blob: 61a9deec299332aa431c13ad15197974306ff117 [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>
Eric Dumazetfa76ce732015-03-19 19:04:20 -070026#include <net/tcp.h>
Craig Gallekc125e802016-02-10 11:50:40 -050027#include <net/sock_reuseport.h>
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070028
29#ifdef INET_CSK_DEBUG
30const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
31EXPORT_SYMBOL(inet_csk_timer_bug_msg);
32#endif
33
Eric W. Biederman0bbf87d2013-09-28 14:10:59 -070034void inet_get_local_port_range(struct net *net, int *low, int *high)
Stephen Hemminger227b60f2007-10-10 17:30:46 -070035{
Eric Dumazet95c96172012-04-15 05:58:06 +000036 unsigned int seq;
37
Stephen Hemminger227b60f2007-10-10 17:30:46 -070038 do {
Cong Wangc9d8f1a2014-05-06 11:02:49 -070039 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
Stephen Hemminger227b60f2007-10-10 17:30:46 -070040
Cong Wangc9d8f1a2014-05-06 11:02:49 -070041 *low = net->ipv4.ip_local_ports.range[0];
42 *high = net->ipv4.ip_local_ports.range[1];
43 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
Stephen Hemminger227b60f2007-10-10 17:30:46 -070044}
45EXPORT_SYMBOL(inet_get_local_port_range);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070046
Arnaldo Carvalho de Melo971af182005-12-13 23:14:47 -080047int inet_csk_bind_conflict(const struct sock *sk,
Alex Copotaacd9282012-04-12 22:21:45 +000048 const struct inet_bind_bucket *tb, bool relax)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070049{
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070050 struct sock *sk2;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070051 int reuse = sk->sk_reuse;
Tom Herbertda5e3632013-01-22 09:50:24 +000052 int reuseport = sk->sk_reuseport;
53 kuid_t uid = sock_i_uid((struct sock *)sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070054
Pavel Emelyanov7477fd2e2008-04-14 02:42:27 -070055 /*
56 * Unlike other sk lookup places we do not check
57 * for sk_net here, since _all_ the socks listed
58 * in tb->owners list belong to the same net - the
59 * one this bucket belongs to.
60 */
61
Sasha Levinb67bfe02013-02-27 17:06:00 -080062 sk_for_each_bound(sk2, &tb->owners) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070063 if (sk != sk2 &&
64 !inet_v6_ipv6only(sk2) &&
65 (!sk->sk_bound_dev_if ||
66 !sk2->sk_bound_dev_if ||
67 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
Tom Herbertda5e3632013-01-22 09:50:24 +000068 if ((!reuse || !sk2->sk_reuse ||
69 sk2->sk_state == TCP_LISTEN) &&
70 (!reuseport || !sk2->sk_reuseport ||
Craig Gallekc125e802016-02-10 11:50:40 -050071 rcu_access_pointer(sk->sk_reuseport_cb) ||
72 (sk2->sk_state != TCP_TIME_WAIT &&
Tom Herbertda5e3632013-01-22 09:50:24 +000073 !uid_eq(uid, sock_i_uid(sk2))))) {
Eric Dumazet50805462013-10-02 04:29:50 -070074
75 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
76 sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070077 break;
David S. Miller8d238b22010-04-28 11:25:59 -070078 }
Alex Copotaacd9282012-04-12 22:21:45 +000079 if (!relax && reuse && sk2->sk_reuse &&
80 sk2->sk_state != TCP_LISTEN) {
Alex Copotaacd9282012-04-12 22:21:45 +000081
Eric Dumazet50805462013-10-02 04:29:50 -070082 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
83 sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
Alex Copotaacd9282012-04-12 22:21:45 +000084 break;
85 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070086 }
87 }
Sasha Levinb67bfe02013-02-27 17:06:00 -080088 return sk2 != NULL;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070089}
Arnaldo Carvalho de Melo971af182005-12-13 23:14:47 -080090EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
91
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070092/* Obtain a reference to a local port for the given sock,
93 * if snum is zero it means select any available local port.
Eric Dumazetea8add22016-02-11 16:28:50 -080094 * We try to allocate an odd port (and leave even ports for connect())
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070095 */
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -080096int inet_csk_get_port(struct sock *sk, unsigned short snum)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070097{
Eric Dumazetea8add22016-02-11 16:28:50 -080098 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
99 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
100 int ret = 1, attempts = 5, port = snum;
101 int smallest_size = -1, smallest_port;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700102 struct inet_bind_hashbucket *head;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900103 struct net *net = sock_net(sk);
Eric Dumazetea8add22016-02-11 16:28:50 -0800104 int i, low, high, attempt_half;
105 struct inet_bind_bucket *tb;
Tom Herbertda5e3632013-01-22 09:50:24 +0000106 kuid_t uid = sock_i_uid(sk);
Eric Dumazetea8add22016-02-11 16:28:50 -0800107 u32 remaining, offset;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700108
Eric Dumazetea8add22016-02-11 16:28:50 -0800109 if (port) {
110have_port:
111 head = &hinfo->bhash[inet_bhashfn(net, port,
112 hinfo->bhash_size)];
113 spin_lock_bh(&head->lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800114 inet_bind_bucket_for_each(tb, &head->chain)
Eric Dumazetea8add22016-02-11 16:28:50 -0800115 if (net_eq(ib_net(tb), net) && tb->port == port)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700116 goto tb_found;
Eric Dumazetea8add22016-02-11 16:28:50 -0800117
118 goto tb_not_found;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700119 }
Eric Dumazetea8add22016-02-11 16:28:50 -0800120again:
121 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
122other_half_scan:
123 inet_get_local_port_range(net, &low, &high);
124 high++; /* [32768, 60999] -> [32768, 61000[ */
125 if (high - low < 4)
126 attempt_half = 0;
127 if (attempt_half) {
128 int half = low + (((high - low) >> 2) << 1);
129
130 if (attempt_half == 1)
131 high = half;
132 else
133 low = half;
134 }
135 remaining = high - low;
136 if (likely(remaining > 1))
137 remaining &= ~1U;
138
139 offset = prandom_u32() % remaining;
140 /* __inet_hash_connect() favors ports having @low parity
141 * We do the opposite to not pollute connect() users.
142 */
143 offset |= 1U;
144 smallest_size = -1;
145 smallest_port = low; /* avoid compiler warning */
146
147other_parity_scan:
148 port = low + offset;
149 for (i = 0; i < remaining; i += 2, port += 2) {
150 if (unlikely(port >= high))
151 port -= remaining;
152 if (inet_is_local_reserved_port(net, port))
153 continue;
154 head = &hinfo->bhash[inet_bhashfn(net, port,
155 hinfo->bhash_size)];
156 spin_lock_bh(&head->lock);
157 inet_bind_bucket_for_each(tb, &head->chain)
158 if (net_eq(ib_net(tb), net) && tb->port == port) {
159 if (((tb->fastreuse > 0 && reuse) ||
160 (tb->fastreuseport > 0 &&
161 sk->sk_reuseport &&
162 !rcu_access_pointer(sk->sk_reuseport_cb) &&
163 uid_eq(tb->fastuid, uid))) &&
164 (tb->num_owners < smallest_size || smallest_size == -1)) {
165 smallest_size = tb->num_owners;
166 smallest_port = port;
167 }
168 if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false))
169 goto tb_found;
170 goto next_port;
171 }
172 goto tb_not_found;
173next_port:
174 spin_unlock_bh(&head->lock);
175 cond_resched();
176 }
177
178 if (smallest_size != -1) {
179 port = smallest_port;
180 goto have_port;
181 }
182 offset--;
183 if (!(offset & 1))
184 goto other_parity_scan;
185
186 if (attempt_half == 1) {
187 /* OK we now try the upper half of the range */
188 attempt_half = 2;
189 goto other_half_scan;
190 }
191 return ret;
192
193tb_not_found:
194 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
195 net, head, port);
196 if (!tb)
197 goto fail_unlock;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700198tb_found:
199 if (!hlist_empty(&tb->owners)) {
Pavel Emelyanov4a17fd52012-04-19 03:39:36 +0000200 if (sk->sk_reuse == SK_FORCE_REUSE)
201 goto success;
202
Eric Dumazetea8add22016-02-11 16:28:50 -0800203 if (((tb->fastreuse > 0 && reuse) ||
Tom Herbertda5e3632013-01-22 09:50:24 +0000204 (tb->fastreuseport > 0 &&
Craig Galleke5fbfc12016-02-22 10:45:29 -0500205 !rcu_access_pointer(sk->sk_reuseport_cb) &&
Eric Dumazetea8add22016-02-11 16:28:50 -0800206 sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
207 smallest_size == -1)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700208 goto success;
Eric Dumazetea8add22016-02-11 16:28:50 -0800209 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
210 if ((reuse ||
211 (tb->fastreuseport > 0 &&
212 sk->sk_reuseport &&
213 !rcu_access_pointer(sk->sk_reuseport_cb) &&
214 uid_eq(tb->fastuid, uid))) &&
215 smallest_size != -1 && --attempts >= 0) {
216 spin_unlock_bh(&head->lock);
217 goto again;
Evgeniy Polyakova9d8f912009-01-19 16:46:02 -0800218 }
Eric Dumazetea8add22016-02-11 16:28:50 -0800219 goto fail_unlock;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700220 }
Eric Dumazetea8add22016-02-11 16:28:50 -0800221 if (!reuse)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700222 tb->fastreuse = 0;
Eric Dumazetea8add22016-02-11 16:28:50 -0800223 if (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid))
224 tb->fastreuseport = 0;
225 } else {
226 tb->fastreuse = reuse;
Tom Herbertda5e3632013-01-22 09:50:24 +0000227 if (sk->sk_reuseport) {
228 tb->fastreuseport = 1;
229 tb->fastuid = uid;
Eric Dumazetea8add22016-02-11 16:28:50 -0800230 } else {
Tom Herbertda5e3632013-01-22 09:50:24 +0000231 tb->fastreuseport = 0;
Eric Dumazetea8add22016-02-11 16:28:50 -0800232 }
Tom Herbertda5e3632013-01-22 09:50:24 +0000233 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700234success:
235 if (!inet_csk(sk)->icsk_bind_hash)
Eric Dumazetea8add22016-02-11 16:28:50 -0800236 inet_bind_hash(sk, tb, port);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700237 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900238 ret = 0;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700239
240fail_unlock:
Eric Dumazetea8add22016-02-11 16:28:50 -0800241 spin_unlock_bh(&head->lock);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700242 return ret;
243}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700244EXPORT_SYMBOL_GPL(inet_csk_get_port);
245
246/*
247 * Wait for an incoming connection, avoid race conditions. This must be called
248 * with the socket locked.
249 */
250static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
251{
252 struct inet_connection_sock *icsk = inet_csk(sk);
253 DEFINE_WAIT(wait);
254 int err;
255
256 /*
257 * True wake-one mechanism for incoming connections: only
258 * one process gets woken up, not the 'whole herd'.
259 * Since we do not 'race & poll' for established sockets
260 * anymore, the common case will execute the loop only once.
261 *
262 * Subtle issue: "add_wait_queue_exclusive()" will be added
263 * after any current non-exclusive waiters, and we know that
264 * it will always _stay_ after any new non-exclusive waiters
265 * because all non-exclusive waiters are added at the
266 * beginning of the wait-queue. As such, it's ok to "drop"
267 * our exclusiveness temporarily when we get woken up without
268 * having to remove and re-insert us on the wait queue.
269 */
270 for (;;) {
Eric Dumazetaa395142010-04-20 13:03:51 +0000271 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700272 TASK_INTERRUPTIBLE);
273 release_sock(sk);
274 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
275 timeo = schedule_timeout(timeo);
Eric Dumazetcb7cf8a2015-03-16 12:19:24 -0700276 sched_annotate_sleep();
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700277 lock_sock(sk);
278 err = 0;
279 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
280 break;
281 err = -EINVAL;
282 if (sk->sk_state != TCP_LISTEN)
283 break;
284 err = sock_intr_errno(timeo);
285 if (signal_pending(current))
286 break;
287 err = -EAGAIN;
288 if (!timeo)
289 break;
290 }
Eric Dumazetaa395142010-04-20 13:03:51 +0000291 finish_wait(sk_sleep(sk), &wait);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700292 return err;
293}
294
295/*
296 * This will accept the next outstanding connection.
297 */
298struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
299{
300 struct inet_connection_sock *icsk = inet_csk(sk);
Jerry Chu83368862012-08-31 12:29:12 +0000301 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
Jerry Chu83368862012-08-31 12:29:12 +0000302 struct request_sock *req;
Eric Dumazete3d95ad2015-03-17 18:32:30 -0700303 struct sock *newsk;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700304 int error;
305
306 lock_sock(sk);
307
308 /* We need to make sure that this socket is listening,
309 * and that it has something pending.
310 */
311 error = -EINVAL;
312 if (sk->sk_state != TCP_LISTEN)
313 goto out_err;
314
315 /* Find already established connection */
Jerry Chu83368862012-08-31 12:29:12 +0000316 if (reqsk_queue_empty(queue)) {
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700317 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
318
319 /* If this is a non blocking socket don't sleep */
320 error = -EAGAIN;
321 if (!timeo)
322 goto out_err;
323
324 error = inet_csk_wait_for_connect(sk, timeo);
325 if (error)
326 goto out_err;
327 }
Eric Dumazetfff1f302015-10-02 11:43:23 -0700328 req = reqsk_queue_remove(queue, sk);
Jerry Chu83368862012-08-31 12:29:12 +0000329 newsk = req->sk;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700330
Eric Dumazete3d95ad2015-03-17 18:32:30 -0700331 if (sk->sk_protocol == IPPROTO_TCP &&
Eric Dumazet0536fcc2015-09-29 07:42:52 -0700332 tcp_rsk(req)->tfo_listener) {
333 spin_lock_bh(&queue->fastopenq.lock);
Eric Dumazet9439ce02015-03-17 18:32:29 -0700334 if (tcp_rsk(req)->tfo_listener) {
Jerry Chu83368862012-08-31 12:29:12 +0000335 /* We are still waiting for the final ACK from 3WHS
336 * so can't free req now. Instead, we set req->sk to
337 * NULL to signify that the child socket is taken
338 * so reqsk_fastopen_remove() will free the req
339 * when 3WHS finishes (or is aborted).
340 */
341 req->sk = NULL;
342 req = NULL;
343 }
Eric Dumazet0536fcc2015-09-29 07:42:52 -0700344 spin_unlock_bh(&queue->fastopenq.lock);
Jerry Chu83368862012-08-31 12:29:12 +0000345 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700346out:
347 release_sock(sk);
Jerry Chu83368862012-08-31 12:29:12 +0000348 if (req)
Eric Dumazet13854e52015-03-15 21:12:16 -0700349 reqsk_put(req);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700350 return newsk;
351out_err:
352 newsk = NULL;
Jerry Chu83368862012-08-31 12:29:12 +0000353 req = NULL;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700354 *err = error;
355 goto out;
356}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700357EXPORT_SYMBOL(inet_csk_accept);
358
359/*
360 * Using different timers for retransmit, delayed acks and probes
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900361 * We may wish use just one timer maintaining a list of expire jiffies
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700362 * to optimize.
363 */
364void inet_csk_init_xmit_timers(struct sock *sk,
365 void (*retransmit_handler)(unsigned long),
366 void (*delack_handler)(unsigned long),
367 void (*keepalive_handler)(unsigned long))
368{
369 struct inet_connection_sock *icsk = inet_csk(sk);
370
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800371 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
372 (unsigned long)sk);
373 setup_timer(&icsk->icsk_delack_timer, delack_handler,
374 (unsigned long)sk);
375 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700376 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
377}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700378EXPORT_SYMBOL(inet_csk_init_xmit_timers);
379
380void inet_csk_clear_xmit_timers(struct sock *sk)
381{
382 struct inet_connection_sock *icsk = inet_csk(sk);
383
384 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
385
386 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
387 sk_stop_timer(sk, &icsk->icsk_delack_timer);
388 sk_stop_timer(sk, &sk->sk_timer);
389}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700390EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
391
392void inet_csk_delete_keepalive_timer(struct sock *sk)
393{
394 sk_stop_timer(sk, &sk->sk_timer);
395}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700396EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
397
398void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
399{
400 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
401}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700402EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
403
Eric Dumazete5895bc2015-09-25 07:39:11 -0700404struct dst_entry *inet_csk_route_req(const struct sock *sk,
David S. Miller6bd023f2011-05-18 18:32:03 -0400405 struct flowi4 *fl4,
David S. Millerba3f7f02012-07-17 14:02:46 -0700406 const struct request_sock *req)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700407{
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700408 const struct inet_request_sock *ireq = inet_rsk(req);
Eric Dumazet8b929ab2015-03-22 10:22:20 -0700409 struct net *net = read_pnet(&ireq->ireq_net);
410 struct ip_options_rcu *opt = ireq->opt;
411 struct rtable *rt;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700412
Eric Dumazet8b929ab2015-03-22 10:22:20 -0700413 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
David S. Millere79d9bc2011-03-31 04:53:20 -0700414 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
Eric Dumazet8b929ab2015-03-22 10:22:20 -0700415 sk->sk_protocol, inet_sk_flowi_flags(sk),
Eric Dumazet634fb9792013-10-09 15:21:29 -0700416 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
Eric Dumazet8b929ab2015-03-22 10:22:20 -0700417 ireq->ir_loc_addr, ireq->ir_rmt_port,
418 htons(ireq->ir_num));
David S. Miller6bd023f2011-05-18 18:32:03 -0400419 security_req_classify_flow(req, flowi4_to_flowi(fl4));
420 rt = ip_route_output_flow(net, fl4, sk);
David S. Millerb23dd4f2011-03-02 14:31:35 -0800421 if (IS_ERR(rt))
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800422 goto no_route;
Julian Anastasov155e8332012-10-08 11:41:18 +0000423 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800424 goto route_err;
Changli Gaod8d1f302010-06-10 23:31:35 -0700425 return &rt->dst;
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800426
427route_err:
428 ip_rt_put(rt);
429no_route:
Eric Dumazetb45386e2016-04-27 16:44:35 -0700430 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
Ilpo Järvinen857a6e02008-12-14 23:13:08 -0800431 return NULL;
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700432}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700433EXPORT_SYMBOL_GPL(inet_csk_route_req);
434
Eric Dumazeta2432c42015-09-29 07:42:43 -0700435struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
David S. Miller77357a92011-05-08 14:34:22 -0700436 struct sock *newsk,
437 const struct request_sock *req)
438{
439 const struct inet_request_sock *ireq = inet_rsk(req);
Eric Dumazet8b929ab2015-03-22 10:22:20 -0700440 struct net *net = read_pnet(&ireq->ireq_net);
David S. Miller77357a92011-05-08 14:34:22 -0700441 struct inet_sock *newinet = inet_sk(newsk);
Christoph Paasch1a7b27c2012-08-20 02:52:09 +0000442 struct ip_options_rcu *opt;
David S. Miller77357a92011-05-08 14:34:22 -0700443 struct flowi4 *fl4;
444 struct rtable *rt;
445
446 fl4 = &newinet->cork.fl.u.ip4;
Christoph Paasch1a7b27c2012-08-20 02:52:09 +0000447
448 rcu_read_lock();
449 opt = rcu_dereference(newinet->inet_opt);
Eric Dumazet8b929ab2015-03-22 10:22:20 -0700450 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
David S. Miller77357a92011-05-08 14:34:22 -0700451 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
452 sk->sk_protocol, inet_sk_flowi_flags(sk),
Eric Dumazet634fb9792013-10-09 15:21:29 -0700453 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
Eric Dumazet8b929ab2015-03-22 10:22:20 -0700454 ireq->ir_loc_addr, ireq->ir_rmt_port,
455 htons(ireq->ir_num));
David S. Miller77357a92011-05-08 14:34:22 -0700456 security_req_classify_flow(req, flowi4_to_flowi(fl4));
457 rt = ip_route_output_flow(net, fl4, sk);
458 if (IS_ERR(rt))
459 goto no_route;
Julian Anastasov155e8332012-10-08 11:41:18 +0000460 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
David S. Miller77357a92011-05-08 14:34:22 -0700461 goto route_err;
Christoph Paasch1a7b27c2012-08-20 02:52:09 +0000462 rcu_read_unlock();
David S. Miller77357a92011-05-08 14:34:22 -0700463 return &rt->dst;
464
465route_err:
466 ip_rt_put(rt);
467no_route:
Christoph Paasch1a7b27c2012-08-20 02:52:09 +0000468 rcu_read_unlock();
Eric Dumazetb45386e2016-04-27 16:44:35 -0700469 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
David S. Miller77357a92011-05-08 14:34:22 -0700470 return NULL;
471}
472EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
473
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000474#if IS_ENABLED(CONFIG_IPV6)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700475#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
476#else
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700477#define AF_INET_FAMILY(fam) true
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700478#endif
479
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000480/* Decide when to expire the request and when to resend SYN-ACK */
481static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
482 const int max_retries,
483 const u8 rskq_defer_accept,
484 int *expire, int *resend)
485{
486 if (!rskq_defer_accept) {
Eric Dumazete6c022a2012-10-27 23:16:46 +0000487 *expire = req->num_timeout >= thresh;
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000488 *resend = 1;
489 return;
490 }
Eric Dumazete6c022a2012-10-27 23:16:46 +0000491 *expire = req->num_timeout >= thresh &&
492 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000493 /*
494 * Do not resend while waiting for data after ACK,
495 * start to resend on end of deferring period to give
496 * last chance for data or ACK to create established socket.
497 */
498 *resend = !inet_rsk(req)->acked ||
Eric Dumazete6c022a2012-10-27 23:16:46 +0000499 req->num_timeout >= rskq_defer_accept - 1;
Julian Anastasov0c3d79b2009-10-19 10:03:58 +0000500}
501
Eric Dumazet1b70e972015-09-25 07:39:24 -0700502int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
Eric Dumazete6c022a2012-10-27 23:16:46 +0000503{
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000504 int err = req->rsk_ops->rtx_syn_ack(parent, req);
Eric Dumazete6c022a2012-10-27 23:16:46 +0000505
506 if (!err)
507 req->num_retrans++;
508 return err;
509}
510EXPORT_SYMBOL(inet_rtx_syn_ack);
511
Eric Dumazet079096f2015-10-02 11:43:32 -0700512/* return true if req was found in the ehash table */
Eric Dumazetb357a362015-04-23 18:03:44 -0700513static bool reqsk_queue_unlink(struct request_sock_queue *queue,
514 struct request_sock *req)
515{
Eric Dumazet079096f2015-10-02 11:43:32 -0700516 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700517 bool found = false;
Eric Dumazetb357a362015-04-23 18:03:44 -0700518
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700519 if (sk_hashed(req_to_sk(req))) {
520 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
Eric Dumazetb357a362015-04-23 18:03:44 -0700521
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700522 spin_lock(lock);
523 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
524 spin_unlock(lock);
525 }
Eric Dumazet83fccfc2015-08-13 15:44:51 -0700526 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
Eric Dumazetb357a362015-04-23 18:03:44 -0700527 reqsk_put(req);
528 return found;
529}
530
531void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
532{
533 if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
534 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
535 reqsk_put(req);
536 }
537}
538EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
539
Eric Dumazetf03f2e12015-10-14 11:16:27 -0700540void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
541{
542 inet_csk_reqsk_queue_drop(sk, req);
543 reqsk_put(req);
544}
545EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
546
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700547static void reqsk_timer_handler(unsigned long data)
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700548{
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700549 struct request_sock *req = (struct request_sock *)data;
550 struct sock *sk_listener = req->rsk_listener;
Nikolay Borisov7c083ec2016-02-03 09:46:50 +0200551 struct net *net = sock_net(sk_listener);
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700552 struct inet_connection_sock *icsk = inet_csk(sk_listener);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700553 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
Eric Dumazet2b41fab2015-03-22 10:22:18 -0700554 int qlen, expire = 0, resend = 0;
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700555 int max_retries, thresh;
Eric Dumazet2b41fab2015-03-22 10:22:18 -0700556 u8 defer_accept;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700557
Eric Dumazet00fd38d2015-11-12 08:43:18 -0800558 if (sk_state_load(sk_listener) != TCP_LISTEN)
Eric Dumazet079096f2015-10-02 11:43:32 -0700559 goto drop;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700560
Nikolay Borisov7c083ec2016-02-03 09:46:50 +0200561 max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700562 thresh = max_retries;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700563 /* Normally all the openreqs are young and become mature
564 * (i.e. converted to established socket) for first timeout.
Eric Dumazetfd4f2ce2012-04-12 19:48:40 +0000565 * If synack was not acknowledged for 1 second, it means
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700566 * one of the following things: synack was lost, ack was lost,
567 * rtt is high or nobody planned to ack (i.e. synflood).
568 * When server is a bit loaded, queue is populated with old
569 * open requests, reducing effective size of queue.
570 * When server is well loaded, queue size reduces to zero
571 * after several minutes of work. It is not synflood,
572 * it is normal operation. The solution is pruning
573 * too old entries overriding normal timeout, when
574 * situation becomes dangerous.
575 *
576 * Essentially, we reserve half of room for young
577 * embrions; and abort old ones without pity, if old
578 * ones are about to clog our table.
579 */
Eric Dumazetaac065c2015-10-02 11:43:24 -0700580 qlen = reqsk_queue_len(queue);
Eric Dumazetacb4a6b2015-10-06 14:49:58 -0700581 if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
Eric Dumazetaac065c2015-10-02 11:43:24 -0700582 int young = reqsk_queue_len_young(queue) << 1;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700583
584 while (thresh > 2) {
Eric Dumazet2b41fab2015-03-22 10:22:18 -0700585 if (qlen < young)
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700586 break;
587 thresh--;
588 young <<= 1;
589 }
590 }
Eric Dumazet2b41fab2015-03-22 10:22:18 -0700591 defer_accept = READ_ONCE(queue->rskq_defer_accept);
592 if (defer_accept)
593 max_retries = defer_accept;
594 syn_ack_recalc(req, thresh, max_retries, defer_accept,
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700595 &expire, &resend);
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700596 req->rsk_ops->syn_ack_timeout(req);
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700597 if (!expire &&
598 (!resend ||
599 !inet_rtx_syn_ack(sk_listener, req) ||
600 inet_rsk(req)->acked)) {
601 unsigned long timeo;
David S. Millerec0a1962008-06-12 16:31:35 -0700602
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700603 if (req->num_timeout++ == 0)
Eric Dumazetaac065c2015-10-02 11:43:24 -0700604 atomic_dec(&queue->young);
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700605 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
Thomas Gleixnerf3438bc2016-07-04 09:50:23 +0000606 mod_timer(&req->rsk_timer, jiffies + timeo);
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700607 return;
608 }
Eric Dumazet079096f2015-10-02 11:43:32 -0700609drop:
Eric Dumazetf03f2e12015-10-14 11:16:27 -0700610 inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700611}
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700612
Eric Dumazet079096f2015-10-02 11:43:32 -0700613static void reqsk_queue_hash_req(struct request_sock *req,
614 unsigned long timeout)
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700615{
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700616 req->num_retrans = 0;
617 req->num_timeout = 0;
618 req->sk = NULL;
619
Thomas Gleixnerf3438bc2016-07-04 09:50:23 +0000620 setup_pinned_timer(&req->rsk_timer, reqsk_timer_handler,
621 (unsigned long)req);
622 mod_timer(&req->rsk_timer, jiffies + timeout);
Eric Dumazet29c68522015-09-19 09:48:04 -0700623
Eric Dumazet079096f2015-10-02 11:43:32 -0700624 inet_ehash_insert(req_to_sk(req), NULL);
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700625 /* before letting lookups find us, make sure all req fields
626 * are committed to memory and refcnt initialized.
627 */
628 smp_wmb();
Eric Dumazetca6fb062015-10-02 11:43:35 -0700629 atomic_set(&req->rsk_refcnt, 2 + 1);
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700630}
Eric Dumazet079096f2015-10-02 11:43:32 -0700631
632void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
633 unsigned long timeout)
634{
635 reqsk_queue_hash_req(req, timeout);
636 inet_csk_reqsk_queue_added(sk);
637}
638EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700639
Eric Dumazete56c57d2011-11-08 17:07:07 -0500640/**
641 * inet_csk_clone_lock - clone an inet socket, and lock its clone
642 * @sk: the socket to clone
643 * @req: request_sock
644 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
645 *
646 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
647 */
648struct sock *inet_csk_clone_lock(const struct sock *sk,
649 const struct request_sock *req,
650 const gfp_t priority)
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700651{
Eric Dumazete56c57d2011-11-08 17:07:07 -0500652 struct sock *newsk = sk_clone_lock(sk, priority);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700653
Ian Morris00db4122015-04-03 09:17:27 +0100654 if (newsk) {
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700655 struct inet_connection_sock *newicsk = inet_csk(newsk);
656
657 newsk->sk_state = TCP_SYN_RECV;
658 newicsk->icsk_bind_hash = NULL;
659
Eric Dumazet634fb9792013-10-09 15:21:29 -0700660 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
Eric Dumazetb44084c2013-10-10 00:04:37 -0700661 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
662 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700663 newsk->sk_write_space = sk_stream_write_space;
664
Eric Dumazet85017862016-04-06 22:07:34 -0700665 /* listeners have SOCK_RCU_FREE, not the children */
666 sock_reset_flag(newsk, SOCK_RCU_FREE);
667
Lorenzo Colitti84f39b02014-05-13 10:17:35 -0700668 newsk->sk_mark = inet_rsk(req)->ir_mark;
Eric Dumazet33cf7c92015-03-11 18:53:14 -0700669 atomic64_set(&newsk->sk_cookie,
670 atomic64_read(&inet_rsk(req)->ir_cookie));
Lorenzo Colitti84f39b02014-05-13 10:17:35 -0700671
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700672 newicsk->icsk_retransmits = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300673 newicsk->icsk_backoff = 0;
674 newicsk->icsk_probes_out = 0;
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700675
676 /* Deinitialize accept_queue to trap illegal accesses. */
677 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700678
679 security_inet_csk_clone(newsk, req);
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700680 }
681 return newsk;
682}
Eric Dumazete56c57d2011-11-08 17:07:07 -0500683EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700684
685/*
686 * At this point, there should be no process reference to this
687 * socket, and thus no user references at all. Therefore we
688 * can assume the socket waitqueue is inactive and nobody will
689 * try to jump onto it.
690 */
691void inet_csk_destroy_sock(struct sock *sk)
692{
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700693 WARN_ON(sk->sk_state != TCP_CLOSE);
694 WARN_ON(!sock_flag(sk, SOCK_DEAD));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700695
696 /* It cannot be in hash table! */
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700697 WARN_ON(!sk_unhashed(sk));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700698
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000699 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
700 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700701
702 sk->sk_prot->destroy(sk);
703
704 sk_stream_kill_queues(sk);
705
706 xfrm_sk_free_policy(sk);
707
708 sk_refcnt_debug_release(sk);
709
Eric Dumazet777c6ae2016-05-04 15:27:29 -0700710 local_bh_disable();
Eric Dumazetdd24c002008-11-25 21:17:14 -0800711 percpu_counter_dec(sk->sk_prot->orphan_count);
Eric Dumazet777c6ae2016-05-04 15:27:29 -0700712 local_bh_enable();
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700713 sock_put(sk);
714}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700715EXPORT_SYMBOL(inet_csk_destroy_sock);
716
Christoph Paasche337e242012-12-14 04:07:58 +0000717/* This function allows to force a closure of a socket after the call to
718 * tcp/dccp_create_openreq_child().
719 */
720void inet_csk_prepare_forced_close(struct sock *sk)
Christoph Paaschc10cb5f2013-03-07 02:34:33 +0000721 __releases(&sk->sk_lock.slock)
Christoph Paasche337e242012-12-14 04:07:58 +0000722{
723 /* sk_clone_lock locked the socket and set refcnt to 2 */
724 bh_unlock_sock(sk);
725 sock_put(sk);
726
727 /* The below has to be done to allow calling inet_csk_destroy_sock */
728 sock_set_flag(sk, SOCK_DEAD);
729 percpu_counter_inc(sk->sk_prot->orphan_count);
730 inet_sk(sk)->inet_num = 0;
731}
732EXPORT_SYMBOL(inet_csk_prepare_forced_close);
733
Eric Dumazetf985c652015-10-14 06:16:49 -0700734int inet_csk_listen_start(struct sock *sk, int backlog)
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700735{
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700736 struct inet_connection_sock *icsk = inet_csk(sk);
Eric Dumazet10cbc8f2015-10-02 11:43:36 -0700737 struct inet_sock *inet = inet_sk(sk);
Craig Gallek086c6532016-02-10 11:50:35 -0500738 int err = -EADDRINUSE;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700739
Eric Dumazetef547f22015-10-02 11:43:37 -0700740 reqsk_queue_alloc(&icsk->icsk_accept_queue);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700741
Eric Dumazetf985c652015-10-14 06:16:49 -0700742 sk->sk_max_ack_backlog = backlog;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700743 sk->sk_ack_backlog = 0;
744 inet_csk_delack_init(sk);
745
746 /* There is race window here: we announce ourselves listening,
747 * but this transition is still not validated by get_port().
748 * It is OK, because this socket enters to hash table only
749 * after validation is complete.
750 */
Eric Dumazet00fd38d2015-11-12 08:43:18 -0800751 sk_state_store(sk, TCP_LISTEN);
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000752 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
753 inet->inet_sport = htons(inet->inet_num);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700754
755 sk_dst_reset(sk);
Craig Gallek086c6532016-02-10 11:50:35 -0500756 err = sk->sk_prot->hash(sk);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700757
Craig Gallek086c6532016-02-10 11:50:35 -0500758 if (likely(!err))
759 return 0;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700760 }
761
762 sk->sk_state = TCP_CLOSE;
Craig Gallek086c6532016-02-10 11:50:35 -0500763 return err;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700764}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700765EXPORT_SYMBOL_GPL(inet_csk_listen_start);
766
Eric Dumazetebb516a2015-10-14 11:16:28 -0700767static void inet_child_forget(struct sock *sk, struct request_sock *req,
768 struct sock *child)
769{
770 sk->sk_prot->disconnect(child, O_NONBLOCK);
771
772 sock_orphan(child);
773
774 percpu_counter_inc(sk->sk_prot->orphan_count);
775
776 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
777 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
778 BUG_ON(sk != req->rsk_listener);
779
780 /* Paranoid, to prevent race condition if
781 * an inbound pkt destined for child is
782 * blocked by sock lock in tcp_v4_rcv().
783 * Also to satisfy an assertion in
784 * tcp_v4_destroy_sock().
785 */
786 tcp_sk(child)->fastopen_rsk = NULL;
787 }
788 inet_csk_destroy_sock(child);
789 reqsk_put(req);
790}
791
Eric Dumazet77166822016-02-18 05:39:18 -0800792struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
793 struct request_sock *req,
794 struct sock *child)
Eric Dumazetebb516a2015-10-14 11:16:28 -0700795{
796 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
797
798 spin_lock(&queue->rskq_lock);
799 if (unlikely(sk->sk_state != TCP_LISTEN)) {
800 inet_child_forget(sk, req, child);
Eric Dumazet77166822016-02-18 05:39:18 -0800801 child = NULL;
Eric Dumazetebb516a2015-10-14 11:16:28 -0700802 } else {
803 req->sk = child;
804 req->dl_next = NULL;
805 if (queue->rskq_accept_head == NULL)
806 queue->rskq_accept_head = req;
807 else
808 queue->rskq_accept_tail->dl_next = req;
809 queue->rskq_accept_tail = req;
810 sk_acceptq_added(sk);
811 }
812 spin_unlock(&queue->rskq_lock);
Eric Dumazet77166822016-02-18 05:39:18 -0800813 return child;
Eric Dumazetebb516a2015-10-14 11:16:28 -0700814}
815EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
816
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700817struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
818 struct request_sock *req, bool own_req)
819{
820 if (own_req) {
821 inet_csk_reqsk_queue_drop(sk, req);
822 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
Eric Dumazet77166822016-02-18 05:39:18 -0800823 if (inet_csk_reqsk_queue_add(sk, req, child))
824 return child;
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700825 }
826 /* Too bad, another child took ownership of the request, undo. */
827 bh_unlock_sock(child);
828 sock_put(child);
829 return NULL;
830}
831EXPORT_SYMBOL(inet_csk_complete_hashdance);
832
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700833/*
834 * This routine closes sockets which have been at least partially
835 * opened, but not yet accepted.
836 */
837void inet_csk_listen_stop(struct sock *sk)
838{
839 struct inet_connection_sock *icsk = inet_csk(sk);
Jerry Chu83368862012-08-31 12:29:12 +0000840 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
Eric Dumazetfff1f302015-10-02 11:43:23 -0700841 struct request_sock *next, *req;
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700842
843 /* Following specs, it would be better either to send FIN
844 * (and enter FIN-WAIT-1, it is normal close)
845 * or to send active reset (abort).
846 * Certainly, it is pretty dangerous while synflood, but it is
847 * bad justification for our negligence 8)
848 * To be honest, we are not able to make either
849 * of the variants now. --ANK
850 */
Eric Dumazetfff1f302015-10-02 11:43:23 -0700851 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700852 struct sock *child = req->sk;
853
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700854 local_bh_disable();
855 bh_lock_sock(child);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700856 WARN_ON(sock_owned_by_user(child));
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700857 sock_hold(child);
858
Eric Dumazetebb516a2015-10-14 11:16:28 -0700859 inet_child_forget(sk, req, child);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700860 bh_unlock_sock(child);
861 local_bh_enable();
862 sock_put(child);
863
Eric Dumazet92d6f172015-10-02 11:43:38 -0700864 cond_resched();
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700865 }
Eric Dumazet0536fcc2015-09-29 07:42:52 -0700866 if (queue->fastopenq.rskq_rst_head) {
Jerry Chu83368862012-08-31 12:29:12 +0000867 /* Free all the reqs queued in rskq_rst_head. */
Eric Dumazet0536fcc2015-09-29 07:42:52 -0700868 spin_lock_bh(&queue->fastopenq.lock);
Eric Dumazetfff1f302015-10-02 11:43:23 -0700869 req = queue->fastopenq.rskq_rst_head;
Eric Dumazet0536fcc2015-09-29 07:42:52 -0700870 queue->fastopenq.rskq_rst_head = NULL;
871 spin_unlock_bh(&queue->fastopenq.lock);
Eric Dumazetfff1f302015-10-02 11:43:23 -0700872 while (req != NULL) {
873 next = req->dl_next;
Eric Dumazet13854e52015-03-15 21:12:16 -0700874 reqsk_put(req);
Eric Dumazetfff1f302015-10-02 11:43:23 -0700875 req = next;
Jerry Chu83368862012-08-31 12:29:12 +0000876 }
877 }
Eric Dumazetebb516a2015-10-14 11:16:28 -0700878 WARN_ON_ONCE(sk->sk_ack_backlog);
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700879}
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700880EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800881
882void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
883{
884 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
885 const struct inet_sock *inet = inet_sk(sk);
886
887 sin->sin_family = AF_INET;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000888 sin->sin_addr.s_addr = inet->inet_daddr;
889 sin->sin_port = inet->inet_dport;
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800890}
Arnaldo Carvalho de Meloaf05dc92005-12-13 23:16:04 -0800891EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
Arnaldo Carvalho de Meloc4d93902006-03-20 22:01:03 -0800892
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800893#ifdef CONFIG_COMPAT
894int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
895 char __user *optval, int __user *optlen)
896{
David S. Millerdbeff122006-03-20 22:52:32 -0800897 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800898
Ian Morris00db4122015-04-03 09:17:27 +0100899 if (icsk->icsk_af_ops->compat_getsockopt)
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800900 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
901 optval, optlen);
902 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
903 optval, optlen);
904}
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800905EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
906
907int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -0700908 char __user *optval, unsigned int optlen)
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800909{
David S. Millerdbeff122006-03-20 22:52:32 -0800910 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800911
Ian Morris00db4122015-04-03 09:17:27 +0100912 if (icsk->icsk_af_ops->compat_setsockopt)
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800913 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
914 optval, optlen);
915 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
916 optval, optlen);
917}
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -0800918EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
919#endif
David S. Miller80d0a692012-07-16 03:28:06 -0700920
921static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
922{
Eric Dumazet5abf7f72012-07-17 22:42:13 +0200923 const struct inet_sock *inet = inet_sk(sk);
924 const struct ip_options_rcu *inet_opt;
David S. Miller80d0a692012-07-16 03:28:06 -0700925 __be32 daddr = inet->inet_daddr;
926 struct flowi4 *fl4;
927 struct rtable *rt;
928
929 rcu_read_lock();
930 inet_opt = rcu_dereference(inet->inet_opt);
931 if (inet_opt && inet_opt->opt.srr)
932 daddr = inet_opt->opt.faddr;
933 fl4 = &fl->u.ip4;
934 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
935 inet->inet_saddr, inet->inet_dport,
936 inet->inet_sport, sk->sk_protocol,
937 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
938 if (IS_ERR(rt))
939 rt = NULL;
940 if (rt)
941 sk_setup_caps(sk, &rt->dst);
942 rcu_read_unlock();
943
944 return &rt->dst;
945}
946
947struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
948{
949 struct dst_entry *dst = __sk_dst_check(sk, 0);
950 struct inet_sock *inet = inet_sk(sk);
951
952 if (!dst) {
953 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
954 if (!dst)
955 goto out;
956 }
David S. Miller6700c272012-07-17 03:29:28 -0700957 dst->ops->update_pmtu(dst, sk, NULL, mtu);
David S. Miller80d0a692012-07-16 03:28:06 -0700958
959 dst = __sk_dst_check(sk, 0);
960 if (!dst)
961 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
962out:
963 return dst;
964}
965EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);