Loading...
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
24extern struct inet_hashinfo tcp_hashinfo;
25
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)
29{
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
41 return __inet6_ehashfn(lhash, lport, fhash, fport,
42 inet6_ehash_secret + net_hash_mix(net));
43}
44
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 */
51struct sock *__inet6_lookup_established(struct net *net,
52 struct inet_hashinfo *hashinfo,
53 const struct in6_addr *saddr,
54 const __be16 sport,
55 const struct in6_addr *daddr,
56 const u16 hnum,
57 const int dif, const int sdif)
58{
59 struct sock *sk;
60 const struct hlist_nulls_node *node;
61 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
62 /* Optimize here for direct hit, only listening connections can
63 * have wildcards anyways.
64 */
65 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
66 unsigned int slot = hash & hashinfo->ehash_mask;
67 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
68
69
70begin:
71 sk_nulls_for_each_rcu(sk, node, &head->chain) {
72 if (sk->sk_hash != hash)
73 continue;
74 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif, sdif))
75 continue;
76 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
77 goto out;
78
79 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif, sdif))) {
80 sock_gen_put(sk);
81 goto begin;
82 }
83 goto found;
84 }
85 if (get_nulls_value(node) != slot)
86 goto begin;
87out:
88 sk = NULL;
89found:
90 return sk;
91}
92EXPORT_SYMBOL(__inet6_lookup_established);
93
94static inline int compute_score(struct sock *sk, struct net *net,
95 const unsigned short hnum,
96 const struct in6_addr *daddr,
97 const int dif, const int sdif)
98{
99 int score = -1;
100
101 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
102 sk->sk_family == PF_INET6) {
103 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
104 return -1;
105
106 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
107 return -1;
108
109 score = sk->sk_bound_dev_if ? 2 : 1;
110 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
111 score++;
112 }
113 return score;
114}
115
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
133/* called with rcu_read_lock() */
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{
141 struct inet_connection_sock *icsk;
142 struct sock *sk, *result = NULL;
143 int score, hiscore = 0;
144
145 inet_lhash2_for_each_icsk_rcu(icsk, &ilb2->head) {
146 sk = (struct sock *)icsk;
147 score = compute_score(sk, net, hnum, daddr, dif, sdif);
148 if (score > hiscore) {
149 result = lookup_reuseport(net, sk, skb, doff,
150 saddr, sport, daddr, hnum);
151 if (result)
152 return result;
153
154 result = sk;
155 hiscore = score;
156 }
157 }
158
159 return result;
160}
161
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,
168 const u16 hnum)
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
176 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_TCP,
177 saddr, sport, daddr, hnum, &sk);
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
187struct sock *inet6_lookup_listener(struct net *net,
188 struct inet_hashinfo *hashinfo,
189 struct sk_buff *skb, int doff,
190 const struct in6_addr *saddr,
191 const __be16 sport, const struct in6_addr *daddr,
192 const unsigned short hnum, const int dif, const int sdif)
193{
194 struct inet_listen_hashbucket *ilb2;
195 struct sock *result = NULL;
196 unsigned int hash2;
197
198 /* Lookup redirect from BPF */
199 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
200 result = inet6_lookup_run_bpf(net, hashinfo, skb, doff,
201 saddr, sport, daddr, hnum);
202 if (result)
203 goto done;
204 }
205
206 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
207 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
208
209 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
210 saddr, sport, daddr, hnum,
211 dif, sdif);
212 if (result)
213 goto done;
214
215 /* Lookup lhash2 with in6addr_any */
216 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
217 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
218
219 result = inet6_lhash2_lookup(net, ilb2, skb, doff,
220 saddr, sport, &in6addr_any, hnum,
221 dif, sdif);
222done:
223 if (IS_ERR(result))
224 return NULL;
225 return result;
226}
227EXPORT_SYMBOL_GPL(inet6_lookup_listener);
228
229struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
230 struct sk_buff *skb, int doff,
231 const struct in6_addr *saddr, const __be16 sport,
232 const struct in6_addr *daddr, const __be16 dport,
233 const int dif)
234{
235 struct sock *sk;
236 bool refcounted;
237
238 sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
239 ntohs(dport), dif, 0, &refcounted);
240 if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
241 sk = NULL;
242 return sk;
243}
244EXPORT_SYMBOL_GPL(inet6_lookup);
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;
251 struct inet_sock *inet = inet_sk(sk);
252 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
253 const struct in6_addr *saddr = &sk->sk_v6_daddr;
254 const int dif = sk->sk_bound_dev_if;
255 struct net *net = sock_net(sk);
256 const int sdif = l3mdev_master_ifindex_by_index(net, dif);
257 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
258 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
259 inet->inet_dport);
260 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
261 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
262 struct sock *sk2;
263 const struct hlist_nulls_node *node;
264 struct inet_timewait_sock *tw = NULL;
265
266 spin_lock(lock);
267
268 sk_nulls_for_each(sk2, node, &head->chain) {
269 if (sk2->sk_hash != hash)
270 continue;
271
272 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports,
273 dif, sdif))) {
274 if (sk2->sk_state == TCP_TIME_WAIT) {
275 tw = inet_twsk(sk2);
276 if (twsk_unique(sk, sk2, twp))
277 break;
278 }
279 goto not_unique;
280 }
281 }
282
283 /* Must record num and sport now. Otherwise we will see
284 * in hash table socket with a funny identity.
285 */
286 inet->inet_num = lport;
287 inet->inet_sport = htons(lport);
288 sk->sk_hash = hash;
289 WARN_ON(!sk_unhashed(sk));
290 __sk_nulls_add_node_rcu(sk, &head->chain);
291 if (tw) {
292 sk_nulls_del_node_init_rcu((struct sock *)tw);
293 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
294 }
295 spin_unlock(lock);
296 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
297
298 if (twp) {
299 *twp = tw;
300 } else if (tw) {
301 /* Silly. Should hash-dance instead... */
302 inet_twsk_deschedule_put(tw);
303 }
304 return 0;
305
306not_unique:
307 spin_unlock(lock);
308 return -EADDRNOTAVAIL;
309}
310
311static u32 inet6_sk_port_offset(const struct sock *sk)
312{
313 const struct inet_sock *inet = inet_sk(sk);
314
315 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
316 sk->sk_v6_daddr.s6_addr32,
317 inet->inet_dport);
318}
319
320int inet6_hash_connect(struct inet_timewait_death_row *death_row,
321 struct sock *sk)
322{
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,
328 __inet6_check_established);
329}
330EXPORT_SYMBOL_GPL(inet6_hash_connect);
331
332int inet6_hash(struct sock *sk)
333{
334 int err = 0;
335
336 if (sk->sk_state != TCP_CLOSE) {
337 local_bh_disable();
338 err = __inet_hash(sk, NULL);
339 local_bh_enable();
340 }
341
342 return err;
343}
344EXPORT_SYMBOL_GPL(inet6_hash);
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
26static unsigned int inet6_ehashfn(struct net *net,
27 const struct in6_addr *laddr,
28 const u16 lport,
29 const struct in6_addr *faddr,
30 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
47static int inet6_sk_ehashfn(const struct sock *sk)
48{
49 const struct inet_sock *inet = inet_sk(sk);
50 const struct in6_addr *laddr = &sk->sk_v6_rcv_saddr;
51 const struct in6_addr *faddr = &sk->sk_v6_daddr;
52 const __u16 lport = inet->inet_num;
53 const __be16 fport = inet->inet_dport;
54 struct net *net = sock_net(sk);
55
56 return inet6_ehashfn(net, laddr, lport, faddr, fport);
57}
58
59int __inet6_hash(struct sock *sk, struct inet_timewait_sock *tw)
60{
61 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
62 int twrefcnt = 0;
63
64 WARN_ON(!sk_unhashed(sk));
65
66 if (sk->sk_state == TCP_LISTEN) {
67 struct inet_listen_hashbucket *ilb;
68
69 ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
70 spin_lock(&ilb->lock);
71 __sk_nulls_add_node_rcu(sk, &ilb->head);
72 spin_unlock(&ilb->lock);
73 } else {
74 unsigned int hash;
75 struct hlist_nulls_head *list;
76 spinlock_t *lock;
77
78 sk->sk_hash = hash = inet6_sk_ehashfn(sk);
79 list = &inet_ehash_bucket(hashinfo, hash)->chain;
80 lock = inet_ehash_lockp(hashinfo, hash);
81 spin_lock(lock);
82 __sk_nulls_add_node_rcu(sk, list);
83 if (tw) {
84 WARN_ON(sk->sk_hash != tw->tw_hash);
85 twrefcnt = inet_twsk_unhash(tw);
86 }
87 spin_unlock(lock);
88 }
89
90 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
91 return twrefcnt;
92}
93EXPORT_SYMBOL(__inet6_hash);
94
95/*
96 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
97 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
98 *
99 * The sockhash lock must be held as a reader here.
100 */
101struct sock *__inet6_lookup_established(struct net *net,
102 struct inet_hashinfo *hashinfo,
103 const struct in6_addr *saddr,
104 const __be16 sport,
105 const struct in6_addr *daddr,
106 const u16 hnum,
107 const int dif)
108{
109 struct sock *sk;
110 const struct hlist_nulls_node *node;
111 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
112 /* Optimize here for direct hit, only listening connections can
113 * have wildcards anyways.
114 */
115 unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
116 unsigned int slot = hash & hashinfo->ehash_mask;
117 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
118
119
120 rcu_read_lock();
121begin:
122 sk_nulls_for_each_rcu(sk, node, &head->chain) {
123 if (sk->sk_hash != hash)
124 continue;
125 if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif))
126 continue;
127 if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
128 goto out;
129
130 if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
131 sock_gen_put(sk);
132 goto begin;
133 }
134 goto found;
135 }
136 if (get_nulls_value(node) != slot)
137 goto begin;
138out:
139 sk = NULL;
140found:
141 rcu_read_unlock();
142 return sk;
143}
144EXPORT_SYMBOL(__inet6_lookup_established);
145
146static inline int compute_score(struct sock *sk, struct net *net,
147 const unsigned short hnum,
148 const struct in6_addr *daddr,
149 const int dif)
150{
151 int score = -1;
152
153 if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
154 sk->sk_family == PF_INET6) {
155
156 score = 1;
157 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
158 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
159 return -1;
160 score++;
161 }
162 if (sk->sk_bound_dev_if) {
163 if (sk->sk_bound_dev_if != dif)
164 return -1;
165 score++;
166 }
167 }
168 return score;
169}
170
171struct sock *inet6_lookup_listener(struct net *net,
172 struct inet_hashinfo *hashinfo, const struct in6_addr *saddr,
173 const __be16 sport, const struct in6_addr *daddr,
174 const unsigned short hnum, const int dif)
175{
176 struct sock *sk;
177 const struct hlist_nulls_node *node;
178 struct sock *result;
179 int score, hiscore, matches = 0, reuseport = 0;
180 u32 phash = 0;
181 unsigned int hash = inet_lhashfn(net, hnum);
182 struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
183
184 rcu_read_lock();
185begin:
186 result = NULL;
187 hiscore = 0;
188 sk_nulls_for_each(sk, node, &ilb->head) {
189 score = compute_score(sk, net, hnum, daddr, dif);
190 if (score > hiscore) {
191 hiscore = score;
192 result = sk;
193 reuseport = sk->sk_reuseport;
194 if (reuseport) {
195 phash = inet6_ehashfn(net, daddr, hnum,
196 saddr, sport);
197 matches = 1;
198 }
199 } else if (score == hiscore && reuseport) {
200 matches++;
201 if (((u64)phash * matches) >> 32 == 0)
202 result = sk;
203 phash = next_pseudo_random32(phash);
204 }
205 }
206 /*
207 * if the nulls value we got at the end of this lookup is
208 * not the expected one, we must restart lookup.
209 * We probably met an item that was moved to another chain.
210 */
211 if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
212 goto begin;
213 if (result) {
214 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
215 result = NULL;
216 else if (unlikely(compute_score(result, net, hnum, daddr,
217 dif) < hiscore)) {
218 sock_put(result);
219 goto begin;
220 }
221 }
222 rcu_read_unlock();
223 return result;
224}
225
226EXPORT_SYMBOL_GPL(inet6_lookup_listener);
227
228struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
229 const struct in6_addr *saddr, const __be16 sport,
230 const struct in6_addr *daddr, const __be16 dport,
231 const int dif)
232{
233 struct sock *sk;
234
235 local_bh_disable();
236 sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
237 local_bh_enable();
238
239 return sk;
240}
241
242EXPORT_SYMBOL_GPL(inet6_lookup);
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;
249 struct inet_sock *inet = inet_sk(sk);
250 const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
251 const struct in6_addr *saddr = &sk->sk_v6_daddr;
252 const int dif = sk->sk_bound_dev_if;
253 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
254 struct net *net = sock_net(sk);
255 const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
256 inet->inet_dport);
257 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
258 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
259 struct sock *sk2;
260 const struct hlist_nulls_node *node;
261 struct inet_timewait_sock *tw = NULL;
262 int twrefcnt = 0;
263
264 spin_lock(lock);
265
266 sk_nulls_for_each(sk2, node, &head->chain) {
267 if (sk2->sk_hash != hash)
268 continue;
269
270 if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) {
271 if (sk2->sk_state == TCP_TIME_WAIT) {
272 tw = inet_twsk(sk2);
273 if (twsk_unique(sk, sk2, twp))
274 break;
275 }
276 goto not_unique;
277 }
278 }
279
280 /* Must record num and sport now. Otherwise we will see
281 * in hash table socket with a funny identity.
282 */
283 inet->inet_num = lport;
284 inet->inet_sport = htons(lport);
285 sk->sk_hash = hash;
286 WARN_ON(!sk_unhashed(sk));
287 __sk_nulls_add_node_rcu(sk, &head->chain);
288 if (tw) {
289 twrefcnt = inet_twsk_unhash(tw);
290 NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
291 }
292 spin_unlock(lock);
293 if (twrefcnt)
294 inet_twsk_put(tw);
295 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
296
297 if (twp) {
298 *twp = tw;
299 } else if (tw) {
300 /* Silly. Should hash-dance instead... */
301 inet_twsk_deschedule(tw, death_row);
302
303 inet_twsk_put(tw);
304 }
305 return 0;
306
307not_unique:
308 spin_unlock(lock);
309 return -EADDRNOTAVAIL;
310}
311
312static inline u32 inet6_sk_port_offset(const struct sock *sk)
313{
314 const struct inet_sock *inet = inet_sk(sk);
315
316 return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
317 sk->sk_v6_daddr.s6_addr32,
318 inet->inet_dport);
319}
320
321int inet6_hash_connect(struct inet_timewait_death_row *death_row,
322 struct sock *sk)
323{
324 return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
325 __inet6_check_established, __inet6_hash);
326}
327
328EXPORT_SYMBOL_GPL(inet6_hash_connect);