Loading...
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 *
8 * Authors: Lotsa people, from code originally in tcp, generalised here
9 * by Arnaldo Carvalho de Melo <acme@mandriva.com>
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
17#include <linux/module.h>
18#include <linux/random.h>
19
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>
25#include <net/ip.h>
26#include <net/sock_reuseport.h>
27
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)
31{
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
43 return __inet6_ehashfn(lhash, lport, fhash, fport,
44 inet6_ehash_secret + net_hash_mix(net));
45}
46
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 */
53struct sock *__inet6_lookup_established(struct net *net,
54 struct inet_hashinfo *hashinfo,
55 const struct in6_addr *saddr,
56 const __be16 sport,
57 const struct in6_addr *daddr,
58 const u16 hnum,
59 const int dif)
60{
61 struct sock *sk;
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.
66 */
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];
70
71
72 rcu_read_lock();
73begin:
74 sk_nulls_for_each_rcu(sk, node, &head->chain) {
75 if (sk->sk_hash != hash)
76 continue;
77 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif))
78 continue;
79 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
80 goto out;
81
82 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
83 sock_gen_put(sk);
84 goto begin;
85 }
86 goto found;
87 }
88 if (get_nulls_value(node) != slot)
89 goto begin;
90out:
91 sk = NULL;
92found:
93 rcu_read_unlock();
94 return sk;
95}
96EXPORT_SYMBOL(__inet6_lookup_established);
97
98static inline int compute_score(struct sock *sk, struct net *net,
99 const unsigned short hnum,
100 const struct in6_addr *daddr,
101 const int dif)
102{
103 int score = -1;
104
105 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
106 sk->sk_family == PF_INET6) {
107
108 score = 1;
109 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
110 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
111 return -1;
112 score++;
113 }
114 if (sk->sk_bound_dev_if) {
115 if (sk->sk_bound_dev_if != dif)
116 return -1;
117 score++;
118 }
119 if (sk->sk_incoming_cpu == raw_smp_processor_id())
120 score++;
121 }
122 return score;
123}
124
125struct 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)
131{
132 struct sock *sk;
133 const struct hlist_nulls_node *node;
134 struct sock *result;
135 int score, hiscore, matches = 0, reuseport = 0;
136 bool select_ok = true;
137 u32 phash = 0;
138 unsigned int hash = inet_lhashfn(net, hnum);
139 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
140
141 rcu_read_lock();
142begin:
143 result = NULL;
144 hiscore = 0;
145 sk_nulls_for_each(sk, node, &ilb->head) {
146 score = compute_score(sk, net, hnum, daddr, dif);
147 if (score > hiscore) {
148 hiscore = score;
149 result = sk;
150 reuseport = sk->sk_reuseport;
151 if (reuseport) {
152 phash = inet6_ehashfn(net, daddr, hnum,
153 saddr, sport);
154 if (select_ok) {
155 struct sock *sk2;
156 sk2 = reuseport_select_sock(sk, phash,
157 skb, doff);
158 if (sk2) {
159 result = sk2;
160 goto found;
161 }
162 }
163 matches = 1;
164 }
165 } else if (score == hiscore && reuseport) {
166 matches++;
167 if (reciprocal_scale(phash, matches) == 0)
168 result = sk;
169 phash = next_pseudo_random32(phash);
170 }
171 }
172 /*
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.
176 */
177 if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
178 goto begin;
179 if (result) {
180found:
181 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
182 result = NULL;
183 else if (unlikely(compute_score(result, net, hnum, daddr,
184 dif) < hiscore)) {
185 sock_put(result);
186 select_ok = false;
187 goto begin;
188 }
189 }
190 rcu_read_unlock();
191 return result;
192}
193EXPORT_SYMBOL_GPL(inet6_lookup_listener);
194
195struct 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,
199 const int dif)
200{
201 struct sock *sk;
202
203 local_bh_disable();
204 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
205 ntohs(dport), dif);
206 local_bh_enable();
207
208 return sk;
209}
210EXPORT_SYMBOL_GPL(inet6_lookup);
211
212static int __inet6_check_established(struct inet_timewait_death_row *death_row,
213 struct sock *sk, const __u16 lport,
214 struct inet_timewait_sock **twp)
215{
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,
224 inet->inet_dport);
225 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
226 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
227 struct sock *sk2;
228 const struct hlist_nulls_node *node;
229 struct inet_timewait_sock *tw = NULL;
230
231 spin_lock(lock);
232
233 sk_nulls_for_each(sk2, node, &head->chain) {
234 if (sk2->sk_hash != hash)
235 continue;
236
237 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) {
238 if (sk2->sk_state == TCP_TIME_WAIT) {
239 tw = inet_twsk(sk2);
240 if (twsk_unique(sk, sk2, twp))
241 break;
242 }
243 goto not_unique;
244 }
245 }
246
247 /* Must record num and sport now. Otherwise we will see
248 * in hash table socket with a funny identity.
249 */
250 inet->inet_num = lport;
251 inet->inet_sport = htons(lport);
252 sk->sk_hash = hash;
253 WARN_ON(!sk_unhashed(sk));
254 __sk_nulls_add_node_rcu(sk, &head->chain);
255 if (tw) {
256 sk_nulls_del_node_init_rcu((struct sock *)tw);
257 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
258 }
259 spin_unlock(lock);
260 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
261
262 if (twp) {
263 *twp = tw;
264 } else if (tw) {
265 /* Silly. Should hash-dance instead... */
266 inet_twsk_deschedule_put(tw);
267 }
268 return 0;
269
270not_unique:
271 spin_unlock(lock);
272 return -EADDRNOTAVAIL;
273}
274
275static u32 inet6_sk_port_offset(const struct sock *sk)
276{
277 const struct inet_sock *inet = inet_sk(sk);
278
279 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
280 sk->sk_v6_daddr.s6_addr32,
281 inet->inet_dport);
282}
283
284int inet6_hash_connect(struct inet_timewait_death_row *death_row,
285 struct sock *sk)
286{
287 u32 port_offset = 0;
288
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);
293}
294EXPORT_SYMBOL_GPL(inet6_hash_connect);
295
296int inet6_hash(struct sock *sk)
297{
298 if (sk->sk_state != TCP_CLOSE) {
299 local_bh_disable();
300 __inet_hash(sk, NULL, ipv6_rcv_saddr_equal);
301 local_bh_enable();
302 }
303
304 return 0;
305}
306EXPORT_SYMBOL_GPL(inet6_hash);
307
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
313 */
314int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
315 bool match_wildcard)
316{
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;
321
322 /* if both are mapped, treat as IPv4 */
323 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
324 if (!sk2_ipv6only) {
325 if (sk->sk_rcv_saddr == sk2->sk_rcv_saddr)
326 return 1;
327 if (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr)
328 return match_wildcard;
329 }
330 return 0;
331 }
332
333 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
334 return 1;
335
336 if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
337 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
338 return 1;
339
340 if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
341 !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED))
342 return 1;
343
344 if (sk2_rcv_saddr6 &&
345 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
346 return 1;
347
348 return 0;
349}
350EXPORT_SYMBOL_GPL(ipv6_rcv_saddr_equal);
1// SPDX-License-Identifier: GPL-2.0-or-later
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 *
9 * Authors: Lotsa people, from code originally in tcp, generalised here
10 * by Arnaldo Carvalho de Melo <acme@mandriva.com>
11 */
12
13#include <linux/module.h>
14#include <linux/random.h>
15
16#include <net/addrconf.h>
17#include <net/inet_connection_sock.h>
18#include <net/inet_hashtables.h>
19#include <net/inet6_hashtables.h>
20#include <net/secure_seq.h>
21#include <net/ip.h>
22#include <net/sock_reuseport.h>
23
24u32 inet6_ehashfn(const struct net *net,
25 const struct in6_addr *laddr, const u16 lport,
26 const struct in6_addr *faddr, const __be16 fport)
27{
28 static u32 inet6_ehash_secret __read_mostly;
29 static u32 ipv6_hash_secret __read_mostly;
30
31 u32 lhash, fhash;
32
33 net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
34 net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
35
36 lhash = (__force u32)laddr->s6_addr32[3];
37 fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
38
39 return __inet6_ehashfn(lhash, lport, fhash, fport,
40 inet6_ehash_secret + net_hash_mix(net));
41}
42
43/*
44 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
45 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
46 *
47 * The sockhash lock must be held as a reader here.
48 */
49struct sock *__inet6_lookup_established(struct net *net,
50 struct inet_hashinfo *hashinfo,
51 const struct in6_addr *saddr,
52 const __be16 sport,
53 const struct in6_addr *daddr,
54 const u16 hnum,
55 const int dif, const int sdif)
56{
57 struct sock *sk;
58 const struct hlist_nulls_node *node;
59 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
60 /* Optimize here for direct hit, only listening connections can
61 * have wildcards anyways.
62 */
63 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
64 unsigned int slot = hash & hashinfo->ehash_mask;
65 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
66
67
68begin:
69 sk_nulls_for_each_rcu(sk, node, &head->chain) {
70 if (sk->sk_hash != hash)
71 continue;
72 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif, sdif))
73 continue;
74 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
75 goto out;
76
77 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif, sdif))) {
78 sock_gen_put(sk);
79 goto begin;
80 }
81 goto found;
82 }
83 if (get_nulls_value(node) != slot)
84 goto begin;
85out:
86 sk = NULL;
87found:
88 return sk;
89}
90EXPORT_SYMBOL(__inet6_lookup_established);
91
92static inline int compute_score(struct sock *sk, struct net *net,
93 const unsigned short hnum,
94 const struct in6_addr *daddr,
95 const int dif, const int sdif, bool exact_dif)
96{
97 int score = -1;
98
99 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
100 sk->sk_family == PF_INET6) {
101 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
102 return -1;
103
104 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
105 return -1;
106
107 score = 1;
108 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
109 score++;
110 }
111 return score;
112}
113
114/* called with rcu_read_lock() */
115static struct sock *inet6_lhash2_lookup(struct net *net,
116 struct inet_listen_hashbucket *ilb2,
117 struct sk_buff *skb, int doff,
118 const struct in6_addr *saddr,
119 const __be16 sport, const struct in6_addr *daddr,
120 const unsigned short hnum, const int dif, const int sdif)
121{
122 bool exact_dif = inet6_exact_dif_match(net, skb);
123 struct inet_connection_sock *icsk;
124 struct sock *sk, *result = NULL;
125 int score, hiscore = 0;
126 u32 phash = 0;
127
128 inet_lhash2_for_each_icsk_rcu(icsk, &ilb2->head) {
129 sk = (struct sock *)icsk;
130 score = compute_score(sk, net, hnum, daddr, dif, sdif,
131 exact_dif);
132 if (score > hiscore) {
133 if (sk->sk_reuseport) {
134 phash = inet6_ehashfn(net, daddr, hnum,
135 saddr, sport);
136 result = reuseport_select_sock(sk, phash,
137 skb, doff);
138 if (result)
139 return result;
140 }
141 result = sk;
142 hiscore = score;
143 }
144 }
145
146 return result;
147}
148
149struct sock *inet6_lookup_listener(struct net *net,
150 struct inet_hashinfo *hashinfo,
151 struct sk_buff *skb, int doff,
152 const struct in6_addr *saddr,
153 const __be16 sport, const struct in6_addr *daddr,
154 const unsigned short hnum, const int dif, const int sdif)
155{
156 struct inet_listen_hashbucket *ilb2;
157 struct sock *result = NULL;
158 unsigned int hash2;
159
160 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
161 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
162
163 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
164 saddr, sport, daddr, hnum,
165 dif, sdif);
166 if (result)
167 goto done;
168
169 /* Lookup lhash2 with in6addr_any */
170 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
171 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
172
173 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
174 saddr, sport, &in6addr_any, hnum,
175 dif, sdif);
176done:
177 if (IS_ERR(result))
178 return NULL;
179 return result;
180}
181EXPORT_SYMBOL_GPL(inet6_lookup_listener);
182
183struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
184 struct sk_buff *skb, int doff,
185 const struct in6_addr *saddr, const __be16 sport,
186 const struct in6_addr *daddr, const __be16 dport,
187 const int dif)
188{
189 struct sock *sk;
190 bool refcounted;
191
192 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
193 ntohs(dport), dif, 0, &refcounted);
194 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
195 sk = NULL;
196 return sk;
197}
198EXPORT_SYMBOL_GPL(inet6_lookup);
199
200static int __inet6_check_established(struct inet_timewait_death_row *death_row,
201 struct sock *sk, const __u16 lport,
202 struct inet_timewait_sock **twp)
203{
204 struct inet_hashinfo *hinfo = death_row->hashinfo;
205 struct inet_sock *inet = inet_sk(sk);
206 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
207 const struct in6_addr *saddr = &sk->sk_v6_daddr;
208 const int dif = sk->sk_bound_dev_if;
209 struct net *net = sock_net(sk);
210 const int sdif = l3mdev_master_ifindex_by_index(net, dif);
211 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
212 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
213 inet->inet_dport);
214 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
215 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
216 struct sock *sk2;
217 const struct hlist_nulls_node *node;
218 struct inet_timewait_sock *tw = NULL;
219
220 spin_lock(lock);
221
222 sk_nulls_for_each(sk2, node, &head->chain) {
223 if (sk2->sk_hash != hash)
224 continue;
225
226 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports,
227 dif, sdif))) {
228 if (sk2->sk_state == TCP_TIME_WAIT) {
229 tw = inet_twsk(sk2);
230 if (twsk_unique(sk, sk2, twp))
231 break;
232 }
233 goto not_unique;
234 }
235 }
236
237 /* Must record num and sport now. Otherwise we will see
238 * in hash table socket with a funny identity.
239 */
240 inet->inet_num = lport;
241 inet->inet_sport = htons(lport);
242 sk->sk_hash = hash;
243 WARN_ON(!sk_unhashed(sk));
244 __sk_nulls_add_node_rcu(sk, &head->chain);
245 if (tw) {
246 sk_nulls_del_node_init_rcu((struct sock *)tw);
247 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
248 }
249 spin_unlock(lock);
250 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
251
252 if (twp) {
253 *twp = tw;
254 } else if (tw) {
255 /* Silly. Should hash-dance instead... */
256 inet_twsk_deschedule_put(tw);
257 }
258 return 0;
259
260not_unique:
261 spin_unlock(lock);
262 return -EADDRNOTAVAIL;
263}
264
265static u32 inet6_sk_port_offset(const struct sock *sk)
266{
267 const struct inet_sock *inet = inet_sk(sk);
268
269 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
270 sk->sk_v6_daddr.s6_addr32,
271 inet->inet_dport);
272}
273
274int inet6_hash_connect(struct inet_timewait_death_row *death_row,
275 struct sock *sk)
276{
277 u32 port_offset = 0;
278
279 if (!inet_sk(sk)->inet_num)
280 port_offset = inet6_sk_port_offset(sk);
281 return __inet_hash_connect(death_row, sk, port_offset,
282 __inet6_check_established);
283}
284EXPORT_SYMBOL_GPL(inet6_hash_connect);
285
286int inet6_hash(struct sock *sk)
287{
288 int err = 0;
289
290 if (sk->sk_state != TCP_CLOSE) {
291 local_bh_disable();
292 err = __inet_hash(sk, NULL);
293 local_bh_enable();
294 }
295
296 return err;
297}
298EXPORT_SYMBOL_GPL(inet6_hash);