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(addr: sk1_rcv_saddr6);
42 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(addr: 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(a1: sk1_rcv_saddr6, a2: 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(sk1_rcv_saddr6: &sk->sk_v6_rcv_saddr,
97 sk2_rcv_saddr6: inet6_rcv_saddr(sk: sk2),
98 sk1_rcv_saddr: sk->sk_rcv_saddr,
99 sk2_rcv_saddr: sk2->sk_rcv_saddr,
100 ipv6_only_sock(sk),
101 ipv6_only_sock(sk2),
102 match_sk1_wildcard: match_wildcard,
103 match_sk2_wildcard: match_wildcard);
104#endif
105 return ipv4_rcv_saddr_equal(sk1_rcv_saddr: sk->sk_rcv_saddr, sk2_rcv_saddr: sk2->sk_rcv_saddr,
106 ipv6_only_sock(sk2), match_sk1_wildcard: match_wildcard,
107 match_sk2_wildcard: 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(a: &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 */
130bool 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(net: sock_net(sk), low: &lo, high: &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}
154EXPORT_SYMBOL(inet_sk_get_local_port_range);
155
156static 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(a: &sk->sk_v6_rcv_saddr))
161 return false;
162
163 if (!ipv6_addr_v4mapped(a: &sk->sk_v6_rcv_saddr))
164 return true;
165 }
166#endif
167 return sk->sk_rcv_saddr != htonl(INADDR_ANY);
168}
169
170static 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(left: uid, right: sk_uid(sk: 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(left: uid, right: sk_uid(sk: sk2)))) {
194 return true;
195 }
196 }
197 return false;
198}
199
200static 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(a: &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
218static 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 */
240static 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 */
287static 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(lock: &head2->lock);
307
308 inet_bind_bucket_for_each(tb2, &head2->chain) {
309 if (!inet_bind2_bucket_match_addr_any(tb: 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(lock: &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 */
328static struct inet_bind_hashbucket *
329inet_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);
343ports_exhausted:
344 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
345other_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(ceil: 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
368other_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, lport: port,
376 bhash_size: hinfo->bhash_size)];
377 spin_lock_bh(lock: &head->lock);
378 if (inet_use_bhash2_on_bind(sk)) {
379 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, reuseport_ok: false))
380 goto next_port;
381 }
382
383 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
384 spin_lock(lock: &head2->lock);
385 tb2 = inet_bind2_bucket_find(head: 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, reuseport_ok: false))
390 goto success;
391 spin_unlock(lock: &head2->lock);
392 goto next_port;
393 }
394 tb = NULL;
395 goto success;
396next_port:
397 spin_unlock_bh(lock: &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;
417success:
418 *port_ret = port;
419 *tb_ret = tb;
420 *tb2_ret = tb2;
421 *head2_ret = head2;
422 return head;
423}
424
425static 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(left: tb->fastuid, right: 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(sk1_rcv_saddr6: &tb->fast_v6_rcv_saddr,
446 sk2_rcv_saddr6: inet6_rcv_saddr(sk),
447 sk1_rcv_saddr: tb->fast_rcv_saddr,
448 sk2_rcv_saddr: sk->sk_rcv_saddr,
449 sk1_ipv6only: tb->fast_ipv6_only,
450 ipv6_only_sock(sk), match_sk1_wildcard: true, match_sk2_wildcard: false);
451#endif
452 return ipv4_rcv_saddr_equal(sk1_rcv_saddr: tb->fast_rcv_saddr, sk2_rcv_saddr: sk->sk_rcv_saddr,
453 ipv6_only_sock(sk), match_sk1_wildcard: true, match_sk2_wildcard: false);
454}
455
456void 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(h: &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 */
514int 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_ret: &tb, tb2_ret: &tb2, head2_ret: &head2, port_ret: &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, lport: port,
541 bhash_size: hinfo->bhash_size)];
542 spin_lock_bh(lock: &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(cachep: hinfo->bind_bucket_cachep, net,
550 head, snum: port, l3mdev);
551 if (!tb)
552 goto fail_unlock;
553 bhash_created = true;
554 }
555
556 if (!found_port) {
557 if (!hlist_empty(h: &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, relax: true, reuseport_ok: true))
566 goto fail_unlock;
567 }
568
569 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
570 spin_lock(lock: &head2->lock);
571 head2_lock_acquired = true;
572 tb2 = inet_bind2_bucket_find(head: head2, net, port, l3mdev, sk);
573 }
574
575 if (!tb2) {
576 tb2 = inet_bind2_bucket_create(cachep: hinfo->bind2_bucket_cachep,
577 net, head: 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, relax: true, reuseport_ok: true))
585 goto fail_unlock;
586 }
587
588success:
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
597fail_unlock:
598 if (ret) {
599 if (bhash2_created)
600 inet_bind2_bucket_destroy(cachep: hinfo->bind2_bucket_cachep, tb: tb2);
601 if (bhash_created)
602 inet_bind_bucket_destroy(tb);
603 }
604 if (head2_lock_acquired)
605 spin_unlock(lock: &head2->lock);
606 spin_unlock_bh(lock: &head->lock);
607 return ret;
608}
609EXPORT_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 */
615static 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(wq_head: sk_sleep(sk), wq_entry: &wait,
637 TASK_INTERRUPTIBLE);
638 release_sock(sk);
639 if (reqsk_queue_empty(queue: &icsk->icsk_accept_queue))
640 timeo = schedule_timeout(timeout: timeo);
641 sched_annotate_sleep();
642 lock_sock(sk);
643 err = 0;
644 if (!reqsk_queue_empty(queue: &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(wq_head: sk_sleep(sk), wq_entry: &wait);
657 return err;
658}
659
660/*
661 * This will accept the next outstanding connection.
662 */
663struct 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, noblock: 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, parent: 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(lock: &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(lock: &queue->fastopenq.lock);
711 }
712
713 release_sock(sk);
714
715 if (mem_cgroup_sockets_enabled) {
716 gfp_t gfp = GFP_KERNEL | __GFP_NOFAIL;
717 int amt = 0;
718
719 /* atomically get the memory usage, set and charge the
720 * newsk->sk_memcg.
721 */
722 lock_sock(sk: newsk);
723
724 mem_cgroup_sk_alloc(sk: newsk);
725 if (mem_cgroup_from_sk(sk: newsk)) {
726 /* The socket has not been accepted yet, no need
727 * to look at newsk->sk_wmem_queued.
728 */
729 amt = sk_mem_pages(amt: newsk->sk_forward_alloc +
730 atomic_read(v: &newsk->sk_rmem_alloc));
731 }
732
733 if (amt)
734 mem_cgroup_sk_charge(sk: newsk, nr_pages: amt, gfp_mask: gfp);
735 kmem_cache_charge(objp: newsk, gfpflags: gfp);
736
737 release_sock(sk: newsk);
738 }
739
740 if (req)
741 reqsk_put(req);
742
743 inet_init_csk_locks(sk: newsk);
744 return newsk;
745
746out_err:
747 release_sock(sk);
748 arg->err = error;
749 return NULL;
750}
751EXPORT_SYMBOL(inet_csk_accept);
752
753/*
754 * Using different timers for retransmit, delayed acks and probes
755 * We may wish use just one timer maintaining a list of expire jiffies
756 * to optimize.
757 */
758void inet_csk_init_xmit_timers(struct sock *sk,
759 void (*retransmit_handler)(struct timer_list *t),
760 void (*delack_handler)(struct timer_list *t),
761 void (*keepalive_handler)(struct timer_list *t))
762{
763 struct inet_connection_sock *icsk = inet_csk(sk);
764
765 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
766 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
767 timer_setup(&sk->sk_timer, keepalive_handler, 0);
768 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
769}
770
771void inet_csk_clear_xmit_timers(struct sock *sk)
772{
773 struct inet_connection_sock *icsk = inet_csk(sk);
774
775 smp_store_release(&icsk->icsk_pending, 0);
776 smp_store_release(&icsk->icsk_ack.pending, 0);
777
778 sk_stop_timer(sk, timer: &icsk->icsk_retransmit_timer);
779 sk_stop_timer(sk, timer: &icsk->icsk_delack_timer);
780 sk_stop_timer(sk, timer: &sk->sk_timer);
781}
782
783void inet_csk_clear_xmit_timers_sync(struct sock *sk)
784{
785 struct inet_connection_sock *icsk = inet_csk(sk);
786
787 /* ongoing timer handlers need to acquire socket lock. */
788 sock_not_owned_by_me(sk);
789
790 smp_store_release(&icsk->icsk_pending, 0);
791 smp_store_release(&icsk->icsk_ack.pending, 0);
792
793 sk_stop_timer_sync(sk, timer: &icsk->icsk_retransmit_timer);
794 sk_stop_timer_sync(sk, timer: &icsk->icsk_delack_timer);
795 sk_stop_timer_sync(sk, timer: &sk->sk_timer);
796}
797
798struct dst_entry *inet_csk_route_req(const struct sock *sk,
799 struct flowi4 *fl4,
800 const struct request_sock *req)
801{
802 const struct inet_request_sock *ireq = inet_rsk(sk: req);
803 struct net *net = read_pnet(pnet: &ireq->ireq_net);
804 struct ip_options_rcu *opt;
805 struct rtable *rt;
806
807 rcu_read_lock();
808 opt = rcu_dereference(ireq->ireq_opt);
809
810 flowi4_init_output(fl4, oif: ireq->ir_iif, mark: ireq->ir_mark,
811 tos: ip_sock_rt_tos(sk), scope: ip_sock_rt_scope(sk),
812 proto: sk->sk_protocol, flags: inet_sk_flowi_flags(sk),
813 daddr: (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
814 saddr: ireq->ir_loc_addr, dport: ireq->ir_rmt_port,
815 htons(ireq->ir_num), uid: sk_uid(sk));
816 security_req_classify_flow(req, flic: flowi4_to_flowi_common(fl4));
817 rt = ip_route_output_flow(net, flp: fl4, sk);
818 if (IS_ERR(ptr: rt))
819 goto no_route;
820 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
821 goto route_err;
822 rcu_read_unlock();
823 return &rt->dst;
824
825route_err:
826 ip_rt_put(rt);
827no_route:
828 rcu_read_unlock();
829 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
830 return NULL;
831}
832
833struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
834 struct sock *newsk,
835 const struct request_sock *req)
836{
837 const struct inet_request_sock *ireq = inet_rsk(sk: req);
838 struct net *net = read_pnet(pnet: &ireq->ireq_net);
839 struct inet_sock *newinet = inet_sk(newsk);
840 struct ip_options_rcu *opt;
841 struct flowi4 *fl4;
842 struct rtable *rt;
843
844 opt = rcu_dereference(ireq->ireq_opt);
845 fl4 = &newinet->cork.fl.u.ip4;
846
847 flowi4_init_output(fl4, oif: ireq->ir_iif, mark: ireq->ir_mark,
848 tos: ip_sock_rt_tos(sk), scope: ip_sock_rt_scope(sk),
849 proto: sk->sk_protocol, flags: inet_sk_flowi_flags(sk),
850 daddr: (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
851 saddr: ireq->ir_loc_addr, dport: ireq->ir_rmt_port,
852 htons(ireq->ir_num), uid: sk_uid(sk));
853 security_req_classify_flow(req, flic: flowi4_to_flowi_common(fl4));
854 rt = ip_route_output_flow(net, flp: fl4, sk);
855 if (IS_ERR(ptr: rt))
856 goto no_route;
857 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
858 goto route_err;
859 return &rt->dst;
860
861route_err:
862 ip_rt_put(rt);
863no_route:
864 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
865 return NULL;
866}
867EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
868
869/* Decide when to expire the request and when to resend SYN-ACK */
870static void syn_ack_recalc(struct request_sock *req,
871 const int max_syn_ack_retries,
872 const u8 rskq_defer_accept,
873 int *expire, int *resend)
874{
875 if (!rskq_defer_accept) {
876 *expire = req->num_timeout >= max_syn_ack_retries;
877 *resend = 1;
878 return;
879 }
880 *expire = req->num_timeout >= max_syn_ack_retries &&
881 (!inet_rsk(sk: req)->acked || req->num_timeout >= rskq_defer_accept);
882 /* Do not resend while waiting for data after ACK,
883 * start to resend on end of deferring period to give
884 * last chance for data or ACK to create established socket.
885 */
886 *resend = !inet_rsk(sk: req)->acked ||
887 req->num_timeout >= rskq_defer_accept - 1;
888}
889
890static struct request_sock *
891reqsk_alloc_noprof(const struct request_sock_ops *ops, struct sock *sk_listener,
892 bool attach_listener)
893{
894 struct request_sock *req;
895
896 req = kmem_cache_alloc_noprof(cachep: ops->slab, GFP_ATOMIC | __GFP_NOWARN);
897 if (!req)
898 return NULL;
899 req->rsk_listener = NULL;
900 if (attach_listener) {
901 if (unlikely(!refcount_inc_not_zero(&sk_listener->sk_refcnt))) {
902 kmem_cache_free(s: ops->slab, objp: req);
903 return NULL;
904 }
905 req->rsk_listener = sk_listener;
906 }
907 req->rsk_ops = ops;
908 req_to_sk(req)->sk_prot = sk_listener->sk_prot;
909 sk_node_init(node: &req_to_sk(req)->sk_node);
910 sk_tx_queue_clear(sk: req_to_sk(req));
911 req->saved_syn = NULL;
912 req->syncookie = 0;
913 req->timeout = 0;
914 req->num_timeout = 0;
915 req->num_retrans = 0;
916 req->sk = NULL;
917 refcount_set(r: &req->rsk_refcnt, n: 0);
918
919 return req;
920}
921#define reqsk_alloc(...) alloc_hooks(reqsk_alloc_noprof(__VA_ARGS__))
922
923struct request_sock *inet_reqsk_alloc(const struct request_sock_ops *ops,
924 struct sock *sk_listener,
925 bool attach_listener)
926{
927 struct request_sock *req = reqsk_alloc(ops, sk_listener,
928 attach_listener);
929
930 if (req) {
931 struct inet_request_sock *ireq = inet_rsk(sk: req);
932
933 ireq->ireq_opt = NULL;
934#if IS_ENABLED(CONFIG_IPV6)
935 ireq->pktopts = NULL;
936#endif
937 atomic64_set(v: &ireq->ir_cookie, i: 0);
938 ireq->ireq_state = TCP_NEW_SYN_RECV;
939 write_pnet(pnet: &ireq->ireq_net, net: sock_net(sk: sk_listener));
940 ireq->ireq_family = sk_listener->sk_family;
941 req->timeout = TCP_TIMEOUT_INIT;
942 }
943
944 return req;
945}
946EXPORT_SYMBOL(inet_reqsk_alloc);
947
948static struct request_sock *inet_reqsk_clone(struct request_sock *req,
949 struct sock *sk)
950{
951 struct sock *req_sk, *nreq_sk;
952 struct request_sock *nreq;
953
954 nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
955 if (!nreq) {
956 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
957
958 /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
959 sock_put(sk);
960 return NULL;
961 }
962
963 req_sk = req_to_sk(req);
964 nreq_sk = req_to_sk(req: nreq);
965
966 memcpy(to: nreq_sk, from: req_sk,
967 offsetof(struct sock, sk_dontcopy_begin));
968 unsafe_memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
969 req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end),
970 /* alloc is larger than struct, see above */);
971
972 sk_node_init(node: &nreq_sk->sk_node);
973 nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
974#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
975 nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
976#endif
977 nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
978
979 nreq->rsk_listener = sk;
980
981 /* We need not acquire fastopenq->lock
982 * because the child socket is locked in inet_csk_listen_stop().
983 */
984 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req: nreq)->tfo_listener)
985 rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
986
987 return nreq;
988}
989
990static void reqsk_queue_migrated(struct request_sock_queue *queue,
991 const struct request_sock *req)
992{
993 if (req->num_timeout == 0)
994 atomic_inc(v: &queue->young);
995 atomic_inc(v: &queue->qlen);
996}
997
998static void reqsk_migrate_reset(struct request_sock *req)
999{
1000 req->saved_syn = NULL;
1001#if IS_ENABLED(CONFIG_IPV6)
1002 inet_rsk(sk: req)->ipv6_opt = NULL;
1003 inet_rsk(sk: req)->pktopts = NULL;
1004#else
1005 inet_rsk(req)->ireq_opt = NULL;
1006#endif
1007}
1008
1009/* return true if req was found in the ehash table */
1010static bool reqsk_queue_unlink(struct request_sock *req)
1011{
1012 struct sock *sk = req_to_sk(req);
1013 bool found = false;
1014
1015 if (sk_hashed(sk)) {
1016 struct inet_hashinfo *hashinfo = tcp_get_hashinfo(sk);
1017 spinlock_t *lock;
1018
1019 lock = inet_ehash_lockp(hashinfo, hash: req->rsk_hash);
1020 spin_lock(lock);
1021 found = __sk_nulls_del_node_init_rcu(sk);
1022 spin_unlock(lock);
1023 }
1024
1025 return found;
1026}
1027
1028static bool __inet_csk_reqsk_queue_drop(struct sock *sk,
1029 struct request_sock *req,
1030 bool from_timer)
1031{
1032 bool unlinked = reqsk_queue_unlink(req);
1033
1034 if (!from_timer && timer_delete_sync(timer: &req->rsk_timer))
1035 reqsk_put(req);
1036
1037 if (unlinked) {
1038 reqsk_queue_removed(queue: &inet_csk(sk)->icsk_accept_queue, req);
1039 reqsk_put(req);
1040 }
1041
1042 return unlinked;
1043}
1044
1045bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
1046{
1047 return __inet_csk_reqsk_queue_drop(sk, req, from_timer: false);
1048}
1049
1050void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
1051{
1052 inet_csk_reqsk_queue_drop(sk, req);
1053 reqsk_put(req);
1054}
1055EXPORT_IPV6_MOD(inet_csk_reqsk_queue_drop_and_put);
1056
1057static void reqsk_timer_handler(struct timer_list *t)
1058{
1059 struct request_sock *req = timer_container_of(req, t, rsk_timer);
1060 struct request_sock *nreq = NULL, *oreq = req;
1061 struct sock *sk_listener = req->rsk_listener;
1062 struct inet_connection_sock *icsk;
1063 struct request_sock_queue *queue;
1064 struct net *net;
1065 int max_syn_ack_retries, qlen, expire = 0, resend = 0;
1066
1067 if (inet_sk_state_load(sk: sk_listener) != TCP_LISTEN) {
1068 struct sock *nsk;
1069
1070 nsk = reuseport_migrate_sock(sk: sk_listener, migrating_sk: req_to_sk(req), NULL);
1071 if (!nsk)
1072 goto drop;
1073
1074 nreq = inet_reqsk_clone(req, sk: nsk);
1075 if (!nreq)
1076 goto drop;
1077
1078 /* The new timer for the cloned req can decrease the 2
1079 * by calling inet_csk_reqsk_queue_drop_and_put(), so
1080 * hold another count to prevent use-after-free and
1081 * call reqsk_put() just before return.
1082 */
1083 refcount_set(r: &nreq->rsk_refcnt, n: 2 + 1);
1084 timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1085 reqsk_queue_migrated(queue: &inet_csk(nsk)->icsk_accept_queue, req);
1086
1087 req = nreq;
1088 sk_listener = nsk;
1089 }
1090
1091 icsk = inet_csk(sk_listener);
1092 net = sock_net(sk: sk_listener);
1093 max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? :
1094 READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1095 /* Normally all the openreqs are young and become mature
1096 * (i.e. converted to established socket) for first timeout.
1097 * If synack was not acknowledged for 1 second, it means
1098 * one of the following things: synack was lost, ack was lost,
1099 * rtt is high or nobody planned to ack (i.e. synflood).
1100 * When server is a bit loaded, queue is populated with old
1101 * open requests, reducing effective size of queue.
1102 * When server is well loaded, queue size reduces to zero
1103 * after several minutes of work. It is not synflood,
1104 * it is normal operation. The solution is pruning
1105 * too old entries overriding normal timeout, when
1106 * situation becomes dangerous.
1107 *
1108 * Essentially, we reserve half of room for young
1109 * embrions; and abort old ones without pity, if old
1110 * ones are about to clog our table.
1111 */
1112 queue = &icsk->icsk_accept_queue;
1113 qlen = reqsk_queue_len(queue);
1114 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1115 int young = reqsk_queue_len_young(queue) << 1;
1116
1117 while (max_syn_ack_retries > 2) {
1118 if (qlen < young)
1119 break;
1120 max_syn_ack_retries--;
1121 young <<= 1;
1122 }
1123 }
1124 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
1125 expire: &expire, resend: &resend);
1126 req->rsk_ops->syn_ack_timeout(req);
1127 if (!expire &&
1128 (!resend ||
1129 !tcp_rtx_synack(sk: sk_listener, req) ||
1130 inet_rsk(sk: req)->acked)) {
1131 if (req->num_timeout++ == 0)
1132 atomic_dec(v: &queue->young);
1133 mod_timer(timer: &req->rsk_timer, expires: jiffies + reqsk_timeout(req, TCP_RTO_MAX));
1134
1135 if (!nreq)
1136 return;
1137
1138 if (!inet_ehash_insert(sk: req_to_sk(req: nreq), osk: req_to_sk(req: oreq), NULL)) {
1139 /* delete timer */
1140 __inet_csk_reqsk_queue_drop(sk: sk_listener, req: nreq, from_timer: true);
1141 goto no_ownership;
1142 }
1143
1144 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1145 reqsk_migrate_reset(req: oreq);
1146 reqsk_queue_removed(queue: &inet_csk(oreq->rsk_listener)->icsk_accept_queue, req: oreq);
1147 reqsk_put(req: oreq);
1148
1149 reqsk_put(req: nreq);
1150 return;
1151 }
1152
1153 /* Even if we can clone the req, we may need not retransmit any more
1154 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1155 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1156 */
1157 if (nreq) {
1158 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1159no_ownership:
1160 reqsk_migrate_reset(req: nreq);
1161 reqsk_queue_removed(queue, req: nreq);
1162 __reqsk_free(req: nreq);
1163 }
1164
1165drop:
1166 __inet_csk_reqsk_queue_drop(sk: sk_listener, req: oreq, from_timer: true);
1167 reqsk_put(req: oreq);
1168}
1169
1170static bool reqsk_queue_hash_req(struct request_sock *req,
1171 unsigned long timeout)
1172{
1173 bool found_dup_sk = false;
1174
1175 if (!inet_ehash_insert(sk: req_to_sk(req), NULL, found_dup_sk: &found_dup_sk))
1176 return false;
1177
1178 /* The timer needs to be setup after a successful insertion. */
1179 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1180 mod_timer(timer: &req->rsk_timer, expires: jiffies + timeout);
1181
1182 /* before letting lookups find us, make sure all req fields
1183 * are committed to memory and refcnt initialized.
1184 */
1185 smp_wmb();
1186 refcount_set(r: &req->rsk_refcnt, n: 2 + 1);
1187 return true;
1188}
1189
1190bool inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1191 unsigned long timeout)
1192{
1193 if (!reqsk_queue_hash_req(req, timeout))
1194 return false;
1195
1196 inet_csk_reqsk_queue_added(sk);
1197 return true;
1198}
1199
1200static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1201 const gfp_t priority)
1202{
1203 struct inet_connection_sock *icsk = inet_csk(newsk);
1204
1205 if (!icsk->icsk_ulp_ops)
1206 return;
1207
1208 icsk->icsk_ulp_ops->clone(req, newsk, priority);
1209}
1210
1211/**
1212 * inet_csk_clone_lock - clone an inet socket, and lock its clone
1213 * @sk: the socket to clone
1214 * @req: request_sock
1215 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1216 *
1217 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1218 */
1219struct sock *inet_csk_clone_lock(const struct sock *sk,
1220 const struct request_sock *req,
1221 const gfp_t priority)
1222{
1223 struct sock *newsk = sk_clone_lock(sk, priority);
1224 struct inet_connection_sock *newicsk;
1225 struct inet_request_sock *ireq;
1226 struct inet_sock *newinet;
1227
1228 if (!newsk)
1229 return NULL;
1230
1231 newicsk = inet_csk(newsk);
1232 newinet = inet_sk(newsk);
1233 ireq = inet_rsk(sk: req);
1234
1235 newicsk->icsk_bind_hash = NULL;
1236 newicsk->icsk_bind2_hash = NULL;
1237
1238 newinet->inet_dport = ireq->ir_rmt_port;
1239 newinet->inet_num = ireq->ir_num;
1240 newinet->inet_sport = htons(ireq->ir_num);
1241
1242 newsk->sk_bound_dev_if = ireq->ir_iif;
1243
1244 newsk->sk_daddr = ireq->ir_rmt_addr;
1245 newsk->sk_rcv_saddr = ireq->ir_loc_addr;
1246 newinet->inet_saddr = ireq->ir_loc_addr;
1247
1248#if IS_ENABLED(CONFIG_IPV6)
1249 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1250 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1251#endif
1252
1253 /* listeners have SOCK_RCU_FREE, not the children */
1254 sock_reset_flag(sk: newsk, flag: SOCK_RCU_FREE);
1255
1256 inet_sk(newsk)->mc_list = NULL;
1257
1258 newsk->sk_mark = inet_rsk(sk: req)->ir_mark;
1259 atomic64_set(v: &newsk->sk_cookie,
1260 i: atomic64_read(v: &inet_rsk(sk: req)->ir_cookie));
1261
1262 newicsk->icsk_retransmits = 0;
1263 newicsk->icsk_backoff = 0;
1264 newicsk->icsk_probes_out = 0;
1265 newicsk->icsk_probes_tstamp = 0;
1266
1267 /* Deinitialize accept_queue to trap illegal accesses. */
1268 memset(s: &newicsk->icsk_accept_queue, c: 0,
1269 n: sizeof(newicsk->icsk_accept_queue));
1270
1271 inet_sk_set_state(sk: newsk, state: TCP_SYN_RECV);
1272
1273 inet_clone_ulp(req, newsk, priority);
1274
1275 security_inet_csk_clone(newsk, req);
1276
1277 return newsk;
1278}
1279
1280/*
1281 * At this point, there should be no process reference to this
1282 * socket, and thus no user references at all. Therefore we
1283 * can assume the socket waitqueue is inactive and nobody will
1284 * try to jump onto it.
1285 */
1286void inet_csk_destroy_sock(struct sock *sk)
1287{
1288 WARN_ON(sk->sk_state != TCP_CLOSE);
1289 WARN_ON(!sock_flag(sk, SOCK_DEAD));
1290
1291 /* It cannot be in hash table! */
1292 WARN_ON(!sk_unhashed(sk));
1293
1294 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1295 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1296
1297 sk->sk_prot->destroy(sk);
1298
1299 sk_stream_kill_queues(sk);
1300
1301 xfrm_sk_free_policy(sk);
1302
1303 tcp_orphan_count_dec();
1304
1305 sock_put(sk);
1306}
1307EXPORT_SYMBOL(inet_csk_destroy_sock);
1308
1309void inet_csk_prepare_for_destroy_sock(struct sock *sk)
1310{
1311 /* The below has to be done to allow calling inet_csk_destroy_sock */
1312 sock_set_flag(sk, flag: SOCK_DEAD);
1313 tcp_orphan_count_inc();
1314}
1315
1316/* This function allows to force a closure of a socket after the call to
1317 * tcp_create_openreq_child().
1318 */
1319void inet_csk_prepare_forced_close(struct sock *sk)
1320 __releases(&sk->sk_lock.slock)
1321{
1322 /* sk_clone_lock locked the socket and set refcnt to 2 */
1323 bh_unlock_sock(sk);
1324 sock_put(sk);
1325 inet_csk_prepare_for_destroy_sock(sk);
1326 inet_sk(sk)->inet_num = 0;
1327}
1328EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1329
1330static int inet_ulp_can_listen(const struct sock *sk)
1331{
1332 const struct inet_connection_sock *icsk = inet_csk(sk);
1333
1334 if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone)
1335 return -EINVAL;
1336
1337 return 0;
1338}
1339
1340int inet_csk_listen_start(struct sock *sk)
1341{
1342 struct inet_connection_sock *icsk = inet_csk(sk);
1343 struct inet_sock *inet = inet_sk(sk);
1344 int err;
1345
1346 err = inet_ulp_can_listen(sk);
1347 if (unlikely(err))
1348 return err;
1349
1350 reqsk_queue_alloc(queue: &icsk->icsk_accept_queue);
1351
1352 sk->sk_ack_backlog = 0;
1353 inet_csk_delack_init(sk);
1354
1355 /* There is race window here: we announce ourselves listening,
1356 * but this transition is still not validated by get_port().
1357 * It is OK, because this socket enters to hash table only
1358 * after validation is complete.
1359 */
1360 inet_sk_state_store(sk, newstate: TCP_LISTEN);
1361 err = sk->sk_prot->get_port(sk, inet->inet_num);
1362 if (!err) {
1363 inet->inet_sport = htons(inet->inet_num);
1364
1365 sk_dst_reset(sk);
1366 err = sk->sk_prot->hash(sk);
1367
1368 if (likely(!err))
1369 return 0;
1370 }
1371
1372 inet_sk_set_state(sk, state: TCP_CLOSE);
1373 return err;
1374}
1375
1376static void inet_child_forget(struct sock *sk, struct request_sock *req,
1377 struct sock *child)
1378{
1379 sk->sk_prot->disconnect(child, O_NONBLOCK);
1380
1381 sock_orphan(sk: child);
1382
1383 tcp_orphan_count_inc();
1384
1385 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1386 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1387 BUG_ON(sk != req->rsk_listener);
1388
1389 /* Paranoid, to prevent race condition if
1390 * an inbound pkt destined for child is
1391 * blocked by sock lock in tcp_v4_rcv().
1392 * Also to satisfy an assertion in
1393 * tcp_v4_destroy_sock().
1394 */
1395 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1396 }
1397 inet_csk_destroy_sock(child);
1398}
1399
1400struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1401 struct request_sock *req,
1402 struct sock *child)
1403{
1404 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1405
1406 spin_lock(lock: &queue->rskq_lock);
1407 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1408 inet_child_forget(sk, req, child);
1409 child = NULL;
1410 } else {
1411 req->sk = child;
1412 req->dl_next = NULL;
1413 if (queue->rskq_accept_head == NULL)
1414 WRITE_ONCE(queue->rskq_accept_head, req);
1415 else
1416 queue->rskq_accept_tail->dl_next = req;
1417 queue->rskq_accept_tail = req;
1418 sk_acceptq_added(sk);
1419 }
1420 spin_unlock(lock: &queue->rskq_lock);
1421 return child;
1422}
1423EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1424
1425struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1426 struct request_sock *req, bool own_req)
1427{
1428 if (own_req) {
1429 inet_csk_reqsk_queue_drop(sk: req->rsk_listener, req);
1430 reqsk_queue_removed(queue: &inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1431
1432 if (sk != req->rsk_listener) {
1433 /* another listening sk has been selected,
1434 * migrate the req to it.
1435 */
1436 struct request_sock *nreq;
1437
1438 /* hold a refcnt for the nreq->rsk_listener
1439 * which is assigned in inet_reqsk_clone()
1440 */
1441 sock_hold(sk);
1442 nreq = inet_reqsk_clone(req, sk);
1443 if (!nreq) {
1444 inet_child_forget(sk, req, child);
1445 goto child_put;
1446 }
1447
1448 refcount_set(r: &nreq->rsk_refcnt, n: 1);
1449 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1450 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1451 reqsk_migrate_reset(req);
1452 reqsk_put(req);
1453 return child;
1454 }
1455
1456 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1457 reqsk_migrate_reset(req: nreq);
1458 __reqsk_free(req: nreq);
1459 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1460 return child;
1461 }
1462 }
1463 /* Too bad, another child took ownership of the request, undo. */
1464child_put:
1465 bh_unlock_sock(child);
1466 sock_put(sk: child);
1467 return NULL;
1468}
1469
1470/*
1471 * This routine closes sockets which have been at least partially
1472 * opened, but not yet accepted.
1473 */
1474void inet_csk_listen_stop(struct sock *sk)
1475{
1476 struct inet_connection_sock *icsk = inet_csk(sk);
1477 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1478 struct request_sock *next, *req;
1479
1480 /* Following specs, it would be better either to send FIN
1481 * (and enter FIN-WAIT-1, it is normal close)
1482 * or to send active reset (abort).
1483 * Certainly, it is pretty dangerous while synflood, but it is
1484 * bad justification for our negligence 8)
1485 * To be honest, we are not able to make either
1486 * of the variants now. --ANK
1487 */
1488 while ((req = reqsk_queue_remove(queue, parent: sk)) != NULL) {
1489 struct sock *child = req->sk, *nsk;
1490 struct request_sock *nreq;
1491
1492 local_bh_disable();
1493 bh_lock_sock(child);
1494 WARN_ON(sock_owned_by_user(child));
1495 sock_hold(sk: child);
1496
1497 nsk = reuseport_migrate_sock(sk, migrating_sk: child, NULL);
1498 if (nsk) {
1499 nreq = inet_reqsk_clone(req, sk: nsk);
1500 if (nreq) {
1501 refcount_set(r: &nreq->rsk_refcnt, n: 1);
1502
1503 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1504 __NET_INC_STATS(sock_net(nsk),
1505 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1506 reqsk_migrate_reset(req);
1507 } else {
1508 __NET_INC_STATS(sock_net(nsk),
1509 LINUX_MIB_TCPMIGRATEREQFAILURE);
1510 reqsk_migrate_reset(req: nreq);
1511 __reqsk_free(req: nreq);
1512 }
1513
1514 /* inet_csk_reqsk_queue_add() has already
1515 * called inet_child_forget() on failure case.
1516 */
1517 goto skip_child_forget;
1518 }
1519 }
1520
1521 inet_child_forget(sk, req, child);
1522skip_child_forget:
1523 reqsk_put(req);
1524 bh_unlock_sock(child);
1525 local_bh_enable();
1526 sock_put(sk: child);
1527
1528 cond_resched();
1529 }
1530 if (queue->fastopenq.rskq_rst_head) {
1531 /* Free all the reqs queued in rskq_rst_head. */
1532 spin_lock_bh(lock: &queue->fastopenq.lock);
1533 req = queue->fastopenq.rskq_rst_head;
1534 queue->fastopenq.rskq_rst_head = NULL;
1535 spin_unlock_bh(lock: &queue->fastopenq.lock);
1536 while (req != NULL) {
1537 next = req->dl_next;
1538 reqsk_put(req);
1539 req = next;
1540 }
1541 }
1542 WARN_ON_ONCE(sk->sk_ack_backlog);
1543}
1544EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1545
1546static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1547{
1548 const struct inet_sock *inet = inet_sk(sk);
1549 struct flowi4 *fl4;
1550 struct rtable *rt;
1551
1552 rcu_read_lock();
1553 fl4 = &fl->u.ip4;
1554 inet_sk_init_flowi4(inet, fl4);
1555 rt = ip_route_output_flow(sock_net(sk), flp: fl4, sk);
1556 if (IS_ERR(ptr: rt))
1557 rt = NULL;
1558 if (rt)
1559 sk_setup_caps(sk, dst: &rt->dst);
1560 rcu_read_unlock();
1561
1562 return &rt->dst;
1563}
1564
1565struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1566{
1567 struct dst_entry *dst = __sk_dst_check(sk, cookie: 0);
1568 struct inet_sock *inet = inet_sk(sk);
1569
1570 if (!dst) {
1571 dst = inet_csk_rebuild_route(sk, fl: &inet->cork.fl);
1572 if (!dst)
1573 goto out;
1574 }
1575 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1576
1577 dst = __sk_dst_check(sk, cookie: 0);
1578 if (!dst)
1579 dst = inet_csk_rebuild_route(sk, fl: &inet->cork.fl);
1580out:
1581 return dst;
1582}
1583