blob: 8da6429269dddd922f9c3fb4eda2b86a7bffe0d2 [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 Emelyanov7477fd2e2008-04-14 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) {
David S. Miller8d238b22010-04-28 11:25:59 -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;
David S. Miller8d238b22010-04-28 11:25:59 -070079 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070080 }
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;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -080096 int ret, attempts = 5;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +090097 struct net *net = sock_net(sk);
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -080098 int smallest_size = -1, smallest_rover;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070099
100 local_bh_disable();
101 if (!snum) {
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700102 int remaining, rover, low, high;
103
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800104again:
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700105 inet_get_local_port_range(&low, &high);
Anton Arapova25de532007-10-18 22:00:17 -0700106 remaining = (high - low) + 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800107 smallest_rover = rover = net_random() % remaining + low;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700108
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800109 smallest_size = -1;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700110 do {
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700111 head = &hashinfo->bhash[inet_bhashfn(net, rover,
112 hashinfo->bhash_size)];
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700113 spin_lock(&head->lock);
114 inet_bind_bucket_for_each(tb, node, &head->chain)
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800115 if (net_eq(ib_net(tb), net) && tb->port == rover) {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800116 if (tb->fastreuse > 0 &&
117 sk->sk_reuse &&
118 sk->sk_state != TCP_LISTEN &&
119 (tb->num_owners < smallest_size || smallest_size == -1)) {
120 smallest_size = tb->num_owners;
121 smallest_rover = rover;
Eric Dumazet24dd1fa2009-02-01 12:31:33 -0800122 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) {
David S. Miller8d238b22010-04-28 11:25:59 -0700123 spin_unlock(&head->lock);
124 snum = smallest_rover;
125 goto have_snum;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800126 }
127 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700128 goto next;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800129 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700130 break;
131 next:
132 spin_unlock(&head->lock);
Stephen Hemminger6df71632005-11-03 16:33:23 -0800133 if (++rover > high)
134 rover = low;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700135 } while (--remaining > 0);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700136
137 /* Exhausted local port range during search? It is not
138 * possible for us to be holding one of the bind hash
139 * locks if this test triggers, because if 'remaining'
140 * drops to zero, we broke out of the do/while loop at
141 * the top level, not from the 'break;' statement.
142 */
143 ret = 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800144 if (remaining <= 0) {
145 if (smallest_size != -1) {
146 snum = smallest_rover;
147 goto have_snum;
148 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700149 goto fail;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800150 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700151 /* OK, here is the one we will use. HEAD is
152 * non-NULL and we hold it's mutex.
153 */
154 snum = rover;
155 } else {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800156have_snum:
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700157 head = &hashinfo->bhash[inet_bhashfn(net, snum,
158 hashinfo->bhash_size)];
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700159 spin_lock(&head->lock);
160 inet_bind_bucket_for_each(tb, node, &head->chain)
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800161 if (net_eq(ib_net(tb), net) && tb->port == snum)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700162 goto tb_found;
163 }
164 tb = NULL;
165 goto tb_not_found;
166tb_found:
167 if (!hlist_empty(&tb->owners)) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700168 if (tb->fastreuse > 0 &&
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800169 sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
170 smallest_size == -1) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700171 goto success;
172 } else {
173 ret = 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800174 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
Stephen Hemminger5add3002009-02-01 01:40:17 -0800175 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
176 smallest_size != -1 && --attempts >= 0) {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800177 spin_unlock(&head->lock);
178 goto again;
179 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700180 goto fail_unlock;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800181 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700182 }
183 }
184tb_not_found:
185 ret = 1;
Pavel Emelyanov941b1d22008-01-31 05:05:50 -0800186 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
187 net, head, snum)) == NULL)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700188 goto fail_unlock;
189 if (hlist_empty(&tb->owners)) {
190 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
191 tb->fastreuse = 1;
192 else
193 tb->fastreuse = 0;
194 } else if (tb->fastreuse &&
195 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
196 tb->fastreuse = 0;
197success:
198 if (!inet_csk(sk)->icsk_bind_hash)
199 inet_bind_hash(sk, tb, snum);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700200 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900201 ret = 0;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700202
203fail_unlock:
204 spin_unlock(&head->lock);
205fail:
206 local_bh_enable();
207 return ret;
208}
209
210EXPORT_SYMBOL_GPL(inet_csk_get_port);
211
212/*
213 * Wait for an incoming connection, avoid race conditions. This must be called
214 * with the socket locked.
215 */
216static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
217{
218 struct inet_connection_sock *icsk = inet_csk(sk);
219 DEFINE_WAIT(wait);
220 int err;
221
222 /*
223 * True wake-one mechanism for incoming connections: only
224 * one process gets woken up, not the 'whole herd'.
225 * Since we do not 'race & poll' for established sockets
226 * anymore, the common case will execute the loop only once.
227 *
228 * Subtle issue: "add_wait_queue_exclusive()" will be added
229 * after any current non-exclusive waiters, and we know that
230 * it will always _stay_ after any new non-exclusive waiters
231 * because all non-exclusive waiters are added at the
232 * beginning of the wait-queue. As such, it's ok to "drop"
233 * our exclusiveness temporarily when we get woken up without
234 * having to remove and re-insert us on the wait queue.
235 */
236 for (;;) {
237 prepare_to_wait_exclusive(sk->sk_sleep, &wait,
238 TASK_INTERRUPTIBLE);
239 release_sock(sk);
240 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
241 timeo = schedule_timeout(timeo);
242 lock_sock(sk);
243 err = 0;
244 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
245 break;
246 err = -EINVAL;
247 if (sk->sk_state != TCP_LISTEN)
248 break;
249 err = sock_intr_errno(timeo);
250 if (signal_pending(current))
251 break;
252 err = -EAGAIN;
253 if (!timeo)
254 break;
255 }
256 finish_wait(sk->sk_sleep, &wait);
257 return err;
258}
259
260/*
261 * This will accept the next outstanding connection.
262 */
263struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
264{
265 struct inet_connection_sock *icsk = inet_csk(sk);
266 struct sock *newsk;
267 int error;
268
269 lock_sock(sk);
270
271 /* We need to make sure that this socket is listening,
272 * and that it has something pending.
273 */
274 error = -EINVAL;
275 if (sk->sk_state != TCP_LISTEN)
276 goto out_err;
277
278 /* Find already established connection */
279 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
280 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
281
282 /* If this is a non blocking socket don't sleep */
283 error = -EAGAIN;
284 if (!timeo)
285 goto out_err;
286
287 error = inet_csk_wait_for_connect(sk, timeo);
288 if (error)
289 goto out_err;
290 }
291
292 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700293 WARN_ON(newsk->sk_state == TCP_SYN_RECV);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700294out:
295 release_sock(sk);
296 return newsk;
297out_err:
298 newsk = NULL;
299 *err = error;
300 goto out;
301}
302
303EXPORT_SYMBOL(inet_csk_accept);
304
305/*
306 * Using different timers for retransmit, delayed acks and probes
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900307 * We may wish use just one timer maintaining a list of expire jiffies
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700308 * to optimize.
309 */
310void inet_csk_init_xmit_timers(struct sock *sk,
311 void (*retransmit_handler)(unsigned long),
312 void (*delack_handler)(unsigned long),
313 void (*keepalive_handler)(unsigned long))
314{
315 struct inet_connection_sock *icsk = inet_csk(sk);
316
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800317 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
318 (unsigned long)sk);
319 setup_timer(&icsk->icsk_delack_timer, delack_handler,
320 (unsigned long)sk);
321 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700322 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
323}
324
325EXPORT_SYMBOL(inet_csk_init_xmit_timers);
326
327void inet_csk_clear_xmit_timers(struct sock *sk)
328{
329 struct inet_connection_sock *icsk = inet_csk(sk);
330
331 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
332
333 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
334 sk_stop_timer(sk, &icsk->icsk_delack_timer);
335 sk_stop_timer(sk, &sk->sk_timer);
336}
337
338EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
339
340void inet_csk_delete_keepalive_timer(struct sock *sk)
341{
342 sk_stop_timer(sk, &sk->sk_timer);
343}
344
345EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
346
347void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
348{
349 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
350}
351
352EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
353
Jianjun Kongd93191002008-11-03 00:23:42 -0800354struct dst_entry *inet_csk_route_req(struct sock *sk,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700355 const struct request_sock *req)
356{
357 struct rtable *rt;
358 const struct inet_request_sock *ireq = inet_rsk(req);
359 struct ip_options *opt = inet_rsk(req)->opt;
360 struct flowi fl = { .oif = sk->sk_bound_dev_if,
Atis Elstsffce9082009-10-07 13:55:57 -0700361 .mark = sk->sk_mark,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700362 .nl_u = { .ip4_u =
363 { .daddr = ((opt && opt->srr) ?
364 opt->faddr :
365 ireq->rmt_addr),
366 .saddr = ireq->loc_addr,
367 .tos = RT_CONN_FLAGS(sk) } },
368 .proto = sk->sk_protocol,
KOVACS Krisztian86b08d82008-10-01 07:44:42 -0700369 .flags = inet_sk_flowi_flags(sk),
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700370 .uli_u = { .ports =
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000371 { .sport = inet_sk(sk)->inet_sport,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700372 .dport = ireq->rmt_port } } };
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700373 struct net *net = sock_net(sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700374
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700375 security_req_classify_flow(req, &fl);
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800376 if (ip_route_output_flow(net, &rt, &fl, sk, 0))
377 goto no_route;
378 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
379 goto route_err;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700380 return &rt->u.dst;
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800381
382route_err:
383 ip_rt_put(rt);
384no_route:
385 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
386 return NULL;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700387}
388
389EXPORT_SYMBOL_GPL(inet_csk_route_req);
390
Al Viro6b729772006-09-27 18:36:59 -0700391static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
Eric Dumazet72a3eff2006-11-16 02:30:37 -0800392 const u32 rnd, const u32 synq_hsize)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700393{
Al Viro6b729772006-09-27 18:36:59 -0700394 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700395}
396
397#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
398#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
399#else
400#define AF_INET_FAMILY(fam) 1
401#endif
402
403struct request_sock *inet_csk_search_req(const struct sock *sk,
404 struct request_sock ***prevp,
Al Viro6b729772006-09-27 18:36:59 -0700405 const __be16 rport, const __be32 raddr,
Al Viro7f25afb2006-09-27 18:27:47 -0700406 const __be32 laddr)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700407{
408 const struct inet_connection_sock *icsk = inet_csk(sk);
409 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
410 struct request_sock *req, **prev;
411
412 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
413 lopt->nr_table_entries)];
414 (req = *prev) != NULL;
415 prev = &req->dl_next) {
416 const struct inet_request_sock *ireq = inet_rsk(req);
417
418 if (ireq->rmt_port == rport &&
419 ireq->rmt_addr == raddr &&
420 ireq->loc_addr == laddr &&
421 AF_INET_FAMILY(req->rsk_ops->family)) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700422 WARN_ON(req->sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700423 *prevp = prev;
424 break;
425 }
426 }
427
428 return req;
429}
430
431EXPORT_SYMBOL_GPL(inet_csk_search_req);
432
433void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
Arnaldo Carvalho de Meloc2977c22005-12-13 23:15:12 -0800434 unsigned long timeout)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700435{
436 struct inet_connection_sock *icsk = inet_csk(sk);
437 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
438 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
439 lopt->hash_rnd, lopt->nr_table_entries);
440
441 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
442 inet_csk_reqsk_queue_added(sk, timeout);
443}
444
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700445/* Only thing we need from tcp.h */
446extern int sysctl_tcp_synack_retries;
447
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700448EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700449
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000450/* Decide when to expire the request and when to resend SYN-ACK */
451static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
452 const int max_retries,
453 const u8 rskq_defer_accept,
454 int *expire, int *resend)
455{
456 if (!rskq_defer_accept) {
457 *expire = req->retrans >= thresh;
458 *resend = 1;
459 return;
460 }
461 *expire = req->retrans >= thresh &&
462 (!inet_rsk(req)->acked || req->retrans >= max_retries);
463 /*
464 * Do not resend while waiting for data after ACK,
465 * start to resend on end of deferring period to give
466 * last chance for data or ACK to create established socket.
467 */
468 *resend = !inet_rsk(req)->acked ||
469 req->retrans >= rskq_defer_accept - 1;
470}
471
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700472void inet_csk_reqsk_queue_prune(struct sock *parent,
473 const unsigned long interval,
474 const unsigned long timeout,
475 const unsigned long max_rto)
476{
477 struct inet_connection_sock *icsk = inet_csk(parent);
478 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
479 struct listen_sock *lopt = queue->listen_opt;
David S. Millerec0a1962008-06-12 16:31:35 -0700480 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
481 int thresh = max_retries;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700482 unsigned long now = jiffies;
483 struct request_sock **reqp, *req;
484 int i, budget;
485
486 if (lopt == NULL || lopt->qlen == 0)
487 return;
488
489 /* Normally all the openreqs are young and become mature
490 * (i.e. converted to established socket) for first timeout.
491 * If synack was not acknowledged for 3 seconds, it means
492 * one of the following things: synack was lost, ack was lost,
493 * rtt is high or nobody planned to ack (i.e. synflood).
494 * When server is a bit loaded, queue is populated with old
495 * open requests, reducing effective size of queue.
496 * When server is well loaded, queue size reduces to zero
497 * after several minutes of work. It is not synflood,
498 * it is normal operation. The solution is pruning
499 * too old entries overriding normal timeout, when
500 * situation becomes dangerous.
501 *
502 * Essentially, we reserve half of room for young
503 * embrions; and abort old ones without pity, if old
504 * ones are about to clog our table.
505 */
506 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
507 int young = (lopt->qlen_young<<1);
508
509 while (thresh > 2) {
510 if (lopt->qlen < young)
511 break;
512 thresh--;
513 young <<= 1;
514 }
515 }
516
David S. Millerec0a1962008-06-12 16:31:35 -0700517 if (queue->rskq_defer_accept)
518 max_retries = queue->rskq_defer_accept;
519
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700520 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
521 i = lopt->clock_hand;
522
523 do {
524 reqp=&lopt->syn_table[i];
525 while ((req = *reqp) != NULL) {
526 if (time_after_eq(now, req->expires)) {
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000527 int expire = 0, resend = 0;
528
529 syn_ack_recalc(req, thresh, max_retries,
530 queue->rskq_defer_accept,
531 &expire, &resend);
Octavian Purdila72659ec2010-01-17 19:09:39 -0800532 if (req->rsk_ops->syn_ack_timeout)
533 req->rsk_ops->syn_ack_timeout(parent, req);
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000534 if (!expire &&
535 (!resend ||
William Allen Simpsone6b4d112009-12-02 18:07:39 +0000536 !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000537 inet_rsk(req)->acked)) {
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700538 unsigned long timeo;
539
540 if (req->retrans++ == 0)
541 lopt->qlen_young--;
542 timeo = min((timeout << req->retrans), max_rto);
543 req->expires = now + timeo;
544 reqp = &req->dl_next;
545 continue;
546 }
547
548 /* Drop this request */
549 inet_csk_reqsk_queue_unlink(parent, req, reqp);
550 reqsk_queue_removed(queue, req);
551 reqsk_free(req);
552 continue;
553 }
554 reqp = &req->dl_next;
555 }
556
557 i = (i + 1) & (lopt->nr_table_entries - 1);
558
559 } while (--budget > 0);
560
561 lopt->clock_hand = i;
562
563 if (lopt->qlen)
564 inet_csk_reset_keepalive_timer(parent, interval);
565}
566
567EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
568
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700569struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
Al Virodd0fc662005-10-07 07:46:04 +0100570 const gfp_t priority)
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700571{
572 struct sock *newsk = sk_clone(sk, priority);
573
574 if (newsk != NULL) {
575 struct inet_connection_sock *newicsk = inet_csk(newsk);
576
577 newsk->sk_state = TCP_SYN_RECV;
578 newicsk->icsk_bind_hash = NULL;
579
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000580 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
581 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
582 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700583 newsk->sk_write_space = sk_stream_write_space;
584
585 newicsk->icsk_retransmits = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300586 newicsk->icsk_backoff = 0;
587 newicsk->icsk_probes_out = 0;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700588
589 /* Deinitialize accept_queue to trap illegal accesses. */
590 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700591
592 security_inet_csk_clone(newsk, req);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700593 }
594 return newsk;
595}
596
597EXPORT_SYMBOL_GPL(inet_csk_clone);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700598
599/*
600 * At this point, there should be no process reference to this
601 * socket, and thus no user references at all. Therefore we
602 * can assume the socket waitqueue is inactive and nobody will
603 * try to jump onto it.
604 */
605void inet_csk_destroy_sock(struct sock *sk)
606{
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700607 WARN_ON(sk->sk_state != TCP_CLOSE);
608 WARN_ON(!sock_flag(sk, SOCK_DEAD));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700609
610 /* It cannot be in hash table! */
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700611 WARN_ON(!sk_unhashed(sk));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700612
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000613 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
614 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700615
616 sk->sk_prot->destroy(sk);
617
618 sk_stream_kill_queues(sk);
619
620 xfrm_sk_free_policy(sk);
621
622 sk_refcnt_debug_release(sk);
623
Eric Dumazetdd24c002008-11-25 21:17:14 -0800624 percpu_counter_dec(sk->sk_prot->orphan_count);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700625 sock_put(sk);
626}
627
628EXPORT_SYMBOL(inet_csk_destroy_sock);
629
630int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
631{
632 struct inet_sock *inet = inet_sk(sk);
633 struct inet_connection_sock *icsk = inet_csk(sk);
634 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
635
636 if (rc != 0)
637 return rc;
638
639 sk->sk_max_ack_backlog = 0;
640 sk->sk_ack_backlog = 0;
641 inet_csk_delack_init(sk);
642
643 /* There is race window here: we announce ourselves listening,
644 * but this transition is still not validated by get_port().
645 * It is OK, because this socket enters to hash table only
646 * after validation is complete.
647 */
648 sk->sk_state = TCP_LISTEN;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000649 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
650 inet->inet_sport = htons(inet->inet_num);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700651
652 sk_dst_reset(sk);
653 sk->sk_prot->hash(sk);
654
655 return 0;
656 }
657
658 sk->sk_state = TCP_CLOSE;
659 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
660 return -EADDRINUSE;
661}
662
663EXPORT_SYMBOL_GPL(inet_csk_listen_start);
664
665/*
666 * This routine closes sockets which have been at least partially
667 * opened, but not yet accepted.
668 */
669void inet_csk_listen_stop(struct sock *sk)
670{
671 struct inet_connection_sock *icsk = inet_csk(sk);
672 struct request_sock *acc_req;
673 struct request_sock *req;
674
675 inet_csk_delete_keepalive_timer(sk);
676
677 /* make all the listen_opt local to us */
678 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
679
680 /* Following specs, it would be better either to send FIN
681 * (and enter FIN-WAIT-1, it is normal close)
682 * or to send active reset (abort).
683 * Certainly, it is pretty dangerous while synflood, but it is
684 * bad justification for our negligence 8)
685 * To be honest, we are not able to make either
686 * of the variants now. --ANK
687 */
688 reqsk_queue_destroy(&icsk->icsk_accept_queue);
689
690 while ((req = acc_req) != NULL) {
691 struct sock *child = req->sk;
692
693 acc_req = req->dl_next;
694
695 local_bh_disable();
696 bh_lock_sock(child);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700697 WARN_ON(sock_owned_by_user(child));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700698 sock_hold(child);
699
700 sk->sk_prot->disconnect(child, O_NONBLOCK);
701
702 sock_orphan(child);
703
Herbert Xueb4dea52008-12-29 23:04:08 -0800704 percpu_counter_inc(sk->sk_prot->orphan_count);
705
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700706 inet_csk_destroy_sock(child);
707
708 bh_unlock_sock(child);
709 local_bh_enable();
710 sock_put(child);
711
712 sk_acceptq_removed(sk);
713 __reqsk_free(req);
714 }
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700715 WARN_ON(sk->sk_ack_backlog);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700716}
717
718EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800719
720void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
721{
722 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
723 const struct inet_sock *inet = inet_sk(sk);
724
725 sin->sin_family = AF_INET;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000726 sin->sin_addr.s_addr = inet->inet_daddr;
727 sin->sin_port = inet->inet_dport;
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800728}
729
730EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
Arnaldo Carvalho de Meloc4d93902006-03-20 22:01:03 -0800731
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800732#ifdef CONFIG_COMPAT
733int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
734 char __user *optval, int __user *optlen)
735{
David S. Millerdbeff122006-03-20 22:52:32 -0800736 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800737
738 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
739 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
740 optval, optlen);
741 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
742 optval, optlen);
743}
744
745EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
746
747int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -0700748 char __user *optval, unsigned int optlen)
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800749{
David S. Millerdbeff122006-03-20 22:52:32 -0800750 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800751
752 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
753 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
754 optval, optlen);
755 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
756 optval, optlen);
757}
758
759EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
760#endif