Loading...
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Support for INET connection oriented protocols.
7 *
8 * Authors: See the TCP sources
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or(at your option) any later version.
14 */
15
16#include <linux/module.h>
17#include <linux/jhash.h>
18
19#include <net/inet_connection_sock.h>
20#include <net/inet_hashtables.h>
21#include <net/inet_timewait_sock.h>
22#include <net/ip.h>
23#include <net/route.h>
24#include <net/tcp_states.h>
25#include <net/xfrm.h>
26#include <net/tcp.h>
27#include <net/sock_reuseport.h>
28#include <net/addrconf.h>
29
30#ifdef INET_CSK_DEBUG
31const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
32EXPORT_SYMBOL(inet_csk_timer_bug_msg);
33#endif
34
35#if IS_ENABLED(CONFIG_IPV6)
36/* match_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
37 * only, and any IPv4 addresses if not IPv6 only
38 * match_wildcard == false: addresses must be exactly the same, i.e.
39 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
40 * and 0.0.0.0 equals to 0.0.0.0 only
41 */
42static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
43 const struct in6_addr *sk2_rcv_saddr6,
44 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
45 bool sk1_ipv6only, bool sk2_ipv6only,
46 bool match_wildcard)
47{
48 int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
49 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
50
51 /* if both are mapped, treat as IPv4 */
52 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
53 if (!sk2_ipv6only) {
54 if (sk1_rcv_saddr == sk2_rcv_saddr)
55 return true;
56 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
57 return match_wildcard;
58 }
59 return false;
60 }
61
62 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
63 return true;
64
65 if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
66 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
67 return true;
68
69 if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
70 !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
71 return true;
72
73 if (sk2_rcv_saddr6 &&
74 ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
75 return true;
76
77 return false;
78}
79#endif
80
81/* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
82 * match_wildcard == false: addresses must be exactly the same, i.e.
83 * 0.0.0.0 only equals to 0.0.0.0
84 */
85static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
86 bool sk2_ipv6only, bool match_wildcard)
87{
88 if (!sk2_ipv6only) {
89 if (sk1_rcv_saddr == sk2_rcv_saddr)
90 return true;
91 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
92 return match_wildcard;
93 }
94 return false;
95}
96
97bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
98 bool match_wildcard)
99{
100#if IS_ENABLED(CONFIG_IPV6)
101 if (sk->sk_family == AF_INET6)
102 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
103 inet6_rcv_saddr(sk2),
104 sk->sk_rcv_saddr,
105 sk2->sk_rcv_saddr,
106 ipv6_only_sock(sk),
107 ipv6_only_sock(sk2),
108 match_wildcard);
109#endif
110 return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
111 ipv6_only_sock(sk2), match_wildcard);
112}
113EXPORT_SYMBOL(inet_rcv_saddr_equal);
114
115void inet_get_local_port_range(struct net *net, int *low, int *high)
116{
117 unsigned int seq;
118
119 do {
120 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
121
122 *low = net->ipv4.ip_local_ports.range[0];
123 *high = net->ipv4.ip_local_ports.range[1];
124 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
125}
126EXPORT_SYMBOL(inet_get_local_port_range);
127
128static int inet_csk_bind_conflict(const struct sock *sk,
129 const struct inet_bind_bucket *tb,
130 bool relax, bool reuseport_ok)
131{
132 struct sock *sk2;
133 bool reuse = sk->sk_reuse;
134 bool reuseport = !!sk->sk_reuseport && reuseport_ok;
135 kuid_t uid = sock_i_uid((struct sock *)sk);
136
137 /*
138 * Unlike other sk lookup places we do not check
139 * for sk_net here, since _all_ the socks listed
140 * in tb->owners list belong to the same net - the
141 * one this bucket belongs to.
142 */
143
144 sk_for_each_bound(sk2, &tb->owners) {
145 if (sk != sk2 &&
146 (!sk->sk_bound_dev_if ||
147 !sk2->sk_bound_dev_if ||
148 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
149 if ((!reuse || !sk2->sk_reuse ||
150 sk2->sk_state == TCP_LISTEN) &&
151 (!reuseport || !sk2->sk_reuseport ||
152 rcu_access_pointer(sk->sk_reuseport_cb) ||
153 (sk2->sk_state != TCP_TIME_WAIT &&
154 !uid_eq(uid, sock_i_uid(sk2))))) {
155 if (inet_rcv_saddr_equal(sk, sk2, true))
156 break;
157 }
158 if (!relax && reuse && sk2->sk_reuse &&
159 sk2->sk_state != TCP_LISTEN) {
160 if (inet_rcv_saddr_equal(sk, sk2, true))
161 break;
162 }
163 }
164 }
165 return sk2 != NULL;
166}
167
168/*
169 * Find an open port number for the socket. Returns with the
170 * inet_bind_hashbucket lock held.
171 */
172static struct inet_bind_hashbucket *
173inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
174{
175 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
176 int port = 0;
177 struct inet_bind_hashbucket *head;
178 struct net *net = sock_net(sk);
179 int i, low, high, attempt_half;
180 struct inet_bind_bucket *tb;
181 u32 remaining, offset;
182
183 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
184other_half_scan:
185 inet_get_local_port_range(net, &low, &high);
186 high++; /* [32768, 60999] -> [32768, 61000[ */
187 if (high - low < 4)
188 attempt_half = 0;
189 if (attempt_half) {
190 int half = low + (((high - low) >> 2) << 1);
191
192 if (attempt_half == 1)
193 high = half;
194 else
195 low = half;
196 }
197 remaining = high - low;
198 if (likely(remaining > 1))
199 remaining &= ~1U;
200
201 offset = prandom_u32() % remaining;
202 /* __inet_hash_connect() favors ports having @low parity
203 * We do the opposite to not pollute connect() users.
204 */
205 offset |= 1U;
206
207other_parity_scan:
208 port = low + offset;
209 for (i = 0; i < remaining; i += 2, port += 2) {
210 if (unlikely(port >= high))
211 port -= remaining;
212 if (inet_is_local_reserved_port(net, port))
213 continue;
214 head = &hinfo->bhash[inet_bhashfn(net, port,
215 hinfo->bhash_size)];
216 spin_lock_bh(&head->lock);
217 inet_bind_bucket_for_each(tb, &head->chain)
218 if (net_eq(ib_net(tb), net) && tb->port == port) {
219 if (!inet_csk_bind_conflict(sk, tb, false, false))
220 goto success;
221 goto next_port;
222 }
223 tb = NULL;
224 goto success;
225next_port:
226 spin_unlock_bh(&head->lock);
227 cond_resched();
228 }
229
230 offset--;
231 if (!(offset & 1))
232 goto other_parity_scan;
233
234 if (attempt_half == 1) {
235 /* OK we now try the upper half of the range */
236 attempt_half = 2;
237 goto other_half_scan;
238 }
239 return NULL;
240success:
241 *port_ret = port;
242 *tb_ret = tb;
243 return head;
244}
245
246static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
247 struct sock *sk)
248{
249 kuid_t uid = sock_i_uid(sk);
250
251 if (tb->fastreuseport <= 0)
252 return 0;
253 if (!sk->sk_reuseport)
254 return 0;
255 if (rcu_access_pointer(sk->sk_reuseport_cb))
256 return 0;
257 if (!uid_eq(tb->fastuid, uid))
258 return 0;
259 /* We only need to check the rcv_saddr if this tb was once marked
260 * without fastreuseport and then was reset, as we can only know that
261 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
262 * owners list.
263 */
264 if (tb->fastreuseport == FASTREUSEPORT_ANY)
265 return 1;
266#if IS_ENABLED(CONFIG_IPV6)
267 if (tb->fast_sk_family == AF_INET6)
268 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
269 inet6_rcv_saddr(sk),
270 tb->fast_rcv_saddr,
271 sk->sk_rcv_saddr,
272 tb->fast_ipv6_only,
273 ipv6_only_sock(sk), true);
274#endif
275 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
276 ipv6_only_sock(sk), true);
277}
278
279/* Obtain a reference to a local port for the given sock,
280 * if snum is zero it means select any available local port.
281 * We try to allocate an odd port (and leave even ports for connect())
282 */
283int inet_csk_get_port(struct sock *sk, unsigned short snum)
284{
285 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
286 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
287 int ret = 1, port = snum;
288 struct inet_bind_hashbucket *head;
289 struct net *net = sock_net(sk);
290 struct inet_bind_bucket *tb = NULL;
291 kuid_t uid = sock_i_uid(sk);
292
293 if (!port) {
294 head = inet_csk_find_open_port(sk, &tb, &port);
295 if (!head)
296 return ret;
297 if (!tb)
298 goto tb_not_found;
299 goto success;
300 }
301 head = &hinfo->bhash[inet_bhashfn(net, port,
302 hinfo->bhash_size)];
303 spin_lock_bh(&head->lock);
304 inet_bind_bucket_for_each(tb, &head->chain)
305 if (net_eq(ib_net(tb), net) && tb->port == port)
306 goto tb_found;
307tb_not_found:
308 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
309 net, head, port);
310 if (!tb)
311 goto fail_unlock;
312tb_found:
313 if (!hlist_empty(&tb->owners)) {
314 if (sk->sk_reuse == SK_FORCE_REUSE)
315 goto success;
316
317 if ((tb->fastreuse > 0 && reuse) ||
318 sk_reuseport_match(tb, sk))
319 goto success;
320 if (inet_csk_bind_conflict(sk, tb, true, true))
321 goto fail_unlock;
322 }
323success:
324 if (hlist_empty(&tb->owners)) {
325 tb->fastreuse = reuse;
326 if (sk->sk_reuseport) {
327 tb->fastreuseport = FASTREUSEPORT_ANY;
328 tb->fastuid = uid;
329 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
330 tb->fast_ipv6_only = ipv6_only_sock(sk);
331 tb->fast_sk_family = sk->sk_family;
332#if IS_ENABLED(CONFIG_IPV6)
333 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
334#endif
335 } else {
336 tb->fastreuseport = 0;
337 }
338 } else {
339 if (!reuse)
340 tb->fastreuse = 0;
341 if (sk->sk_reuseport) {
342 /* We didn't match or we don't have fastreuseport set on
343 * the tb, but we have sk_reuseport set on this socket
344 * and we know that there are no bind conflicts with
345 * this socket in this tb, so reset our tb's reuseport
346 * settings so that any subsequent sockets that match
347 * our current socket will be put on the fast path.
348 *
349 * If we reset we need to set FASTREUSEPORT_STRICT so we
350 * do extra checking for all subsequent sk_reuseport
351 * socks.
352 */
353 if (!sk_reuseport_match(tb, sk)) {
354 tb->fastreuseport = FASTREUSEPORT_STRICT;
355 tb->fastuid = uid;
356 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
357 tb->fast_ipv6_only = ipv6_only_sock(sk);
358 tb->fast_sk_family = sk->sk_family;
359#if IS_ENABLED(CONFIG_IPV6)
360 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
361#endif
362 }
363 } else {
364 tb->fastreuseport = 0;
365 }
366 }
367 if (!inet_csk(sk)->icsk_bind_hash)
368 inet_bind_hash(sk, tb, port);
369 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
370 ret = 0;
371
372fail_unlock:
373 spin_unlock_bh(&head->lock);
374 return ret;
375}
376EXPORT_SYMBOL_GPL(inet_csk_get_port);
377
378/*
379 * Wait for an incoming connection, avoid race conditions. This must be called
380 * with the socket locked.
381 */
382static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
383{
384 struct inet_connection_sock *icsk = inet_csk(sk);
385 DEFINE_WAIT(wait);
386 int err;
387
388 /*
389 * True wake-one mechanism for incoming connections: only
390 * one process gets woken up, not the 'whole herd'.
391 * Since we do not 'race & poll' for established sockets
392 * anymore, the common case will execute the loop only once.
393 *
394 * Subtle issue: "add_wait_queue_exclusive()" will be added
395 * after any current non-exclusive waiters, and we know that
396 * it will always _stay_ after any new non-exclusive waiters
397 * because all non-exclusive waiters are added at the
398 * beginning of the wait-queue. As such, it's ok to "drop"
399 * our exclusiveness temporarily when we get woken up without
400 * having to remove and re-insert us on the wait queue.
401 */
402 for (;;) {
403 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
404 TASK_INTERRUPTIBLE);
405 release_sock(sk);
406 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
407 timeo = schedule_timeout(timeo);
408 sched_annotate_sleep();
409 lock_sock(sk);
410 err = 0;
411 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
412 break;
413 err = -EINVAL;
414 if (sk->sk_state != TCP_LISTEN)
415 break;
416 err = sock_intr_errno(timeo);
417 if (signal_pending(current))
418 break;
419 err = -EAGAIN;
420 if (!timeo)
421 break;
422 }
423 finish_wait(sk_sleep(sk), &wait);
424 return err;
425}
426
427/*
428 * This will accept the next outstanding connection.
429 */
430struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
431{
432 struct inet_connection_sock *icsk = inet_csk(sk);
433 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
434 struct request_sock *req;
435 struct sock *newsk;
436 int error;
437
438 lock_sock(sk);
439
440 /* We need to make sure that this socket is listening,
441 * and that it has something pending.
442 */
443 error = -EINVAL;
444 if (sk->sk_state != TCP_LISTEN)
445 goto out_err;
446
447 /* Find already established connection */
448 if (reqsk_queue_empty(queue)) {
449 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
450
451 /* If this is a non blocking socket don't sleep */
452 error = -EAGAIN;
453 if (!timeo)
454 goto out_err;
455
456 error = inet_csk_wait_for_connect(sk, timeo);
457 if (error)
458 goto out_err;
459 }
460 req = reqsk_queue_remove(queue, sk);
461 newsk = req->sk;
462
463 if (sk->sk_protocol == IPPROTO_TCP &&
464 tcp_rsk(req)->tfo_listener) {
465 spin_lock_bh(&queue->fastopenq.lock);
466 if (tcp_rsk(req)->tfo_listener) {
467 /* We are still waiting for the final ACK from 3WHS
468 * so can't free req now. Instead, we set req->sk to
469 * NULL to signify that the child socket is taken
470 * so reqsk_fastopen_remove() will free the req
471 * when 3WHS finishes (or is aborted).
472 */
473 req->sk = NULL;
474 req = NULL;
475 }
476 spin_unlock_bh(&queue->fastopenq.lock);
477 }
478out:
479 release_sock(sk);
480 if (req)
481 reqsk_put(req);
482 return newsk;
483out_err:
484 newsk = NULL;
485 req = NULL;
486 *err = error;
487 goto out;
488}
489EXPORT_SYMBOL(inet_csk_accept);
490
491/*
492 * Using different timers for retransmit, delayed acks and probes
493 * We may wish use just one timer maintaining a list of expire jiffies
494 * to optimize.
495 */
496void inet_csk_init_xmit_timers(struct sock *sk,
497 void (*retransmit_handler)(struct timer_list *t),
498 void (*delack_handler)(struct timer_list *t),
499 void (*keepalive_handler)(struct timer_list *t))
500{
501 struct inet_connection_sock *icsk = inet_csk(sk);
502
503 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
504 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
505 timer_setup(&sk->sk_timer, keepalive_handler, 0);
506 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
507}
508EXPORT_SYMBOL(inet_csk_init_xmit_timers);
509
510void inet_csk_clear_xmit_timers(struct sock *sk)
511{
512 struct inet_connection_sock *icsk = inet_csk(sk);
513
514 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
515
516 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
517 sk_stop_timer(sk, &icsk->icsk_delack_timer);
518 sk_stop_timer(sk, &sk->sk_timer);
519}
520EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
521
522void inet_csk_delete_keepalive_timer(struct sock *sk)
523{
524 sk_stop_timer(sk, &sk->sk_timer);
525}
526EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
527
528void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
529{
530 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
531}
532EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
533
534struct dst_entry *inet_csk_route_req(const struct sock *sk,
535 struct flowi4 *fl4,
536 const struct request_sock *req)
537{
538 const struct inet_request_sock *ireq = inet_rsk(req);
539 struct net *net = read_pnet(&ireq->ireq_net);
540 struct ip_options_rcu *opt;
541 struct rtable *rt;
542
543 opt = ireq_opt_deref(ireq);
544
545 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
546 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
547 sk->sk_protocol, inet_sk_flowi_flags(sk),
548 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
549 ireq->ir_loc_addr, ireq->ir_rmt_port,
550 htons(ireq->ir_num), sk->sk_uid);
551 security_req_classify_flow(req, flowi4_to_flowi(fl4));
552 rt = ip_route_output_flow(net, fl4, sk);
553 if (IS_ERR(rt))
554 goto no_route;
555 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
556 goto route_err;
557 return &rt->dst;
558
559route_err:
560 ip_rt_put(rt);
561no_route:
562 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
563 return NULL;
564}
565EXPORT_SYMBOL_GPL(inet_csk_route_req);
566
567struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
568 struct sock *newsk,
569 const struct request_sock *req)
570{
571 const struct inet_request_sock *ireq = inet_rsk(req);
572 struct net *net = read_pnet(&ireq->ireq_net);
573 struct inet_sock *newinet = inet_sk(newsk);
574 struct ip_options_rcu *opt;
575 struct flowi4 *fl4;
576 struct rtable *rt;
577
578 opt = rcu_dereference(ireq->ireq_opt);
579 fl4 = &newinet->cork.fl.u.ip4;
580
581 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
582 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
583 sk->sk_protocol, inet_sk_flowi_flags(sk),
584 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
585 ireq->ir_loc_addr, ireq->ir_rmt_port,
586 htons(ireq->ir_num), sk->sk_uid);
587 security_req_classify_flow(req, flowi4_to_flowi(fl4));
588 rt = ip_route_output_flow(net, fl4, sk);
589 if (IS_ERR(rt))
590 goto no_route;
591 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
592 goto route_err;
593 return &rt->dst;
594
595route_err:
596 ip_rt_put(rt);
597no_route:
598 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
599 return NULL;
600}
601EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
602
603#if IS_ENABLED(CONFIG_IPV6)
604#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
605#else
606#define AF_INET_FAMILY(fam) true
607#endif
608
609/* Decide when to expire the request and when to resend SYN-ACK */
610static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
611 const int max_retries,
612 const u8 rskq_defer_accept,
613 int *expire, int *resend)
614{
615 if (!rskq_defer_accept) {
616 *expire = req->num_timeout >= thresh;
617 *resend = 1;
618 return;
619 }
620 *expire = req->num_timeout >= thresh &&
621 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
622 /*
623 * Do not resend while waiting for data after ACK,
624 * start to resend on end of deferring period to give
625 * last chance for data or ACK to create established socket.
626 */
627 *resend = !inet_rsk(req)->acked ||
628 req->num_timeout >= rskq_defer_accept - 1;
629}
630
631int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
632{
633 int err = req->rsk_ops->rtx_syn_ack(parent, req);
634
635 if (!err)
636 req->num_retrans++;
637 return err;
638}
639EXPORT_SYMBOL(inet_rtx_syn_ack);
640
641/* return true if req was found in the ehash table */
642static bool reqsk_queue_unlink(struct request_sock_queue *queue,
643 struct request_sock *req)
644{
645 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
646 bool found = false;
647
648 if (sk_hashed(req_to_sk(req))) {
649 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
650
651 spin_lock(lock);
652 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
653 spin_unlock(lock);
654 }
655 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
656 reqsk_put(req);
657 return found;
658}
659
660void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
661{
662 if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
663 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
664 reqsk_put(req);
665 }
666}
667EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
668
669void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
670{
671 inet_csk_reqsk_queue_drop(sk, req);
672 reqsk_put(req);
673}
674EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
675
676static void reqsk_timer_handler(struct timer_list *t)
677{
678 struct request_sock *req = from_timer(req, t, rsk_timer);
679 struct sock *sk_listener = req->rsk_listener;
680 struct net *net = sock_net(sk_listener);
681 struct inet_connection_sock *icsk = inet_csk(sk_listener);
682 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
683 int qlen, expire = 0, resend = 0;
684 int max_retries, thresh;
685 u8 defer_accept;
686
687 if (inet_sk_state_load(sk_listener) != TCP_LISTEN)
688 goto drop;
689
690 max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
691 thresh = max_retries;
692 /* Normally all the openreqs are young and become mature
693 * (i.e. converted to established socket) for first timeout.
694 * If synack was not acknowledged for 1 second, it means
695 * one of the following things: synack was lost, ack was lost,
696 * rtt is high or nobody planned to ack (i.e. synflood).
697 * When server is a bit loaded, queue is populated with old
698 * open requests, reducing effective size of queue.
699 * When server is well loaded, queue size reduces to zero
700 * after several minutes of work. It is not synflood,
701 * it is normal operation. The solution is pruning
702 * too old entries overriding normal timeout, when
703 * situation becomes dangerous.
704 *
705 * Essentially, we reserve half of room for young
706 * embrions; and abort old ones without pity, if old
707 * ones are about to clog our table.
708 */
709 qlen = reqsk_queue_len(queue);
710 if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
711 int young = reqsk_queue_len_young(queue) << 1;
712
713 while (thresh > 2) {
714 if (qlen < young)
715 break;
716 thresh--;
717 young <<= 1;
718 }
719 }
720 defer_accept = READ_ONCE(queue->rskq_defer_accept);
721 if (defer_accept)
722 max_retries = defer_accept;
723 syn_ack_recalc(req, thresh, max_retries, defer_accept,
724 &expire, &resend);
725 req->rsk_ops->syn_ack_timeout(req);
726 if (!expire &&
727 (!resend ||
728 !inet_rtx_syn_ack(sk_listener, req) ||
729 inet_rsk(req)->acked)) {
730 unsigned long timeo;
731
732 if (req->num_timeout++ == 0)
733 atomic_dec(&queue->young);
734 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
735 mod_timer(&req->rsk_timer, jiffies + timeo);
736 return;
737 }
738drop:
739 inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
740}
741
742static void reqsk_queue_hash_req(struct request_sock *req,
743 unsigned long timeout)
744{
745 req->num_retrans = 0;
746 req->num_timeout = 0;
747 req->sk = NULL;
748
749 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
750 mod_timer(&req->rsk_timer, jiffies + timeout);
751
752 inet_ehash_insert(req_to_sk(req), NULL);
753 /* before letting lookups find us, make sure all req fields
754 * are committed to memory and refcnt initialized.
755 */
756 smp_wmb();
757 refcount_set(&req->rsk_refcnt, 2 + 1);
758}
759
760void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
761 unsigned long timeout)
762{
763 reqsk_queue_hash_req(req, timeout);
764 inet_csk_reqsk_queue_added(sk);
765}
766EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
767
768/**
769 * inet_csk_clone_lock - clone an inet socket, and lock its clone
770 * @sk: the socket to clone
771 * @req: request_sock
772 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
773 *
774 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
775 */
776struct sock *inet_csk_clone_lock(const struct sock *sk,
777 const struct request_sock *req,
778 const gfp_t priority)
779{
780 struct sock *newsk = sk_clone_lock(sk, priority);
781
782 if (newsk) {
783 struct inet_connection_sock *newicsk = inet_csk(newsk);
784
785 inet_sk_set_state(newsk, TCP_SYN_RECV);
786 newicsk->icsk_bind_hash = NULL;
787
788 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
789 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
790 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
791
792 /* listeners have SOCK_RCU_FREE, not the children */
793 sock_reset_flag(newsk, SOCK_RCU_FREE);
794
795 inet_sk(newsk)->mc_list = NULL;
796
797 newsk->sk_mark = inet_rsk(req)->ir_mark;
798 atomic64_set(&newsk->sk_cookie,
799 atomic64_read(&inet_rsk(req)->ir_cookie));
800
801 newicsk->icsk_retransmits = 0;
802 newicsk->icsk_backoff = 0;
803 newicsk->icsk_probes_out = 0;
804
805 /* Deinitialize accept_queue to trap illegal accesses. */
806 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
807
808 security_inet_csk_clone(newsk, req);
809 }
810 return newsk;
811}
812EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
813
814/*
815 * At this point, there should be no process reference to this
816 * socket, and thus no user references at all. Therefore we
817 * can assume the socket waitqueue is inactive and nobody will
818 * try to jump onto it.
819 */
820void inet_csk_destroy_sock(struct sock *sk)
821{
822 WARN_ON(sk->sk_state != TCP_CLOSE);
823 WARN_ON(!sock_flag(sk, SOCK_DEAD));
824
825 /* It cannot be in hash table! */
826 WARN_ON(!sk_unhashed(sk));
827
828 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
829 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
830
831 sk->sk_prot->destroy(sk);
832
833 sk_stream_kill_queues(sk);
834
835 xfrm_sk_free_policy(sk);
836
837 sk_refcnt_debug_release(sk);
838
839 percpu_counter_dec(sk->sk_prot->orphan_count);
840
841 sock_put(sk);
842}
843EXPORT_SYMBOL(inet_csk_destroy_sock);
844
845/* This function allows to force a closure of a socket after the call to
846 * tcp/dccp_create_openreq_child().
847 */
848void inet_csk_prepare_forced_close(struct sock *sk)
849 __releases(&sk->sk_lock.slock)
850{
851 /* sk_clone_lock locked the socket and set refcnt to 2 */
852 bh_unlock_sock(sk);
853 sock_put(sk);
854
855 /* The below has to be done to allow calling inet_csk_destroy_sock */
856 sock_set_flag(sk, SOCK_DEAD);
857 percpu_counter_inc(sk->sk_prot->orphan_count);
858 inet_sk(sk)->inet_num = 0;
859}
860EXPORT_SYMBOL(inet_csk_prepare_forced_close);
861
862int inet_csk_listen_start(struct sock *sk, int backlog)
863{
864 struct inet_connection_sock *icsk = inet_csk(sk);
865 struct inet_sock *inet = inet_sk(sk);
866 int err = -EADDRINUSE;
867
868 reqsk_queue_alloc(&icsk->icsk_accept_queue);
869
870 sk->sk_max_ack_backlog = backlog;
871 sk->sk_ack_backlog = 0;
872 inet_csk_delack_init(sk);
873
874 /* There is race window here: we announce ourselves listening,
875 * but this transition is still not validated by get_port().
876 * It is OK, because this socket enters to hash table only
877 * after validation is complete.
878 */
879 inet_sk_state_store(sk, TCP_LISTEN);
880 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
881 inet->inet_sport = htons(inet->inet_num);
882
883 sk_dst_reset(sk);
884 err = sk->sk_prot->hash(sk);
885
886 if (likely(!err))
887 return 0;
888 }
889
890 inet_sk_set_state(sk, TCP_CLOSE);
891 return err;
892}
893EXPORT_SYMBOL_GPL(inet_csk_listen_start);
894
895static void inet_child_forget(struct sock *sk, struct request_sock *req,
896 struct sock *child)
897{
898 sk->sk_prot->disconnect(child, O_NONBLOCK);
899
900 sock_orphan(child);
901
902 percpu_counter_inc(sk->sk_prot->orphan_count);
903
904 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
905 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
906 BUG_ON(sk != req->rsk_listener);
907
908 /* Paranoid, to prevent race condition if
909 * an inbound pkt destined for child is
910 * blocked by sock lock in tcp_v4_rcv().
911 * Also to satisfy an assertion in
912 * tcp_v4_destroy_sock().
913 */
914 tcp_sk(child)->fastopen_rsk = NULL;
915 }
916 inet_csk_destroy_sock(child);
917}
918
919struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
920 struct request_sock *req,
921 struct sock *child)
922{
923 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
924
925 spin_lock(&queue->rskq_lock);
926 if (unlikely(sk->sk_state != TCP_LISTEN)) {
927 inet_child_forget(sk, req, child);
928 child = NULL;
929 } else {
930 req->sk = child;
931 req->dl_next = NULL;
932 if (queue->rskq_accept_head == NULL)
933 queue->rskq_accept_head = req;
934 else
935 queue->rskq_accept_tail->dl_next = req;
936 queue->rskq_accept_tail = req;
937 sk_acceptq_added(sk);
938 }
939 spin_unlock(&queue->rskq_lock);
940 return child;
941}
942EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
943
944struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
945 struct request_sock *req, bool own_req)
946{
947 if (own_req) {
948 inet_csk_reqsk_queue_drop(sk, req);
949 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
950 if (inet_csk_reqsk_queue_add(sk, req, child))
951 return child;
952 }
953 /* Too bad, another child took ownership of the request, undo. */
954 bh_unlock_sock(child);
955 sock_put(child);
956 return NULL;
957}
958EXPORT_SYMBOL(inet_csk_complete_hashdance);
959
960/*
961 * This routine closes sockets which have been at least partially
962 * opened, but not yet accepted.
963 */
964void inet_csk_listen_stop(struct sock *sk)
965{
966 struct inet_connection_sock *icsk = inet_csk(sk);
967 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
968 struct request_sock *next, *req;
969
970 /* Following specs, it would be better either to send FIN
971 * (and enter FIN-WAIT-1, it is normal close)
972 * or to send active reset (abort).
973 * Certainly, it is pretty dangerous while synflood, but it is
974 * bad justification for our negligence 8)
975 * To be honest, we are not able to make either
976 * of the variants now. --ANK
977 */
978 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
979 struct sock *child = req->sk;
980
981 local_bh_disable();
982 bh_lock_sock(child);
983 WARN_ON(sock_owned_by_user(child));
984 sock_hold(child);
985
986 inet_child_forget(sk, req, child);
987 reqsk_put(req);
988 bh_unlock_sock(child);
989 local_bh_enable();
990 sock_put(child);
991
992 cond_resched();
993 }
994 if (queue->fastopenq.rskq_rst_head) {
995 /* Free all the reqs queued in rskq_rst_head. */
996 spin_lock_bh(&queue->fastopenq.lock);
997 req = queue->fastopenq.rskq_rst_head;
998 queue->fastopenq.rskq_rst_head = NULL;
999 spin_unlock_bh(&queue->fastopenq.lock);
1000 while (req != NULL) {
1001 next = req->dl_next;
1002 reqsk_put(req);
1003 req = next;
1004 }
1005 }
1006 WARN_ON_ONCE(sk->sk_ack_backlog);
1007}
1008EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1009
1010void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1011{
1012 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1013 const struct inet_sock *inet = inet_sk(sk);
1014
1015 sin->sin_family = AF_INET;
1016 sin->sin_addr.s_addr = inet->inet_daddr;
1017 sin->sin_port = inet->inet_dport;
1018}
1019EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1020
1021#ifdef CONFIG_COMPAT
1022int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
1023 char __user *optval, int __user *optlen)
1024{
1025 const struct inet_connection_sock *icsk = inet_csk(sk);
1026
1027 if (icsk->icsk_af_ops->compat_getsockopt)
1028 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
1029 optval, optlen);
1030 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
1031 optval, optlen);
1032}
1033EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
1034
1035int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
1036 char __user *optval, unsigned int optlen)
1037{
1038 const struct inet_connection_sock *icsk = inet_csk(sk);
1039
1040 if (icsk->icsk_af_ops->compat_setsockopt)
1041 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
1042 optval, optlen);
1043 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
1044 optval, optlen);
1045}
1046EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
1047#endif
1048
1049static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1050{
1051 const struct inet_sock *inet = inet_sk(sk);
1052 const struct ip_options_rcu *inet_opt;
1053 __be32 daddr = inet->inet_daddr;
1054 struct flowi4 *fl4;
1055 struct rtable *rt;
1056
1057 rcu_read_lock();
1058 inet_opt = rcu_dereference(inet->inet_opt);
1059 if (inet_opt && inet_opt->opt.srr)
1060 daddr = inet_opt->opt.faddr;
1061 fl4 = &fl->u.ip4;
1062 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1063 inet->inet_saddr, inet->inet_dport,
1064 inet->inet_sport, sk->sk_protocol,
1065 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1066 if (IS_ERR(rt))
1067 rt = NULL;
1068 if (rt)
1069 sk_setup_caps(sk, &rt->dst);
1070 rcu_read_unlock();
1071
1072 return &rt->dst;
1073}
1074
1075struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1076{
1077 struct dst_entry *dst = __sk_dst_check(sk, 0);
1078 struct inet_sock *inet = inet_sk(sk);
1079
1080 if (!dst) {
1081 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1082 if (!dst)
1083 goto out;
1084 }
1085 dst->ops->update_pmtu(dst, sk, NULL, mtu);
1086
1087 dst = __sk_dst_check(sk, 0);
1088 if (!dst)
1089 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1090out:
1091 return dst;
1092}
1093EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
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_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
28 * only, and any IPv4 addresses if not IPv6 only
29 * match_wildcard == false: addresses must be exactly the same, i.e.
30 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
31 * and 0.0.0.0 equals to 0.0.0.0 only
32 */
33static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
34 const struct in6_addr *sk2_rcv_saddr6,
35 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
36 bool sk1_ipv6only, bool sk2_ipv6only,
37 bool match_wildcard)
38{
39 int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
40 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
41
42 /* if both are mapped, treat as IPv4 */
43 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
44 if (!sk2_ipv6only) {
45 if (sk1_rcv_saddr == sk2_rcv_saddr)
46 return true;
47 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
48 return match_wildcard;
49 }
50 return false;
51 }
52
53 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
54 return true;
55
56 if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
57 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
58 return true;
59
60 if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
61 !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
62 return true;
63
64 if (sk2_rcv_saddr6 &&
65 ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
66 return true;
67
68 return false;
69}
70#endif
71
72/* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
73 * match_wildcard == false: addresses must be exactly the same, i.e.
74 * 0.0.0.0 only equals to 0.0.0.0
75 */
76static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
77 bool sk2_ipv6only, bool match_wildcard)
78{
79 if (!sk2_ipv6only) {
80 if (sk1_rcv_saddr == sk2_rcv_saddr)
81 return true;
82 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
83 return match_wildcard;
84 }
85 return false;
86}
87
88bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
89 bool match_wildcard)
90{
91#if IS_ENABLED(CONFIG_IPV6)
92 if (sk->sk_family == AF_INET6)
93 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
94 inet6_rcv_saddr(sk2),
95 sk->sk_rcv_saddr,
96 sk2->sk_rcv_saddr,
97 ipv6_only_sock(sk),
98 ipv6_only_sock(sk2),
99 match_wildcard);
100#endif
101 return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
102 ipv6_only_sock(sk2), match_wildcard);
103}
104EXPORT_SYMBOL(inet_rcv_saddr_equal);
105
106bool inet_rcv_saddr_any(const struct sock *sk)
107{
108#if IS_ENABLED(CONFIG_IPV6)
109 if (sk->sk_family == AF_INET6)
110 return ipv6_addr_any(&sk->sk_v6_rcv_saddr);
111#endif
112 return !sk->sk_rcv_saddr;
113}
114
115void inet_get_local_port_range(struct net *net, int *low, int *high)
116{
117 unsigned int seq;
118
119 do {
120 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
121
122 *low = net->ipv4.ip_local_ports.range[0];
123 *high = net->ipv4.ip_local_ports.range[1];
124 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
125}
126EXPORT_SYMBOL(inet_get_local_port_range);
127
128static int inet_csk_bind_conflict(const struct sock *sk,
129 const struct inet_bind_bucket *tb,
130 bool relax, bool reuseport_ok)
131{
132 struct sock *sk2;
133 bool reuse = sk->sk_reuse;
134 bool reuseport = !!sk->sk_reuseport && reuseport_ok;
135 kuid_t uid = sock_i_uid((struct sock *)sk);
136
137 /*
138 * Unlike other sk lookup places we do not check
139 * for sk_net here, since _all_ the socks listed
140 * in tb->owners list belong to the same net - the
141 * one this bucket belongs to.
142 */
143
144 sk_for_each_bound(sk2, &tb->owners) {
145 if (sk != sk2 &&
146 (!sk->sk_bound_dev_if ||
147 !sk2->sk_bound_dev_if ||
148 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
149 if ((!reuse || !sk2->sk_reuse ||
150 sk2->sk_state == TCP_LISTEN) &&
151 (!reuseport || !sk2->sk_reuseport ||
152 rcu_access_pointer(sk->sk_reuseport_cb) ||
153 (sk2->sk_state != TCP_TIME_WAIT &&
154 !uid_eq(uid, sock_i_uid(sk2))))) {
155 if (inet_rcv_saddr_equal(sk, sk2, true))
156 break;
157 }
158 if (!relax && reuse && sk2->sk_reuse &&
159 sk2->sk_state != TCP_LISTEN) {
160 if (inet_rcv_saddr_equal(sk, sk2, true))
161 break;
162 }
163 }
164 }
165 return sk2 != NULL;
166}
167
168/*
169 * Find an open port number for the socket. Returns with the
170 * inet_bind_hashbucket lock held.
171 */
172static struct inet_bind_hashbucket *
173inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
174{
175 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
176 int port = 0;
177 struct inet_bind_hashbucket *head;
178 struct net *net = sock_net(sk);
179 int i, low, high, attempt_half;
180 struct inet_bind_bucket *tb;
181 u32 remaining, offset;
182 int l3mdev;
183
184 l3mdev = inet_sk_bound_l3mdev(sk);
185 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
186other_half_scan:
187 inet_get_local_port_range(net, &low, &high);
188 high++; /* [32768, 60999] -> [32768, 61000[ */
189 if (high - low < 4)
190 attempt_half = 0;
191 if (attempt_half) {
192 int half = low + (((high - low) >> 2) << 1);
193
194 if (attempt_half == 1)
195 high = half;
196 else
197 low = half;
198 }
199 remaining = high - low;
200 if (likely(remaining > 1))
201 remaining &= ~1U;
202
203 offset = prandom_u32() % remaining;
204 /* __inet_hash_connect() favors ports having @low parity
205 * We do the opposite to not pollute connect() users.
206 */
207 offset |= 1U;
208
209other_parity_scan:
210 port = low + offset;
211 for (i = 0; i < remaining; i += 2, port += 2) {
212 if (unlikely(port >= high))
213 port -= remaining;
214 if (inet_is_local_reserved_port(net, port))
215 continue;
216 head = &hinfo->bhash[inet_bhashfn(net, port,
217 hinfo->bhash_size)];
218 spin_lock_bh(&head->lock);
219 inet_bind_bucket_for_each(tb, &head->chain)
220 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
221 tb->port == port) {
222 if (!inet_csk_bind_conflict(sk, tb, false, false))
223 goto success;
224 goto next_port;
225 }
226 tb = NULL;
227 goto success;
228next_port:
229 spin_unlock_bh(&head->lock);
230 cond_resched();
231 }
232
233 offset--;
234 if (!(offset & 1))
235 goto other_parity_scan;
236
237 if (attempt_half == 1) {
238 /* OK we now try the upper half of the range */
239 attempt_half = 2;
240 goto other_half_scan;
241 }
242 return NULL;
243success:
244 *port_ret = port;
245 *tb_ret = tb;
246 return head;
247}
248
249static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
250 struct sock *sk)
251{
252 kuid_t uid = sock_i_uid(sk);
253
254 if (tb->fastreuseport <= 0)
255 return 0;
256 if (!sk->sk_reuseport)
257 return 0;
258 if (rcu_access_pointer(sk->sk_reuseport_cb))
259 return 0;
260 if (!uid_eq(tb->fastuid, uid))
261 return 0;
262 /* We only need to check the rcv_saddr if this tb was once marked
263 * without fastreuseport and then was reset, as we can only know that
264 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
265 * owners list.
266 */
267 if (tb->fastreuseport == FASTREUSEPORT_ANY)
268 return 1;
269#if IS_ENABLED(CONFIG_IPV6)
270 if (tb->fast_sk_family == AF_INET6)
271 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
272 inet6_rcv_saddr(sk),
273 tb->fast_rcv_saddr,
274 sk->sk_rcv_saddr,
275 tb->fast_ipv6_only,
276 ipv6_only_sock(sk), true);
277#endif
278 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
279 ipv6_only_sock(sk), true);
280}
281
282/* Obtain a reference to a local port for the given sock,
283 * if snum is zero it means select any available local port.
284 * We try to allocate an odd port (and leave even ports for connect())
285 */
286int inet_csk_get_port(struct sock *sk, unsigned short snum)
287{
288 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
289 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
290 int ret = 1, port = snum;
291 struct inet_bind_hashbucket *head;
292 struct net *net = sock_net(sk);
293 struct inet_bind_bucket *tb = NULL;
294 kuid_t uid = sock_i_uid(sk);
295 int l3mdev;
296
297 l3mdev = inet_sk_bound_l3mdev(sk);
298
299 if (!port) {
300 head = inet_csk_find_open_port(sk, &tb, &port);
301 if (!head)
302 return ret;
303 if (!tb)
304 goto tb_not_found;
305 goto success;
306 }
307 head = &hinfo->bhash[inet_bhashfn(net, port,
308 hinfo->bhash_size)];
309 spin_lock_bh(&head->lock);
310 inet_bind_bucket_for_each(tb, &head->chain)
311 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
312 tb->port == port)
313 goto tb_found;
314tb_not_found:
315 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
316 net, head, port, l3mdev);
317 if (!tb)
318 goto fail_unlock;
319tb_found:
320 if (!hlist_empty(&tb->owners)) {
321 if (sk->sk_reuse == SK_FORCE_REUSE)
322 goto success;
323
324 if ((tb->fastreuse > 0 && reuse) ||
325 sk_reuseport_match(tb, sk))
326 goto success;
327 if (inet_csk_bind_conflict(sk, tb, true, true))
328 goto fail_unlock;
329 }
330success:
331 if (hlist_empty(&tb->owners)) {
332 tb->fastreuse = reuse;
333 if (sk->sk_reuseport) {
334 tb->fastreuseport = FASTREUSEPORT_ANY;
335 tb->fastuid = uid;
336 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
337 tb->fast_ipv6_only = ipv6_only_sock(sk);
338 tb->fast_sk_family = sk->sk_family;
339#if IS_ENABLED(CONFIG_IPV6)
340 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
341#endif
342 } else {
343 tb->fastreuseport = 0;
344 }
345 } else {
346 if (!reuse)
347 tb->fastreuse = 0;
348 if (sk->sk_reuseport) {
349 /* We didn't match or we don't have fastreuseport set on
350 * the tb, but we have sk_reuseport set on this socket
351 * and we know that there are no bind conflicts with
352 * this socket in this tb, so reset our tb's reuseport
353 * settings so that any subsequent sockets that match
354 * our current socket will be put on the fast path.
355 *
356 * If we reset we need to set FASTREUSEPORT_STRICT so we
357 * do extra checking for all subsequent sk_reuseport
358 * socks.
359 */
360 if (!sk_reuseport_match(tb, sk)) {
361 tb->fastreuseport = FASTREUSEPORT_STRICT;
362 tb->fastuid = uid;
363 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
364 tb->fast_ipv6_only = ipv6_only_sock(sk);
365 tb->fast_sk_family = sk->sk_family;
366#if IS_ENABLED(CONFIG_IPV6)
367 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
368#endif
369 }
370 } else {
371 tb->fastreuseport = 0;
372 }
373 }
374 if (!inet_csk(sk)->icsk_bind_hash)
375 inet_bind_hash(sk, tb, port);
376 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
377 ret = 0;
378
379fail_unlock:
380 spin_unlock_bh(&head->lock);
381 return ret;
382}
383EXPORT_SYMBOL_GPL(inet_csk_get_port);
384
385/*
386 * Wait for an incoming connection, avoid race conditions. This must be called
387 * with the socket locked.
388 */
389static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
390{
391 struct inet_connection_sock *icsk = inet_csk(sk);
392 DEFINE_WAIT(wait);
393 int err;
394
395 /*
396 * True wake-one mechanism for incoming connections: only
397 * one process gets woken up, not the 'whole herd'.
398 * Since we do not 'race & poll' for established sockets
399 * anymore, the common case will execute the loop only once.
400 *
401 * Subtle issue: "add_wait_queue_exclusive()" will be added
402 * after any current non-exclusive waiters, and we know that
403 * it will always _stay_ after any new non-exclusive waiters
404 * because all non-exclusive waiters are added at the
405 * beginning of the wait-queue. As such, it's ok to "drop"
406 * our exclusiveness temporarily when we get woken up without
407 * having to remove and re-insert us on the wait queue.
408 */
409 for (;;) {
410 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
411 TASK_INTERRUPTIBLE);
412 release_sock(sk);
413 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
414 timeo = schedule_timeout(timeo);
415 sched_annotate_sleep();
416 lock_sock(sk);
417 err = 0;
418 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
419 break;
420 err = -EINVAL;
421 if (sk->sk_state != TCP_LISTEN)
422 break;
423 err = sock_intr_errno(timeo);
424 if (signal_pending(current))
425 break;
426 err = -EAGAIN;
427 if (!timeo)
428 break;
429 }
430 finish_wait(sk_sleep(sk), &wait);
431 return err;
432}
433
434/*
435 * This will accept the next outstanding connection.
436 */
437struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
438{
439 struct inet_connection_sock *icsk = inet_csk(sk);
440 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
441 struct request_sock *req;
442 struct sock *newsk;
443 int error;
444
445 lock_sock(sk);
446
447 /* We need to make sure that this socket is listening,
448 * and that it has something pending.
449 */
450 error = -EINVAL;
451 if (sk->sk_state != TCP_LISTEN)
452 goto out_err;
453
454 /* Find already established connection */
455 if (reqsk_queue_empty(queue)) {
456 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
457
458 /* If this is a non blocking socket don't sleep */
459 error = -EAGAIN;
460 if (!timeo)
461 goto out_err;
462
463 error = inet_csk_wait_for_connect(sk, timeo);
464 if (error)
465 goto out_err;
466 }
467 req = reqsk_queue_remove(queue, sk);
468 newsk = req->sk;
469
470 if (sk->sk_protocol == IPPROTO_TCP &&
471 tcp_rsk(req)->tfo_listener) {
472 spin_lock_bh(&queue->fastopenq.lock);
473 if (tcp_rsk(req)->tfo_listener) {
474 /* We are still waiting for the final ACK from 3WHS
475 * so can't free req now. Instead, we set req->sk to
476 * NULL to signify that the child socket is taken
477 * so reqsk_fastopen_remove() will free the req
478 * when 3WHS finishes (or is aborted).
479 */
480 req->sk = NULL;
481 req = NULL;
482 }
483 spin_unlock_bh(&queue->fastopenq.lock);
484 }
485out:
486 release_sock(sk);
487 if (req)
488 reqsk_put(req);
489 return newsk;
490out_err:
491 newsk = NULL;
492 req = NULL;
493 *err = error;
494 goto out;
495}
496EXPORT_SYMBOL(inet_csk_accept);
497
498/*
499 * Using different timers for retransmit, delayed acks and probes
500 * We may wish use just one timer maintaining a list of expire jiffies
501 * to optimize.
502 */
503void inet_csk_init_xmit_timers(struct sock *sk,
504 void (*retransmit_handler)(struct timer_list *t),
505 void (*delack_handler)(struct timer_list *t),
506 void (*keepalive_handler)(struct timer_list *t))
507{
508 struct inet_connection_sock *icsk = inet_csk(sk);
509
510 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
511 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
512 timer_setup(&sk->sk_timer, keepalive_handler, 0);
513 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
514}
515EXPORT_SYMBOL(inet_csk_init_xmit_timers);
516
517void inet_csk_clear_xmit_timers(struct sock *sk)
518{
519 struct inet_connection_sock *icsk = inet_csk(sk);
520
521 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
522
523 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
524 sk_stop_timer(sk, &icsk->icsk_delack_timer);
525 sk_stop_timer(sk, &sk->sk_timer);
526}
527EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
528
529void inet_csk_delete_keepalive_timer(struct sock *sk)
530{
531 sk_stop_timer(sk, &sk->sk_timer);
532}
533EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
534
535void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
536{
537 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
538}
539EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
540
541struct dst_entry *inet_csk_route_req(const struct sock *sk,
542 struct flowi4 *fl4,
543 const struct request_sock *req)
544{
545 const struct inet_request_sock *ireq = inet_rsk(req);
546 struct net *net = read_pnet(&ireq->ireq_net);
547 struct ip_options_rcu *opt;
548 struct rtable *rt;
549
550 rcu_read_lock();
551 opt = rcu_dereference(ireq->ireq_opt);
552
553 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
554 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
555 sk->sk_protocol, inet_sk_flowi_flags(sk),
556 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
557 ireq->ir_loc_addr, ireq->ir_rmt_port,
558 htons(ireq->ir_num), sk->sk_uid);
559 security_req_classify_flow(req, flowi4_to_flowi(fl4));
560 rt = ip_route_output_flow(net, fl4, sk);
561 if (IS_ERR(rt))
562 goto no_route;
563 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
564 goto route_err;
565 rcu_read_unlock();
566 return &rt->dst;
567
568route_err:
569 ip_rt_put(rt);
570no_route:
571 rcu_read_unlock();
572 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
573 return NULL;
574}
575EXPORT_SYMBOL_GPL(inet_csk_route_req);
576
577struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
578 struct sock *newsk,
579 const struct request_sock *req)
580{
581 const struct inet_request_sock *ireq = inet_rsk(req);
582 struct net *net = read_pnet(&ireq->ireq_net);
583 struct inet_sock *newinet = inet_sk(newsk);
584 struct ip_options_rcu *opt;
585 struct flowi4 *fl4;
586 struct rtable *rt;
587
588 opt = rcu_dereference(ireq->ireq_opt);
589 fl4 = &newinet->cork.fl.u.ip4;
590
591 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
592 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
593 sk->sk_protocol, inet_sk_flowi_flags(sk),
594 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
595 ireq->ir_loc_addr, ireq->ir_rmt_port,
596 htons(ireq->ir_num), sk->sk_uid);
597 security_req_classify_flow(req, flowi4_to_flowi(fl4));
598 rt = ip_route_output_flow(net, fl4, sk);
599 if (IS_ERR(rt))
600 goto no_route;
601 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
602 goto route_err;
603 return &rt->dst;
604
605route_err:
606 ip_rt_put(rt);
607no_route:
608 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
609 return NULL;
610}
611EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
612
613#if IS_ENABLED(CONFIG_IPV6)
614#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
615#else
616#define AF_INET_FAMILY(fam) true
617#endif
618
619/* Decide when to expire the request and when to resend SYN-ACK */
620static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
621 const int max_retries,
622 const u8 rskq_defer_accept,
623 int *expire, int *resend)
624{
625 if (!rskq_defer_accept) {
626 *expire = req->num_timeout >= thresh;
627 *resend = 1;
628 return;
629 }
630 *expire = req->num_timeout >= thresh &&
631 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
632 /*
633 * Do not resend while waiting for data after ACK,
634 * start to resend on end of deferring period to give
635 * last chance for data or ACK to create established socket.
636 */
637 *resend = !inet_rsk(req)->acked ||
638 req->num_timeout >= rskq_defer_accept - 1;
639}
640
641int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
642{
643 int err = req->rsk_ops->rtx_syn_ack(parent, req);
644
645 if (!err)
646 req->num_retrans++;
647 return err;
648}
649EXPORT_SYMBOL(inet_rtx_syn_ack);
650
651/* return true if req was found in the ehash table */
652static bool reqsk_queue_unlink(struct request_sock *req)
653{
654 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
655 bool found = false;
656
657 if (sk_hashed(req_to_sk(req))) {
658 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
659
660 spin_lock(lock);
661 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
662 spin_unlock(lock);
663 }
664 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
665 reqsk_put(req);
666 return found;
667}
668
669void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
670{
671 if (reqsk_queue_unlink(req)) {
672 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
673 reqsk_put(req);
674 }
675}
676EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
677
678void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
679{
680 inet_csk_reqsk_queue_drop(sk, req);
681 reqsk_put(req);
682}
683EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
684
685static void reqsk_timer_handler(struct timer_list *t)
686{
687 struct request_sock *req = from_timer(req, t, rsk_timer);
688 struct sock *sk_listener = req->rsk_listener;
689 struct net *net = sock_net(sk_listener);
690 struct inet_connection_sock *icsk = inet_csk(sk_listener);
691 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
692 int qlen, expire = 0, resend = 0;
693 int max_retries, thresh;
694 u8 defer_accept;
695
696 if (inet_sk_state_load(sk_listener) != TCP_LISTEN)
697 goto drop;
698
699 max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
700 thresh = max_retries;
701 /* Normally all the openreqs are young and become mature
702 * (i.e. converted to established socket) for first timeout.
703 * If synack was not acknowledged for 1 second, it means
704 * one of the following things: synack was lost, ack was lost,
705 * rtt is high or nobody planned to ack (i.e. synflood).
706 * When server is a bit loaded, queue is populated with old
707 * open requests, reducing effective size of queue.
708 * When server is well loaded, queue size reduces to zero
709 * after several minutes of work. It is not synflood,
710 * it is normal operation. The solution is pruning
711 * too old entries overriding normal timeout, when
712 * situation becomes dangerous.
713 *
714 * Essentially, we reserve half of room for young
715 * embrions; and abort old ones without pity, if old
716 * ones are about to clog our table.
717 */
718 qlen = reqsk_queue_len(queue);
719 if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
720 int young = reqsk_queue_len_young(queue) << 1;
721
722 while (thresh > 2) {
723 if (qlen < young)
724 break;
725 thresh--;
726 young <<= 1;
727 }
728 }
729 defer_accept = READ_ONCE(queue->rskq_defer_accept);
730 if (defer_accept)
731 max_retries = defer_accept;
732 syn_ack_recalc(req, thresh, max_retries, defer_accept,
733 &expire, &resend);
734 req->rsk_ops->syn_ack_timeout(req);
735 if (!expire &&
736 (!resend ||
737 !inet_rtx_syn_ack(sk_listener, req) ||
738 inet_rsk(req)->acked)) {
739 unsigned long timeo;
740
741 if (req->num_timeout++ == 0)
742 atomic_dec(&queue->young);
743 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
744 mod_timer(&req->rsk_timer, jiffies + timeo);
745 return;
746 }
747drop:
748 inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
749}
750
751static void reqsk_queue_hash_req(struct request_sock *req,
752 unsigned long timeout)
753{
754 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
755 mod_timer(&req->rsk_timer, jiffies + timeout);
756
757 inet_ehash_insert(req_to_sk(req), NULL);
758 /* before letting lookups find us, make sure all req fields
759 * are committed to memory and refcnt initialized.
760 */
761 smp_wmb();
762 refcount_set(&req->rsk_refcnt, 2 + 1);
763}
764
765void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
766 unsigned long timeout)
767{
768 reqsk_queue_hash_req(req, timeout);
769 inet_csk_reqsk_queue_added(sk);
770}
771EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
772
773/**
774 * inet_csk_clone_lock - clone an inet socket, and lock its clone
775 * @sk: the socket to clone
776 * @req: request_sock
777 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
778 *
779 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
780 */
781struct sock *inet_csk_clone_lock(const struct sock *sk,
782 const struct request_sock *req,
783 const gfp_t priority)
784{
785 struct sock *newsk = sk_clone_lock(sk, priority);
786
787 if (newsk) {
788 struct inet_connection_sock *newicsk = inet_csk(newsk);
789
790 inet_sk_set_state(newsk, TCP_SYN_RECV);
791 newicsk->icsk_bind_hash = NULL;
792
793 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
794 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
795 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
796
797 /* listeners have SOCK_RCU_FREE, not the children */
798 sock_reset_flag(newsk, SOCK_RCU_FREE);
799
800 inet_sk(newsk)->mc_list = NULL;
801
802 newsk->sk_mark = inet_rsk(req)->ir_mark;
803 atomic64_set(&newsk->sk_cookie,
804 atomic64_read(&inet_rsk(req)->ir_cookie));
805
806 newicsk->icsk_retransmits = 0;
807 newicsk->icsk_backoff = 0;
808 newicsk->icsk_probes_out = 0;
809
810 /* Deinitialize accept_queue to trap illegal accesses. */
811 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
812
813 security_inet_csk_clone(newsk, req);
814 }
815 return newsk;
816}
817EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
818
819/*
820 * At this point, there should be no process reference to this
821 * socket, and thus no user references at all. Therefore we
822 * can assume the socket waitqueue is inactive and nobody will
823 * try to jump onto it.
824 */
825void inet_csk_destroy_sock(struct sock *sk)
826{
827 WARN_ON(sk->sk_state != TCP_CLOSE);
828 WARN_ON(!sock_flag(sk, SOCK_DEAD));
829
830 /* It cannot be in hash table! */
831 WARN_ON(!sk_unhashed(sk));
832
833 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
834 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
835
836 sk->sk_prot->destroy(sk);
837
838 sk_stream_kill_queues(sk);
839
840 xfrm_sk_free_policy(sk);
841
842 sk_refcnt_debug_release(sk);
843
844 percpu_counter_dec(sk->sk_prot->orphan_count);
845
846 sock_put(sk);
847}
848EXPORT_SYMBOL(inet_csk_destroy_sock);
849
850/* This function allows to force a closure of a socket after the call to
851 * tcp/dccp_create_openreq_child().
852 */
853void inet_csk_prepare_forced_close(struct sock *sk)
854 __releases(&sk->sk_lock.slock)
855{
856 /* sk_clone_lock locked the socket and set refcnt to 2 */
857 bh_unlock_sock(sk);
858 sock_put(sk);
859
860 /* The below has to be done to allow calling inet_csk_destroy_sock */
861 sock_set_flag(sk, SOCK_DEAD);
862 percpu_counter_inc(sk->sk_prot->orphan_count);
863 inet_sk(sk)->inet_num = 0;
864}
865EXPORT_SYMBOL(inet_csk_prepare_forced_close);
866
867int inet_csk_listen_start(struct sock *sk, int backlog)
868{
869 struct inet_connection_sock *icsk = inet_csk(sk);
870 struct inet_sock *inet = inet_sk(sk);
871 int err = -EADDRINUSE;
872
873 reqsk_queue_alloc(&icsk->icsk_accept_queue);
874
875 sk->sk_ack_backlog = 0;
876 inet_csk_delack_init(sk);
877
878 /* There is race window here: we announce ourselves listening,
879 * but this transition is still not validated by get_port().
880 * It is OK, because this socket enters to hash table only
881 * after validation is complete.
882 */
883 inet_sk_state_store(sk, TCP_LISTEN);
884 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
885 inet->inet_sport = htons(inet->inet_num);
886
887 sk_dst_reset(sk);
888 err = sk->sk_prot->hash(sk);
889
890 if (likely(!err))
891 return 0;
892 }
893
894 inet_sk_set_state(sk, TCP_CLOSE);
895 return err;
896}
897EXPORT_SYMBOL_GPL(inet_csk_listen_start);
898
899static void inet_child_forget(struct sock *sk, struct request_sock *req,
900 struct sock *child)
901{
902 sk->sk_prot->disconnect(child, O_NONBLOCK);
903
904 sock_orphan(child);
905
906 percpu_counter_inc(sk->sk_prot->orphan_count);
907
908 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
909 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
910 BUG_ON(sk != req->rsk_listener);
911
912 /* Paranoid, to prevent race condition if
913 * an inbound pkt destined for child is
914 * blocked by sock lock in tcp_v4_rcv().
915 * Also to satisfy an assertion in
916 * tcp_v4_destroy_sock().
917 */
918 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
919 }
920 inet_csk_destroy_sock(child);
921}
922
923struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
924 struct request_sock *req,
925 struct sock *child)
926{
927 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
928
929 spin_lock(&queue->rskq_lock);
930 if (unlikely(sk->sk_state != TCP_LISTEN)) {
931 inet_child_forget(sk, req, child);
932 child = NULL;
933 } else {
934 req->sk = child;
935 req->dl_next = NULL;
936 if (queue->rskq_accept_head == NULL)
937 WRITE_ONCE(queue->rskq_accept_head, req);
938 else
939 queue->rskq_accept_tail->dl_next = req;
940 queue->rskq_accept_tail = req;
941 sk_acceptq_added(sk);
942 }
943 spin_unlock(&queue->rskq_lock);
944 return child;
945}
946EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
947
948struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
949 struct request_sock *req, bool own_req)
950{
951 if (own_req) {
952 inet_csk_reqsk_queue_drop(sk, req);
953 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
954 if (inet_csk_reqsk_queue_add(sk, req, child))
955 return child;
956 }
957 /* Too bad, another child took ownership of the request, undo. */
958 bh_unlock_sock(child);
959 sock_put(child);
960 return NULL;
961}
962EXPORT_SYMBOL(inet_csk_complete_hashdance);
963
964/*
965 * This routine closes sockets which have been at least partially
966 * opened, but not yet accepted.
967 */
968void inet_csk_listen_stop(struct sock *sk)
969{
970 struct inet_connection_sock *icsk = inet_csk(sk);
971 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
972 struct request_sock *next, *req;
973
974 /* Following specs, it would be better either to send FIN
975 * (and enter FIN-WAIT-1, it is normal close)
976 * or to send active reset (abort).
977 * Certainly, it is pretty dangerous while synflood, but it is
978 * bad justification for our negligence 8)
979 * To be honest, we are not able to make either
980 * of the variants now. --ANK
981 */
982 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
983 struct sock *child = req->sk;
984
985 local_bh_disable();
986 bh_lock_sock(child);
987 WARN_ON(sock_owned_by_user(child));
988 sock_hold(child);
989
990 inet_child_forget(sk, req, child);
991 reqsk_put(req);
992 bh_unlock_sock(child);
993 local_bh_enable();
994 sock_put(child);
995
996 cond_resched();
997 }
998 if (queue->fastopenq.rskq_rst_head) {
999 /* Free all the reqs queued in rskq_rst_head. */
1000 spin_lock_bh(&queue->fastopenq.lock);
1001 req = queue->fastopenq.rskq_rst_head;
1002 queue->fastopenq.rskq_rst_head = NULL;
1003 spin_unlock_bh(&queue->fastopenq.lock);
1004 while (req != NULL) {
1005 next = req->dl_next;
1006 reqsk_put(req);
1007 req = next;
1008 }
1009 }
1010 WARN_ON_ONCE(sk->sk_ack_backlog);
1011}
1012EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1013
1014void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1015{
1016 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1017 const struct inet_sock *inet = inet_sk(sk);
1018
1019 sin->sin_family = AF_INET;
1020 sin->sin_addr.s_addr = inet->inet_daddr;
1021 sin->sin_port = inet->inet_dport;
1022}
1023EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1024
1025#ifdef CONFIG_COMPAT
1026int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
1027 char __user *optval, int __user *optlen)
1028{
1029 const struct inet_connection_sock *icsk = inet_csk(sk);
1030
1031 if (icsk->icsk_af_ops->compat_getsockopt)
1032 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
1033 optval, optlen);
1034 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
1035 optval, optlen);
1036}
1037EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
1038
1039int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
1040 char __user *optval, unsigned int optlen)
1041{
1042 const struct inet_connection_sock *icsk = inet_csk(sk);
1043
1044 if (icsk->icsk_af_ops->compat_setsockopt)
1045 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
1046 optval, optlen);
1047 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
1048 optval, optlen);
1049}
1050EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
1051#endif
1052
1053static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1054{
1055 const struct inet_sock *inet = inet_sk(sk);
1056 const struct ip_options_rcu *inet_opt;
1057 __be32 daddr = inet->inet_daddr;
1058 struct flowi4 *fl4;
1059 struct rtable *rt;
1060
1061 rcu_read_lock();
1062 inet_opt = rcu_dereference(inet->inet_opt);
1063 if (inet_opt && inet_opt->opt.srr)
1064 daddr = inet_opt->opt.faddr;
1065 fl4 = &fl->u.ip4;
1066 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1067 inet->inet_saddr, inet->inet_dport,
1068 inet->inet_sport, sk->sk_protocol,
1069 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1070 if (IS_ERR(rt))
1071 rt = NULL;
1072 if (rt)
1073 sk_setup_caps(sk, &rt->dst);
1074 rcu_read_unlock();
1075
1076 return &rt->dst;
1077}
1078
1079struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1080{
1081 struct dst_entry *dst = __sk_dst_check(sk, 0);
1082 struct inet_sock *inet = inet_sk(sk);
1083
1084 if (!dst) {
1085 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1086 if (!dst)
1087 goto out;
1088 }
1089 dst->ops->update_pmtu(dst, sk, NULL, mtu);
1090
1091 dst = __sk_dst_check(sk, 0);
1092 if (!dst)
1093 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1094out:
1095 return dst;
1096}
1097EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);