1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Support for INET connection oriented protocols.
8 *
9 * Authors: See the TCP sources
10 */
11
12 #include <linux/module.h>
13 #include <linux/jhash.h>
14
15 #include <net/inet_connection_sock.h>
16 #include <net/inet_hashtables.h>
17 #include <net/inet_timewait_sock.h>
18 #include <net/ip.h>
19 #include <net/route.h>
20 #include <net/tcp_states.h>
21 #include <net/xfrm.h>
22 #include <net/tcp.h>
23 #include <net/sock_reuseport.h>
24 #include <net/addrconf.h>
25
26 #if IS_ENABLED(CONFIG_IPV6)
27 /* match_sk*_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses
28 * if IPv6 only, and any IPv4 addresses
29 * if not IPv6 only
30 * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
31 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
32 * and 0.0.0.0 equals to 0.0.0.0 only
33 */
34 static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
35 const struct in6_addr *sk2_rcv_saddr6,
36 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
37 bool sk1_ipv6only, bool sk2_ipv6only,
38 bool match_sk1_wildcard,
39 bool match_sk2_wildcard)
40 {
41 int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
42 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
43
44 /* if both are mapped, treat as IPv4 */
45 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
46 if (!sk2_ipv6only) {
47 if (sk1_rcv_saddr == sk2_rcv_saddr)
48 return true;
49 return (match_sk1_wildcard && !sk1_rcv_saddr) ||
50 (match_sk2_wildcard && !sk2_rcv_saddr);
51 }
52 return false;
53 }
54
55 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
56 return true;
57
58 if (addr_type2 == IPV6_ADDR_ANY && match_sk2_wildcard &&
59 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
60 return true;
61
62 if (addr_type == IPV6_ADDR_ANY && match_sk1_wildcard &&
63 !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
64 return true;
65
66 if (sk2_rcv_saddr6 &&
67 ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
68 return true;
69
70 return false;
71 }
72 #endif
73
74 /* match_sk*_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
75 * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
76 * 0.0.0.0 only equals to 0.0.0.0
77 */
78 static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
79 bool sk2_ipv6only, bool match_sk1_wildcard,
80 bool match_sk2_wildcard)
81 {
82 if (!sk2_ipv6only) {
83 if (sk1_rcv_saddr == sk2_rcv_saddr)
84 return true;
85 return (match_sk1_wildcard && !sk1_rcv_saddr) ||
86 (match_sk2_wildcard && !sk2_rcv_saddr);
87 }
88 return false;
89 }
90
91 bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
92 bool match_wildcard)
93 {
94 #if IS_ENABLED(CONFIG_IPV6)
95 if (sk->sk_family == AF_INET6)
96 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
97 inet6_rcv_saddr(sk2),
98 sk->sk_rcv_saddr,
99 sk2->sk_rcv_saddr,
100 ipv6_only_sock(sk),
101 ipv6_only_sock(sk2),
102 match_wildcard,
103 match_wildcard);
104 #endif
105 return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
106 ipv6_only_sock(sk2), match_wildcard,
107 match_wildcard);
108 }
109 EXPORT_SYMBOL(inet_rcv_saddr_equal);
110
111 bool inet_rcv_saddr_any(const struct sock *sk)
112 {
113 #if IS_ENABLED(CONFIG_IPV6)
114 if (sk->sk_family == AF_INET6)
115 return ipv6_addr_any(&sk->sk_v6_rcv_saddr);
116 #endif
117 return !sk->sk_rcv_saddr;
118 }
119
120 /**
121 * inet_sk_get_local_port_range - fetch ephemeral ports range
122 * @sk: socket
123 * @low: pointer to low port
124 * @high: pointer to high port
125 *
126 * Fetch netns port range (/proc/sys/net/ipv4/ip_local_port_range)
127 * Range can be overridden if socket got IP_LOCAL_PORT_RANGE option.
128 * Returns true if IP_LOCAL_PORT_RANGE was set on this socket.
129 */
130 bool inet_sk_get_local_port_range(const struct sock *sk, int *low, int *high)
131 {
132 int lo, hi, sk_lo, sk_hi;
133 bool local_range = false;
134 u32 sk_range;
135
136 inet_get_local_port_range(sock_net(sk), &lo, &hi);
137
138 sk_range = READ_ONCE(inet_sk(sk)->local_port_range);
139 if (unlikely(sk_range)) {
140 sk_lo = sk_range & 0xffff;
141 sk_hi = sk_range >> 16;
142
143 if (lo <= sk_lo && sk_lo <= hi)
144 lo = sk_lo;
145 if (lo <= sk_hi && sk_hi <= hi)
146 hi = sk_hi;
147 local_range = true;
148 }
149
150 *low = lo;
151 *high = hi;
152 return local_range;
153 }
154 EXPORT_SYMBOL(inet_sk_get_local_port_range);
155
156 static bool inet_use_bhash2_on_bind(const struct sock *sk)
157 {
158 #if IS_ENABLED(CONFIG_IPV6)
159 if (sk->sk_family == AF_INET6) {
160 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
161
162 if (addr_type == IPV6_ADDR_ANY)
163 return false;
164
165 if (addr_type != IPV6_ADDR_MAPPED)
166 return true;
167 }
168 #endif
169 return sk->sk_rcv_saddr != htonl(INADDR_ANY);
170 }
171
172 static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2,
173 kuid_t sk_uid, bool relax,
174 bool reuseport_cb_ok, bool reuseport_ok)
175 {
176 int bound_dev_if2;
177
178 if (sk == sk2)
179 return false;
180
181 bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
182
183 if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
184 sk->sk_bound_dev_if == bound_dev_if2) {
185 if (sk->sk_reuse && sk2->sk_reuse &&
186 sk2->sk_state != TCP_LISTEN) {
187 if (!relax || (!reuseport_ok && sk->sk_reuseport &&
188 sk2->sk_reuseport && reuseport_cb_ok &&
189 (sk2->sk_state == TCP_TIME_WAIT ||
190 uid_eq(sk_uid, sock_i_uid(sk2)))))
191 return true;
192 } else if (!reuseport_ok || !sk->sk_reuseport ||
193 !sk2->sk_reuseport || !reuseport_cb_ok ||
194 (sk2->sk_state != TCP_TIME_WAIT &&
195 !uid_eq(sk_uid, sock_i_uid(sk2)))) {
196 return true;
197 }
198 }
199 return false;
200 }
201
202 static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2,
203 kuid_t sk_uid, bool relax,
204 bool reuseport_cb_ok, bool reuseport_ok)
205 {
206 if (ipv6_only_sock(sk2)) {
207 if (sk->sk_family == AF_INET)
208 return false;
209
210 #if IS_ENABLED(CONFIG_IPV6)
211 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
212 return false;
213 #endif
214 }
215
216 return inet_bind_conflict(sk, sk2, sk_uid, relax,
217 reuseport_cb_ok, reuseport_ok);
218 }
219
220 static bool inet_bhash2_conflict(const struct sock *sk,
221 const struct inet_bind2_bucket *tb2,
222 kuid_t sk_uid,
223 bool relax, bool reuseport_cb_ok,
224 bool reuseport_ok)
225 {
226 struct sock *sk2;
227
228 sk_for_each_bound(sk2, &tb2->owners) {
229 if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
230 reuseport_cb_ok, reuseport_ok))
231 return true;
232 }
233
234 return false;
235 }
236
237 #define sk_for_each_bound_bhash(__sk, __tb2, __tb) \
238 hlist_for_each_entry(__tb2, &(__tb)->bhash2, bhash_node) \
239 sk_for_each_bound(sk2, &(__tb2)->owners)
240
241 /* This should be called only when the tb and tb2 hashbuckets' locks are held */
242 static int inet_csk_bind_conflict(const struct sock *sk,
243 const struct inet_bind_bucket *tb,
244 const struct inet_bind2_bucket *tb2, /* may be null */
245 bool relax, bool reuseport_ok)
246 {
247 kuid_t uid = sock_i_uid((struct sock *)sk);
248 struct sock_reuseport *reuseport_cb;
249 bool reuseport_cb_ok;
250 struct sock *sk2;
251
252 rcu_read_lock();
253 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
254 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
255 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
256 rcu_read_unlock();
257
258 /* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if
259 * ipv4) should have been checked already. We need to do these two
260 * checks separately because their spinlocks have to be acquired/released
261 * independently of each other, to prevent possible deadlocks
262 */
263 if (inet_use_bhash2_on_bind(sk))
264 return tb2 && inet_bhash2_conflict(sk, tb2, uid, relax,
265 reuseport_cb_ok, reuseport_ok);
266
267 /* Unlike other sk lookup places we do not check
268 * for sk_net here, since _all_ the socks listed
269 * in tb->owners and tb2->owners list belong
270 * to the same net - the one this bucket belongs to.
271 */
272 sk_for_each_bound_bhash(sk2, tb2, tb) {
273 if (!inet_bind_conflict(sk, sk2, uid, relax, reuseport_cb_ok, reuseport_ok))
274 continue;
275
276 if (inet_rcv_saddr_equal(sk, sk2, true))
277 return true;
278 }
279
280 return false;
281 }
282
283 /* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or
284 * INADDR_ANY (if ipv4) socket.
285 *
286 * Caller must hold bhash hashbucket lock with local bh disabled, to protect
287 * against concurrent binds on the port for addr any
288 */
289 static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev,
290 bool relax, bool reuseport_ok)
291 {
292 kuid_t uid = sock_i_uid((struct sock *)sk);
293 const struct net *net = sock_net(sk);
294 struct sock_reuseport *reuseport_cb;
295 struct inet_bind_hashbucket *head2;
296 struct inet_bind2_bucket *tb2;
297 bool conflict = false;
298 bool reuseport_cb_ok;
299
300 rcu_read_lock();
301 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
302 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
303 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
304 rcu_read_unlock();
305
306 head2 = inet_bhash2_addr_any_hashbucket(sk, net, port);
307
308 spin_lock(&head2->lock);
309
310 inet_bind_bucket_for_each(tb2, &head2->chain) {
311 if (!inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
312 continue;
313
314 if (!inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok, reuseport_ok))
315 continue;
316
317 conflict = true;
318 break;
319 }
320
321 spin_unlock(&head2->lock);
322
323 return conflict;
324 }
325
326 /*
327 * Find an open port number for the socket. Returns with the
328 * inet_bind_hashbucket locks held if successful.
329 */
330 static struct inet_bind_hashbucket *
331 inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret,
332 struct inet_bind2_bucket **tb2_ret,
333 struct inet_bind_hashbucket **head2_ret, int *port_ret)
334 {
335 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
336 int i, low, high, attempt_half, port, l3mdev;
337 struct inet_bind_hashbucket *head, *head2;
338 struct net *net = sock_net(sk);
339 struct inet_bind2_bucket *tb2;
340 struct inet_bind_bucket *tb;
341 u32 remaining, offset;
342 bool relax = false;
343
344 l3mdev = inet_sk_bound_l3mdev(sk);
345 ports_exhausted:
346 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
347 other_half_scan:
348 inet_sk_get_local_port_range(sk, &low, &high);
349 high++; /* [32768, 60999] -> [32768, 61000[ */
350 if (high - low < 4)
351 attempt_half = 0;
352 if (attempt_half) {
353 int half = low + (((high - low) >> 2) << 1);
354
355 if (attempt_half == 1)
356 high = half;
357 else
358 low = half;
359 }
360 remaining = high - low;
361 if (likely(remaining > 1))
362 remaining &= ~1U;
363
364 offset = get_random_u32_below(remaining);
365 /* __inet_hash_connect() favors ports having @low parity
366 * We do the opposite to not pollute connect() users.
367 */
368 offset |= 1U;
369
370 other_parity_scan:
371 port = low + offset;
372 for (i = 0; i < remaining; i += 2, port += 2) {
373 if (unlikely(port >= high))
374 port -= remaining;
375 if (inet_is_local_reserved_port(net, port))
376 continue;
377 head = &hinfo->bhash[inet_bhashfn(net, port,
378 hinfo->bhash_size)];
379 spin_lock_bh(&head->lock);
380 if (inet_use_bhash2_on_bind(sk)) {
381 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, false))
382 goto next_port;
383 }
384
385 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
386 spin_lock(&head2->lock);
387 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
388 inet_bind_bucket_for_each(tb, &head->chain)
389 if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
390 if (!inet_csk_bind_conflict(sk, tb, tb2,
391 relax, false))
392 goto success;
393 spin_unlock(&head2->lock);
394 goto next_port;
395 }
396 tb = NULL;
397 goto success;
398 next_port:
399 spin_unlock_bh(&head->lock);
400 cond_resched();
401 }
402
403 offset--;
404 if (!(offset & 1))
405 goto other_parity_scan;
406
407 if (attempt_half == 1) {
408 /* OK we now try the upper half of the range */
409 attempt_half = 2;
410 goto other_half_scan;
411 }
412
413 if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
414 /* We still have a chance to connect to different destinations */
415 relax = true;
416 goto ports_exhausted;
417 }
418 return NULL;
419 success:
420 *port_ret = port;
421 *tb_ret = tb;
422 *tb2_ret = tb2;
423 *head2_ret = head2;
424 return head;
425 }
426
427 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
428 struct sock *sk)
429 {
430 kuid_t uid = sock_i_uid(sk);
431
432 if (tb->fastreuseport <= 0)
433 return 0;
434 if (!sk->sk_reuseport)
435 return 0;
436 if (rcu_access_pointer(sk->sk_reuseport_cb))
437 return 0;
438 if (!uid_eq(tb->fastuid, uid))
439 return 0;
440 /* We only need to check the rcv_saddr if this tb was once marked
441 * without fastreuseport and then was reset, as we can only know that
442 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
443 * owners list.
444 */
445 if (tb->fastreuseport == FASTREUSEPORT_ANY)
446 return 1;
447 #if IS_ENABLED(CONFIG_IPV6)
448 if (tb->fast_sk_family == AF_INET6)
449 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
450 inet6_rcv_saddr(sk),
451 tb->fast_rcv_saddr,
452 sk->sk_rcv_saddr,
453 tb->fast_ipv6_only,
454 ipv6_only_sock(sk), true, false);
455 #endif
456 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
457 ipv6_only_sock(sk), true, false);
458 }
459
460 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
461 struct sock *sk)
462 {
463 kuid_t uid = sock_i_uid(sk);
464 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
465
466 if (hlist_empty(&tb->bhash2)) {
467 tb->fastreuse = reuse;
468 if (sk->sk_reuseport) {
469 tb->fastreuseport = FASTREUSEPORT_ANY;
470 tb->fastuid = uid;
471 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
472 tb->fast_ipv6_only = ipv6_only_sock(sk);
473 tb->fast_sk_family = sk->sk_family;
474 #if IS_ENABLED(CONFIG_IPV6)
475 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
476 #endif
477 } else {
478 tb->fastreuseport = 0;
479 }
480 } else {
481 if (!reuse)
482 tb->fastreuse = 0;
483 if (sk->sk_reuseport) {
484 /* We didn't match or we don't have fastreuseport set on
485 * the tb, but we have sk_reuseport set on this socket
486 * and we know that there are no bind conflicts with
487 * this socket in this tb, so reset our tb's reuseport
488 * settings so that any subsequent sockets that match
489 * our current socket will be put on the fast path.
490 *
491 * If we reset we need to set FASTREUSEPORT_STRICT so we
492 * do extra checking for all subsequent sk_reuseport
493 * socks.
494 */
495 if (!sk_reuseport_match(tb, sk)) {
496 tb->fastreuseport = FASTREUSEPORT_STRICT;
497 tb->fastuid = uid;
498 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
499 tb->fast_ipv6_only = ipv6_only_sock(sk);
500 tb->fast_sk_family = sk->sk_family;
501 #if IS_ENABLED(CONFIG_IPV6)
502 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
503 #endif
504 }
505 } else {
506 tb->fastreuseport = 0;
507 }
508 }
509 }
510
511 /* Obtain a reference to a local port for the given sock,
512 * if snum is zero it means select any available local port.
513 * We try to allocate an odd port (and leave even ports for connect())
514 */
515 int inet_csk_get_port(struct sock *sk, unsigned short snum)
516 {
517 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
518 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
519 bool found_port = false, check_bind_conflict = true;
520 bool bhash_created = false, bhash2_created = false;
521 int ret = -EADDRINUSE, port = snum, l3mdev;
522 struct inet_bind_hashbucket *head, *head2;
523 struct inet_bind2_bucket *tb2 = NULL;
524 struct inet_bind_bucket *tb = NULL;
525 bool head2_lock_acquired = false;
526 struct net *net = sock_net(sk);
527
528 l3mdev = inet_sk_bound_l3mdev(sk);
529
530 if (!port) {
531 head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
532 if (!head)
533 return ret;
534
535 head2_lock_acquired = true;
536
537 if (tb && tb2)
538 goto success;
539 found_port = true;
540 } else {
541 head = &hinfo->bhash[inet_bhashfn(net, port,
542 hinfo->bhash_size)];
543 spin_lock_bh(&head->lock);
544 inet_bind_bucket_for_each(tb, &head->chain)
545 if (inet_bind_bucket_match(tb, net, port, l3mdev))
546 break;
547 }
548
549 if (!tb) {
550 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
551 head, port, l3mdev);
552 if (!tb)
553 goto fail_unlock;
554 bhash_created = true;
555 }
556
557 if (!found_port) {
558 if (!hlist_empty(&tb->bhash2)) {
559 if (sk->sk_reuse == SK_FORCE_REUSE ||
560 (tb->fastreuse > 0 && reuse) ||
561 sk_reuseport_match(tb, sk))
562 check_bind_conflict = false;
563 }
564
565 if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) {
566 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, true, true))
567 goto fail_unlock;
568 }
569
570 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
571 spin_lock(&head2->lock);
572 head2_lock_acquired = true;
573 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
574 }
575
576 if (!tb2) {
577 tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
578 net, head2, tb, sk);
579 if (!tb2)
580 goto fail_unlock;
581 bhash2_created = true;
582 }
583
584 if (!found_port && check_bind_conflict) {
585 if (inet_csk_bind_conflict(sk, tb, tb2, true, true))
586 goto fail_unlock;
587 }
588
589 success:
590 inet_csk_update_fastreuse(tb, sk);
591
592 if (!inet_csk(sk)->icsk_bind_hash)
593 inet_bind_hash(sk, tb, tb2, port);
594 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
595 WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
596 ret = 0;
597
598 fail_unlock:
599 if (ret) {
600 if (bhash2_created)
601 inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep, tb2);
602 if (bhash_created)
603 inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
604 }
605 if (head2_lock_acquired)
606 spin_unlock(&head2->lock);
607 spin_unlock_bh(&head->lock);
608 return ret;
609 }
610 EXPORT_SYMBOL_GPL(inet_csk_get_port);
611
612 /*
613 * Wait for an incoming connection, avoid race conditions. This must be called
614 * with the socket locked.
615 */
616 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
617 {
618 struct inet_connection_sock *icsk = inet_csk(sk);
619 DEFINE_WAIT(wait);
620 int err;
621
622 /*
623 * True wake-one mechanism for incoming connections: only
624 * one process gets woken up, not the 'whole herd'.
625 * Since we do not 'race & poll' for established sockets
626 * anymore, the common case will execute the loop only once.
627 *
628 * Subtle issue: "add_wait_queue_exclusive()" will be added
629 * after any current non-exclusive waiters, and we know that
630 * it will always _stay_ after any new non-exclusive waiters
631 * because all non-exclusive waiters are added at the
632 * beginning of the wait-queue. As such, it's ok to "drop"
633 * our exclusiveness temporarily when we get woken up without
634 * having to remove and re-insert us on the wait queue.
635 */
636 for (;;) {
637 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
638 TASK_INTERRUPTIBLE);
639 release_sock(sk);
640 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
641 timeo = schedule_timeout(timeo);
642 sched_annotate_sleep();
643 lock_sock(sk);
644 err = 0;
645 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
646 break;
647 err = -EINVAL;
648 if (sk->sk_state != TCP_LISTEN)
649 break;
650 err = sock_intr_errno(timeo);
651 if (signal_pending(current))
652 break;
653 err = -EAGAIN;
654 if (!timeo)
655 break;
656 }
657 finish_wait(sk_sleep(sk), &wait);
658 return err;
659 }
660
661 /*
662 * This will accept the next outstanding connection.
663 */
664 struct sock *inet_csk_accept(struct sock *sk, struct proto_accept_arg *arg)
665 {
666 struct inet_connection_sock *icsk = inet_csk(sk);
667 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
668 struct request_sock *req;
669 struct sock *newsk;
670 int error;
671
672 lock_sock(sk);
673
674 /* We need to make sure that this socket is listening,
675 * and that it has something pending.
676 */
677 error = -EINVAL;
678 if (sk->sk_state != TCP_LISTEN)
679 goto out_err;
680
681 /* Find already established connection */
682 if (reqsk_queue_empty(queue)) {
683 long timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK);
684
685 /* If this is a non blocking socket don't sleep */
686 error = -EAGAIN;
687 if (!timeo)
688 goto out_err;
689
690 error = inet_csk_wait_for_connect(sk, timeo);
691 if (error)
692 goto out_err;
693 }
694 req = reqsk_queue_remove(queue, sk);
695 arg->is_empty = reqsk_queue_empty(queue);
696 newsk = req->sk;
697
698 if (sk->sk_protocol == IPPROTO_TCP &&
699 tcp_rsk(req)->tfo_listener) {
700 spin_lock_bh(&queue->fastopenq.lock);
701 if (tcp_rsk(req)->tfo_listener) {
702 /* We are still waiting for the final ACK from 3WHS
703 * so can't free req now. Instead, we set req->sk to
704 * NULL to signify that the child socket is taken
705 * so reqsk_fastopen_remove() will free the req
706 * when 3WHS finishes (or is aborted).
707 */
708 req->sk = NULL;
709 req = NULL;
710 }
711 spin_unlock_bh(&queue->fastopenq.lock);
712 }
713
714 out:
715 release_sock(sk);
716 if (newsk && mem_cgroup_sockets_enabled) {
717 int amt = 0;
718
719 /* atomically get the memory usage, set and charge the
720 * newsk->sk_memcg.
721 */
722 lock_sock(newsk);
723
724 mem_cgroup_sk_alloc(newsk);
725 if (newsk->sk_memcg) {
726 /* The socket has not been accepted yet, no need
727 * to look at newsk->sk_wmem_queued.
728 */
729 amt = sk_mem_pages(newsk->sk_forward_alloc +
730 atomic_read(&newsk->sk_rmem_alloc));
731 }
732
733 if (amt)
734 mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
735 GFP_KERNEL | __GFP_NOFAIL);
736
737 release_sock(newsk);
738 }
739 if (req)
740 reqsk_put(req);
741
742 if (newsk)
743 inet_init_csk_locks(newsk);
744
745 return newsk;
746 out_err:
747 newsk = NULL;
748 req = NULL;
749 arg->err = error;
750 goto out;
751 }
752 EXPORT_SYMBOL(inet_csk_accept);
753
754 /*
755 * Using different timers for retransmit, delayed acks and probes
756 * We may wish use just one timer maintaining a list of expire jiffies
757 * to optimize.
758 */
759 void inet_csk_init_xmit_timers(struct sock *sk,
760 void (*retransmit_handler)(struct timer_list *t),
761 void (*delack_handler)(struct timer_list *t),
762 void (*keepalive_handler)(struct timer_list *t))
763 {
764 struct inet_connection_sock *icsk = inet_csk(sk);
765
766 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
767 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
768 timer_setup(&sk->sk_timer, keepalive_handler, 0);
769 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
770 }
771 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
772
773 void inet_csk_clear_xmit_timers(struct sock *sk)
774 {
775 struct inet_connection_sock *icsk = inet_csk(sk);
776
777 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
778
779 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
780 sk_stop_timer(sk, &icsk->icsk_delack_timer);
781 sk_stop_timer(sk, &sk->sk_timer);
782 }
783 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
784
785 void inet_csk_clear_xmit_timers_sync(struct sock *sk)
786 {
787 struct inet_connection_sock *icsk = inet_csk(sk);
788
789 /* ongoing timer handlers need to acquire socket lock. */
790 sock_not_owned_by_me(sk);
791
792 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
793
794 sk_stop_timer_sync(sk, &icsk->icsk_retransmit_timer);
795 sk_stop_timer_sync(sk, &icsk->icsk_delack_timer);
796 sk_stop_timer_sync(sk, &sk->sk_timer);
797 }
798
799 void inet_csk_delete_keepalive_timer(struct sock *sk)
800 {
801 sk_stop_timer(sk, &sk->sk_timer);
802 }
803 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
804
805 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
806 {
807 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
808 }
809 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
810
811 struct dst_entry *inet_csk_route_req(const struct sock *sk,
812 struct flowi4 *fl4,
813 const struct request_sock *req)
814 {
815 const struct inet_request_sock *ireq = inet_rsk(req);
816 struct net *net = read_pnet(&ireq->ireq_net);
817 struct ip_options_rcu *opt;
818 struct rtable *rt;
819
820 rcu_read_lock();
821 opt = rcu_dereference(ireq->ireq_opt);
822
823 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
824 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
825 sk->sk_protocol, inet_sk_flowi_flags(sk),
826 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
827 ireq->ir_loc_addr, ireq->ir_rmt_port,
828 htons(ireq->ir_num), sk->sk_uid);
829 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
830 rt = ip_route_output_flow(net, fl4, sk);
831 if (IS_ERR(rt))
832 goto no_route;
833 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
834 goto route_err;
835 rcu_read_unlock();
836 return &rt->dst;
837
838 route_err:
839 ip_rt_put(rt);
840 no_route:
841 rcu_read_unlock();
842 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
843 return NULL;
844 }
845 EXPORT_SYMBOL_GPL(inet_csk_route_req);
846
847 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
848 struct sock *newsk,
849 const struct request_sock *req)
850 {
851 const struct inet_request_sock *ireq = inet_rsk(req);
852 struct net *net = read_pnet(&ireq->ireq_net);
853 struct inet_sock *newinet = inet_sk(newsk);
854 struct ip_options_rcu *opt;
855 struct flowi4 *fl4;
856 struct rtable *rt;
857
858 opt = rcu_dereference(ireq->ireq_opt);
859 fl4 = &newinet->cork.fl.u.ip4;
860
861 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
862 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
863 sk->sk_protocol, inet_sk_flowi_flags(sk),
864 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
865 ireq->ir_loc_addr, ireq->ir_rmt_port,
866 htons(ireq->ir_num), sk->sk_uid);
867 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
868 rt = ip_route_output_flow(net, fl4, sk);
869 if (IS_ERR(rt))
870 goto no_route;
871 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
872 goto route_err;
873 return &rt->dst;
874
875 route_err:
876 ip_rt_put(rt);
877 no_route:
878 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
879 return NULL;
880 }
881 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
882
883 /* Decide when to expire the request and when to resend SYN-ACK */
884 static void syn_ack_recalc(struct request_sock *req,
885 const int max_syn_ack_retries,
886 const u8 rskq_defer_accept,
887 int *expire, int *resend)
888 {
889 if (!rskq_defer_accept) {
890 *expire = req->num_timeout >= max_syn_ack_retries;
891 *resend = 1;
892 return;
893 }
894 *expire = req->num_timeout >= max_syn_ack_retries &&
895 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
896 /* Do not resend while waiting for data after ACK,
897 * start to resend on end of deferring period to give
898 * last chance for data or ACK to create established socket.
899 */
900 *resend = !inet_rsk(req)->acked ||
901 req->num_timeout >= rskq_defer_accept - 1;
902 }
903
904 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
905 {
906 int err = req->rsk_ops->rtx_syn_ack(parent, req);
907
908 if (!err)
909 req->num_retrans++;
910 return err;
911 }
912 EXPORT_SYMBOL(inet_rtx_syn_ack);
913
914 static struct request_sock *
915 reqsk_alloc_noprof(const struct request_sock_ops *ops, struct sock *sk_listener,
916 bool attach_listener)
917 {
918 struct request_sock *req;
919
920 req = kmem_cache_alloc_noprof(ops->slab, GFP_ATOMIC | __GFP_NOWARN);
921 if (!req)
922 return NULL;
923 req->rsk_listener = NULL;
924 if (attach_listener) {
925 if (unlikely(!refcount_inc_not_zero(&sk_listener->sk_refcnt))) {
926 kmem_cache_free(ops->slab, req);
927 return NULL;
928 }
929 req->rsk_listener = sk_listener;
930 }
931 req->rsk_ops = ops;
932 req_to_sk(req)->sk_prot = sk_listener->sk_prot;
933 sk_node_init(&req_to_sk(req)->sk_node);
934 sk_tx_queue_clear(req_to_sk(req));
935 req->saved_syn = NULL;
936 req->syncookie = 0;
937 req->timeout = 0;
938 req->num_timeout = 0;
939 req->num_retrans = 0;
940 req->sk = NULL;
941 refcount_set(&req->rsk_refcnt, 0);
942
943 return req;
944 }
945 #define reqsk_alloc(...) alloc_hooks(reqsk_alloc_noprof(__VA_ARGS__))
946
947 struct request_sock *inet_reqsk_alloc(const struct request_sock_ops *ops,
948 struct sock *sk_listener,
949 bool attach_listener)
950 {
951 struct request_sock *req = reqsk_alloc(ops, sk_listener,
952 attach_listener);
953
954 if (req) {
955 struct inet_request_sock *ireq = inet_rsk(req);
956
957 ireq->ireq_opt = NULL;
958 #if IS_ENABLED(CONFIG_IPV6)
959 ireq->pktopts = NULL;
960 #endif
961 atomic64_set(&ireq->ir_cookie, 0);
962 ireq->ireq_state = TCP_NEW_SYN_RECV;
963 write_pnet(&ireq->ireq_net, sock_net(sk_listener));
964 ireq->ireq_family = sk_listener->sk_family;
965 req->timeout = TCP_TIMEOUT_INIT;
966 }
967
968 return req;
969 }
970 EXPORT_SYMBOL(inet_reqsk_alloc);
971
972 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
973 struct sock *sk)
974 {
975 struct sock *req_sk, *nreq_sk;
976 struct request_sock *nreq;
977
978 nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
979 if (!nreq) {
980 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
981
982 /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
983 sock_put(sk);
984 return NULL;
985 }
986
987 req_sk = req_to_sk(req);
988 nreq_sk = req_to_sk(nreq);
989
990 memcpy(nreq_sk, req_sk,
991 offsetof(struct sock, sk_dontcopy_begin));
992 unsafe_memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
993 req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end),
994 /* alloc is larger than struct, see above */);
995
996 sk_node_init(&nreq_sk->sk_node);
997 nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
998 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
999 nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
1000 #endif
1001 nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
1002
1003 nreq->rsk_listener = sk;
1004
1005 /* We need not acquire fastopenq->lock
1006 * because the child socket is locked in inet_csk_listen_stop().
1007 */
1008 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
1009 rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
1010
1011 return nreq;
1012 }
1013
1014 static void reqsk_queue_migrated(struct request_sock_queue *queue,
1015 const struct request_sock *req)
1016 {
1017 if (req->num_timeout == 0)
1018 atomic_inc(&queue->young);
1019 atomic_inc(&queue->qlen);
1020 }
1021
1022 static void reqsk_migrate_reset(struct request_sock *req)
1023 {
1024 req->saved_syn = NULL;
1025 #if IS_ENABLED(CONFIG_IPV6)
1026 inet_rsk(req)->ipv6_opt = NULL;
1027 inet_rsk(req)->pktopts = NULL;
1028 #else
1029 inet_rsk(req)->ireq_opt = NULL;
1030 #endif
1031 }
1032
1033 /* return true if req was found in the ehash table */
1034 static bool reqsk_queue_unlink(struct request_sock *req)
1035 {
1036 struct sock *sk = req_to_sk(req);
1037 bool found = false;
1038
1039 if (sk_hashed(sk)) {
1040 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
1041 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
1042
1043 spin_lock(lock);
1044 found = __sk_nulls_del_node_init_rcu(sk);
1045 spin_unlock(lock);
1046 }
1047 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
1048 reqsk_put(req);
1049 return found;
1050 }
1051
1052 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
1053 {
1054 bool unlinked = reqsk_queue_unlink(req);
1055
1056 if (unlinked) {
1057 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
1058 reqsk_put(req);
1059 }
1060 return unlinked;
1061 }
1062 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
1063
1064 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
1065 {
1066 inet_csk_reqsk_queue_drop(sk, req);
1067 reqsk_put(req);
1068 }
1069 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
1070
1071 static void reqsk_timer_handler(struct timer_list *t)
1072 {
1073 struct request_sock *req = from_timer(req, t, rsk_timer);
1074 struct request_sock *nreq = NULL, *oreq = req;
1075 struct sock *sk_listener = req->rsk_listener;
1076 struct inet_connection_sock *icsk;
1077 struct request_sock_queue *queue;
1078 struct net *net;
1079 int max_syn_ack_retries, qlen, expire = 0, resend = 0;
1080
1081 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
1082 struct sock *nsk;
1083
1084 nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
1085 if (!nsk)
1086 goto drop;
1087
1088 nreq = inet_reqsk_clone(req, nsk);
1089 if (!nreq)
1090 goto drop;
1091
1092 /* The new timer for the cloned req can decrease the 2
1093 * by calling inet_csk_reqsk_queue_drop_and_put(), so
1094 * hold another count to prevent use-after-free and
1095 * call reqsk_put() just before return.
1096 */
1097 refcount_set(&nreq->rsk_refcnt, 2 + 1);
1098 timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1099 reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
1100
1101 req = nreq;
1102 sk_listener = nsk;
1103 }
1104
1105 icsk = inet_csk(sk_listener);
1106 net = sock_net(sk_listener);
1107 max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? :
1108 READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1109 /* Normally all the openreqs are young and become mature
1110 * (i.e. converted to established socket) for first timeout.
1111 * If synack was not acknowledged for 1 second, it means
1112 * one of the following things: synack was lost, ack was lost,
1113 * rtt is high or nobody planned to ack (i.e. synflood).
1114 * When server is a bit loaded, queue is populated with old
1115 * open requests, reducing effective size of queue.
1116 * When server is well loaded, queue size reduces to zero
1117 * after several minutes of work. It is not synflood,
1118 * it is normal operation. The solution is pruning
1119 * too old entries overriding normal timeout, when
1120 * situation becomes dangerous.
1121 *
1122 * Essentially, we reserve half of room for young
1123 * embrions; and abort old ones without pity, if old
1124 * ones are about to clog our table.
1125 */
1126 queue = &icsk->icsk_accept_queue;
1127 qlen = reqsk_queue_len(queue);
1128 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1129 int young = reqsk_queue_len_young(queue) << 1;
1130
1131 while (max_syn_ack_retries > 2) {
1132 if (qlen < young)
1133 break;
1134 max_syn_ack_retries--;
1135 young <<= 1;
1136 }
1137 }
1138 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
1139 &expire, &resend);
1140 req->rsk_ops->syn_ack_timeout(req);
1141 if (!expire &&
1142 (!resend ||
1143 !inet_rtx_syn_ack(sk_listener, req) ||
1144 inet_rsk(req)->acked)) {
1145 if (req->num_timeout++ == 0)
1146 atomic_dec(&queue->young);
1147 mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
1148
1149 if (!nreq)
1150 return;
1151
1152 if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
1153 /* delete timer */
1154 inet_csk_reqsk_queue_drop(sk_listener, nreq);
1155 goto no_ownership;
1156 }
1157
1158 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1159 reqsk_migrate_reset(oreq);
1160 reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1161 reqsk_put(oreq);
1162
1163 reqsk_put(nreq);
1164 return;
1165 }
1166
1167 /* Even if we can clone the req, we may need not retransmit any more
1168 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1169 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1170 */
1171 if (nreq) {
1172 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1173 no_ownership:
1174 reqsk_migrate_reset(nreq);
1175 reqsk_queue_removed(queue, nreq);
1176 __reqsk_free(nreq);
1177 }
1178
1179 drop:
1180 inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
1181 }
1182
1183 static bool reqsk_queue_hash_req(struct request_sock *req,
1184 unsigned long timeout)
1185 {
1186 bool found_dup_sk = false;
1187
1188 if (!inet_ehash_insert(req_to_sk(req), NULL, &found_dup_sk))
1189 return false;
1190
1191 /* The timer needs to be setup after a successful insertion. */
1192 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1193 mod_timer(&req->rsk_timer, jiffies + timeout);
1194
1195 /* before letting lookups find us, make sure all req fields
1196 * are committed to memory and refcnt initialized.
1197 */
1198 smp_wmb();
1199 refcount_set(&req->rsk_refcnt, 2 + 1);
1200 return true;
1201 }
1202
1203 bool inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1204 unsigned long timeout)
1205 {
1206 if (!reqsk_queue_hash_req(req, timeout))
1207 return false;
1208
1209 inet_csk_reqsk_queue_added(sk);
1210 return true;
1211 }
1212 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1213
1214 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1215 const gfp_t priority)
1216 {
1217 struct inet_connection_sock *icsk = inet_csk(newsk);
1218
1219 if (!icsk->icsk_ulp_ops)
1220 return;
1221
1222 icsk->icsk_ulp_ops->clone(req, newsk, priority);
1223 }
1224
1225 /**
1226 * inet_csk_clone_lock - clone an inet socket, and lock its clone
1227 * @sk: the socket to clone
1228 * @req: request_sock
1229 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1230 *
1231 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1232 */
1233 struct sock *inet_csk_clone_lock(const struct sock *sk,
1234 const struct request_sock *req,
1235 const gfp_t priority)
1236 {
1237 struct sock *newsk = sk_clone_lock(sk, priority);
1238
1239 if (newsk) {
1240 struct inet_connection_sock *newicsk = inet_csk(newsk);
1241
1242 inet_sk_set_state(newsk, TCP_SYN_RECV);
1243 newicsk->icsk_bind_hash = NULL;
1244 newicsk->icsk_bind2_hash = NULL;
1245
1246 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
1247 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
1248 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
1249
1250 /* listeners have SOCK_RCU_FREE, not the children */
1251 sock_reset_flag(newsk, SOCK_RCU_FREE);
1252
1253 inet_sk(newsk)->mc_list = NULL;
1254
1255 newsk->sk_mark = inet_rsk(req)->ir_mark;
1256 atomic64_set(&newsk->sk_cookie,
1257 atomic64_read(&inet_rsk(req)->ir_cookie));
1258
1259 newicsk->icsk_retransmits = 0;
1260 newicsk->icsk_backoff = 0;
1261 newicsk->icsk_probes_out = 0;
1262 newicsk->icsk_probes_tstamp = 0;
1263
1264 /* Deinitialize accept_queue to trap illegal accesses. */
1265 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
1266
1267 inet_clone_ulp(req, newsk, priority);
1268
1269 security_inet_csk_clone(newsk, req);
1270 }
1271 return newsk;
1272 }
1273 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
1274
1275 /*
1276 * At this point, there should be no process reference to this
1277 * socket, and thus no user references at all. Therefore we
1278 * can assume the socket waitqueue is inactive and nobody will
1279 * try to jump onto it.
1280 */
1281 void inet_csk_destroy_sock(struct sock *sk)
1282 {
1283 WARN_ON(sk->sk_state != TCP_CLOSE);
1284 WARN_ON(!sock_flag(sk, SOCK_DEAD));
1285
1286 /* It cannot be in hash table! */
1287 WARN_ON(!sk_unhashed(sk));
1288
1289 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1290 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1291
1292 sk->sk_prot->destroy(sk);
1293
1294 sk_stream_kill_queues(sk);
1295
1296 xfrm_sk_free_policy(sk);
1297
1298 this_cpu_dec(*sk->sk_prot->orphan_count);
1299
1300 sock_put(sk);
1301 }
1302 EXPORT_SYMBOL(inet_csk_destroy_sock);
1303
1304 /* This function allows to force a closure of a socket after the call to
1305 * tcp/dccp_create_openreq_child().
1306 */
1307 void inet_csk_prepare_forced_close(struct sock *sk)
1308 __releases(&sk->sk_lock.slock)
1309 {
1310 /* sk_clone_lock locked the socket and set refcnt to 2 */
1311 bh_unlock_sock(sk);
1312 sock_put(sk);
1313 inet_csk_prepare_for_destroy_sock(sk);
1314 inet_sk(sk)->inet_num = 0;
1315 }
1316 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1317
1318 static int inet_ulp_can_listen(const struct sock *sk)
1319 {
1320 const struct inet_connection_sock *icsk = inet_csk(sk);
1321
1322 if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone)
1323 return -EINVAL;
1324
1325 return 0;
1326 }
1327
1328 int inet_csk_listen_start(struct sock *sk)
1329 {
1330 struct inet_connection_sock *icsk = inet_csk(sk);
1331 struct inet_sock *inet = inet_sk(sk);
1332 int err;
1333
1334 err = inet_ulp_can_listen(sk);
1335 if (unlikely(err))
1336 return err;
1337
1338 reqsk_queue_alloc(&icsk->icsk_accept_queue);
1339
1340 sk->sk_ack_backlog = 0;
1341 inet_csk_delack_init(sk);
1342
1343 /* There is race window here: we announce ourselves listening,
1344 * but this transition is still not validated by get_port().
1345 * It is OK, because this socket enters to hash table only
1346 * after validation is complete.
1347 */
1348 inet_sk_state_store(sk, TCP_LISTEN);
1349 err = sk->sk_prot->get_port(sk, inet->inet_num);
1350 if (!err) {
1351 inet->inet_sport = htons(inet->inet_num);
1352
1353 sk_dst_reset(sk);
1354 err = sk->sk_prot->hash(sk);
1355
1356 if (likely(!err))
1357 return 0;
1358 }
1359
1360 inet_sk_set_state(sk, TCP_CLOSE);
1361 return err;
1362 }
1363 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1364
1365 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1366 struct sock *child)
1367 {
1368 sk->sk_prot->disconnect(child, O_NONBLOCK);
1369
1370 sock_orphan(child);
1371
1372 this_cpu_inc(*sk->sk_prot->orphan_count);
1373
1374 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1375 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1376 BUG_ON(sk != req->rsk_listener);
1377
1378 /* Paranoid, to prevent race condition if
1379 * an inbound pkt destined for child is
1380 * blocked by sock lock in tcp_v4_rcv().
1381 * Also to satisfy an assertion in
1382 * tcp_v4_destroy_sock().
1383 */
1384 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1385 }
1386 inet_csk_destroy_sock(child);
1387 }
1388
1389 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1390 struct request_sock *req,
1391 struct sock *child)
1392 {
1393 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1394
1395 spin_lock(&queue->rskq_lock);
1396 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1397 inet_child_forget(sk, req, child);
1398 child = NULL;
1399 } else {
1400 req->sk = child;
1401 req->dl_next = NULL;
1402 if (queue->rskq_accept_head == NULL)
1403 WRITE_ONCE(queue->rskq_accept_head, req);
1404 else
1405 queue->rskq_accept_tail->dl_next = req;
1406 queue->rskq_accept_tail = req;
1407 sk_acceptq_added(sk);
1408 }
1409 spin_unlock(&queue->rskq_lock);
1410 return child;
1411 }
1412 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1413
1414 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1415 struct request_sock *req, bool own_req)
1416 {
1417 if (own_req) {
1418 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1419 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1420
1421 if (sk != req->rsk_listener) {
1422 /* another listening sk has been selected,
1423 * migrate the req to it.
1424 */
1425 struct request_sock *nreq;
1426
1427 /* hold a refcnt for the nreq->rsk_listener
1428 * which is assigned in inet_reqsk_clone()
1429 */
1430 sock_hold(sk);
1431 nreq = inet_reqsk_clone(req, sk);
1432 if (!nreq) {
1433 inet_child_forget(sk, req, child);
1434 goto child_put;
1435 }
1436
1437 refcount_set(&nreq->rsk_refcnt, 1);
1438 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1439 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1440 reqsk_migrate_reset(req);
1441 reqsk_put(req);
1442 return child;
1443 }
1444
1445 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1446 reqsk_migrate_reset(nreq);
1447 __reqsk_free(nreq);
1448 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1449 return child;
1450 }
1451 }
1452 /* Too bad, another child took ownership of the request, undo. */
1453 child_put:
1454 bh_unlock_sock(child);
1455 sock_put(child);
1456 return NULL;
1457 }
1458 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1459
1460 /*
1461 * This routine closes sockets which have been at least partially
1462 * opened, but not yet accepted.
1463 */
1464 void inet_csk_listen_stop(struct sock *sk)
1465 {
1466 struct inet_connection_sock *icsk = inet_csk(sk);
1467 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1468 struct request_sock *next, *req;
1469
1470 /* Following specs, it would be better either to send FIN
1471 * (and enter FIN-WAIT-1, it is normal close)
1472 * or to send active reset (abort).
1473 * Certainly, it is pretty dangerous while synflood, but it is
1474 * bad justification for our negligence 8)
1475 * To be honest, we are not able to make either
1476 * of the variants now. --ANK
1477 */
1478 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1479 struct sock *child = req->sk, *nsk;
1480 struct request_sock *nreq;
1481
1482 local_bh_disable();
1483 bh_lock_sock(child);
1484 WARN_ON(sock_owned_by_user(child));
1485 sock_hold(child);
1486
1487 nsk = reuseport_migrate_sock(sk, child, NULL);
1488 if (nsk) {
1489 nreq = inet_reqsk_clone(req, nsk);
1490 if (nreq) {
1491 refcount_set(&nreq->rsk_refcnt, 1);
1492
1493 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1494 __NET_INC_STATS(sock_net(nsk),
1495 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1496 reqsk_migrate_reset(req);
1497 } else {
1498 __NET_INC_STATS(sock_net(nsk),
1499 LINUX_MIB_TCPMIGRATEREQFAILURE);
1500 reqsk_migrate_reset(nreq);
1501 __reqsk_free(nreq);
1502 }
1503
1504 /* inet_csk_reqsk_queue_add() has already
1505 * called inet_child_forget() on failure case.
1506 */
1507 goto skip_child_forget;
1508 }
1509 }
1510
1511 inet_child_forget(sk, req, child);
1512 skip_child_forget:
1513 reqsk_put(req);
1514 bh_unlock_sock(child);
1515 local_bh_enable();
1516 sock_put(child);
1517
1518 cond_resched();
1519 }
1520 if (queue->fastopenq.rskq_rst_head) {
1521 /* Free all the reqs queued in rskq_rst_head. */
1522 spin_lock_bh(&queue->fastopenq.lock);
1523 req = queue->fastopenq.rskq_rst_head;
1524 queue->fastopenq.rskq_rst_head = NULL;
1525 spin_unlock_bh(&queue->fastopenq.lock);
1526 while (req != NULL) {
1527 next = req->dl_next;
1528 reqsk_put(req);
1529 req = next;
1530 }
1531 }
1532 WARN_ON_ONCE(sk->sk_ack_backlog);
1533 }
1534 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1535
1536 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1537 {
1538 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1539 const struct inet_sock *inet = inet_sk(sk);
1540
1541 sin->sin_family = AF_INET;
1542 sin->sin_addr.s_addr = inet->inet_daddr;
1543 sin->sin_port = inet->inet_dport;
1544 }
1545 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1546
1547 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1548 {
1549 const struct inet_sock *inet = inet_sk(sk);
1550 const struct ip_options_rcu *inet_opt;
1551 __be32 daddr = inet->inet_daddr;
1552 struct flowi4 *fl4;
1553 struct rtable *rt;
1554
1555 rcu_read_lock();
1556 inet_opt = rcu_dereference(inet->inet_opt);
1557 if (inet_opt && inet_opt->opt.srr)
1558 daddr = inet_opt->opt.faddr;
1559 fl4 = &fl->u.ip4;
1560 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1561 inet->inet_saddr, inet->inet_dport,
1562 inet->inet_sport, sk->sk_protocol,
1563 ip_sock_rt_tos(sk), sk->sk_bound_dev_if);
1564 if (IS_ERR(rt))
1565 rt = NULL;
1566 if (rt)
1567 sk_setup_caps(sk, &rt->dst);
1568 rcu_read_unlock();
1569
1570 return &rt->dst;
1571 }
1572
1573 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1574 {
1575 struct dst_entry *dst = __sk_dst_check(sk, 0);
1576 struct inet_sock *inet = inet_sk(sk);
1577
1578 if (!dst) {
1579 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1580 if (!dst)
1581 goto out;
1582 }
1583 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1584
1585 dst = __sk_dst_check(sk, 0);
1586 if (!dst)
1587 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1588 out:
1589 return dst;
1590 }
1591 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
1592
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.