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> |
77d8bf9c | 21 | |
463c84b9 | 22 | #include <net/inet_connection_sock.h> |
77d8bf9c | 23 | #include <net/inet_hashtables.h> |
6e5714ea | 24 | #include <net/secure_seq.h> |
a7f5e7f1 | 25 | #include <net/ip.h> |
77d8bf9c | 26 | |
6eada011 ED |
27 | static u32 inet_ehashfn(const struct net *net, const __be32 laddr, |
28 | const __u16 lport, const __be32 faddr, | |
29 | const __be16 fport) | |
65cd8033 | 30 | { |
1bbdceef HFS |
31 | static u32 inet_ehash_secret __read_mostly; |
32 | ||
33 | net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret)); | |
34 | ||
65cd8033 HFS |
35 | return __inet_ehashfn(laddr, lport, faddr, fport, |
36 | inet_ehash_secret + net_hash_mix(net)); | |
37 | } | |
38 | ||
d1e559d0 ED |
39 | /* This function handles inet_sock, but also timewait and request sockets |
40 | * for IPv4/IPv6. | |
41 | */ | |
5b441f76 | 42 | u32 sk_ehashfn(const struct sock *sk) |
65cd8033 | 43 | { |
d1e559d0 ED |
44 | #if IS_ENABLED(CONFIG_IPV6) |
45 | if (sk->sk_family == AF_INET6 && | |
46 | !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) | |
47 | return inet6_ehashfn(sock_net(sk), | |
48 | &sk->sk_v6_rcv_saddr, sk->sk_num, | |
49 | &sk->sk_v6_daddr, sk->sk_dport); | |
50 | #endif | |
5b441f76 ED |
51 | return inet_ehashfn(sock_net(sk), |
52 | sk->sk_rcv_saddr, sk->sk_num, | |
53 | sk->sk_daddr, sk->sk_dport); | |
65cd8033 HFS |
54 | } |
55 | ||
77d8bf9c ACM |
56 | /* |
57 | * Allocate and initialize a new local port bind bucket. | |
58 | * The bindhash mutex for snum's hash chain must be held here. | |
59 | */ | |
e18b890b | 60 | struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep, |
941b1d22 | 61 | struct net *net, |
77d8bf9c ACM |
62 | struct inet_bind_hashbucket *head, |
63 | const unsigned short snum) | |
64 | { | |
54e6ecb2 | 65 | struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC); |
77d8bf9c | 66 | |
00db4124 | 67 | if (tb) { |
efd7ef1c | 68 | write_pnet(&tb->ib_net, net); |
77d8bf9c ACM |
69 | tb->port = snum; |
70 | tb->fastreuse = 0; | |
da5e3630 | 71 | tb->fastreuseport = 0; |
a9d8f911 | 72 | tb->num_owners = 0; |
77d8bf9c ACM |
73 | INIT_HLIST_HEAD(&tb->owners); |
74 | hlist_add_head(&tb->node, &head->chain); | |
75 | } | |
76 | return tb; | |
77 | } | |
78 | ||
77d8bf9c ACM |
79 | /* |
80 | * Caller must hold hashbucket lock for this tb with local BH disabled | |
81 | */ | |
e18b890b | 82 | void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb) |
77d8bf9c ACM |
83 | { |
84 | if (hlist_empty(&tb->owners)) { | |
85 | __hlist_del(&tb->node); | |
86 | kmem_cache_free(cachep, tb); | |
87 | } | |
88 | } | |
2d8c4ce5 ACM |
89 | |
90 | void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, | |
91 | const unsigned short snum) | |
92 | { | |
a9d8f911 EP |
93 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
94 | ||
24dd1fa1 | 95 | atomic_inc(&hashinfo->bsockets); |
a9d8f911 | 96 | |
c720c7e8 | 97 | inet_sk(sk)->inet_num = snum; |
2d8c4ce5 | 98 | sk_add_bind_node(sk, &tb->owners); |
a9d8f911 | 99 | tb->num_owners++; |
463c84b9 | 100 | inet_csk(sk)->icsk_bind_hash = tb; |
2d8c4ce5 ACM |
101 | } |
102 | ||
2d8c4ce5 ACM |
103 | /* |
104 | * Get rid of any references to a local port held by the given sock. | |
105 | */ | |
ab1e0a13 | 106 | static void __inet_put_port(struct sock *sk) |
2d8c4ce5 | 107 | { |
39d8cda7 | 108 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
c720c7e8 | 109 | const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num, |
7f635ab7 | 110 | hashinfo->bhash_size); |
2d8c4ce5 ACM |
111 | struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash]; |
112 | struct inet_bind_bucket *tb; | |
113 | ||
24dd1fa1 | 114 | atomic_dec(&hashinfo->bsockets); |
a9d8f911 | 115 | |
2d8c4ce5 | 116 | spin_lock(&head->lock); |
463c84b9 | 117 | tb = inet_csk(sk)->icsk_bind_hash; |
2d8c4ce5 | 118 | __sk_del_bind_node(sk); |
a9d8f911 | 119 | tb->num_owners--; |
463c84b9 | 120 | inet_csk(sk)->icsk_bind_hash = NULL; |
c720c7e8 | 121 | inet_sk(sk)->inet_num = 0; |
2d8c4ce5 ACM |
122 | inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb); |
123 | spin_unlock(&head->lock); | |
124 | } | |
125 | ||
ab1e0a13 | 126 | void inet_put_port(struct sock *sk) |
2d8c4ce5 ACM |
127 | { |
128 | local_bh_disable(); | |
ab1e0a13 | 129 | __inet_put_port(sk); |
2d8c4ce5 ACM |
130 | local_bh_enable(); |
131 | } | |
2d8c4ce5 | 132 | EXPORT_SYMBOL(inet_put_port); |
f3f05f70 | 133 | |
093d2823 | 134 | int __inet_inherit_port(struct sock *sk, struct sock *child) |
53083773 PE |
135 | { |
136 | struct inet_hashinfo *table = sk->sk_prot->h.hashinfo; | |
093d2823 BS |
137 | unsigned short port = inet_sk(child)->inet_num; |
138 | const int bhash = inet_bhashfn(sock_net(sk), port, | |
7f635ab7 | 139 | table->bhash_size); |
53083773 PE |
140 | struct inet_bind_hashbucket *head = &table->bhash[bhash]; |
141 | struct inet_bind_bucket *tb; | |
142 | ||
143 | spin_lock(&head->lock); | |
144 | tb = inet_csk(sk)->icsk_bind_hash; | |
093d2823 BS |
145 | if (tb->port != port) { |
146 | /* NOTE: using tproxy and redirecting skbs to a proxy | |
147 | * on a different listener port breaks the assumption | |
148 | * that the listener socket's icsk_bind_hash is the same | |
149 | * as that of the child socket. We have to look up or | |
150 | * create a new bind bucket for the child here. */ | |
b67bfe0d | 151 | inet_bind_bucket_for_each(tb, &head->chain) { |
093d2823 BS |
152 | if (net_eq(ib_net(tb), sock_net(sk)) && |
153 | tb->port == port) | |
154 | break; | |
155 | } | |
b67bfe0d | 156 | if (!tb) { |
093d2823 BS |
157 | tb = inet_bind_bucket_create(table->bind_bucket_cachep, |
158 | sock_net(sk), head, port); | |
159 | if (!tb) { | |
160 | spin_unlock(&head->lock); | |
161 | return -ENOMEM; | |
162 | } | |
163 | } | |
164 | } | |
b4ff3c90 | 165 | inet_bind_hash(child, tb, port); |
53083773 | 166 | spin_unlock(&head->lock); |
093d2823 BS |
167 | |
168 | return 0; | |
53083773 | 169 | } |
53083773 PE |
170 | EXPORT_SYMBOL_GPL(__inet_inherit_port); |
171 | ||
c25eb3bf ED |
172 | static inline int compute_score(struct sock *sk, struct net *net, |
173 | const unsigned short hnum, const __be32 daddr, | |
174 | const int dif) | |
175 | { | |
176 | int score = -1; | |
177 | struct inet_sock *inet = inet_sk(sk); | |
178 | ||
c720c7e8 | 179 | if (net_eq(sock_net(sk), net) && inet->inet_num == hnum && |
c25eb3bf | 180 | !ipv6_only_sock(sk)) { |
c720c7e8 | 181 | __be32 rcv_saddr = inet->inet_rcv_saddr; |
da5e3630 | 182 | score = sk->sk_family == PF_INET ? 2 : 1; |
c25eb3bf ED |
183 | if (rcv_saddr) { |
184 | if (rcv_saddr != daddr) | |
185 | return -1; | |
da5e3630 | 186 | score += 4; |
c25eb3bf ED |
187 | } |
188 | if (sk->sk_bound_dev_if) { | |
189 | if (sk->sk_bound_dev_if != dif) | |
190 | return -1; | |
da5e3630 | 191 | score += 4; |
c25eb3bf ED |
192 | } |
193 | } | |
194 | return score; | |
195 | } | |
196 | ||
33b62231 ACM |
197 | /* |
198 | * Don't inline this cruft. Here are some nice properties to exploit here. The | |
199 | * BSD API does not allow a listening sock to specify the remote port nor the | |
200 | * remote address for the connection. So always assume those are both | |
201 | * wildcarded during the search since they can never be otherwise. | |
202 | */ | |
e48c414e | 203 | |
c25eb3bf | 204 | |
c67499c0 PE |
205 | struct sock *__inet_lookup_listener(struct net *net, |
206 | struct inet_hashinfo *hashinfo, | |
da5e3630 | 207 | const __be32 saddr, __be16 sport, |
fb99c848 | 208 | const __be32 daddr, const unsigned short hnum, |
8f491069 | 209 | const int dif) |
99a92ff5 | 210 | { |
c25eb3bf ED |
211 | struct sock *sk, *result; |
212 | struct hlist_nulls_node *node; | |
213 | unsigned int hash = inet_lhashfn(net, hnum); | |
214 | struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash]; | |
da5e3630 TH |
215 | int score, hiscore, matches = 0, reuseport = 0; |
216 | u32 phash = 0; | |
99a92ff5 | 217 | |
c25eb3bf ED |
218 | rcu_read_lock(); |
219 | begin: | |
220 | result = NULL; | |
da5e3630 | 221 | hiscore = 0; |
c25eb3bf ED |
222 | sk_nulls_for_each_rcu(sk, node, &ilb->head) { |
223 | score = compute_score(sk, net, hnum, daddr, dif); | |
224 | if (score > hiscore) { | |
225 | result = sk; | |
226 | hiscore = score; | |
da5e3630 TH |
227 | reuseport = sk->sk_reuseport; |
228 | if (reuseport) { | |
229 | phash = inet_ehashfn(net, daddr, hnum, | |
230 | saddr, sport); | |
231 | matches = 1; | |
232 | } | |
233 | } else if (score == hiscore && reuseport) { | |
234 | matches++; | |
8fc54f68 | 235 | if (reciprocal_scale(phash, matches) == 0) |
da5e3630 TH |
236 | result = sk; |
237 | phash = next_pseudo_random32(phash); | |
c25eb3bf | 238 | } |
99a92ff5 | 239 | } |
c25eb3bf ED |
240 | /* |
241 | * if the nulls value we got at the end of this lookup is | |
242 | * not the expected one, we must restart lookup. | |
243 | * We probably met an item that was moved to another chain. | |
244 | */ | |
245 | if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE) | |
246 | goto begin; | |
247 | if (result) { | |
248 | if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt))) | |
249 | result = NULL; | |
250 | else if (unlikely(compute_score(result, net, hnum, daddr, | |
251 | dif) < hiscore)) { | |
252 | sock_put(result); | |
253 | goto begin; | |
254 | } | |
99a92ff5 | 255 | } |
c25eb3bf ED |
256 | rcu_read_unlock(); |
257 | return result; | |
99a92ff5 | 258 | } |
8f491069 | 259 | EXPORT_SYMBOL_GPL(__inet_lookup_listener); |
a7f5e7f1 | 260 | |
05dbc7b5 ED |
261 | /* All sockets share common refcount, but have different destructors */ |
262 | void sock_gen_put(struct sock *sk) | |
263 | { | |
264 | if (!atomic_dec_and_test(&sk->sk_refcnt)) | |
265 | return; | |
266 | ||
267 | if (sk->sk_state == TCP_TIME_WAIT) | |
268 | inet_twsk_free(inet_twsk(sk)); | |
41b822c5 ED |
269 | else if (sk->sk_state == TCP_NEW_SYN_RECV) |
270 | reqsk_free(inet_reqsk(sk)); | |
05dbc7b5 ED |
271 | else |
272 | sk_free(sk); | |
273 | } | |
274 | EXPORT_SYMBOL_GPL(sock_gen_put); | |
275 | ||
2c13270b ED |
276 | void sock_edemux(struct sk_buff *skb) |
277 | { | |
278 | sock_gen_put(skb->sk); | |
279 | } | |
280 | EXPORT_SYMBOL(sock_edemux); | |
281 | ||
5e73ea1a | 282 | struct sock *__inet_lookup_established(struct net *net, |
c67499c0 | 283 | struct inet_hashinfo *hashinfo, |
77a5ba55 PE |
284 | const __be32 saddr, const __be16 sport, |
285 | const __be32 daddr, const u16 hnum, | |
286 | const int dif) | |
287 | { | |
c7228317 | 288 | INET_ADDR_COOKIE(acookie, saddr, daddr); |
77a5ba55 PE |
289 | const __portpair ports = INET_COMBINED_PORTS(sport, hnum); |
290 | struct sock *sk; | |
3ab5aee7 | 291 | const struct hlist_nulls_node *node; |
77a5ba55 PE |
292 | /* Optimize here for direct hit, only listening connections can |
293 | * have wildcards anyways. | |
294 | */ | |
9f26b3ad | 295 | unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport); |
f373b53b | 296 | unsigned int slot = hash & hashinfo->ehash_mask; |
3ab5aee7 | 297 | struct inet_ehash_bucket *head = &hashinfo->ehash[slot]; |
77a5ba55 | 298 | |
3ab5aee7 ED |
299 | rcu_read_lock(); |
300 | begin: | |
301 | sk_nulls_for_each_rcu(sk, node, &head->chain) { | |
ce43b03e ED |
302 | if (sk->sk_hash != hash) |
303 | continue; | |
304 | if (likely(INET_MATCH(sk, net, acookie, | |
305 | saddr, daddr, ports, dif))) { | |
3ab5aee7 | 306 | if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) |
05dbc7b5 | 307 | goto out; |
ce43b03e ED |
308 | if (unlikely(!INET_MATCH(sk, net, acookie, |
309 | saddr, daddr, ports, dif))) { | |
05dbc7b5 | 310 | sock_gen_put(sk); |
3ab5aee7 ED |
311 | goto begin; |
312 | } | |
05dbc7b5 | 313 | goto found; |
3ab5aee7 | 314 | } |
77a5ba55 | 315 | } |
3ab5aee7 ED |
316 | /* |
317 | * if the nulls value we got at the end of this lookup is | |
318 | * not the expected one, we must restart lookup. | |
319 | * We probably met an item that was moved to another chain. | |
320 | */ | |
321 | if (get_nulls_value(node) != slot) | |
322 | goto begin; | |
77a5ba55 | 323 | out: |
05dbc7b5 ED |
324 | sk = NULL; |
325 | found: | |
3ab5aee7 | 326 | rcu_read_unlock(); |
77a5ba55 | 327 | return sk; |
77a5ba55 PE |
328 | } |
329 | EXPORT_SYMBOL_GPL(__inet_lookup_established); | |
330 | ||
a7f5e7f1 ACM |
331 | /* called with local bh disabled */ |
332 | static int __inet_check_established(struct inet_timewait_death_row *death_row, | |
333 | struct sock *sk, __u16 lport, | |
334 | struct inet_timewait_sock **twp) | |
335 | { | |
336 | struct inet_hashinfo *hinfo = death_row->hashinfo; | |
337 | struct inet_sock *inet = inet_sk(sk); | |
c720c7e8 ED |
338 | __be32 daddr = inet->inet_rcv_saddr; |
339 | __be32 saddr = inet->inet_daddr; | |
a7f5e7f1 | 340 | int dif = sk->sk_bound_dev_if; |
c7228317 | 341 | INET_ADDR_COOKIE(acookie, saddr, daddr); |
c720c7e8 | 342 | const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport); |
9f26b3ad | 343 | struct net *net = sock_net(sk); |
c720c7e8 ED |
344 | unsigned int hash = inet_ehashfn(net, daddr, lport, |
345 | saddr, inet->inet_dport); | |
a7f5e7f1 | 346 | struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash); |
9db66bdc | 347 | spinlock_t *lock = inet_ehash_lockp(hinfo, hash); |
a7f5e7f1 | 348 | struct sock *sk2; |
3ab5aee7 | 349 | const struct hlist_nulls_node *node; |
05dbc7b5 | 350 | struct inet_timewait_sock *tw = NULL; |
13475a30 | 351 | int twrefcnt = 0; |
a7f5e7f1 | 352 | |
9db66bdc | 353 | spin_lock(lock); |
a7f5e7f1 | 354 | |
3ab5aee7 | 355 | sk_nulls_for_each(sk2, node, &head->chain) { |
ce43b03e ED |
356 | if (sk2->sk_hash != hash) |
357 | continue; | |
05dbc7b5 | 358 | |
ce43b03e | 359 | if (likely(INET_MATCH(sk2, net, acookie, |
05dbc7b5 ED |
360 | saddr, daddr, ports, dif))) { |
361 | if (sk2->sk_state == TCP_TIME_WAIT) { | |
362 | tw = inet_twsk(sk2); | |
363 | if (twsk_unique(sk, sk2, twp)) | |
364 | break; | |
365 | } | |
a7f5e7f1 | 366 | goto not_unique; |
05dbc7b5 | 367 | } |
a7f5e7f1 ACM |
368 | } |
369 | ||
a7f5e7f1 | 370 | /* Must record num and sport now. Otherwise we will see |
05dbc7b5 ED |
371 | * in hash table socket with a funny identity. |
372 | */ | |
c720c7e8 ED |
373 | inet->inet_num = lport; |
374 | inet->inet_sport = htons(lport); | |
a7f5e7f1 | 375 | sk->sk_hash = hash; |
547b792c | 376 | WARN_ON(!sk_unhashed(sk)); |
3ab5aee7 | 377 | __sk_nulls_add_node_rcu(sk, &head->chain); |
13475a30 ED |
378 | if (tw) { |
379 | twrefcnt = inet_twsk_unhash(tw); | |
380 | NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED); | |
381 | } | |
9db66bdc | 382 | spin_unlock(lock); |
13475a30 ED |
383 | if (twrefcnt) |
384 | inet_twsk_put(tw); | |
c29a0bc4 | 385 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
a7f5e7f1 ACM |
386 | |
387 | if (twp) { | |
388 | *twp = tw; | |
a7f5e7f1 ACM |
389 | } else if (tw) { |
390 | /* Silly. Should hash-dance instead... */ | |
789f558c | 391 | inet_twsk_deschedule(tw); |
a7f5e7f1 ACM |
392 | |
393 | inet_twsk_put(tw); | |
394 | } | |
a7f5e7f1 ACM |
395 | return 0; |
396 | ||
397 | not_unique: | |
9db66bdc | 398 | spin_unlock(lock); |
a7f5e7f1 ACM |
399 | return -EADDRNOTAVAIL; |
400 | } | |
401 | ||
402 | static inline u32 inet_sk_port_offset(const struct sock *sk) | |
403 | { | |
404 | const struct inet_sock *inet = inet_sk(sk); | |
c720c7e8 ED |
405 | return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr, |
406 | inet->inet_daddr, | |
407 | inet->inet_dport); | |
a7f5e7f1 ACM |
408 | } |
409 | ||
9327f705 | 410 | int __inet_hash_nolisten(struct sock *sk, struct inet_timewait_sock *tw) |
152da81d | 411 | { |
39d8cda7 | 412 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
3ab5aee7 | 413 | struct hlist_nulls_head *list; |
152da81d | 414 | struct inet_ehash_bucket *head; |
5b441f76 | 415 | spinlock_t *lock; |
9327f705 | 416 | int twrefcnt = 0; |
152da81d | 417 | |
547b792c | 418 | WARN_ON(!sk_unhashed(sk)); |
152da81d | 419 | |
5b441f76 | 420 | sk->sk_hash = sk_ehashfn(sk); |
152da81d PE |
421 | head = inet_ehash_bucket(hashinfo, sk->sk_hash); |
422 | list = &head->chain; | |
423 | lock = inet_ehash_lockp(hashinfo, sk->sk_hash); | |
424 | ||
9db66bdc | 425 | spin_lock(lock); |
3ab5aee7 | 426 | __sk_nulls_add_node_rcu(sk, list); |
9327f705 ED |
427 | if (tw) { |
428 | WARN_ON(sk->sk_hash != tw->tw_hash); | |
429 | twrefcnt = inet_twsk_unhash(tw); | |
430 | } | |
9db66bdc | 431 | spin_unlock(lock); |
c29a0bc4 | 432 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
9327f705 | 433 | return twrefcnt; |
152da81d PE |
434 | } |
435 | EXPORT_SYMBOL_GPL(__inet_hash_nolisten); | |
436 | ||
77a6a471 | 437 | int __inet_hash(struct sock *sk, struct inet_timewait_sock *tw) |
152da81d | 438 | { |
39d8cda7 | 439 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
5caea4ea | 440 | struct inet_listen_hashbucket *ilb; |
152da81d | 441 | |
77a6a471 ED |
442 | if (sk->sk_state != TCP_LISTEN) |
443 | return __inet_hash_nolisten(sk, tw); | |
152da81d | 444 | |
547b792c | 445 | WARN_ON(!sk_unhashed(sk)); |
5caea4ea | 446 | ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)]; |
152da81d | 447 | |
5caea4ea | 448 | spin_lock(&ilb->lock); |
c25eb3bf | 449 | __sk_nulls_add_node_rcu(sk, &ilb->head); |
c29a0bc4 | 450 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
5caea4ea | 451 | spin_unlock(&ilb->lock); |
77a6a471 | 452 | return 0; |
152da81d | 453 | } |
77a6a471 | 454 | EXPORT_SYMBOL(__inet_hash); |
ab1e0a13 ACM |
455 | |
456 | void inet_hash(struct sock *sk) | |
457 | { | |
458 | if (sk->sk_state != TCP_CLOSE) { | |
459 | local_bh_disable(); | |
77a6a471 | 460 | __inet_hash(sk, NULL); |
ab1e0a13 ACM |
461 | local_bh_enable(); |
462 | } | |
463 | } | |
464 | EXPORT_SYMBOL_GPL(inet_hash); | |
465 | ||
466 | void inet_unhash(struct sock *sk) | |
467 | { | |
39d8cda7 | 468 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
c25eb3bf ED |
469 | spinlock_t *lock; |
470 | int done; | |
ab1e0a13 ACM |
471 | |
472 | if (sk_unhashed(sk)) | |
5caea4ea | 473 | return; |
ab1e0a13 | 474 | |
c25eb3bf ED |
475 | if (sk->sk_state == TCP_LISTEN) |
476 | lock = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)].lock; | |
477 | else | |
478 | lock = inet_ehash_lockp(hashinfo, sk->sk_hash); | |
5caea4ea | 479 | |
c25eb3bf | 480 | spin_lock_bh(lock); |
3b8ccd44 | 481 | done = __sk_nulls_del_node_init_rcu(sk); |
c25eb3bf ED |
482 | if (done) |
483 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); | |
920de804 | 484 | spin_unlock_bh(lock); |
ab1e0a13 ACM |
485 | } |
486 | EXPORT_SYMBOL_GPL(inet_unhash); | |
152da81d | 487 | |
5ee31fc1 | 488 | int __inet_hash_connect(struct inet_timewait_death_row *death_row, |
5d8c0aa9 | 489 | struct sock *sk, u32 port_offset, |
5ee31fc1 | 490 | int (*check_established)(struct inet_timewait_death_row *, |
b4d6444e | 491 | struct sock *, __u16, struct inet_timewait_sock **)) |
a7f5e7f1 ACM |
492 | { |
493 | struct inet_hashinfo *hinfo = death_row->hashinfo; | |
c720c7e8 | 494 | const unsigned short snum = inet_sk(sk)->inet_num; |
e905a9ed YH |
495 | struct inet_bind_hashbucket *head; |
496 | struct inet_bind_bucket *tb; | |
a7f5e7f1 | 497 | int ret; |
3b1e0a65 | 498 | struct net *net = sock_net(sk); |
9327f705 | 499 | int twrefcnt = 1; |
a7f5e7f1 | 500 | |
e905a9ed | 501 | if (!snum) { |
227b60f5 | 502 | int i, remaining, low, high, port; |
a7f5e7f1 | 503 | static u32 hint; |
5d8c0aa9 | 504 | u32 offset = hint + port_offset; |
e905a9ed | 505 | struct inet_timewait_sock *tw = NULL; |
a7f5e7f1 | 506 | |
0bbf87d8 | 507 | inet_get_local_port_range(net, &low, &high); |
a25de534 | 508 | remaining = (high - low) + 1; |
227b60f5 | 509 | |
e905a9ed | 510 | local_bh_disable(); |
227b60f5 SH |
511 | for (i = 1; i <= remaining; i++) { |
512 | port = low + (i + offset) % remaining; | |
122ff243 | 513 | if (inet_is_local_reserved_port(net, port)) |
e3826f1e | 514 | continue; |
7f635ab7 PE |
515 | head = &hinfo->bhash[inet_bhashfn(net, port, |
516 | hinfo->bhash_size)]; | |
e905a9ed | 517 | spin_lock(&head->lock); |
a7f5e7f1 | 518 | |
e905a9ed YH |
519 | /* Does not bother with rcv_saddr checks, |
520 | * because the established check is already | |
521 | * unique enough. | |
522 | */ | |
b67bfe0d | 523 | inet_bind_bucket_for_each(tb, &head->chain) { |
09ad9bc7 OP |
524 | if (net_eq(ib_net(tb), net) && |
525 | tb->port == port) { | |
da5e3630 TH |
526 | if (tb->fastreuse >= 0 || |
527 | tb->fastreuseport >= 0) | |
e905a9ed | 528 | goto next_port; |
a9d8f911 | 529 | WARN_ON(hlist_empty(&tb->owners)); |
5ee31fc1 PE |
530 | if (!check_established(death_row, sk, |
531 | port, &tw)) | |
e905a9ed YH |
532 | goto ok; |
533 | goto next_port; | |
534 | } | |
535 | } | |
536 | ||
941b1d22 PE |
537 | tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, |
538 | net, head, port); | |
e905a9ed YH |
539 | if (!tb) { |
540 | spin_unlock(&head->lock); | |
541 | break; | |
542 | } | |
543 | tb->fastreuse = -1; | |
da5e3630 | 544 | tb->fastreuseport = -1; |
e905a9ed YH |
545 | goto ok; |
546 | ||
547 | next_port: | |
548 | spin_unlock(&head->lock); | |
549 | } | |
550 | local_bh_enable(); | |
551 | ||
552 | return -EADDRNOTAVAIL; | |
a7f5e7f1 ACM |
553 | |
554 | ok: | |
555 | hint += i; | |
556 | ||
e905a9ed YH |
557 | /* Head lock still held and bh's disabled */ |
558 | inet_bind_hash(sk, tb, port); | |
a7f5e7f1 | 559 | if (sk_unhashed(sk)) { |
c720c7e8 | 560 | inet_sk(sk)->inet_sport = htons(port); |
b4d6444e | 561 | twrefcnt += __inet_hash_nolisten(sk, tw); |
e905a9ed | 562 | } |
3cdaedae ED |
563 | if (tw) |
564 | twrefcnt += inet_twsk_bind_unhash(tw, hinfo); | |
e905a9ed | 565 | spin_unlock(&head->lock); |
a7f5e7f1 | 566 | |
e905a9ed | 567 | if (tw) { |
789f558c | 568 | inet_twsk_deschedule(tw); |
9327f705 ED |
569 | while (twrefcnt) { |
570 | twrefcnt--; | |
571 | inet_twsk_put(tw); | |
572 | } | |
e905a9ed | 573 | } |
a7f5e7f1 ACM |
574 | |
575 | ret = 0; | |
576 | goto out; | |
e905a9ed | 577 | } |
a7f5e7f1 | 578 | |
7f635ab7 | 579 | head = &hinfo->bhash[inet_bhashfn(net, snum, hinfo->bhash_size)]; |
e905a9ed | 580 | tb = inet_csk(sk)->icsk_bind_hash; |
a7f5e7f1 ACM |
581 | spin_lock_bh(&head->lock); |
582 | if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) { | |
b4d6444e | 583 | __inet_hash_nolisten(sk, NULL); |
a7f5e7f1 ACM |
584 | spin_unlock_bh(&head->lock); |
585 | return 0; | |
586 | } else { | |
587 | spin_unlock(&head->lock); | |
588 | /* No definite answer... Walk to established hash table */ | |
5ee31fc1 | 589 | ret = check_established(death_row, sk, snum, NULL); |
a7f5e7f1 ACM |
590 | out: |
591 | local_bh_enable(); | |
592 | return ret; | |
593 | } | |
594 | } | |
5ee31fc1 PE |
595 | |
596 | /* | |
597 | * Bind a port for a connect operation and hash it. | |
598 | */ | |
599 | int inet_hash_connect(struct inet_timewait_death_row *death_row, | |
600 | struct sock *sk) | |
601 | { | |
5d8c0aa9 | 602 | return __inet_hash_connect(death_row, sk, inet_sk_port_offset(sk), |
b4d6444e | 603 | __inet_check_established); |
5ee31fc1 | 604 | } |
a7f5e7f1 | 605 | EXPORT_SYMBOL_GPL(inet_hash_connect); |
5caea4ea ED |
606 | |
607 | void inet_hashinfo_init(struct inet_hashinfo *h) | |
608 | { | |
609 | int i; | |
610 | ||
24dd1fa1 | 611 | atomic_set(&h->bsockets, 0); |
c25eb3bf | 612 | for (i = 0; i < INET_LHTABLE_SIZE; i++) { |
5caea4ea | 613 | spin_lock_init(&h->listening_hash[i].lock); |
c25eb3bf ED |
614 | INIT_HLIST_NULLS_HEAD(&h->listening_hash[i].head, |
615 | i + LISTENING_NULLS_BASE); | |
616 | } | |
5caea4ea | 617 | } |
5caea4ea | 618 | EXPORT_SYMBOL_GPL(inet_hashinfo_init); |