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// 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}
42EXPORT_SYMBOL_GPL(inet6_ehashfn);
43
44/*
45 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
46 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
47 *
48 * The sockhash lock must be held as a reader here.
49 */
50struct sock *__inet6_lookup_established(struct net *net,
51 struct inet_hashinfo *hashinfo,
52 const struct in6_addr *saddr,
53 const __be16 sport,
54 const struct in6_addr *daddr,
55 const u16 hnum,
56 const int dif, const int sdif)
57{
58 struct sock *sk;
59 const struct hlist_nulls_node *node;
60 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
61 /* Optimize here for direct hit, only listening connections can
62 * have wildcards anyways.
63 */
64 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
65 unsigned int slot = hash & hashinfo->ehash_mask;
66 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
67
68
69begin:
70 sk_nulls_for_each_rcu(sk, node, &head->chain) {
71 if (sk->sk_hash != hash)
72 continue;
73 if (!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))
74 continue;
75 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
76 goto out;
77
78 if (unlikely(!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))) {
79 sock_gen_put(sk);
80 goto begin;
81 }
82 goto found;
83 }
84 if (get_nulls_value(node) != slot)
85 goto begin;
86out:
87 sk = NULL;
88found:
89 return sk;
90}
91EXPORT_SYMBOL(__inet6_lookup_established);
92
93static inline int compute_score(struct sock *sk, struct net *net,
94 const unsigned short hnum,
95 const struct in6_addr *daddr,
96 const int dif, const int sdif)
97{
98 int score = -1;
99
100 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
101 sk->sk_family == PF_INET6) {
102 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
103 return -1;
104
105 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
106 return -1;
107
108 score = sk->sk_bound_dev_if ? 2 : 1;
109 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
110 score++;
111 }
112 return score;
113}
114
115/**
116 * inet6_lookup_reuseport() - execute reuseport logic on AF_INET6 socket if necessary.
117 * @net: network namespace.
118 * @sk: AF_INET6 socket, must be in TCP_LISTEN state for TCP or TCP_CLOSE for UDP.
119 * @skb: context for a potential SK_REUSEPORT program.
120 * @doff: header offset.
121 * @saddr: source address.
122 * @sport: source port.
123 * @daddr: destination address.
124 * @hnum: destination port in host byte order.
125 * @ehashfn: hash function used to generate the fallback hash.
126 *
127 * Return: NULL if sk doesn't have SO_REUSEPORT set, otherwise a pointer to
128 * the selected sock or an error.
129 */
130struct sock *inet6_lookup_reuseport(struct net *net, struct sock *sk,
131 struct sk_buff *skb, int doff,
132 const struct in6_addr *saddr,
133 __be16 sport,
134 const struct in6_addr *daddr,
135 unsigned short hnum,
136 inet6_ehashfn_t *ehashfn)
137{
138 struct sock *reuse_sk = NULL;
139 u32 phash;
140
141 if (sk->sk_reuseport) {
142 phash = INDIRECT_CALL_INET(ehashfn, udp6_ehashfn, inet6_ehashfn,
143 net, daddr, hnum, saddr, sport);
144 reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
145 }
146 return reuse_sk;
147}
148EXPORT_SYMBOL_GPL(inet6_lookup_reuseport);
149
150/* called with rcu_read_lock() */
151static struct sock *inet6_lhash2_lookup(struct net *net,
152 struct inet_listen_hashbucket *ilb2,
153 struct sk_buff *skb, int doff,
154 const struct in6_addr *saddr,
155 const __be16 sport, const struct in6_addr *daddr,
156 const unsigned short hnum, const int dif, const int sdif)
157{
158 struct sock *sk, *result = NULL;
159 struct hlist_nulls_node *node;
160 int score, hiscore = 0;
161
162 sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) {
163 score = compute_score(sk, net, hnum, daddr, dif, sdif);
164 if (score > hiscore) {
165 result = inet6_lookup_reuseport(net, sk, skb, doff,
166 saddr, sport, daddr, hnum, inet6_ehashfn);
167 if (result)
168 return result;
169
170 result = sk;
171 hiscore = score;
172 }
173 }
174
175 return result;
176}
177
178struct sock *inet6_lookup_run_sk_lookup(struct net *net,
179 int protocol,
180 struct sk_buff *skb, int doff,
181 const struct in6_addr *saddr,
182 const __be16 sport,
183 const struct in6_addr *daddr,
184 const u16 hnum, const int dif,
185 inet6_ehashfn_t *ehashfn)
186{
187 struct sock *sk, *reuse_sk;
188 bool no_reuseport;
189
190 no_reuseport = bpf_sk_lookup_run_v6(net, protocol, saddr, sport,
191 daddr, hnum, dif, &sk);
192 if (no_reuseport || IS_ERR_OR_NULL(sk))
193 return sk;
194
195 reuse_sk = inet6_lookup_reuseport(net, sk, skb, doff,
196 saddr, sport, daddr, hnum, ehashfn);
197 if (reuse_sk)
198 sk = reuse_sk;
199 return sk;
200}
201EXPORT_SYMBOL_GPL(inet6_lookup_run_sk_lookup);
202
203struct sock *inet6_lookup_listener(struct net *net,
204 struct inet_hashinfo *hashinfo,
205 struct sk_buff *skb, int doff,
206 const struct in6_addr *saddr,
207 const __be16 sport, const struct in6_addr *daddr,
208 const unsigned short hnum, const int dif, const int sdif)
209{
210 struct inet_listen_hashbucket *ilb2;
211 struct sock *result = NULL;
212 unsigned int hash2;
213
214 /* Lookup redirect from BPF */
215 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
216 hashinfo == net->ipv4.tcp_death_row.hashinfo) {
217 result = inet6_lookup_run_sk_lookup(net, IPPROTO_TCP, skb, doff,
218 saddr, sport, daddr, hnum, dif,
219 inet6_ehashfn);
220 if (result)
221 goto done;
222 }
223
224 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
225 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
226
227 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
228 saddr, sport, daddr, hnum,
229 dif, sdif);
230 if (result)
231 goto done;
232
233 /* Lookup lhash2 with in6addr_any */
234 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
235 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
236
237 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
238 saddr, sport, &in6addr_any, hnum,
239 dif, sdif);
240done:
241 if (IS_ERR(result))
242 return NULL;
243 return result;
244}
245EXPORT_SYMBOL_GPL(inet6_lookup_listener);
246
247struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
248 struct sk_buff *skb, int doff,
249 const struct in6_addr *saddr, const __be16 sport,
250 const struct in6_addr *daddr, const __be16 dport,
251 const int dif)
252{
253 struct sock *sk;
254 bool refcounted;
255
256 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
257 ntohs(dport), dif, 0, &refcounted);
258 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
259 sk = NULL;
260 return sk;
261}
262EXPORT_SYMBOL_GPL(inet6_lookup);
263
264static int __inet6_check_established(struct inet_timewait_death_row *death_row,
265 struct sock *sk, const __u16 lport,
266 struct inet_timewait_sock **twp)
267{
268 struct inet_hashinfo *hinfo = death_row->hashinfo;
269 struct inet_sock *inet = inet_sk(sk);
270 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
271 const struct in6_addr *saddr = &sk->sk_v6_daddr;
272 const int dif = sk->sk_bound_dev_if;
273 struct net *net = sock_net(sk);
274 const int sdif = l3mdev_master_ifindex_by_index(net, dif);
275 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
276 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
277 inet->inet_dport);
278 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
279 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
280 struct sock *sk2;
281 const struct hlist_nulls_node *node;
282 struct inet_timewait_sock *tw = NULL;
283
284 spin_lock(lock);
285
286 sk_nulls_for_each(sk2, node, &head->chain) {
287 if (sk2->sk_hash != hash)
288 continue;
289
290 if (likely(inet6_match(net, sk2, saddr, daddr, ports,
291 dif, sdif))) {
292 if (sk2->sk_state == TCP_TIME_WAIT) {
293 tw = inet_twsk(sk2);
294 if (twsk_unique(sk, sk2, twp))
295 break;
296 }
297 goto not_unique;
298 }
299 }
300
301 /* Must record num and sport now. Otherwise we will see
302 * in hash table socket with a funny identity.
303 */
304 inet->inet_num = lport;
305 inet->inet_sport = htons(lport);
306 sk->sk_hash = hash;
307 WARN_ON(!sk_unhashed(sk));
308 __sk_nulls_add_node_rcu(sk, &head->chain);
309 if (tw) {
310 sk_nulls_del_node_init_rcu((struct sock *)tw);
311 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
312 }
313 spin_unlock(lock);
314 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
315
316 if (twp) {
317 *twp = tw;
318 } else if (tw) {
319 /* Silly. Should hash-dance instead... */
320 inet_twsk_deschedule_put(tw);
321 }
322 return 0;
323
324not_unique:
325 spin_unlock(lock);
326 return -EADDRNOTAVAIL;
327}
328
329static u64 inet6_sk_port_offset(const struct sock *sk)
330{
331 const struct inet_sock *inet = inet_sk(sk);
332
333 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
334 sk->sk_v6_daddr.s6_addr32,
335 inet->inet_dport);
336}
337
338int inet6_hash_connect(struct inet_timewait_death_row *death_row,
339 struct sock *sk)
340{
341 u64 port_offset = 0;
342
343 if (!inet_sk(sk)->inet_num)
344 port_offset = inet6_sk_port_offset(sk);
345 return __inet_hash_connect(death_row, sk, port_offset,
346 __inet6_check_established);
347}
348EXPORT_SYMBOL_GPL(inet6_hash_connect);
349
350int inet6_hash(struct sock *sk)
351{
352 int err = 0;
353
354 if (sk->sk_state != TCP_CLOSE)
355 err = __inet_hash(sk, NULL);
356
357 return err;
358}
359EXPORT_SYMBOL_GPL(inet6_hash);