1 /*
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 
16 #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>
25 #include <net/xfrm.h>
26 
27 #ifdef INET_CSK_DEBUG
28 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
29 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
30 #endif
31 
32 unsigned long *sysctl_local_reserved_ports;
33 EXPORT_SYMBOL(sysctl_local_reserved_ports);
34 
35 void inet_get_local_port_range(struct net *net, int *low, int *high)
36 {
37 	unsigned int seq;
38 
39 	do {
40 		seq = read_seqbegin(&net->ipv4.sysctl_local_ports.lock);
41 
42 		*low = net->ipv4.sysctl_local_ports.range[0];
43 		*high = net->ipv4.sysctl_local_ports.range[1];
44 	} while (read_seqretry(&net->ipv4.sysctl_local_ports.lock, seq));
45 }
46 EXPORT_SYMBOL(inet_get_local_port_range);
47 
48 int inet_csk_bind_conflict(const struct sock *sk,
49 			   const struct inet_bind_bucket *tb, bool relax)
50 {
51 	struct sock *sk2;
52 	int reuse = sk->sk_reuse;
53 	int reuseport = sk->sk_reuseport;
54 	kuid_t uid = sock_i_uid((struct sock *)sk);
55 
56 	/*
57 	 * Unlike other sk lookup places we do not check
58 	 * for sk_net here, since _all_ the socks listed
59 	 * in tb->owners list belong to the same net - the
60 	 * one this bucket belongs to.
61 	 */
62 
63 	sk_for_each_bound(sk2, &tb->owners) {
64 		if (sk != sk2 &&
65 		    !inet_v6_ipv6only(sk2) &&
66 		    (!sk->sk_bound_dev_if ||
67 		     !sk2->sk_bound_dev_if ||
68 		     sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
69 			if ((!reuse || !sk2->sk_reuse ||
70 			    sk2->sk_state == TCP_LISTEN) &&
71 			    (!reuseport || !sk2->sk_reuseport ||
72 			    (sk2->sk_state != TCP_TIME_WAIT &&
73 			     !uid_eq(uid, sock_i_uid(sk2))))) {
74 
75 				if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
76 				    sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
77 					break;
78 			}
79 			if (!relax && reuse && sk2->sk_reuse &&
80 			    sk2->sk_state != TCP_LISTEN) {
81 
82 				if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
83 				    sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
84 					break;
85 			}
86 		}
87 	}
88 	return sk2 != NULL;
89 }
90 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
91 
92 /* Obtain a reference to a local port for the given sock,
93  * if snum is zero it means select any available local port.
94  */
95 int inet_csk_get_port(struct sock *sk, unsigned short snum)
96 {
97 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
98 	struct inet_bind_hashbucket *head;
99 	struct inet_bind_bucket *tb;
100 	int ret, attempts = 5;
101 	struct net *net = sock_net(sk);
102 	int smallest_size = -1, smallest_rover;
103 	kuid_t uid = sock_i_uid(sk);
104 
105 	local_bh_disable();
106 	if (!snum) {
107 		int remaining, rover, low, high;
108 
109 again:
110 		inet_get_local_port_range(net, &low, &high);
111 		remaining = (high - low) + 1;
112 		smallest_rover = rover = net_random() % remaining + low;
113 
114 		smallest_size = -1;
115 		do {
116 			if (inet_is_reserved_local_port(rover))
117 				goto next_nolock;
118 			head = &hashinfo->bhash[inet_bhashfn(net, rover,
119 					hashinfo->bhash_size)];
120 			spin_lock(&head->lock);
121 			inet_bind_bucket_for_each(tb, &head->chain)
122 				if (net_eq(ib_net(tb), net) && tb->port == rover) {
123 					if (((tb->fastreuse > 0 &&
124 					      sk->sk_reuse &&
125 					      sk->sk_state != TCP_LISTEN) ||
126 					     (tb->fastreuseport > 0 &&
127 					      sk->sk_reuseport &&
128 					      uid_eq(tb->fastuid, uid))) &&
129 					    (tb->num_owners < smallest_size || smallest_size == -1)) {
130 						smallest_size = tb->num_owners;
131 						smallest_rover = rover;
132 						if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&
133 						    !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
134 							snum = smallest_rover;
135 							goto tb_found;
136 						}
137 					}
138 					if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
139 						snum = rover;
140 						goto tb_found;
141 					}
142 					goto next;
143 				}
144 			break;
145 		next:
146 			spin_unlock(&head->lock);
147 		next_nolock:
148 			if (++rover > high)
149 				rover = low;
150 		} while (--remaining > 0);
151 
152 		/* Exhausted local port range during search?  It is not
153 		 * possible for us to be holding one of the bind hash
154 		 * locks if this test triggers, because if 'remaining'
155 		 * drops to zero, we broke out of the do/while loop at
156 		 * the top level, not from the 'break;' statement.
157 		 */
158 		ret = 1;
159 		if (remaining <= 0) {
160 			if (smallest_size != -1) {
161 				snum = smallest_rover;
162 				goto have_snum;
163 			}
164 			goto fail;
165 		}
166 		/* OK, here is the one we will use.  HEAD is
167 		 * non-NULL and we hold it's mutex.
168 		 */
169 		snum = rover;
170 	} else {
171 have_snum:
172 		head = &hashinfo->bhash[inet_bhashfn(net, snum,
173 				hashinfo->bhash_size)];
174 		spin_lock(&head->lock);
175 		inet_bind_bucket_for_each(tb, &head->chain)
176 			if (net_eq(ib_net(tb), net) && tb->port == snum)
177 				goto tb_found;
178 	}
179 	tb = NULL;
180 	goto tb_not_found;
181 tb_found:
182 	if (!hlist_empty(&tb->owners)) {
183 		if (sk->sk_reuse == SK_FORCE_REUSE)
184 			goto success;
185 
186 		if (((tb->fastreuse > 0 &&
187 		      sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
188 		     (tb->fastreuseport > 0 &&
189 		      sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
190 		    smallest_size == -1) {
191 			goto success;
192 		} else {
193 			ret = 1;
194 			if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
195 				if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
196 				     (tb->fastreuseport > 0 &&
197 				      sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
198 				    smallest_size != -1 && --attempts >= 0) {
199 					spin_unlock(&head->lock);
200 					goto again;
201 				}
202 
203 				goto fail_unlock;
204 			}
205 		}
206 	}
207 tb_not_found:
208 	ret = 1;
209 	if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
210 					net, head, snum)) == NULL)
211 		goto fail_unlock;
212 	if (hlist_empty(&tb->owners)) {
213 		if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
214 			tb->fastreuse = 1;
215 		else
216 			tb->fastreuse = 0;
217 		if (sk->sk_reuseport) {
218 			tb->fastreuseport = 1;
219 			tb->fastuid = uid;
220 		} else
221 			tb->fastreuseport = 0;
222 	} else {
223 		if (tb->fastreuse &&
224 		    (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
225 			tb->fastreuse = 0;
226 		if (tb->fastreuseport &&
227 		    (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid)))
228 			tb->fastreuseport = 0;
229 	}
230 success:
231 	if (!inet_csk(sk)->icsk_bind_hash)
232 		inet_bind_hash(sk, tb, snum);
233 	WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
234 	ret = 0;
235 
236 fail_unlock:
237 	spin_unlock(&head->lock);
238 fail:
239 	local_bh_enable();
240 	return ret;
241 }
242 EXPORT_SYMBOL_GPL(inet_csk_get_port);
243 
244 /*
245  * Wait for an incoming connection, avoid race conditions. This must be called
246  * with the socket locked.
247  */
248 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
249 {
250 	struct inet_connection_sock *icsk = inet_csk(sk);
251 	DEFINE_WAIT(wait);
252 	int err;
253 
254 	/*
255 	 * True wake-one mechanism for incoming connections: only
256 	 * one process gets woken up, not the 'whole herd'.
257 	 * Since we do not 'race & poll' for established sockets
258 	 * anymore, the common case will execute the loop only once.
259 	 *
260 	 * Subtle issue: "add_wait_queue_exclusive()" will be added
261 	 * after any current non-exclusive waiters, and we know that
262 	 * it will always _stay_ after any new non-exclusive waiters
263 	 * because all non-exclusive waiters are added at the
264 	 * beginning of the wait-queue. As such, it's ok to "drop"
265 	 * our exclusiveness temporarily when we get woken up without
266 	 * having to remove and re-insert us on the wait queue.
267 	 */
268 	for (;;) {
269 		prepare_to_wait_exclusive(sk_sleep(sk), &wait,
270 					  TASK_INTERRUPTIBLE);
271 		release_sock(sk);
272 		if (reqsk_queue_empty(&icsk->icsk_accept_queue))
273 			timeo = schedule_timeout(timeo);
274 		lock_sock(sk);
275 		err = 0;
276 		if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
277 			break;
278 		err = -EINVAL;
279 		if (sk->sk_state != TCP_LISTEN)
280 			break;
281 		err = sock_intr_errno(timeo);
282 		if (signal_pending(current))
283 			break;
284 		err = -EAGAIN;
285 		if (!timeo)
286 			break;
287 	}
288 	finish_wait(sk_sleep(sk), &wait);
289 	return err;
290 }
291 
292 /*
293  * This will accept the next outstanding connection.
294  */
295 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
296 {
297 	struct inet_connection_sock *icsk = inet_csk(sk);
298 	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
299 	struct sock *newsk;
300 	struct request_sock *req;
301 	int error;
302 
303 	lock_sock(sk);
304 
305 	/* We need to make sure that this socket is listening,
306 	 * and that it has something pending.
307 	 */
308 	error = -EINVAL;
309 	if (sk->sk_state != TCP_LISTEN)
310 		goto out_err;
311 
312 	/* Find already established connection */
313 	if (reqsk_queue_empty(queue)) {
314 		long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
315 
316 		/* If this is a non blocking socket don't sleep */
317 		error = -EAGAIN;
318 		if (!timeo)
319 			goto out_err;
320 
321 		error = inet_csk_wait_for_connect(sk, timeo);
322 		if (error)
323 			goto out_err;
324 	}
325 	req = reqsk_queue_remove(queue);
326 	newsk = req->sk;
327 
328 	sk_acceptq_removed(sk);
329 	if (sk->sk_protocol == IPPROTO_TCP && queue->fastopenq != NULL) {
330 		spin_lock_bh(&queue->fastopenq->lock);
331 		if (tcp_rsk(req)->listener) {
332 			/* We are still waiting for the final ACK from 3WHS
333 			 * so can't free req now. Instead, we set req->sk to
334 			 * NULL to signify that the child socket is taken
335 			 * so reqsk_fastopen_remove() will free the req
336 			 * when 3WHS finishes (or is aborted).
337 			 */
338 			req->sk = NULL;
339 			req = NULL;
340 		}
341 		spin_unlock_bh(&queue->fastopenq->lock);
342 	}
343 out:
344 	release_sock(sk);
345 	if (req)
346 		__reqsk_free(req);
347 	return newsk;
348 out_err:
349 	newsk = NULL;
350 	req = NULL;
351 	*err = error;
352 	goto out;
353 }
354 EXPORT_SYMBOL(inet_csk_accept);
355 
356 /*
357  * Using different timers for retransmit, delayed acks and probes
358  * We may wish use just one timer maintaining a list of expire jiffies
359  * to optimize.
360  */
361 void inet_csk_init_xmit_timers(struct sock *sk,
362 			       void (*retransmit_handler)(unsigned long),
363 			       void (*delack_handler)(unsigned long),
364 			       void (*keepalive_handler)(unsigned long))
365 {
366 	struct inet_connection_sock *icsk = inet_csk(sk);
367 
368 	setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
369 			(unsigned long)sk);
370 	setup_timer(&icsk->icsk_delack_timer, delack_handler,
371 			(unsigned long)sk);
372 	setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
373 	icsk->icsk_pending = icsk->icsk_ack.pending = 0;
374 }
375 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
376 
377 void inet_csk_clear_xmit_timers(struct sock *sk)
378 {
379 	struct inet_connection_sock *icsk = inet_csk(sk);
380 
381 	icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
382 
383 	sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
384 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
385 	sk_stop_timer(sk, &sk->sk_timer);
386 }
387 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
388 
389 void inet_csk_delete_keepalive_timer(struct sock *sk)
390 {
391 	sk_stop_timer(sk, &sk->sk_timer);
392 }
393 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
394 
395 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
396 {
397 	sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
398 }
399 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
400 
401 struct dst_entry *inet_csk_route_req(struct sock *sk,
402 				     struct flowi4 *fl4,
403 				     const struct request_sock *req)
404 {
405 	struct rtable *rt;
406 	const struct inet_request_sock *ireq = inet_rsk(req);
407 	struct ip_options_rcu *opt = inet_rsk(req)->opt;
408 	struct net *net = sock_net(sk);
409 	int flags = inet_sk_flowi_flags(sk);
410 
411 	flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
412 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
413 			   sk->sk_protocol,
414 			   flags,
415 			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
416 			   ireq->ir_loc_addr, ireq->ir_rmt_port, inet_sk(sk)->inet_sport);
417 	security_req_classify_flow(req, flowi4_to_flowi(fl4));
418 	rt = ip_route_output_flow(net, fl4, sk);
419 	if (IS_ERR(rt))
420 		goto no_route;
421 	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
422 		goto route_err;
423 	return &rt->dst;
424 
425 route_err:
426 	ip_rt_put(rt);
427 no_route:
428 	IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
429 	return NULL;
430 }
431 EXPORT_SYMBOL_GPL(inet_csk_route_req);
432 
433 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
434 					    struct sock *newsk,
435 					    const struct request_sock *req)
436 {
437 	const struct inet_request_sock *ireq = inet_rsk(req);
438 	struct inet_sock *newinet = inet_sk(newsk);
439 	struct ip_options_rcu *opt;
440 	struct net *net = sock_net(sk);
441 	struct flowi4 *fl4;
442 	struct rtable *rt;
443 
444 	fl4 = &newinet->cork.fl.u.ip4;
445 
446 	rcu_read_lock();
447 	opt = rcu_dereference(newinet->inet_opt);
448 	flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
449 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
450 			   sk->sk_protocol, inet_sk_flowi_flags(sk),
451 			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
452 			   ireq->ir_loc_addr, ireq->ir_rmt_port, inet_sk(sk)->inet_sport);
453 	security_req_classify_flow(req, flowi4_to_flowi(fl4));
454 	rt = ip_route_output_flow(net, fl4, sk);
455 	if (IS_ERR(rt))
456 		goto no_route;
457 	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
458 		goto route_err;
459 	rcu_read_unlock();
460 	return &rt->dst;
461 
462 route_err:
463 	ip_rt_put(rt);
464 no_route:
465 	rcu_read_unlock();
466 	IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
467 	return NULL;
468 }
469 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
470 
471 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
472 				 const u32 rnd, const u32 synq_hsize)
473 {
474 	return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
475 }
476 
477 #if IS_ENABLED(CONFIG_IPV6)
478 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
479 #else
480 #define AF_INET_FAMILY(fam) 1
481 #endif
482 
483 struct request_sock *inet_csk_search_req(const struct sock *sk,
484 					 struct request_sock ***prevp,
485 					 const __be16 rport, const __be32 raddr,
486 					 const __be32 laddr)
487 {
488 	const struct inet_connection_sock *icsk = inet_csk(sk);
489 	struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
490 	struct request_sock *req, **prev;
491 
492 	for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
493 						    lopt->nr_table_entries)];
494 	     (req = *prev) != NULL;
495 	     prev = &req->dl_next) {
496 		const struct inet_request_sock *ireq = inet_rsk(req);
497 
498 		if (ireq->ir_rmt_port == rport &&
499 		    ireq->ir_rmt_addr == raddr &&
500 		    ireq->ir_loc_addr == laddr &&
501 		    AF_INET_FAMILY(req->rsk_ops->family)) {
502 			WARN_ON(req->sk);
503 			*prevp = prev;
504 			break;
505 		}
506 	}
507 
508 	return req;
509 }
510 EXPORT_SYMBOL_GPL(inet_csk_search_req);
511 
512 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
513 				   unsigned long timeout)
514 {
515 	struct inet_connection_sock *icsk = inet_csk(sk);
516 	struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
517 	const u32 h = inet_synq_hash(inet_rsk(req)->ir_rmt_addr,
518 				     inet_rsk(req)->ir_rmt_port,
519 				     lopt->hash_rnd, lopt->nr_table_entries);
520 
521 	reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
522 	inet_csk_reqsk_queue_added(sk, timeout);
523 }
524 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
525 
526 /* Only thing we need from tcp.h */
527 extern int sysctl_tcp_synack_retries;
528 
529 
530 /* Decide when to expire the request and when to resend SYN-ACK */
531 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
532 				  const int max_retries,
533 				  const u8 rskq_defer_accept,
534 				  int *expire, int *resend)
535 {
536 	if (!rskq_defer_accept) {
537 		*expire = req->num_timeout >= thresh;
538 		*resend = 1;
539 		return;
540 	}
541 	*expire = req->num_timeout >= thresh &&
542 		  (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
543 	/*
544 	 * Do not resend while waiting for data after ACK,
545 	 * start to resend on end of deferring period to give
546 	 * last chance for data or ACK to create established socket.
547 	 */
548 	*resend = !inet_rsk(req)->acked ||
549 		  req->num_timeout >= rskq_defer_accept - 1;
550 }
551 
552 int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req)
553 {
554 	int err = req->rsk_ops->rtx_syn_ack(parent, req);
555 
556 	if (!err)
557 		req->num_retrans++;
558 	return err;
559 }
560 EXPORT_SYMBOL(inet_rtx_syn_ack);
561 
562 void inet_csk_reqsk_queue_prune(struct sock *parent,
563 				const unsigned long interval,
564 				const unsigned long timeout,
565 				const unsigned long max_rto)
566 {
567 	struct inet_connection_sock *icsk = inet_csk(parent);
568 	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
569 	struct listen_sock *lopt = queue->listen_opt;
570 	int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
571 	int thresh = max_retries;
572 	unsigned long now = jiffies;
573 	struct request_sock **reqp, *req;
574 	int i, budget;
575 
576 	if (lopt == NULL || lopt->qlen == 0)
577 		return;
578 
579 	/* Normally all the openreqs are young and become mature
580 	 * (i.e. converted to established socket) for first timeout.
581 	 * If synack was not acknowledged for 1 second, it means
582 	 * one of the following things: synack was lost, ack was lost,
583 	 * rtt is high or nobody planned to ack (i.e. synflood).
584 	 * When server is a bit loaded, queue is populated with old
585 	 * open requests, reducing effective size of queue.
586 	 * When server is well loaded, queue size reduces to zero
587 	 * after several minutes of work. It is not synflood,
588 	 * it is normal operation. The solution is pruning
589 	 * too old entries overriding normal timeout, when
590 	 * situation becomes dangerous.
591 	 *
592 	 * Essentially, we reserve half of room for young
593 	 * embrions; and abort old ones without pity, if old
594 	 * ones are about to clog our table.
595 	 */
596 	if (lopt->qlen>>(lopt->max_qlen_log-1)) {
597 		int young = (lopt->qlen_young<<1);
598 
599 		while (thresh > 2) {
600 			if (lopt->qlen < young)
601 				break;
602 			thresh--;
603 			young <<= 1;
604 		}
605 	}
606 
607 	if (queue->rskq_defer_accept)
608 		max_retries = queue->rskq_defer_accept;
609 
610 	budget = 2 * (lopt->nr_table_entries / (timeout / interval));
611 	i = lopt->clock_hand;
612 
613 	do {
614 		reqp=&lopt->syn_table[i];
615 		while ((req = *reqp) != NULL) {
616 			if (time_after_eq(now, req->expires)) {
617 				int expire = 0, resend = 0;
618 
619 				syn_ack_recalc(req, thresh, max_retries,
620 					       queue->rskq_defer_accept,
621 					       &expire, &resend);
622 				req->rsk_ops->syn_ack_timeout(parent, req);
623 				if (!expire &&
624 				    (!resend ||
625 				     !inet_rtx_syn_ack(parent, req) ||
626 				     inet_rsk(req)->acked)) {
627 					unsigned long timeo;
628 
629 					if (req->num_timeout++ == 0)
630 						lopt->qlen_young--;
631 					timeo = min(timeout << req->num_timeout,
632 						    max_rto);
633 					req->expires = now + timeo;
634 					reqp = &req->dl_next;
635 					continue;
636 				}
637 
638 				/* Drop this request */
639 				inet_csk_reqsk_queue_unlink(parent, req, reqp);
640 				reqsk_queue_removed(queue, req);
641 				reqsk_free(req);
642 				continue;
643 			}
644 			reqp = &req->dl_next;
645 		}
646 
647 		i = (i + 1) & (lopt->nr_table_entries - 1);
648 
649 	} while (--budget > 0);
650 
651 	lopt->clock_hand = i;
652 
653 	if (lopt->qlen)
654 		inet_csk_reset_keepalive_timer(parent, interval);
655 }
656 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
657 
658 /**
659  *	inet_csk_clone_lock - clone an inet socket, and lock its clone
660  *	@sk: the socket to clone
661  *	@req: request_sock
662  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
663  *
664  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
665  */
666 struct sock *inet_csk_clone_lock(const struct sock *sk,
667 				 const struct request_sock *req,
668 				 const gfp_t priority)
669 {
670 	struct sock *newsk = sk_clone_lock(sk, priority);
671 
672 	if (newsk != NULL) {
673 		struct inet_connection_sock *newicsk = inet_csk(newsk);
674 
675 		newsk->sk_state = TCP_SYN_RECV;
676 		newicsk->icsk_bind_hash = NULL;
677 
678 		inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
679 		inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
680 		inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
681 		newsk->sk_write_space = sk_stream_write_space;
682 
683 		newicsk->icsk_retransmits = 0;
684 		newicsk->icsk_backoff	  = 0;
685 		newicsk->icsk_probes_out  = 0;
686 
687 		/* Deinitialize accept_queue to trap illegal accesses. */
688 		memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
689 
690 		security_inet_csk_clone(newsk, req);
691 	}
692 	return newsk;
693 }
694 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
695 
696 /*
697  * At this point, there should be no process reference to this
698  * socket, and thus no user references at all.  Therefore we
699  * can assume the socket waitqueue is inactive and nobody will
700  * try to jump onto it.
701  */
702 void inet_csk_destroy_sock(struct sock *sk)
703 {
704 	WARN_ON(sk->sk_state != TCP_CLOSE);
705 	WARN_ON(!sock_flag(sk, SOCK_DEAD));
706 
707 	/* It cannot be in hash table! */
708 	WARN_ON(!sk_unhashed(sk));
709 
710 	/* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
711 	WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
712 
713 	sk->sk_prot->destroy(sk);
714 
715 	sk_stream_kill_queues(sk);
716 
717 	xfrm_sk_free_policy(sk);
718 
719 	sk_refcnt_debug_release(sk);
720 
721 	percpu_counter_dec(sk->sk_prot->orphan_count);
722 	sock_put(sk);
723 }
724 EXPORT_SYMBOL(inet_csk_destroy_sock);
725 
726 /* This function allows to force a closure of a socket after the call to
727  * tcp/dccp_create_openreq_child().
728  */
729 void inet_csk_prepare_forced_close(struct sock *sk)
730 	__releases(&sk->sk_lock.slock)
731 {
732 	/* sk_clone_lock locked the socket and set refcnt to 2 */
733 	bh_unlock_sock(sk);
734 	sock_put(sk);
735 
736 	/* The below has to be done to allow calling inet_csk_destroy_sock */
737 	sock_set_flag(sk, SOCK_DEAD);
738 	percpu_counter_inc(sk->sk_prot->orphan_count);
739 	inet_sk(sk)->inet_num = 0;
740 }
741 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
742 
743 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
744 {
745 	struct inet_sock *inet = inet_sk(sk);
746 	struct inet_connection_sock *icsk = inet_csk(sk);
747 	int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
748 
749 	if (rc != 0)
750 		return rc;
751 
752 	sk->sk_max_ack_backlog = 0;
753 	sk->sk_ack_backlog = 0;
754 	inet_csk_delack_init(sk);
755 
756 	/* There is race window here: we announce ourselves listening,
757 	 * but this transition is still not validated by get_port().
758 	 * It is OK, because this socket enters to hash table only
759 	 * after validation is complete.
760 	 */
761 	sk->sk_state = TCP_LISTEN;
762 	if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
763 		inet->inet_sport = htons(inet->inet_num);
764 
765 		sk_dst_reset(sk);
766 		sk->sk_prot->hash(sk);
767 
768 		return 0;
769 	}
770 
771 	sk->sk_state = TCP_CLOSE;
772 	__reqsk_queue_destroy(&icsk->icsk_accept_queue);
773 	return -EADDRINUSE;
774 }
775 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
776 
777 /*
778  *	This routine closes sockets which have been at least partially
779  *	opened, but not yet accepted.
780  */
781 void inet_csk_listen_stop(struct sock *sk)
782 {
783 	struct inet_connection_sock *icsk = inet_csk(sk);
784 	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
785 	struct request_sock *acc_req;
786 	struct request_sock *req;
787 
788 	inet_csk_delete_keepalive_timer(sk);
789 
790 	/* make all the listen_opt local to us */
791 	acc_req = reqsk_queue_yank_acceptq(queue);
792 
793 	/* Following specs, it would be better either to send FIN
794 	 * (and enter FIN-WAIT-1, it is normal close)
795 	 * or to send active reset (abort).
796 	 * Certainly, it is pretty dangerous while synflood, but it is
797 	 * bad justification for our negligence 8)
798 	 * To be honest, we are not able to make either
799 	 * of the variants now.			--ANK
800 	 */
801 	reqsk_queue_destroy(queue);
802 
803 	while ((req = acc_req) != NULL) {
804 		struct sock *child = req->sk;
805 
806 		acc_req = req->dl_next;
807 
808 		local_bh_disable();
809 		bh_lock_sock(child);
810 		WARN_ON(sock_owned_by_user(child));
811 		sock_hold(child);
812 
813 		sk->sk_prot->disconnect(child, O_NONBLOCK);
814 
815 		sock_orphan(child);
816 
817 		percpu_counter_inc(sk->sk_prot->orphan_count);
818 
819 		if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->listener) {
820 			BUG_ON(tcp_sk(child)->fastopen_rsk != req);
821 			BUG_ON(sk != tcp_rsk(req)->listener);
822 
823 			/* Paranoid, to prevent race condition if
824 			 * an inbound pkt destined for child is
825 			 * blocked by sock lock in tcp_v4_rcv().
826 			 * Also to satisfy an assertion in
827 			 * tcp_v4_destroy_sock().
828 			 */
829 			tcp_sk(child)->fastopen_rsk = NULL;
830 			sock_put(sk);
831 		}
832 		inet_csk_destroy_sock(child);
833 
834 		bh_unlock_sock(child);
835 		local_bh_enable();
836 		sock_put(child);
837 
838 		sk_acceptq_removed(sk);
839 		__reqsk_free(req);
840 	}
841 	if (queue->fastopenq != NULL) {
842 		/* Free all the reqs queued in rskq_rst_head. */
843 		spin_lock_bh(&queue->fastopenq->lock);
844 		acc_req = queue->fastopenq->rskq_rst_head;
845 		queue->fastopenq->rskq_rst_head = NULL;
846 		spin_unlock_bh(&queue->fastopenq->lock);
847 		while ((req = acc_req) != NULL) {
848 			acc_req = req->dl_next;
849 			__reqsk_free(req);
850 		}
851 	}
852 	WARN_ON(sk->sk_ack_backlog);
853 }
854 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
855 
856 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
857 {
858 	struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
859 	const struct inet_sock *inet = inet_sk(sk);
860 
861 	sin->sin_family		= AF_INET;
862 	sin->sin_addr.s_addr	= inet->inet_daddr;
863 	sin->sin_port		= inet->inet_dport;
864 }
865 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
866 
867 #ifdef CONFIG_COMPAT
868 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
869 			       char __user *optval, int __user *optlen)
870 {
871 	const struct inet_connection_sock *icsk = inet_csk(sk);
872 
873 	if (icsk->icsk_af_ops->compat_getsockopt != NULL)
874 		return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
875 							    optval, optlen);
876 	return icsk->icsk_af_ops->getsockopt(sk, level, optname,
877 					     optval, optlen);
878 }
879 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
880 
881 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
882 			       char __user *optval, unsigned int optlen)
883 {
884 	const struct inet_connection_sock *icsk = inet_csk(sk);
885 
886 	if (icsk->icsk_af_ops->compat_setsockopt != NULL)
887 		return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
888 							    optval, optlen);
889 	return icsk->icsk_af_ops->setsockopt(sk, level, optname,
890 					     optval, optlen);
891 }
892 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
893 #endif
894 
895 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
896 {
897 	const struct inet_sock *inet = inet_sk(sk);
898 	const struct ip_options_rcu *inet_opt;
899 	__be32 daddr = inet->inet_daddr;
900 	struct flowi4 *fl4;
901 	struct rtable *rt;
902 
903 	rcu_read_lock();
904 	inet_opt = rcu_dereference(inet->inet_opt);
905 	if (inet_opt && inet_opt->opt.srr)
906 		daddr = inet_opt->opt.faddr;
907 	fl4 = &fl->u.ip4;
908 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
909 				   inet->inet_saddr, inet->inet_dport,
910 				   inet->inet_sport, sk->sk_protocol,
911 				   RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
912 	if (IS_ERR(rt))
913 		rt = NULL;
914 	if (rt)
915 		sk_setup_caps(sk, &rt->dst);
916 	rcu_read_unlock();
917 
918 	return &rt->dst;
919 }
920 
921 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
922 {
923 	struct dst_entry *dst = __sk_dst_check(sk, 0);
924 	struct inet_sock *inet = inet_sk(sk);
925 
926 	if (!dst) {
927 		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
928 		if (!dst)
929 			goto out;
930 	}
931 	dst->ops->update_pmtu(dst, sk, NULL, mtu);
932 
933 	dst = __sk_dst_check(sk, 0);
934 	if (!dst)
935 		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
936 out:
937 	return dst;
938 }
939 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
940