Merge https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
[linux-block.git] / net / ipv6 / inet6_hashtables.c
CommitLineData
2874c5fd 1// SPDX-License-Identifier: GPL-2.0-or-later
5324a040
ACM
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic INET6 transport hashtables
8 *
d8313f5c 9 * Authors: Lotsa people, from code originally in tcp, generalised here
67ba4152 10 * by Arnaldo Carvalho de Melo <acme@mandriva.com>
5324a040
ACM
11 */
12
5324a040 13#include <linux/module.h>
d8313f5c 14#include <linux/random.h>
5324a040 15
c125e80b 16#include <net/addrconf.h>
5324a040
ACM
17#include <net/inet_connection_sock.h>
18#include <net/inet_hashtables.h>
19#include <net/inet6_hashtables.h>
6e5714ea 20#include <net/secure_seq.h>
d8313f5c 21#include <net/ip.h>
c125e80b 22#include <net/sock_reuseport.h>
5324a040 23
1122702f
JS
24extern struct inet_hashinfo tcp_hashinfo;
25
d1e559d0
ED
26u32 inet6_ehashfn(const struct net *net,
27 const struct in6_addr *laddr, const u16 lport,
28 const struct in6_addr *faddr, const __be16 fport)
b50026b5 29{
1bbdceef
HFS
30 static u32 inet6_ehash_secret __read_mostly;
31 static u32 ipv6_hash_secret __read_mostly;
32
33 u32 lhash, fhash;
34
35 net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
36 net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
37
38 lhash = (__force u32)laddr->s6_addr32[3];
39 fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
40
b50026b5 41 return __inet6_ehashfn(lhash, lport, fhash, fport,
1bbdceef 42 inet6_ehash_secret + net_hash_mix(net));
b50026b5
HFS
43}
44
b1a7ffcb
DV
45/*
46 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
47 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
48 *
49 * The sockhash lock must be held as a reader here.
50 */
d86e0dac
PE
51struct sock *__inet6_lookup_established(struct net *net,
52 struct inet_hashinfo *hashinfo,
b1a7ffcb 53 const struct in6_addr *saddr,
d2ecd9cc 54 const __be16 sport,
b1a7ffcb
DV
55 const struct in6_addr *daddr,
56 const u16 hnum,
4297a0ef 57 const int dif, const int sdif)
b1a7ffcb
DV
58{
59 struct sock *sk;
3ab5aee7 60 const struct hlist_nulls_node *node;
4f765d84 61 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
b1a7ffcb
DV
62 /* Optimize here for direct hit, only listening connections can
63 * have wildcards anyways.
64 */
33de014c 65 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
f373b53b 66 unsigned int slot = hash & hashinfo->ehash_mask;
3ab5aee7 67 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
b1a7ffcb 68
3ab5aee7 69
3ab5aee7
ED
70begin:
71 sk_nulls_for_each_rcu(sk, node, &head->chain) {
ce43b03e
ED
72 if (sk->sk_hash != hash)
73 continue;
4297a0ef 74 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif, sdif))
efe4208f 75 continue;
41c6d650 76 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
05dbc7b5
ED
77 goto out;
78
4297a0ef 79 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif, sdif))) {
efe4208f
ED
80 sock_gen_put(sk);
81 goto begin;
3ab5aee7 82 }
efe4208f 83 goto found;
b1a7ffcb 84 }
3ab5aee7
ED
85 if (get_nulls_value(node) != slot)
86 goto begin;
3ab5aee7 87out:
05dbc7b5
ED
88 sk = NULL;
89found:
b1a7ffcb
DV
90 return sk;
91}
92EXPORT_SYMBOL(__inet6_lookup_established);
93
42b16b3f 94static inline int compute_score(struct sock *sk, struct net *net,
c25eb3bf
ED
95 const unsigned short hnum,
96 const struct in6_addr *daddr,
3f7d820b 97 const int dif, const int sdif)
c25eb3bf
ED
98{
99 int score = -1;
100
c720c7e8 101 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
c25eb3bf 102 sk->sk_family == PF_INET6) {
0ee58dad
PO
103 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
104 return -1;
c25eb3bf 105
0ee58dad 106 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
e7819058 107 return -1;
4297a0ef 108
8d6c414c 109 score = sk->sk_bound_dev_if ? 2 : 1;
7170a977 110 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
70da268b 111 score++;
c25eb3bf
ED
112 }
113 return score;
114}
115
5df65312
JS
116static inline struct sock *lookup_reuseport(struct net *net, struct sock *sk,
117 struct sk_buff *skb, int doff,
118 const struct in6_addr *saddr,
119 __be16 sport,
120 const struct in6_addr *daddr,
121 unsigned short hnum)
122{
123 struct sock *reuse_sk = NULL;
124 u32 phash;
125
126 if (sk->sk_reuseport) {
127 phash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
128 reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
129 }
130 return reuse_sk;
131}
132
3b24d854 133/* called with rcu_read_lock() */
61b7c691
MKL
134static struct sock *inet6_lhash2_lookup(struct net *net,
135 struct inet_listen_hashbucket *ilb2,
136 struct sk_buff *skb, int doff,
137 const struct in6_addr *saddr,
138 const __be16 sport, const struct in6_addr *daddr,
139 const unsigned short hnum, const int dif, const int sdif)
140{
61b7c691
MKL
141 struct inet_connection_sock *icsk;
142 struct sock *sk, *result = NULL;
143 int score, hiscore = 0;
61b7c691
MKL
144
145 inet_lhash2_for_each_icsk_rcu(icsk, &ilb2->head) {
146 sk = (struct sock *)icsk;
3f7d820b 147 score = compute_score(sk, net, hnum, daddr, dif, sdif);
61b7c691 148 if (score > hiscore) {
5df65312
JS
149 result = lookup_reuseport(net, sk, skb, doff,
150 saddr, sport, daddr, hnum);
151 if (result)
152 return result;
153
61b7c691
MKL
154 result = sk;
155 hiscore = score;
156 }
157 }
158
159 return result;
160}
161
1122702f
JS
162static inline struct sock *inet6_lookup_run_bpf(struct net *net,
163 struct inet_hashinfo *hashinfo,
164 struct sk_buff *skb, int doff,
165 const struct in6_addr *saddr,
166 const __be16 sport,
167 const struct in6_addr *daddr,
f8931565 168 const u16 hnum, const int dif)
1122702f
JS
169{
170 struct sock *sk, *reuse_sk;
171 bool no_reuseport;
172
173 if (hashinfo != &tcp_hashinfo)
174 return NULL; /* only TCP is supported */
175
f8931565
MP
176 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_TCP, saddr, sport,
177 daddr, hnum, dif, &sk);
1122702f
JS
178 if (no_reuseport || IS_ERR_OR_NULL(sk))
179 return sk;
180
181 reuse_sk = lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum);
182 if (reuse_sk)
183 sk = reuse_sk;
184 return sk;
185}
186
d86e0dac 187struct sock *inet6_lookup_listener(struct net *net,
a583636a
CG
188 struct inet_hashinfo *hashinfo,
189 struct sk_buff *skb, int doff,
190 const struct in6_addr *saddr,
5ba24953 191 const __be16 sport, const struct in6_addr *daddr,
4297a0ef 192 const unsigned short hnum, const int dif, const int sdif)
5324a040 193{
61b7c691 194 struct inet_listen_hashbucket *ilb2;
0ee58dad 195 struct sock *result = NULL;
61b7c691 196 unsigned int hash2;
61b7c691 197
1122702f
JS
198 /* Lookup redirect from BPF */
199 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
200 result = inet6_lookup_run_bpf(net, hashinfo, skb, doff,
f8931565 201 saddr, sport, daddr, hnum, dif);
1122702f
JS
202 if (result)
203 goto done;
204 }
205
61b7c691
MKL
206 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
207 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
61b7c691
MKL
208
209 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
210 saddr, sport, daddr, hnum,
211 dif, sdif);
212 if (result)
8217ca65 213 goto done;
61b7c691
MKL
214
215 /* Lookup lhash2 with in6addr_any */
61b7c691
MKL
216 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
217 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
61b7c691 218
8217ca65 219 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
0ee58dad 220 saddr, sport, &in6addr_any, hnum,
8217ca65 221 dif, sdif);
8217ca65 222done:
26f8113c 223 if (IS_ERR(result))
8217ca65 224 return NULL;
5324a040
ACM
225 return result;
226}
5324a040
ACM
227EXPORT_SYMBOL_GPL(inet6_lookup_listener);
228
d86e0dac 229struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
a583636a 230 struct sk_buff *skb, int doff,
d2ecd9cc
AV
231 const struct in6_addr *saddr, const __be16 sport,
232 const struct in6_addr *daddr, const __be16 dport,
5324a040
ACM
233 const int dif)
234{
235 struct sock *sk;
3b24d854 236 bool refcounted;
5324a040 237
a583636a 238 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
4297a0ef 239 ntohs(dport), dif, 0, &refcounted);
41c6d650 240 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
3b24d854 241 sk = NULL;
5324a040
ACM
242 return sk;
243}
5324a040 244EXPORT_SYMBOL_GPL(inet6_lookup);
d8313f5c
ACM
245
246static int __inet6_check_established(struct inet_timewait_death_row *death_row,
247 struct sock *sk, const __u16 lport,
248 struct inet_timewait_sock **twp)
249{
250 struct inet_hashinfo *hinfo = death_row->hashinfo;
3759fa9c 251 struct inet_sock *inet = inet_sk(sk);
efe4208f
ED
252 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
253 const struct in6_addr *saddr = &sk->sk_v6_daddr;
d8313f5c 254 const int dif = sk->sk_bound_dev_if;
33de014c 255 struct net *net = sock_net(sk);
4297a0ef
DA
256 const int sdif = l3mdev_master_ifindex_by_index(net, dif);
257 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
33de014c 258 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
c720c7e8 259 inet->inet_dport);
d8313f5c 260 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
9db66bdc 261 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
d8313f5c 262 struct sock *sk2;
3ab5aee7 263 const struct hlist_nulls_node *node;
05dbc7b5 264 struct inet_timewait_sock *tw = NULL;
d8313f5c 265
9db66bdc 266 spin_lock(lock);
d8313f5c 267
05dbc7b5 268 sk_nulls_for_each(sk2, node, &head->chain) {
ce43b03e
ED
269 if (sk2->sk_hash != hash)
270 continue;
d8313f5c 271
4297a0ef
DA
272 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports,
273 dif, sdif))) {
efe4208f 274 if (sk2->sk_state == TCP_TIME_WAIT) {
05dbc7b5
ED
275 tw = inet_twsk(sk2);
276 if (twsk_unique(sk, sk2, twp))
efe4208f 277 break;
05dbc7b5 278 }
d8313f5c 279 goto not_unique;
efe4208f 280 }
d8313f5c
ACM
281 }
282
3759fa9c 283 /* Must record num and sport now. Otherwise we will see
efe4208f
ED
284 * in hash table socket with a funny identity.
285 */
c720c7e8
ED
286 inet->inet_num = lport;
287 inet->inet_sport = htons(lport);
13475a30 288 sk->sk_hash = hash;
547b792c 289 WARN_ON(!sk_unhashed(sk));
3ab5aee7 290 __sk_nulls_add_node_rcu(sk, &head->chain);
13475a30 291 if (tw) {
fc01538f 292 sk_nulls_del_node_init_rcu((struct sock *)tw);
02a1d6e7 293 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
13475a30 294 }
9db66bdc 295 spin_unlock(lock);
c29a0bc4 296 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
d8313f5c 297
13475a30 298 if (twp) {
d8313f5c 299 *twp = tw;
13475a30 300 } else if (tw) {
d8313f5c 301 /* Silly. Should hash-dance instead... */
dbe7faa4 302 inet_twsk_deschedule_put(tw);
d8313f5c
ACM
303 }
304 return 0;
305
306not_unique:
9db66bdc 307 spin_unlock(lock);
d8313f5c
ACM
308 return -EADDRNOTAVAIL;
309}
310
e2baad9e 311static u32 inet6_sk_port_offset(const struct sock *sk)
d8313f5c
ACM
312{
313 const struct inet_sock *inet = inet_sk(sk);
efe4208f
ED
314
315 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
316 sk->sk_v6_daddr.s6_addr32,
c720c7e8 317 inet->inet_dport);
d8313f5c
ACM
318}
319
320int inet6_hash_connect(struct inet_timewait_death_row *death_row,
321 struct sock *sk)
322{
e2baad9e
ED
323 u32 port_offset = 0;
324
325 if (!inet_sk(sk)->inet_num)
326 port_offset = inet6_sk_port_offset(sk);
327 return __inet_hash_connect(death_row, sk, port_offset,
b4d6444e 328 __inet6_check_established);
d8313f5c 329}
d8313f5c 330EXPORT_SYMBOL_GPL(inet6_hash_connect);
496611d7
CG
331
332int inet6_hash(struct sock *sk)
333{
e4cabca5
CG
334 int err = 0;
335
496611d7
CG
336 if (sk->sk_state != TCP_CLOSE) {
337 local_bh_disable();
fe38d2a1 338 err = __inet_hash(sk, NULL);
496611d7
CG
339 local_bh_enable();
340 }
341
e4cabca5 342 return err;
496611d7
CG
343}
344EXPORT_SYMBOL_GPL(inet6_hash);