]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - net/ipv6/inet6_hashtables.c
tcp: Don't acquire inet_listen_hashbucket::lock with disabled BH.
[mirror_ubuntu-jammy-kernel.git] / net / ipv6 / inet6_hashtables.c
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
24 extern struct inet_hashinfo tcp_hashinfo;
25
26 u32 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 */
51 struct 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
70 begin:
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;
87 out:
88 sk = NULL;
89 found:
90 return sk;
91 }
92 EXPORT_SYMBOL(__inet6_lookup_established);
93
94 static 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
116 static 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() */
134 static 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
162 static 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
187 struct 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);
222 done:
223 if (IS_ERR(result))
224 return NULL;
225 return result;
226 }
227 EXPORT_SYMBOL_GPL(inet6_lookup_listener);
228
229 struct 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 }
244 EXPORT_SYMBOL_GPL(inet6_lookup);
245
246 static 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
306 not_unique:
307 spin_unlock(lock);
308 return -EADDRNOTAVAIL;
309 }
310
311 static 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
320 int 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 }
330 EXPORT_SYMBOL_GPL(inet6_hash_connect);
331
332 int inet6_hash(struct sock *sk)
333 {
334 int err = 0;
335
336 if (sk->sk_state != TCP_CLOSE)
337 err = __inet_hash(sk, NULL);
338
339 return err;
340 }
341 EXPORT_SYMBOL_GPL(inet6_hash);