Commit | Line | Data |
---|---|---|
77d8bf9c ACM |
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 INET transport hashtables | |
7 | * | |
8 | * Authors: Lotsa people, from code originally in tcp | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public License | |
12 | * as published by the Free Software Foundation; either version | |
13 | * 2 of the License, or (at your option) any later version. | |
14 | */ | |
15 | ||
2d8c4ce5 | 16 | #include <linux/module.h> |
a7f5e7f1 | 17 | #include <linux/random.h> |
f3f05f70 | 18 | #include <linux/sched.h> |
77d8bf9c | 19 | #include <linux/slab.h> |
f3f05f70 | 20 | #include <linux/wait.h> |
095dc8e0 | 21 | #include <linux/vmalloc.h> |
77d8bf9c | 22 | |
463c84b9 | 23 | #include <net/inet_connection_sock.h> |
77d8bf9c | 24 | #include <net/inet_hashtables.h> |
6e5714ea | 25 | #include <net/secure_seq.h> |
a7f5e7f1 | 26 | #include <net/ip.h> |
77d8bf9c | 27 | |
6eada011 ED |
28 | static u32 inet_ehashfn(const struct net *net, const __be32 laddr, |
29 | const __u16 lport, const __be32 faddr, | |
30 | const __be16 fport) | |
65cd8033 | 31 | { |
1bbdceef HFS |
32 | static u32 inet_ehash_secret __read_mostly; |
33 | ||
34 | net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret)); | |
35 | ||
65cd8033 HFS |
36 | return __inet_ehashfn(laddr, lport, faddr, fport, |
37 | inet_ehash_secret + net_hash_mix(net)); | |
38 | } | |
39 | ||
d1e559d0 ED |
40 | /* This function handles inet_sock, but also timewait and request sockets |
41 | * for IPv4/IPv6. | |
42 | */ | |
5b441f76 | 43 | u32 sk_ehashfn(const struct sock *sk) |
65cd8033 | 44 | { |
d1e559d0 ED |
45 | #if IS_ENABLED(CONFIG_IPV6) |
46 | if (sk->sk_family == AF_INET6 && | |
47 | !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) | |
48 | return inet6_ehashfn(sock_net(sk), | |
49 | &sk->sk_v6_rcv_saddr, sk->sk_num, | |
50 | &sk->sk_v6_daddr, sk->sk_dport); | |
51 | #endif | |
5b441f76 ED |
52 | return inet_ehashfn(sock_net(sk), |
53 | sk->sk_rcv_saddr, sk->sk_num, | |
54 | sk->sk_daddr, sk->sk_dport); | |
65cd8033 HFS |
55 | } |
56 | ||
77d8bf9c ACM |
57 | /* |
58 | * Allocate and initialize a new local port bind bucket. | |
59 | * The bindhash mutex for snum's hash chain must be held here. | |
60 | */ | |
e18b890b | 61 | struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep, |
941b1d22 | 62 | struct net *net, |
77d8bf9c ACM |
63 | struct inet_bind_hashbucket *head, |
64 | const unsigned short snum) | |
65 | { | |
54e6ecb2 | 66 | struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC); |
77d8bf9c | 67 | |
00db4124 | 68 | if (tb) { |
efd7ef1c | 69 | write_pnet(&tb->ib_net, net); |
77d8bf9c ACM |
70 | tb->port = snum; |
71 | tb->fastreuse = 0; | |
da5e3630 | 72 | tb->fastreuseport = 0; |
a9d8f911 | 73 | tb->num_owners = 0; |
77d8bf9c ACM |
74 | INIT_HLIST_HEAD(&tb->owners); |
75 | hlist_add_head(&tb->node, &head->chain); | |
76 | } | |
77 | return tb; | |
78 | } | |
79 | ||
77d8bf9c ACM |
80 | /* |
81 | * Caller must hold hashbucket lock for this tb with local BH disabled | |
82 | */ | |
e18b890b | 83 | void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb) |
77d8bf9c ACM |
84 | { |
85 | if (hlist_empty(&tb->owners)) { | |
86 | __hlist_del(&tb->node); | |
87 | kmem_cache_free(cachep, tb); | |
88 | } | |
89 | } | |
2d8c4ce5 ACM |
90 | |
91 | void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, | |
92 | const unsigned short snum) | |
93 | { | |
c720c7e8 | 94 | inet_sk(sk)->inet_num = snum; |
2d8c4ce5 | 95 | sk_add_bind_node(sk, &tb->owners); |
a9d8f911 | 96 | tb->num_owners++; |
463c84b9 | 97 | inet_csk(sk)->icsk_bind_hash = tb; |
2d8c4ce5 ACM |
98 | } |
99 | ||
2d8c4ce5 ACM |
100 | /* |
101 | * Get rid of any references to a local port held by the given sock. | |
102 | */ | |
ab1e0a13 | 103 | static void __inet_put_port(struct sock *sk) |
2d8c4ce5 | 104 | { |
39d8cda7 | 105 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
c720c7e8 | 106 | const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num, |
7f635ab7 | 107 | hashinfo->bhash_size); |
2d8c4ce5 ACM |
108 | struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash]; |
109 | struct inet_bind_bucket *tb; | |
110 | ||
111 | spin_lock(&head->lock); | |
463c84b9 | 112 | tb = inet_csk(sk)->icsk_bind_hash; |
2d8c4ce5 | 113 | __sk_del_bind_node(sk); |
a9d8f911 | 114 | tb->num_owners--; |
463c84b9 | 115 | inet_csk(sk)->icsk_bind_hash = NULL; |
c720c7e8 | 116 | inet_sk(sk)->inet_num = 0; |
2d8c4ce5 ACM |
117 | inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb); |
118 | spin_unlock(&head->lock); | |
119 | } | |
120 | ||
ab1e0a13 | 121 | void inet_put_port(struct sock *sk) |
2d8c4ce5 ACM |
122 | { |
123 | local_bh_disable(); | |
ab1e0a13 | 124 | __inet_put_port(sk); |
2d8c4ce5 ACM |
125 | local_bh_enable(); |
126 | } | |
2d8c4ce5 | 127 | EXPORT_SYMBOL(inet_put_port); |
f3f05f70 | 128 | |
093d2823 | 129 | int __inet_inherit_port(struct sock *sk, struct sock *child) |
53083773 PE |
130 | { |
131 | struct inet_hashinfo *table = sk->sk_prot->h.hashinfo; | |
093d2823 BS |
132 | unsigned short port = inet_sk(child)->inet_num; |
133 | const int bhash = inet_bhashfn(sock_net(sk), port, | |
7f635ab7 | 134 | table->bhash_size); |
53083773 PE |
135 | struct inet_bind_hashbucket *head = &table->bhash[bhash]; |
136 | struct inet_bind_bucket *tb; | |
137 | ||
138 | spin_lock(&head->lock); | |
139 | tb = inet_csk(sk)->icsk_bind_hash; | |
093d2823 BS |
140 | if (tb->port != port) { |
141 | /* NOTE: using tproxy and redirecting skbs to a proxy | |
142 | * on a different listener port breaks the assumption | |
143 | * that the listener socket's icsk_bind_hash is the same | |
144 | * as that of the child socket. We have to look up or | |
145 | * create a new bind bucket for the child here. */ | |
b67bfe0d | 146 | inet_bind_bucket_for_each(tb, &head->chain) { |
093d2823 BS |
147 | if (net_eq(ib_net(tb), sock_net(sk)) && |
148 | tb->port == port) | |
149 | break; | |
150 | } | |
b67bfe0d | 151 | if (!tb) { |
093d2823 BS |
152 | tb = inet_bind_bucket_create(table->bind_bucket_cachep, |
153 | sock_net(sk), head, port); | |
154 | if (!tb) { | |
155 | spin_unlock(&head->lock); | |
156 | return -ENOMEM; | |
157 | } | |
158 | } | |
159 | } | |
b4ff3c90 | 160 | inet_bind_hash(child, tb, port); |
53083773 | 161 | spin_unlock(&head->lock); |
093d2823 BS |
162 | |
163 | return 0; | |
53083773 | 164 | } |
53083773 PE |
165 | EXPORT_SYMBOL_GPL(__inet_inherit_port); |
166 | ||
c25eb3bf ED |
167 | static inline int compute_score(struct sock *sk, struct net *net, |
168 | const unsigned short hnum, const __be32 daddr, | |
169 | const int dif) | |
170 | { | |
171 | int score = -1; | |
172 | struct inet_sock *inet = inet_sk(sk); | |
173 | ||
c720c7e8 | 174 | if (net_eq(sock_net(sk), net) && inet->inet_num == hnum && |
c25eb3bf | 175 | !ipv6_only_sock(sk)) { |
c720c7e8 | 176 | __be32 rcv_saddr = inet->inet_rcv_saddr; |
da5e3630 | 177 | score = sk->sk_family == PF_INET ? 2 : 1; |
c25eb3bf ED |
178 | if (rcv_saddr) { |
179 | if (rcv_saddr != daddr) | |
180 | return -1; | |
da5e3630 | 181 | score += 4; |
c25eb3bf ED |
182 | } |
183 | if (sk->sk_bound_dev_if) { | |
184 | if (sk->sk_bound_dev_if != dif) | |
185 | return -1; | |
da5e3630 | 186 | score += 4; |
c25eb3bf ED |
187 | } |
188 | } | |
189 | return score; | |
190 | } | |
191 | ||
33b62231 ACM |
192 | /* |
193 | * Don't inline this cruft. Here are some nice properties to exploit here. The | |
194 | * BSD API does not allow a listening sock to specify the remote port nor the | |
195 | * remote address for the connection. So always assume those are both | |
196 | * wildcarded during the search since they can never be otherwise. | |
197 | */ | |
e48c414e | 198 | |
c25eb3bf | 199 | |
c67499c0 PE |
200 | struct sock *__inet_lookup_listener(struct net *net, |
201 | struct inet_hashinfo *hashinfo, | |
da5e3630 | 202 | const __be32 saddr, __be16 sport, |
fb99c848 | 203 | const __be32 daddr, const unsigned short hnum, |
8f491069 | 204 | const int dif) |
99a92ff5 | 205 | { |
c25eb3bf ED |
206 | struct sock *sk, *result; |
207 | struct hlist_nulls_node *node; | |
208 | unsigned int hash = inet_lhashfn(net, hnum); | |
209 | struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash]; | |
da5e3630 TH |
210 | int score, hiscore, matches = 0, reuseport = 0; |
211 | u32 phash = 0; | |
99a92ff5 | 212 | |
c25eb3bf ED |
213 | rcu_read_lock(); |
214 | begin: | |
215 | result = NULL; | |
da5e3630 | 216 | hiscore = 0; |
c25eb3bf ED |
217 | sk_nulls_for_each_rcu(sk, node, &ilb->head) { |
218 | score = compute_score(sk, net, hnum, daddr, dif); | |
219 | if (score > hiscore) { | |
220 | result = sk; | |
221 | hiscore = score; | |
da5e3630 TH |
222 | reuseport = sk->sk_reuseport; |
223 | if (reuseport) { | |
224 | phash = inet_ehashfn(net, daddr, hnum, | |
225 | saddr, sport); | |
226 | matches = 1; | |
227 | } | |
228 | } else if (score == hiscore && reuseport) { | |
229 | matches++; | |
8fc54f68 | 230 | if (reciprocal_scale(phash, matches) == 0) |
da5e3630 TH |
231 | result = sk; |
232 | phash = next_pseudo_random32(phash); | |
c25eb3bf | 233 | } |
99a92ff5 | 234 | } |
c25eb3bf ED |
235 | /* |
236 | * if the nulls value we got at the end of this lookup is | |
237 | * not the expected one, we must restart lookup. | |
238 | * We probably met an item that was moved to another chain. | |
239 | */ | |
240 | if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE) | |
241 | goto begin; | |
242 | if (result) { | |
243 | if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) | |
244 | result = NULL; | |
245 | else if (unlikely(compute_score(result, net, hnum, daddr, | |
246 | dif) < hiscore)) { | |
247 | sock_put(result); | |
248 | goto begin; | |
249 | } | |
99a92ff5 | 250 | } |
c25eb3bf ED |
251 | rcu_read_unlock(); |
252 | return result; | |
99a92ff5 | 253 | } |
8f491069 | 254 | EXPORT_SYMBOL_GPL(__inet_lookup_listener); |
a7f5e7f1 | 255 | |
05dbc7b5 ED |
256 | /* All sockets share common refcount, but have different destructors */ |
257 | void sock_gen_put(struct sock *sk) | |
258 | { | |
259 | if (!atomic_dec_and_test(&sk->sk_refcnt)) | |
260 | return; | |
261 | ||
262 | if (sk->sk_state == TCP_TIME_WAIT) | |
263 | inet_twsk_free(inet_twsk(sk)); | |
41b822c5 ED |
264 | else if (sk->sk_state == TCP_NEW_SYN_RECV) |
265 | reqsk_free(inet_reqsk(sk)); | |
05dbc7b5 ED |
266 | else |
267 | sk_free(sk); | |
268 | } | |
269 | EXPORT_SYMBOL_GPL(sock_gen_put); | |
270 | ||
2c13270b ED |
271 | void sock_edemux(struct sk_buff *skb) |
272 | { | |
273 | sock_gen_put(skb->sk); | |
274 | } | |
275 | EXPORT_SYMBOL(sock_edemux); | |
276 | ||
5e73ea1a | 277 | struct sock *__inet_lookup_established(struct net *net, |
c67499c0 | 278 | struct inet_hashinfo *hashinfo, |
77a5ba55 PE |
279 | const __be32 saddr, const __be16 sport, |
280 | const __be32 daddr, const u16 hnum, | |
281 | const int dif) | |
282 | { | |
c7228317 | 283 | INET_ADDR_COOKIE(acookie, saddr, daddr); |
77a5ba55 PE |
284 | const __portpair ports = INET_COMBINED_PORTS(sport, hnum); |
285 | struct sock *sk; | |
3ab5aee7 | 286 | const struct hlist_nulls_node *node; |
77a5ba55 PE |
287 | /* Optimize here for direct hit, only listening connections can |
288 | * have wildcards anyways. | |
289 | */ | |
9f26b3ad | 290 | unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport); |
f373b53b | 291 | unsigned int slot = hash & hashinfo->ehash_mask; |
3ab5aee7 | 292 | struct inet_ehash_bucket *head = &hashinfo->ehash[slot]; |
77a5ba55 | 293 | |
3ab5aee7 ED |
294 | rcu_read_lock(); |
295 | begin: | |
296 | sk_nulls_for_each_rcu(sk, node, &head->chain) { | |
ce43b03e ED |
297 | if (sk->sk_hash != hash) |
298 | continue; | |
299 | if (likely(INET_MATCH(sk, net, acookie, | |
300 | saddr, daddr, ports, dif))) { | |
3ab5aee7 | 301 | if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) |
05dbc7b5 | 302 | goto out; |
ce43b03e ED |
303 | if (unlikely(!INET_MATCH(sk, net, acookie, |
304 | saddr, daddr, ports, dif))) { | |
05dbc7b5 | 305 | sock_gen_put(sk); |
3ab5aee7 ED |
306 | goto begin; |
307 | } | |
05dbc7b5 | 308 | goto found; |
3ab5aee7 | 309 | } |
77a5ba55 | 310 | } |
3ab5aee7 ED |
311 | /* |
312 | * if the nulls value we got at the end of this lookup is | |
313 | * not the expected one, we must restart lookup. | |
314 | * We probably met an item that was moved to another chain. | |
315 | */ | |
316 | if (get_nulls_value(node) != slot) | |
317 | goto begin; | |
77a5ba55 | 318 | out: |
05dbc7b5 ED |
319 | sk = NULL; |
320 | found: | |
3ab5aee7 | 321 | rcu_read_unlock(); |
77a5ba55 | 322 | return sk; |
77a5ba55 PE |
323 | } |
324 | EXPORT_SYMBOL_GPL(__inet_lookup_established); | |
325 | ||
a7f5e7f1 ACM |
326 | /* called with local bh disabled */ |
327 | static int __inet_check_established(struct inet_timewait_death_row *death_row, | |
328 | struct sock *sk, __u16 lport, | |
329 | struct inet_timewait_sock **twp) | |
330 | { | |
331 | struct inet_hashinfo *hinfo = death_row->hashinfo; | |
332 | struct inet_sock *inet = inet_sk(sk); | |
c720c7e8 ED |
333 | __be32 daddr = inet->inet_rcv_saddr; |
334 | __be32 saddr = inet->inet_daddr; | |
a7f5e7f1 | 335 | int dif = sk->sk_bound_dev_if; |
c7228317 | 336 | INET_ADDR_COOKIE(acookie, saddr, daddr); |
c720c7e8 | 337 | const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport); |
9f26b3ad | 338 | struct net *net = sock_net(sk); |
c720c7e8 ED |
339 | unsigned int hash = inet_ehashfn(net, daddr, lport, |
340 | saddr, inet->inet_dport); | |
a7f5e7f1 | 341 | struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash); |
9db66bdc | 342 | spinlock_t *lock = inet_ehash_lockp(hinfo, hash); |
a7f5e7f1 | 343 | struct sock *sk2; |
3ab5aee7 | 344 | const struct hlist_nulls_node *node; |
05dbc7b5 | 345 | struct inet_timewait_sock *tw = NULL; |
a7f5e7f1 | 346 | |
9db66bdc | 347 | spin_lock(lock); |
a7f5e7f1 | 348 | |
3ab5aee7 | 349 | sk_nulls_for_each(sk2, node, &head->chain) { |
ce43b03e ED |
350 | if (sk2->sk_hash != hash) |
351 | continue; | |
05dbc7b5 | 352 | |
ce43b03e | 353 | if (likely(INET_MATCH(sk2, net, acookie, |
05dbc7b5 ED |
354 | saddr, daddr, ports, dif))) { |
355 | if (sk2->sk_state == TCP_TIME_WAIT) { | |
356 | tw = inet_twsk(sk2); | |
357 | if (twsk_unique(sk, sk2, twp)) | |
358 | break; | |
359 | } | |
a7f5e7f1 | 360 | goto not_unique; |
05dbc7b5 | 361 | } |
a7f5e7f1 ACM |
362 | } |
363 | ||
a7f5e7f1 | 364 | /* Must record num and sport now. Otherwise we will see |
05dbc7b5 ED |
365 | * in hash table socket with a funny identity. |
366 | */ | |
c720c7e8 ED |
367 | inet->inet_num = lport; |
368 | inet->inet_sport = htons(lport); | |
a7f5e7f1 | 369 | sk->sk_hash = hash; |
547b792c | 370 | WARN_ON(!sk_unhashed(sk)); |
3ab5aee7 | 371 | __sk_nulls_add_node_rcu(sk, &head->chain); |
13475a30 | 372 | if (tw) { |
fc01538f | 373 | sk_nulls_del_node_init_rcu((struct sock *)tw); |
13475a30 ED |
374 | NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED); |
375 | } | |
9db66bdc | 376 | spin_unlock(lock); |
c29a0bc4 | 377 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
a7f5e7f1 ACM |
378 | |
379 | if (twp) { | |
380 | *twp = tw; | |
a7f5e7f1 ACM |
381 | } else if (tw) { |
382 | /* Silly. Should hash-dance instead... */ | |
dbe7faa4 | 383 | inet_twsk_deschedule_put(tw); |
a7f5e7f1 | 384 | } |
a7f5e7f1 ACM |
385 | return 0; |
386 | ||
387 | not_unique: | |
9db66bdc | 388 | spin_unlock(lock); |
a7f5e7f1 ACM |
389 | return -EADDRNOTAVAIL; |
390 | } | |
391 | ||
e2baad9e | 392 | static u32 inet_sk_port_offset(const struct sock *sk) |
a7f5e7f1 ACM |
393 | { |
394 | const struct inet_sock *inet = inet_sk(sk); | |
e2baad9e | 395 | |
c720c7e8 ED |
396 | return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr, |
397 | inet->inet_daddr, | |
398 | inet->inet_dport); | |
a7f5e7f1 ACM |
399 | } |
400 | ||
fc01538f | 401 | void __inet_hash_nolisten(struct sock *sk, struct sock *osk) |
152da81d | 402 | { |
39d8cda7 | 403 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
3ab5aee7 | 404 | struct hlist_nulls_head *list; |
152da81d | 405 | struct inet_ehash_bucket *head; |
5b441f76 | 406 | spinlock_t *lock; |
152da81d | 407 | |
547b792c | 408 | WARN_ON(!sk_unhashed(sk)); |
152da81d | 409 | |
5b441f76 | 410 | sk->sk_hash = sk_ehashfn(sk); |
152da81d PE |
411 | head = inet_ehash_bucket(hashinfo, sk->sk_hash); |
412 | list = &head->chain; | |
413 | lock = inet_ehash_lockp(hashinfo, sk->sk_hash); | |
414 | ||
9db66bdc | 415 | spin_lock(lock); |
3ab5aee7 | 416 | __sk_nulls_add_node_rcu(sk, list); |
fc01538f ED |
417 | if (osk) { |
418 | WARN_ON(sk->sk_hash != osk->sk_hash); | |
419 | sk_nulls_del_node_init_rcu(osk); | |
9327f705 | 420 | } |
9db66bdc | 421 | spin_unlock(lock); |
c29a0bc4 | 422 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
152da81d PE |
423 | } |
424 | EXPORT_SYMBOL_GPL(__inet_hash_nolisten); | |
425 | ||
fc01538f | 426 | void __inet_hash(struct sock *sk, struct sock *osk) |
152da81d | 427 | { |
39d8cda7 | 428 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
5caea4ea | 429 | struct inet_listen_hashbucket *ilb; |
152da81d | 430 | |
77a6a471 | 431 | if (sk->sk_state != TCP_LISTEN) |
fc01538f | 432 | return __inet_hash_nolisten(sk, osk); |
152da81d | 433 | |
547b792c | 434 | WARN_ON(!sk_unhashed(sk)); |
5caea4ea | 435 | ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)]; |
152da81d | 436 | |
5caea4ea | 437 | spin_lock(&ilb->lock); |
c25eb3bf | 438 | __sk_nulls_add_node_rcu(sk, &ilb->head); |
c29a0bc4 | 439 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
5caea4ea | 440 | spin_unlock(&ilb->lock); |
152da81d | 441 | } |
77a6a471 | 442 | EXPORT_SYMBOL(__inet_hash); |
ab1e0a13 ACM |
443 | |
444 | void inet_hash(struct sock *sk) | |
445 | { | |
446 | if (sk->sk_state != TCP_CLOSE) { | |
447 | local_bh_disable(); | |
77a6a471 | 448 | __inet_hash(sk, NULL); |
ab1e0a13 ACM |
449 | local_bh_enable(); |
450 | } | |
451 | } | |
452 | EXPORT_SYMBOL_GPL(inet_hash); | |
453 | ||
454 | void inet_unhash(struct sock *sk) | |
455 | { | |
39d8cda7 | 456 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
c25eb3bf ED |
457 | spinlock_t *lock; |
458 | int done; | |
ab1e0a13 ACM |
459 | |
460 | if (sk_unhashed(sk)) | |
5caea4ea | 461 | return; |
ab1e0a13 | 462 | |
c25eb3bf ED |
463 | if (sk->sk_state == TCP_LISTEN) |
464 | lock = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)].lock; | |
465 | else | |
466 | lock = inet_ehash_lockp(hashinfo, sk->sk_hash); | |
5caea4ea | 467 | |
c25eb3bf | 468 | spin_lock_bh(lock); |
3b8ccd44 | 469 | done = __sk_nulls_del_node_init_rcu(sk); |
c25eb3bf ED |
470 | if (done) |
471 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); | |
920de804 | 472 | spin_unlock_bh(lock); |
ab1e0a13 ACM |
473 | } |
474 | EXPORT_SYMBOL_GPL(inet_unhash); | |
152da81d | 475 | |
5ee31fc1 | 476 | int __inet_hash_connect(struct inet_timewait_death_row *death_row, |
5d8c0aa9 | 477 | struct sock *sk, u32 port_offset, |
5ee31fc1 | 478 | int (*check_established)(struct inet_timewait_death_row *, |
b4d6444e | 479 | struct sock *, __u16, struct inet_timewait_sock **)) |
a7f5e7f1 ACM |
480 | { |
481 | struct inet_hashinfo *hinfo = death_row->hashinfo; | |
c720c7e8 | 482 | const unsigned short snum = inet_sk(sk)->inet_num; |
e905a9ed YH |
483 | struct inet_bind_hashbucket *head; |
484 | struct inet_bind_bucket *tb; | |
a7f5e7f1 | 485 | int ret; |
3b1e0a65 | 486 | struct net *net = sock_net(sk); |
a7f5e7f1 | 487 | |
e905a9ed | 488 | if (!snum) { |
227b60f5 | 489 | int i, remaining, low, high, port; |
a7f5e7f1 | 490 | static u32 hint; |
5d8c0aa9 | 491 | u32 offset = hint + port_offset; |
e905a9ed | 492 | struct inet_timewait_sock *tw = NULL; |
a7f5e7f1 | 493 | |
0bbf87d8 | 494 | inet_get_local_port_range(net, &low, &high); |
a25de534 | 495 | remaining = (high - low) + 1; |
227b60f5 | 496 | |
07f4c900 ED |
497 | /* By starting with offset being an even number, |
498 | * we tend to leave about 50% of ports for other uses, | |
499 | * like bind(0). | |
500 | */ | |
501 | offset &= ~1; | |
502 | ||
e905a9ed | 503 | local_bh_disable(); |
07f4c900 | 504 | for (i = 0; i < remaining; i++) { |
227b60f5 | 505 | port = low + (i + offset) % remaining; |
122ff243 | 506 | if (inet_is_local_reserved_port(net, port)) |
e3826f1e | 507 | continue; |
7f635ab7 PE |
508 | head = &hinfo->bhash[inet_bhashfn(net, port, |
509 | hinfo->bhash_size)]; | |
e905a9ed | 510 | spin_lock(&head->lock); |
a7f5e7f1 | 511 | |
e905a9ed YH |
512 | /* Does not bother with rcv_saddr checks, |
513 | * because the established check is already | |
514 | * unique enough. | |
515 | */ | |
b67bfe0d | 516 | inet_bind_bucket_for_each(tb, &head->chain) { |
09ad9bc7 OP |
517 | if (net_eq(ib_net(tb), net) && |
518 | tb->port == port) { | |
da5e3630 TH |
519 | if (tb->fastreuse >= 0 || |
520 | tb->fastreuseport >= 0) | |
e905a9ed | 521 | goto next_port; |
a9d8f911 | 522 | WARN_ON(hlist_empty(&tb->owners)); |
5ee31fc1 PE |
523 | if (!check_established(death_row, sk, |
524 | port, &tw)) | |
e905a9ed YH |
525 | goto ok; |
526 | goto next_port; | |
527 | } | |
528 | } | |
529 | ||
941b1d22 PE |
530 | tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, |
531 | net, head, port); | |
e905a9ed YH |
532 | if (!tb) { |
533 | spin_unlock(&head->lock); | |
534 | break; | |
535 | } | |
536 | tb->fastreuse = -1; | |
da5e3630 | 537 | tb->fastreuseport = -1; |
e905a9ed YH |
538 | goto ok; |
539 | ||
540 | next_port: | |
541 | spin_unlock(&head->lock); | |
542 | } | |
543 | local_bh_enable(); | |
544 | ||
545 | return -EADDRNOTAVAIL; | |
a7f5e7f1 ACM |
546 | |
547 | ok: | |
07f4c900 | 548 | hint += (i + 2) & ~1; |
a7f5e7f1 | 549 | |
e905a9ed YH |
550 | /* Head lock still held and bh's disabled */ |
551 | inet_bind_hash(sk, tb, port); | |
a7f5e7f1 | 552 | if (sk_unhashed(sk)) { |
c720c7e8 | 553 | inet_sk(sk)->inet_sport = htons(port); |
fc01538f | 554 | __inet_hash_nolisten(sk, (struct sock *)tw); |
e905a9ed | 555 | } |
3cdaedae | 556 | if (tw) |
fc01538f | 557 | inet_twsk_bind_unhash(tw, hinfo); |
e905a9ed | 558 | spin_unlock(&head->lock); |
a7f5e7f1 | 559 | |
dbe7faa4 ED |
560 | if (tw) |
561 | inet_twsk_deschedule_put(tw); | |
a7f5e7f1 ACM |
562 | |
563 | ret = 0; | |
564 | goto out; | |
e905a9ed | 565 | } |
a7f5e7f1 | 566 | |
7f635ab7 | 567 | head = &hinfo->bhash[inet_bhashfn(net, snum, hinfo->bhash_size)]; |
e905a9ed | 568 | tb = inet_csk(sk)->icsk_bind_hash; |
a7f5e7f1 ACM |
569 | spin_lock_bh(&head->lock); |
570 | if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) { | |
b4d6444e | 571 | __inet_hash_nolisten(sk, NULL); |
a7f5e7f1 ACM |
572 | spin_unlock_bh(&head->lock); |
573 | return 0; | |
574 | } else { | |
575 | spin_unlock(&head->lock); | |
576 | /* No definite answer... Walk to established hash table */ | |
5ee31fc1 | 577 | ret = check_established(death_row, sk, snum, NULL); |
a7f5e7f1 ACM |
578 | out: |
579 | local_bh_enable(); | |
580 | return ret; | |
581 | } | |
582 | } | |
5ee31fc1 PE |
583 | |
584 | /* | |
585 | * Bind a port for a connect operation and hash it. | |
586 | */ | |
587 | int inet_hash_connect(struct inet_timewait_death_row *death_row, | |
588 | struct sock *sk) | |
589 | { | |
e2baad9e ED |
590 | u32 port_offset = 0; |
591 | ||
592 | if (!inet_sk(sk)->inet_num) | |
593 | port_offset = inet_sk_port_offset(sk); | |
594 | return __inet_hash_connect(death_row, sk, port_offset, | |
b4d6444e | 595 | __inet_check_established); |
5ee31fc1 | 596 | } |
a7f5e7f1 | 597 | EXPORT_SYMBOL_GPL(inet_hash_connect); |
5caea4ea ED |
598 | |
599 | void inet_hashinfo_init(struct inet_hashinfo *h) | |
600 | { | |
601 | int i; | |
602 | ||
c25eb3bf | 603 | for (i = 0; i < INET_LHTABLE_SIZE; i++) { |
5caea4ea | 604 | spin_lock_init(&h->listening_hash[i].lock); |
c25eb3bf ED |
605 | INIT_HLIST_NULLS_HEAD(&h->listening_hash[i].head, |
606 | i + LISTENING_NULLS_BASE); | |
607 | } | |
5caea4ea | 608 | } |
5caea4ea | 609 | EXPORT_SYMBOL_GPL(inet_hashinfo_init); |
095dc8e0 ED |
610 | |
611 | int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo) | |
612 | { | |
89e478a2 | 613 | unsigned int locksz = sizeof(spinlock_t); |
095dc8e0 ED |
614 | unsigned int i, nblocks = 1; |
615 | ||
89e478a2 | 616 | if (locksz != 0) { |
095dc8e0 | 617 | /* allocate 2 cache lines or at least one spinlock per cpu */ |
89e478a2 | 618 | nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U); |
095dc8e0 ED |
619 | nblocks = roundup_pow_of_two(nblocks * num_possible_cpus()); |
620 | ||
621 | /* no more locks than number of hash buckets */ | |
622 | nblocks = min(nblocks, hashinfo->ehash_mask + 1); | |
623 | ||
89e478a2 | 624 | hashinfo->ehash_locks = kmalloc_array(nblocks, locksz, |
095dc8e0 ED |
625 | GFP_KERNEL | __GFP_NOWARN); |
626 | if (!hashinfo->ehash_locks) | |
89e478a2 | 627 | hashinfo->ehash_locks = vmalloc(nblocks * locksz); |
095dc8e0 ED |
628 | |
629 | if (!hashinfo->ehash_locks) | |
630 | return -ENOMEM; | |
631 | ||
632 | for (i = 0; i < nblocks; i++) | |
633 | spin_lock_init(&hashinfo->ehash_locks[i]); | |
634 | } | |
635 | hashinfo->ehash_locks_mask = nblocks - 1; | |
636 | return 0; | |
637 | } | |
638 | EXPORT_SYMBOL_GPL(inet_ehash_locks_alloc); |