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> |
61b7c691 | 22 | #include <linux/bootmem.h> |
77d8bf9c | 23 | |
c125e80b | 24 | #include <net/addrconf.h> |
463c84b9 | 25 | #include <net/inet_connection_sock.h> |
77d8bf9c | 26 | #include <net/inet_hashtables.h> |
6e5714ea | 27 | #include <net/secure_seq.h> |
a7f5e7f1 | 28 | #include <net/ip.h> |
a04a480d | 29 | #include <net/tcp.h> |
c125e80b | 30 | #include <net/sock_reuseport.h> |
77d8bf9c | 31 | |
6eada011 ED |
32 | static u32 inet_ehashfn(const struct net *net, const __be32 laddr, |
33 | const __u16 lport, const __be32 faddr, | |
34 | const __be16 fport) | |
65cd8033 | 35 | { |
1bbdceef HFS |
36 | static u32 inet_ehash_secret __read_mostly; |
37 | ||
38 | net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret)); | |
39 | ||
65cd8033 HFS |
40 | return __inet_ehashfn(laddr, lport, faddr, fport, |
41 | inet_ehash_secret + net_hash_mix(net)); | |
42 | } | |
43 | ||
d1e559d0 ED |
44 | /* This function handles inet_sock, but also timewait and request sockets |
45 | * for IPv4/IPv6. | |
46 | */ | |
784c372a | 47 | static u32 sk_ehashfn(const struct sock *sk) |
65cd8033 | 48 | { |
d1e559d0 ED |
49 | #if IS_ENABLED(CONFIG_IPV6) |
50 | if (sk->sk_family == AF_INET6 && | |
51 | !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) | |
52 | return inet6_ehashfn(sock_net(sk), | |
53 | &sk->sk_v6_rcv_saddr, sk->sk_num, | |
54 | &sk->sk_v6_daddr, sk->sk_dport); | |
55 | #endif | |
5b441f76 ED |
56 | return inet_ehashfn(sock_net(sk), |
57 | sk->sk_rcv_saddr, sk->sk_num, | |
58 | sk->sk_daddr, sk->sk_dport); | |
65cd8033 HFS |
59 | } |
60 | ||
77d8bf9c ACM |
61 | /* |
62 | * Allocate and initialize a new local port bind bucket. | |
63 | * The bindhash mutex for snum's hash chain must be held here. | |
64 | */ | |
e18b890b | 65 | struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep, |
941b1d22 | 66 | struct net *net, |
77d8bf9c ACM |
67 | struct inet_bind_hashbucket *head, |
68 | const unsigned short snum) | |
69 | { | |
54e6ecb2 | 70 | struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC); |
77d8bf9c | 71 | |
00db4124 | 72 | if (tb) { |
efd7ef1c | 73 | write_pnet(&tb->ib_net, net); |
77d8bf9c ACM |
74 | tb->port = snum; |
75 | tb->fastreuse = 0; | |
da5e3630 | 76 | tb->fastreuseport = 0; |
77d8bf9c ACM |
77 | INIT_HLIST_HEAD(&tb->owners); |
78 | hlist_add_head(&tb->node, &head->chain); | |
79 | } | |
80 | return tb; | |
81 | } | |
82 | ||
77d8bf9c ACM |
83 | /* |
84 | * Caller must hold hashbucket lock for this tb with local BH disabled | |
85 | */ | |
e18b890b | 86 | void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb) |
77d8bf9c ACM |
87 | { |
88 | if (hlist_empty(&tb->owners)) { | |
89 | __hlist_del(&tb->node); | |
90 | kmem_cache_free(cachep, tb); | |
91 | } | |
92 | } | |
2d8c4ce5 ACM |
93 | |
94 | void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, | |
95 | const unsigned short snum) | |
96 | { | |
c720c7e8 | 97 | inet_sk(sk)->inet_num = snum; |
2d8c4ce5 | 98 | sk_add_bind_node(sk, &tb->owners); |
463c84b9 | 99 | inet_csk(sk)->icsk_bind_hash = tb; |
2d8c4ce5 ACM |
100 | } |
101 | ||
2d8c4ce5 ACM |
102 | /* |
103 | * Get rid of any references to a local port held by the given sock. | |
104 | */ | |
ab1e0a13 | 105 | static void __inet_put_port(struct sock *sk) |
2d8c4ce5 | 106 | { |
39d8cda7 | 107 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
c720c7e8 | 108 | const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num, |
7f635ab7 | 109 | hashinfo->bhash_size); |
2d8c4ce5 ACM |
110 | struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash]; |
111 | struct inet_bind_bucket *tb; | |
112 | ||
113 | spin_lock(&head->lock); | |
463c84b9 | 114 | tb = inet_csk(sk)->icsk_bind_hash; |
2d8c4ce5 | 115 | __sk_del_bind_node(sk); |
463c84b9 | 116 | inet_csk(sk)->icsk_bind_hash = NULL; |
c720c7e8 | 117 | inet_sk(sk)->inet_num = 0; |
2d8c4ce5 ACM |
118 | inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb); |
119 | spin_unlock(&head->lock); | |
120 | } | |
121 | ||
ab1e0a13 | 122 | void inet_put_port(struct sock *sk) |
2d8c4ce5 ACM |
123 | { |
124 | local_bh_disable(); | |
ab1e0a13 | 125 | __inet_put_port(sk); |
2d8c4ce5 ACM |
126 | local_bh_enable(); |
127 | } | |
2d8c4ce5 | 128 | EXPORT_SYMBOL(inet_put_port); |
f3f05f70 | 129 | |
1ce31c9e | 130 | int __inet_inherit_port(const struct sock *sk, struct sock *child) |
53083773 PE |
131 | { |
132 | struct inet_hashinfo *table = sk->sk_prot->h.hashinfo; | |
093d2823 BS |
133 | unsigned short port = inet_sk(child)->inet_num; |
134 | const int bhash = inet_bhashfn(sock_net(sk), port, | |
7f635ab7 | 135 | table->bhash_size); |
53083773 PE |
136 | struct inet_bind_hashbucket *head = &table->bhash[bhash]; |
137 | struct inet_bind_bucket *tb; | |
138 | ||
139 | spin_lock(&head->lock); | |
140 | tb = inet_csk(sk)->icsk_bind_hash; | |
c2f34a65 ED |
141 | if (unlikely(!tb)) { |
142 | spin_unlock(&head->lock); | |
143 | return -ENOENT; | |
144 | } | |
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 | ||
61b7c691 MKL |
172 | static struct inet_listen_hashbucket * |
173 | inet_lhash2_bucket_sk(struct inet_hashinfo *h, struct sock *sk) | |
174 | { | |
175 | u32 hash; | |
176 | ||
177 | #if IS_ENABLED(CONFIG_IPV6) | |
178 | if (sk->sk_family == AF_INET6) | |
179 | hash = ipv6_portaddr_hash(sock_net(sk), | |
180 | &sk->sk_v6_rcv_saddr, | |
181 | inet_sk(sk)->inet_num); | |
182 | else | |
183 | #endif | |
184 | hash = ipv4_portaddr_hash(sock_net(sk), | |
185 | inet_sk(sk)->inet_rcv_saddr, | |
186 | inet_sk(sk)->inet_num); | |
187 | return inet_lhash2_bucket(h, hash); | |
188 | } | |
189 | ||
190 | static void inet_hash2(struct inet_hashinfo *h, struct sock *sk) | |
191 | { | |
192 | struct inet_listen_hashbucket *ilb2; | |
193 | ||
194 | if (!h->lhash2) | |
195 | return; | |
196 | ||
197 | ilb2 = inet_lhash2_bucket_sk(h, sk); | |
198 | ||
199 | spin_lock(&ilb2->lock); | |
200 | if (sk->sk_reuseport && sk->sk_family == AF_INET6) | |
201 | hlist_add_tail_rcu(&inet_csk(sk)->icsk_listen_portaddr_node, | |
202 | &ilb2->head); | |
203 | else | |
204 | hlist_add_head_rcu(&inet_csk(sk)->icsk_listen_portaddr_node, | |
205 | &ilb2->head); | |
206 | ilb2->count++; | |
207 | spin_unlock(&ilb2->lock); | |
208 | } | |
209 | ||
210 | static void inet_unhash2(struct inet_hashinfo *h, struct sock *sk) | |
211 | { | |
212 | struct inet_listen_hashbucket *ilb2; | |
213 | ||
214 | if (!h->lhash2 || | |
215 | WARN_ON_ONCE(hlist_unhashed(&inet_csk(sk)->icsk_listen_portaddr_node))) | |
216 | return; | |
217 | ||
218 | ilb2 = inet_lhash2_bucket_sk(h, sk); | |
219 | ||
220 | spin_lock(&ilb2->lock); | |
221 | hlist_del_init_rcu(&inet_csk(sk)->icsk_listen_portaddr_node); | |
222 | ilb2->count--; | |
223 | spin_unlock(&ilb2->lock); | |
224 | } | |
225 | ||
c25eb3bf ED |
226 | static inline int compute_score(struct sock *sk, struct net *net, |
227 | const unsigned short hnum, const __be32 daddr, | |
3fa6f616 | 228 | const int dif, const int sdif, bool exact_dif) |
c25eb3bf ED |
229 | { |
230 | int score = -1; | |
231 | struct inet_sock *inet = inet_sk(sk); | |
232 | ||
c720c7e8 | 233 | if (net_eq(sock_net(sk), net) && inet->inet_num == hnum && |
c25eb3bf | 234 | !ipv6_only_sock(sk)) { |
c720c7e8 | 235 | __be32 rcv_saddr = inet->inet_rcv_saddr; |
da5e3630 | 236 | score = sk->sk_family == PF_INET ? 2 : 1; |
c25eb3bf ED |
237 | if (rcv_saddr) { |
238 | if (rcv_saddr != daddr) | |
239 | return -1; | |
da5e3630 | 240 | score += 4; |
c25eb3bf | 241 | } |
a04a480d | 242 | if (sk->sk_bound_dev_if || exact_dif) { |
3fa6f616 DA |
243 | bool dev_match = (sk->sk_bound_dev_if == dif || |
244 | sk->sk_bound_dev_if == sdif); | |
245 | ||
246 | if (exact_dif && !dev_match) | |
c25eb3bf | 247 | return -1; |
3fa6f616 DA |
248 | if (sk->sk_bound_dev_if && dev_match) |
249 | score += 4; | |
c25eb3bf | 250 | } |
70da268b ED |
251 | if (sk->sk_incoming_cpu == raw_smp_processor_id()) |
252 | score++; | |
c25eb3bf ED |
253 | } |
254 | return score; | |
255 | } | |
256 | ||
33b62231 | 257 | /* |
3b24d854 ED |
258 | * Here are some nice properties to exploit here. The BSD API |
259 | * does not allow a listening sock to specify the remote port nor the | |
33b62231 ACM |
260 | * remote address for the connection. So always assume those are both |
261 | * wildcarded during the search since they can never be otherwise. | |
262 | */ | |
e48c414e | 263 | |
3b24d854 | 264 | /* called with rcu_read_lock() : No refcount taken on the socket */ |
61b7c691 MKL |
265 | static struct sock *inet_lhash2_lookup(struct net *net, |
266 | struct inet_listen_hashbucket *ilb2, | |
267 | struct sk_buff *skb, int doff, | |
268 | const __be32 saddr, __be16 sport, | |
269 | const __be32 daddr, const unsigned short hnum, | |
270 | const int dif, const int sdif) | |
271 | { | |
272 | bool exact_dif = inet_exact_dif_match(net, skb); | |
273 | struct inet_connection_sock *icsk; | |
274 | struct sock *sk, *result = NULL; | |
275 | int score, hiscore = 0; | |
276 | u32 phash = 0; | |
277 | ||
278 | inet_lhash2_for_each_icsk_rcu(icsk, &ilb2->head) { | |
279 | sk = (struct sock *)icsk; | |
280 | score = compute_score(sk, net, hnum, daddr, | |
281 | dif, sdif, exact_dif); | |
282 | if (score > hiscore) { | |
283 | if (sk->sk_reuseport) { | |
284 | phash = inet_ehashfn(net, daddr, hnum, | |
285 | saddr, sport); | |
286 | result = reuseport_select_sock(sk, phash, | |
287 | skb, doff); | |
288 | if (result) | |
289 | return result; | |
290 | } | |
291 | result = sk; | |
292 | hiscore = score; | |
293 | } | |
294 | } | |
295 | ||
296 | return result; | |
297 | } | |
298 | ||
c67499c0 PE |
299 | struct sock *__inet_lookup_listener(struct net *net, |
300 | struct inet_hashinfo *hashinfo, | |
a583636a | 301 | struct sk_buff *skb, int doff, |
da5e3630 | 302 | const __be32 saddr, __be16 sport, |
fb99c848 | 303 | const __be32 daddr, const unsigned short hnum, |
3fa6f616 | 304 | const int dif, const int sdif) |
99a92ff5 | 305 | { |
c25eb3bf ED |
306 | unsigned int hash = inet_lhashfn(net, hnum); |
307 | struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash]; | |
a04a480d | 308 | bool exact_dif = inet_exact_dif_match(net, skb); |
61b7c691 | 309 | struct inet_listen_hashbucket *ilb2; |
3b24d854 | 310 | struct sock *sk, *result = NULL; |
e94a62f5 | 311 | int score, hiscore = 0; |
61b7c691 | 312 | unsigned int hash2; |
da5e3630 | 313 | u32 phash = 0; |
99a92ff5 | 314 | |
61b7c691 MKL |
315 | if (ilb->count <= 10 || !hashinfo->lhash2) |
316 | goto port_lookup; | |
317 | ||
318 | /* Too many sk in the ilb bucket (which is hashed by port alone). | |
319 | * Try lhash2 (which is hashed by port and addr) instead. | |
320 | */ | |
321 | ||
322 | hash2 = ipv4_portaddr_hash(net, daddr, hnum); | |
323 | ilb2 = inet_lhash2_bucket(hashinfo, hash2); | |
324 | if (ilb2->count > ilb->count) | |
325 | goto port_lookup; | |
326 | ||
327 | result = inet_lhash2_lookup(net, ilb2, skb, doff, | |
328 | saddr, sport, daddr, hnum, | |
329 | dif, sdif); | |
330 | if (result) | |
331 | return result; | |
332 | ||
333 | /* Lookup lhash2 with INADDR_ANY */ | |
334 | ||
335 | hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); | |
336 | ilb2 = inet_lhash2_bucket(hashinfo, hash2); | |
337 | if (ilb2->count > ilb->count) | |
338 | goto port_lookup; | |
339 | ||
340 | return inet_lhash2_lookup(net, ilb2, skb, doff, | |
341 | saddr, sport, daddr, hnum, | |
342 | dif, sdif); | |
343 | ||
344 | port_lookup: | |
3b24d854 | 345 | sk_for_each_rcu(sk, &ilb->head) { |
3fa6f616 DA |
346 | score = compute_score(sk, net, hnum, daddr, |
347 | dif, sdif, exact_dif); | |
c25eb3bf | 348 | if (score > hiscore) { |
e94a62f5 | 349 | if (sk->sk_reuseport) { |
da5e3630 TH |
350 | phash = inet_ehashfn(net, daddr, hnum, |
351 | saddr, sport); | |
3b24d854 ED |
352 | result = reuseport_select_sock(sk, phash, |
353 | skb, doff); | |
354 | if (result) | |
355 | return result; | |
da5e3630 | 356 | } |
3b24d854 ED |
357 | result = sk; |
358 | hiscore = score; | |
c25eb3bf | 359 | } |
99a92ff5 | 360 | } |
c25eb3bf | 361 | return result; |
99a92ff5 | 362 | } |
8f491069 | 363 | EXPORT_SYMBOL_GPL(__inet_lookup_listener); |
a7f5e7f1 | 364 | |
05dbc7b5 ED |
365 | /* All sockets share common refcount, but have different destructors */ |
366 | void sock_gen_put(struct sock *sk) | |
367 | { | |
41c6d650 | 368 | if (!refcount_dec_and_test(&sk->sk_refcnt)) |
05dbc7b5 ED |
369 | return; |
370 | ||
371 | if (sk->sk_state == TCP_TIME_WAIT) | |
372 | inet_twsk_free(inet_twsk(sk)); | |
41b822c5 ED |
373 | else if (sk->sk_state == TCP_NEW_SYN_RECV) |
374 | reqsk_free(inet_reqsk(sk)); | |
05dbc7b5 ED |
375 | else |
376 | sk_free(sk); | |
377 | } | |
378 | EXPORT_SYMBOL_GPL(sock_gen_put); | |
379 | ||
2c13270b ED |
380 | void sock_edemux(struct sk_buff *skb) |
381 | { | |
382 | sock_gen_put(skb->sk); | |
383 | } | |
384 | EXPORT_SYMBOL(sock_edemux); | |
385 | ||
5e73ea1a | 386 | struct sock *__inet_lookup_established(struct net *net, |
c67499c0 | 387 | struct inet_hashinfo *hashinfo, |
77a5ba55 PE |
388 | const __be32 saddr, const __be16 sport, |
389 | const __be32 daddr, const u16 hnum, | |
3fa6f616 | 390 | const int dif, const int sdif) |
77a5ba55 | 391 | { |
c7228317 | 392 | INET_ADDR_COOKIE(acookie, saddr, daddr); |
77a5ba55 PE |
393 | const __portpair ports = INET_COMBINED_PORTS(sport, hnum); |
394 | struct sock *sk; | |
3ab5aee7 | 395 | const struct hlist_nulls_node *node; |
77a5ba55 PE |
396 | /* Optimize here for direct hit, only listening connections can |
397 | * have wildcards anyways. | |
398 | */ | |
9f26b3ad | 399 | unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport); |
f373b53b | 400 | unsigned int slot = hash & hashinfo->ehash_mask; |
3ab5aee7 | 401 | struct inet_ehash_bucket *head = &hashinfo->ehash[slot]; |
77a5ba55 | 402 | |
3ab5aee7 ED |
403 | begin: |
404 | sk_nulls_for_each_rcu(sk, node, &head->chain) { | |
ce43b03e ED |
405 | if (sk->sk_hash != hash) |
406 | continue; | |
407 | if (likely(INET_MATCH(sk, net, acookie, | |
3fa6f616 | 408 | saddr, daddr, ports, dif, sdif))) { |
41c6d650 | 409 | if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) |
05dbc7b5 | 410 | goto out; |
ce43b03e | 411 | if (unlikely(!INET_MATCH(sk, net, acookie, |
3fa6f616 DA |
412 | saddr, daddr, ports, |
413 | dif, sdif))) { | |
05dbc7b5 | 414 | sock_gen_put(sk); |
3ab5aee7 ED |
415 | goto begin; |
416 | } | |
05dbc7b5 | 417 | goto found; |
3ab5aee7 | 418 | } |
77a5ba55 | 419 | } |
3ab5aee7 ED |
420 | /* |
421 | * if the nulls value we got at the end of this lookup is | |
422 | * not the expected one, we must restart lookup. | |
423 | * We probably met an item that was moved to another chain. | |
424 | */ | |
425 | if (get_nulls_value(node) != slot) | |
426 | goto begin; | |
77a5ba55 | 427 | out: |
05dbc7b5 ED |
428 | sk = NULL; |
429 | found: | |
77a5ba55 | 430 | return sk; |
77a5ba55 PE |
431 | } |
432 | EXPORT_SYMBOL_GPL(__inet_lookup_established); | |
433 | ||
a7f5e7f1 ACM |
434 | /* called with local bh disabled */ |
435 | static int __inet_check_established(struct inet_timewait_death_row *death_row, | |
436 | struct sock *sk, __u16 lport, | |
437 | struct inet_timewait_sock **twp) | |
438 | { | |
439 | struct inet_hashinfo *hinfo = death_row->hashinfo; | |
440 | struct inet_sock *inet = inet_sk(sk); | |
c720c7e8 ED |
441 | __be32 daddr = inet->inet_rcv_saddr; |
442 | __be32 saddr = inet->inet_daddr; | |
a7f5e7f1 | 443 | int dif = sk->sk_bound_dev_if; |
3fa6f616 DA |
444 | struct net *net = sock_net(sk); |
445 | int sdif = l3mdev_master_ifindex_by_index(net, dif); | |
c7228317 | 446 | INET_ADDR_COOKIE(acookie, saddr, daddr); |
c720c7e8 | 447 | const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport); |
c720c7e8 ED |
448 | unsigned int hash = inet_ehashfn(net, daddr, lport, |
449 | saddr, inet->inet_dport); | |
a7f5e7f1 | 450 | struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash); |
9db66bdc | 451 | spinlock_t *lock = inet_ehash_lockp(hinfo, hash); |
a7f5e7f1 | 452 | struct sock *sk2; |
3ab5aee7 | 453 | const struct hlist_nulls_node *node; |
05dbc7b5 | 454 | struct inet_timewait_sock *tw = NULL; |
a7f5e7f1 | 455 | |
9db66bdc | 456 | spin_lock(lock); |
a7f5e7f1 | 457 | |
3ab5aee7 | 458 | sk_nulls_for_each(sk2, node, &head->chain) { |
ce43b03e ED |
459 | if (sk2->sk_hash != hash) |
460 | continue; | |
05dbc7b5 | 461 | |
ce43b03e | 462 | if (likely(INET_MATCH(sk2, net, acookie, |
3fa6f616 | 463 | saddr, daddr, ports, dif, sdif))) { |
05dbc7b5 ED |
464 | if (sk2->sk_state == TCP_TIME_WAIT) { |
465 | tw = inet_twsk(sk2); | |
466 | if (twsk_unique(sk, sk2, twp)) | |
467 | break; | |
468 | } | |
a7f5e7f1 | 469 | goto not_unique; |
05dbc7b5 | 470 | } |
a7f5e7f1 ACM |
471 | } |
472 | ||
a7f5e7f1 | 473 | /* Must record num and sport now. Otherwise we will see |
05dbc7b5 ED |
474 | * in hash table socket with a funny identity. |
475 | */ | |
c720c7e8 ED |
476 | inet->inet_num = lport; |
477 | inet->inet_sport = htons(lport); | |
a7f5e7f1 | 478 | sk->sk_hash = hash; |
547b792c | 479 | WARN_ON(!sk_unhashed(sk)); |
3ab5aee7 | 480 | __sk_nulls_add_node_rcu(sk, &head->chain); |
13475a30 | 481 | if (tw) { |
fc01538f | 482 | sk_nulls_del_node_init_rcu((struct sock *)tw); |
02a1d6e7 | 483 | __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED); |
13475a30 | 484 | } |
9db66bdc | 485 | spin_unlock(lock); |
c29a0bc4 | 486 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
a7f5e7f1 ACM |
487 | |
488 | if (twp) { | |
489 | *twp = tw; | |
a7f5e7f1 ACM |
490 | } else if (tw) { |
491 | /* Silly. Should hash-dance instead... */ | |
dbe7faa4 | 492 | inet_twsk_deschedule_put(tw); |
a7f5e7f1 | 493 | } |
a7f5e7f1 ACM |
494 | return 0; |
495 | ||
496 | not_unique: | |
9db66bdc | 497 | spin_unlock(lock); |
a7f5e7f1 ACM |
498 | return -EADDRNOTAVAIL; |
499 | } | |
500 | ||
e2baad9e | 501 | static u32 inet_sk_port_offset(const struct sock *sk) |
a7f5e7f1 ACM |
502 | { |
503 | const struct inet_sock *inet = inet_sk(sk); | |
e2baad9e | 504 | |
c720c7e8 ED |
505 | return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr, |
506 | inet->inet_daddr, | |
507 | inet->inet_dport); | |
a7f5e7f1 ACM |
508 | } |
509 | ||
079096f1 ED |
510 | /* insert a socket into ehash, and eventually remove another one |
511 | * (The another one can be a SYN_RECV or TIMEWAIT | |
512 | */ | |
5e0724d0 | 513 | bool inet_ehash_insert(struct sock *sk, struct sock *osk) |
152da81d | 514 | { |
39d8cda7 | 515 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
3ab5aee7 | 516 | struct hlist_nulls_head *list; |
152da81d | 517 | struct inet_ehash_bucket *head; |
5b441f76 | 518 | spinlock_t *lock; |
5e0724d0 | 519 | bool ret = true; |
152da81d | 520 | |
079096f1 | 521 | WARN_ON_ONCE(!sk_unhashed(sk)); |
152da81d | 522 | |
5b441f76 | 523 | sk->sk_hash = sk_ehashfn(sk); |
152da81d PE |
524 | head = inet_ehash_bucket(hashinfo, sk->sk_hash); |
525 | list = &head->chain; | |
526 | lock = inet_ehash_lockp(hashinfo, sk->sk_hash); | |
527 | ||
9db66bdc | 528 | spin_lock(lock); |
fc01538f | 529 | if (osk) { |
5e0724d0 ED |
530 | WARN_ON_ONCE(sk->sk_hash != osk->sk_hash); |
531 | ret = sk_nulls_del_node_init_rcu(osk); | |
9327f705 | 532 | } |
5e0724d0 ED |
533 | if (ret) |
534 | __sk_nulls_add_node_rcu(sk, list); | |
9db66bdc | 535 | spin_unlock(lock); |
079096f1 ED |
536 | return ret; |
537 | } | |
538 | ||
5e0724d0 | 539 | bool inet_ehash_nolisten(struct sock *sk, struct sock *osk) |
079096f1 | 540 | { |
5e0724d0 ED |
541 | bool ok = inet_ehash_insert(sk, osk); |
542 | ||
543 | if (ok) { | |
544 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); | |
545 | } else { | |
546 | percpu_counter_inc(sk->sk_prot->orphan_count); | |
563e0bb0 | 547 | inet_sk_set_state(sk, TCP_CLOSE); |
5e0724d0 ED |
548 | sock_set_flag(sk, SOCK_DEAD); |
549 | inet_csk_destroy_sock(sk); | |
550 | } | |
551 | return ok; | |
152da81d | 552 | } |
5e0724d0 | 553 | EXPORT_SYMBOL_GPL(inet_ehash_nolisten); |
152da81d | 554 | |
c125e80b | 555 | static int inet_reuseport_add_sock(struct sock *sk, |
fe38d2a1 | 556 | struct inet_listen_hashbucket *ilb) |
c125e80b | 557 | { |
90e5d0db | 558 | struct inet_bind_bucket *tb = inet_csk(sk)->icsk_bind_hash; |
c125e80b | 559 | struct sock *sk2; |
c125e80b CG |
560 | kuid_t uid = sock_i_uid(sk); |
561 | ||
85017869 | 562 | sk_for_each_rcu(sk2, &ilb->head) { |
c125e80b CG |
563 | if (sk2 != sk && |
564 | sk2->sk_family == sk->sk_family && | |
565 | ipv6_only_sock(sk2) == ipv6_only_sock(sk) && | |
566 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if && | |
90e5d0db | 567 | inet_csk(sk2)->icsk_bind_hash == tb && |
c125e80b | 568 | sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && |
fe38d2a1 | 569 | inet_rcv_saddr_equal(sk, sk2, false)) |
c125e80b CG |
570 | return reuseport_add_sock(sk, sk2); |
571 | } | |
572 | ||
1b5f962e | 573 | return reuseport_alloc(sk); |
c125e80b CG |
574 | } |
575 | ||
fe38d2a1 | 576 | int __inet_hash(struct sock *sk, struct sock *osk) |
152da81d | 577 | { |
39d8cda7 | 578 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
5caea4ea | 579 | struct inet_listen_hashbucket *ilb; |
c125e80b | 580 | int err = 0; |
152da81d | 581 | |
5e0724d0 ED |
582 | if (sk->sk_state != TCP_LISTEN) { |
583 | inet_ehash_nolisten(sk, osk); | |
c125e80b | 584 | return 0; |
5e0724d0 | 585 | } |
547b792c | 586 | WARN_ON(!sk_unhashed(sk)); |
5caea4ea | 587 | ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)]; |
152da81d | 588 | |
5caea4ea | 589 | spin_lock(&ilb->lock); |
c125e80b | 590 | if (sk->sk_reuseport) { |
fe38d2a1 | 591 | err = inet_reuseport_add_sock(sk, ilb); |
c125e80b CG |
592 | if (err) |
593 | goto unlock; | |
594 | } | |
d296ba60 CG |
595 | if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && |
596 | sk->sk_family == AF_INET6) | |
597 | hlist_add_tail_rcu(&sk->sk_node, &ilb->head); | |
598 | else | |
599 | hlist_add_head_rcu(&sk->sk_node, &ilb->head); | |
61b7c691 | 600 | inet_hash2(hashinfo, sk); |
76d013b2 | 601 | ilb->count++; |
3b24d854 | 602 | sock_set_flag(sk, SOCK_RCU_FREE); |
c29a0bc4 | 603 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
c125e80b | 604 | unlock: |
5caea4ea | 605 | spin_unlock(&ilb->lock); |
c125e80b CG |
606 | |
607 | return err; | |
152da81d | 608 | } |
77a6a471 | 609 | EXPORT_SYMBOL(__inet_hash); |
ab1e0a13 | 610 | |
086c653f | 611 | int inet_hash(struct sock *sk) |
ab1e0a13 | 612 | { |
c125e80b CG |
613 | int err = 0; |
614 | ||
ab1e0a13 ACM |
615 | if (sk->sk_state != TCP_CLOSE) { |
616 | local_bh_disable(); | |
fe38d2a1 | 617 | err = __inet_hash(sk, NULL); |
ab1e0a13 ACM |
618 | local_bh_enable(); |
619 | } | |
086c653f | 620 | |
c125e80b | 621 | return err; |
ab1e0a13 ACM |
622 | } |
623 | EXPORT_SYMBOL_GPL(inet_hash); | |
624 | ||
625 | void inet_unhash(struct sock *sk) | |
626 | { | |
39d8cda7 | 627 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
0ba98718 | 628 | struct inet_listen_hashbucket *ilb = NULL; |
c25eb3bf | 629 | spinlock_t *lock; |
ab1e0a13 ACM |
630 | |
631 | if (sk_unhashed(sk)) | |
5caea4ea | 632 | return; |
ab1e0a13 | 633 | |
3b24d854 | 634 | if (sk->sk_state == TCP_LISTEN) { |
76d013b2 MKL |
635 | ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)]; |
636 | lock = &ilb->lock; | |
3b24d854 | 637 | } else { |
c25eb3bf | 638 | lock = inet_ehash_lockp(hashinfo, sk->sk_hash); |
3b24d854 | 639 | } |
c25eb3bf | 640 | spin_lock_bh(lock); |
61b7c691 MKL |
641 | if (sk_unhashed(sk)) |
642 | goto unlock; | |
643 | ||
c125e80b CG |
644 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
645 | reuseport_detach_sock(sk); | |
0ba98718 | 646 | if (ilb) { |
61b7c691 MKL |
647 | inet_unhash2(hashinfo, sk); |
648 | __sk_del_node_init(sk); | |
649 | ilb->count--; | |
650 | } else { | |
651 | __sk_nulls_del_node_init_rcu(sk); | |
76d013b2 | 652 | } |
61b7c691 MKL |
653 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); |
654 | unlock: | |
920de804 | 655 | spin_unlock_bh(lock); |
ab1e0a13 ACM |
656 | } |
657 | EXPORT_SYMBOL_GPL(inet_unhash); | |
152da81d | 658 | |
5ee31fc1 | 659 | int __inet_hash_connect(struct inet_timewait_death_row *death_row, |
5d8c0aa9 | 660 | struct sock *sk, u32 port_offset, |
5ee31fc1 | 661 | int (*check_established)(struct inet_timewait_death_row *, |
b4d6444e | 662 | struct sock *, __u16, struct inet_timewait_sock **)) |
a7f5e7f1 ACM |
663 | { |
664 | struct inet_hashinfo *hinfo = death_row->hashinfo; | |
1580ab63 | 665 | struct inet_timewait_sock *tw = NULL; |
e905a9ed | 666 | struct inet_bind_hashbucket *head; |
1580ab63 | 667 | int port = inet_sk(sk)->inet_num; |
3b1e0a65 | 668 | struct net *net = sock_net(sk); |
1580ab63 ED |
669 | struct inet_bind_bucket *tb; |
670 | u32 remaining, offset; | |
671 | int ret, i, low, high; | |
672 | static u32 hint; | |
673 | ||
674 | if (port) { | |
675 | head = &hinfo->bhash[inet_bhashfn(net, port, | |
676 | hinfo->bhash_size)]; | |
677 | tb = inet_csk(sk)->icsk_bind_hash; | |
678 | spin_lock_bh(&head->lock); | |
679 | if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) { | |
680 | inet_ehash_nolisten(sk, NULL); | |
681 | spin_unlock_bh(&head->lock); | |
682 | return 0; | |
683 | } | |
684 | spin_unlock(&head->lock); | |
685 | /* No definite answer... Walk to established hash table */ | |
686 | ret = check_established(death_row, sk, port, NULL); | |
687 | local_bh_enable(); | |
688 | return ret; | |
689 | } | |
a7f5e7f1 | 690 | |
1580ab63 ED |
691 | inet_get_local_port_range(net, &low, &high); |
692 | high++; /* [32768, 60999] -> [32768, 61000[ */ | |
693 | remaining = high - low; | |
694 | if (likely(remaining > 1)) | |
695 | remaining &= ~1U; | |
a7f5e7f1 | 696 | |
1580ab63 ED |
697 | offset = (hint + port_offset) % remaining; |
698 | /* In first pass we try ports of @low parity. | |
699 | * inet_csk_get_port() does the opposite choice. | |
700 | */ | |
701 | offset &= ~1U; | |
702 | other_parity_scan: | |
703 | port = low + offset; | |
704 | for (i = 0; i < remaining; i += 2, port += 2) { | |
705 | if (unlikely(port >= high)) | |
706 | port -= remaining; | |
707 | if (inet_is_local_reserved_port(net, port)) | |
708 | continue; | |
709 | head = &hinfo->bhash[inet_bhashfn(net, port, | |
710 | hinfo->bhash_size)]; | |
711 | spin_lock_bh(&head->lock); | |
227b60f5 | 712 | |
1580ab63 ED |
713 | /* Does not bother with rcv_saddr checks, because |
714 | * the established check is already unique enough. | |
07f4c900 | 715 | */ |
1580ab63 ED |
716 | inet_bind_bucket_for_each(tb, &head->chain) { |
717 | if (net_eq(ib_net(tb), net) && tb->port == port) { | |
718 | if (tb->fastreuse >= 0 || | |
719 | tb->fastreuseport >= 0) | |
e905a9ed | 720 | goto next_port; |
1580ab63 ED |
721 | WARN_ON(hlist_empty(&tb->owners)); |
722 | if (!check_established(death_row, sk, | |
723 | port, &tw)) | |
724 | goto ok; | |
725 | goto next_port; | |
e905a9ed | 726 | } |
e905a9ed | 727 | } |
a7f5e7f1 | 728 | |
1580ab63 ED |
729 | tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, |
730 | net, head, port); | |
731 | if (!tb) { | |
732 | spin_unlock_bh(&head->lock); | |
733 | return -ENOMEM; | |
e905a9ed | 734 | } |
1580ab63 ED |
735 | tb->fastreuse = -1; |
736 | tb->fastreuseport = -1; | |
737 | goto ok; | |
738 | next_port: | |
739 | spin_unlock_bh(&head->lock); | |
740 | cond_resched(); | |
741 | } | |
a7f5e7f1 | 742 | |
1580ab63 ED |
743 | offset++; |
744 | if ((offset & 1) && remaining > 1) | |
745 | goto other_parity_scan; | |
a7f5e7f1 | 746 | |
1580ab63 | 747 | return -EADDRNOTAVAIL; |
a7f5e7f1 | 748 | |
1580ab63 ED |
749 | ok: |
750 | hint += i + 2; | |
751 | ||
752 | /* Head lock still held and bh's disabled */ | |
753 | inet_bind_hash(sk, tb, port); | |
754 | if (sk_unhashed(sk)) { | |
755 | inet_sk(sk)->inet_sport = htons(port); | |
756 | inet_ehash_nolisten(sk, (struct sock *)tw); | |
a7f5e7f1 | 757 | } |
1580ab63 ED |
758 | if (tw) |
759 | inet_twsk_bind_unhash(tw, hinfo); | |
760 | spin_unlock(&head->lock); | |
761 | if (tw) | |
762 | inet_twsk_deschedule_put(tw); | |
763 | local_bh_enable(); | |
764 | return 0; | |
a7f5e7f1 | 765 | } |
5ee31fc1 PE |
766 | |
767 | /* | |
768 | * Bind a port for a connect operation and hash it. | |
769 | */ | |
770 | int inet_hash_connect(struct inet_timewait_death_row *death_row, | |
771 | struct sock *sk) | |
772 | { | |
e2baad9e ED |
773 | u32 port_offset = 0; |
774 | ||
775 | if (!inet_sk(sk)->inet_num) | |
776 | port_offset = inet_sk_port_offset(sk); | |
777 | return __inet_hash_connect(death_row, sk, port_offset, | |
b4d6444e | 778 | __inet_check_established); |
5ee31fc1 | 779 | } |
a7f5e7f1 | 780 | EXPORT_SYMBOL_GPL(inet_hash_connect); |
5caea4ea ED |
781 | |
782 | void inet_hashinfo_init(struct inet_hashinfo *h) | |
783 | { | |
784 | int i; | |
785 | ||
c25eb3bf | 786 | for (i = 0; i < INET_LHTABLE_SIZE; i++) { |
5caea4ea | 787 | spin_lock_init(&h->listening_hash[i].lock); |
3b24d854 | 788 | INIT_HLIST_HEAD(&h->listening_hash[i].head); |
76d013b2 | 789 | h->listening_hash[i].count = 0; |
3b24d854 | 790 | } |
61b7c691 MKL |
791 | |
792 | h->lhash2 = NULL; | |
5caea4ea | 793 | } |
5caea4ea | 794 | EXPORT_SYMBOL_GPL(inet_hashinfo_init); |
095dc8e0 | 795 | |
61b7c691 MKL |
796 | void __init inet_hashinfo2_init(struct inet_hashinfo *h, const char *name, |
797 | unsigned long numentries, int scale, | |
798 | unsigned long low_limit, | |
799 | unsigned long high_limit) | |
800 | { | |
801 | unsigned int i; | |
802 | ||
803 | h->lhash2 = alloc_large_system_hash(name, | |
804 | sizeof(*h->lhash2), | |
805 | numentries, | |
806 | scale, | |
807 | 0, | |
808 | NULL, | |
809 | &h->lhash2_mask, | |
810 | low_limit, | |
811 | high_limit); | |
812 | ||
813 | for (i = 0; i <= h->lhash2_mask; i++) { | |
814 | spin_lock_init(&h->lhash2[i].lock); | |
815 | INIT_HLIST_HEAD(&h->lhash2[i].head); | |
816 | h->lhash2[i].count = 0; | |
817 | } | |
818 | } | |
819 | ||
095dc8e0 ED |
820 | int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo) |
821 | { | |
89e478a2 | 822 | unsigned int locksz = sizeof(spinlock_t); |
095dc8e0 ED |
823 | unsigned int i, nblocks = 1; |
824 | ||
89e478a2 | 825 | if (locksz != 0) { |
095dc8e0 | 826 | /* allocate 2 cache lines or at least one spinlock per cpu */ |
89e478a2 | 827 | nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U); |
095dc8e0 ED |
828 | nblocks = roundup_pow_of_two(nblocks * num_possible_cpus()); |
829 | ||
830 | /* no more locks than number of hash buckets */ | |
831 | nblocks = min(nblocks, hashinfo->ehash_mask + 1); | |
832 | ||
752ade68 | 833 | hashinfo->ehash_locks = kvmalloc_array(nblocks, locksz, GFP_KERNEL); |
095dc8e0 ED |
834 | if (!hashinfo->ehash_locks) |
835 | return -ENOMEM; | |
836 | ||
837 | for (i = 0; i < nblocks; i++) | |
838 | spin_lock_init(&hashinfo->ehash_locks[i]); | |
839 | } | |
840 | hashinfo->ehash_locks_mask = nblocks - 1; | |
841 | return 0; | |
842 | } | |
843 | EXPORT_SYMBOL_GPL(inet_ehash_locks_alloc); |