 162struct sock *inet6_lookup_listener(struct net *net,
a583636a
CG
163 struct inet_hashinfo *hashinfo,
164 struct sk_buff *skb, int doff,
165 const struct in6_addr *saddr,
5ba24953 166 const __be16 sport, const struct in6_addr *daddr,
4297a0ef 167 const unsigned short hnum, const int dif, const int sdif)
5324a040 168{
c25eb3bf
ED
169 unsigned int hash = inet_lhashfn(net, hnum);
170 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
a04a480d 171 bool exact_dif = inet6_exact_dif_match(net, skb);
61b7c691 172 struct inet_listen_hashbucket *ilb2;
3b24d854 173 struct sock *sk, *result = NULL;
e94a62f5 174 int score, hiscore = 0;
61b7c691 175 unsigned int hash2;
3b24d854 176 u32 phash = 0;
c25eb3bf 177
61b7c691
MKL
178 if (ilb->count <= 10 || !hashinfo->lhash2)
179 goto port_lookup;
180
181 /* Too many sk in the ilb bucket (which is hashed by port alone).
182 * Try lhash2 (which is hashed by port and addr) instead.
183 */
184
185 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
186 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
187 if (ilb2->count > ilb->count)
188 goto port_lookup;
189
190 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
191 saddr, sport, daddr, hnum,
192 dif, sdif);
193 if (result)
194 return result;
195
196 /* Lookup lhash2 with in6addr_any */
197
198 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
199 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
200 if (ilb2->count > ilb->count)
201 goto port_lookup;
202
203 return inet6_lhash2_lookup(net, ilb2, skb, doff,
204 saddr, sport, daddr, hnum,
205 dif, sdif);
206
207port_lookup:
3b24d854 208 sk_for_each(sk, &ilb->head) {
4297a0ef 209 score = compute_score(sk, net, hnum, daddr, dif, sdif, exact_dif);
c25eb3bf 210 if (score > hiscore) {
e94a62f5 211 if (sk->sk_reuseport) {
5ba24953
TH
212 phash = inet6_ehashfn(net, daddr, hnum,
213 saddr, sport);
3b24d854
ED
214 result = reuseport_select_sock(sk, phash,
215 skb, doff);
216 if (result)
217 return result;
5ba24953 218 }
3b24d854 219 result = sk;
03c5b534 220 hiscore = score;
5324a040
ACM
221 }
222 }
5324a040
ACM
223 return result;
224}
5324a040
ACM
225EXPORT_SYMBOL_GPL(inet6_lookup_listener);
226
d86e0dac 227struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
a583636a 228 struct sk_buff *skb, int doff,
d2ecd9cc
AV
229 const struct in6_addr *saddr, const __be16 sport,
230 const struct in6_addr *daddr, const __be16 dport,
5324a040
ACM
231 const int dif)
232{
233 struct sock *sk;
3b24d854 234 bool refcounted;
5324a040 235
a583636a 236 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
4297a0ef 237 ntohs(dport), dif, 0, &refcounted);
41c6d650 238 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
3b24d854 239 sk = NULL;
5324a040
ACM
240 return sk;
241}
5324a040 242EXPORT_SYMBOL_GPL(inet6_lookup);
d8313f5c
ACM
243
244static int __inet6_check_established(struct inet_timewait_death_row *death_row,
245 struct sock *sk, const __u16 lport,
246 struct inet_timewait_sock **twp)
247{
248 struct inet_hashinfo *hinfo = death_row->hashinfo;
3759fa9c 249 struct inet_sock *inet = inet_sk(sk);
efe4208f
ED
250 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
251 const struct in6_addr *saddr = &sk->sk_v6_daddr;
d8313f5c 252 const int dif = sk->sk_bound_dev_if;
33de014c 253 struct net *net = sock_net(sk);
4297a0ef
DA
254 const int sdif = l3mdev_master_ifindex_by_index(net, dif);
255 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
33de014c 256 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
c720c7e8 257 inet->inet_dport);
d8313f5c 258 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
9db66bdc 259 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
d8313f5c 260 struct sock *sk2;
3ab5aee7 261 const struct hlist_nulls_node *node;
05dbc7b5 262 struct inet_timewait_sock *tw = NULL;
d8313f5c 263
9db66bdc 264 spin_lock(lock);
d8313f5c 265
05dbc7b5 266 sk_nulls_for_each(sk2, node, &head->chain) {
ce43b03e
ED
267 if (sk2->sk_hash != hash)
268 continue;
d8313f5c 269
4297a0ef
DA
270 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports,
271 dif, sdif))) {
efe4208f 272 if (sk2->sk_state == TCP_TIME_WAIT) {
05dbc7b5
ED
273 tw = inet_twsk(sk2);
274 if (twsk_unique(sk, sk2, twp))
efe4208f 275 break;
05dbc7b5 276 }
d8313f5c 277 goto not_unique;
efe4208f 278 }
d8313f5c
ACM
279 }
280
3759fa9c 281 /* Must record num and sport now. Otherwise we will see
efe4208f
ED
282 * in hash table socket with a funny identity.
283 */
c720c7e8
ED
284 inet->inet_num = lport;
285 inet->inet_sport = htons(lport);
13475a30 286 sk->sk_hash = hash;
547b792c 287 WARN_ON(!sk_unhashed(sk));
3ab5aee7 288 __sk_nulls_add_node_rcu(sk, &head->chain);
13475a30 289 if (tw) {
fc01538f 290 sk_nulls_del_node_init_rcu((struct sock *)tw);
02a1d6e7 291 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
13475a30 292 }
9db66bdc 293 spin_unlock(lock);
c29a0bc4 294 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
d8313f5c 295
13475a30 296 if (twp) {
d8313f5c 297 *twp = tw;
13475a30 298 } else if (tw) {
d8313f5c 299 /* Silly. Should hash-dance instead... */
dbe7faa4 300 inet_twsk_deschedule_put(tw);
d8313f5c
ACM
301 }
302 return 0;
303
304not_unique:
9db66bdc 305 spin_unlock(lock);
d8313f5c
ACM
306 return -EADDRNOTAVAIL;
307}
308
e2baad9e 309static u32 inet6_sk_port_offset(const struct sock *sk)
d8313f5c
ACM
310{
311 const struct inet_sock *inet = inet_sk(sk);
efe4208f
ED
312
313 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
314 sk->sk_v6_daddr.s6_addr32,
c720c7e8 315 inet->inet_dport);
d8313f5c
ACM
316}
317
318int inet6_hash_connect(struct inet_timewait_death_row *death_row,
319 struct sock *sk)
320{
e2baad9e
ED
321 u32 port_offset = 0;
322
323 if (!inet_sk(sk)->inet_num)
324 port_offset = inet6_sk_port_offset(sk);
325 return __inet_hash_connect(death_row, sk, port_offset,
b4d6444e 326 __inet6_check_established);
d8313f5c 327}
d8313f5c 328EXPORT_SYMBOL_GPL(inet6_hash_connect);
496611d7
CG
329
330int inet6_hash(struct sock *sk)
331{
e4cabca5
CG
332 int err = 0;
333
496611d7
CG
334 if (sk->sk_state != TCP_CLOSE) {
335 local_bh_disable();
fe38d2a1 336 err = __inet_hash(sk, NULL);
496611d7
CG
337 local_bh_enable();
338 }
339
e4cabca5 340 return err;
496611d7
CG
341}
342EXPORT_SYMBOL_GPL(inet6_hash);