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/inet_connection_sock.h>
21#include <net/inet_hashtables.h>
22#include <net/inet6_hashtables.h>
23#include <net/secure_seq.h>
24#include <net/ip.h>
25
26int __inet6_hash(struct sock *sk, struct inet_timewait_sock *tw)
27{
28 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
29 int twrefcnt = 0;
30
31 WARN_ON(!sk_unhashed(sk));
32
33 if (sk->sk_state == TCP_LISTEN) {
34 struct inet_listen_hashbucket *ilb;
35
36 ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
37 spin_lock(&ilb->lock);
38 __sk_nulls_add_node_rcu(sk, &ilb->head);
39 spin_unlock(&ilb->lock);
40 } else {
41 unsigned int hash;
42 struct hlist_nulls_head *list;
43 spinlock_t *lock;
44
45 sk->sk_hash = hash = inet6_sk_ehashfn(sk);
46 list = &inet_ehash_bucket(hashinfo, hash)->chain;
47 lock = inet_ehash_lockp(hashinfo, hash);
48 spin_lock(lock);
49 __sk_nulls_add_node_rcu(sk, list);
50 if (tw) {
51 WARN_ON(sk->sk_hash != tw->tw_hash);
52 twrefcnt = inet_twsk_unhash(tw);
53 }
54 spin_unlock(lock);
55 }
56
57 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
58 return twrefcnt;
59}
60EXPORT_SYMBOL(__inet6_hash);
61
62/*
63 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
64 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
65 *
66 * The sockhash lock must be held as a reader here.
67 */
68struct sock *__inet6_lookup_established(struct net *net,
69 struct inet_hashinfo *hashinfo,
70 const struct in6_addr *saddr,
71 const __be16 sport,
72 const struct in6_addr *daddr,
73 const u16 hnum,
74 const int dif)
75{
76 struct sock *sk;
77 const struct hlist_nulls_node *node;
78 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
79 /* Optimize here for direct hit, only listening connections can
80 * have wildcards anyways.
81 */
82 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
83 unsigned int slot = hash & hashinfo->ehash_mask;
84 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
85
86
87 rcu_read_lock();
88begin:
89 sk_nulls_for_each_rcu(sk, node, &head->chain) {
90 /* For IPV6 do the cheaper port and family tests first. */
91 if (INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
92 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
93 goto begintw;
94 if (!INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
95 sock_put(sk);
96 goto begin;
97 }
98 goto out;
99 }
100 }
101 if (get_nulls_value(node) != slot)
102 goto begin;
103
104begintw:
105 /* Must check for a TIME_WAIT'er before going to listener hash. */
106 sk_nulls_for_each_rcu(sk, node, &head->twchain) {
107 if (INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
108 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) {
109 sk = NULL;
110 goto out;
111 }
112 if (!INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
113 sock_put(sk);
114 goto begintw;
115 }
116 goto out;
117 }
118 }
119 if (get_nulls_value(node) != slot)
120 goto begintw;
121 sk = NULL;
122out:
123 rcu_read_unlock();
124 return sk;
125}
126EXPORT_SYMBOL(__inet6_lookup_established);
127
128static inline int compute_score(struct sock *sk, struct net *net,
129 const unsigned short hnum,
130 const struct in6_addr *daddr,
131 const int dif)
132{
133 int score = -1;
134
135 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
136 sk->sk_family == PF_INET6) {
137 const struct ipv6_pinfo *np = inet6_sk(sk);
138
139 score = 1;
140 if (!ipv6_addr_any(&np->rcv_saddr)) {
141 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
142 return -1;
143 score++;
144 }
145 if (sk->sk_bound_dev_if) {
146 if (sk->sk_bound_dev_if != dif)
147 return -1;
148 score++;
149 }
150 }
151 return score;
152}
153
154struct sock *inet6_lookup_listener(struct net *net,
155 struct inet_hashinfo *hashinfo, const struct in6_addr *daddr,
156 const unsigned short hnum, const int dif)
157{
158 struct sock *sk;
159 const struct hlist_nulls_node *node;
160 struct sock *result;
161 int score, hiscore;
162 unsigned int hash = inet_lhashfn(net, hnum);
163 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
164
165 rcu_read_lock();
166begin:
167 result = NULL;
168 hiscore = -1;
169 sk_nulls_for_each(sk, node, &ilb->head) {
170 score = compute_score(sk, net, hnum, daddr, dif);
171 if (score > hiscore) {
172 hiscore = score;
173 result = sk;
174 }
175 }
176 /*
177 * if the nulls value we got at the end of this lookup is
178 * not the expected one, we must restart lookup.
179 * We probably met an item that was moved to another chain.
180 */
181 if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
182 goto begin;
183 if (result) {
184 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
185 result = NULL;
186 else if (unlikely(compute_score(result, net, hnum, daddr,
187 dif) < hiscore)) {
188 sock_put(result);
189 goto begin;
190 }
191 }
192 rcu_read_unlock();
193 return result;
194}
195
196EXPORT_SYMBOL_GPL(inet6_lookup_listener);
197
198struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
199 const struct in6_addr *saddr, const __be16 sport,
200 const struct in6_addr *daddr, const __be16 dport,
201 const int dif)
202{
203 struct sock *sk;
204
205 local_bh_disable();
206 sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
207 local_bh_enable();
208
209 return sk;
210}
211
212EXPORT_SYMBOL_GPL(inet6_lookup);
213
214static int __inet6_check_established(struct inet_timewait_death_row *death_row,
215 struct sock *sk, const __u16 lport,
216 struct inet_timewait_sock **twp)
217{
218 struct inet_hashinfo *hinfo = death_row->hashinfo;
219 struct inet_sock *inet = inet_sk(sk);
220 const struct ipv6_pinfo *np = inet6_sk(sk);
221 const struct in6_addr *daddr = &np->rcv_saddr;
222 const struct in6_addr *saddr = &np->daddr;
223 const int dif = sk->sk_bound_dev_if;
224 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
225 struct net *net = sock_net(sk);
226 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
227 inet->inet_dport);
228 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
229 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
230 struct sock *sk2;
231 const struct hlist_nulls_node *node;
232 struct inet_timewait_sock *tw;
233 int twrefcnt = 0;
234
235 spin_lock(lock);
236
237 /* Check TIME-WAIT sockets first. */
238 sk_nulls_for_each(sk2, node, &head->twchain) {
239 tw = inet_twsk(sk2);
240
241 if (INET6_TW_MATCH(sk2, net, hash, saddr, daddr, ports, dif)) {
242 if (twsk_unique(sk, sk2, twp))
243 goto unique;
244 else
245 goto not_unique;
246 }
247 }
248 tw = NULL;
249
250 /* And established part... */
251 sk_nulls_for_each(sk2, node, &head->chain) {
252 if (INET6_MATCH(sk2, net, hash, saddr, daddr, ports, dif))
253 goto not_unique;
254 }
255
256unique:
257 /* Must record num and sport now. Otherwise we will see
258 * in hash table socket with a funny identity. */
259 inet->inet_num = lport;
260 inet->inet_sport = htons(lport);
261 sk->sk_hash = hash;
262 WARN_ON(!sk_unhashed(sk));
263 __sk_nulls_add_node_rcu(sk, &head->chain);
264 if (tw) {
265 twrefcnt = inet_twsk_unhash(tw);
266 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
267 }
268 spin_unlock(lock);
269 if (twrefcnt)
270 inet_twsk_put(tw);
271 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
272
273 if (twp) {
274 *twp = tw;
275 } else if (tw) {
276 /* Silly. Should hash-dance instead... */
277 inet_twsk_deschedule(tw, death_row);
278
279 inet_twsk_put(tw);
280 }
281 return 0;
282
283not_unique:
284 spin_unlock(lock);
285 return -EADDRNOTAVAIL;
286}
287
288static inline u32 inet6_sk_port_offset(const struct sock *sk)
289{
290 const struct inet_sock *inet = inet_sk(sk);
291 const struct ipv6_pinfo *np = inet6_sk(sk);
292 return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
293 np->daddr.s6_addr32,
294 inet->inet_dport);
295}
296
297int inet6_hash_connect(struct inet_timewait_death_row *death_row,
298 struct sock *sk)
299{
300 return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
301 __inet6_check_established, __inet6_hash);
302}
303
304EXPORT_SYMBOL_GPL(inet6_hash_connect);
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
72begin:
73 sk_nulls_for_each_rcu(sk, node, &head->chain) {
74 if (sk->sk_hash != hash)
75 continue;
76 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif))
77 continue;
78 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
79 goto out;
80
81 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
82 sock_gen_put(sk);
83 goto begin;
84 }
85 goto found;
86 }
87 if (get_nulls_value(node) != slot)
88 goto begin;
89out:
90 sk = NULL;
91found:
92 return sk;
93}
94EXPORT_SYMBOL(__inet6_lookup_established);
95
96static 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, bool exact_dif)
100{
101 int score = -1;
102
103 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
104 sk->sk_family == PF_INET6) {
105
106 score = 1;
107 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
108 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
109 return -1;
110 score++;
111 }
112 if (sk->sk_bound_dev_if || exact_dif) {
113 if (sk->sk_bound_dev_if != dif)
114 return -1;
115 score++;
116 }
117 if (sk->sk_incoming_cpu == raw_smp_processor_id())
118 score++;
119 }
120 return score;
121}
122
123/* called with rcu_read_lock() */
124struct sock *inet6_lookup_listener(struct net *net,
125 struct inet_hashinfo *hashinfo,
126 struct sk_buff *skb, int doff,
127 const struct in6_addr *saddr,
128 const __be16 sport, const struct in6_addr *daddr,
129 const unsigned short hnum, const int dif)
130{
131 unsigned int hash = inet_lhashfn(net, hnum);
132 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
133 int score, hiscore = 0, matches = 0, reuseport = 0;
134 bool exact_dif = inet6_exact_dif_match(net, skb);
135 struct sock *sk, *result = NULL;
136 u32 phash = 0;
137
138 sk_for_each(sk, &ilb->head) {
139 score = compute_score(sk, net, hnum, daddr, dif, exact_dif);
140 if (score > hiscore) {
141 reuseport = sk->sk_reuseport;
142 if (reuseport) {
143 phash = inet6_ehashfn(net, daddr, hnum,
144 saddr, sport);
145 result = reuseport_select_sock(sk, phash,
146 skb, doff);
147 if (result)
148 return result;
149 matches = 1;
150 }
151 result = sk;
152 hiscore = score;
153 } else if (score == hiscore && reuseport) {
154 matches++;
155 if (reciprocal_scale(phash, matches) == 0)
156 result = sk;
157 phash = next_pseudo_random32(phash);
158 }
159 }
160 return result;
161}
162EXPORT_SYMBOL_GPL(inet6_lookup_listener);
163
164struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
165 struct sk_buff *skb, int doff,
166 const struct in6_addr *saddr, const __be16 sport,
167 const struct in6_addr *daddr, const __be16 dport,
168 const int dif)
169{
170 struct sock *sk;
171 bool refcounted;
172
173 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
174 ntohs(dport), dif, &refcounted);
175 if (sk && !refcounted && !atomic_inc_not_zero(&sk->sk_refcnt))
176 sk = NULL;
177 return sk;
178}
179EXPORT_SYMBOL_GPL(inet6_lookup);
180
181static int __inet6_check_established(struct inet_timewait_death_row *death_row,
182 struct sock *sk, const __u16 lport,
183 struct inet_timewait_sock **twp)
184{
185 struct inet_hashinfo *hinfo = death_row->hashinfo;
186 struct inet_sock *inet = inet_sk(sk);
187 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
188 const struct in6_addr *saddr = &sk->sk_v6_daddr;
189 const int dif = sk->sk_bound_dev_if;
190 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
191 struct net *net = sock_net(sk);
192 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
193 inet->inet_dport);
194 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
195 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
196 struct sock *sk2;
197 const struct hlist_nulls_node *node;
198 struct inet_timewait_sock *tw = NULL;
199
200 spin_lock(lock);
201
202 sk_nulls_for_each(sk2, node, &head->chain) {
203 if (sk2->sk_hash != hash)
204 continue;
205
206 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) {
207 if (sk2->sk_state == TCP_TIME_WAIT) {
208 tw = inet_twsk(sk2);
209 if (twsk_unique(sk, sk2, twp))
210 break;
211 }
212 goto not_unique;
213 }
214 }
215
216 /* Must record num and sport now. Otherwise we will see
217 * in hash table socket with a funny identity.
218 */
219 inet->inet_num = lport;
220 inet->inet_sport = htons(lport);
221 sk->sk_hash = hash;
222 WARN_ON(!sk_unhashed(sk));
223 __sk_nulls_add_node_rcu(sk, &head->chain);
224 if (tw) {
225 sk_nulls_del_node_init_rcu((struct sock *)tw);
226 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
227 }
228 spin_unlock(lock);
229 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
230
231 if (twp) {
232 *twp = tw;
233 } else if (tw) {
234 /* Silly. Should hash-dance instead... */
235 inet_twsk_deschedule_put(tw);
236 }
237 return 0;
238
239not_unique:
240 spin_unlock(lock);
241 return -EADDRNOTAVAIL;
242}
243
244static u32 inet6_sk_port_offset(const struct sock *sk)
245{
246 const struct inet_sock *inet = inet_sk(sk);
247
248 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
249 sk->sk_v6_daddr.s6_addr32,
250 inet->inet_dport);
251}
252
253int inet6_hash_connect(struct inet_timewait_death_row *death_row,
254 struct sock *sk)
255{
256 u32 port_offset = 0;
257
258 if (!inet_sk(sk)->inet_num)
259 port_offset = inet6_sk_port_offset(sk);
260 return __inet_hash_connect(death_row, sk, port_offset,
261 __inet6_check_established);
262}
263EXPORT_SYMBOL_GPL(inet6_hash_connect);
264
265int inet6_hash(struct sock *sk)
266{
267 int err = 0;
268
269 if (sk->sk_state != TCP_CLOSE) {
270 local_bh_disable();
271 err = __inet_hash(sk, NULL, ipv6_rcv_saddr_equal);
272 local_bh_enable();
273 }
274
275 return err;
276}
277EXPORT_SYMBOL_GPL(inet6_hash);
278
279/* match_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
280 * only, and any IPv4 addresses if not IPv6 only
281 * match_wildcard == false: addresses must be exactly the same, i.e.
282 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
283 * and 0.0.0.0 equals to 0.0.0.0 only
284 */
285int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
286 bool match_wildcard)
287{
288 const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
289 int sk2_ipv6only = inet_v6_ipv6only(sk2);
290 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
291 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
292
293 /* if both are mapped, treat as IPv4 */
294 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
295 if (!sk2_ipv6only) {
296 if (sk->sk_rcv_saddr == sk2->sk_rcv_saddr)
297 return 1;
298 if (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr)
299 return match_wildcard;
300 }
301 return 0;
302 }
303
304 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
305 return 1;
306
307 if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
308 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
309 return 1;
310
311 if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
312 !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED))
313 return 1;
314
315 if (sk2_rcv_saddr6 &&
316 ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
317 return 1;
318
319 return 0;
320}
321EXPORT_SYMBOL_GPL(ipv6_rcv_saddr_equal);