blob: 70eb3507c40645a028d0ecd0421911f10f0a541d [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
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{
Al Viro82103232006-09-27 18:44:10 -070058 const __be32 sk_rcv_saddr = inet_rcv_saddr(sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070059 struct sock *sk2;
60 struct hlist_node *node;
61 int reuse = sk->sk_reuse;
62
Pavel Emelyanov7477fd2e2008-04-14 02:42:27 -070063 /*
64 * Unlike other sk lookup places we do not check
65 * for sk_net here, since _all_ the socks listed
66 * in tb->owners list belong to the same net - the
67 * one this bucket belongs to.
68 */
69
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070070 sk_for_each_bound(sk2, node, &tb->owners) {
71 if (sk != sk2 &&
72 !inet_v6_ipv6only(sk2) &&
73 (!sk->sk_bound_dev_if ||
74 !sk2->sk_bound_dev_if ||
75 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
76 if (!reuse || !sk2->sk_reuse ||
77 sk2->sk_state == TCP_LISTEN) {
David S. Miller8d238b22010-04-28 11:25:59 -070078 const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070079 if (!sk2_rcv_saddr || !sk_rcv_saddr ||
80 sk2_rcv_saddr == sk_rcv_saddr)
81 break;
David S. Miller8d238b22010-04-28 11:25:59 -070082 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070083 }
84 }
85 return node != NULL;
86}
87
Arnaldo Carvalho de Melo971af182005-12-13 23:14:47 -080088EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
89
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070090/* Obtain a reference to a local port for the given sock,
91 * if snum is zero it means select any available local port.
92 */
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -080093int inet_csk_get_port(struct sock *sk, unsigned short snum)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070094{
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -070095 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070096 struct inet_bind_hashbucket *head;
97 struct hlist_node *node;
98 struct inet_bind_bucket *tb;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -080099 int ret, attempts = 5;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900100 struct net *net = sock_net(sk);
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800101 int smallest_size = -1, smallest_rover;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700102
103 local_bh_disable();
104 if (!snum) {
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700105 int remaining, rover, low, high;
106
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800107again:
Stephen Hemminger227b60f2007-10-10 17:30:46 -0700108 inet_get_local_port_range(&low, &high);
Anton Arapova25de532007-10-18 22:00:17 -0700109 remaining = (high - low) + 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800110 smallest_rover = rover = net_random() % remaining + low;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700111
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800112 smallest_size = -1;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700113 do {
Amerigo Wange3826f12010-05-05 00:27:06 +0000114 if (inet_is_reserved_local_port(rover))
115 goto next_nolock;
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700116 head = &hashinfo->bhash[inet_bhashfn(net, rover,
117 hashinfo->bhash_size)];
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700118 spin_lock(&head->lock);
119 inet_bind_bucket_for_each(tb, node, &head->chain)
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800120 if (net_eq(ib_net(tb), net) && tb->port == rover) {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800121 if (tb->fastreuse > 0 &&
122 sk->sk_reuse &&
123 sk->sk_state != TCP_LISTEN &&
124 (tb->num_owners < smallest_size || smallest_size == -1)) {
125 smallest_size = tb->num_owners;
126 smallest_rover = rover;
Eric Dumazet24dd1fa2009-02-01 12:31:33 -0800127 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) {
David S. Miller8d238b22010-04-28 11:25:59 -0700128 spin_unlock(&head->lock);
129 snum = smallest_rover;
130 goto have_snum;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800131 }
132 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700133 goto next;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800134 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700135 break;
136 next:
137 spin_unlock(&head->lock);
Amerigo Wange3826f12010-05-05 00:27:06 +0000138 next_nolock:
Stephen Hemminger6df71632005-11-03 16:33:23 -0800139 if (++rover > high)
140 rover = low;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700141 } while (--remaining > 0);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700142
143 /* Exhausted local port range during search? It is not
144 * possible for us to be holding one of the bind hash
145 * locks if this test triggers, because if 'remaining'
146 * drops to zero, we broke out of the do/while loop at
147 * the top level, not from the 'break;' statement.
148 */
149 ret = 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800150 if (remaining <= 0) {
151 if (smallest_size != -1) {
152 snum = smallest_rover;
153 goto have_snum;
154 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700155 goto fail;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800156 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700157 /* OK, here is the one we will use. HEAD is
158 * non-NULL and we hold it's mutex.
159 */
160 snum = rover;
161 } else {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800162have_snum:
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700163 head = &hashinfo->bhash[inet_bhashfn(net, snum,
164 hashinfo->bhash_size)];
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700165 spin_lock(&head->lock);
166 inet_bind_bucket_for_each(tb, node, &head->chain)
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800167 if (net_eq(ib_net(tb), net) && tb->port == snum)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700168 goto tb_found;
169 }
170 tb = NULL;
171 goto tb_not_found;
172tb_found:
173 if (!hlist_empty(&tb->owners)) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700174 if (tb->fastreuse > 0 &&
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800175 sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
176 smallest_size == -1) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700177 goto success;
178 } else {
179 ret = 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800180 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
Stephen Hemminger5add3002009-02-01 01:40:17 -0800181 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
182 smallest_size != -1 && --attempts >= 0) {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800183 spin_unlock(&head->lock);
184 goto again;
185 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700186 goto fail_unlock;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800187 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700188 }
189 }
190tb_not_found:
191 ret = 1;
Pavel Emelyanov941b1d22008-01-31 05:05:50 -0800192 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
193 net, head, snum)) == NULL)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700194 goto fail_unlock;
195 if (hlist_empty(&tb->owners)) {
196 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
197 tb->fastreuse = 1;
198 else
199 tb->fastreuse = 0;
200 } else if (tb->fastreuse &&
201 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
202 tb->fastreuse = 0;
203success:
204 if (!inet_csk(sk)->icsk_bind_hash)
205 inet_bind_hash(sk, tb, snum);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700206 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900207 ret = 0;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700208
209fail_unlock:
210 spin_unlock(&head->lock);
211fail:
212 local_bh_enable();
213 return ret;
214}
215
216EXPORT_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}
308
309EXPORT_SYMBOL(inet_csk_accept);
310
311/*
312 * Using different timers for retransmit, delayed acks and probes
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900313 * We may wish use just one timer maintaining a list of expire jiffies
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700314 * to optimize.
315 */
316void inet_csk_init_xmit_timers(struct sock *sk,
317 void (*retransmit_handler)(unsigned long),
318 void (*delack_handler)(unsigned long),
319 void (*keepalive_handler)(unsigned long))
320{
321 struct inet_connection_sock *icsk = inet_csk(sk);
322
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800323 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
324 (unsigned long)sk);
325 setup_timer(&icsk->icsk_delack_timer, delack_handler,
326 (unsigned long)sk);
327 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700328 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
329}
330
331EXPORT_SYMBOL(inet_csk_init_xmit_timers);
332
333void inet_csk_clear_xmit_timers(struct sock *sk)
334{
335 struct inet_connection_sock *icsk = inet_csk(sk);
336
337 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
338
339 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
340 sk_stop_timer(sk, &icsk->icsk_delack_timer);
341 sk_stop_timer(sk, &sk->sk_timer);
342}
343
344EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
345
346void inet_csk_delete_keepalive_timer(struct sock *sk)
347{
348 sk_stop_timer(sk, &sk->sk_timer);
349}
350
351EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
352
353void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
354{
355 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
356}
357
358EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
359
Jianjun Kongd93191002008-11-03 00:23:42 -0800360struct dst_entry *inet_csk_route_req(struct sock *sk,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700361 const struct request_sock *req)
362{
363 struct rtable *rt;
364 const struct inet_request_sock *ireq = inet_rsk(req);
365 struct ip_options *opt = inet_rsk(req)->opt;
366 struct flowi fl = { .oif = sk->sk_bound_dev_if,
Atis Elstsffce9082009-10-07 13:55:57 -0700367 .mark = sk->sk_mark,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700368 .nl_u = { .ip4_u =
369 { .daddr = ((opt && opt->srr) ?
370 opt->faddr :
371 ireq->rmt_addr),
372 .saddr = ireq->loc_addr,
373 .tos = RT_CONN_FLAGS(sk) } },
374 .proto = sk->sk_protocol,
KOVACS Krisztian86b08d82008-10-01 07:44:42 -0700375 .flags = inet_sk_flowi_flags(sk),
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700376 .uli_u = { .ports =
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000377 { .sport = inet_sk(sk)->inet_sport,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700378 .dport = ireq->rmt_port } } };
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700379 struct net *net = sock_net(sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700380
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700381 security_req_classify_flow(req, &fl);
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800382 if (ip_route_output_flow(net, &rt, &fl, sk, 0))
383 goto no_route;
384 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
385 goto route_err;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700386 return &rt->u.dst;
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800387
388route_err:
389 ip_rt_put(rt);
390no_route:
391 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
392 return NULL;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700393}
394
395EXPORT_SYMBOL_GPL(inet_csk_route_req);
396
Al Viro6b729772006-09-27 18:36:59 -0700397static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
Eric Dumazet72a3eff2006-11-16 02:30:37 -0800398 const u32 rnd, const u32 synq_hsize)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700399{
Al Viro6b729772006-09-27 18:36:59 -0700400 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700401}
402
403#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
404#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
405#else
406#define AF_INET_FAMILY(fam) 1
407#endif
408
409struct request_sock *inet_csk_search_req(const struct sock *sk,
410 struct request_sock ***prevp,
Al Viro6b729772006-09-27 18:36:59 -0700411 const __be16 rport, const __be32 raddr,
Al Viro7f25afb2006-09-27 18:27:47 -0700412 const __be32 laddr)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700413{
414 const struct inet_connection_sock *icsk = inet_csk(sk);
415 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
416 struct request_sock *req, **prev;
417
418 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
419 lopt->nr_table_entries)];
420 (req = *prev) != NULL;
421 prev = &req->dl_next) {
422 const struct inet_request_sock *ireq = inet_rsk(req);
423
424 if (ireq->rmt_port == rport &&
425 ireq->rmt_addr == raddr &&
426 ireq->loc_addr == laddr &&
427 AF_INET_FAMILY(req->rsk_ops->family)) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700428 WARN_ON(req->sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700429 *prevp = prev;
430 break;
431 }
432 }
433
434 return req;
435}
436
437EXPORT_SYMBOL_GPL(inet_csk_search_req);
438
439void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
Arnaldo Carvalho de Meloc2977c22005-12-13 23:15:12 -0800440 unsigned long timeout)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700441{
442 struct inet_connection_sock *icsk = inet_csk(sk);
443 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
444 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
445 lopt->hash_rnd, lopt->nr_table_entries);
446
447 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
448 inet_csk_reqsk_queue_added(sk, timeout);
449}
450
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700451/* Only thing we need from tcp.h */
452extern int sysctl_tcp_synack_retries;
453
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700454EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700455
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000456/* Decide when to expire the request and when to resend SYN-ACK */
457static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
458 const int max_retries,
459 const u8 rskq_defer_accept,
460 int *expire, int *resend)
461{
462 if (!rskq_defer_accept) {
463 *expire = req->retrans >= thresh;
464 *resend = 1;
465 return;
466 }
467 *expire = req->retrans >= thresh &&
468 (!inet_rsk(req)->acked || req->retrans >= max_retries);
469 /*
470 * Do not resend while waiting for data after ACK,
471 * start to resend on end of deferring period to give
472 * last chance for data or ACK to create established socket.
473 */
474 *resend = !inet_rsk(req)->acked ||
475 req->retrans >= rskq_defer_accept - 1;
476}
477
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700478void inet_csk_reqsk_queue_prune(struct sock *parent,
479 const unsigned long interval,
480 const unsigned long timeout,
481 const unsigned long max_rto)
482{
483 struct inet_connection_sock *icsk = inet_csk(parent);
484 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
485 struct listen_sock *lopt = queue->listen_opt;
David S. Millerec0a1962008-06-12 16:31:35 -0700486 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
487 int thresh = max_retries;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700488 unsigned long now = jiffies;
489 struct request_sock **reqp, *req;
490 int i, budget;
491
492 if (lopt == NULL || lopt->qlen == 0)
493 return;
494
495 /* Normally all the openreqs are young and become mature
496 * (i.e. converted to established socket) for first timeout.
497 * If synack was not acknowledged for 3 seconds, it means
498 * one of the following things: synack was lost, ack was lost,
499 * rtt is high or nobody planned to ack (i.e. synflood).
500 * When server is a bit loaded, queue is populated with old
501 * open requests, reducing effective size of queue.
502 * When server is well loaded, queue size reduces to zero
503 * after several minutes of work. It is not synflood,
504 * it is normal operation. The solution is pruning
505 * too old entries overriding normal timeout, when
506 * situation becomes dangerous.
507 *
508 * Essentially, we reserve half of room for young
509 * embrions; and abort old ones without pity, if old
510 * ones are about to clog our table.
511 */
512 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
513 int young = (lopt->qlen_young<<1);
514
515 while (thresh > 2) {
516 if (lopt->qlen < young)
517 break;
518 thresh--;
519 young <<= 1;
520 }
521 }
522
David S. Millerec0a1962008-06-12 16:31:35 -0700523 if (queue->rskq_defer_accept)
524 max_retries = queue->rskq_defer_accept;
525
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700526 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
527 i = lopt->clock_hand;
528
529 do {
530 reqp=&lopt->syn_table[i];
531 while ((req = *reqp) != NULL) {
532 if (time_after_eq(now, req->expires)) {
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000533 int expire = 0, resend = 0;
534
535 syn_ack_recalc(req, thresh, max_retries,
536 queue->rskq_defer_accept,
537 &expire, &resend);
Octavian Purdila72659ec2010-01-17 19:09:39 -0800538 if (req->rsk_ops->syn_ack_timeout)
539 req->rsk_ops->syn_ack_timeout(parent, req);
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000540 if (!expire &&
541 (!resend ||
William Allen Simpsone6b4d112009-12-02 18:07:39 +0000542 !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000543 inet_rsk(req)->acked)) {
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700544 unsigned long timeo;
545
546 if (req->retrans++ == 0)
547 lopt->qlen_young--;
548 timeo = min((timeout << req->retrans), max_rto);
549 req->expires = now + timeo;
550 reqp = &req->dl_next;
551 continue;
552 }
553
554 /* Drop this request */
555 inet_csk_reqsk_queue_unlink(parent, req, reqp);
556 reqsk_queue_removed(queue, req);
557 reqsk_free(req);
558 continue;
559 }
560 reqp = &req->dl_next;
561 }
562
563 i = (i + 1) & (lopt->nr_table_entries - 1);
564
565 } while (--budget > 0);
566
567 lopt->clock_hand = i;
568
569 if (lopt->qlen)
570 inet_csk_reset_keepalive_timer(parent, interval);
571}
572
573EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
574
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700575struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
Al Virodd0fc662005-10-07 07:46:04 +0100576 const gfp_t priority)
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700577{
578 struct sock *newsk = sk_clone(sk, priority);
579
580 if (newsk != NULL) {
581 struct inet_connection_sock *newicsk = inet_csk(newsk);
582
583 newsk->sk_state = TCP_SYN_RECV;
584 newicsk->icsk_bind_hash = NULL;
585
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000586 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
587 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
588 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700589 newsk->sk_write_space = sk_stream_write_space;
590
591 newicsk->icsk_retransmits = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300592 newicsk->icsk_backoff = 0;
593 newicsk->icsk_probes_out = 0;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700594
595 /* Deinitialize accept_queue to trap illegal accesses. */
596 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700597
598 security_inet_csk_clone(newsk, req);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700599 }
600 return newsk;
601}
602
603EXPORT_SYMBOL_GPL(inet_csk_clone);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700604
605/*
606 * At this point, there should be no process reference to this
607 * socket, and thus no user references at all. Therefore we
608 * can assume the socket waitqueue is inactive and nobody will
609 * try to jump onto it.
610 */
611void inet_csk_destroy_sock(struct sock *sk)
612{
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700613 WARN_ON(sk->sk_state != TCP_CLOSE);
614 WARN_ON(!sock_flag(sk, SOCK_DEAD));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700615
616 /* It cannot be in hash table! */
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700617 WARN_ON(!sk_unhashed(sk));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700618
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000619 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
620 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700621
622 sk->sk_prot->destroy(sk);
623
624 sk_stream_kill_queues(sk);
625
626 xfrm_sk_free_policy(sk);
627
628 sk_refcnt_debug_release(sk);
629
Eric Dumazetdd24c002008-11-25 21:17:14 -0800630 percpu_counter_dec(sk->sk_prot->orphan_count);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700631 sock_put(sk);
632}
633
634EXPORT_SYMBOL(inet_csk_destroy_sock);
635
636int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
637{
638 struct inet_sock *inet = inet_sk(sk);
639 struct inet_connection_sock *icsk = inet_csk(sk);
640 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
641
642 if (rc != 0)
643 return rc;
644
645 sk->sk_max_ack_backlog = 0;
646 sk->sk_ack_backlog = 0;
647 inet_csk_delack_init(sk);
648
649 /* There is race window here: we announce ourselves listening,
650 * but this transition is still not validated by get_port().
651 * It is OK, because this socket enters to hash table only
652 * after validation is complete.
653 */
654 sk->sk_state = TCP_LISTEN;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000655 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
656 inet->inet_sport = htons(inet->inet_num);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700657
658 sk_dst_reset(sk);
659 sk->sk_prot->hash(sk);
660
661 return 0;
662 }
663
664 sk->sk_state = TCP_CLOSE;
665 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
666 return -EADDRINUSE;
667}
668
669EXPORT_SYMBOL_GPL(inet_csk_listen_start);
670
671/*
672 * This routine closes sockets which have been at least partially
673 * opened, but not yet accepted.
674 */
675void inet_csk_listen_stop(struct sock *sk)
676{
677 struct inet_connection_sock *icsk = inet_csk(sk);
678 struct request_sock *acc_req;
679 struct request_sock *req;
680
681 inet_csk_delete_keepalive_timer(sk);
682
683 /* make all the listen_opt local to us */
684 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
685
686 /* Following specs, it would be better either to send FIN
687 * (and enter FIN-WAIT-1, it is normal close)
688 * or to send active reset (abort).
689 * Certainly, it is pretty dangerous while synflood, but it is
690 * bad justification for our negligence 8)
691 * To be honest, we are not able to make either
692 * of the variants now. --ANK
693 */
694 reqsk_queue_destroy(&icsk->icsk_accept_queue);
695
696 while ((req = acc_req) != NULL) {
697 struct sock *child = req->sk;
698
699 acc_req = req->dl_next;
700
701 local_bh_disable();
702 bh_lock_sock(child);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700703 WARN_ON(sock_owned_by_user(child));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700704 sock_hold(child);
705
706 sk->sk_prot->disconnect(child, O_NONBLOCK);
707
708 sock_orphan(child);
709
Herbert Xueb4dea52008-12-29 23:04:08 -0800710 percpu_counter_inc(sk->sk_prot->orphan_count);
711
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700712 inet_csk_destroy_sock(child);
713
714 bh_unlock_sock(child);
715 local_bh_enable();
716 sock_put(child);
717
718 sk_acceptq_removed(sk);
719 __reqsk_free(req);
720 }
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700721 WARN_ON(sk->sk_ack_backlog);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700722}
723
724EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800725
726void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
727{
728 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
729 const struct inet_sock *inet = inet_sk(sk);
730
731 sin->sin_family = AF_INET;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000732 sin->sin_addr.s_addr = inet->inet_daddr;
733 sin->sin_port = inet->inet_dport;
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800734}
735
736EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
Arnaldo Carvalho de Meloc4d93902006-03-20 22:01:03 -0800737
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800738#ifdef CONFIG_COMPAT
739int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
740 char __user *optval, int __user *optlen)
741{
David S. Millerdbeff122006-03-20 22:52:32 -0800742 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800743
744 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
745 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
746 optval, optlen);
747 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
748 optval, optlen);
749}
750
751EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
752
753int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -0700754 char __user *optval, unsigned int optlen)
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800755{
David S. Millerdbeff122006-03-20 22:52:32 -0800756 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800757
758 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
759 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
760 optval, optlen);
761 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
762 optval, optlen);
763}
764
765EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
766#endif