1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Support for INET connection oriented protocols.
8  *
9  * Authors:	See the TCP sources
10  */
11 
12 #include <linux/module.h>
13 #include <linux/jhash.h>
14 
15 #include <net/inet_connection_sock.h>
16 #include <net/inet_hashtables.h>
17 #include <net/inet_timewait_sock.h>
18 #include <net/ip.h>
19 #include <net/route.h>
20 #include <net/tcp_states.h>
21 #include <net/xfrm.h>
22 #include <net/tcp.h>
23 #include <net/sock_reuseport.h>
24 #include <net/addrconf.h>
25 
26 #if IS_ENABLED(CONFIG_IPV6)
27 /* match_sk*_wildcard == true:  IPV6_ADDR_ANY equals to any IPv6 addresses
28  *				if IPv6 only, and any IPv4 addresses
29  *				if not IPv6 only
30  * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
31  *				IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
32  *				and 0.0.0.0 equals to 0.0.0.0 only
33  */
34 static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
35 				 const struct in6_addr *sk2_rcv_saddr6,
36 				 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
37 				 bool sk1_ipv6only, bool sk2_ipv6only,
38 				 bool match_sk1_wildcard,
39 				 bool match_sk2_wildcard)
40 {
41 	int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
42 	int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
43 
44 	/* if both are mapped, treat as IPv4 */
45 	if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
46 		if (!sk2_ipv6only) {
47 			if (sk1_rcv_saddr == sk2_rcv_saddr)
48 				return true;
49 			return (match_sk1_wildcard && !sk1_rcv_saddr) ||
50 				(match_sk2_wildcard && !sk2_rcv_saddr);
51 		}
52 		return false;
53 	}
54 
55 	if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
56 		return true;
57 
58 	if (addr_type2 == IPV6_ADDR_ANY && match_sk2_wildcard &&
59 	    !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
60 		return true;
61 
62 	if (addr_type == IPV6_ADDR_ANY && match_sk1_wildcard &&
63 	    !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
64 		return true;
65 
66 	if (sk2_rcv_saddr6 &&
67 	    ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
68 		return true;
69 
70 	return false;
71 }
72 #endif
73 
74 /* match_sk*_wildcard == true:  0.0.0.0 equals to any IPv4 addresses
75  * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
76  *				0.0.0.0 only equals to 0.0.0.0
77  */
78 static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
79 				 bool sk2_ipv6only, bool match_sk1_wildcard,
80 				 bool match_sk2_wildcard)
81 {
82 	if (!sk2_ipv6only) {
83 		if (sk1_rcv_saddr == sk2_rcv_saddr)
84 			return true;
85 		return (match_sk1_wildcard && !sk1_rcv_saddr) ||
86 			(match_sk2_wildcard && !sk2_rcv_saddr);
87 	}
88 	return false;
89 }
90 
91 bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
92 			  bool match_wildcard)
93 {
94 #if IS_ENABLED(CONFIG_IPV6)
95 	if (sk->sk_family == AF_INET6)
96 		return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
97 					    inet6_rcv_saddr(sk2),
98 					    sk->sk_rcv_saddr,
99 					    sk2->sk_rcv_saddr,
100 					    ipv6_only_sock(sk),
101 					    ipv6_only_sock(sk2),
102 					    match_wildcard,
103 					    match_wildcard);
104 #endif
105 	return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
106 				    ipv6_only_sock(sk2), match_wildcard,
107 				    match_wildcard);
108 }
109 EXPORT_SYMBOL(inet_rcv_saddr_equal);
110 
111 bool inet_rcv_saddr_any(const struct sock *sk)
112 {
113 #if IS_ENABLED(CONFIG_IPV6)
114 	if (sk->sk_family == AF_INET6)
115 		return ipv6_addr_any(&sk->sk_v6_rcv_saddr);
116 #endif
117 	return !sk->sk_rcv_saddr;
118 }
119 
120 static bool use_bhash2_on_bind(const struct sock *sk)
121 {
122 #if IS_ENABLED(CONFIG_IPV6)
123 	int addr_type;
124 
125 	if (sk->sk_family == AF_INET6) {
126 		addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
127 		return addr_type != IPV6_ADDR_ANY &&
128 			addr_type != IPV6_ADDR_MAPPED;
129 	}
130 #endif
131 	return sk->sk_rcv_saddr != htonl(INADDR_ANY);
132 }
133 
134 static u32 get_bhash2_nulladdr_hash(const struct sock *sk, struct net *net,
135 				    int port)
136 {
137 #if IS_ENABLED(CONFIG_IPV6)
138 	struct in6_addr nulladdr = {};
139 
140 	if (sk->sk_family == AF_INET6)
141 		return ipv6_portaddr_hash(net, &nulladdr, port);
142 #endif
143 	return ipv4_portaddr_hash(net, 0, port);
144 }
145 
146 void inet_get_local_port_range(struct net *net, int *low, int *high)
147 {
148 	unsigned int seq;
149 
150 	do {
151 		seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
152 
153 		*low = net->ipv4.ip_local_ports.range[0];
154 		*high = net->ipv4.ip_local_ports.range[1];
155 	} while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
156 }
157 EXPORT_SYMBOL(inet_get_local_port_range);
158 
159 static bool bind_conflict_exist(const struct sock *sk, struct sock *sk2,
160 				kuid_t sk_uid, bool relax,
161 				bool reuseport_cb_ok, bool reuseport_ok)
162 {
163 	int bound_dev_if2;
164 
165 	if (sk == sk2)
166 		return false;
167 
168 	bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
169 
170 	if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
171 	    sk->sk_bound_dev_if == bound_dev_if2) {
172 		if (sk->sk_reuse && sk2->sk_reuse &&
173 		    sk2->sk_state != TCP_LISTEN) {
174 			if (!relax || (!reuseport_ok && sk->sk_reuseport &&
175 				       sk2->sk_reuseport && reuseport_cb_ok &&
176 				       (sk2->sk_state == TCP_TIME_WAIT ||
177 					uid_eq(sk_uid, sock_i_uid(sk2)))))
178 				return true;
179 		} else if (!reuseport_ok || !sk->sk_reuseport ||
180 			   !sk2->sk_reuseport || !reuseport_cb_ok ||
181 			   (sk2->sk_state != TCP_TIME_WAIT &&
182 			    !uid_eq(sk_uid, sock_i_uid(sk2)))) {
183 			return true;
184 		}
185 	}
186 	return false;
187 }
188 
189 static bool check_bhash2_conflict(const struct sock *sk,
190 				  struct inet_bind2_bucket *tb2, kuid_t sk_uid,
191 				  bool relax, bool reuseport_cb_ok,
192 				  bool reuseport_ok)
193 {
194 	struct sock *sk2;
195 
196 	sk_for_each_bound_bhash2(sk2, &tb2->owners) {
197 		if (sk->sk_family == AF_INET && ipv6_only_sock(sk2))
198 			continue;
199 
200 		if (bind_conflict_exist(sk, sk2, sk_uid, relax,
201 					reuseport_cb_ok, reuseport_ok))
202 			return true;
203 	}
204 	return false;
205 }
206 
207 /* This should be called only when the corresponding inet_bind_bucket spinlock
208  * is held
209  */
210 static int inet_csk_bind_conflict(const struct sock *sk, int port,
211 				  struct inet_bind_bucket *tb,
212 				  struct inet_bind2_bucket *tb2, /* may be null */
213 				  bool relax, bool reuseport_ok)
214 {
215 	struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
216 	kuid_t uid = sock_i_uid((struct sock *)sk);
217 	struct sock_reuseport *reuseport_cb;
218 	struct inet_bind2_hashbucket *head2;
219 	bool reuseport_cb_ok;
220 	struct sock *sk2;
221 	struct net *net;
222 	int l3mdev;
223 	u32 hash;
224 
225 	rcu_read_lock();
226 	reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
227 	/* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
228 	reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
229 	rcu_read_unlock();
230 
231 	/*
232 	 * Unlike other sk lookup places we do not check
233 	 * for sk_net here, since _all_ the socks listed
234 	 * in tb->owners and tb2->owners list belong
235 	 * to the same net
236 	 */
237 
238 	if (!use_bhash2_on_bind(sk)) {
239 		sk_for_each_bound(sk2, &tb->owners)
240 			if (bind_conflict_exist(sk, sk2, uid, relax,
241 						reuseport_cb_ok, reuseport_ok) &&
242 			    inet_rcv_saddr_equal(sk, sk2, true))
243 				return true;
244 
245 		return false;
246 	}
247 
248 	if (tb2 && check_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
249 					 reuseport_ok))
250 		return true;
251 
252 	net = sock_net(sk);
253 
254 	/* check there's no conflict with an existing IPV6_ADDR_ANY (if ipv6) or
255 	 * INADDR_ANY (if ipv4) socket.
256 	 */
257 	hash = get_bhash2_nulladdr_hash(sk, net, port);
258 	head2 = &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
259 
260 	l3mdev = inet_sk_bound_l3mdev(sk);
261 	inet_bind_bucket_for_each(tb2, &head2->chain)
262 		if (check_bind2_bucket_match_nulladdr(tb2, net, port, l3mdev, sk))
263 			break;
264 
265 	if (tb2 && check_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
266 					 reuseport_ok))
267 		return true;
268 
269 	return false;
270 }
271 
272 /*
273  * Find an open port number for the socket.  Returns with the
274  * inet_bind_hashbucket lock held.
275  */
276 static struct inet_bind_hashbucket *
277 inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret,
278 			struct inet_bind2_bucket **tb2_ret,
279 			struct inet_bind2_hashbucket **head2_ret, int *port_ret)
280 {
281 	struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
282 	struct inet_bind2_hashbucket *head2;
283 	struct inet_bind_hashbucket *head;
284 	struct net *net = sock_net(sk);
285 	int i, low, high, attempt_half;
286 	struct inet_bind2_bucket *tb2;
287 	struct inet_bind_bucket *tb;
288 	u32 remaining, offset;
289 	bool relax = false;
290 	int port = 0;
291 	int l3mdev;
292 
293 	l3mdev = inet_sk_bound_l3mdev(sk);
294 ports_exhausted:
295 	attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
296 other_half_scan:
297 	inet_get_local_port_range(net, &low, &high);
298 	high++; /* [32768, 60999] -> [32768, 61000[ */
299 	if (high - low < 4)
300 		attempt_half = 0;
301 	if (attempt_half) {
302 		int half = low + (((high - low) >> 2) << 1);
303 
304 		if (attempt_half == 1)
305 			high = half;
306 		else
307 			low = half;
308 	}
309 	remaining = high - low;
310 	if (likely(remaining > 1))
311 		remaining &= ~1U;
312 
313 	offset = prandom_u32() % remaining;
314 	/* __inet_hash_connect() favors ports having @low parity
315 	 * We do the opposite to not pollute connect() users.
316 	 */
317 	offset |= 1U;
318 
319 other_parity_scan:
320 	port = low + offset;
321 	for (i = 0; i < remaining; i += 2, port += 2) {
322 		if (unlikely(port >= high))
323 			port -= remaining;
324 		if (inet_is_local_reserved_port(net, port))
325 			continue;
326 		head = &hinfo->bhash[inet_bhashfn(net, port,
327 						  hinfo->bhash_size)];
328 		spin_lock_bh(&head->lock);
329 		tb2 = inet_bind2_bucket_find(hinfo, net, port, l3mdev, sk,
330 					     &head2);
331 		inet_bind_bucket_for_each(tb, &head->chain)
332 			if (check_bind_bucket_match(tb, net, port, l3mdev)) {
333 				if (!inet_csk_bind_conflict(sk, port, tb, tb2,
334 							    relax, false))
335 					goto success;
336 				goto next_port;
337 			}
338 		tb = NULL;
339 		goto success;
340 next_port:
341 		spin_unlock_bh(&head->lock);
342 		cond_resched();
343 	}
344 
345 	offset--;
346 	if (!(offset & 1))
347 		goto other_parity_scan;
348 
349 	if (attempt_half == 1) {
350 		/* OK we now try the upper half of the range */
351 		attempt_half = 2;
352 		goto other_half_scan;
353 	}
354 
355 	if (net->ipv4.sysctl_ip_autobind_reuse && !relax) {
356 		/* We still have a chance to connect to different destinations */
357 		relax = true;
358 		goto ports_exhausted;
359 	}
360 	return NULL;
361 success:
362 	*port_ret = port;
363 	*tb_ret = tb;
364 	*tb2_ret = tb2;
365 	*head2_ret = head2;
366 	return head;
367 }
368 
369 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
370 				     struct sock *sk)
371 {
372 	kuid_t uid = sock_i_uid(sk);
373 
374 	if (tb->fastreuseport <= 0)
375 		return 0;
376 	if (!sk->sk_reuseport)
377 		return 0;
378 	if (rcu_access_pointer(sk->sk_reuseport_cb))
379 		return 0;
380 	if (!uid_eq(tb->fastuid, uid))
381 		return 0;
382 	/* We only need to check the rcv_saddr if this tb was once marked
383 	 * without fastreuseport and then was reset, as we can only know that
384 	 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
385 	 * owners list.
386 	 */
387 	if (tb->fastreuseport == FASTREUSEPORT_ANY)
388 		return 1;
389 #if IS_ENABLED(CONFIG_IPV6)
390 	if (tb->fast_sk_family == AF_INET6)
391 		return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
392 					    inet6_rcv_saddr(sk),
393 					    tb->fast_rcv_saddr,
394 					    sk->sk_rcv_saddr,
395 					    tb->fast_ipv6_only,
396 					    ipv6_only_sock(sk), true, false);
397 #endif
398 	return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
399 				    ipv6_only_sock(sk), true, false);
400 }
401 
402 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
403 			       struct sock *sk)
404 {
405 	kuid_t uid = sock_i_uid(sk);
406 	bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
407 
408 	if (hlist_empty(&tb->owners)) {
409 		tb->fastreuse = reuse;
410 		if (sk->sk_reuseport) {
411 			tb->fastreuseport = FASTREUSEPORT_ANY;
412 			tb->fastuid = uid;
413 			tb->fast_rcv_saddr = sk->sk_rcv_saddr;
414 			tb->fast_ipv6_only = ipv6_only_sock(sk);
415 			tb->fast_sk_family = sk->sk_family;
416 #if IS_ENABLED(CONFIG_IPV6)
417 			tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
418 #endif
419 		} else {
420 			tb->fastreuseport = 0;
421 		}
422 	} else {
423 		if (!reuse)
424 			tb->fastreuse = 0;
425 		if (sk->sk_reuseport) {
426 			/* We didn't match or we don't have fastreuseport set on
427 			 * the tb, but we have sk_reuseport set on this socket
428 			 * and we know that there are no bind conflicts with
429 			 * this socket in this tb, so reset our tb's reuseport
430 			 * settings so that any subsequent sockets that match
431 			 * our current socket will be put on the fast path.
432 			 *
433 			 * If we reset we need to set FASTREUSEPORT_STRICT so we
434 			 * do extra checking for all subsequent sk_reuseport
435 			 * socks.
436 			 */
437 			if (!sk_reuseport_match(tb, sk)) {
438 				tb->fastreuseport = FASTREUSEPORT_STRICT;
439 				tb->fastuid = uid;
440 				tb->fast_rcv_saddr = sk->sk_rcv_saddr;
441 				tb->fast_ipv6_only = ipv6_only_sock(sk);
442 				tb->fast_sk_family = sk->sk_family;
443 #if IS_ENABLED(CONFIG_IPV6)
444 				tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
445 #endif
446 			}
447 		} else {
448 			tb->fastreuseport = 0;
449 		}
450 	}
451 }
452 
453 /* Obtain a reference to a local port for the given sock,
454  * if snum is zero it means select any available local port.
455  * We try to allocate an odd port (and leave even ports for connect())
456  */
457 int inet_csk_get_port(struct sock *sk, unsigned short snum)
458 {
459 	bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
460 	struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
461 	bool bhash_created = false, bhash2_created = false;
462 	struct inet_bind2_bucket *tb2 = NULL;
463 	struct inet_bind2_hashbucket *head2;
464 	struct inet_bind_bucket *tb = NULL;
465 	struct inet_bind_hashbucket *head;
466 	struct net *net = sock_net(sk);
467 	int ret = 1, port = snum;
468 	bool found_port = false;
469 	int l3mdev;
470 
471 	l3mdev = inet_sk_bound_l3mdev(sk);
472 
473 	if (!port) {
474 		head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
475 		if (!head)
476 			return ret;
477 		if (tb && tb2)
478 			goto success;
479 		found_port = true;
480 	} else {
481 		head = &hinfo->bhash[inet_bhashfn(net, port,
482 						  hinfo->bhash_size)];
483 		spin_lock_bh(&head->lock);
484 		inet_bind_bucket_for_each(tb, &head->chain)
485 			if (check_bind_bucket_match(tb, net, port, l3mdev))
486 				break;
487 
488 		tb2 = inet_bind2_bucket_find(hinfo, net, port, l3mdev, sk,
489 					     &head2);
490 	}
491 
492 	if (!tb) {
493 		tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
494 					     head, port, l3mdev);
495 		if (!tb)
496 			goto fail_unlock;
497 		bhash_created = true;
498 	}
499 
500 	if (!tb2) {
501 		tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
502 					       net, head2, port, l3mdev, sk);
503 		if (!tb2)
504 			goto fail_unlock;
505 		bhash2_created = true;
506 	}
507 
508 	/* If we had to find an open port, we already checked for conflicts */
509 	if (!found_port && !hlist_empty(&tb->owners)) {
510 		if (sk->sk_reuse == SK_FORCE_REUSE)
511 			goto success;
512 
513 		if ((tb->fastreuse > 0 && reuse) ||
514 		    sk_reuseport_match(tb, sk))
515 			goto success;
516 		if (inet_csk_bind_conflict(sk, port, tb, tb2, true, true))
517 			goto fail_unlock;
518 	}
519 success:
520 	inet_csk_update_fastreuse(tb, sk);
521 
522 	if (!inet_csk(sk)->icsk_bind_hash)
523 		inet_bind_hash(sk, tb, tb2, port);
524 	WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
525 	WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
526 	ret = 0;
527 
528 fail_unlock:
529 	if (ret) {
530 		if (bhash_created)
531 			inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
532 		if (bhash2_created)
533 			inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep,
534 						  tb2);
535 	}
536 	spin_unlock_bh(&head->lock);
537 	return ret;
538 }
539 EXPORT_SYMBOL_GPL(inet_csk_get_port);
540 
541 /*
542  * Wait for an incoming connection, avoid race conditions. This must be called
543  * with the socket locked.
544  */
545 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
546 {
547 	struct inet_connection_sock *icsk = inet_csk(sk);
548 	DEFINE_WAIT(wait);
549 	int err;
550 
551 	/*
552 	 * True wake-one mechanism for incoming connections: only
553 	 * one process gets woken up, not the 'whole herd'.
554 	 * Since we do not 'race & poll' for established sockets
555 	 * anymore, the common case will execute the loop only once.
556 	 *
557 	 * Subtle issue: "add_wait_queue_exclusive()" will be added
558 	 * after any current non-exclusive waiters, and we know that
559 	 * it will always _stay_ after any new non-exclusive waiters
560 	 * because all non-exclusive waiters are added at the
561 	 * beginning of the wait-queue. As such, it's ok to "drop"
562 	 * our exclusiveness temporarily when we get woken up without
563 	 * having to remove and re-insert us on the wait queue.
564 	 */
565 	for (;;) {
566 		prepare_to_wait_exclusive(sk_sleep(sk), &wait,
567 					  TASK_INTERRUPTIBLE);
568 		release_sock(sk);
569 		if (reqsk_queue_empty(&icsk->icsk_accept_queue))
570 			timeo = schedule_timeout(timeo);
571 		sched_annotate_sleep();
572 		lock_sock(sk);
573 		err = 0;
574 		if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
575 			break;
576 		err = -EINVAL;
577 		if (sk->sk_state != TCP_LISTEN)
578 			break;
579 		err = sock_intr_errno(timeo);
580 		if (signal_pending(current))
581 			break;
582 		err = -EAGAIN;
583 		if (!timeo)
584 			break;
585 	}
586 	finish_wait(sk_sleep(sk), &wait);
587 	return err;
588 }
589 
590 /*
591  * This will accept the next outstanding connection.
592  */
593 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
594 {
595 	struct inet_connection_sock *icsk = inet_csk(sk);
596 	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
597 	struct request_sock *req;
598 	struct sock *newsk;
599 	int error;
600 
601 	lock_sock(sk);
602 
603 	/* We need to make sure that this socket is listening,
604 	 * and that it has something pending.
605 	 */
606 	error = -EINVAL;
607 	if (sk->sk_state != TCP_LISTEN)
608 		goto out_err;
609 
610 	/* Find already established connection */
611 	if (reqsk_queue_empty(queue)) {
612 		long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
613 
614 		/* If this is a non blocking socket don't sleep */
615 		error = -EAGAIN;
616 		if (!timeo)
617 			goto out_err;
618 
619 		error = inet_csk_wait_for_connect(sk, timeo);
620 		if (error)
621 			goto out_err;
622 	}
623 	req = reqsk_queue_remove(queue, sk);
624 	newsk = req->sk;
625 
626 	if (sk->sk_protocol == IPPROTO_TCP &&
627 	    tcp_rsk(req)->tfo_listener) {
628 		spin_lock_bh(&queue->fastopenq.lock);
629 		if (tcp_rsk(req)->tfo_listener) {
630 			/* We are still waiting for the final ACK from 3WHS
631 			 * so can't free req now. Instead, we set req->sk to
632 			 * NULL to signify that the child socket is taken
633 			 * so reqsk_fastopen_remove() will free the req
634 			 * when 3WHS finishes (or is aborted).
635 			 */
636 			req->sk = NULL;
637 			req = NULL;
638 		}
639 		spin_unlock_bh(&queue->fastopenq.lock);
640 	}
641 
642 out:
643 	release_sock(sk);
644 	if (newsk && mem_cgroup_sockets_enabled) {
645 		int amt;
646 
647 		/* atomically get the memory usage, set and charge the
648 		 * newsk->sk_memcg.
649 		 */
650 		lock_sock(newsk);
651 
652 		/* The socket has not been accepted yet, no need to look at
653 		 * newsk->sk_wmem_queued.
654 		 */
655 		amt = sk_mem_pages(newsk->sk_forward_alloc +
656 				   atomic_read(&newsk->sk_rmem_alloc));
657 		mem_cgroup_sk_alloc(newsk);
658 		if (newsk->sk_memcg && amt)
659 			mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
660 						GFP_KERNEL | __GFP_NOFAIL);
661 
662 		release_sock(newsk);
663 	}
664 	if (req)
665 		reqsk_put(req);
666 	return newsk;
667 out_err:
668 	newsk = NULL;
669 	req = NULL;
670 	*err = error;
671 	goto out;
672 }
673 EXPORT_SYMBOL(inet_csk_accept);
674 
675 /*
676  * Using different timers for retransmit, delayed acks and probes
677  * We may wish use just one timer maintaining a list of expire jiffies
678  * to optimize.
679  */
680 void inet_csk_init_xmit_timers(struct sock *sk,
681 			       void (*retransmit_handler)(struct timer_list *t),
682 			       void (*delack_handler)(struct timer_list *t),
683 			       void (*keepalive_handler)(struct timer_list *t))
684 {
685 	struct inet_connection_sock *icsk = inet_csk(sk);
686 
687 	timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
688 	timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
689 	timer_setup(&sk->sk_timer, keepalive_handler, 0);
690 	icsk->icsk_pending = icsk->icsk_ack.pending = 0;
691 }
692 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
693 
694 void inet_csk_clear_xmit_timers(struct sock *sk)
695 {
696 	struct inet_connection_sock *icsk = inet_csk(sk);
697 
698 	icsk->icsk_pending = icsk->icsk_ack.pending = 0;
699 
700 	sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
701 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
702 	sk_stop_timer(sk, &sk->sk_timer);
703 }
704 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
705 
706 void inet_csk_delete_keepalive_timer(struct sock *sk)
707 {
708 	sk_stop_timer(sk, &sk->sk_timer);
709 }
710 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
711 
712 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
713 {
714 	sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
715 }
716 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
717 
718 struct dst_entry *inet_csk_route_req(const struct sock *sk,
719 				     struct flowi4 *fl4,
720 				     const struct request_sock *req)
721 {
722 	const struct inet_request_sock *ireq = inet_rsk(req);
723 	struct net *net = read_pnet(&ireq->ireq_net);
724 	struct ip_options_rcu *opt;
725 	struct rtable *rt;
726 
727 	rcu_read_lock();
728 	opt = rcu_dereference(ireq->ireq_opt);
729 
730 	flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
731 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
732 			   sk->sk_protocol, inet_sk_flowi_flags(sk),
733 			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
734 			   ireq->ir_loc_addr, ireq->ir_rmt_port,
735 			   htons(ireq->ir_num), sk->sk_uid);
736 	security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
737 	rt = ip_route_output_flow(net, fl4, sk);
738 	if (IS_ERR(rt))
739 		goto no_route;
740 	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
741 		goto route_err;
742 	rcu_read_unlock();
743 	return &rt->dst;
744 
745 route_err:
746 	ip_rt_put(rt);
747 no_route:
748 	rcu_read_unlock();
749 	__IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
750 	return NULL;
751 }
752 EXPORT_SYMBOL_GPL(inet_csk_route_req);
753 
754 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
755 					    struct sock *newsk,
756 					    const struct request_sock *req)
757 {
758 	const struct inet_request_sock *ireq = inet_rsk(req);
759 	struct net *net = read_pnet(&ireq->ireq_net);
760 	struct inet_sock *newinet = inet_sk(newsk);
761 	struct ip_options_rcu *opt;
762 	struct flowi4 *fl4;
763 	struct rtable *rt;
764 
765 	opt = rcu_dereference(ireq->ireq_opt);
766 	fl4 = &newinet->cork.fl.u.ip4;
767 
768 	flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
769 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
770 			   sk->sk_protocol, inet_sk_flowi_flags(sk),
771 			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
772 			   ireq->ir_loc_addr, ireq->ir_rmt_port,
773 			   htons(ireq->ir_num), sk->sk_uid);
774 	security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
775 	rt = ip_route_output_flow(net, fl4, sk);
776 	if (IS_ERR(rt))
777 		goto no_route;
778 	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
779 		goto route_err;
780 	return &rt->dst;
781 
782 route_err:
783 	ip_rt_put(rt);
784 no_route:
785 	__IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
786 	return NULL;
787 }
788 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
789 
790 /* Decide when to expire the request and when to resend SYN-ACK */
791 static void syn_ack_recalc(struct request_sock *req,
792 			   const int max_syn_ack_retries,
793 			   const u8 rskq_defer_accept,
794 			   int *expire, int *resend)
795 {
796 	if (!rskq_defer_accept) {
797 		*expire = req->num_timeout >= max_syn_ack_retries;
798 		*resend = 1;
799 		return;
800 	}
801 	*expire = req->num_timeout >= max_syn_ack_retries &&
802 		  (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
803 	/* Do not resend while waiting for data after ACK,
804 	 * start to resend on end of deferring period to give
805 	 * last chance for data or ACK to create established socket.
806 	 */
807 	*resend = !inet_rsk(req)->acked ||
808 		  req->num_timeout >= rskq_defer_accept - 1;
809 }
810 
811 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
812 {
813 	int err = req->rsk_ops->rtx_syn_ack(parent, req);
814 
815 	if (!err)
816 		req->num_retrans++;
817 	return err;
818 }
819 EXPORT_SYMBOL(inet_rtx_syn_ack);
820 
821 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
822 					     struct sock *sk)
823 {
824 	struct sock *req_sk, *nreq_sk;
825 	struct request_sock *nreq;
826 
827 	nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
828 	if (!nreq) {
829 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
830 
831 		/* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
832 		sock_put(sk);
833 		return NULL;
834 	}
835 
836 	req_sk = req_to_sk(req);
837 	nreq_sk = req_to_sk(nreq);
838 
839 	memcpy(nreq_sk, req_sk,
840 	       offsetof(struct sock, sk_dontcopy_begin));
841 	memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
842 	       req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
843 
844 	sk_node_init(&nreq_sk->sk_node);
845 	nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
846 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
847 	nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
848 #endif
849 	nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
850 
851 	nreq->rsk_listener = sk;
852 
853 	/* We need not acquire fastopenq->lock
854 	 * because the child socket is locked in inet_csk_listen_stop().
855 	 */
856 	if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
857 		rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
858 
859 	return nreq;
860 }
861 
862 static void reqsk_queue_migrated(struct request_sock_queue *queue,
863 				 const struct request_sock *req)
864 {
865 	if (req->num_timeout == 0)
866 		atomic_inc(&queue->young);
867 	atomic_inc(&queue->qlen);
868 }
869 
870 static void reqsk_migrate_reset(struct request_sock *req)
871 {
872 	req->saved_syn = NULL;
873 #if IS_ENABLED(CONFIG_IPV6)
874 	inet_rsk(req)->ipv6_opt = NULL;
875 	inet_rsk(req)->pktopts = NULL;
876 #else
877 	inet_rsk(req)->ireq_opt = NULL;
878 #endif
879 }
880 
881 /* return true if req was found in the ehash table */
882 static bool reqsk_queue_unlink(struct request_sock *req)
883 {
884 	struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
885 	bool found = false;
886 
887 	if (sk_hashed(req_to_sk(req))) {
888 		spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
889 
890 		spin_lock(lock);
891 		found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
892 		spin_unlock(lock);
893 	}
894 	if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
895 		reqsk_put(req);
896 	return found;
897 }
898 
899 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
900 {
901 	bool unlinked = reqsk_queue_unlink(req);
902 
903 	if (unlinked) {
904 		reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
905 		reqsk_put(req);
906 	}
907 	return unlinked;
908 }
909 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
910 
911 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
912 {
913 	inet_csk_reqsk_queue_drop(sk, req);
914 	reqsk_put(req);
915 }
916 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
917 
918 static void reqsk_timer_handler(struct timer_list *t)
919 {
920 	struct request_sock *req = from_timer(req, t, rsk_timer);
921 	struct request_sock *nreq = NULL, *oreq = req;
922 	struct sock *sk_listener = req->rsk_listener;
923 	struct inet_connection_sock *icsk;
924 	struct request_sock_queue *queue;
925 	struct net *net;
926 	int max_syn_ack_retries, qlen, expire = 0, resend = 0;
927 
928 	if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
929 		struct sock *nsk;
930 
931 		nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
932 		if (!nsk)
933 			goto drop;
934 
935 		nreq = inet_reqsk_clone(req, nsk);
936 		if (!nreq)
937 			goto drop;
938 
939 		/* The new timer for the cloned req can decrease the 2
940 		 * by calling inet_csk_reqsk_queue_drop_and_put(), so
941 		 * hold another count to prevent use-after-free and
942 		 * call reqsk_put() just before return.
943 		 */
944 		refcount_set(&nreq->rsk_refcnt, 2 + 1);
945 		timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
946 		reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
947 
948 		req = nreq;
949 		sk_listener = nsk;
950 	}
951 
952 	icsk = inet_csk(sk_listener);
953 	net = sock_net(sk_listener);
954 	max_syn_ack_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
955 	/* Normally all the openreqs are young and become mature
956 	 * (i.e. converted to established socket) for first timeout.
957 	 * If synack was not acknowledged for 1 second, it means
958 	 * one of the following things: synack was lost, ack was lost,
959 	 * rtt is high or nobody planned to ack (i.e. synflood).
960 	 * When server is a bit loaded, queue is populated with old
961 	 * open requests, reducing effective size of queue.
962 	 * When server is well loaded, queue size reduces to zero
963 	 * after several minutes of work. It is not synflood,
964 	 * it is normal operation. The solution is pruning
965 	 * too old entries overriding normal timeout, when
966 	 * situation becomes dangerous.
967 	 *
968 	 * Essentially, we reserve half of room for young
969 	 * embrions; and abort old ones without pity, if old
970 	 * ones are about to clog our table.
971 	 */
972 	queue = &icsk->icsk_accept_queue;
973 	qlen = reqsk_queue_len(queue);
974 	if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
975 		int young = reqsk_queue_len_young(queue) << 1;
976 
977 		while (max_syn_ack_retries > 2) {
978 			if (qlen < young)
979 				break;
980 			max_syn_ack_retries--;
981 			young <<= 1;
982 		}
983 	}
984 	syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
985 		       &expire, &resend);
986 	req->rsk_ops->syn_ack_timeout(req);
987 	if (!expire &&
988 	    (!resend ||
989 	     !inet_rtx_syn_ack(sk_listener, req) ||
990 	     inet_rsk(req)->acked)) {
991 		if (req->num_timeout++ == 0)
992 			atomic_dec(&queue->young);
993 		mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
994 
995 		if (!nreq)
996 			return;
997 
998 		if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
999 			/* delete timer */
1000 			inet_csk_reqsk_queue_drop(sk_listener, nreq);
1001 			goto no_ownership;
1002 		}
1003 
1004 		__NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1005 		reqsk_migrate_reset(oreq);
1006 		reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1007 		reqsk_put(oreq);
1008 
1009 		reqsk_put(nreq);
1010 		return;
1011 	}
1012 
1013 	/* Even if we can clone the req, we may need not retransmit any more
1014 	 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1015 	 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1016 	 */
1017 	if (nreq) {
1018 		__NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1019 no_ownership:
1020 		reqsk_migrate_reset(nreq);
1021 		reqsk_queue_removed(queue, nreq);
1022 		__reqsk_free(nreq);
1023 	}
1024 
1025 drop:
1026 	inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
1027 }
1028 
1029 static void reqsk_queue_hash_req(struct request_sock *req,
1030 				 unsigned long timeout)
1031 {
1032 	timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1033 	mod_timer(&req->rsk_timer, jiffies + timeout);
1034 
1035 	inet_ehash_insert(req_to_sk(req), NULL, NULL);
1036 	/* before letting lookups find us, make sure all req fields
1037 	 * are committed to memory and refcnt initialized.
1038 	 */
1039 	smp_wmb();
1040 	refcount_set(&req->rsk_refcnt, 2 + 1);
1041 }
1042 
1043 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1044 				   unsigned long timeout)
1045 {
1046 	reqsk_queue_hash_req(req, timeout);
1047 	inet_csk_reqsk_queue_added(sk);
1048 }
1049 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1050 
1051 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1052 			   const gfp_t priority)
1053 {
1054 	struct inet_connection_sock *icsk = inet_csk(newsk);
1055 
1056 	if (!icsk->icsk_ulp_ops)
1057 		return;
1058 
1059 	if (icsk->icsk_ulp_ops->clone)
1060 		icsk->icsk_ulp_ops->clone(req, newsk, priority);
1061 }
1062 
1063 /**
1064  *	inet_csk_clone_lock - clone an inet socket, and lock its clone
1065  *	@sk: the socket to clone
1066  *	@req: request_sock
1067  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1068  *
1069  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1070  */
1071 struct sock *inet_csk_clone_lock(const struct sock *sk,
1072 				 const struct request_sock *req,
1073 				 const gfp_t priority)
1074 {
1075 	struct sock *newsk = sk_clone_lock(sk, priority);
1076 
1077 	if (newsk) {
1078 		struct inet_connection_sock *newicsk = inet_csk(newsk);
1079 
1080 		inet_sk_set_state(newsk, TCP_SYN_RECV);
1081 		newicsk->icsk_bind_hash = NULL;
1082 		newicsk->icsk_bind2_hash = NULL;
1083 
1084 		inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
1085 		inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
1086 		inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
1087 
1088 		/* listeners have SOCK_RCU_FREE, not the children */
1089 		sock_reset_flag(newsk, SOCK_RCU_FREE);
1090 
1091 		inet_sk(newsk)->mc_list = NULL;
1092 
1093 		newsk->sk_mark = inet_rsk(req)->ir_mark;
1094 		atomic64_set(&newsk->sk_cookie,
1095 			     atomic64_read(&inet_rsk(req)->ir_cookie));
1096 
1097 		newicsk->icsk_retransmits = 0;
1098 		newicsk->icsk_backoff	  = 0;
1099 		newicsk->icsk_probes_out  = 0;
1100 		newicsk->icsk_probes_tstamp = 0;
1101 
1102 		/* Deinitialize accept_queue to trap illegal accesses. */
1103 		memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
1104 
1105 		inet_clone_ulp(req, newsk, priority);
1106 
1107 		security_inet_csk_clone(newsk, req);
1108 	}
1109 	return newsk;
1110 }
1111 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
1112 
1113 /*
1114  * At this point, there should be no process reference to this
1115  * socket, and thus no user references at all.  Therefore we
1116  * can assume the socket waitqueue is inactive and nobody will
1117  * try to jump onto it.
1118  */
1119 void inet_csk_destroy_sock(struct sock *sk)
1120 {
1121 	WARN_ON(sk->sk_state != TCP_CLOSE);
1122 	WARN_ON(!sock_flag(sk, SOCK_DEAD));
1123 
1124 	/* It cannot be in hash table! */
1125 	WARN_ON(!sk_unhashed(sk));
1126 
1127 	/* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1128 	WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1129 
1130 	sk->sk_prot->destroy(sk);
1131 
1132 	sk_stream_kill_queues(sk);
1133 
1134 	xfrm_sk_free_policy(sk);
1135 
1136 	sk_refcnt_debug_release(sk);
1137 
1138 	this_cpu_dec(*sk->sk_prot->orphan_count);
1139 
1140 	sock_put(sk);
1141 }
1142 EXPORT_SYMBOL(inet_csk_destroy_sock);
1143 
1144 /* This function allows to force a closure of a socket after the call to
1145  * tcp/dccp_create_openreq_child().
1146  */
1147 void inet_csk_prepare_forced_close(struct sock *sk)
1148 	__releases(&sk->sk_lock.slock)
1149 {
1150 	/* sk_clone_lock locked the socket and set refcnt to 2 */
1151 	bh_unlock_sock(sk);
1152 	sock_put(sk);
1153 	inet_csk_prepare_for_destroy_sock(sk);
1154 	inet_sk(sk)->inet_num = 0;
1155 }
1156 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1157 
1158 int inet_csk_listen_start(struct sock *sk)
1159 {
1160 	struct inet_connection_sock *icsk = inet_csk(sk);
1161 	struct inet_sock *inet = inet_sk(sk);
1162 	int err = -EADDRINUSE;
1163 
1164 	reqsk_queue_alloc(&icsk->icsk_accept_queue);
1165 
1166 	sk->sk_ack_backlog = 0;
1167 	inet_csk_delack_init(sk);
1168 
1169 	if (sk->sk_txrehash == SOCK_TXREHASH_DEFAULT)
1170 		sk->sk_txrehash = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1171 
1172 	/* There is race window here: we announce ourselves listening,
1173 	 * but this transition is still not validated by get_port().
1174 	 * It is OK, because this socket enters to hash table only
1175 	 * after validation is complete.
1176 	 */
1177 	inet_sk_state_store(sk, TCP_LISTEN);
1178 	if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
1179 		inet->inet_sport = htons(inet->inet_num);
1180 
1181 		sk_dst_reset(sk);
1182 		err = sk->sk_prot->hash(sk);
1183 
1184 		if (likely(!err))
1185 			return 0;
1186 	}
1187 
1188 	inet_sk_set_state(sk, TCP_CLOSE);
1189 	return err;
1190 }
1191 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1192 
1193 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1194 			      struct sock *child)
1195 {
1196 	sk->sk_prot->disconnect(child, O_NONBLOCK);
1197 
1198 	sock_orphan(child);
1199 
1200 	this_cpu_inc(*sk->sk_prot->orphan_count);
1201 
1202 	if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1203 		BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1204 		BUG_ON(sk != req->rsk_listener);
1205 
1206 		/* Paranoid, to prevent race condition if
1207 		 * an inbound pkt destined for child is
1208 		 * blocked by sock lock in tcp_v4_rcv().
1209 		 * Also to satisfy an assertion in
1210 		 * tcp_v4_destroy_sock().
1211 		 */
1212 		RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1213 	}
1214 	inet_csk_destroy_sock(child);
1215 }
1216 
1217 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1218 				      struct request_sock *req,
1219 				      struct sock *child)
1220 {
1221 	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1222 
1223 	spin_lock(&queue->rskq_lock);
1224 	if (unlikely(sk->sk_state != TCP_LISTEN)) {
1225 		inet_child_forget(sk, req, child);
1226 		child = NULL;
1227 	} else {
1228 		req->sk = child;
1229 		req->dl_next = NULL;
1230 		if (queue->rskq_accept_head == NULL)
1231 			WRITE_ONCE(queue->rskq_accept_head, req);
1232 		else
1233 			queue->rskq_accept_tail->dl_next = req;
1234 		queue->rskq_accept_tail = req;
1235 		sk_acceptq_added(sk);
1236 	}
1237 	spin_unlock(&queue->rskq_lock);
1238 	return child;
1239 }
1240 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1241 
1242 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1243 					 struct request_sock *req, bool own_req)
1244 {
1245 	if (own_req) {
1246 		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1247 		reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1248 
1249 		if (sk != req->rsk_listener) {
1250 			/* another listening sk has been selected,
1251 			 * migrate the req to it.
1252 			 */
1253 			struct request_sock *nreq;
1254 
1255 			/* hold a refcnt for the nreq->rsk_listener
1256 			 * which is assigned in inet_reqsk_clone()
1257 			 */
1258 			sock_hold(sk);
1259 			nreq = inet_reqsk_clone(req, sk);
1260 			if (!nreq) {
1261 				inet_child_forget(sk, req, child);
1262 				goto child_put;
1263 			}
1264 
1265 			refcount_set(&nreq->rsk_refcnt, 1);
1266 			if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1267 				__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1268 				reqsk_migrate_reset(req);
1269 				reqsk_put(req);
1270 				return child;
1271 			}
1272 
1273 			__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1274 			reqsk_migrate_reset(nreq);
1275 			__reqsk_free(nreq);
1276 		} else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1277 			return child;
1278 		}
1279 	}
1280 	/* Too bad, another child took ownership of the request, undo. */
1281 child_put:
1282 	bh_unlock_sock(child);
1283 	sock_put(child);
1284 	return NULL;
1285 }
1286 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1287 
1288 /*
1289  *	This routine closes sockets which have been at least partially
1290  *	opened, but not yet accepted.
1291  */
1292 void inet_csk_listen_stop(struct sock *sk)
1293 {
1294 	struct inet_connection_sock *icsk = inet_csk(sk);
1295 	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1296 	struct request_sock *next, *req;
1297 
1298 	/* Following specs, it would be better either to send FIN
1299 	 * (and enter FIN-WAIT-1, it is normal close)
1300 	 * or to send active reset (abort).
1301 	 * Certainly, it is pretty dangerous while synflood, but it is
1302 	 * bad justification for our negligence 8)
1303 	 * To be honest, we are not able to make either
1304 	 * of the variants now.			--ANK
1305 	 */
1306 	while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1307 		struct sock *child = req->sk, *nsk;
1308 		struct request_sock *nreq;
1309 
1310 		local_bh_disable();
1311 		bh_lock_sock(child);
1312 		WARN_ON(sock_owned_by_user(child));
1313 		sock_hold(child);
1314 
1315 		nsk = reuseport_migrate_sock(sk, child, NULL);
1316 		if (nsk) {
1317 			nreq = inet_reqsk_clone(req, nsk);
1318 			if (nreq) {
1319 				refcount_set(&nreq->rsk_refcnt, 1);
1320 
1321 				if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1322 					__NET_INC_STATS(sock_net(nsk),
1323 							LINUX_MIB_TCPMIGRATEREQSUCCESS);
1324 					reqsk_migrate_reset(req);
1325 				} else {
1326 					__NET_INC_STATS(sock_net(nsk),
1327 							LINUX_MIB_TCPMIGRATEREQFAILURE);
1328 					reqsk_migrate_reset(nreq);
1329 					__reqsk_free(nreq);
1330 				}
1331 
1332 				/* inet_csk_reqsk_queue_add() has already
1333 				 * called inet_child_forget() on failure case.
1334 				 */
1335 				goto skip_child_forget;
1336 			}
1337 		}
1338 
1339 		inet_child_forget(sk, req, child);
1340 skip_child_forget:
1341 		reqsk_put(req);
1342 		bh_unlock_sock(child);
1343 		local_bh_enable();
1344 		sock_put(child);
1345 
1346 		cond_resched();
1347 	}
1348 	if (queue->fastopenq.rskq_rst_head) {
1349 		/* Free all the reqs queued in rskq_rst_head. */
1350 		spin_lock_bh(&queue->fastopenq.lock);
1351 		req = queue->fastopenq.rskq_rst_head;
1352 		queue->fastopenq.rskq_rst_head = NULL;
1353 		spin_unlock_bh(&queue->fastopenq.lock);
1354 		while (req != NULL) {
1355 			next = req->dl_next;
1356 			reqsk_put(req);
1357 			req = next;
1358 		}
1359 	}
1360 	WARN_ON_ONCE(sk->sk_ack_backlog);
1361 }
1362 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1363 
1364 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1365 {
1366 	struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1367 	const struct inet_sock *inet = inet_sk(sk);
1368 
1369 	sin->sin_family		= AF_INET;
1370 	sin->sin_addr.s_addr	= inet->inet_daddr;
1371 	sin->sin_port		= inet->inet_dport;
1372 }
1373 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1374 
1375 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1376 {
1377 	const struct inet_sock *inet = inet_sk(sk);
1378 	const struct ip_options_rcu *inet_opt;
1379 	__be32 daddr = inet->inet_daddr;
1380 	struct flowi4 *fl4;
1381 	struct rtable *rt;
1382 
1383 	rcu_read_lock();
1384 	inet_opt = rcu_dereference(inet->inet_opt);
1385 	if (inet_opt && inet_opt->opt.srr)
1386 		daddr = inet_opt->opt.faddr;
1387 	fl4 = &fl->u.ip4;
1388 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1389 				   inet->inet_saddr, inet->inet_dport,
1390 				   inet->inet_sport, sk->sk_protocol,
1391 				   RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1392 	if (IS_ERR(rt))
1393 		rt = NULL;
1394 	if (rt)
1395 		sk_setup_caps(sk, &rt->dst);
1396 	rcu_read_unlock();
1397 
1398 	return &rt->dst;
1399 }
1400 
1401 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1402 {
1403 	struct dst_entry *dst = __sk_dst_check(sk, 0);
1404 	struct inet_sock *inet = inet_sk(sk);
1405 
1406 	if (!dst) {
1407 		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1408 		if (!dst)
1409 			goto out;
1410 	}
1411 	dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1412 
1413 	dst = __sk_dst_check(sk, 0);
1414 	if (!dst)
1415 		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1416 out:
1417 	return dst;
1418 }
1419 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
1420