blob: f26ab38680de00a5b77c303e34774a078022d499 [file] [log] [blame]
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -07001/*
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
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070016#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>
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -070025#include <net/xfrm.h>
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070026
27#ifdef INET_CSK_DEBUG
28const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
29EXPORT_SYMBOL(inet_csk_timer_bug_msg);
30#endif
31
32/*
Eric Dumazet3c689b72008-10-08 14:18:04 -070033 * This struct holds the first and last local port number.
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070034 */
Eric Dumazet3c689b72008-10-08 14:18:04 -070035struct local_ports sysctl_local_ports __read_mostly = {
36 .lock = SEQLOCK_UNLOCKED,
37 .range = { 32768, 61000 },
38};
Stephen Hemminger227b60f2007-10-10 17:30:46 -070039
40void inet_get_local_port_range(int *low, int *high)
41{
42 unsigned seq;
43 do {
Eric Dumazet3c689b72008-10-08 14:18:04 -070044 seq = read_seqbegin(&sysctl_local_ports.lock);
Stephen Hemminger227b60f2007-10-10 17:30:46 -070045
Eric Dumazet3c689b72008-10-08 14:18:04 -070046 *low = sysctl_local_ports.range[0];
47 *high = sysctl_local_ports.range[1];
48 } while (read_seqretry(&sysctl_local_ports.lock, seq));
Stephen Hemminger227b60f2007-10-10 17:30:46 -070049}
50EXPORT_SYMBOL(inet_get_local_port_range);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070051
Arnaldo Carvalho de Melo971af182005-12-13 23:14:47 -080052int inet_csk_bind_conflict(const struct sock *sk,
53 const struct inet_bind_bucket *tb)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070054{
Al Viro82103232006-09-27 18:44:10 -070055 const __be32 sk_rcv_saddr = inet_rcv_saddr(sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070056 struct sock *sk2;
57 struct hlist_node *node;
58 int reuse = sk->sk_reuse;
59
Pavel Emelyanov7477fd22008-04-14 02:42:27 -070060 /*
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
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070067 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) {
Al Viro82103232006-09-27 18:44:10 -070075 const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070076 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
Arnaldo Carvalho de Melo971af182005-12-13 23:14:47 -080085EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
86
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070087/* Obtain a reference to a local port for the given sock,
88 * if snum is zero it means select any available local port.
89 */
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -080090int inet_csk_get_port(struct sock *sk, unsigned short snum)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070091{
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -070092 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070093 struct inet_bind_hashbucket *head;
94 struct hlist_node *node;
95 struct inet_bind_bucket *tb;
96 int ret;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +090097 struct net *net = sock_net(sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070098
99 local_bh_disable();
100 if (!snum) {
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700101 int remaining, rover, low, high;
102
103 inet_get_local_port_range(&low, &high);
Anton Arapova25de532007-10-18 22:00:17 -0700104 remaining = (high - low) + 1;
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700105 rover = net_random() % remaining + low;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700106
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700107 do {
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700108 head = &hashinfo->bhash[inet_bhashfn(net, rover,
109 hashinfo->bhash_size)];
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700110 spin_lock(&head->lock);
111 inet_bind_bucket_for_each(tb, node, &head->chain)
Eric Dumazet7a9546e2008-11-12 00:54:20 -0800112 if (ib_net(tb) == net && tb->port == rover)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700113 goto next;
114 break;
115 next:
116 spin_unlock(&head->lock);
Stephen Hemminger6df71632005-11-03 16:33:23 -0800117 if (++rover > high)
118 rover = low;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700119 } while (--remaining > 0);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700120
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 {
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700136 head = &hashinfo->bhash[inet_bhashfn(net, snum,
137 hashinfo->bhash_size)];
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700138 spin_lock(&head->lock);
139 inet_bind_bucket_for_each(tb, node, &head->chain)
Eric Dumazet7a9546e2008-11-12 00:54:20 -0800140 if (ib_net(tb) == net && tb->port == snum)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700141 goto tb_found;
142 }
143 tb = NULL;
144 goto tb_not_found;
145tb_found:
146 if (!hlist_empty(&tb->owners)) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700147 if (tb->fastreuse > 0 &&
148 sk->sk_reuse && sk->sk_state != TCP_LISTEN) {
149 goto success;
150 } else {
151 ret = 1;
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -0800152 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb))
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700153 goto fail_unlock;
154 }
155 }
156tb_not_found:
157 ret = 1;
Pavel Emelyanov941b1d22008-01-31 05:05:50 -0800158 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
159 net, head, snum)) == NULL)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700160 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;
169success:
170 if (!inet_csk(sk)->icsk_bind_hash)
171 inet_bind_hash(sk, tb, snum);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700172 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900173 ret = 0;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700174
175fail_unlock:
176 spin_unlock(&head->lock);
177fail:
178 local_bh_enable();
179 return ret;
180}
181
182EXPORT_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 */
188static 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 */
235struct 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);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700265 WARN_ON(newsk->sk_state == TCP_SYN_RECV);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700266out:
267 release_sock(sk);
268 return newsk;
269out_err:
270 newsk = NULL;
271 *err = error;
272 goto out;
273}
274
275EXPORT_SYMBOL(inet_csk_accept);
276
277/*
278 * Using different timers for retransmit, delayed acks and probes
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900279 * We may wish use just one timer maintaining a list of expire jiffies
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700280 * to optimize.
281 */
282void 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
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800289 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);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700294 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
295}
296
297EXPORT_SYMBOL(inet_csk_init_xmit_timers);
298
299void 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
310EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
311
312void inet_csk_delete_keepalive_timer(struct sock *sk)
313{
314 sk_stop_timer(sk, &sk->sk_timer);
315}
316
317EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
318
319void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
320{
321 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
322}
323
324EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
325
Jianjun Kongd9319102008-11-03 00:23:42 -0800326struct dst_entry *inet_csk_route_req(struct sock *sk,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700327 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,
KOVACS Krisztian86b08d82008-10-01 07:44:42 -0700340 .flags = inet_sk_flowi_flags(sk),
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700341 .uli_u = { .ports =
342 { .sport = inet_sk(sk)->sport,
343 .dport = ireq->rmt_port } } };
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700344 struct net *net = sock_net(sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700345
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700346 security_req_classify_flow(req, &fl);
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800347 if (ip_route_output_flow(net, &rt, &fl, sk, 0))
348 goto no_route;
349 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
350 goto route_err;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700351 return &rt->u.dst;
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800352
353route_err:
354 ip_rt_put(rt);
355no_route:
356 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
357 return NULL;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700358}
359
360EXPORT_SYMBOL_GPL(inet_csk_route_req);
361
Al Viro6b729772006-09-27 18:36:59 -0700362static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
Eric Dumazet72a3eff2006-11-16 02:30:37 -0800363 const u32 rnd, const u32 synq_hsize)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700364{
Al Viro6b729772006-09-27 18:36:59 -0700365 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700366}
367
368#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
369#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
370#else
371#define AF_INET_FAMILY(fam) 1
372#endif
373
374struct request_sock *inet_csk_search_req(const struct sock *sk,
375 struct request_sock ***prevp,
Al Viro6b729772006-09-27 18:36:59 -0700376 const __be16 rport, const __be32 raddr,
Al Viro7f25afb2006-09-27 18:27:47 -0700377 const __be32 laddr)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700378{
379 const struct inet_connection_sock *icsk = inet_csk(sk);
380 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
381 struct request_sock *req, **prev;
382
383 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
384 lopt->nr_table_entries)];
385 (req = *prev) != NULL;
386 prev = &req->dl_next) {
387 const struct inet_request_sock *ireq = inet_rsk(req);
388
389 if (ireq->rmt_port == rport &&
390 ireq->rmt_addr == raddr &&
391 ireq->loc_addr == laddr &&
392 AF_INET_FAMILY(req->rsk_ops->family)) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700393 WARN_ON(req->sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700394 *prevp = prev;
395 break;
396 }
397 }
398
399 return req;
400}
401
402EXPORT_SYMBOL_GPL(inet_csk_search_req);
403
404void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
Arnaldo Carvalho de Meloc2977c22005-12-13 23:15:12 -0800405 unsigned long timeout)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700406{
407 struct inet_connection_sock *icsk = inet_csk(sk);
408 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
409 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
410 lopt->hash_rnd, lopt->nr_table_entries);
411
412 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
413 inet_csk_reqsk_queue_added(sk, timeout);
414}
415
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700416/* Only thing we need from tcp.h */
417extern int sysctl_tcp_synack_retries;
418
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700419EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700420
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700421void inet_csk_reqsk_queue_prune(struct sock *parent,
422 const unsigned long interval,
423 const unsigned long timeout,
424 const unsigned long max_rto)
425{
426 struct inet_connection_sock *icsk = inet_csk(parent);
427 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
428 struct listen_sock *lopt = queue->listen_opt;
David S. Millerec0a1962008-06-12 16:31:35 -0700429 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
430 int thresh = max_retries;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700431 unsigned long now = jiffies;
432 struct request_sock **reqp, *req;
433 int i, budget;
434
435 if (lopt == NULL || lopt->qlen == 0)
436 return;
437
438 /* Normally all the openreqs are young and become mature
439 * (i.e. converted to established socket) for first timeout.
440 * If synack was not acknowledged for 3 seconds, it means
441 * one of the following things: synack was lost, ack was lost,
442 * rtt is high or nobody planned to ack (i.e. synflood).
443 * When server is a bit loaded, queue is populated with old
444 * open requests, reducing effective size of queue.
445 * When server is well loaded, queue size reduces to zero
446 * after several minutes of work. It is not synflood,
447 * it is normal operation. The solution is pruning
448 * too old entries overriding normal timeout, when
449 * situation becomes dangerous.
450 *
451 * Essentially, we reserve half of room for young
452 * embrions; and abort old ones without pity, if old
453 * ones are about to clog our table.
454 */
455 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
456 int young = (lopt->qlen_young<<1);
457
458 while (thresh > 2) {
459 if (lopt->qlen < young)
460 break;
461 thresh--;
462 young <<= 1;
463 }
464 }
465
David S. Millerec0a1962008-06-12 16:31:35 -0700466 if (queue->rskq_defer_accept)
467 max_retries = queue->rskq_defer_accept;
468
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700469 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
470 i = lopt->clock_hand;
471
472 do {
473 reqp=&lopt->syn_table[i];
474 while ((req = *reqp) != NULL) {
475 if (time_after_eq(now, req->expires)) {
David S. Miller93653e02008-06-16 16:57:40 -0700476 if ((req->retrans < thresh ||
477 (inet_rsk(req)->acked && req->retrans < max_retries))
478 && !req->rsk_ops->rtx_syn_ack(parent, req)) {
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700479 unsigned long timeo;
480
481 if (req->retrans++ == 0)
482 lopt->qlen_young--;
483 timeo = min((timeout << req->retrans), max_rto);
484 req->expires = now + timeo;
485 reqp = &req->dl_next;
486 continue;
487 }
488
489 /* Drop this request */
490 inet_csk_reqsk_queue_unlink(parent, req, reqp);
491 reqsk_queue_removed(queue, req);
492 reqsk_free(req);
493 continue;
494 }
495 reqp = &req->dl_next;
496 }
497
498 i = (i + 1) & (lopt->nr_table_entries - 1);
499
500 } while (--budget > 0);
501
502 lopt->clock_hand = i;
503
504 if (lopt->qlen)
505 inet_csk_reset_keepalive_timer(parent, interval);
506}
507
508EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
509
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700510struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
Al Virodd0fc662005-10-07 07:46:04 +0100511 const gfp_t priority)
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700512{
513 struct sock *newsk = sk_clone(sk, priority);
514
515 if (newsk != NULL) {
516 struct inet_connection_sock *newicsk = inet_csk(newsk);
517
518 newsk->sk_state = TCP_SYN_RECV;
519 newicsk->icsk_bind_hash = NULL;
520
521 inet_sk(newsk)->dport = inet_rsk(req)->rmt_port;
KOVACS Krisztiana3116ac2008-10-01 07:46:49 -0700522 inet_sk(newsk)->num = ntohs(inet_rsk(req)->loc_port);
523 inet_sk(newsk)->sport = inet_rsk(req)->loc_port;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700524 newsk->sk_write_space = sk_stream_write_space;
525
526 newicsk->icsk_retransmits = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300527 newicsk->icsk_backoff = 0;
528 newicsk->icsk_probes_out = 0;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700529
530 /* Deinitialize accept_queue to trap illegal accesses. */
531 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700532
533 security_inet_csk_clone(newsk, req);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700534 }
535 return newsk;
536}
537
538EXPORT_SYMBOL_GPL(inet_csk_clone);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700539
540/*
541 * At this point, there should be no process reference to this
542 * socket, and thus no user references at all. Therefore we
543 * can assume the socket waitqueue is inactive and nobody will
544 * try to jump onto it.
545 */
546void inet_csk_destroy_sock(struct sock *sk)
547{
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700548 WARN_ON(sk->sk_state != TCP_CLOSE);
549 WARN_ON(!sock_flag(sk, SOCK_DEAD));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700550
551 /* It cannot be in hash table! */
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700552 WARN_ON(!sk_unhashed(sk));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700553
554 /* If it has not 0 inet_sk(sk)->num, it must be bound */
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700555 WARN_ON(inet_sk(sk)->num && !inet_csk(sk)->icsk_bind_hash);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700556
557 sk->sk_prot->destroy(sk);
558
559 sk_stream_kill_queues(sk);
560
561 xfrm_sk_free_policy(sk);
562
563 sk_refcnt_debug_release(sk);
564
Eric Dumazetdd24c002008-11-25 21:17:14 -0800565 percpu_counter_dec(sk->sk_prot->orphan_count);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700566 sock_put(sk);
567}
568
569EXPORT_SYMBOL(inet_csk_destroy_sock);
570
571int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
572{
573 struct inet_sock *inet = inet_sk(sk);
574 struct inet_connection_sock *icsk = inet_csk(sk);
575 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
576
577 if (rc != 0)
578 return rc;
579
580 sk->sk_max_ack_backlog = 0;
581 sk->sk_ack_backlog = 0;
582 inet_csk_delack_init(sk);
583
584 /* There is race window here: we announce ourselves listening,
585 * but this transition is still not validated by get_port().
586 * It is OK, because this socket enters to hash table only
587 * after validation is complete.
588 */
589 sk->sk_state = TCP_LISTEN;
590 if (!sk->sk_prot->get_port(sk, inet->num)) {
591 inet->sport = htons(inet->num);
592
593 sk_dst_reset(sk);
594 sk->sk_prot->hash(sk);
595
596 return 0;
597 }
598
599 sk->sk_state = TCP_CLOSE;
600 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
601 return -EADDRINUSE;
602}
603
604EXPORT_SYMBOL_GPL(inet_csk_listen_start);
605
606/*
607 * This routine closes sockets which have been at least partially
608 * opened, but not yet accepted.
609 */
610void inet_csk_listen_stop(struct sock *sk)
611{
612 struct inet_connection_sock *icsk = inet_csk(sk);
613 struct request_sock *acc_req;
614 struct request_sock *req;
615
616 inet_csk_delete_keepalive_timer(sk);
617
618 /* make all the listen_opt local to us */
619 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
620
621 /* Following specs, it would be better either to send FIN
622 * (and enter FIN-WAIT-1, it is normal close)
623 * or to send active reset (abort).
624 * Certainly, it is pretty dangerous while synflood, but it is
625 * bad justification for our negligence 8)
626 * To be honest, we are not able to make either
627 * of the variants now. --ANK
628 */
629 reqsk_queue_destroy(&icsk->icsk_accept_queue);
630
631 while ((req = acc_req) != NULL) {
632 struct sock *child = req->sk;
633
634 acc_req = req->dl_next;
635
636 local_bh_disable();
637 bh_lock_sock(child);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700638 WARN_ON(sock_owned_by_user(child));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700639 sock_hold(child);
640
641 sk->sk_prot->disconnect(child, O_NONBLOCK);
642
643 sock_orphan(child);
644
Herbert Xueb4dea52008-12-29 23:04:08 -0800645 percpu_counter_inc(sk->sk_prot->orphan_count);
646
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700647 inet_csk_destroy_sock(child);
648
649 bh_unlock_sock(child);
650 local_bh_enable();
651 sock_put(child);
652
653 sk_acceptq_removed(sk);
654 __reqsk_free(req);
655 }
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700656 WARN_ON(sk->sk_ack_backlog);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700657}
658
659EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800660
661void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
662{
663 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
664 const struct inet_sock *inet = inet_sk(sk);
665
666 sin->sin_family = AF_INET;
667 sin->sin_addr.s_addr = inet->daddr;
668 sin->sin_port = inet->dport;
669}
670
671EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
Arnaldo Carvalho de Meloc4d93902006-03-20 22:01:03 -0800672
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800673#ifdef CONFIG_COMPAT
674int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
675 char __user *optval, int __user *optlen)
676{
David S. Millerdbeff122006-03-20 22:52:32 -0800677 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800678
679 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
680 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
681 optval, optlen);
682 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
683 optval, optlen);
684}
685
686EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
687
688int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
689 char __user *optval, int optlen)
690{
David S. Millerdbeff122006-03-20 22:52:32 -0800691 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800692
693 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
694 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
695 optval, optlen);
696 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
697 optval, optlen);
698}
699
700EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
701#endif