blob: 19d66cefd7d34beae89c617a10f195f2f097770f [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 = {
Eric Dumazetc4dbe542011-05-24 14:08:08 +020036 .lock = __SEQLOCK_UNLOCKED(sysctl_local_ports.lock),
Eric Dumazet3c689b72008-10-08 14:18:04 -070037 .range = { 32768, 61000 },
38};
Stephen Hemminger227b60f2007-10-10 17:30:46 -070039
Amerigo Wange3826f12010-05-05 00:27:06 +000040unsigned long *sysctl_local_reserved_ports;
41EXPORT_SYMBOL(sysctl_local_reserved_ports);
42
Stephen Hemminger227b60f2007-10-10 17:30:46 -070043void inet_get_local_port_range(int *low, int *high)
44{
45 unsigned seq;
46 do {
Eric Dumazet3c689b72008-10-08 14:18:04 -070047 seq = read_seqbegin(&sysctl_local_ports.lock);
Stephen Hemminger227b60f2007-10-10 17:30:46 -070048
Eric Dumazet3c689b72008-10-08 14:18:04 -070049 *low = sysctl_local_ports.range[0];
50 *high = sysctl_local_ports.range[1];
51 } while (read_seqretry(&sysctl_local_ports.lock, seq));
Stephen Hemminger227b60f2007-10-10 17:30:46 -070052}
53EXPORT_SYMBOL(inet_get_local_port_range);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070054
Arnaldo Carvalho de Melo971af182005-12-13 23:14:47 -080055int inet_csk_bind_conflict(const struct sock *sk,
56 const struct inet_bind_bucket *tb)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070057{
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070058 struct sock *sk2;
59 struct hlist_node *node;
60 int reuse = sk->sk_reuse;
61
Pavel Emelyanov7477fd22008-04-14 02:42:27 -070062 /*
63 * Unlike other sk lookup places we do not check
64 * for sk_net here, since _all_ the socks listed
65 * in tb->owners list belong to the same net - the
66 * one this bucket belongs to.
67 */
68
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070069 sk_for_each_bound(sk2, node, &tb->owners) {
70 if (sk != sk2 &&
71 !inet_v6_ipv6only(sk2) &&
72 (!sk->sk_bound_dev_if ||
73 !sk2->sk_bound_dev_if ||
74 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
75 if (!reuse || !sk2->sk_reuse ||
David S. Miller3e8c8062011-04-13 12:01:14 -070076 sk2->sk_state == TCP_LISTEN) {
Eric Dumazet68835ab2010-11-30 19:04:07 +000077 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
78 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
79 sk2_rcv_saddr == sk_rcv_saddr(sk))
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070080 break;
David S. Miller8d238b22010-04-28 11:25:59 -070081 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070082 }
83 }
84 return node != NULL;
85}
Arnaldo Carvalho de Melo971af182005-12-13 23:14:47 -080086EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
87
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070088/* Obtain a reference to a local port for the given sock,
89 * if snum is zero it means select any available local port.
90 */
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -080091int inet_csk_get_port(struct sock *sk, unsigned short snum)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070092{
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -070093 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070094 struct inet_bind_hashbucket *head;
95 struct hlist_node *node;
96 struct inet_bind_bucket *tb;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -080097 int ret, attempts = 5;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +090098 struct net *net = sock_net(sk);
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -080099 int smallest_size = -1, smallest_rover;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700100
101 local_bh_disable();
102 if (!snum) {
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700103 int remaining, rover, low, high;
104
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800105again:
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700106 inet_get_local_port_range(&low, &high);
Anton Arapova25de532007-10-18 22:00:17 -0700107 remaining = (high - low) + 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800108 smallest_rover = rover = net_random() % remaining + low;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700109
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800110 smallest_size = -1;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700111 do {
Amerigo Wange3826f12010-05-05 00:27:06 +0000112 if (inet_is_reserved_local_port(rover))
113 goto next_nolock;
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700114 head = &hashinfo->bhash[inet_bhashfn(net, rover,
115 hashinfo->bhash_size)];
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700116 spin_lock(&head->lock);
117 inet_bind_bucket_for_each(tb, node, &head->chain)
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800118 if (net_eq(ib_net(tb), net) && tb->port == rover) {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800119 if (tb->fastreuse > 0 &&
120 sk->sk_reuse &&
121 sk->sk_state != TCP_LISTEN &&
122 (tb->num_owners < smallest_size || smallest_size == -1)) {
123 smallest_size = tb->num_owners;
124 smallest_rover = rover;
David S. Miller3e8c8062011-04-13 12:01:14 -0700125 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) {
David S. Miller8d238b22010-04-28 11:25:59 -0700126 snum = smallest_rover;
Flavio Leitnerfddb7b52012-01-25 08:34:52 +0000127 goto tb_found;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800128 }
129 }
Flavio Leitner2b05ad32012-01-25 08:34:51 +0000130 if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
Flavio Leitner2b05ad32012-01-25 08:34:51 +0000131 snum = rover;
Flavio Leitnerfddb7b52012-01-25 08:34:52 +0000132 goto tb_found;
Flavio Leitner2b05ad32012-01-25 08:34:51 +0000133 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700134 goto next;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800135 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700136 break;
137 next:
138 spin_unlock(&head->lock);
Amerigo Wange3826f12010-05-05 00:27:06 +0000139 next_nolock:
Stephen Hemminger6df71632005-11-03 16:33:23 -0800140 if (++rover > high)
141 rover = low;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700142 } while (--remaining > 0);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700143
144 /* Exhausted local port range during search? It is not
145 * possible for us to be holding one of the bind hash
146 * locks if this test triggers, because if 'remaining'
147 * drops to zero, we broke out of the do/while loop at
148 * the top level, not from the 'break;' statement.
149 */
150 ret = 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800151 if (remaining <= 0) {
152 if (smallest_size != -1) {
153 snum = smallest_rover;
154 goto have_snum;
155 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700156 goto fail;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800157 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700158 /* OK, here is the one we will use. HEAD is
159 * non-NULL and we hold it's mutex.
160 */
161 snum = rover;
162 } else {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800163have_snum:
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700164 head = &hashinfo->bhash[inet_bhashfn(net, snum,
165 hashinfo->bhash_size)];
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700166 spin_lock(&head->lock);
167 inet_bind_bucket_for_each(tb, node, &head->chain)
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800168 if (net_eq(ib_net(tb), net) && tb->port == snum)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700169 goto tb_found;
170 }
171 tb = NULL;
172 goto tb_not_found;
173tb_found:
174 if (!hlist_empty(&tb->owners)) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700175 if (tb->fastreuse > 0 &&
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800176 sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
177 smallest_size == -1) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700178 goto success;
179 } else {
180 ret = 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800181 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
Stephen Hemminger5add3002009-02-01 01:40:17 -0800182 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
183 smallest_size != -1 && --attempts >= 0) {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800184 spin_unlock(&head->lock);
185 goto again;
186 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700187 goto fail_unlock;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800188 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700189 }
190 }
191tb_not_found:
192 ret = 1;
Pavel Emelyanov941b1d22008-01-31 05:05:50 -0800193 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
194 net, head, snum)) == NULL)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700195 goto fail_unlock;
196 if (hlist_empty(&tb->owners)) {
197 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
198 tb->fastreuse = 1;
199 else
200 tb->fastreuse = 0;
201 } else if (tb->fastreuse &&
202 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
203 tb->fastreuse = 0;
204success:
205 if (!inet_csk(sk)->icsk_bind_hash)
206 inet_bind_hash(sk, tb, snum);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700207 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900208 ret = 0;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700209
210fail_unlock:
211 spin_unlock(&head->lock);
212fail:
213 local_bh_enable();
214 return ret;
215}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700216EXPORT_SYMBOL_GPL(inet_csk_get_port);
217
218/*
219 * Wait for an incoming connection, avoid race conditions. This must be called
220 * with the socket locked.
221 */
222static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
223{
224 struct inet_connection_sock *icsk = inet_csk(sk);
225 DEFINE_WAIT(wait);
226 int err;
227
228 /*
229 * True wake-one mechanism for incoming connections: only
230 * one process gets woken up, not the 'whole herd'.
231 * Since we do not 'race & poll' for established sockets
232 * anymore, the common case will execute the loop only once.
233 *
234 * Subtle issue: "add_wait_queue_exclusive()" will be added
235 * after any current non-exclusive waiters, and we know that
236 * it will always _stay_ after any new non-exclusive waiters
237 * because all non-exclusive waiters are added at the
238 * beginning of the wait-queue. As such, it's ok to "drop"
239 * our exclusiveness temporarily when we get woken up without
240 * having to remove and re-insert us on the wait queue.
241 */
242 for (;;) {
Eric Dumazetaa395142010-04-20 13:03:51 +0000243 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700244 TASK_INTERRUPTIBLE);
245 release_sock(sk);
246 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
247 timeo = schedule_timeout(timeo);
248 lock_sock(sk);
249 err = 0;
250 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
251 break;
252 err = -EINVAL;
253 if (sk->sk_state != TCP_LISTEN)
254 break;
255 err = sock_intr_errno(timeo);
256 if (signal_pending(current))
257 break;
258 err = -EAGAIN;
259 if (!timeo)
260 break;
261 }
Eric Dumazetaa395142010-04-20 13:03:51 +0000262 finish_wait(sk_sleep(sk), &wait);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700263 return err;
264}
265
266/*
267 * This will accept the next outstanding connection.
268 */
269struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
270{
271 struct inet_connection_sock *icsk = inet_csk(sk);
272 struct sock *newsk;
273 int error;
274
275 lock_sock(sk);
276
277 /* We need to make sure that this socket is listening,
278 * and that it has something pending.
279 */
280 error = -EINVAL;
281 if (sk->sk_state != TCP_LISTEN)
282 goto out_err;
283
284 /* Find already established connection */
285 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
286 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
287
288 /* If this is a non blocking socket don't sleep */
289 error = -EAGAIN;
290 if (!timeo)
291 goto out_err;
292
293 error = inet_csk_wait_for_connect(sk, timeo);
294 if (error)
295 goto out_err;
296 }
297
298 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700299 WARN_ON(newsk->sk_state == TCP_SYN_RECV);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700300out:
301 release_sock(sk);
302 return newsk;
303out_err:
304 newsk = NULL;
305 *err = error;
306 goto out;
307}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700308EXPORT_SYMBOL(inet_csk_accept);
309
310/*
311 * Using different timers for retransmit, delayed acks and probes
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900312 * We may wish use just one timer maintaining a list of expire jiffies
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700313 * to optimize.
314 */
315void inet_csk_init_xmit_timers(struct sock *sk,
316 void (*retransmit_handler)(unsigned long),
317 void (*delack_handler)(unsigned long),
318 void (*keepalive_handler)(unsigned long))
319{
320 struct inet_connection_sock *icsk = inet_csk(sk);
321
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800322 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
323 (unsigned long)sk);
324 setup_timer(&icsk->icsk_delack_timer, delack_handler,
325 (unsigned long)sk);
326 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700327 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
328}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700329EXPORT_SYMBOL(inet_csk_init_xmit_timers);
330
331void inet_csk_clear_xmit_timers(struct sock *sk)
332{
333 struct inet_connection_sock *icsk = inet_csk(sk);
334
335 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
336
337 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
338 sk_stop_timer(sk, &icsk->icsk_delack_timer);
339 sk_stop_timer(sk, &sk->sk_timer);
340}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700341EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
342
343void inet_csk_delete_keepalive_timer(struct sock *sk)
344{
345 sk_stop_timer(sk, &sk->sk_timer);
346}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700347EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
348
349void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
350{
351 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
352}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700353EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
354
Jianjun Kongd9319102008-11-03 00:23:42 -0800355struct dst_entry *inet_csk_route_req(struct sock *sk,
David S. Miller6bd023f2011-05-18 18:32:03 -0400356 struct flowi4 *fl4,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700357 const struct request_sock *req)
358{
359 struct rtable *rt;
360 const struct inet_request_sock *ireq = inet_rsk(req);
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000361 struct ip_options_rcu *opt = inet_rsk(req)->opt;
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700362 struct net *net = sock_net(sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700363
David S. Miller6bd023f2011-05-18 18:32:03 -0400364 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
David S. Millere79d9bc2011-03-31 04:53:20 -0700365 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
366 sk->sk_protocol, inet_sk_flowi_flags(sk),
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000367 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
David S. Millere79d9bc2011-03-31 04:53:20 -0700368 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
David S. Miller6bd023f2011-05-18 18:32:03 -0400369 security_req_classify_flow(req, flowi4_to_flowi(fl4));
370 rt = ip_route_output_flow(net, fl4, sk);
David S. Millerb23dd4f2011-03-02 14:31:35 -0800371 if (IS_ERR(rt))
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800372 goto no_route;
David S. Miller6bd023f2011-05-18 18:32:03 -0400373 if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800374 goto route_err;
Changli Gaod8d1f302010-06-10 23:31:35 -0700375 return &rt->dst;
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800376
377route_err:
378 ip_rt_put(rt);
379no_route:
380 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
381 return NULL;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700382}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700383EXPORT_SYMBOL_GPL(inet_csk_route_req);
384
David S. Miller77357a92011-05-08 14:34:22 -0700385struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
386 struct sock *newsk,
387 const struct request_sock *req)
388{
389 const struct inet_request_sock *ireq = inet_rsk(req);
390 struct inet_sock *newinet = inet_sk(newsk);
391 struct ip_options_rcu *opt = ireq->opt;
392 struct net *net = sock_net(sk);
393 struct flowi4 *fl4;
394 struct rtable *rt;
395
396 fl4 = &newinet->cork.fl.u.ip4;
397 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
398 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
399 sk->sk_protocol, inet_sk_flowi_flags(sk),
400 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
401 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
402 security_req_classify_flow(req, flowi4_to_flowi(fl4));
403 rt = ip_route_output_flow(net, fl4, sk);
404 if (IS_ERR(rt))
405 goto no_route;
406 if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
407 goto route_err;
408 return &rt->dst;
409
410route_err:
411 ip_rt_put(rt);
412no_route:
413 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
414 return NULL;
415}
416EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
417
Al Viro6b729772006-09-27 18:36:59 -0700418static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
Eric Dumazet72a3eff2006-11-16 02:30:37 -0800419 const u32 rnd, const u32 synq_hsize)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700420{
Al Viro6b729772006-09-27 18:36:59 -0700421 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700422}
423
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000424#if IS_ENABLED(CONFIG_IPV6)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700425#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
426#else
427#define AF_INET_FAMILY(fam) 1
428#endif
429
430struct request_sock *inet_csk_search_req(const struct sock *sk,
431 struct request_sock ***prevp,
Al Viro6b729772006-09-27 18:36:59 -0700432 const __be16 rport, const __be32 raddr,
Al Viro7f25afb2006-09-27 18:27:47 -0700433 const __be32 laddr)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700434{
435 const struct inet_connection_sock *icsk = inet_csk(sk);
436 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
437 struct request_sock *req, **prev;
438
439 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
440 lopt->nr_table_entries)];
441 (req = *prev) != NULL;
442 prev = &req->dl_next) {
443 const struct inet_request_sock *ireq = inet_rsk(req);
444
445 if (ireq->rmt_port == rport &&
446 ireq->rmt_addr == raddr &&
447 ireq->loc_addr == laddr &&
448 AF_INET_FAMILY(req->rsk_ops->family)) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700449 WARN_ON(req->sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700450 *prevp = prev;
451 break;
452 }
453 }
454
455 return req;
456}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700457EXPORT_SYMBOL_GPL(inet_csk_search_req);
458
459void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
Arnaldo Carvalho de Meloc2977c22005-12-13 23:15:12 -0800460 unsigned long timeout)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700461{
462 struct inet_connection_sock *icsk = inet_csk(sk);
463 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
464 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
465 lopt->hash_rnd, lopt->nr_table_entries);
466
467 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
468 inet_csk_reqsk_queue_added(sk, timeout);
469}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000470EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700471
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700472/* Only thing we need from tcp.h */
473extern int sysctl_tcp_synack_retries;
474
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700475
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000476/* Decide when to expire the request and when to resend SYN-ACK */
477static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
478 const int max_retries,
479 const u8 rskq_defer_accept,
480 int *expire, int *resend)
481{
482 if (!rskq_defer_accept) {
483 *expire = req->retrans >= thresh;
484 *resend = 1;
485 return;
486 }
487 *expire = req->retrans >= thresh &&
488 (!inet_rsk(req)->acked || req->retrans >= max_retries);
489 /*
490 * Do not resend while waiting for data after ACK,
491 * start to resend on end of deferring period to give
492 * last chance for data or ACK to create established socket.
493 */
494 *resend = !inet_rsk(req)->acked ||
495 req->retrans >= rskq_defer_accept - 1;
496}
497
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700498void inet_csk_reqsk_queue_prune(struct sock *parent,
499 const unsigned long interval,
500 const unsigned long timeout,
501 const unsigned long max_rto)
502{
503 struct inet_connection_sock *icsk = inet_csk(parent);
504 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
505 struct listen_sock *lopt = queue->listen_opt;
David S. Millerec0a1962008-06-12 16:31:35 -0700506 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
507 int thresh = max_retries;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700508 unsigned long now = jiffies;
509 struct request_sock **reqp, *req;
510 int i, budget;
511
512 if (lopt == NULL || lopt->qlen == 0)
513 return;
514
515 /* Normally all the openreqs are young and become mature
516 * (i.e. converted to established socket) for first timeout.
517 * If synack was not acknowledged for 3 seconds, it means
518 * one of the following things: synack was lost, ack was lost,
519 * rtt is high or nobody planned to ack (i.e. synflood).
520 * When server is a bit loaded, queue is populated with old
521 * open requests, reducing effective size of queue.
522 * When server is well loaded, queue size reduces to zero
523 * after several minutes of work. It is not synflood,
524 * it is normal operation. The solution is pruning
525 * too old entries overriding normal timeout, when
526 * situation becomes dangerous.
527 *
528 * Essentially, we reserve half of room for young
529 * embrions; and abort old ones without pity, if old
530 * ones are about to clog our table.
531 */
532 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
533 int young = (lopt->qlen_young<<1);
534
535 while (thresh > 2) {
536 if (lopt->qlen < young)
537 break;
538 thresh--;
539 young <<= 1;
540 }
541 }
542
David S. Millerec0a1962008-06-12 16:31:35 -0700543 if (queue->rskq_defer_accept)
544 max_retries = queue->rskq_defer_accept;
545
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700546 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
547 i = lopt->clock_hand;
548
549 do {
550 reqp=&lopt->syn_table[i];
551 while ((req = *reqp) != NULL) {
552 if (time_after_eq(now, req->expires)) {
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000553 int expire = 0, resend = 0;
554
555 syn_ack_recalc(req, thresh, max_retries,
556 queue->rskq_defer_accept,
557 &expire, &resend);
Octavian Purdila72659ec2010-01-17 19:09:39 -0800558 if (req->rsk_ops->syn_ack_timeout)
559 req->rsk_ops->syn_ack_timeout(parent, req);
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000560 if (!expire &&
561 (!resend ||
William Allen Simpsone6b4d112009-12-02 18:07:39 +0000562 !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000563 inet_rsk(req)->acked)) {
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700564 unsigned long timeo;
565
566 if (req->retrans++ == 0)
567 lopt->qlen_young--;
568 timeo = min((timeout << req->retrans), max_rto);
569 req->expires = now + timeo;
570 reqp = &req->dl_next;
571 continue;
572 }
573
574 /* Drop this request */
575 inet_csk_reqsk_queue_unlink(parent, req, reqp);
576 reqsk_queue_removed(queue, req);
577 reqsk_free(req);
578 continue;
579 }
580 reqp = &req->dl_next;
581 }
582
583 i = (i + 1) & (lopt->nr_table_entries - 1);
584
585 } while (--budget > 0);
586
587 lopt->clock_hand = i;
588
589 if (lopt->qlen)
590 inet_csk_reset_keepalive_timer(parent, interval);
591}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700592EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
593
Eric Dumazete56c57d2011-11-08 17:07:07 -0500594/**
595 * inet_csk_clone_lock - clone an inet socket, and lock its clone
596 * @sk: the socket to clone
597 * @req: request_sock
598 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
599 *
600 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
601 */
602struct sock *inet_csk_clone_lock(const struct sock *sk,
603 const struct request_sock *req,
604 const gfp_t priority)
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700605{
Eric Dumazete56c57d2011-11-08 17:07:07 -0500606 struct sock *newsk = sk_clone_lock(sk, priority);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700607
608 if (newsk != NULL) {
609 struct inet_connection_sock *newicsk = inet_csk(newsk);
610
611 newsk->sk_state = TCP_SYN_RECV;
612 newicsk->icsk_bind_hash = NULL;
613
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000614 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
615 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
616 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700617 newsk->sk_write_space = sk_stream_write_space;
618
619 newicsk->icsk_retransmits = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300620 newicsk->icsk_backoff = 0;
621 newicsk->icsk_probes_out = 0;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700622
623 /* Deinitialize accept_queue to trap illegal accesses. */
624 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700625
626 security_inet_csk_clone(newsk, req);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700627 }
628 return newsk;
629}
Eric Dumazete56c57d2011-11-08 17:07:07 -0500630EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700631
632/*
633 * At this point, there should be no process reference to this
634 * socket, and thus no user references at all. Therefore we
635 * can assume the socket waitqueue is inactive and nobody will
636 * try to jump onto it.
637 */
638void inet_csk_destroy_sock(struct sock *sk)
639{
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700640 WARN_ON(sk->sk_state != TCP_CLOSE);
641 WARN_ON(!sock_flag(sk, SOCK_DEAD));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700642
643 /* It cannot be in hash table! */
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700644 WARN_ON(!sk_unhashed(sk));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700645
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000646 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
647 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700648
649 sk->sk_prot->destroy(sk);
650
651 sk_stream_kill_queues(sk);
652
653 xfrm_sk_free_policy(sk);
654
655 sk_refcnt_debug_release(sk);
656
Eric Dumazetdd24c002008-11-25 21:17:14 -0800657 percpu_counter_dec(sk->sk_prot->orphan_count);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700658 sock_put(sk);
659}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700660EXPORT_SYMBOL(inet_csk_destroy_sock);
661
662int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
663{
664 struct inet_sock *inet = inet_sk(sk);
665 struct inet_connection_sock *icsk = inet_csk(sk);
666 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
667
668 if (rc != 0)
669 return rc;
670
671 sk->sk_max_ack_backlog = 0;
672 sk->sk_ack_backlog = 0;
673 inet_csk_delack_init(sk);
674
675 /* There is race window here: we announce ourselves listening,
676 * but this transition is still not validated by get_port().
677 * It is OK, because this socket enters to hash table only
678 * after validation is complete.
679 */
680 sk->sk_state = TCP_LISTEN;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000681 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
682 inet->inet_sport = htons(inet->inet_num);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700683
684 sk_dst_reset(sk);
685 sk->sk_prot->hash(sk);
686
687 return 0;
688 }
689
690 sk->sk_state = TCP_CLOSE;
691 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
692 return -EADDRINUSE;
693}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700694EXPORT_SYMBOL_GPL(inet_csk_listen_start);
695
696/*
697 * This routine closes sockets which have been at least partially
698 * opened, but not yet accepted.
699 */
700void inet_csk_listen_stop(struct sock *sk)
701{
702 struct inet_connection_sock *icsk = inet_csk(sk);
703 struct request_sock *acc_req;
704 struct request_sock *req;
705
706 inet_csk_delete_keepalive_timer(sk);
707
708 /* make all the listen_opt local to us */
709 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
710
711 /* Following specs, it would be better either to send FIN
712 * (and enter FIN-WAIT-1, it is normal close)
713 * or to send active reset (abort).
714 * Certainly, it is pretty dangerous while synflood, but it is
715 * bad justification for our negligence 8)
716 * To be honest, we are not able to make either
717 * of the variants now. --ANK
718 */
719 reqsk_queue_destroy(&icsk->icsk_accept_queue);
720
721 while ((req = acc_req) != NULL) {
722 struct sock *child = req->sk;
723
724 acc_req = req->dl_next;
725
726 local_bh_disable();
727 bh_lock_sock(child);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700728 WARN_ON(sock_owned_by_user(child));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700729 sock_hold(child);
730
731 sk->sk_prot->disconnect(child, O_NONBLOCK);
732
733 sock_orphan(child);
734
Herbert Xueb4dea52008-12-29 23:04:08 -0800735 percpu_counter_inc(sk->sk_prot->orphan_count);
736
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700737 inet_csk_destroy_sock(child);
738
739 bh_unlock_sock(child);
740 local_bh_enable();
741 sock_put(child);
742
743 sk_acceptq_removed(sk);
744 __reqsk_free(req);
745 }
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700746 WARN_ON(sk->sk_ack_backlog);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700747}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700748EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800749
750void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
751{
752 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
753 const struct inet_sock *inet = inet_sk(sk);
754
755 sin->sin_family = AF_INET;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000756 sin->sin_addr.s_addr = inet->inet_daddr;
757 sin->sin_port = inet->inet_dport;
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800758}
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800759EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
Arnaldo Carvalho de Meloc4d93902006-03-20 22:01:03 -0800760
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800761#ifdef CONFIG_COMPAT
762int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
763 char __user *optval, int __user *optlen)
764{
David S. Millerdbeff122006-03-20 22:52:32 -0800765 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800766
767 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
768 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
769 optval, optlen);
770 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
771 optval, optlen);
772}
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800773EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
774
775int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -0700776 char __user *optval, unsigned int optlen)
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800777{
David S. Millerdbeff122006-03-20 22:52:32 -0800778 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800779
780 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
781 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
782 optval, optlen);
783 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
784 optval, optlen);
785}
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800786EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
787#endif