blob: 97e5fb76526500d6355b921be6359680bf7a8720 [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{
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 Emelyanov7477fd2e2008-04-14 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 ||
Eric Dumazetc191a832011-01-11 01:14:22 +000076 ((1 << sk2->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))) {
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;
Eric Dumazetc191a832011-01-11 01:14:22 +0000125 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&
126 !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
David S. Miller8d238b22010-04-28 11:25:59 -0700127 spin_unlock(&head->lock);
128 snum = smallest_rover;
129 goto have_snum;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800130 }
131 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700132 goto next;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800133 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700134 break;
135 next:
136 spin_unlock(&head->lock);
Amerigo Wange3826f12010-05-05 00:27:06 +0000137 next_nolock:
Stephen Hemminger6df71632005-11-03 16:33:23 -0800138 if (++rover > high)
139 rover = low;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700140 } while (--remaining > 0);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700141
142 /* Exhausted local port range during search? It is not
143 * possible for us to be holding one of the bind hash
144 * locks if this test triggers, because if 'remaining'
145 * drops to zero, we broke out of the do/while loop at
146 * the top level, not from the 'break;' statement.
147 */
148 ret = 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800149 if (remaining <= 0) {
150 if (smallest_size != -1) {
151 snum = smallest_rover;
152 goto have_snum;
153 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700154 goto fail;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800155 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700156 /* OK, here is the one we will use. HEAD is
157 * non-NULL and we hold it's mutex.
158 */
159 snum = rover;
160 } else {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800161have_snum:
Pavel Emelyanov7f635ab2008-06-16 17:12:49 -0700162 head = &hashinfo->bhash[inet_bhashfn(net, snum,
163 hashinfo->bhash_size)];
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700164 spin_lock(&head->lock);
165 inet_bind_bucket_for_each(tb, node, &head->chain)
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800166 if (net_eq(ib_net(tb), net) && tb->port == snum)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700167 goto tb_found;
168 }
169 tb = NULL;
170 goto tb_not_found;
171tb_found:
172 if (!hlist_empty(&tb->owners)) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700173 if (tb->fastreuse > 0 &&
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800174 sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
175 smallest_size == -1) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700176 goto success;
177 } else {
178 ret = 1;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800179 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
Stephen Hemminger5add3002009-02-01 01:40:17 -0800180 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
181 smallest_size != -1 && --attempts >= 0) {
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800182 spin_unlock(&head->lock);
183 goto again;
184 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700185 goto fail_unlock;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800186 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700187 }
188 }
189tb_not_found:
190 ret = 1;
Pavel Emelyanov941b1d22008-01-31 05:05:50 -0800191 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
192 net, head, snum)) == NULL)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700193 goto fail_unlock;
194 if (hlist_empty(&tb->owners)) {
195 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
196 tb->fastreuse = 1;
197 else
198 tb->fastreuse = 0;
199 } else if (tb->fastreuse &&
200 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
201 tb->fastreuse = 0;
202success:
203 if (!inet_csk(sk)->icsk_bind_hash)
204 inet_bind_hash(sk, tb, snum);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700205 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900206 ret = 0;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700207
208fail_unlock:
209 spin_unlock(&head->lock);
210fail:
211 local_bh_enable();
212 return ret;
213}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700214EXPORT_SYMBOL_GPL(inet_csk_get_port);
215
216/*
217 * Wait for an incoming connection, avoid race conditions. This must be called
218 * with the socket locked.
219 */
220static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
221{
222 struct inet_connection_sock *icsk = inet_csk(sk);
223 DEFINE_WAIT(wait);
224 int err;
225
226 /*
227 * True wake-one mechanism for incoming connections: only
228 * one process gets woken up, not the 'whole herd'.
229 * Since we do not 'race & poll' for established sockets
230 * anymore, the common case will execute the loop only once.
231 *
232 * Subtle issue: "add_wait_queue_exclusive()" will be added
233 * after any current non-exclusive waiters, and we know that
234 * it will always _stay_ after any new non-exclusive waiters
235 * because all non-exclusive waiters are added at the
236 * beginning of the wait-queue. As such, it's ok to "drop"
237 * our exclusiveness temporarily when we get woken up without
238 * having to remove and re-insert us on the wait queue.
239 */
240 for (;;) {
Eric Dumazetaa395142010-04-20 13:03:51 +0000241 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700242 TASK_INTERRUPTIBLE);
243 release_sock(sk);
244 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
245 timeo = schedule_timeout(timeo);
246 lock_sock(sk);
247 err = 0;
248 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
249 break;
250 err = -EINVAL;
251 if (sk->sk_state != TCP_LISTEN)
252 break;
253 err = sock_intr_errno(timeo);
254 if (signal_pending(current))
255 break;
256 err = -EAGAIN;
257 if (!timeo)
258 break;
259 }
Eric Dumazetaa395142010-04-20 13:03:51 +0000260 finish_wait(sk_sleep(sk), &wait);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700261 return err;
262}
263
264/*
265 * This will accept the next outstanding connection.
266 */
267struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
268{
269 struct inet_connection_sock *icsk = inet_csk(sk);
270 struct sock *newsk;
271 int error;
272
273 lock_sock(sk);
274
275 /* We need to make sure that this socket is listening,
276 * and that it has something pending.
277 */
278 error = -EINVAL;
279 if (sk->sk_state != TCP_LISTEN)
280 goto out_err;
281
282 /* Find already established connection */
283 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
284 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
285
286 /* If this is a non blocking socket don't sleep */
287 error = -EAGAIN;
288 if (!timeo)
289 goto out_err;
290
291 error = inet_csk_wait_for_connect(sk, timeo);
292 if (error)
293 goto out_err;
294 }
295
296 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700297 WARN_ON(newsk->sk_state == TCP_SYN_RECV);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700298out:
299 release_sock(sk);
300 return newsk;
301out_err:
302 newsk = NULL;
303 *err = error;
304 goto out;
305}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700306EXPORT_SYMBOL(inet_csk_accept);
307
308/*
309 * Using different timers for retransmit, delayed acks and probes
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900310 * We may wish use just one timer maintaining a list of expire jiffies
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700311 * to optimize.
312 */
313void inet_csk_init_xmit_timers(struct sock *sk,
314 void (*retransmit_handler)(unsigned long),
315 void (*delack_handler)(unsigned long),
316 void (*keepalive_handler)(unsigned long))
317{
318 struct inet_connection_sock *icsk = inet_csk(sk);
319
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800320 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
321 (unsigned long)sk);
322 setup_timer(&icsk->icsk_delack_timer, delack_handler,
323 (unsigned long)sk);
324 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700325 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
326}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700327EXPORT_SYMBOL(inet_csk_init_xmit_timers);
328
329void inet_csk_clear_xmit_timers(struct sock *sk)
330{
331 struct inet_connection_sock *icsk = inet_csk(sk);
332
333 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
334
335 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
336 sk_stop_timer(sk, &icsk->icsk_delack_timer);
337 sk_stop_timer(sk, &sk->sk_timer);
338}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700339EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
340
341void inet_csk_delete_keepalive_timer(struct sock *sk)
342{
343 sk_stop_timer(sk, &sk->sk_timer);
344}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700345EXPORT_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}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700351EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
352
Jianjun Kongd93191002008-11-03 00:23:42 -0800353struct dst_entry *inet_csk_route_req(struct sock *sk,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700354 const struct request_sock *req)
355{
356 struct rtable *rt;
357 const struct inet_request_sock *ireq = inet_rsk(req);
358 struct ip_options *opt = inet_rsk(req)->opt;
359 struct flowi fl = { .oif = sk->sk_bound_dev_if,
Atis Elstsffce9082009-10-07 13:55:57 -0700360 .mark = sk->sk_mark,
Changli Gao58116622010-11-12 18:43:55 +0000361 .fl4_dst = ((opt && opt->srr) ?
362 opt->faddr : ireq->rmt_addr),
363 .fl4_src = ireq->loc_addr,
364 .fl4_tos = RT_CONN_FLAGS(sk),
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700365 .proto = sk->sk_protocol,
KOVACS Krisztian86b08d82008-10-01 07:44:42 -0700366 .flags = inet_sk_flowi_flags(sk),
Changli Gao58116622010-11-12 18:43:55 +0000367 .fl_ip_sport = inet_sk(sk)->inet_sport,
368 .fl_ip_dport = ireq->rmt_port };
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700369 struct net *net = sock_net(sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700370
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700371 security_req_classify_flow(req, &fl);
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800372 if (ip_route_output_flow(net, &rt, &fl, sk, 0))
373 goto no_route;
374 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
375 goto route_err;
Changli Gaod8d1f302010-06-10 23:31:35 -0700376 return &rt->dst;
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800377
378route_err:
379 ip_rt_put(rt);
380no_route:
381 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
382 return NULL;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700383}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700384EXPORT_SYMBOL_GPL(inet_csk_route_req);
385
Al Viro6b729772006-09-27 18:36:59 -0700386static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
Eric Dumazet72a3eff2006-11-16 02:30:37 -0800387 const u32 rnd, const u32 synq_hsize)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700388{
Al Viro6b729772006-09-27 18:36:59 -0700389 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700390}
391
392#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
393#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
394#else
395#define AF_INET_FAMILY(fam) 1
396#endif
397
398struct request_sock *inet_csk_search_req(const struct sock *sk,
399 struct request_sock ***prevp,
Al Viro6b729772006-09-27 18:36:59 -0700400 const __be16 rport, const __be32 raddr,
Al Viro7f25afb2006-09-27 18:27:47 -0700401 const __be32 laddr)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700402{
403 const struct inet_connection_sock *icsk = inet_csk(sk);
404 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
405 struct request_sock *req, **prev;
406
407 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
408 lopt->nr_table_entries)];
409 (req = *prev) != NULL;
410 prev = &req->dl_next) {
411 const struct inet_request_sock *ireq = inet_rsk(req);
412
413 if (ireq->rmt_port == rport &&
414 ireq->rmt_addr == raddr &&
415 ireq->loc_addr == laddr &&
416 AF_INET_FAMILY(req->rsk_ops->family)) {
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700417 WARN_ON(req->sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700418 *prevp = prev;
419 break;
420 }
421 }
422
423 return req;
424}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700425EXPORT_SYMBOL_GPL(inet_csk_search_req);
426
427void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
Arnaldo Carvalho de Meloc2977c22005-12-13 23:15:12 -0800428 unsigned long timeout)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700429{
430 struct inet_connection_sock *icsk = inet_csk(sk);
431 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
432 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
433 lopt->hash_rnd, lopt->nr_table_entries);
434
435 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
436 inet_csk_reqsk_queue_added(sk, timeout);
437}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000438EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700439
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700440/* Only thing we need from tcp.h */
441extern int sysctl_tcp_synack_retries;
442
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700443
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000444/* Decide when to expire the request and when to resend SYN-ACK */
445static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
446 const int max_retries,
447 const u8 rskq_defer_accept,
448 int *expire, int *resend)
449{
450 if (!rskq_defer_accept) {
451 *expire = req->retrans >= thresh;
452 *resend = 1;
453 return;
454 }
455 *expire = req->retrans >= thresh &&
456 (!inet_rsk(req)->acked || req->retrans >= max_retries);
457 /*
458 * Do not resend while waiting for data after ACK,
459 * start to resend on end of deferring period to give
460 * last chance for data or ACK to create established socket.
461 */
462 *resend = !inet_rsk(req)->acked ||
463 req->retrans >= rskq_defer_accept - 1;
464}
465
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700466void inet_csk_reqsk_queue_prune(struct sock *parent,
467 const unsigned long interval,
468 const unsigned long timeout,
469 const unsigned long max_rto)
470{
471 struct inet_connection_sock *icsk = inet_csk(parent);
472 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
473 struct listen_sock *lopt = queue->listen_opt;
David S. Millerec0a1962008-06-12 16:31:35 -0700474 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
475 int thresh = max_retries;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700476 unsigned long now = jiffies;
477 struct request_sock **reqp, *req;
478 int i, budget;
479
480 if (lopt == NULL || lopt->qlen == 0)
481 return;
482
483 /* Normally all the openreqs are young and become mature
484 * (i.e. converted to established socket) for first timeout.
485 * If synack was not acknowledged for 3 seconds, it means
486 * one of the following things: synack was lost, ack was lost,
487 * rtt is high or nobody planned to ack (i.e. synflood).
488 * When server is a bit loaded, queue is populated with old
489 * open requests, reducing effective size of queue.
490 * When server is well loaded, queue size reduces to zero
491 * after several minutes of work. It is not synflood,
492 * it is normal operation. The solution is pruning
493 * too old entries overriding normal timeout, when
494 * situation becomes dangerous.
495 *
496 * Essentially, we reserve half of room for young
497 * embrions; and abort old ones without pity, if old
498 * ones are about to clog our table.
499 */
500 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
501 int young = (lopt->qlen_young<<1);
502
503 while (thresh > 2) {
504 if (lopt->qlen < young)
505 break;
506 thresh--;
507 young <<= 1;
508 }
509 }
510
David S. Millerec0a1962008-06-12 16:31:35 -0700511 if (queue->rskq_defer_accept)
512 max_retries = queue->rskq_defer_accept;
513
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700514 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
515 i = lopt->clock_hand;
516
517 do {
518 reqp=&lopt->syn_table[i];
519 while ((req = *reqp) != NULL) {
520 if (time_after_eq(now, req->expires)) {
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000521 int expire = 0, resend = 0;
522
523 syn_ack_recalc(req, thresh, max_retries,
524 queue->rskq_defer_accept,
525 &expire, &resend);
Octavian Purdila72659ec2010-01-17 19:09:39 -0800526 if (req->rsk_ops->syn_ack_timeout)
527 req->rsk_ops->syn_ack_timeout(parent, req);
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000528 if (!expire &&
529 (!resend ||
William Allen Simpsone6b4d112009-12-02 18:07:39 +0000530 !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000531 inet_rsk(req)->acked)) {
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700532 unsigned long timeo;
533
534 if (req->retrans++ == 0)
535 lopt->qlen_young--;
536 timeo = min((timeout << req->retrans), max_rto);
537 req->expires = now + timeo;
538 reqp = &req->dl_next;
539 continue;
540 }
541
542 /* Drop this request */
543 inet_csk_reqsk_queue_unlink(parent, req, reqp);
544 reqsk_queue_removed(queue, req);
545 reqsk_free(req);
546 continue;
547 }
548 reqp = &req->dl_next;
549 }
550
551 i = (i + 1) & (lopt->nr_table_entries - 1);
552
553 } while (--budget > 0);
554
555 lopt->clock_hand = i;
556
557 if (lopt->qlen)
558 inet_csk_reset_keepalive_timer(parent, interval);
559}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700560EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
561
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700562struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
Al Virodd0fc662005-10-07 07:46:04 +0100563 const gfp_t priority)
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700564{
565 struct sock *newsk = sk_clone(sk, priority);
566
567 if (newsk != NULL) {
568 struct inet_connection_sock *newicsk = inet_csk(newsk);
569
570 newsk->sk_state = TCP_SYN_RECV;
571 newicsk->icsk_bind_hash = NULL;
572
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000573 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
574 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
575 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700576 newsk->sk_write_space = sk_stream_write_space;
577
578 newicsk->icsk_retransmits = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300579 newicsk->icsk_backoff = 0;
580 newicsk->icsk_probes_out = 0;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700581
582 /* Deinitialize accept_queue to trap illegal accesses. */
583 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700584
585 security_inet_csk_clone(newsk, req);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700586 }
587 return newsk;
588}
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700589EXPORT_SYMBOL_GPL(inet_csk_clone);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700590
591/*
592 * At this point, there should be no process reference to this
593 * socket, and thus no user references at all. Therefore we
594 * can assume the socket waitqueue is inactive and nobody will
595 * try to jump onto it.
596 */
597void inet_csk_destroy_sock(struct sock *sk)
598{
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700599 WARN_ON(sk->sk_state != TCP_CLOSE);
600 WARN_ON(!sock_flag(sk, SOCK_DEAD));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700601
602 /* It cannot be in hash table! */
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700603 WARN_ON(!sk_unhashed(sk));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700604
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000605 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
606 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700607
608 sk->sk_prot->destroy(sk);
609
610 sk_stream_kill_queues(sk);
611
612 xfrm_sk_free_policy(sk);
613
614 sk_refcnt_debug_release(sk);
615
Eric Dumazetdd24c002008-11-25 21:17:14 -0800616 percpu_counter_dec(sk->sk_prot->orphan_count);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700617 sock_put(sk);
618}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700619EXPORT_SYMBOL(inet_csk_destroy_sock);
620
621int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
622{
623 struct inet_sock *inet = inet_sk(sk);
624 struct inet_connection_sock *icsk = inet_csk(sk);
625 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
626
627 if (rc != 0)
628 return rc;
629
630 sk->sk_max_ack_backlog = 0;
631 sk->sk_ack_backlog = 0;
632 inet_csk_delack_init(sk);
633
634 /* There is race window here: we announce ourselves listening,
635 * but this transition is still not validated by get_port().
636 * It is OK, because this socket enters to hash table only
637 * after validation is complete.
638 */
639 sk->sk_state = TCP_LISTEN;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000640 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
641 inet->inet_sport = htons(inet->inet_num);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700642
643 sk_dst_reset(sk);
644 sk->sk_prot->hash(sk);
645
646 return 0;
647 }
648
649 sk->sk_state = TCP_CLOSE;
650 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
651 return -EADDRINUSE;
652}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700653EXPORT_SYMBOL_GPL(inet_csk_listen_start);
654
655/*
656 * This routine closes sockets which have been at least partially
657 * opened, but not yet accepted.
658 */
659void inet_csk_listen_stop(struct sock *sk)
660{
661 struct inet_connection_sock *icsk = inet_csk(sk);
662 struct request_sock *acc_req;
663 struct request_sock *req;
664
665 inet_csk_delete_keepalive_timer(sk);
666
667 /* make all the listen_opt local to us */
668 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
669
670 /* Following specs, it would be better either to send FIN
671 * (and enter FIN-WAIT-1, it is normal close)
672 * or to send active reset (abort).
673 * Certainly, it is pretty dangerous while synflood, but it is
674 * bad justification for our negligence 8)
675 * To be honest, we are not able to make either
676 * of the variants now. --ANK
677 */
678 reqsk_queue_destroy(&icsk->icsk_accept_queue);
679
680 while ((req = acc_req) != NULL) {
681 struct sock *child = req->sk;
682
683 acc_req = req->dl_next;
684
685 local_bh_disable();
686 bh_lock_sock(child);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700687 WARN_ON(sock_owned_by_user(child));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700688 sock_hold(child);
689
690 sk->sk_prot->disconnect(child, O_NONBLOCK);
691
692 sock_orphan(child);
693
Herbert Xueb4dea52008-12-29 23:04:08 -0800694 percpu_counter_inc(sk->sk_prot->orphan_count);
695
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700696 inet_csk_destroy_sock(child);
697
698 bh_unlock_sock(child);
699 local_bh_enable();
700 sock_put(child);
701
702 sk_acceptq_removed(sk);
703 __reqsk_free(req);
704 }
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700705 WARN_ON(sk->sk_ack_backlog);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700706}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700707EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800708
709void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
710{
711 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
712 const struct inet_sock *inet = inet_sk(sk);
713
714 sin->sin_family = AF_INET;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000715 sin->sin_addr.s_addr = inet->inet_daddr;
716 sin->sin_port = inet->inet_dport;
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800717}
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800718EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
Arnaldo Carvalho de Meloc4d93902006-03-20 22:01:03 -0800719
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800720#ifdef CONFIG_COMPAT
721int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
722 char __user *optval, int __user *optlen)
723{
David S. Millerdbeff122006-03-20 22:52:32 -0800724 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800725
726 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
727 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
728 optval, optlen);
729 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
730 optval, optlen);
731}
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800732EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
733
734int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -0700735 char __user *optval, unsigned int optlen)
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800736{
David S. Millerdbeff122006-03-20 22:52:32 -0800737 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800738
739 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
740 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
741 optval, optlen);
742 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
743 optval, optlen);
744}
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800745EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
746#endif