Commit | Line | Data |
---|---|---|
3f421baa 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 | * Support for INET connection oriented protocols. | |
7 | * | |
8 | * Authors: See the TCP sources | |
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 | ||
3f421baa ACM |
16 | #include <linux/module.h> |
17 | #include <linux/jhash.h> | |
18 | ||
19 | #include <net/inet_connection_sock.h> | |
20 | #include <net/inet_hashtables.h> | |
21 | #include <net/inet_timewait_sock.h> | |
22 | #include <net/ip.h> | |
23 | #include <net/route.h> | |
24 | #include <net/tcp_states.h> | |
a019d6fe | 25 | #include <net/xfrm.h> |
3f421baa ACM |
26 | |
27 | #ifdef INET_CSK_DEBUG | |
28 | const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n"; | |
29 | EXPORT_SYMBOL(inet_csk_timer_bug_msg); | |
30 | #endif | |
31 | ||
32 | /* | |
3c689b73 | 33 | * This struct holds the first and last local port number. |
3f421baa | 34 | */ |
3c689b73 ED |
35 | struct local_ports sysctl_local_ports __read_mostly = { |
36 | .lock = SEQLOCK_UNLOCKED, | |
37 | .range = { 32768, 61000 }, | |
38 | }; | |
227b60f5 SH |
39 | |
40 | void inet_get_local_port_range(int *low, int *high) | |
41 | { | |
42 | unsigned seq; | |
43 | do { | |
3c689b73 | 44 | seq = read_seqbegin(&sysctl_local_ports.lock); |
227b60f5 | 45 | |
3c689b73 ED |
46 | *low = sysctl_local_ports.range[0]; |
47 | *high = sysctl_local_ports.range[1]; | |
48 | } while (read_seqretry(&sysctl_local_ports.lock, seq)); | |
227b60f5 SH |
49 | } |
50 | EXPORT_SYMBOL(inet_get_local_port_range); | |
3f421baa | 51 | |
971af18b ACM |
52 | int inet_csk_bind_conflict(const struct sock *sk, |
53 | const struct inet_bind_bucket *tb) | |
3f421baa | 54 | { |
82103232 | 55 | const __be32 sk_rcv_saddr = inet_rcv_saddr(sk); |
3f421baa ACM |
56 | struct sock *sk2; |
57 | struct hlist_node *node; | |
58 | int reuse = sk->sk_reuse; | |
59 | ||
7477fd2e PE |
60 | /* |
61 | * Unlike other sk lookup places we do not check | |
62 | * for sk_net here, since _all_ the socks listed | |
63 | * in tb->owners list belong to the same net - the | |
64 | * one this bucket belongs to. | |
65 | */ | |
66 | ||
3f421baa ACM |
67 | sk_for_each_bound(sk2, node, &tb->owners) { |
68 | if (sk != sk2 && | |
69 | !inet_v6_ipv6only(sk2) && | |
70 | (!sk->sk_bound_dev_if || | |
71 | !sk2->sk_bound_dev_if || | |
72 | sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { | |
73 | if (!reuse || !sk2->sk_reuse || | |
74 | sk2->sk_state == TCP_LISTEN) { | |
82103232 | 75 | const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2); |
3f421baa ACM |
76 | if (!sk2_rcv_saddr || !sk_rcv_saddr || |
77 | sk2_rcv_saddr == sk_rcv_saddr) | |
78 | break; | |
79 | } | |
80 | } | |
81 | } | |
82 | return node != NULL; | |
83 | } | |
84 | ||
971af18b ACM |
85 | EXPORT_SYMBOL_GPL(inet_csk_bind_conflict); |
86 | ||
3f421baa ACM |
87 | /* Obtain a reference to a local port for the given sock, |
88 | * if snum is zero it means select any available local port. | |
89 | */ | |
ab1e0a13 | 90 | int inet_csk_get_port(struct sock *sk, unsigned short snum) |
3f421baa | 91 | { |
39d8cda7 | 92 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
3f421baa ACM |
93 | struct inet_bind_hashbucket *head; |
94 | struct hlist_node *node; | |
95 | struct inet_bind_bucket *tb; | |
96 | int ret; | |
3b1e0a65 | 97 | struct net *net = sock_net(sk); |
3f421baa ACM |
98 | |
99 | local_bh_disable(); | |
100 | if (!snum) { | |
227b60f5 SH |
101 | int remaining, rover, low, high; |
102 | ||
103 | inet_get_local_port_range(&low, &high); | |
a25de534 | 104 | remaining = (high - low) + 1; |
227b60f5 | 105 | rover = net_random() % remaining + low; |
3f421baa | 106 | |
3f421baa | 107 | do { |
7f635ab7 PE |
108 | head = &hashinfo->bhash[inet_bhashfn(net, rover, |
109 | hashinfo->bhash_size)]; | |
3f421baa ACM |
110 | spin_lock(&head->lock); |
111 | inet_bind_bucket_for_each(tb, node, &head->chain) | |
941b1d22 | 112 | if (tb->ib_net == net && tb->port == rover) |
3f421baa ACM |
113 | goto next; |
114 | break; | |
115 | next: | |
116 | spin_unlock(&head->lock); | |
6df71634 SH |
117 | if (++rover > high) |
118 | rover = low; | |
3f421baa | 119 | } while (--remaining > 0); |
3f421baa ACM |
120 | |
121 | /* Exhausted local port range during search? It is not | |
122 | * possible for us to be holding one of the bind hash | |
123 | * locks if this test triggers, because if 'remaining' | |
124 | * drops to zero, we broke out of the do/while loop at | |
125 | * the top level, not from the 'break;' statement. | |
126 | */ | |
127 | ret = 1; | |
128 | if (remaining <= 0) | |
129 | goto fail; | |
130 | ||
131 | /* OK, here is the one we will use. HEAD is | |
132 | * non-NULL and we hold it's mutex. | |
133 | */ | |
134 | snum = rover; | |
135 | } else { | |
7f635ab7 PE |
136 | head = &hashinfo->bhash[inet_bhashfn(net, snum, |
137 | hashinfo->bhash_size)]; | |
3f421baa ACM |
138 | spin_lock(&head->lock); |
139 | inet_bind_bucket_for_each(tb, node, &head->chain) | |
941b1d22 | 140 | if (tb->ib_net == net && tb->port == snum) |
3f421baa ACM |
141 | goto tb_found; |
142 | } | |
143 | tb = NULL; | |
144 | goto tb_not_found; | |
145 | tb_found: | |
146 | if (!hlist_empty(&tb->owners)) { | |
3f421baa ACM |
147 | if (tb->fastreuse > 0 && |
148 | sk->sk_reuse && sk->sk_state != TCP_LISTEN) { | |
149 | goto success; | |
150 | } else { | |
151 | ret = 1; | |
ab1e0a13 | 152 | if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) |
3f421baa ACM |
153 | goto fail_unlock; |
154 | } | |
155 | } | |
156 | tb_not_found: | |
157 | ret = 1; | |
941b1d22 PE |
158 | if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep, |
159 | net, head, snum)) == NULL) | |
3f421baa ACM |
160 | goto fail_unlock; |
161 | if (hlist_empty(&tb->owners)) { | |
162 | if (sk->sk_reuse && sk->sk_state != TCP_LISTEN) | |
163 | tb->fastreuse = 1; | |
164 | else | |
165 | tb->fastreuse = 0; | |
166 | } else if (tb->fastreuse && | |
167 | (!sk->sk_reuse || sk->sk_state == TCP_LISTEN)) | |
168 | tb->fastreuse = 0; | |
169 | success: | |
170 | if (!inet_csk(sk)->icsk_bind_hash) | |
171 | inet_bind_hash(sk, tb, snum); | |
547b792c | 172 | WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); |
e905a9ed | 173 | ret = 0; |
3f421baa ACM |
174 | |
175 | fail_unlock: | |
176 | spin_unlock(&head->lock); | |
177 | fail: | |
178 | local_bh_enable(); | |
179 | return ret; | |
180 | } | |
181 | ||
182 | EXPORT_SYMBOL_GPL(inet_csk_get_port); | |
183 | ||
184 | /* | |
185 | * Wait for an incoming connection, avoid race conditions. This must be called | |
186 | * with the socket locked. | |
187 | */ | |
188 | static int inet_csk_wait_for_connect(struct sock *sk, long timeo) | |
189 | { | |
190 | struct inet_connection_sock *icsk = inet_csk(sk); | |
191 | DEFINE_WAIT(wait); | |
192 | int err; | |
193 | ||
194 | /* | |
195 | * True wake-one mechanism for incoming connections: only | |
196 | * one process gets woken up, not the 'whole herd'. | |
197 | * Since we do not 'race & poll' for established sockets | |
198 | * anymore, the common case will execute the loop only once. | |
199 | * | |
200 | * Subtle issue: "add_wait_queue_exclusive()" will be added | |
201 | * after any current non-exclusive waiters, and we know that | |
202 | * it will always _stay_ after any new non-exclusive waiters | |
203 | * because all non-exclusive waiters are added at the | |
204 | * beginning of the wait-queue. As such, it's ok to "drop" | |
205 | * our exclusiveness temporarily when we get woken up without | |
206 | * having to remove and re-insert us on the wait queue. | |
207 | */ | |
208 | for (;;) { | |
209 | prepare_to_wait_exclusive(sk->sk_sleep, &wait, | |
210 | TASK_INTERRUPTIBLE); | |
211 | release_sock(sk); | |
212 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
213 | timeo = schedule_timeout(timeo); | |
214 | lock_sock(sk); | |
215 | err = 0; | |
216 | if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
217 | break; | |
218 | err = -EINVAL; | |
219 | if (sk->sk_state != TCP_LISTEN) | |
220 | break; | |
221 | err = sock_intr_errno(timeo); | |
222 | if (signal_pending(current)) | |
223 | break; | |
224 | err = -EAGAIN; | |
225 | if (!timeo) | |
226 | break; | |
227 | } | |
228 | finish_wait(sk->sk_sleep, &wait); | |
229 | return err; | |
230 | } | |
231 | ||
232 | /* | |
233 | * This will accept the next outstanding connection. | |
234 | */ | |
235 | struct sock *inet_csk_accept(struct sock *sk, int flags, int *err) | |
236 | { | |
237 | struct inet_connection_sock *icsk = inet_csk(sk); | |
238 | struct sock *newsk; | |
239 | int error; | |
240 | ||
241 | lock_sock(sk); | |
242 | ||
243 | /* We need to make sure that this socket is listening, | |
244 | * and that it has something pending. | |
245 | */ | |
246 | error = -EINVAL; | |
247 | if (sk->sk_state != TCP_LISTEN) | |
248 | goto out_err; | |
249 | ||
250 | /* Find already established connection */ | |
251 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) { | |
252 | long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); | |
253 | ||
254 | /* If this is a non blocking socket don't sleep */ | |
255 | error = -EAGAIN; | |
256 | if (!timeo) | |
257 | goto out_err; | |
258 | ||
259 | error = inet_csk_wait_for_connect(sk, timeo); | |
260 | if (error) | |
261 | goto out_err; | |
262 | } | |
263 | ||
264 | newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk); | |
547b792c | 265 | WARN_ON(newsk->sk_state == TCP_SYN_RECV); |
3f421baa ACM |
266 | out: |
267 | release_sock(sk); | |
268 | return newsk; | |
269 | out_err: | |
270 | newsk = NULL; | |
271 | *err = error; | |
272 | goto out; | |
273 | } | |
274 | ||
275 | EXPORT_SYMBOL(inet_csk_accept); | |
276 | ||
277 | /* | |
278 | * Using different timers for retransmit, delayed acks and probes | |
e905a9ed | 279 | * We may wish use just one timer maintaining a list of expire jiffies |
3f421baa ACM |
280 | * to optimize. |
281 | */ | |
282 | void inet_csk_init_xmit_timers(struct sock *sk, | |
283 | void (*retransmit_handler)(unsigned long), | |
284 | void (*delack_handler)(unsigned long), | |
285 | void (*keepalive_handler)(unsigned long)) | |
286 | { | |
287 | struct inet_connection_sock *icsk = inet_csk(sk); | |
288 | ||
b24b8a24 PE |
289 | setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler, |
290 | (unsigned long)sk); | |
291 | setup_timer(&icsk->icsk_delack_timer, delack_handler, | |
292 | (unsigned long)sk); | |
293 | setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk); | |
3f421baa ACM |
294 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
295 | } | |
296 | ||
297 | EXPORT_SYMBOL(inet_csk_init_xmit_timers); | |
298 | ||
299 | void inet_csk_clear_xmit_timers(struct sock *sk) | |
300 | { | |
301 | struct inet_connection_sock *icsk = inet_csk(sk); | |
302 | ||
303 | icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0; | |
304 | ||
305 | sk_stop_timer(sk, &icsk->icsk_retransmit_timer); | |
306 | sk_stop_timer(sk, &icsk->icsk_delack_timer); | |
307 | sk_stop_timer(sk, &sk->sk_timer); | |
308 | } | |
309 | ||
310 | EXPORT_SYMBOL(inet_csk_clear_xmit_timers); | |
311 | ||
312 | void inet_csk_delete_keepalive_timer(struct sock *sk) | |
313 | { | |
314 | sk_stop_timer(sk, &sk->sk_timer); | |
315 | } | |
316 | ||
317 | EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); | |
318 | ||
319 | void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) | |
320 | { | |
321 | sk_reset_timer(sk, &sk->sk_timer, jiffies + len); | |
322 | } | |
323 | ||
324 | EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); | |
325 | ||
326 | struct dst_entry* inet_csk_route_req(struct sock *sk, | |
327 | const struct request_sock *req) | |
328 | { | |
329 | struct rtable *rt; | |
330 | const struct inet_request_sock *ireq = inet_rsk(req); | |
331 | struct ip_options *opt = inet_rsk(req)->opt; | |
332 | struct flowi fl = { .oif = sk->sk_bound_dev_if, | |
333 | .nl_u = { .ip4_u = | |
334 | { .daddr = ((opt && opt->srr) ? | |
335 | opt->faddr : | |
336 | ireq->rmt_addr), | |
337 | .saddr = ireq->loc_addr, | |
338 | .tos = RT_CONN_FLAGS(sk) } }, | |
339 | .proto = sk->sk_protocol, | |
86b08d86 | 340 | .flags = inet_sk_flowi_flags(sk), |
3f421baa ACM |
341 | .uli_u = { .ports = |
342 | { .sport = inet_sk(sk)->sport, | |
343 | .dport = ireq->rmt_port } } }; | |
84a3aa00 | 344 | struct net *net = sock_net(sk); |
3f421baa | 345 | |
4237c75c | 346 | security_req_classify_flow(req, &fl); |
84a3aa00 | 347 | if (ip_route_output_flow(net, &rt, &fl, sk, 0)) { |
7c73a6fa | 348 | IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); |
3f421baa ACM |
349 | return NULL; |
350 | } | |
351 | if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) { | |
352 | ip_rt_put(rt); | |
7c73a6fa | 353 | IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); |
3f421baa ACM |
354 | return NULL; |
355 | } | |
356 | return &rt->u.dst; | |
357 | } | |
358 | ||
359 | EXPORT_SYMBOL_GPL(inet_csk_route_req); | |
360 | ||
6b72977b | 361 | static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport, |
72a3effa | 362 | const u32 rnd, const u32 synq_hsize) |
3f421baa | 363 | { |
6b72977b | 364 | return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1); |
3f421baa ACM |
365 | } |
366 | ||
367 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
368 | #define AF_INET_FAMILY(fam) ((fam) == AF_INET) | |
369 | #else | |
370 | #define AF_INET_FAMILY(fam) 1 | |
371 | #endif | |
372 | ||
373 | struct request_sock *inet_csk_search_req(const struct sock *sk, | |
374 | struct request_sock ***prevp, | |
6b72977b | 375 | const __be16 rport, const __be32 raddr, |
7f25afbb | 376 | const __be32 laddr) |
3f421baa ACM |
377 | { |
378 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
379 | struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; | |
380 | struct request_sock *req, **prev; | |
381 | ||
382 | for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd, | |
383 | lopt->nr_table_entries)]; | |
384 | (req = *prev) != NULL; | |
385 | prev = &req->dl_next) { | |
386 | const struct inet_request_sock *ireq = inet_rsk(req); | |
387 | ||
388 | if (ireq->rmt_port == rport && | |
389 | ireq->rmt_addr == raddr && | |
390 | ireq->loc_addr == laddr && | |
391 | AF_INET_FAMILY(req->rsk_ops->family)) { | |
547b792c | 392 | WARN_ON(req->sk); |
3f421baa ACM |
393 | *prevp = prev; |
394 | break; | |
395 | } | |
396 | } | |
397 | ||
398 | return req; | |
399 | } | |
400 | ||
401 | EXPORT_SYMBOL_GPL(inet_csk_search_req); | |
402 | ||
403 | void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, | |
c2977c22 | 404 | unsigned long timeout) |
3f421baa ACM |
405 | { |
406 | struct inet_connection_sock *icsk = inet_csk(sk); | |
407 | struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; | |
408 | const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port, | |
409 | lopt->hash_rnd, lopt->nr_table_entries); | |
410 | ||
411 | reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout); | |
412 | inet_csk_reqsk_queue_added(sk, timeout); | |
413 | } | |
414 | ||
a019d6fe ACM |
415 | /* Only thing we need from tcp.h */ |
416 | extern int sysctl_tcp_synack_retries; | |
417 | ||
3f421baa | 418 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); |
9f1d2604 | 419 | |
a019d6fe ACM |
420 | void inet_csk_reqsk_queue_prune(struct sock *parent, |
421 | const unsigned long interval, | |
422 | const unsigned long timeout, | |
423 | const unsigned long max_rto) | |
424 | { | |
425 | struct inet_connection_sock *icsk = inet_csk(parent); | |
426 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; | |
427 | struct listen_sock *lopt = queue->listen_opt; | |
ec0a1966 DM |
428 | int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries; |
429 | int thresh = max_retries; | |
a019d6fe ACM |
430 | unsigned long now = jiffies; |
431 | struct request_sock **reqp, *req; | |
432 | int i, budget; | |
433 | ||
434 | if (lopt == NULL || lopt->qlen == 0) | |
435 | return; | |
436 | ||
437 | /* Normally all the openreqs are young and become mature | |
438 | * (i.e. converted to established socket) for first timeout. | |
439 | * If synack was not acknowledged for 3 seconds, it means | |
440 | * one of the following things: synack was lost, ack was lost, | |
441 | * rtt is high or nobody planned to ack (i.e. synflood). | |
442 | * When server is a bit loaded, queue is populated with old | |
443 | * open requests, reducing effective size of queue. | |
444 | * When server is well loaded, queue size reduces to zero | |
445 | * after several minutes of work. It is not synflood, | |
446 | * it is normal operation. The solution is pruning | |
447 | * too old entries overriding normal timeout, when | |
448 | * situation becomes dangerous. | |
449 | * | |
450 | * Essentially, we reserve half of room for young | |
451 | * embrions; and abort old ones without pity, if old | |
452 | * ones are about to clog our table. | |
453 | */ | |
454 | if (lopt->qlen>>(lopt->max_qlen_log-1)) { | |
455 | int young = (lopt->qlen_young<<1); | |
456 | ||
457 | while (thresh > 2) { | |
458 | if (lopt->qlen < young) | |
459 | break; | |
460 | thresh--; | |
461 | young <<= 1; | |
462 | } | |
463 | } | |
464 | ||
ec0a1966 DM |
465 | if (queue->rskq_defer_accept) |
466 | max_retries = queue->rskq_defer_accept; | |
467 | ||
a019d6fe ACM |
468 | budget = 2 * (lopt->nr_table_entries / (timeout / interval)); |
469 | i = lopt->clock_hand; | |
470 | ||
471 | do { | |
472 | reqp=&lopt->syn_table[i]; | |
473 | while ((req = *reqp) != NULL) { | |
474 | if (time_after_eq(now, req->expires)) { | |
93653e04 DM |
475 | if ((req->retrans < thresh || |
476 | (inet_rsk(req)->acked && req->retrans < max_retries)) | |
477 | && !req->rsk_ops->rtx_syn_ack(parent, req)) { | |
a019d6fe ACM |
478 | unsigned long timeo; |
479 | ||
480 | if (req->retrans++ == 0) | |
481 | lopt->qlen_young--; | |
482 | timeo = min((timeout << req->retrans), max_rto); | |
483 | req->expires = now + timeo; | |
484 | reqp = &req->dl_next; | |
485 | continue; | |
486 | } | |
487 | ||
488 | /* Drop this request */ | |
489 | inet_csk_reqsk_queue_unlink(parent, req, reqp); | |
490 | reqsk_queue_removed(queue, req); | |
491 | reqsk_free(req); | |
492 | continue; | |
493 | } | |
494 | reqp = &req->dl_next; | |
495 | } | |
496 | ||
497 | i = (i + 1) & (lopt->nr_table_entries - 1); | |
498 | ||
499 | } while (--budget > 0); | |
500 | ||
501 | lopt->clock_hand = i; | |
502 | ||
503 | if (lopt->qlen) | |
504 | inet_csk_reset_keepalive_timer(parent, interval); | |
505 | } | |
506 | ||
507 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune); | |
508 | ||
9f1d2604 | 509 | struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req, |
dd0fc66f | 510 | const gfp_t priority) |
9f1d2604 ACM |
511 | { |
512 | struct sock *newsk = sk_clone(sk, priority); | |
513 | ||
514 | if (newsk != NULL) { | |
515 | struct inet_connection_sock *newicsk = inet_csk(newsk); | |
516 | ||
517 | newsk->sk_state = TCP_SYN_RECV; | |
518 | newicsk->icsk_bind_hash = NULL; | |
519 | ||
520 | inet_sk(newsk)->dport = inet_rsk(req)->rmt_port; | |
a3116ac5 KK |
521 | inet_sk(newsk)->num = ntohs(inet_rsk(req)->loc_port); |
522 | inet_sk(newsk)->sport = inet_rsk(req)->loc_port; | |
9f1d2604 ACM |
523 | newsk->sk_write_space = sk_stream_write_space; |
524 | ||
525 | newicsk->icsk_retransmits = 0; | |
6687e988 ACM |
526 | newicsk->icsk_backoff = 0; |
527 | newicsk->icsk_probes_out = 0; | |
9f1d2604 ACM |
528 | |
529 | /* Deinitialize accept_queue to trap illegal accesses. */ | |
530 | memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); | |
4237c75c VY |
531 | |
532 | security_inet_csk_clone(newsk, req); | |
9f1d2604 ACM |
533 | } |
534 | return newsk; | |
535 | } | |
536 | ||
537 | EXPORT_SYMBOL_GPL(inet_csk_clone); | |
a019d6fe ACM |
538 | |
539 | /* | |
540 | * At this point, there should be no process reference to this | |
541 | * socket, and thus no user references at all. Therefore we | |
542 | * can assume the socket waitqueue is inactive and nobody will | |
543 | * try to jump onto it. | |
544 | */ | |
545 | void inet_csk_destroy_sock(struct sock *sk) | |
546 | { | |
547b792c IJ |
547 | WARN_ON(sk->sk_state != TCP_CLOSE); |
548 | WARN_ON(!sock_flag(sk, SOCK_DEAD)); | |
a019d6fe ACM |
549 | |
550 | /* It cannot be in hash table! */ | |
547b792c | 551 | WARN_ON(!sk_unhashed(sk)); |
a019d6fe ACM |
552 | |
553 | /* If it has not 0 inet_sk(sk)->num, it must be bound */ | |
547b792c | 554 | WARN_ON(inet_sk(sk)->num && !inet_csk(sk)->icsk_bind_hash); |
a019d6fe ACM |
555 | |
556 | sk->sk_prot->destroy(sk); | |
557 | ||
558 | sk_stream_kill_queues(sk); | |
559 | ||
560 | xfrm_sk_free_policy(sk); | |
561 | ||
562 | sk_refcnt_debug_release(sk); | |
563 | ||
564 | atomic_dec(sk->sk_prot->orphan_count); | |
565 | sock_put(sk); | |
566 | } | |
567 | ||
568 | EXPORT_SYMBOL(inet_csk_destroy_sock); | |
569 | ||
570 | int inet_csk_listen_start(struct sock *sk, const int nr_table_entries) | |
571 | { | |
572 | struct inet_sock *inet = inet_sk(sk); | |
573 | struct inet_connection_sock *icsk = inet_csk(sk); | |
574 | int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries); | |
575 | ||
576 | if (rc != 0) | |
577 | return rc; | |
578 | ||
579 | sk->sk_max_ack_backlog = 0; | |
580 | sk->sk_ack_backlog = 0; | |
581 | inet_csk_delack_init(sk); | |
582 | ||
583 | /* There is race window here: we announce ourselves listening, | |
584 | * but this transition is still not validated by get_port(). | |
585 | * It is OK, because this socket enters to hash table only | |
586 | * after validation is complete. | |
587 | */ | |
588 | sk->sk_state = TCP_LISTEN; | |
589 | if (!sk->sk_prot->get_port(sk, inet->num)) { | |
590 | inet->sport = htons(inet->num); | |
591 | ||
592 | sk_dst_reset(sk); | |
593 | sk->sk_prot->hash(sk); | |
594 | ||
595 | return 0; | |
596 | } | |
597 | ||
598 | sk->sk_state = TCP_CLOSE; | |
599 | __reqsk_queue_destroy(&icsk->icsk_accept_queue); | |
600 | return -EADDRINUSE; | |
601 | } | |
602 | ||
603 | EXPORT_SYMBOL_GPL(inet_csk_listen_start); | |
604 | ||
605 | /* | |
606 | * This routine closes sockets which have been at least partially | |
607 | * opened, but not yet accepted. | |
608 | */ | |
609 | void inet_csk_listen_stop(struct sock *sk) | |
610 | { | |
611 | struct inet_connection_sock *icsk = inet_csk(sk); | |
612 | struct request_sock *acc_req; | |
613 | struct request_sock *req; | |
614 | ||
615 | inet_csk_delete_keepalive_timer(sk); | |
616 | ||
617 | /* make all the listen_opt local to us */ | |
618 | acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue); | |
619 | ||
620 | /* Following specs, it would be better either to send FIN | |
621 | * (and enter FIN-WAIT-1, it is normal close) | |
622 | * or to send active reset (abort). | |
623 | * Certainly, it is pretty dangerous while synflood, but it is | |
624 | * bad justification for our negligence 8) | |
625 | * To be honest, we are not able to make either | |
626 | * of the variants now. --ANK | |
627 | */ | |
628 | reqsk_queue_destroy(&icsk->icsk_accept_queue); | |
629 | ||
630 | while ((req = acc_req) != NULL) { | |
631 | struct sock *child = req->sk; | |
632 | ||
633 | acc_req = req->dl_next; | |
634 | ||
635 | local_bh_disable(); | |
636 | bh_lock_sock(child); | |
547b792c | 637 | WARN_ON(sock_owned_by_user(child)); |
a019d6fe ACM |
638 | sock_hold(child); |
639 | ||
640 | sk->sk_prot->disconnect(child, O_NONBLOCK); | |
641 | ||
642 | sock_orphan(child); | |
643 | ||
644 | atomic_inc(sk->sk_prot->orphan_count); | |
645 | ||
646 | inet_csk_destroy_sock(child); | |
647 | ||
648 | bh_unlock_sock(child); | |
649 | local_bh_enable(); | |
650 | sock_put(child); | |
651 | ||
652 | sk_acceptq_removed(sk); | |
653 | __reqsk_free(req); | |
654 | } | |
547b792c | 655 | WARN_ON(sk->sk_ack_backlog); |
a019d6fe ACM |
656 | } |
657 | ||
658 | EXPORT_SYMBOL_GPL(inet_csk_listen_stop); | |
af05dc93 ACM |
659 | |
660 | void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) | |
661 | { | |
662 | struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; | |
663 | const struct inet_sock *inet = inet_sk(sk); | |
664 | ||
665 | sin->sin_family = AF_INET; | |
666 | sin->sin_addr.s_addr = inet->daddr; | |
667 | sin->sin_port = inet->dport; | |
668 | } | |
669 | ||
670 | EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); | |
c4d93909 | 671 | |
dec73ff0 ACM |
672 | #ifdef CONFIG_COMPAT |
673 | int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname, | |
674 | char __user *optval, int __user *optlen) | |
675 | { | |
dbeff12b | 676 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 ACM |
677 | |
678 | if (icsk->icsk_af_ops->compat_getsockopt != NULL) | |
679 | return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname, | |
680 | optval, optlen); | |
681 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
682 | optval, optlen); | |
683 | } | |
684 | ||
685 | EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt); | |
686 | ||
687 | int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname, | |
688 | char __user *optval, int optlen) | |
689 | { | |
dbeff12b | 690 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 ACM |
691 | |
692 | if (icsk->icsk_af_ops->compat_setsockopt != NULL) | |
693 | return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname, | |
694 | optval, optlen); | |
695 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
696 | optval, optlen); | |
697 | } | |
698 | ||
699 | EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt); | |
700 | #endif |