Loading...
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 */
34static 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 */
78static 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
91bool 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}
109EXPORT_SYMBOL(inet_rcv_saddr_equal);
110
111bool 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
120void 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}
131EXPORT_SYMBOL(inet_get_local_port_range);
132
133static 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
146static 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
176static 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
187static 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 */
214static 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 */
263static 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 */
301static struct inet_bind_hashbucket *
302inet_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);
316ports_exhausted:
317 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
318other_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
341other_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;
369next_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;
390success:
391 *port_ret = port;
392 *tb_ret = tb;
393 *tb2_ret = tb2;
394 *head2_ret = head2;
395 return head;
396}
397
398static 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
431void 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 */
486int 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
560success:
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
569fail_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}
582EXPORT_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 */
588static 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 */
636struct 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
685out:
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;
710out_err:
711 newsk = NULL;
712 req = NULL;
713 *err = error;
714 goto out;
715}
716EXPORT_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 */
723void 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}
735EXPORT_SYMBOL(inet_csk_init_xmit_timers);
736
737void 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}
747EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
748
749void inet_csk_delete_keepalive_timer(struct sock *sk)
750{
751 sk_stop_timer(sk, &sk->sk_timer);
752}
753EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
754
755void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
756{
757 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
758}
759EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
760
761struct 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
788route_err:
789 ip_rt_put(rt);
790no_route:
791 rcu_read_unlock();
792 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
793 return NULL;
794}
795EXPORT_SYMBOL_GPL(inet_csk_route_req);
796
797struct 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
825route_err:
826 ip_rt_put(rt);
827no_route:
828 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
829 return NULL;
830}
831EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
832
833/* Decide when to expire the request and when to resend SYN-ACK */
834static 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
854int 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}
862EXPORT_SYMBOL(inet_rtx_syn_ack);
863
864static 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
905static 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
913static 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 */
925static 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
943bool 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}
953EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
954
955void 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}
960EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
961
962static 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);
1064no_ownership:
1065 reqsk_migrate_reset(nreq);
1066 reqsk_queue_removed(queue, nreq);
1067 __reqsk_free(nreq);
1068 }
1069
1070drop:
1071 inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
1072}
1073
1074static 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
1088void 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}
1094EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1095
1096static 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 */
1116struct 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}
1156EXPORT_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 */
1164void 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}
1187EXPORT_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 */
1192void 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}
1201EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1202
1203static 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
1213int 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 /* There is race window here: we announce ourselves listening,
1229 * but this transition is still not validated by get_port().
1230 * It is OK, because this socket enters to hash table only
1231 * after validation is complete.
1232 */
1233 inet_sk_state_store(sk, TCP_LISTEN);
1234 err = sk->sk_prot->get_port(sk, inet->inet_num);
1235 if (!err) {
1236 inet->inet_sport = htons(inet->inet_num);
1237
1238 sk_dst_reset(sk);
1239 err = sk->sk_prot->hash(sk);
1240
1241 if (likely(!err))
1242 return 0;
1243 }
1244
1245 inet_sk_set_state(sk, TCP_CLOSE);
1246 return err;
1247}
1248EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1249
1250static void inet_child_forget(struct sock *sk, struct request_sock *req,
1251 struct sock *child)
1252{
1253 sk->sk_prot->disconnect(child, O_NONBLOCK);
1254
1255 sock_orphan(child);
1256
1257 this_cpu_inc(*sk->sk_prot->orphan_count);
1258
1259 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1260 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1261 BUG_ON(sk != req->rsk_listener);
1262
1263 /* Paranoid, to prevent race condition if
1264 * an inbound pkt destined for child is
1265 * blocked by sock lock in tcp_v4_rcv().
1266 * Also to satisfy an assertion in
1267 * tcp_v4_destroy_sock().
1268 */
1269 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1270 }
1271 inet_csk_destroy_sock(child);
1272}
1273
1274struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1275 struct request_sock *req,
1276 struct sock *child)
1277{
1278 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1279
1280 spin_lock(&queue->rskq_lock);
1281 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1282 inet_child_forget(sk, req, child);
1283 child = NULL;
1284 } else {
1285 req->sk = child;
1286 req->dl_next = NULL;
1287 if (queue->rskq_accept_head == NULL)
1288 WRITE_ONCE(queue->rskq_accept_head, req);
1289 else
1290 queue->rskq_accept_tail->dl_next = req;
1291 queue->rskq_accept_tail = req;
1292 sk_acceptq_added(sk);
1293 }
1294 spin_unlock(&queue->rskq_lock);
1295 return child;
1296}
1297EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1298
1299struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1300 struct request_sock *req, bool own_req)
1301{
1302 if (own_req) {
1303 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1304 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1305
1306 if (sk != req->rsk_listener) {
1307 /* another listening sk has been selected,
1308 * migrate the req to it.
1309 */
1310 struct request_sock *nreq;
1311
1312 /* hold a refcnt for the nreq->rsk_listener
1313 * which is assigned in inet_reqsk_clone()
1314 */
1315 sock_hold(sk);
1316 nreq = inet_reqsk_clone(req, sk);
1317 if (!nreq) {
1318 inet_child_forget(sk, req, child);
1319 goto child_put;
1320 }
1321
1322 refcount_set(&nreq->rsk_refcnt, 1);
1323 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1324 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1325 reqsk_migrate_reset(req);
1326 reqsk_put(req);
1327 return child;
1328 }
1329
1330 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1331 reqsk_migrate_reset(nreq);
1332 __reqsk_free(nreq);
1333 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1334 return child;
1335 }
1336 }
1337 /* Too bad, another child took ownership of the request, undo. */
1338child_put:
1339 bh_unlock_sock(child);
1340 sock_put(child);
1341 return NULL;
1342}
1343EXPORT_SYMBOL(inet_csk_complete_hashdance);
1344
1345/*
1346 * This routine closes sockets which have been at least partially
1347 * opened, but not yet accepted.
1348 */
1349void inet_csk_listen_stop(struct sock *sk)
1350{
1351 struct inet_connection_sock *icsk = inet_csk(sk);
1352 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1353 struct request_sock *next, *req;
1354
1355 /* Following specs, it would be better either to send FIN
1356 * (and enter FIN-WAIT-1, it is normal close)
1357 * or to send active reset (abort).
1358 * Certainly, it is pretty dangerous while synflood, but it is
1359 * bad justification for our negligence 8)
1360 * To be honest, we are not able to make either
1361 * of the variants now. --ANK
1362 */
1363 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1364 struct sock *child = req->sk, *nsk;
1365 struct request_sock *nreq;
1366
1367 local_bh_disable();
1368 bh_lock_sock(child);
1369 WARN_ON(sock_owned_by_user(child));
1370 sock_hold(child);
1371
1372 nsk = reuseport_migrate_sock(sk, child, NULL);
1373 if (nsk) {
1374 nreq = inet_reqsk_clone(req, nsk);
1375 if (nreq) {
1376 refcount_set(&nreq->rsk_refcnt, 1);
1377
1378 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1379 __NET_INC_STATS(sock_net(nsk),
1380 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1381 reqsk_migrate_reset(req);
1382 } else {
1383 __NET_INC_STATS(sock_net(nsk),
1384 LINUX_MIB_TCPMIGRATEREQFAILURE);
1385 reqsk_migrate_reset(nreq);
1386 __reqsk_free(nreq);
1387 }
1388
1389 /* inet_csk_reqsk_queue_add() has already
1390 * called inet_child_forget() on failure case.
1391 */
1392 goto skip_child_forget;
1393 }
1394 }
1395
1396 inet_child_forget(sk, req, child);
1397skip_child_forget:
1398 reqsk_put(req);
1399 bh_unlock_sock(child);
1400 local_bh_enable();
1401 sock_put(child);
1402
1403 cond_resched();
1404 }
1405 if (queue->fastopenq.rskq_rst_head) {
1406 /* Free all the reqs queued in rskq_rst_head. */
1407 spin_lock_bh(&queue->fastopenq.lock);
1408 req = queue->fastopenq.rskq_rst_head;
1409 queue->fastopenq.rskq_rst_head = NULL;
1410 spin_unlock_bh(&queue->fastopenq.lock);
1411 while (req != NULL) {
1412 next = req->dl_next;
1413 reqsk_put(req);
1414 req = next;
1415 }
1416 }
1417 WARN_ON_ONCE(sk->sk_ack_backlog);
1418}
1419EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1420
1421void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1422{
1423 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1424 const struct inet_sock *inet = inet_sk(sk);
1425
1426 sin->sin_family = AF_INET;
1427 sin->sin_addr.s_addr = inet->inet_daddr;
1428 sin->sin_port = inet->inet_dport;
1429}
1430EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1431
1432static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1433{
1434 const struct inet_sock *inet = inet_sk(sk);
1435 const struct ip_options_rcu *inet_opt;
1436 __be32 daddr = inet->inet_daddr;
1437 struct flowi4 *fl4;
1438 struct rtable *rt;
1439
1440 rcu_read_lock();
1441 inet_opt = rcu_dereference(inet->inet_opt);
1442 if (inet_opt && inet_opt->opt.srr)
1443 daddr = inet_opt->opt.faddr;
1444 fl4 = &fl->u.ip4;
1445 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1446 inet->inet_saddr, inet->inet_dport,
1447 inet->inet_sport, sk->sk_protocol,
1448 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1449 if (IS_ERR(rt))
1450 rt = NULL;
1451 if (rt)
1452 sk_setup_caps(sk, &rt->dst);
1453 rcu_read_unlock();
1454
1455 return &rt->dst;
1456}
1457
1458struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1459{
1460 struct dst_entry *dst = __sk_dst_check(sk, 0);
1461 struct inet_sock *inet = inet_sk(sk);
1462
1463 if (!dst) {
1464 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1465 if (!dst)
1466 goto out;
1467 }
1468 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1469
1470 dst = __sk_dst_check(sk, 0);
1471 if (!dst)
1472 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1473out:
1474 return dst;
1475}
1476EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Support for INET connection oriented protocols.
7 *
8 * Authors: See the TCP sources
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or(at your option) any later version.
14 */
15
16#include <linux/module.h>
17#include <linux/jhash.h>
18
19#include <net/inet_connection_sock.h>
20#include <net/inet_hashtables.h>
21#include <net/inet_timewait_sock.h>
22#include <net/ip.h>
23#include <net/route.h>
24#include <net/tcp_states.h>
25#include <net/xfrm.h>
26#include <net/tcp.h>
27#include <net/sock_reuseport.h>
28
29#ifdef INET_CSK_DEBUG
30const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
31EXPORT_SYMBOL(inet_csk_timer_bug_msg);
32#endif
33
34void inet_get_local_port_range(struct net *net, int *low, int *high)
35{
36 unsigned int seq;
37
38 do {
39 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
40
41 *low = net->ipv4.ip_local_ports.range[0];
42 *high = net->ipv4.ip_local_ports.range[1];
43 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
44}
45EXPORT_SYMBOL(inet_get_local_port_range);
46
47int inet_csk_bind_conflict(const struct sock *sk,
48 const struct inet_bind_bucket *tb, bool relax)
49{
50 struct sock *sk2;
51 int reuse = sk->sk_reuse;
52 int reuseport = sk->sk_reuseport;
53 kuid_t uid = sock_i_uid((struct sock *)sk);
54
55 /*
56 * Unlike other sk lookup places we do not check
57 * for sk_net here, since _all_ the socks listed
58 * in tb->owners list belong to the same net - the
59 * one this bucket belongs to.
60 */
61
62 sk_for_each_bound(sk2, &tb->owners) {
63 if (sk != sk2 &&
64 !inet_v6_ipv6only(sk2) &&
65 (!sk->sk_bound_dev_if ||
66 !sk2->sk_bound_dev_if ||
67 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
68 if ((!reuse || !sk2->sk_reuse ||
69 sk2->sk_state == TCP_LISTEN) &&
70 (!reuseport || !sk2->sk_reuseport ||
71 rcu_access_pointer(sk->sk_reuseport_cb) ||
72 (sk2->sk_state != TCP_TIME_WAIT &&
73 !uid_eq(uid, sock_i_uid(sk2))))) {
74
75 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
76 sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
77 break;
78 }
79 if (!relax && reuse && sk2->sk_reuse &&
80 sk2->sk_state != TCP_LISTEN) {
81
82 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
83 sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
84 break;
85 }
86 }
87 }
88 return sk2 != NULL;
89}
90EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
91
92/* Obtain a reference to a local port for the given sock,
93 * if snum is zero it means select any available local port.
94 * We try to allocate an odd port (and leave even ports for connect())
95 */
96int inet_csk_get_port(struct sock *sk, unsigned short snum)
97{
98 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
99 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
100 int ret = 1, attempts = 5, port = snum;
101 int smallest_size = -1, smallest_port;
102 struct inet_bind_hashbucket *head;
103 struct net *net = sock_net(sk);
104 int i, low, high, attempt_half;
105 struct inet_bind_bucket *tb;
106 kuid_t uid = sock_i_uid(sk);
107 u32 remaining, offset;
108
109 if (port) {
110have_port:
111 head = &hinfo->bhash[inet_bhashfn(net, port,
112 hinfo->bhash_size)];
113 spin_lock_bh(&head->lock);
114 inet_bind_bucket_for_each(tb, &head->chain)
115 if (net_eq(ib_net(tb), net) && tb->port == port)
116 goto tb_found;
117
118 goto tb_not_found;
119 }
120again:
121 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
122other_half_scan:
123 inet_get_local_port_range(net, &low, &high);
124 high++; /* [32768, 60999] -> [32768, 61000[ */
125 if (high - low < 4)
126 attempt_half = 0;
127 if (attempt_half) {
128 int half = low + (((high - low) >> 2) << 1);
129
130 if (attempt_half == 1)
131 high = half;
132 else
133 low = half;
134 }
135 remaining = high - low;
136 if (likely(remaining > 1))
137 remaining &= ~1U;
138
139 offset = prandom_u32() % remaining;
140 /* __inet_hash_connect() favors ports having @low parity
141 * We do the opposite to not pollute connect() users.
142 */
143 offset |= 1U;
144 smallest_size = -1;
145 smallest_port = low; /* avoid compiler warning */
146
147other_parity_scan:
148 port = low + offset;
149 for (i = 0; i < remaining; i += 2, port += 2) {
150 if (unlikely(port >= high))
151 port -= remaining;
152 if (inet_is_local_reserved_port(net, port))
153 continue;
154 head = &hinfo->bhash[inet_bhashfn(net, port,
155 hinfo->bhash_size)];
156 spin_lock_bh(&head->lock);
157 inet_bind_bucket_for_each(tb, &head->chain)
158 if (net_eq(ib_net(tb), net) && tb->port == port) {
159 if (((tb->fastreuse > 0 && reuse) ||
160 (tb->fastreuseport > 0 &&
161 sk->sk_reuseport &&
162 !rcu_access_pointer(sk->sk_reuseport_cb) &&
163 uid_eq(tb->fastuid, uid))) &&
164 (tb->num_owners < smallest_size || smallest_size == -1)) {
165 smallest_size = tb->num_owners;
166 smallest_port = port;
167 }
168 if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false))
169 goto tb_found;
170 goto next_port;
171 }
172 goto tb_not_found;
173next_port:
174 spin_unlock_bh(&head->lock);
175 cond_resched();
176 }
177
178 if (smallest_size != -1) {
179 port = smallest_port;
180 goto have_port;
181 }
182 offset--;
183 if (!(offset & 1))
184 goto other_parity_scan;
185
186 if (attempt_half == 1) {
187 /* OK we now try the upper half of the range */
188 attempt_half = 2;
189 goto other_half_scan;
190 }
191 return ret;
192
193tb_not_found:
194 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
195 net, head, port);
196 if (!tb)
197 goto fail_unlock;
198tb_found:
199 if (!hlist_empty(&tb->owners)) {
200 if (sk->sk_reuse == SK_FORCE_REUSE)
201 goto success;
202
203 if (((tb->fastreuse > 0 && reuse) ||
204 (tb->fastreuseport > 0 &&
205 !rcu_access_pointer(sk->sk_reuseport_cb) &&
206 sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
207 smallest_size == -1)
208 goto success;
209 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
210 if ((reuse ||
211 (tb->fastreuseport > 0 &&
212 sk->sk_reuseport &&
213 !rcu_access_pointer(sk->sk_reuseport_cb) &&
214 uid_eq(tb->fastuid, uid))) &&
215 smallest_size != -1 && --attempts >= 0) {
216 spin_unlock_bh(&head->lock);
217 goto again;
218 }
219 goto fail_unlock;
220 }
221 if (!reuse)
222 tb->fastreuse = 0;
223 if (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid))
224 tb->fastreuseport = 0;
225 } else {
226 tb->fastreuse = reuse;
227 if (sk->sk_reuseport) {
228 tb->fastreuseport = 1;
229 tb->fastuid = uid;
230 } else {
231 tb->fastreuseport = 0;
232 }
233 }
234success:
235 if (!inet_csk(sk)->icsk_bind_hash)
236 inet_bind_hash(sk, tb, port);
237 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
238 ret = 0;
239
240fail_unlock:
241 spin_unlock_bh(&head->lock);
242 return ret;
243}
244EXPORT_SYMBOL_GPL(inet_csk_get_port);
245
246/*
247 * Wait for an incoming connection, avoid race conditions. This must be called
248 * with the socket locked.
249 */
250static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
251{
252 struct inet_connection_sock *icsk = inet_csk(sk);
253 DEFINE_WAIT(wait);
254 int err;
255
256 /*
257 * True wake-one mechanism for incoming connections: only
258 * one process gets woken up, not the 'whole herd'.
259 * Since we do not 'race & poll' for established sockets
260 * anymore, the common case will execute the loop only once.
261 *
262 * Subtle issue: "add_wait_queue_exclusive()" will be added
263 * after any current non-exclusive waiters, and we know that
264 * it will always _stay_ after any new non-exclusive waiters
265 * because all non-exclusive waiters are added at the
266 * beginning of the wait-queue. As such, it's ok to "drop"
267 * our exclusiveness temporarily when we get woken up without
268 * having to remove and re-insert us on the wait queue.
269 */
270 for (;;) {
271 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
272 TASK_INTERRUPTIBLE);
273 release_sock(sk);
274 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
275 timeo = schedule_timeout(timeo);
276 sched_annotate_sleep();
277 lock_sock(sk);
278 err = 0;
279 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
280 break;
281 err = -EINVAL;
282 if (sk->sk_state != TCP_LISTEN)
283 break;
284 err = sock_intr_errno(timeo);
285 if (signal_pending(current))
286 break;
287 err = -EAGAIN;
288 if (!timeo)
289 break;
290 }
291 finish_wait(sk_sleep(sk), &wait);
292 return err;
293}
294
295/*
296 * This will accept the next outstanding connection.
297 */
298struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
299{
300 struct inet_connection_sock *icsk = inet_csk(sk);
301 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
302 struct request_sock *req;
303 struct sock *newsk;
304 int error;
305
306 lock_sock(sk);
307
308 /* We need to make sure that this socket is listening,
309 * and that it has something pending.
310 */
311 error = -EINVAL;
312 if (sk->sk_state != TCP_LISTEN)
313 goto out_err;
314
315 /* Find already established connection */
316 if (reqsk_queue_empty(queue)) {
317 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
318
319 /* If this is a non blocking socket don't sleep */
320 error = -EAGAIN;
321 if (!timeo)
322 goto out_err;
323
324 error = inet_csk_wait_for_connect(sk, timeo);
325 if (error)
326 goto out_err;
327 }
328 req = reqsk_queue_remove(queue, sk);
329 newsk = req->sk;
330
331 if (sk->sk_protocol == IPPROTO_TCP &&
332 tcp_rsk(req)->tfo_listener) {
333 spin_lock_bh(&queue->fastopenq.lock);
334 if (tcp_rsk(req)->tfo_listener) {
335 /* We are still waiting for the final ACK from 3WHS
336 * so can't free req now. Instead, we set req->sk to
337 * NULL to signify that the child socket is taken
338 * so reqsk_fastopen_remove() will free the req
339 * when 3WHS finishes (or is aborted).
340 */
341 req->sk = NULL;
342 req = NULL;
343 }
344 spin_unlock_bh(&queue->fastopenq.lock);
345 }
346out:
347 release_sock(sk);
348 if (req)
349 reqsk_put(req);
350 return newsk;
351out_err:
352 newsk = NULL;
353 req = NULL;
354 *err = error;
355 goto out;
356}
357EXPORT_SYMBOL(inet_csk_accept);
358
359/*
360 * Using different timers for retransmit, delayed acks and probes
361 * We may wish use just one timer maintaining a list of expire jiffies
362 * to optimize.
363 */
364void inet_csk_init_xmit_timers(struct sock *sk,
365 void (*retransmit_handler)(unsigned long),
366 void (*delack_handler)(unsigned long),
367 void (*keepalive_handler)(unsigned long))
368{
369 struct inet_connection_sock *icsk = inet_csk(sk);
370
371 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
372 (unsigned long)sk);
373 setup_timer(&icsk->icsk_delack_timer, delack_handler,
374 (unsigned long)sk);
375 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
376 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
377}
378EXPORT_SYMBOL(inet_csk_init_xmit_timers);
379
380void inet_csk_clear_xmit_timers(struct sock *sk)
381{
382 struct inet_connection_sock *icsk = inet_csk(sk);
383
384 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
385
386 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
387 sk_stop_timer(sk, &icsk->icsk_delack_timer);
388 sk_stop_timer(sk, &sk->sk_timer);
389}
390EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
391
392void inet_csk_delete_keepalive_timer(struct sock *sk)
393{
394 sk_stop_timer(sk, &sk->sk_timer);
395}
396EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
397
398void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
399{
400 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
401}
402EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
403
404struct dst_entry *inet_csk_route_req(const struct sock *sk,
405 struct flowi4 *fl4,
406 const struct request_sock *req)
407{
408 const struct inet_request_sock *ireq = inet_rsk(req);
409 struct net *net = read_pnet(&ireq->ireq_net);
410 struct ip_options_rcu *opt = ireq->opt;
411 struct rtable *rt;
412
413 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
414 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
415 sk->sk_protocol, inet_sk_flowi_flags(sk),
416 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
417 ireq->ir_loc_addr, ireq->ir_rmt_port,
418 htons(ireq->ir_num));
419 security_req_classify_flow(req, flowi4_to_flowi(fl4));
420 rt = ip_route_output_flow(net, fl4, sk);
421 if (IS_ERR(rt))
422 goto no_route;
423 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
424 goto route_err;
425 return &rt->dst;
426
427route_err:
428 ip_rt_put(rt);
429no_route:
430 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
431 return NULL;
432}
433EXPORT_SYMBOL_GPL(inet_csk_route_req);
434
435struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
436 struct sock *newsk,
437 const struct request_sock *req)
438{
439 const struct inet_request_sock *ireq = inet_rsk(req);
440 struct net *net = read_pnet(&ireq->ireq_net);
441 struct inet_sock *newinet = inet_sk(newsk);
442 struct ip_options_rcu *opt;
443 struct flowi4 *fl4;
444 struct rtable *rt;
445
446 fl4 = &newinet->cork.fl.u.ip4;
447
448 rcu_read_lock();
449 opt = rcu_dereference(newinet->inet_opt);
450 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
451 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
452 sk->sk_protocol, inet_sk_flowi_flags(sk),
453 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
454 ireq->ir_loc_addr, ireq->ir_rmt_port,
455 htons(ireq->ir_num));
456 security_req_classify_flow(req, flowi4_to_flowi(fl4));
457 rt = ip_route_output_flow(net, fl4, sk);
458 if (IS_ERR(rt))
459 goto no_route;
460 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
461 goto route_err;
462 rcu_read_unlock();
463 return &rt->dst;
464
465route_err:
466 ip_rt_put(rt);
467no_route:
468 rcu_read_unlock();
469 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
470 return NULL;
471}
472EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
473
474#if IS_ENABLED(CONFIG_IPV6)
475#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
476#else
477#define AF_INET_FAMILY(fam) true
478#endif
479
480/* Decide when to expire the request and when to resend SYN-ACK */
481static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
482 const int max_retries,
483 const u8 rskq_defer_accept,
484 int *expire, int *resend)
485{
486 if (!rskq_defer_accept) {
487 *expire = req->num_timeout >= thresh;
488 *resend = 1;
489 return;
490 }
491 *expire = req->num_timeout >= thresh &&
492 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
493 /*
494 * Do not resend while waiting for data after ACK,
495 * start to resend on end of deferring period to give
496 * last chance for data or ACK to create established socket.
497 */
498 *resend = !inet_rsk(req)->acked ||
499 req->num_timeout >= rskq_defer_accept - 1;
500}
501
502int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
503{
504 int err = req->rsk_ops->rtx_syn_ack(parent, req);
505
506 if (!err)
507 req->num_retrans++;
508 return err;
509}
510EXPORT_SYMBOL(inet_rtx_syn_ack);
511
512/* return true if req was found in the ehash table */
513static bool reqsk_queue_unlink(struct request_sock_queue *queue,
514 struct request_sock *req)
515{
516 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
517 bool found = false;
518
519 if (sk_hashed(req_to_sk(req))) {
520 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
521
522 spin_lock(lock);
523 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
524 spin_unlock(lock);
525 }
526 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
527 reqsk_put(req);
528 return found;
529}
530
531void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
532{
533 if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
534 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
535 reqsk_put(req);
536 }
537}
538EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
539
540void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
541{
542 inet_csk_reqsk_queue_drop(sk, req);
543 reqsk_put(req);
544}
545EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
546
547static void reqsk_timer_handler(unsigned long data)
548{
549 struct request_sock *req = (struct request_sock *)data;
550 struct sock *sk_listener = req->rsk_listener;
551 struct net *net = sock_net(sk_listener);
552 struct inet_connection_sock *icsk = inet_csk(sk_listener);
553 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
554 int qlen, expire = 0, resend = 0;
555 int max_retries, thresh;
556 u8 defer_accept;
557
558 if (sk_state_load(sk_listener) != TCP_LISTEN)
559 goto drop;
560
561 max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
562 thresh = max_retries;
563 /* Normally all the openreqs are young and become mature
564 * (i.e. converted to established socket) for first timeout.
565 * If synack was not acknowledged for 1 second, it means
566 * one of the following things: synack was lost, ack was lost,
567 * rtt is high or nobody planned to ack (i.e. synflood).
568 * When server is a bit loaded, queue is populated with old
569 * open requests, reducing effective size of queue.
570 * When server is well loaded, queue size reduces to zero
571 * after several minutes of work. It is not synflood,
572 * it is normal operation. The solution is pruning
573 * too old entries overriding normal timeout, when
574 * situation becomes dangerous.
575 *
576 * Essentially, we reserve half of room for young
577 * embrions; and abort old ones without pity, if old
578 * ones are about to clog our table.
579 */
580 qlen = reqsk_queue_len(queue);
581 if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
582 int young = reqsk_queue_len_young(queue) << 1;
583
584 while (thresh > 2) {
585 if (qlen < young)
586 break;
587 thresh--;
588 young <<= 1;
589 }
590 }
591 defer_accept = READ_ONCE(queue->rskq_defer_accept);
592 if (defer_accept)
593 max_retries = defer_accept;
594 syn_ack_recalc(req, thresh, max_retries, defer_accept,
595 &expire, &resend);
596 req->rsk_ops->syn_ack_timeout(req);
597 if (!expire &&
598 (!resend ||
599 !inet_rtx_syn_ack(sk_listener, req) ||
600 inet_rsk(req)->acked)) {
601 unsigned long timeo;
602
603 if (req->num_timeout++ == 0)
604 atomic_dec(&queue->young);
605 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
606 mod_timer_pinned(&req->rsk_timer, jiffies + timeo);
607 return;
608 }
609drop:
610 inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
611}
612
613static void reqsk_queue_hash_req(struct request_sock *req,
614 unsigned long timeout)
615{
616 req->num_retrans = 0;
617 req->num_timeout = 0;
618 req->sk = NULL;
619
620 setup_timer(&req->rsk_timer, reqsk_timer_handler, (unsigned long)req);
621 mod_timer_pinned(&req->rsk_timer, jiffies + timeout);
622
623 inet_ehash_insert(req_to_sk(req), NULL);
624 /* before letting lookups find us, make sure all req fields
625 * are committed to memory and refcnt initialized.
626 */
627 smp_wmb();
628 atomic_set(&req->rsk_refcnt, 2 + 1);
629}
630
631void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
632 unsigned long timeout)
633{
634 reqsk_queue_hash_req(req, timeout);
635 inet_csk_reqsk_queue_added(sk);
636}
637EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
638
639/**
640 * inet_csk_clone_lock - clone an inet socket, and lock its clone
641 * @sk: the socket to clone
642 * @req: request_sock
643 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
644 *
645 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
646 */
647struct sock *inet_csk_clone_lock(const struct sock *sk,
648 const struct request_sock *req,
649 const gfp_t priority)
650{
651 struct sock *newsk = sk_clone_lock(sk, priority);
652
653 if (newsk) {
654 struct inet_connection_sock *newicsk = inet_csk(newsk);
655
656 newsk->sk_state = TCP_SYN_RECV;
657 newicsk->icsk_bind_hash = NULL;
658
659 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
660 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
661 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
662 newsk->sk_write_space = sk_stream_write_space;
663
664 newsk->sk_mark = inet_rsk(req)->ir_mark;
665 atomic64_set(&newsk->sk_cookie,
666 atomic64_read(&inet_rsk(req)->ir_cookie));
667
668 newicsk->icsk_retransmits = 0;
669 newicsk->icsk_backoff = 0;
670 newicsk->icsk_probes_out = 0;
671
672 /* Deinitialize accept_queue to trap illegal accesses. */
673 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
674
675 security_inet_csk_clone(newsk, req);
676 }
677 return newsk;
678}
679EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
680
681/*
682 * At this point, there should be no process reference to this
683 * socket, and thus no user references at all. Therefore we
684 * can assume the socket waitqueue is inactive and nobody will
685 * try to jump onto it.
686 */
687void inet_csk_destroy_sock(struct sock *sk)
688{
689 WARN_ON(sk->sk_state != TCP_CLOSE);
690 WARN_ON(!sock_flag(sk, SOCK_DEAD));
691
692 /* It cannot be in hash table! */
693 WARN_ON(!sk_unhashed(sk));
694
695 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
696 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
697
698 sk->sk_prot->destroy(sk);
699
700 sk_stream_kill_queues(sk);
701
702 xfrm_sk_free_policy(sk);
703
704 sk_refcnt_debug_release(sk);
705
706 percpu_counter_dec(sk->sk_prot->orphan_count);
707 sock_put(sk);
708}
709EXPORT_SYMBOL(inet_csk_destroy_sock);
710
711/* This function allows to force a closure of a socket after the call to
712 * tcp/dccp_create_openreq_child().
713 */
714void inet_csk_prepare_forced_close(struct sock *sk)
715 __releases(&sk->sk_lock.slock)
716{
717 /* sk_clone_lock locked the socket and set refcnt to 2 */
718 bh_unlock_sock(sk);
719 sock_put(sk);
720
721 /* The below has to be done to allow calling inet_csk_destroy_sock */
722 sock_set_flag(sk, SOCK_DEAD);
723 percpu_counter_inc(sk->sk_prot->orphan_count);
724 inet_sk(sk)->inet_num = 0;
725}
726EXPORT_SYMBOL(inet_csk_prepare_forced_close);
727
728int inet_csk_listen_start(struct sock *sk, int backlog)
729{
730 struct inet_connection_sock *icsk = inet_csk(sk);
731 struct inet_sock *inet = inet_sk(sk);
732 int err = -EADDRINUSE;
733
734 reqsk_queue_alloc(&icsk->icsk_accept_queue);
735
736 sk->sk_max_ack_backlog = backlog;
737 sk->sk_ack_backlog = 0;
738 inet_csk_delack_init(sk);
739
740 /* There is race window here: we announce ourselves listening,
741 * but this transition is still not validated by get_port().
742 * It is OK, because this socket enters to hash table only
743 * after validation is complete.
744 */
745 sk_state_store(sk, TCP_LISTEN);
746 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
747 inet->inet_sport = htons(inet->inet_num);
748
749 sk_dst_reset(sk);
750 err = sk->sk_prot->hash(sk);
751
752 if (likely(!err))
753 return 0;
754 }
755
756 sk->sk_state = TCP_CLOSE;
757 return err;
758}
759EXPORT_SYMBOL_GPL(inet_csk_listen_start);
760
761static void inet_child_forget(struct sock *sk, struct request_sock *req,
762 struct sock *child)
763{
764 sk->sk_prot->disconnect(child, O_NONBLOCK);
765
766 sock_orphan(child);
767
768 percpu_counter_inc(sk->sk_prot->orphan_count);
769
770 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
771 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
772 BUG_ON(sk != req->rsk_listener);
773
774 /* Paranoid, to prevent race condition if
775 * an inbound pkt destined for child is
776 * blocked by sock lock in tcp_v4_rcv().
777 * Also to satisfy an assertion in
778 * tcp_v4_destroy_sock().
779 */
780 tcp_sk(child)->fastopen_rsk = NULL;
781 }
782 inet_csk_destroy_sock(child);
783 reqsk_put(req);
784}
785
786struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
787 struct request_sock *req,
788 struct sock *child)
789{
790 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
791
792 spin_lock(&queue->rskq_lock);
793 if (unlikely(sk->sk_state != TCP_LISTEN)) {
794 inet_child_forget(sk, req, child);
795 child = NULL;
796 } else {
797 req->sk = child;
798 req->dl_next = NULL;
799 if (queue->rskq_accept_head == NULL)
800 queue->rskq_accept_head = req;
801 else
802 queue->rskq_accept_tail->dl_next = req;
803 queue->rskq_accept_tail = req;
804 sk_acceptq_added(sk);
805 }
806 spin_unlock(&queue->rskq_lock);
807 return child;
808}
809EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
810
811struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
812 struct request_sock *req, bool own_req)
813{
814 if (own_req) {
815 inet_csk_reqsk_queue_drop(sk, req);
816 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
817 if (inet_csk_reqsk_queue_add(sk, req, child))
818 return child;
819 }
820 /* Too bad, another child took ownership of the request, undo. */
821 bh_unlock_sock(child);
822 sock_put(child);
823 return NULL;
824}
825EXPORT_SYMBOL(inet_csk_complete_hashdance);
826
827/*
828 * This routine closes sockets which have been at least partially
829 * opened, but not yet accepted.
830 */
831void inet_csk_listen_stop(struct sock *sk)
832{
833 struct inet_connection_sock *icsk = inet_csk(sk);
834 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
835 struct request_sock *next, *req;
836
837 /* Following specs, it would be better either to send FIN
838 * (and enter FIN-WAIT-1, it is normal close)
839 * or to send active reset (abort).
840 * Certainly, it is pretty dangerous while synflood, but it is
841 * bad justification for our negligence 8)
842 * To be honest, we are not able to make either
843 * of the variants now. --ANK
844 */
845 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
846 struct sock *child = req->sk;
847
848 local_bh_disable();
849 bh_lock_sock(child);
850 WARN_ON(sock_owned_by_user(child));
851 sock_hold(child);
852
853 inet_child_forget(sk, req, child);
854 bh_unlock_sock(child);
855 local_bh_enable();
856 sock_put(child);
857
858 cond_resched();
859 }
860 if (queue->fastopenq.rskq_rst_head) {
861 /* Free all the reqs queued in rskq_rst_head. */
862 spin_lock_bh(&queue->fastopenq.lock);
863 req = queue->fastopenq.rskq_rst_head;
864 queue->fastopenq.rskq_rst_head = NULL;
865 spin_unlock_bh(&queue->fastopenq.lock);
866 while (req != NULL) {
867 next = req->dl_next;
868 reqsk_put(req);
869 req = next;
870 }
871 }
872 WARN_ON_ONCE(sk->sk_ack_backlog);
873}
874EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
875
876void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
877{
878 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
879 const struct inet_sock *inet = inet_sk(sk);
880
881 sin->sin_family = AF_INET;
882 sin->sin_addr.s_addr = inet->inet_daddr;
883 sin->sin_port = inet->inet_dport;
884}
885EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
886
887#ifdef CONFIG_COMPAT
888int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
889 char __user *optval, int __user *optlen)
890{
891 const struct inet_connection_sock *icsk = inet_csk(sk);
892
893 if (icsk->icsk_af_ops->compat_getsockopt)
894 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
895 optval, optlen);
896 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
897 optval, optlen);
898}
899EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
900
901int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
902 char __user *optval, unsigned int optlen)
903{
904 const struct inet_connection_sock *icsk = inet_csk(sk);
905
906 if (icsk->icsk_af_ops->compat_setsockopt)
907 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
908 optval, optlen);
909 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
910 optval, optlen);
911}
912EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
913#endif
914
915static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
916{
917 const struct inet_sock *inet = inet_sk(sk);
918 const struct ip_options_rcu *inet_opt;
919 __be32 daddr = inet->inet_daddr;
920 struct flowi4 *fl4;
921 struct rtable *rt;
922
923 rcu_read_lock();
924 inet_opt = rcu_dereference(inet->inet_opt);
925 if (inet_opt && inet_opt->opt.srr)
926 daddr = inet_opt->opt.faddr;
927 fl4 = &fl->u.ip4;
928 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
929 inet->inet_saddr, inet->inet_dport,
930 inet->inet_sport, sk->sk_protocol,
931 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
932 if (IS_ERR(rt))
933 rt = NULL;
934 if (rt)
935 sk_setup_caps(sk, &rt->dst);
936 rcu_read_unlock();
937
938 return &rt->dst;
939}
940
941struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
942{
943 struct dst_entry *dst = __sk_dst_check(sk, 0);
944 struct inet_sock *inet = inet_sk(sk);
945
946 if (!dst) {
947 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
948 if (!dst)
949 goto out;
950 }
951 dst->ops->update_pmtu(dst, sk, NULL, mtu);
952
953 dst = __sk_dst_check(sk, 0);
954 if (!dst)
955 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
956out:
957 return dst;
958}
959EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);