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 void inet_get_local_port_range(struct net *net, int *low, int *high)
121 {
122 	unsigned int seq;
123 
124 	do {
125 		seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
126 
127 		*low = net->ipv4.ip_local_ports.range[0];
128 		*high = net->ipv4.ip_local_ports.range[1];
129 	} while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
130 }
131 EXPORT_SYMBOL(inet_get_local_port_range);
132 
133 static bool inet_use_bhash2_on_bind(const struct sock *sk)
134 {
135 #if IS_ENABLED(CONFIG_IPV6)
136 	if (sk->sk_family == AF_INET6) {
137 		int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
138 
139 		return addr_type != IPV6_ADDR_ANY &&
140 			addr_type != IPV6_ADDR_MAPPED;
141 	}
142 #endif
143 	return sk->sk_rcv_saddr != htonl(INADDR_ANY);
144 }
145 
146 static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2,
147 			       kuid_t sk_uid, bool relax,
148 			       bool reuseport_cb_ok, bool reuseport_ok)
149 {
150 	int bound_dev_if2;
151 
152 	if (sk == sk2)
153 		return false;
154 
155 	bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
156 
157 	if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
158 	    sk->sk_bound_dev_if == bound_dev_if2) {
159 		if (sk->sk_reuse && sk2->sk_reuse &&
160 		    sk2->sk_state != TCP_LISTEN) {
161 			if (!relax || (!reuseport_ok && sk->sk_reuseport &&
162 				       sk2->sk_reuseport && reuseport_cb_ok &&
163 				       (sk2->sk_state == TCP_TIME_WAIT ||
164 					uid_eq(sk_uid, sock_i_uid(sk2)))))
165 				return true;
166 		} else if (!reuseport_ok || !sk->sk_reuseport ||
167 			   !sk2->sk_reuseport || !reuseport_cb_ok ||
168 			   (sk2->sk_state != TCP_TIME_WAIT &&
169 			    !uid_eq(sk_uid, sock_i_uid(sk2)))) {
170 			return true;
171 		}
172 	}
173 	return false;
174 }
175 
176 static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2,
177 				   kuid_t sk_uid, bool relax,
178 				   bool reuseport_cb_ok, bool reuseport_ok)
179 {
180 	if (sk->sk_family == AF_INET && ipv6_only_sock(sk2))
181 		return false;
182 
183 	return inet_bind_conflict(sk, sk2, sk_uid, relax,
184 				  reuseport_cb_ok, reuseport_ok);
185 }
186 
187 static bool inet_bhash2_conflict(const struct sock *sk,
188 				 const struct inet_bind2_bucket *tb2,
189 				 kuid_t sk_uid,
190 				 bool relax, bool reuseport_cb_ok,
191 				 bool reuseport_ok)
192 {
193 	struct inet_timewait_sock *tw2;
194 	struct sock *sk2;
195 
196 	sk_for_each_bound_bhash2(sk2, &tb2->owners) {
197 		if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
198 					   reuseport_cb_ok, reuseport_ok))
199 			return true;
200 	}
201 
202 	twsk_for_each_bound_bhash2(tw2, &tb2->deathrow) {
203 		sk2 = (struct sock *)tw2;
204 
205 		if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
206 					   reuseport_cb_ok, reuseport_ok))
207 			return true;
208 	}
209 
210 	return false;
211 }
212 
213 /* This should be called only when the tb and tb2 hashbuckets' locks are held */
214 static int inet_csk_bind_conflict(const struct sock *sk,
215 				  const struct inet_bind_bucket *tb,
216 				  const struct inet_bind2_bucket *tb2, /* may be null */
217 				  bool relax, bool reuseport_ok)
218 {
219 	bool reuseport_cb_ok;
220 	struct sock_reuseport *reuseport_cb;
221 	kuid_t uid = sock_i_uid((struct sock *)sk);
222 
223 	rcu_read_lock();
224 	reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
225 	/* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
226 	reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
227 	rcu_read_unlock();
228 
229 	/*
230 	 * Unlike other sk lookup places we do not check
231 	 * for sk_net here, since _all_ the socks listed
232 	 * in tb->owners and tb2->owners list belong
233 	 * to the same net - the one this bucket belongs to.
234 	 */
235 
236 	if (!inet_use_bhash2_on_bind(sk)) {
237 		struct sock *sk2;
238 
239 		sk_for_each_bound(sk2, &tb->owners)
240 			if (inet_bind_conflict(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 	/* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if
249 	 * ipv4) should have been checked already. We need to do these two
250 	 * checks separately because their spinlocks have to be acquired/released
251 	 * independently of each other, to prevent possible deadlocks
252 	 */
253 	return tb2 && inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
254 					   reuseport_ok);
255 }
256 
257 /* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or
258  * INADDR_ANY (if ipv4) socket.
259  *
260  * Caller must hold bhash hashbucket lock with local bh disabled, to protect
261  * against concurrent binds on the port for addr any
262  */
263 static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev,
264 					  bool relax, bool reuseport_ok)
265 {
266 	kuid_t uid = sock_i_uid((struct sock *)sk);
267 	const struct net *net = sock_net(sk);
268 	struct sock_reuseport *reuseport_cb;
269 	struct inet_bind_hashbucket *head2;
270 	struct inet_bind2_bucket *tb2;
271 	bool reuseport_cb_ok;
272 
273 	rcu_read_lock();
274 	reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
275 	/* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
276 	reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
277 	rcu_read_unlock();
278 
279 	head2 = inet_bhash2_addr_any_hashbucket(sk, net, port);
280 
281 	spin_lock(&head2->lock);
282 
283 	inet_bind_bucket_for_each(tb2, &head2->chain)
284 		if (inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
285 			break;
286 
287 	if (tb2 && inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
288 					reuseport_ok)) {
289 		spin_unlock(&head2->lock);
290 		return true;
291 	}
292 
293 	spin_unlock(&head2->lock);
294 	return false;
295 }
296 
297 /*
298  * Find an open port number for the socket.  Returns with the
299  * inet_bind_hashbucket locks held if successful.
300  */
301 static struct inet_bind_hashbucket *
302 inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret,
303 			struct inet_bind2_bucket **tb2_ret,
304 			struct inet_bind_hashbucket **head2_ret, int *port_ret)
305 {
306 	struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
307 	int i, low, high, attempt_half, port, l3mdev;
308 	struct inet_bind_hashbucket *head, *head2;
309 	struct net *net = sock_net(sk);
310 	struct inet_bind2_bucket *tb2;
311 	struct inet_bind_bucket *tb;
312 	u32 remaining, offset;
313 	bool relax = false;
314 
315 	l3mdev = inet_sk_bound_l3mdev(sk);
316 ports_exhausted:
317 	attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
318 other_half_scan:
319 	inet_get_local_port_range(net, &low, &high);
320 	high++; /* [32768, 60999] -> [32768, 61000[ */
321 	if (high - low < 4)
322 		attempt_half = 0;
323 	if (attempt_half) {
324 		int half = low + (((high - low) >> 2) << 1);
325 
326 		if (attempt_half == 1)
327 			high = half;
328 		else
329 			low = half;
330 	}
331 	remaining = high - low;
332 	if (likely(remaining > 1))
333 		remaining &= ~1U;
334 
335 	offset = get_random_u32_below(remaining);
336 	/* __inet_hash_connect() favors ports having @low parity
337 	 * We do the opposite to not pollute connect() users.
338 	 */
339 	offset |= 1U;
340 
341 other_parity_scan:
342 	port = low + offset;
343 	for (i = 0; i < remaining; i += 2, port += 2) {
344 		if (unlikely(port >= high))
345 			port -= remaining;
346 		if (inet_is_local_reserved_port(net, port))
347 			continue;
348 		head = &hinfo->bhash[inet_bhashfn(net, port,
349 						  hinfo->bhash_size)];
350 		spin_lock_bh(&head->lock);
351 		if (inet_use_bhash2_on_bind(sk)) {
352 			if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, false))
353 				goto next_port;
354 		}
355 
356 		head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
357 		spin_lock(&head2->lock);
358 		tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
359 		inet_bind_bucket_for_each(tb, &head->chain)
360 			if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
361 				if (!inet_csk_bind_conflict(sk, tb, tb2,
362 							    relax, false))
363 					goto success;
364 				spin_unlock(&head2->lock);
365 				goto next_port;
366 			}
367 		tb = NULL;
368 		goto success;
369 next_port:
370 		spin_unlock_bh(&head->lock);
371 		cond_resched();
372 	}
373 
374 	offset--;
375 	if (!(offset & 1))
376 		goto other_parity_scan;
377 
378 	if (attempt_half == 1) {
379 		/* OK we now try the upper half of the range */
380 		attempt_half = 2;
381 		goto other_half_scan;
382 	}
383 
384 	if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
385 		/* We still have a chance to connect to different destinations */
386 		relax = true;
387 		goto ports_exhausted;
388 	}
389 	return NULL;
390 success:
391 	*port_ret = port;
392 	*tb_ret = tb;
393 	*tb2_ret = tb2;
394 	*head2_ret = head2;
395 	return head;
396 }
397 
398 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
399 				     struct sock *sk)
400 {
401 	kuid_t uid = sock_i_uid(sk);
402 
403 	if (tb->fastreuseport <= 0)
404 		return 0;
405 	if (!sk->sk_reuseport)
406 		return 0;
407 	if (rcu_access_pointer(sk->sk_reuseport_cb))
408 		return 0;
409 	if (!uid_eq(tb->fastuid, uid))
410 		return 0;
411 	/* We only need to check the rcv_saddr if this tb was once marked
412 	 * without fastreuseport and then was reset, as we can only know that
413 	 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
414 	 * owners list.
415 	 */
416 	if (tb->fastreuseport == FASTREUSEPORT_ANY)
417 		return 1;
418 #if IS_ENABLED(CONFIG_IPV6)
419 	if (tb->fast_sk_family == AF_INET6)
420 		return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
421 					    inet6_rcv_saddr(sk),
422 					    tb->fast_rcv_saddr,
423 					    sk->sk_rcv_saddr,
424 					    tb->fast_ipv6_only,
425 					    ipv6_only_sock(sk), true, false);
426 #endif
427 	return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
428 				    ipv6_only_sock(sk), true, false);
429 }
430 
431 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
432 			       struct sock *sk)
433 {
434 	kuid_t uid = sock_i_uid(sk);
435 	bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
436 
437 	if (hlist_empty(&tb->owners)) {
438 		tb->fastreuse = reuse;
439 		if (sk->sk_reuseport) {
440 			tb->fastreuseport = FASTREUSEPORT_ANY;
441 			tb->fastuid = uid;
442 			tb->fast_rcv_saddr = sk->sk_rcv_saddr;
443 			tb->fast_ipv6_only = ipv6_only_sock(sk);
444 			tb->fast_sk_family = sk->sk_family;
445 #if IS_ENABLED(CONFIG_IPV6)
446 			tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
447 #endif
448 		} else {
449 			tb->fastreuseport = 0;
450 		}
451 	} else {
452 		if (!reuse)
453 			tb->fastreuse = 0;
454 		if (sk->sk_reuseport) {
455 			/* We didn't match or we don't have fastreuseport set on
456 			 * the tb, but we have sk_reuseport set on this socket
457 			 * and we know that there are no bind conflicts with
458 			 * this socket in this tb, so reset our tb's reuseport
459 			 * settings so that any subsequent sockets that match
460 			 * our current socket will be put on the fast path.
461 			 *
462 			 * If we reset we need to set FASTREUSEPORT_STRICT so we
463 			 * do extra checking for all subsequent sk_reuseport
464 			 * socks.
465 			 */
466 			if (!sk_reuseport_match(tb, sk)) {
467 				tb->fastreuseport = FASTREUSEPORT_STRICT;
468 				tb->fastuid = uid;
469 				tb->fast_rcv_saddr = sk->sk_rcv_saddr;
470 				tb->fast_ipv6_only = ipv6_only_sock(sk);
471 				tb->fast_sk_family = sk->sk_family;
472 #if IS_ENABLED(CONFIG_IPV6)
473 				tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
474 #endif
475 			}
476 		} else {
477 			tb->fastreuseport = 0;
478 		}
479 	}
480 }
481 
482 /* Obtain a reference to a local port for the given sock,
483  * if snum is zero it means select any available local port.
484  * We try to allocate an odd port (and leave even ports for connect())
485  */
486 int inet_csk_get_port(struct sock *sk, unsigned short snum)
487 {
488 	struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
489 	bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
490 	bool found_port = false, check_bind_conflict = true;
491 	bool bhash_created = false, bhash2_created = false;
492 	int ret = -EADDRINUSE, port = snum, l3mdev;
493 	struct inet_bind_hashbucket *head, *head2;
494 	struct inet_bind2_bucket *tb2 = NULL;
495 	struct inet_bind_bucket *tb = NULL;
496 	bool head2_lock_acquired = false;
497 	struct net *net = sock_net(sk);
498 
499 	l3mdev = inet_sk_bound_l3mdev(sk);
500 
501 	if (!port) {
502 		head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
503 		if (!head)
504 			return ret;
505 
506 		head2_lock_acquired = true;
507 
508 		if (tb && tb2)
509 			goto success;
510 		found_port = true;
511 	} else {
512 		head = &hinfo->bhash[inet_bhashfn(net, port,
513 						  hinfo->bhash_size)];
514 		spin_lock_bh(&head->lock);
515 		inet_bind_bucket_for_each(tb, &head->chain)
516 			if (inet_bind_bucket_match(tb, net, port, l3mdev))
517 				break;
518 	}
519 
520 	if (!tb) {
521 		tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
522 					     head, port, l3mdev);
523 		if (!tb)
524 			goto fail_unlock;
525 		bhash_created = true;
526 	}
527 
528 	if (!found_port) {
529 		if (!hlist_empty(&tb->owners)) {
530 			if (sk->sk_reuse == SK_FORCE_REUSE ||
531 			    (tb->fastreuse > 0 && reuse) ||
532 			    sk_reuseport_match(tb, sk))
533 				check_bind_conflict = false;
534 		}
535 
536 		if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) {
537 			if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, true, true))
538 				goto fail_unlock;
539 		}
540 
541 		head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
542 		spin_lock(&head2->lock);
543 		head2_lock_acquired = true;
544 		tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
545 	}
546 
547 	if (!tb2) {
548 		tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
549 					       net, head2, port, l3mdev, sk);
550 		if (!tb2)
551 			goto fail_unlock;
552 		bhash2_created = true;
553 	}
554 
555 	if (!found_port && check_bind_conflict) {
556 		if (inet_csk_bind_conflict(sk, tb, tb2, true, true))
557 			goto fail_unlock;
558 	}
559 
560 success:
561 	inet_csk_update_fastreuse(tb, sk);
562 
563 	if (!inet_csk(sk)->icsk_bind_hash)
564 		inet_bind_hash(sk, tb, tb2, port);
565 	WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
566 	WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
567 	ret = 0;
568 
569 fail_unlock:
570 	if (ret) {
571 		if (bhash_created)
572 			inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
573 		if (bhash2_created)
574 			inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep,
575 						  tb2);
576 	}
577 	if (head2_lock_acquired)
578 		spin_unlock(&head2->lock);
579 	spin_unlock_bh(&head->lock);
580 	return ret;
581 }
582 EXPORT_SYMBOL_GPL(inet_csk_get_port);
583 
584 /*
585  * Wait for an incoming connection, avoid race conditions. This must be called
586  * with the socket locked.
587  */
588 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
589 {
590 	struct inet_connection_sock *icsk = inet_csk(sk);
591 	DEFINE_WAIT(wait);
592 	int err;
593 
594 	/*
595 	 * True wake-one mechanism for incoming connections: only
596 	 * one process gets woken up, not the 'whole herd'.
597 	 * Since we do not 'race & poll' for established sockets
598 	 * anymore, the common case will execute the loop only once.
599 	 *
600 	 * Subtle issue: "add_wait_queue_exclusive()" will be added
601 	 * after any current non-exclusive waiters, and we know that
602 	 * it will always _stay_ after any new non-exclusive waiters
603 	 * because all non-exclusive waiters are added at the
604 	 * beginning of the wait-queue. As such, it's ok to "drop"
605 	 * our exclusiveness temporarily when we get woken up without
606 	 * having to remove and re-insert us on the wait queue.
607 	 */
608 	for (;;) {
609 		prepare_to_wait_exclusive(sk_sleep(sk), &wait,
610 					  TASK_INTERRUPTIBLE);
611 		release_sock(sk);
612 		if (reqsk_queue_empty(&icsk->icsk_accept_queue))
613 			timeo = schedule_timeout(timeo);
614 		sched_annotate_sleep();
615 		lock_sock(sk);
616 		err = 0;
617 		if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
618 			break;
619 		err = -EINVAL;
620 		if (sk->sk_state != TCP_LISTEN)
621 			break;
622 		err = sock_intr_errno(timeo);
623 		if (signal_pending(current))
624 			break;
625 		err = -EAGAIN;
626 		if (!timeo)
627 			break;
628 	}
629 	finish_wait(sk_sleep(sk), &wait);
630 	return err;
631 }
632 
633 /*
634  * This will accept the next outstanding connection.
635  */
636 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
637 {
638 	struct inet_connection_sock *icsk = inet_csk(sk);
639 	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
640 	struct request_sock *req;
641 	struct sock *newsk;
642 	int error;
643 
644 	lock_sock(sk);
645 
646 	/* We need to make sure that this socket is listening,
647 	 * and that it has something pending.
648 	 */
649 	error = -EINVAL;
650 	if (sk->sk_state != TCP_LISTEN)
651 		goto out_err;
652 
653 	/* Find already established connection */
654 	if (reqsk_queue_empty(queue)) {
655 		long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
656 
657 		/* If this is a non blocking socket don't sleep */
658 		error = -EAGAIN;
659 		if (!timeo)
660 			goto out_err;
661 
662 		error = inet_csk_wait_for_connect(sk, timeo);
663 		if (error)
664 			goto out_err;
665 	}
666 	req = reqsk_queue_remove(queue, sk);
667 	newsk = req->sk;
668 
669 	if (sk->sk_protocol == IPPROTO_TCP &&
670 	    tcp_rsk(req)->tfo_listener) {
671 		spin_lock_bh(&queue->fastopenq.lock);
672 		if (tcp_rsk(req)->tfo_listener) {
673 			/* We are still waiting for the final ACK from 3WHS
674 			 * so can't free req now. Instead, we set req->sk to
675 			 * NULL to signify that the child socket is taken
676 			 * so reqsk_fastopen_remove() will free the req
677 			 * when 3WHS finishes (or is aborted).
678 			 */
679 			req->sk = NULL;
680 			req = NULL;
681 		}
682 		spin_unlock_bh(&queue->fastopenq.lock);
683 	}
684 
685 out:
686 	release_sock(sk);
687 	if (newsk && mem_cgroup_sockets_enabled) {
688 		int amt;
689 
690 		/* atomically get the memory usage, set and charge the
691 		 * newsk->sk_memcg.
692 		 */
693 		lock_sock(newsk);
694 
695 		/* The socket has not been accepted yet, no need to look at
696 		 * newsk->sk_wmem_queued.
697 		 */
698 		amt = sk_mem_pages(newsk->sk_forward_alloc +
699 				   atomic_read(&newsk->sk_rmem_alloc));
700 		mem_cgroup_sk_alloc(newsk);
701 		if (newsk->sk_memcg && amt)
702 			mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
703 						GFP_KERNEL | __GFP_NOFAIL);
704 
705 		release_sock(newsk);
706 	}
707 	if (req)
708 		reqsk_put(req);
709 	return newsk;
710 out_err:
711 	newsk = NULL;
712 	req = NULL;
713 	*err = error;
714 	goto out;
715 }
716 EXPORT_SYMBOL(inet_csk_accept);
717 
718 /*
719  * Using different timers for retransmit, delayed acks and probes
720  * We may wish use just one timer maintaining a list of expire jiffies
721  * to optimize.
722  */
723 void inet_csk_init_xmit_timers(struct sock *sk,
724 			       void (*retransmit_handler)(struct timer_list *t),
725 			       void (*delack_handler)(struct timer_list *t),
726 			       void (*keepalive_handler)(struct timer_list *t))
727 {
728 	struct inet_connection_sock *icsk = inet_csk(sk);
729 
730 	timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
731 	timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
732 	timer_setup(&sk->sk_timer, keepalive_handler, 0);
733 	icsk->icsk_pending = icsk->icsk_ack.pending = 0;
734 }
735 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
736 
737 void inet_csk_clear_xmit_timers(struct sock *sk)
738 {
739 	struct inet_connection_sock *icsk = inet_csk(sk);
740 
741 	icsk->icsk_pending = icsk->icsk_ack.pending = 0;
742 
743 	sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
744 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
745 	sk_stop_timer(sk, &sk->sk_timer);
746 }
747 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
748 
749 void inet_csk_delete_keepalive_timer(struct sock *sk)
750 {
751 	sk_stop_timer(sk, &sk->sk_timer);
752 }
753 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
754 
755 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
756 {
757 	sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
758 }
759 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
760 
761 struct dst_entry *inet_csk_route_req(const struct sock *sk,
762 				     struct flowi4 *fl4,
763 				     const struct request_sock *req)
764 {
765 	const struct inet_request_sock *ireq = inet_rsk(req);
766 	struct net *net = read_pnet(&ireq->ireq_net);
767 	struct ip_options_rcu *opt;
768 	struct rtable *rt;
769 
770 	rcu_read_lock();
771 	opt = rcu_dereference(ireq->ireq_opt);
772 
773 	flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
774 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
775 			   sk->sk_protocol, inet_sk_flowi_flags(sk),
776 			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
777 			   ireq->ir_loc_addr, ireq->ir_rmt_port,
778 			   htons(ireq->ir_num), sk->sk_uid);
779 	security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
780 	rt = ip_route_output_flow(net, fl4, sk);
781 	if (IS_ERR(rt))
782 		goto no_route;
783 	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
784 		goto route_err;
785 	rcu_read_unlock();
786 	return &rt->dst;
787 
788 route_err:
789 	ip_rt_put(rt);
790 no_route:
791 	rcu_read_unlock();
792 	__IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
793 	return NULL;
794 }
795 EXPORT_SYMBOL_GPL(inet_csk_route_req);
796 
797 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
798 					    struct sock *newsk,
799 					    const struct request_sock *req)
800 {
801 	const struct inet_request_sock *ireq = inet_rsk(req);
802 	struct net *net = read_pnet(&ireq->ireq_net);
803 	struct inet_sock *newinet = inet_sk(newsk);
804 	struct ip_options_rcu *opt;
805 	struct flowi4 *fl4;
806 	struct rtable *rt;
807 
808 	opt = rcu_dereference(ireq->ireq_opt);
809 	fl4 = &newinet->cork.fl.u.ip4;
810 
811 	flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
812 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
813 			   sk->sk_protocol, inet_sk_flowi_flags(sk),
814 			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
815 			   ireq->ir_loc_addr, ireq->ir_rmt_port,
816 			   htons(ireq->ir_num), sk->sk_uid);
817 	security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
818 	rt = ip_route_output_flow(net, fl4, sk);
819 	if (IS_ERR(rt))
820 		goto no_route;
821 	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
822 		goto route_err;
823 	return &rt->dst;
824 
825 route_err:
826 	ip_rt_put(rt);
827 no_route:
828 	__IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
829 	return NULL;
830 }
831 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
832 
833 /* Decide when to expire the request and when to resend SYN-ACK */
834 static void syn_ack_recalc(struct request_sock *req,
835 			   const int max_syn_ack_retries,
836 			   const u8 rskq_defer_accept,
837 			   int *expire, int *resend)
838 {
839 	if (!rskq_defer_accept) {
840 		*expire = req->num_timeout >= max_syn_ack_retries;
841 		*resend = 1;
842 		return;
843 	}
844 	*expire = req->num_timeout >= max_syn_ack_retries &&
845 		  (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
846 	/* Do not resend while waiting for data after ACK,
847 	 * start to resend on end of deferring period to give
848 	 * last chance for data or ACK to create established socket.
849 	 */
850 	*resend = !inet_rsk(req)->acked ||
851 		  req->num_timeout >= rskq_defer_accept - 1;
852 }
853 
854 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
855 {
856 	int err = req->rsk_ops->rtx_syn_ack(parent, req);
857 
858 	if (!err)
859 		req->num_retrans++;
860 	return err;
861 }
862 EXPORT_SYMBOL(inet_rtx_syn_ack);
863 
864 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
865 					     struct sock *sk)
866 {
867 	struct sock *req_sk, *nreq_sk;
868 	struct request_sock *nreq;
869 
870 	nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
871 	if (!nreq) {
872 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
873 
874 		/* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
875 		sock_put(sk);
876 		return NULL;
877 	}
878 
879 	req_sk = req_to_sk(req);
880 	nreq_sk = req_to_sk(nreq);
881 
882 	memcpy(nreq_sk, req_sk,
883 	       offsetof(struct sock, sk_dontcopy_begin));
884 	memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
885 	       req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
886 
887 	sk_node_init(&nreq_sk->sk_node);
888 	nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
889 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
890 	nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
891 #endif
892 	nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
893 
894 	nreq->rsk_listener = sk;
895 
896 	/* We need not acquire fastopenq->lock
897 	 * because the child socket is locked in inet_csk_listen_stop().
898 	 */
899 	if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
900 		rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
901 
902 	return nreq;
903 }
904 
905 static void reqsk_queue_migrated(struct request_sock_queue *queue,
906 				 const struct request_sock *req)
907 {
908 	if (req->num_timeout == 0)
909 		atomic_inc(&queue->young);
910 	atomic_inc(&queue->qlen);
911 }
912 
913 static void reqsk_migrate_reset(struct request_sock *req)
914 {
915 	req->saved_syn = NULL;
916 #if IS_ENABLED(CONFIG_IPV6)
917 	inet_rsk(req)->ipv6_opt = NULL;
918 	inet_rsk(req)->pktopts = NULL;
919 #else
920 	inet_rsk(req)->ireq_opt = NULL;
921 #endif
922 }
923 
924 /* return true if req was found in the ehash table */
925 static bool reqsk_queue_unlink(struct request_sock *req)
926 {
927 	struct sock *sk = req_to_sk(req);
928 	bool found = false;
929 
930 	if (sk_hashed(sk)) {
931 		struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
932 		spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
933 
934 		spin_lock(lock);
935 		found = __sk_nulls_del_node_init_rcu(sk);
936 		spin_unlock(lock);
937 	}
938 	if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
939 		reqsk_put(req);
940 	return found;
941 }
942 
943 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
944 {
945 	bool unlinked = reqsk_queue_unlink(req);
946 
947 	if (unlinked) {
948 		reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
949 		reqsk_put(req);
950 	}
951 	return unlinked;
952 }
953 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
954 
955 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
956 {
957 	inet_csk_reqsk_queue_drop(sk, req);
958 	reqsk_put(req);
959 }
960 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
961 
962 static void reqsk_timer_handler(struct timer_list *t)
963 {
964 	struct request_sock *req = from_timer(req, t, rsk_timer);
965 	struct request_sock *nreq = NULL, *oreq = req;
966 	struct sock *sk_listener = req->rsk_listener;
967 	struct inet_connection_sock *icsk;
968 	struct request_sock_queue *queue;
969 	struct net *net;
970 	int max_syn_ack_retries, qlen, expire = 0, resend = 0;
971 
972 	if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
973 		struct sock *nsk;
974 
975 		nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
976 		if (!nsk)
977 			goto drop;
978 
979 		nreq = inet_reqsk_clone(req, nsk);
980 		if (!nreq)
981 			goto drop;
982 
983 		/* The new timer for the cloned req can decrease the 2
984 		 * by calling inet_csk_reqsk_queue_drop_and_put(), so
985 		 * hold another count to prevent use-after-free and
986 		 * call reqsk_put() just before return.
987 		 */
988 		refcount_set(&nreq->rsk_refcnt, 2 + 1);
989 		timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
990 		reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
991 
992 		req = nreq;
993 		sk_listener = nsk;
994 	}
995 
996 	icsk = inet_csk(sk_listener);
997 	net = sock_net(sk_listener);
998 	max_syn_ack_retries = icsk->icsk_syn_retries ? :
999 		READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1000 	/* Normally all the openreqs are young and become mature
1001 	 * (i.e. converted to established socket) for first timeout.
1002 	 * If synack was not acknowledged for 1 second, it means
1003 	 * one of the following things: synack was lost, ack was lost,
1004 	 * rtt is high or nobody planned to ack (i.e. synflood).
1005 	 * When server is a bit loaded, queue is populated with old
1006 	 * open requests, reducing effective size of queue.
1007 	 * When server is well loaded, queue size reduces to zero
1008 	 * after several minutes of work. It is not synflood,
1009 	 * it is normal operation. The solution is pruning
1010 	 * too old entries overriding normal timeout, when
1011 	 * situation becomes dangerous.
1012 	 *
1013 	 * Essentially, we reserve half of room for young
1014 	 * embrions; and abort old ones without pity, if old
1015 	 * ones are about to clog our table.
1016 	 */
1017 	queue = &icsk->icsk_accept_queue;
1018 	qlen = reqsk_queue_len(queue);
1019 	if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1020 		int young = reqsk_queue_len_young(queue) << 1;
1021 
1022 		while (max_syn_ack_retries > 2) {
1023 			if (qlen < young)
1024 				break;
1025 			max_syn_ack_retries--;
1026 			young <<= 1;
1027 		}
1028 	}
1029 	syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
1030 		       &expire, &resend);
1031 	req->rsk_ops->syn_ack_timeout(req);
1032 	if (!expire &&
1033 	    (!resend ||
1034 	     !inet_rtx_syn_ack(sk_listener, req) ||
1035 	     inet_rsk(req)->acked)) {
1036 		if (req->num_timeout++ == 0)
1037 			atomic_dec(&queue->young);
1038 		mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
1039 
1040 		if (!nreq)
1041 			return;
1042 
1043 		if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
1044 			/* delete timer */
1045 			inet_csk_reqsk_queue_drop(sk_listener, nreq);
1046 			goto no_ownership;
1047 		}
1048 
1049 		__NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1050 		reqsk_migrate_reset(oreq);
1051 		reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1052 		reqsk_put(oreq);
1053 
1054 		reqsk_put(nreq);
1055 		return;
1056 	}
1057 
1058 	/* Even if we can clone the req, we may need not retransmit any more
1059 	 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1060 	 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1061 	 */
1062 	if (nreq) {
1063 		__NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1064 no_ownership:
1065 		reqsk_migrate_reset(nreq);
1066 		reqsk_queue_removed(queue, nreq);
1067 		__reqsk_free(nreq);
1068 	}
1069 
1070 drop:
1071 	inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
1072 }
1073 
1074 static void reqsk_queue_hash_req(struct request_sock *req,
1075 				 unsigned long timeout)
1076 {
1077 	timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1078 	mod_timer(&req->rsk_timer, jiffies + timeout);
1079 
1080 	inet_ehash_insert(req_to_sk(req), NULL, NULL);
1081 	/* before letting lookups find us, make sure all req fields
1082 	 * are committed to memory and refcnt initialized.
1083 	 */
1084 	smp_wmb();
1085 	refcount_set(&req->rsk_refcnt, 2 + 1);
1086 }
1087 
1088 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1089 				   unsigned long timeout)
1090 {
1091 	reqsk_queue_hash_req(req, timeout);
1092 	inet_csk_reqsk_queue_added(sk);
1093 }
1094 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1095 
1096 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1097 			   const gfp_t priority)
1098 {
1099 	struct inet_connection_sock *icsk = inet_csk(newsk);
1100 
1101 	if (!icsk->icsk_ulp_ops)
1102 		return;
1103 
1104 	if (icsk->icsk_ulp_ops->clone)
1105 		icsk->icsk_ulp_ops->clone(req, newsk, priority);
1106 }
1107 
1108 /**
1109  *	inet_csk_clone_lock - clone an inet socket, and lock its clone
1110  *	@sk: the socket to clone
1111  *	@req: request_sock
1112  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1113  *
1114  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1115  */
1116 struct sock *inet_csk_clone_lock(const struct sock *sk,
1117 				 const struct request_sock *req,
1118 				 const gfp_t priority)
1119 {
1120 	struct sock *newsk = sk_clone_lock(sk, priority);
1121 
1122 	if (newsk) {
1123 		struct inet_connection_sock *newicsk = inet_csk(newsk);
1124 
1125 		inet_sk_set_state(newsk, TCP_SYN_RECV);
1126 		newicsk->icsk_bind_hash = NULL;
1127 		newicsk->icsk_bind2_hash = NULL;
1128 
1129 		inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
1130 		inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
1131 		inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
1132 
1133 		/* listeners have SOCK_RCU_FREE, not the children */
1134 		sock_reset_flag(newsk, SOCK_RCU_FREE);
1135 
1136 		inet_sk(newsk)->mc_list = NULL;
1137 
1138 		newsk->sk_mark = inet_rsk(req)->ir_mark;
1139 		atomic64_set(&newsk->sk_cookie,
1140 			     atomic64_read(&inet_rsk(req)->ir_cookie));
1141 
1142 		newicsk->icsk_retransmits = 0;
1143 		newicsk->icsk_backoff	  = 0;
1144 		newicsk->icsk_probes_out  = 0;
1145 		newicsk->icsk_probes_tstamp = 0;
1146 
1147 		/* Deinitialize accept_queue to trap illegal accesses. */
1148 		memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
1149 
1150 		inet_clone_ulp(req, newsk, priority);
1151 
1152 		security_inet_csk_clone(newsk, req);
1153 	}
1154 	return newsk;
1155 }
1156 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
1157 
1158 /*
1159  * At this point, there should be no process reference to this
1160  * socket, and thus no user references at all.  Therefore we
1161  * can assume the socket waitqueue is inactive and nobody will
1162  * try to jump onto it.
1163  */
1164 void inet_csk_destroy_sock(struct sock *sk)
1165 {
1166 	WARN_ON(sk->sk_state != TCP_CLOSE);
1167 	WARN_ON(!sock_flag(sk, SOCK_DEAD));
1168 
1169 	/* It cannot be in hash table! */
1170 	WARN_ON(!sk_unhashed(sk));
1171 
1172 	/* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1173 	WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1174 
1175 	sk->sk_prot->destroy(sk);
1176 
1177 	sk_stream_kill_queues(sk);
1178 
1179 	xfrm_sk_free_policy(sk);
1180 
1181 	sk_refcnt_debug_release(sk);
1182 
1183 	this_cpu_dec(*sk->sk_prot->orphan_count);
1184 
1185 	sock_put(sk);
1186 }
1187 EXPORT_SYMBOL(inet_csk_destroy_sock);
1188 
1189 /* This function allows to force a closure of a socket after the call to
1190  * tcp/dccp_create_openreq_child().
1191  */
1192 void inet_csk_prepare_forced_close(struct sock *sk)
1193 	__releases(&sk->sk_lock.slock)
1194 {
1195 	/* sk_clone_lock locked the socket and set refcnt to 2 */
1196 	bh_unlock_sock(sk);
1197 	sock_put(sk);
1198 	inet_csk_prepare_for_destroy_sock(sk);
1199 	inet_sk(sk)->inet_num = 0;
1200 }
1201 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1202 
1203 static int inet_ulp_can_listen(const struct sock *sk)
1204 {
1205 	const struct inet_connection_sock *icsk = inet_csk(sk);
1206 
1207 	if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone)
1208 		return -EINVAL;
1209 
1210 	return 0;
1211 }
1212 
1213 int inet_csk_listen_start(struct sock *sk)
1214 {
1215 	struct inet_connection_sock *icsk = inet_csk(sk);
1216 	struct inet_sock *inet = inet_sk(sk);
1217 	int err;
1218 
1219 	err = inet_ulp_can_listen(sk);
1220 	if (unlikely(err))
1221 		return err;
1222 
1223 	reqsk_queue_alloc(&icsk->icsk_accept_queue);
1224 
1225 	sk->sk_ack_backlog = 0;
1226 	inet_csk_delack_init(sk);
1227 
1228 	if (sk->sk_txrehash == SOCK_TXREHASH_DEFAULT)
1229 		sk->sk_txrehash = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1230 
1231 	/* There is race window here: we announce ourselves listening,
1232 	 * but this transition is still not validated by get_port().
1233 	 * It is OK, because this socket enters to hash table only
1234 	 * after validation is complete.
1235 	 */
1236 	inet_sk_state_store(sk, TCP_LISTEN);
1237 	err = sk->sk_prot->get_port(sk, inet->inet_num);
1238 	if (!err) {
1239 		inet->inet_sport = htons(inet->inet_num);
1240 
1241 		sk_dst_reset(sk);
1242 		err = sk->sk_prot->hash(sk);
1243 
1244 		if (likely(!err))
1245 			return 0;
1246 	}
1247 
1248 	inet_sk_set_state(sk, TCP_CLOSE);
1249 	return err;
1250 }
1251 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1252 
1253 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1254 			      struct sock *child)
1255 {
1256 	sk->sk_prot->disconnect(child, O_NONBLOCK);
1257 
1258 	sock_orphan(child);
1259 
1260 	this_cpu_inc(*sk->sk_prot->orphan_count);
1261 
1262 	if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1263 		BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1264 		BUG_ON(sk != req->rsk_listener);
1265 
1266 		/* Paranoid, to prevent race condition if
1267 		 * an inbound pkt destined for child is
1268 		 * blocked by sock lock in tcp_v4_rcv().
1269 		 * Also to satisfy an assertion in
1270 		 * tcp_v4_destroy_sock().
1271 		 */
1272 		RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1273 	}
1274 	inet_csk_destroy_sock(child);
1275 }
1276 
1277 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1278 				      struct request_sock *req,
1279 				      struct sock *child)
1280 {
1281 	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1282 
1283 	spin_lock(&queue->rskq_lock);
1284 	if (unlikely(sk->sk_state != TCP_LISTEN)) {
1285 		inet_child_forget(sk, req, child);
1286 		child = NULL;
1287 	} else {
1288 		req->sk = child;
1289 		req->dl_next = NULL;
1290 		if (queue->rskq_accept_head == NULL)
1291 			WRITE_ONCE(queue->rskq_accept_head, req);
1292 		else
1293 			queue->rskq_accept_tail->dl_next = req;
1294 		queue->rskq_accept_tail = req;
1295 		sk_acceptq_added(sk);
1296 	}
1297 	spin_unlock(&queue->rskq_lock);
1298 	return child;
1299 }
1300 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1301 
1302 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1303 					 struct request_sock *req, bool own_req)
1304 {
1305 	if (own_req) {
1306 		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1307 		reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1308 
1309 		if (sk != req->rsk_listener) {
1310 			/* another listening sk has been selected,
1311 			 * migrate the req to it.
1312 			 */
1313 			struct request_sock *nreq;
1314 
1315 			/* hold a refcnt for the nreq->rsk_listener
1316 			 * which is assigned in inet_reqsk_clone()
1317 			 */
1318 			sock_hold(sk);
1319 			nreq = inet_reqsk_clone(req, sk);
1320 			if (!nreq) {
1321 				inet_child_forget(sk, req, child);
1322 				goto child_put;
1323 			}
1324 
1325 			refcount_set(&nreq->rsk_refcnt, 1);
1326 			if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1327 				__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1328 				reqsk_migrate_reset(req);
1329 				reqsk_put(req);
1330 				return child;
1331 			}
1332 
1333 			__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1334 			reqsk_migrate_reset(nreq);
1335 			__reqsk_free(nreq);
1336 		} else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1337 			return child;
1338 		}
1339 	}
1340 	/* Too bad, another child took ownership of the request, undo. */
1341 child_put:
1342 	bh_unlock_sock(child);
1343 	sock_put(child);
1344 	return NULL;
1345 }
1346 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1347 
1348 /*
1349  *	This routine closes sockets which have been at least partially
1350  *	opened, but not yet accepted.
1351  */
1352 void inet_csk_listen_stop(struct sock *sk)
1353 {
1354 	struct inet_connection_sock *icsk = inet_csk(sk);
1355 	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1356 	struct request_sock *next, *req;
1357 
1358 	/* Following specs, it would be better either to send FIN
1359 	 * (and enter FIN-WAIT-1, it is normal close)
1360 	 * or to send active reset (abort).
1361 	 * Certainly, it is pretty dangerous while synflood, but it is
1362 	 * bad justification for our negligence 8)
1363 	 * To be honest, we are not able to make either
1364 	 * of the variants now.			--ANK
1365 	 */
1366 	while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1367 		struct sock *child = req->sk, *nsk;
1368 		struct request_sock *nreq;
1369 
1370 		local_bh_disable();
1371 		bh_lock_sock(child);
1372 		WARN_ON(sock_owned_by_user(child));
1373 		sock_hold(child);
1374 
1375 		nsk = reuseport_migrate_sock(sk, child, NULL);
1376 		if (nsk) {
1377 			nreq = inet_reqsk_clone(req, nsk);
1378 			if (nreq) {
1379 				refcount_set(&nreq->rsk_refcnt, 1);
1380 
1381 				if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1382 					__NET_INC_STATS(sock_net(nsk),
1383 							LINUX_MIB_TCPMIGRATEREQSUCCESS);
1384 					reqsk_migrate_reset(req);
1385 				} else {
1386 					__NET_INC_STATS(sock_net(nsk),
1387 							LINUX_MIB_TCPMIGRATEREQFAILURE);
1388 					reqsk_migrate_reset(nreq);
1389 					__reqsk_free(nreq);
1390 				}
1391 
1392 				/* inet_csk_reqsk_queue_add() has already
1393 				 * called inet_child_forget() on failure case.
1394 				 */
1395 				goto skip_child_forget;
1396 			}
1397 		}
1398 
1399 		inet_child_forget(sk, req, child);
1400 skip_child_forget:
1401 		reqsk_put(req);
1402 		bh_unlock_sock(child);
1403 		local_bh_enable();
1404 		sock_put(child);
1405 
1406 		cond_resched();
1407 	}
1408 	if (queue->fastopenq.rskq_rst_head) {
1409 		/* Free all the reqs queued in rskq_rst_head. */
1410 		spin_lock_bh(&queue->fastopenq.lock);
1411 		req = queue->fastopenq.rskq_rst_head;
1412 		queue->fastopenq.rskq_rst_head = NULL;
1413 		spin_unlock_bh(&queue->fastopenq.lock);
1414 		while (req != NULL) {
1415 			next = req->dl_next;
1416 			reqsk_put(req);
1417 			req = next;
1418 		}
1419 	}
1420 	WARN_ON_ONCE(sk->sk_ack_backlog);
1421 }
1422 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1423 
1424 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1425 {
1426 	struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1427 	const struct inet_sock *inet = inet_sk(sk);
1428 
1429 	sin->sin_family		= AF_INET;
1430 	sin->sin_addr.s_addr	= inet->inet_daddr;
1431 	sin->sin_port		= inet->inet_dport;
1432 }
1433 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1434 
1435 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1436 {
1437 	const struct inet_sock *inet = inet_sk(sk);
1438 	const struct ip_options_rcu *inet_opt;
1439 	__be32 daddr = inet->inet_daddr;
1440 	struct flowi4 *fl4;
1441 	struct rtable *rt;
1442 
1443 	rcu_read_lock();
1444 	inet_opt = rcu_dereference(inet->inet_opt);
1445 	if (inet_opt && inet_opt->opt.srr)
1446 		daddr = inet_opt->opt.faddr;
1447 	fl4 = &fl->u.ip4;
1448 	rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1449 				   inet->inet_saddr, inet->inet_dport,
1450 				   inet->inet_sport, sk->sk_protocol,
1451 				   RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1452 	if (IS_ERR(rt))
1453 		rt = NULL;
1454 	if (rt)
1455 		sk_setup_caps(sk, &rt->dst);
1456 	rcu_read_unlock();
1457 
1458 	return &rt->dst;
1459 }
1460 
1461 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1462 {
1463 	struct dst_entry *dst = __sk_dst_check(sk, 0);
1464 	struct inet_sock *inet = inet_sk(sk);
1465 
1466 	if (!dst) {
1467 		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1468 		if (!dst)
1469 			goto out;
1470 	}
1471 	dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1472 
1473 	dst = __sk_dst_check(sk, 0);
1474 	if (!dst)
1475 		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1476 out:
1477 	return dst;
1478 }
1479 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
1480