xref: /linux/net/ipv4/inet_connection_sock.c (revision 6dfafbd0299a60bfb5d5e277fdf100037c7ded07)
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 /**
121  *	inet_sk_get_local_port_range - fetch ephemeral ports range
122  *	@sk: socket
123  *	@low: pointer to low port
124  *	@high: pointer to high port
125  *
126  *	Fetch netns port range (/proc/sys/net/ipv4/ip_local_port_range)
127  *	Range can be overridden if socket got IP_LOCAL_PORT_RANGE option.
128  *	Returns true if IP_LOCAL_PORT_RANGE was set on this socket.
129  */
130 bool inet_sk_get_local_port_range(const struct sock *sk, int *low, int *high)
131 {
132 	int lo, hi, sk_lo, sk_hi;
133 	bool local_range = false;
134 	u32 sk_range;
135 
136 	inet_get_local_port_range(sock_net(sk), &lo, &hi);
137 
138 	sk_range = READ_ONCE(inet_sk(sk)->local_port_range);
139 	if (unlikely(sk_range)) {
140 		sk_lo = sk_range & 0xffff;
141 		sk_hi = sk_range >> 16;
142 
143 		if (lo <= sk_lo && sk_lo <= hi)
144 			lo = sk_lo;
145 		if (lo <= sk_hi && sk_hi <= hi)
146 			hi = sk_hi;
147 		local_range = true;
148 	}
149 
150 	*low = lo;
151 	*high = hi;
152 	return local_range;
153 }
154 EXPORT_SYMBOL(inet_sk_get_local_port_range);
155 
156 static bool inet_use_bhash2_on_bind(const struct sock *sk)
157 {
158 #if IS_ENABLED(CONFIG_IPV6)
159 	if (sk->sk_family == AF_INET6) {
160 		if (ipv6_addr_any(&sk->sk_v6_rcv_saddr))
161 			return false;
162 
163 		if (!ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
164 			return true;
165 	}
166 #endif
167 	return sk->sk_rcv_saddr != htonl(INADDR_ANY);
168 }
169 
170 static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2,
171 			       kuid_t uid, bool relax,
172 			       bool reuseport_cb_ok, bool reuseport_ok)
173 {
174 	int bound_dev_if2;
175 
176 	if (sk == sk2)
177 		return false;
178 
179 	bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
180 
181 	if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
182 	    sk->sk_bound_dev_if == bound_dev_if2) {
183 		if (sk->sk_reuse && sk2->sk_reuse &&
184 		    sk2->sk_state != TCP_LISTEN) {
185 			if (!relax || (!reuseport_ok && sk->sk_reuseport &&
186 				       sk2->sk_reuseport && reuseport_cb_ok &&
187 				       (sk2->sk_state == TCP_TIME_WAIT ||
188 					uid_eq(uid, sk_uid(sk2)))))
189 				return true;
190 		} else if (!reuseport_ok || !sk->sk_reuseport ||
191 			   !sk2->sk_reuseport || !reuseport_cb_ok ||
192 			   (sk2->sk_state != TCP_TIME_WAIT &&
193 			    !uid_eq(uid, sk_uid(sk2)))) {
194 			return true;
195 		}
196 	}
197 	return false;
198 }
199 
200 static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2,
201 				   kuid_t uid, bool relax,
202 				   bool reuseport_cb_ok, bool reuseport_ok)
203 {
204 	if (ipv6_only_sock(sk2)) {
205 		if (sk->sk_family == AF_INET)
206 			return false;
207 
208 #if IS_ENABLED(CONFIG_IPV6)
209 		if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
210 			return false;
211 #endif
212 	}
213 
214 	return inet_bind_conflict(sk, sk2, uid, relax,
215 				  reuseport_cb_ok, reuseport_ok);
216 }
217 
218 static bool inet_bhash2_conflict(const struct sock *sk,
219 				 const struct inet_bind2_bucket *tb2,
220 				 kuid_t uid,
221 				 bool relax, bool reuseport_cb_ok,
222 				 bool reuseport_ok)
223 {
224 	struct sock *sk2;
225 
226 	sk_for_each_bound(sk2, &tb2->owners) {
227 		if (__inet_bhash2_conflict(sk, sk2, uid, relax,
228 					   reuseport_cb_ok, reuseport_ok))
229 			return true;
230 	}
231 
232 	return false;
233 }
234 
235 #define sk_for_each_bound_bhash(__sk, __tb2, __tb)			\
236 	hlist_for_each_entry(__tb2, &(__tb)->bhash2, bhash_node)	\
237 		sk_for_each_bound((__sk), &(__tb2)->owners)
238 
239 /* This should be called only when the tb and tb2 hashbuckets' locks are held */
240 static int inet_csk_bind_conflict(const struct sock *sk,
241 				  const struct inet_bind_bucket *tb,
242 				  const struct inet_bind2_bucket *tb2, /* may be null */
243 				  bool relax, bool reuseport_ok)
244 {
245 	struct sock_reuseport *reuseport_cb;
246 	kuid_t uid = sk_uid(sk);
247 	bool reuseport_cb_ok;
248 	struct sock *sk2;
249 
250 	rcu_read_lock();
251 	reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
252 	/* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
253 	reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
254 	rcu_read_unlock();
255 
256 	/* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if
257 	 * ipv4) should have been checked already. We need to do these two
258 	 * checks separately because their spinlocks have to be acquired/released
259 	 * independently of each other, to prevent possible deadlocks
260 	 */
261 	if (inet_use_bhash2_on_bind(sk))
262 		return tb2 && inet_bhash2_conflict(sk, tb2, uid, relax,
263 						   reuseport_cb_ok, reuseport_ok);
264 
265 	/* Unlike other sk lookup places we do not check
266 	 * for sk_net here, since _all_ the socks listed
267 	 * in tb->owners and tb2->owners list belong
268 	 * to the same net - the one this bucket belongs to.
269 	 */
270 	sk_for_each_bound_bhash(sk2, tb2, tb) {
271 		if (!inet_bind_conflict(sk, sk2, uid, relax, reuseport_cb_ok, reuseport_ok))
272 			continue;
273 
274 		if (inet_rcv_saddr_equal(sk, sk2, true))
275 			return true;
276 	}
277 
278 	return false;
279 }
280 
281 /* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or
282  * INADDR_ANY (if ipv4) socket.
283  *
284  * Caller must hold bhash hashbucket lock with local bh disabled, to protect
285  * against concurrent binds on the port for addr any
286  */
287 static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev,
288 					  bool relax, bool reuseport_ok)
289 {
290 	const struct net *net = sock_net(sk);
291 	struct sock_reuseport *reuseport_cb;
292 	struct inet_bind_hashbucket *head2;
293 	struct inet_bind2_bucket *tb2;
294 	kuid_t uid = sk_uid(sk);
295 	bool conflict = false;
296 	bool reuseport_cb_ok;
297 
298 	rcu_read_lock();
299 	reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
300 	/* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
301 	reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
302 	rcu_read_unlock();
303 
304 	head2 = inet_bhash2_addr_any_hashbucket(sk, net, port);
305 
306 	spin_lock(&head2->lock);
307 
308 	inet_bind_bucket_for_each(tb2, &head2->chain) {
309 		if (!inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
310 			continue;
311 
312 		if (!inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,	reuseport_ok))
313 			continue;
314 
315 		conflict = true;
316 		break;
317 	}
318 
319 	spin_unlock(&head2->lock);
320 
321 	return conflict;
322 }
323 
324 /*
325  * Find an open port number for the socket.  Returns with the
326  * inet_bind_hashbucket locks held if successful.
327  */
328 static struct inet_bind_hashbucket *
329 inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret,
330 			struct inet_bind2_bucket **tb2_ret,
331 			struct inet_bind_hashbucket **head2_ret, int *port_ret)
332 {
333 	struct inet_hashinfo *hinfo = tcp_get_hashinfo(sk);
334 	int i, low, high, attempt_half, port, l3mdev;
335 	struct inet_bind_hashbucket *head, *head2;
336 	struct net *net = sock_net(sk);
337 	struct inet_bind2_bucket *tb2;
338 	struct inet_bind_bucket *tb;
339 	u32 remaining, offset;
340 	bool relax = false;
341 
342 	l3mdev = inet_sk_bound_l3mdev(sk);
343 ports_exhausted:
344 	attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
345 other_half_scan:
346 	inet_sk_get_local_port_range(sk, &low, &high);
347 	high++; /* [32768, 60999] -> [32768, 61000[ */
348 	if (high - low < 4)
349 		attempt_half = 0;
350 	if (attempt_half) {
351 		int half = low + (((high - low) >> 2) << 1);
352 
353 		if (attempt_half == 1)
354 			high = half;
355 		else
356 			low = half;
357 	}
358 	remaining = high - low;
359 	if (likely(remaining > 1))
360 		remaining &= ~1U;
361 
362 	offset = get_random_u32_below(remaining);
363 	/* __inet_hash_connect() favors ports having @low parity
364 	 * We do the opposite to not pollute connect() users.
365 	 */
366 	offset |= 1U;
367 
368 other_parity_scan:
369 	port = low + offset;
370 	for (i = 0; i < remaining; i += 2, port += 2) {
371 		if (unlikely(port >= high))
372 			port -= remaining;
373 		if (inet_is_local_reserved_port(net, port))
374 			continue;
375 		head = &hinfo->bhash[inet_bhashfn(net, port,
376 						  hinfo->bhash_size)];
377 		spin_lock_bh(&head->lock);
378 		if (inet_use_bhash2_on_bind(sk)) {
379 			if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, false))
380 				goto next_port;
381 		}
382 
383 		head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
384 		spin_lock(&head2->lock);
385 		tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
386 		inet_bind_bucket_for_each(tb, &head->chain)
387 			if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
388 				if (!inet_csk_bind_conflict(sk, tb, tb2,
389 							    relax, false))
390 					goto success;
391 				spin_unlock(&head2->lock);
392 				goto next_port;
393 			}
394 		tb = NULL;
395 		goto success;
396 next_port:
397 		spin_unlock_bh(&head->lock);
398 		cond_resched();
399 	}
400 
401 	offset--;
402 	if (!(offset & 1))
403 		goto other_parity_scan;
404 
405 	if (attempt_half == 1) {
406 		/* OK we now try the upper half of the range */
407 		attempt_half = 2;
408 		goto other_half_scan;
409 	}
410 
411 	if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
412 		/* We still have a chance to connect to different destinations */
413 		relax = true;
414 		goto ports_exhausted;
415 	}
416 	return NULL;
417 success:
418 	*port_ret = port;
419 	*tb_ret = tb;
420 	*tb2_ret = tb2;
421 	*head2_ret = head2;
422 	return head;
423 }
424 
425 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
426 				     const struct sock *sk)
427 {
428 	if (tb->fastreuseport <= 0)
429 		return 0;
430 	if (!sk->sk_reuseport)
431 		return 0;
432 	if (rcu_access_pointer(sk->sk_reuseport_cb))
433 		return 0;
434 	if (!uid_eq(tb->fastuid, sk_uid(sk)))
435 		return 0;
436 	/* We only need to check the rcv_saddr if this tb was once marked
437 	 * without fastreuseport and then was reset, as we can only know that
438 	 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
439 	 * owners list.
440 	 */
441 	if (tb->fastreuseport == FASTREUSEPORT_ANY)
442 		return 1;
443 #if IS_ENABLED(CONFIG_IPV6)
444 	if (tb->fast_sk_family == AF_INET6)
445 		return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
446 					    inet6_rcv_saddr(sk),
447 					    tb->fast_rcv_saddr,
448 					    sk->sk_rcv_saddr,
449 					    tb->fast_ipv6_only,
450 					    ipv6_only_sock(sk), true, false);
451 #endif
452 	return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
453 				    ipv6_only_sock(sk), true, false);
454 }
455 
456 void inet_csk_update_fastreuse(const struct sock *sk,
457 			       struct inet_bind_bucket *tb,
458 			       struct inet_bind2_bucket *tb2)
459 {
460 	bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
461 
462 	if (hlist_empty(&tb->bhash2)) {
463 		tb->fastreuse = reuse;
464 		if (sk->sk_reuseport) {
465 			tb->fastreuseport = FASTREUSEPORT_ANY;
466 			tb->fastuid = sk_uid(sk);
467 			tb->fast_rcv_saddr = sk->sk_rcv_saddr;
468 			tb->fast_ipv6_only = ipv6_only_sock(sk);
469 			tb->fast_sk_family = sk->sk_family;
470 #if IS_ENABLED(CONFIG_IPV6)
471 			tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
472 #endif
473 		} else {
474 			tb->fastreuseport = 0;
475 		}
476 	} else {
477 		if (!reuse)
478 			tb->fastreuse = 0;
479 		if (sk->sk_reuseport) {
480 			/* We didn't match or we don't have fastreuseport set on
481 			 * the tb, but we have sk_reuseport set on this socket
482 			 * and we know that there are no bind conflicts with
483 			 * this socket in this tb, so reset our tb's reuseport
484 			 * settings so that any subsequent sockets that match
485 			 * our current socket will be put on the fast path.
486 			 *
487 			 * If we reset we need to set FASTREUSEPORT_STRICT so we
488 			 * do extra checking for all subsequent sk_reuseport
489 			 * socks.
490 			 */
491 			if (!sk_reuseport_match(tb, sk)) {
492 				tb->fastreuseport = FASTREUSEPORT_STRICT;
493 				tb->fastuid = sk_uid(sk);
494 				tb->fast_rcv_saddr = sk->sk_rcv_saddr;
495 				tb->fast_ipv6_only = ipv6_only_sock(sk);
496 				tb->fast_sk_family = sk->sk_family;
497 #if IS_ENABLED(CONFIG_IPV6)
498 				tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
499 #endif
500 			}
501 		} else {
502 			tb->fastreuseport = 0;
503 		}
504 	}
505 
506 	tb2->fastreuse = tb->fastreuse;
507 	tb2->fastreuseport = tb->fastreuseport;
508 }
509 
510 /* Obtain a reference to a local port for the given sock,
511  * if snum is zero it means select any available local port.
512  * We try to allocate an odd port (and leave even ports for connect())
513  */
514 int inet_csk_get_port(struct sock *sk, unsigned short snum)
515 {
516 	bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
517 	bool found_port = false, check_bind_conflict = true;
518 	bool bhash_created = false, bhash2_created = false;
519 	struct inet_hashinfo *hinfo = tcp_get_hashinfo(sk);
520 	int ret = -EADDRINUSE, port = snum, l3mdev;
521 	struct inet_bind_hashbucket *head, *head2;
522 	struct inet_bind2_bucket *tb2 = NULL;
523 	struct inet_bind_bucket *tb = NULL;
524 	bool head2_lock_acquired = false;
525 	struct net *net = sock_net(sk);
526 
527 	l3mdev = inet_sk_bound_l3mdev(sk);
528 
529 	if (!port) {
530 		head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
531 		if (!head)
532 			return ret;
533 
534 		head2_lock_acquired = true;
535 
536 		if (tb && tb2)
537 			goto success;
538 		found_port = true;
539 	} else {
540 		head = &hinfo->bhash[inet_bhashfn(net, port,
541 						  hinfo->bhash_size)];
542 		spin_lock_bh(&head->lock);
543 		inet_bind_bucket_for_each(tb, &head->chain)
544 			if (inet_bind_bucket_match(tb, net, port, l3mdev))
545 				break;
546 	}
547 
548 	if (!tb) {
549 		tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
550 					     head, port, l3mdev);
551 		if (!tb)
552 			goto fail_unlock;
553 		bhash_created = true;
554 	}
555 
556 	if (!found_port) {
557 		if (!hlist_empty(&tb->bhash2)) {
558 			if (sk->sk_reuse == SK_FORCE_REUSE ||
559 			    (tb->fastreuse > 0 && reuse) ||
560 			    sk_reuseport_match(tb, sk))
561 				check_bind_conflict = false;
562 		}
563 
564 		if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) {
565 			if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, true, true))
566 				goto fail_unlock;
567 		}
568 
569 		head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
570 		spin_lock(&head2->lock);
571 		head2_lock_acquired = true;
572 		tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
573 	}
574 
575 	if (!tb2) {
576 		tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
577 					       net, head2, tb, sk);
578 		if (!tb2)
579 			goto fail_unlock;
580 		bhash2_created = true;
581 	}
582 
583 	if (!found_port && check_bind_conflict) {
584 		if (inet_csk_bind_conflict(sk, tb, tb2, true, true))
585 			goto fail_unlock;
586 	}
587 
588 success:
589 	inet_csk_update_fastreuse(sk, tb, tb2);
590 
591 	if (!inet_csk(sk)->icsk_bind_hash)
592 		inet_bind_hash(sk, tb, tb2, port);
593 	WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
594 	WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
595 	ret = 0;
596 
597 fail_unlock:
598 	if (ret) {
599 		if (bhash2_created)
600 			inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep, tb2);
601 		if (bhash_created)
602 			inet_bind_bucket_destroy(tb);
603 	}
604 	if (head2_lock_acquired)
605 		spin_unlock(&head2->lock);
606 	spin_unlock_bh(&head->lock);
607 	return ret;
608 }
609 EXPORT_SYMBOL_GPL(inet_csk_get_port);
610 
611 /*
612  * Wait for an incoming connection, avoid race conditions. This must be called
613  * with the socket locked.
614  */
615 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
616 {
617 	struct inet_connection_sock *icsk = inet_csk(sk);
618 	DEFINE_WAIT(wait);
619 	int err;
620 
621 	/*
622 	 * True wake-one mechanism for incoming connections: only
623 	 * one process gets woken up, not the 'whole herd'.
624 	 * Since we do not 'race & poll' for established sockets
625 	 * anymore, the common case will execute the loop only once.
626 	 *
627 	 * Subtle issue: "add_wait_queue_exclusive()" will be added
628 	 * after any current non-exclusive waiters, and we know that
629 	 * it will always _stay_ after any new non-exclusive waiters
630 	 * because all non-exclusive waiters are added at the
631 	 * beginning of the wait-queue. As such, it's ok to "drop"
632 	 * our exclusiveness temporarily when we get woken up without
633 	 * having to remove and re-insert us on the wait queue.
634 	 */
635 	for (;;) {
636 		prepare_to_wait_exclusive(sk_sleep(sk), &wait,
637 					  TASK_INTERRUPTIBLE);
638 		release_sock(sk);
639 		if (reqsk_queue_empty(&icsk->icsk_accept_queue))
640 			timeo = schedule_timeout(timeo);
641 		sched_annotate_sleep();
642 		lock_sock(sk);
643 		err = 0;
644 		if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
645 			break;
646 		err = -EINVAL;
647 		if (sk->sk_state != TCP_LISTEN)
648 			break;
649 		err = sock_intr_errno(timeo);
650 		if (signal_pending(current))
651 			break;
652 		err = -EAGAIN;
653 		if (!timeo)
654 			break;
655 	}
656 	finish_wait(sk_sleep(sk), &wait);
657 	return err;
658 }
659 
660 /*
661  * This will accept the next outstanding connection.
662  */
663 struct sock *inet_csk_accept(struct sock *sk, struct proto_accept_arg *arg)
664 {
665 	struct inet_connection_sock *icsk = inet_csk(sk);
666 	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
667 	struct request_sock *req;
668 	struct sock *newsk;
669 	int error;
670 
671 	lock_sock(sk);
672 
673 	/* We need to make sure that this socket is listening,
674 	 * and that it has something pending.
675 	 */
676 	error = -EINVAL;
677 	if (sk->sk_state != TCP_LISTEN)
678 		goto out_err;
679 
680 	/* Find already established connection */
681 	if (reqsk_queue_empty(queue)) {
682 		long timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK);
683 
684 		/* If this is a non blocking socket don't sleep */
685 		error = -EAGAIN;
686 		if (!timeo)
687 			goto out_err;
688 
689 		error = inet_csk_wait_for_connect(sk, timeo);
690 		if (error)
691 			goto out_err;
692 	}
693 	req = reqsk_queue_remove(queue, sk);
694 	arg->is_empty = reqsk_queue_empty(queue);
695 	newsk = req->sk;
696 
697 	if (sk->sk_protocol == IPPROTO_TCP &&
698 	    tcp_rsk(req)->tfo_listener) {
699 		spin_lock_bh(&queue->fastopenq.lock);
700 		if (tcp_rsk(req)->tfo_listener) {
701 			/* We are still waiting for the final ACK from 3WHS
702 			 * so can't free req now. Instead, we set req->sk to
703 			 * NULL to signify that the child socket is taken
704 			 * so reqsk_fastopen_remove() will free the req
705 			 * when 3WHS finishes (or is aborted).
706 			 */
707 			req->sk = NULL;
708 			req = NULL;
709 		}
710 		spin_unlock_bh(&queue->fastopenq.lock);
711 	}
712 
713 	release_sock(sk);
714 
715 	if (req)
716 		reqsk_put(req);
717 
718 	inet_init_csk_locks(newsk);
719 	return newsk;
720 
721 out_err:
722 	release_sock(sk);
723 	arg->err = error;
724 	return NULL;
725 }
726 EXPORT_SYMBOL(inet_csk_accept);
727 
728 /*
729  * Using different timers for retransmit, delayed acks and probes
730  * We may wish use just one timer maintaining a list of expire jiffies
731  * to optimize.
732  */
733 void inet_csk_init_xmit_timers(struct sock *sk,
734 			       void (*retransmit_handler)(struct timer_list *t),
735 			       void (*delack_handler)(struct timer_list *t),
736 			       void (*keepalive_handler)(struct timer_list *t))
737 {
738 	struct inet_connection_sock *icsk = inet_csk(sk);
739 
740 	timer_setup(&sk->tcp_retransmit_timer, retransmit_handler, 0);
741 	timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
742 	timer_setup(&icsk->icsk_keepalive_timer, keepalive_handler, 0);
743 	icsk->icsk_pending = icsk->icsk_ack.pending = 0;
744 }
745 
746 void inet_csk_clear_xmit_timers(struct sock *sk)
747 {
748 	struct inet_connection_sock *icsk = inet_csk(sk);
749 
750 	smp_store_release(&icsk->icsk_pending, 0);
751 	smp_store_release(&icsk->icsk_ack.pending, 0);
752 
753 	sk_stop_timer(sk, &sk->tcp_retransmit_timer);
754 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
755 	sk_stop_timer(sk, &icsk->icsk_keepalive_timer);
756 }
757 
758 void inet_csk_clear_xmit_timers_sync(struct sock *sk)
759 {
760 	struct inet_connection_sock *icsk = inet_csk(sk);
761 
762 	/* ongoing timer handlers need to acquire socket lock. */
763 	sock_not_owned_by_me(sk);
764 
765 	smp_store_release(&icsk->icsk_pending, 0);
766 	smp_store_release(&icsk->icsk_ack.pending, 0);
767 
768 	sk_stop_timer_sync(sk, &sk->tcp_retransmit_timer);
769 	sk_stop_timer_sync(sk, &icsk->icsk_delack_timer);
770 	sk_stop_timer_sync(sk, &icsk->icsk_keepalive_timer);
771 }
772 
773 struct dst_entry *inet_csk_route_req(const struct sock *sk,
774 				     struct flowi4 *fl4,
775 				     const struct request_sock *req)
776 {
777 	const struct inet_request_sock *ireq = inet_rsk(req);
778 	struct net *net = read_pnet(&ireq->ireq_net);
779 	struct ip_options_rcu *opt;
780 	struct rtable *rt;
781 
782 	rcu_read_lock();
783 	opt = rcu_dereference(ireq->ireq_opt);
784 
785 	flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
786 			   ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
787 			   sk->sk_protocol, inet_sk_flowi_flags(sk),
788 			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
789 			   ireq->ir_loc_addr, ireq->ir_rmt_port,
790 			   htons(ireq->ir_num), sk_uid(sk));
791 	security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
792 	rt = ip_route_output_flow(net, fl4, sk);
793 	if (IS_ERR(rt))
794 		goto no_route;
795 	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
796 		goto route_err;
797 	rcu_read_unlock();
798 	return &rt->dst;
799 
800 route_err:
801 	ip_rt_put(rt);
802 no_route:
803 	rcu_read_unlock();
804 	__IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
805 	return NULL;
806 }
807 
808 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
809 					    struct sock *newsk,
810 					    const struct request_sock *req)
811 {
812 	const struct inet_request_sock *ireq = inet_rsk(req);
813 	struct net *net = read_pnet(&ireq->ireq_net);
814 	struct inet_sock *newinet = inet_sk(newsk);
815 	struct ip_options_rcu *opt;
816 	struct flowi4 *fl4;
817 	struct rtable *rt;
818 
819 	opt = rcu_dereference(ireq->ireq_opt);
820 	fl4 = &newinet->cork.fl.u.ip4;
821 
822 	flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
823 			   ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
824 			   sk->sk_protocol, inet_sk_flowi_flags(sk),
825 			   (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
826 			   ireq->ir_loc_addr, ireq->ir_rmt_port,
827 			   htons(ireq->ir_num), sk_uid(sk));
828 	security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
829 	rt = ip_route_output_flow(net, fl4, sk);
830 	if (IS_ERR(rt))
831 		goto no_route;
832 	if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
833 		goto route_err;
834 	return &rt->dst;
835 
836 route_err:
837 	ip_rt_put(rt);
838 no_route:
839 	__IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
840 	return NULL;
841 }
842 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
843 
844 /* Decide when to expire the request and when to resend SYN-ACK */
845 static void syn_ack_recalc(struct request_sock *req,
846 			   const int max_syn_ack_retries,
847 			   const u8 rskq_defer_accept,
848 			   int *expire, int *resend)
849 {
850 	if (!rskq_defer_accept) {
851 		*expire = req->num_timeout >= max_syn_ack_retries;
852 		*resend = 1;
853 		return;
854 	}
855 	*expire = req->num_timeout >= max_syn_ack_retries &&
856 		  (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
857 	/* Do not resend while waiting for data after ACK,
858 	 * start to resend on end of deferring period to give
859 	 * last chance for data or ACK to create established socket.
860 	 */
861 	*resend = !inet_rsk(req)->acked ||
862 		  req->num_timeout >= rskq_defer_accept - 1;
863 }
864 
865 static struct request_sock *
866 reqsk_alloc_noprof(const struct request_sock_ops *ops, struct sock *sk_listener,
867 		   bool attach_listener)
868 {
869 	struct request_sock *req;
870 
871 	req = kmem_cache_alloc_noprof(ops->slab, GFP_ATOMIC | __GFP_NOWARN);
872 	if (!req)
873 		return NULL;
874 	req->rsk_listener = NULL;
875 	if (attach_listener) {
876 		if (unlikely(!refcount_inc_not_zero(&sk_listener->sk_refcnt))) {
877 			kmem_cache_free(ops->slab, req);
878 			return NULL;
879 		}
880 		req->rsk_listener = sk_listener;
881 	}
882 	req->rsk_ops = ops;
883 	req_to_sk(req)->sk_prot = sk_listener->sk_prot;
884 	sk_node_init(&req_to_sk(req)->sk_node);
885 	sk_tx_queue_clear(req_to_sk(req));
886 	req->saved_syn = NULL;
887 	req->syncookie = 0;
888 	req->num_timeout = 0;
889 	req->num_retrans = 0;
890 	req->sk = NULL;
891 	refcount_set(&req->rsk_refcnt, 0);
892 
893 	return req;
894 }
895 #define reqsk_alloc(...)	alloc_hooks(reqsk_alloc_noprof(__VA_ARGS__))
896 
897 struct request_sock *inet_reqsk_alloc(const struct request_sock_ops *ops,
898 				      struct sock *sk_listener,
899 				      bool attach_listener)
900 {
901 	struct request_sock *req = reqsk_alloc(ops, sk_listener,
902 					       attach_listener);
903 
904 	if (req) {
905 		struct inet_request_sock *ireq = inet_rsk(req);
906 
907 		ireq->ireq_opt = NULL;
908 #if IS_ENABLED(CONFIG_IPV6)
909 		ireq->pktopts = NULL;
910 #endif
911 		atomic64_set(&ireq->ir_cookie, 0);
912 		ireq->ireq_state = TCP_NEW_SYN_RECV;
913 		write_pnet(&ireq->ireq_net, sock_net(sk_listener));
914 		ireq->ireq_family = sk_listener->sk_family;
915 	}
916 
917 	return req;
918 }
919 EXPORT_SYMBOL(inet_reqsk_alloc);
920 
921 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
922 					     struct sock *sk)
923 {
924 	struct sock *req_sk, *nreq_sk;
925 	struct request_sock *nreq;
926 
927 	nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
928 	if (!nreq) {
929 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
930 
931 		/* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
932 		sock_put(sk);
933 		return NULL;
934 	}
935 
936 	req_sk = req_to_sk(req);
937 	nreq_sk = req_to_sk(nreq);
938 
939 	memcpy(nreq_sk, req_sk,
940 	       offsetof(struct sock, sk_dontcopy_begin));
941 	unsafe_memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
942 		      req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end),
943 		      /* alloc is larger than struct, see above */);
944 
945 	sk_node_init(&nreq_sk->sk_node);
946 	nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
947 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
948 	nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
949 #endif
950 	nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
951 
952 	nreq->rsk_listener = sk;
953 
954 	/* We need not acquire fastopenq->lock
955 	 * because the child socket is locked in inet_csk_listen_stop().
956 	 */
957 	if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
958 		rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
959 
960 	return nreq;
961 }
962 
963 static void reqsk_queue_migrated(struct request_sock_queue *queue,
964 				 const struct request_sock *req)
965 {
966 	if (req->num_timeout == 0)
967 		atomic_inc(&queue->young);
968 	atomic_inc(&queue->qlen);
969 }
970 
971 static void reqsk_migrate_reset(struct request_sock *req)
972 {
973 	req->saved_syn = NULL;
974 #if IS_ENABLED(CONFIG_IPV6)
975 	inet_rsk(req)->ipv6_opt = NULL;
976 	inet_rsk(req)->pktopts = NULL;
977 #else
978 	inet_rsk(req)->ireq_opt = NULL;
979 #endif
980 }
981 
982 /* return true if req was found in the ehash table */
983 static bool reqsk_queue_unlink(struct request_sock *req)
984 {
985 	struct sock *sk = req_to_sk(req);
986 	bool found = false;
987 
988 	if (sk_hashed(sk)) {
989 		struct inet_hashinfo *hashinfo = tcp_get_hashinfo(sk);
990 		spinlock_t *lock;
991 
992 		lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
993 		spin_lock(lock);
994 		found = __sk_nulls_del_node_init_rcu(sk);
995 		spin_unlock(lock);
996 	}
997 
998 	return found;
999 }
1000 
1001 static bool __inet_csk_reqsk_queue_drop(struct sock *sk,
1002 					struct request_sock *req,
1003 					bool from_timer)
1004 {
1005 	bool unlinked = reqsk_queue_unlink(req);
1006 
1007 	if (!from_timer && timer_delete_sync(&req->rsk_timer))
1008 		reqsk_put(req);
1009 
1010 	if (unlinked) {
1011 		reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
1012 		reqsk_put(req);
1013 	}
1014 
1015 	return unlinked;
1016 }
1017 
1018 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
1019 {
1020 	return __inet_csk_reqsk_queue_drop(sk, req, false);
1021 }
1022 
1023 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
1024 {
1025 	inet_csk_reqsk_queue_drop(sk, req);
1026 	reqsk_put(req);
1027 }
1028 EXPORT_IPV6_MOD(inet_csk_reqsk_queue_drop_and_put);
1029 
1030 static void reqsk_timer_handler(struct timer_list *t)
1031 {
1032 	struct request_sock *req = timer_container_of(req, t, rsk_timer);
1033 	struct request_sock *nreq = NULL, *oreq = req;
1034 	struct sock *sk_listener = req->rsk_listener;
1035 	struct inet_connection_sock *icsk;
1036 	struct request_sock_queue *queue;
1037 	struct net *net;
1038 	int max_syn_ack_retries, qlen, expire = 0, resend = 0;
1039 
1040 	if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
1041 		struct sock *nsk;
1042 
1043 		nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
1044 		if (!nsk)
1045 			goto drop;
1046 
1047 		nreq = inet_reqsk_clone(req, nsk);
1048 		if (!nreq)
1049 			goto drop;
1050 
1051 		/* The new timer for the cloned req can decrease the 2
1052 		 * by calling inet_csk_reqsk_queue_drop_and_put(), so
1053 		 * hold another count to prevent use-after-free and
1054 		 * call reqsk_put() just before return.
1055 		 */
1056 		refcount_set(&nreq->rsk_refcnt, 2 + 1);
1057 		timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1058 		reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
1059 
1060 		req = nreq;
1061 		sk_listener = nsk;
1062 	}
1063 
1064 	icsk = inet_csk(sk_listener);
1065 	net = sock_net(sk_listener);
1066 	max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? :
1067 		READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1068 	/* Normally all the openreqs are young and become mature
1069 	 * (i.e. converted to established socket) for first timeout.
1070 	 * If synack was not acknowledged for 1 second, it means
1071 	 * one of the following things: synack was lost, ack was lost,
1072 	 * rtt is high or nobody planned to ack (i.e. synflood).
1073 	 * When server is a bit loaded, queue is populated with old
1074 	 * open requests, reducing effective size of queue.
1075 	 * When server is well loaded, queue size reduces to zero
1076 	 * after several minutes of work. It is not synflood,
1077 	 * it is normal operation. The solution is pruning
1078 	 * too old entries overriding normal timeout, when
1079 	 * situation becomes dangerous.
1080 	 *
1081 	 * Essentially, we reserve half of room for young
1082 	 * embrions; and abort old ones without pity, if old
1083 	 * ones are about to clog our table.
1084 	 */
1085 	queue = &icsk->icsk_accept_queue;
1086 	qlen = reqsk_queue_len(queue);
1087 	if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1088 		int young = reqsk_queue_len_young(queue) << 1;
1089 
1090 		while (max_syn_ack_retries > 2) {
1091 			if (qlen < young)
1092 				break;
1093 			max_syn_ack_retries--;
1094 			young <<= 1;
1095 		}
1096 	}
1097 
1098 	syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
1099 		       &expire, &resend);
1100 	tcp_syn_ack_timeout(req);
1101 
1102 	if (!expire &&
1103 	    (!resend ||
1104 	     !tcp_rtx_synack(sk_listener, req) ||
1105 	     inet_rsk(req)->acked)) {
1106 		if (req->num_timeout++ == 0)
1107 			atomic_dec(&queue->young);
1108 		mod_timer(&req->rsk_timer, jiffies + tcp_reqsk_timeout(req));
1109 
1110 		if (!nreq)
1111 			return;
1112 
1113 		if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
1114 			/* delete timer */
1115 			__inet_csk_reqsk_queue_drop(sk_listener, nreq, true);
1116 			goto no_ownership;
1117 		}
1118 
1119 		__NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1120 		reqsk_migrate_reset(oreq);
1121 		reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1122 		reqsk_put(oreq);
1123 
1124 		reqsk_put(nreq);
1125 		return;
1126 	}
1127 
1128 	/* Even if we can clone the req, we may need not retransmit any more
1129 	 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1130 	 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1131 	 */
1132 	if (nreq) {
1133 		__NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1134 no_ownership:
1135 		reqsk_migrate_reset(nreq);
1136 		reqsk_queue_removed(queue, nreq);
1137 		__reqsk_free(nreq);
1138 	}
1139 
1140 drop:
1141 	__inet_csk_reqsk_queue_drop(sk_listener, oreq, true);
1142 	reqsk_put(oreq);
1143 }
1144 
1145 static bool reqsk_queue_hash_req(struct request_sock *req)
1146 {
1147 	bool found_dup_sk = false;
1148 
1149 	if (!inet_ehash_insert(req_to_sk(req), NULL, &found_dup_sk))
1150 		return false;
1151 
1152 	/* The timer needs to be setup after a successful insertion. */
1153 	req->timeout = tcp_timeout_init((struct sock *)req);
1154 	timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1155 	mod_timer(&req->rsk_timer, jiffies + req->timeout);
1156 
1157 	/* before letting lookups find us, make sure all req fields
1158 	 * are committed to memory and refcnt initialized.
1159 	 */
1160 	smp_wmb();
1161 	refcount_set(&req->rsk_refcnt, 2 + 1);
1162 	return true;
1163 }
1164 
1165 bool inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req)
1166 {
1167 	if (!reqsk_queue_hash_req(req))
1168 		return false;
1169 
1170 	inet_csk_reqsk_queue_added(sk);
1171 	return true;
1172 }
1173 
1174 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1175 			   const gfp_t priority)
1176 {
1177 	struct inet_connection_sock *icsk = inet_csk(newsk);
1178 
1179 	if (!icsk->icsk_ulp_ops)
1180 		return;
1181 
1182 	icsk->icsk_ulp_ops->clone(req, newsk, priority);
1183 }
1184 
1185 /**
1186  *	inet_csk_clone_lock - clone an inet socket, and lock its clone
1187  *	@sk: the socket to clone
1188  *	@req: request_sock
1189  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1190  *
1191  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1192  */
1193 struct sock *inet_csk_clone_lock(const struct sock *sk,
1194 				 const struct request_sock *req,
1195 				 const gfp_t priority)
1196 {
1197 	struct sock *newsk = sk_clone_lock(sk, priority);
1198 	struct inet_connection_sock *newicsk;
1199 	struct inet_request_sock *ireq;
1200 	struct inet_sock *newinet;
1201 
1202 	if (!newsk)
1203 		return NULL;
1204 
1205 	newicsk = inet_csk(newsk);
1206 	newinet = inet_sk(newsk);
1207 	ireq = inet_rsk(req);
1208 
1209 	newicsk->icsk_bind_hash = NULL;
1210 	newicsk->icsk_bind2_hash = NULL;
1211 
1212 	newinet->inet_dport = ireq->ir_rmt_port;
1213 	newinet->inet_num = ireq->ir_num;
1214 	newinet->inet_sport = htons(ireq->ir_num);
1215 
1216 	newsk->sk_bound_dev_if = ireq->ir_iif;
1217 
1218 	newsk->sk_daddr = ireq->ir_rmt_addr;
1219 	newsk->sk_rcv_saddr = ireq->ir_loc_addr;
1220 	newinet->inet_saddr = ireq->ir_loc_addr;
1221 
1222 #if IS_ENABLED(CONFIG_IPV6)
1223 	newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1224 	newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1225 #endif
1226 
1227 	/* listeners have SOCK_RCU_FREE, not the children */
1228 	sock_reset_flag(newsk, SOCK_RCU_FREE);
1229 
1230 	inet_sk(newsk)->mc_list = NULL;
1231 
1232 	newsk->sk_mark = inet_rsk(req)->ir_mark;
1233 	atomic64_set(&newsk->sk_cookie,
1234 		     atomic64_read(&inet_rsk(req)->ir_cookie));
1235 
1236 	newicsk->icsk_retransmits = 0;
1237 	newicsk->icsk_backoff	  = 0;
1238 	newicsk->icsk_probes_out  = 0;
1239 	newicsk->icsk_probes_tstamp = 0;
1240 
1241 	/* Deinitialize accept_queue to trap illegal accesses. */
1242 	memset(&newicsk->icsk_accept_queue, 0,
1243 	       sizeof(newicsk->icsk_accept_queue));
1244 
1245 	inet_sk_set_state(newsk, TCP_SYN_RECV);
1246 
1247 	inet_clone_ulp(req, newsk, priority);
1248 
1249 	security_inet_csk_clone(newsk, req);
1250 
1251 	return newsk;
1252 }
1253 
1254 /*
1255  * At this point, there should be no process reference to this
1256  * socket, and thus no user references at all.  Therefore we
1257  * can assume the socket waitqueue is inactive and nobody will
1258  * try to jump onto it.
1259  */
1260 void inet_csk_destroy_sock(struct sock *sk)
1261 {
1262 	WARN_ON(sk->sk_state != TCP_CLOSE);
1263 	WARN_ON(!sock_flag(sk, SOCK_DEAD));
1264 
1265 	/* It cannot be in hash table! */
1266 	WARN_ON(!sk_unhashed(sk));
1267 
1268 	/* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1269 	WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1270 
1271 	sk->sk_prot->destroy(sk);
1272 
1273 	sk_stream_kill_queues(sk);
1274 
1275 	xfrm_sk_free_policy(sk);
1276 
1277 	tcp_orphan_count_dec();
1278 
1279 	sock_put(sk);
1280 }
1281 EXPORT_SYMBOL(inet_csk_destroy_sock);
1282 
1283 void inet_csk_prepare_for_destroy_sock(struct sock *sk)
1284 {
1285 	/* The below has to be done to allow calling inet_csk_destroy_sock */
1286 	sock_set_flag(sk, SOCK_DEAD);
1287 	tcp_orphan_count_inc();
1288 }
1289 
1290 /* This function allows to force a closure of a socket after the call to
1291  * tcp_create_openreq_child().
1292  */
1293 void inet_csk_prepare_forced_close(struct sock *sk)
1294 	__releases(&sk->sk_lock.slock)
1295 {
1296 	/* sk_clone_lock locked the socket and set refcnt to 2 */
1297 	bh_unlock_sock(sk);
1298 	sock_put(sk);
1299 	inet_csk_prepare_for_destroy_sock(sk);
1300 	inet_sk(sk)->inet_num = 0;
1301 }
1302 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1303 
1304 static int inet_ulp_can_listen(const struct sock *sk)
1305 {
1306 	const struct inet_connection_sock *icsk = inet_csk(sk);
1307 
1308 	if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone)
1309 		return -EINVAL;
1310 
1311 	return 0;
1312 }
1313 
1314 int inet_csk_listen_start(struct sock *sk)
1315 {
1316 	struct inet_connection_sock *icsk = inet_csk(sk);
1317 	struct inet_sock *inet = inet_sk(sk);
1318 	int err;
1319 
1320 	err = inet_ulp_can_listen(sk);
1321 	if (unlikely(err))
1322 		return err;
1323 
1324 	reqsk_queue_alloc(&icsk->icsk_accept_queue);
1325 
1326 	sk->sk_ack_backlog = 0;
1327 	inet_csk_delack_init(sk);
1328 
1329 	/* There is race window here: we announce ourselves listening,
1330 	 * but this transition is still not validated by get_port().
1331 	 * It is OK, because this socket enters to hash table only
1332 	 * after validation is complete.
1333 	 */
1334 	inet_sk_state_store(sk, TCP_LISTEN);
1335 	err = sk->sk_prot->get_port(sk, inet->inet_num);
1336 	if (!err) {
1337 		inet->inet_sport = htons(inet->inet_num);
1338 
1339 		sk_dst_reset(sk);
1340 		err = sk->sk_prot->hash(sk);
1341 
1342 		if (likely(!err))
1343 			return 0;
1344 	}
1345 
1346 	inet_sk_set_state(sk, TCP_CLOSE);
1347 	return err;
1348 }
1349 
1350 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1351 			      struct sock *child)
1352 {
1353 	sk->sk_prot->disconnect(child, O_NONBLOCK);
1354 
1355 	sock_orphan(child);
1356 
1357 	tcp_orphan_count_inc();
1358 
1359 	if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1360 		BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1361 		BUG_ON(sk != req->rsk_listener);
1362 
1363 		/* Paranoid, to prevent race condition if
1364 		 * an inbound pkt destined for child is
1365 		 * blocked by sock lock in tcp_v4_rcv().
1366 		 * Also to satisfy an assertion in
1367 		 * tcp_v4_destroy_sock().
1368 		 */
1369 		RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1370 	}
1371 	inet_csk_destroy_sock(child);
1372 }
1373 
1374 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1375 				      struct request_sock *req,
1376 				      struct sock *child)
1377 {
1378 	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1379 
1380 	spin_lock(&queue->rskq_lock);
1381 	if (unlikely(sk->sk_state != TCP_LISTEN)) {
1382 		inet_child_forget(sk, req, child);
1383 		child = NULL;
1384 	} else {
1385 		req->sk = child;
1386 		req->dl_next = NULL;
1387 		if (queue->rskq_accept_head == NULL)
1388 			WRITE_ONCE(queue->rskq_accept_head, req);
1389 		else
1390 			queue->rskq_accept_tail->dl_next = req;
1391 		queue->rskq_accept_tail = req;
1392 		sk_acceptq_added(sk);
1393 	}
1394 	spin_unlock(&queue->rskq_lock);
1395 	return child;
1396 }
1397 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1398 
1399 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1400 					 struct request_sock *req, bool own_req)
1401 {
1402 	if (own_req) {
1403 		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1404 		reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1405 
1406 		if (sk != req->rsk_listener) {
1407 			/* another listening sk has been selected,
1408 			 * migrate the req to it.
1409 			 */
1410 			struct request_sock *nreq;
1411 
1412 			/* hold a refcnt for the nreq->rsk_listener
1413 			 * which is assigned in inet_reqsk_clone()
1414 			 */
1415 			sock_hold(sk);
1416 			nreq = inet_reqsk_clone(req, sk);
1417 			if (!nreq) {
1418 				inet_child_forget(sk, req, child);
1419 				goto child_put;
1420 			}
1421 
1422 			refcount_set(&nreq->rsk_refcnt, 1);
1423 			if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1424 				__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1425 				reqsk_migrate_reset(req);
1426 				reqsk_put(req);
1427 				return child;
1428 			}
1429 
1430 			__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1431 			reqsk_migrate_reset(nreq);
1432 			__reqsk_free(nreq);
1433 		} else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1434 			return child;
1435 		}
1436 	}
1437 	/* Too bad, another child took ownership of the request, undo. */
1438 child_put:
1439 	bh_unlock_sock(child);
1440 	sock_put(child);
1441 	return NULL;
1442 }
1443 
1444 /*
1445  *	This routine closes sockets which have been at least partially
1446  *	opened, but not yet accepted.
1447  */
1448 void inet_csk_listen_stop(struct sock *sk)
1449 {
1450 	struct inet_connection_sock *icsk = inet_csk(sk);
1451 	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1452 	struct request_sock *next, *req;
1453 
1454 	/* Following specs, it would be better either to send FIN
1455 	 * (and enter FIN-WAIT-1, it is normal close)
1456 	 * or to send active reset (abort).
1457 	 * Certainly, it is pretty dangerous while synflood, but it is
1458 	 * bad justification for our negligence 8)
1459 	 * To be honest, we are not able to make either
1460 	 * of the variants now.			--ANK
1461 	 */
1462 	while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1463 		struct sock *child = req->sk, *nsk;
1464 		struct request_sock *nreq;
1465 
1466 		local_bh_disable();
1467 		bh_lock_sock(child);
1468 		WARN_ON(sock_owned_by_user(child));
1469 		sock_hold(child);
1470 
1471 		nsk = reuseport_migrate_sock(sk, child, NULL);
1472 		if (nsk) {
1473 			nreq = inet_reqsk_clone(req, nsk);
1474 			if (nreq) {
1475 				refcount_set(&nreq->rsk_refcnt, 1);
1476 
1477 				if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1478 					__NET_INC_STATS(sock_net(nsk),
1479 							LINUX_MIB_TCPMIGRATEREQSUCCESS);
1480 					reqsk_migrate_reset(req);
1481 				} else {
1482 					__NET_INC_STATS(sock_net(nsk),
1483 							LINUX_MIB_TCPMIGRATEREQFAILURE);
1484 					reqsk_migrate_reset(nreq);
1485 					__reqsk_free(nreq);
1486 				}
1487 
1488 				/* inet_csk_reqsk_queue_add() has already
1489 				 * called inet_child_forget() on failure case.
1490 				 */
1491 				goto skip_child_forget;
1492 			}
1493 		}
1494 
1495 		inet_child_forget(sk, req, child);
1496 skip_child_forget:
1497 		reqsk_put(req);
1498 		bh_unlock_sock(child);
1499 		local_bh_enable();
1500 		sock_put(child);
1501 
1502 		cond_resched();
1503 	}
1504 	if (queue->fastopenq.rskq_rst_head) {
1505 		/* Free all the reqs queued in rskq_rst_head. */
1506 		spin_lock_bh(&queue->fastopenq.lock);
1507 		req = queue->fastopenq.rskq_rst_head;
1508 		queue->fastopenq.rskq_rst_head = NULL;
1509 		spin_unlock_bh(&queue->fastopenq.lock);
1510 		while (req != NULL) {
1511 			next = req->dl_next;
1512 			reqsk_put(req);
1513 			req = next;
1514 		}
1515 	}
1516 	WARN_ON_ONCE(sk->sk_ack_backlog);
1517 }
1518 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1519 
1520 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1521 {
1522 	const struct inet_sock *inet = inet_sk(sk);
1523 	struct flowi4 *fl4;
1524 	struct rtable *rt;
1525 
1526 	rcu_read_lock();
1527 	fl4 = &fl->u.ip4;
1528 	inet_sk_init_flowi4(inet, fl4);
1529 	rt = ip_route_output_flow(sock_net(sk), fl4, sk);
1530 	if (IS_ERR(rt))
1531 		rt = NULL;
1532 	if (rt)
1533 		sk_setup_caps(sk, &rt->dst);
1534 	rcu_read_unlock();
1535 
1536 	return &rt->dst;
1537 }
1538 
1539 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1540 {
1541 	struct dst_entry *dst = __sk_dst_check(sk, 0);
1542 	struct inet_sock *inet = inet_sk(sk);
1543 
1544 	if (!dst) {
1545 		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1546 		if (!dst)
1547 			goto out;
1548 	}
1549 	dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1550 
1551 	dst = __sk_dst_check(sk, 0);
1552 	if (!dst)
1553 		dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1554 out:
1555 	return dst;
1556 }
1557