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#if IS_ENABLED(CONFIG_IPV6)
31/* match_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
32 * only, and any IPv4 addresses if not IPv6 only
33 * match_wildcard == false: addresses must be exactly the same, i.e.
34 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
35 * and 0.0.0.0 equals to 0.0.0.0 only
36 */
37static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
38 const struct in6_addr *sk2_rcv_saddr6,
39 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
40 bool sk1_ipv6only, bool sk2_ipv6only,
41 bool match_wildcard)
42{
43 int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
44 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
45
46 /* if both are mapped, treat as IPv4 */
47 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
48 if (!sk2_ipv6only) {
49 if (sk1_rcv_saddr == sk2_rcv_saddr)
50 return true;
51 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
52 return match_wildcard;
53 }
54 return false;
55 }
56
57 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
58 return true;
59
60 if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
61 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
62 return true;
63
64 if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
65 !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
66 return true;
67
68 if (sk2_rcv_saddr6 &&
69 ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
70 return true;
71
72 return false;
73}
74#endif
75
76/* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
77 * match_wildcard == false: addresses must be exactly the same, i.e.
78 * 0.0.0.0 only equals to 0.0.0.0
79 */
80static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
81 bool sk2_ipv6only, bool match_wildcard)
82{
83 if (!sk2_ipv6only) {
84 if (sk1_rcv_saddr == sk2_rcv_saddr)
85 return true;
86 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
87 return match_wildcard;
88 }
89 return false;
90}
91
92bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
93 bool match_wildcard)
94{
95#if IS_ENABLED(CONFIG_IPV6)
96 if (sk->sk_family == AF_INET6)
97 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
98 inet6_rcv_saddr(sk2),
99 sk->sk_rcv_saddr,
100 sk2->sk_rcv_saddr,
101 ipv6_only_sock(sk),
102 ipv6_only_sock(sk2),
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}
108EXPORT_SYMBOL(inet_rcv_saddr_equal);
109
110bool inet_rcv_saddr_any(const struct sock *sk)
111{
112#if IS_ENABLED(CONFIG_IPV6)
113 if (sk->sk_family == AF_INET6)
114 return ipv6_addr_any(&sk->sk_v6_rcv_saddr);
115#endif
116 return !sk->sk_rcv_saddr;
117}
118
119void inet_get_local_port_range(struct net *net, int *low, int *high)
120{
121 unsigned int seq;
122
123 do {
124 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
125
126 *low = net->ipv4.ip_local_ports.range[0];
127 *high = net->ipv4.ip_local_ports.range[1];
128 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
129}
130EXPORT_SYMBOL(inet_get_local_port_range);
131
132static int inet_csk_bind_conflict(const struct sock *sk,
133 const struct inet_bind_bucket *tb,
134 bool relax, bool reuseport_ok)
135{
136 struct sock *sk2;
137 bool reuse = sk->sk_reuse;
138 bool reuseport = !!sk->sk_reuseport && reuseport_ok;
139 kuid_t uid = sock_i_uid((struct sock *)sk);
140
141 /*
142 * Unlike other sk lookup places we do not check
143 * for sk_net here, since _all_ the socks listed
144 * in tb->owners list belong to the same net - the
145 * one this bucket belongs to.
146 */
147
148 sk_for_each_bound(sk2, &tb->owners) {
149 if (sk != sk2 &&
150 (!sk->sk_bound_dev_if ||
151 !sk2->sk_bound_dev_if ||
152 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
153 if ((!reuse || !sk2->sk_reuse ||
154 sk2->sk_state == TCP_LISTEN) &&
155 (!reuseport || !sk2->sk_reuseport ||
156 rcu_access_pointer(sk->sk_reuseport_cb) ||
157 (sk2->sk_state != TCP_TIME_WAIT &&
158 !uid_eq(uid, sock_i_uid(sk2))))) {
159 if (inet_rcv_saddr_equal(sk, sk2, true))
160 break;
161 }
162 if (!relax && reuse && sk2->sk_reuse &&
163 sk2->sk_state != TCP_LISTEN) {
164 if (inet_rcv_saddr_equal(sk, sk2, true))
165 break;
166 }
167 }
168 }
169 return sk2 != NULL;
170}
171
172/*
173 * Find an open port number for the socket. Returns with the
174 * inet_bind_hashbucket lock held.
175 */
176static struct inet_bind_hashbucket *
177inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
178{
179 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
180 int port = 0;
181 struct inet_bind_hashbucket *head;
182 struct net *net = sock_net(sk);
183 int i, low, high, attempt_half;
184 struct inet_bind_bucket *tb;
185 u32 remaining, offset;
186 int l3mdev;
187
188 l3mdev = inet_sk_bound_l3mdev(sk);
189 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
190other_half_scan:
191 inet_get_local_port_range(net, &low, &high);
192 high++; /* [32768, 60999] -> [32768, 61000[ */
193 if (high - low < 4)
194 attempt_half = 0;
195 if (attempt_half) {
196 int half = low + (((high - low) >> 2) << 1);
197
198 if (attempt_half == 1)
199 high = half;
200 else
201 low = half;
202 }
203 remaining = high - low;
204 if (likely(remaining > 1))
205 remaining &= ~1U;
206
207 offset = prandom_u32() % remaining;
208 /* __inet_hash_connect() favors ports having @low parity
209 * We do the opposite to not pollute connect() users.
210 */
211 offset |= 1U;
212
213other_parity_scan:
214 port = low + offset;
215 for (i = 0; i < remaining; i += 2, port += 2) {
216 if (unlikely(port >= high))
217 port -= remaining;
218 if (inet_is_local_reserved_port(net, port))
219 continue;
220 head = &hinfo->bhash[inet_bhashfn(net, port,
221 hinfo->bhash_size)];
222 spin_lock_bh(&head->lock);
223 inet_bind_bucket_for_each(tb, &head->chain)
224 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
225 tb->port == port) {
226 if (!inet_csk_bind_conflict(sk, tb, false, false))
227 goto success;
228 goto next_port;
229 }
230 tb = NULL;
231 goto success;
232next_port:
233 spin_unlock_bh(&head->lock);
234 cond_resched();
235 }
236
237 offset--;
238 if (!(offset & 1))
239 goto other_parity_scan;
240
241 if (attempt_half == 1) {
242 /* OK we now try the upper half of the range */
243 attempt_half = 2;
244 goto other_half_scan;
245 }
246 return NULL;
247success:
248 *port_ret = port;
249 *tb_ret = tb;
250 return head;
251}
252
253static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
254 struct sock *sk)
255{
256 kuid_t uid = sock_i_uid(sk);
257
258 if (tb->fastreuseport <= 0)
259 return 0;
260 if (!sk->sk_reuseport)
261 return 0;
262 if (rcu_access_pointer(sk->sk_reuseport_cb))
263 return 0;
264 if (!uid_eq(tb->fastuid, uid))
265 return 0;
266 /* We only need to check the rcv_saddr if this tb was once marked
267 * without fastreuseport and then was reset, as we can only know that
268 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
269 * owners list.
270 */
271 if (tb->fastreuseport == FASTREUSEPORT_ANY)
272 return 1;
273#if IS_ENABLED(CONFIG_IPV6)
274 if (tb->fast_sk_family == AF_INET6)
275 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
276 inet6_rcv_saddr(sk),
277 tb->fast_rcv_saddr,
278 sk->sk_rcv_saddr,
279 tb->fast_ipv6_only,
280 ipv6_only_sock(sk), true);
281#endif
282 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
283 ipv6_only_sock(sk), true);
284}
285
286/* Obtain a reference to a local port for the given sock,
287 * if snum is zero it means select any available local port.
288 * We try to allocate an odd port (and leave even ports for connect())
289 */
290int inet_csk_get_port(struct sock *sk, unsigned short snum)
291{
292 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
293 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
294 int ret = 1, port = snum;
295 struct inet_bind_hashbucket *head;
296 struct net *net = sock_net(sk);
297 struct inet_bind_bucket *tb = NULL;
298 kuid_t uid = sock_i_uid(sk);
299 int l3mdev;
300
301 l3mdev = inet_sk_bound_l3mdev(sk);
302
303 if (!port) {
304 head = inet_csk_find_open_port(sk, &tb, &port);
305 if (!head)
306 return ret;
307 if (!tb)
308 goto tb_not_found;
309 goto success;
310 }
311 head = &hinfo->bhash[inet_bhashfn(net, port,
312 hinfo->bhash_size)];
313 spin_lock_bh(&head->lock);
314 inet_bind_bucket_for_each(tb, &head->chain)
315 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
316 tb->port == port)
317 goto tb_found;
318tb_not_found:
319 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
320 net, head, port, l3mdev);
321 if (!tb)
322 goto fail_unlock;
323tb_found:
324 if (!hlist_empty(&tb->owners)) {
325 if (sk->sk_reuse == SK_FORCE_REUSE)
326 goto success;
327
328 if ((tb->fastreuse > 0 && reuse) ||
329 sk_reuseport_match(tb, sk))
330 goto success;
331 if (inet_csk_bind_conflict(sk, tb, true, true))
332 goto fail_unlock;
333 }
334success:
335 if (hlist_empty(&tb->owners)) {
336 tb->fastreuse = reuse;
337 if (sk->sk_reuseport) {
338 tb->fastreuseport = FASTREUSEPORT_ANY;
339 tb->fastuid = uid;
340 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
341 tb->fast_ipv6_only = ipv6_only_sock(sk);
342 tb->fast_sk_family = sk->sk_family;
343#if IS_ENABLED(CONFIG_IPV6)
344 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
345#endif
346 } else {
347 tb->fastreuseport = 0;
348 }
349 } else {
350 if (!reuse)
351 tb->fastreuse = 0;
352 if (sk->sk_reuseport) {
353 /* We didn't match or we don't have fastreuseport set on
354 * the tb, but we have sk_reuseport set on this socket
355 * and we know that there are no bind conflicts with
356 * this socket in this tb, so reset our tb's reuseport
357 * settings so that any subsequent sockets that match
358 * our current socket will be put on the fast path.
359 *
360 * If we reset we need to set FASTREUSEPORT_STRICT so we
361 * do extra checking for all subsequent sk_reuseport
362 * socks.
363 */
364 if (!sk_reuseport_match(tb, sk)) {
365 tb->fastreuseport = FASTREUSEPORT_STRICT;
366 tb->fastuid = uid;
367 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
368 tb->fast_ipv6_only = ipv6_only_sock(sk);
369 tb->fast_sk_family = sk->sk_family;
370#if IS_ENABLED(CONFIG_IPV6)
371 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
372#endif
373 }
374 } else {
375 tb->fastreuseport = 0;
376 }
377 }
378 if (!inet_csk(sk)->icsk_bind_hash)
379 inet_bind_hash(sk, tb, port);
380 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
381 ret = 0;
382
383fail_unlock:
384 spin_unlock_bh(&head->lock);
385 return ret;
386}
387EXPORT_SYMBOL_GPL(inet_csk_get_port);
388
389/*
390 * Wait for an incoming connection, avoid race conditions. This must be called
391 * with the socket locked.
392 */
393static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
394{
395 struct inet_connection_sock *icsk = inet_csk(sk);
396 DEFINE_WAIT(wait);
397 int err;
398
399 /*
400 * True wake-one mechanism for incoming connections: only
401 * one process gets woken up, not the 'whole herd'.
402 * Since we do not 'race & poll' for established sockets
403 * anymore, the common case will execute the loop only once.
404 *
405 * Subtle issue: "add_wait_queue_exclusive()" will be added
406 * after any current non-exclusive waiters, and we know that
407 * it will always _stay_ after any new non-exclusive waiters
408 * because all non-exclusive waiters are added at the
409 * beginning of the wait-queue. As such, it's ok to "drop"
410 * our exclusiveness temporarily when we get woken up without
411 * having to remove and re-insert us on the wait queue.
412 */
413 for (;;) {
414 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
415 TASK_INTERRUPTIBLE);
416 release_sock(sk);
417 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
418 timeo = schedule_timeout(timeo);
419 sched_annotate_sleep();
420 lock_sock(sk);
421 err = 0;
422 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
423 break;
424 err = -EINVAL;
425 if (sk->sk_state != TCP_LISTEN)
426 break;
427 err = sock_intr_errno(timeo);
428 if (signal_pending(current))
429 break;
430 err = -EAGAIN;
431 if (!timeo)
432 break;
433 }
434 finish_wait(sk_sleep(sk), &wait);
435 return err;
436}
437
438/*
439 * This will accept the next outstanding connection.
440 */
441struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
442{
443 struct inet_connection_sock *icsk = inet_csk(sk);
444 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
445 struct request_sock *req;
446 struct sock *newsk;
447 int error;
448
449 lock_sock(sk);
450
451 /* We need to make sure that this socket is listening,
452 * and that it has something pending.
453 */
454 error = -EINVAL;
455 if (sk->sk_state != TCP_LISTEN)
456 goto out_err;
457
458 /* Find already established connection */
459 if (reqsk_queue_empty(queue)) {
460 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
461
462 /* If this is a non blocking socket don't sleep */
463 error = -EAGAIN;
464 if (!timeo)
465 goto out_err;
466
467 error = inet_csk_wait_for_connect(sk, timeo);
468 if (error)
469 goto out_err;
470 }
471 req = reqsk_queue_remove(queue, sk);
472 newsk = req->sk;
473
474 if (sk->sk_protocol == IPPROTO_TCP &&
475 tcp_rsk(req)->tfo_listener) {
476 spin_lock_bh(&queue->fastopenq.lock);
477 if (tcp_rsk(req)->tfo_listener) {
478 /* We are still waiting for the final ACK from 3WHS
479 * so can't free req now. Instead, we set req->sk to
480 * NULL to signify that the child socket is taken
481 * so reqsk_fastopen_remove() will free the req
482 * when 3WHS finishes (or is aborted).
483 */
484 req->sk = NULL;
485 req = NULL;
486 }
487 spin_unlock_bh(&queue->fastopenq.lock);
488 }
489out:
490 release_sock(sk);
491 if (req)
492 reqsk_put(req);
493 return newsk;
494out_err:
495 newsk = NULL;
496 req = NULL;
497 *err = error;
498 goto out;
499}
500EXPORT_SYMBOL(inet_csk_accept);
501
502/*
503 * Using different timers for retransmit, delayed acks and probes
504 * We may wish use just one timer maintaining a list of expire jiffies
505 * to optimize.
506 */
507void inet_csk_init_xmit_timers(struct sock *sk,
508 void (*retransmit_handler)(struct timer_list *t),
509 void (*delack_handler)(struct timer_list *t),
510 void (*keepalive_handler)(struct timer_list *t))
511{
512 struct inet_connection_sock *icsk = inet_csk(sk);
513
514 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
515 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
516 timer_setup(&sk->sk_timer, keepalive_handler, 0);
517 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
518}
519EXPORT_SYMBOL(inet_csk_init_xmit_timers);
520
521void inet_csk_clear_xmit_timers(struct sock *sk)
522{
523 struct inet_connection_sock *icsk = inet_csk(sk);
524
525 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
526
527 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
528 sk_stop_timer(sk, &icsk->icsk_delack_timer);
529 sk_stop_timer(sk, &sk->sk_timer);
530}
531EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
532
533void inet_csk_delete_keepalive_timer(struct sock *sk)
534{
535 sk_stop_timer(sk, &sk->sk_timer);
536}
537EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
538
539void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
540{
541 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
542}
543EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
544
545struct dst_entry *inet_csk_route_req(const struct sock *sk,
546 struct flowi4 *fl4,
547 const struct request_sock *req)
548{
549 const struct inet_request_sock *ireq = inet_rsk(req);
550 struct net *net = read_pnet(&ireq->ireq_net);
551 struct ip_options_rcu *opt;
552 struct rtable *rt;
553
554 rcu_read_lock();
555 opt = rcu_dereference(ireq->ireq_opt);
556
557 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
558 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
559 sk->sk_protocol, inet_sk_flowi_flags(sk),
560 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
561 ireq->ir_loc_addr, ireq->ir_rmt_port,
562 htons(ireq->ir_num), sk->sk_uid);
563 security_req_classify_flow(req, flowi4_to_flowi(fl4));
564 rt = ip_route_output_flow(net, fl4, sk);
565 if (IS_ERR(rt))
566 goto no_route;
567 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
568 goto route_err;
569 rcu_read_unlock();
570 return &rt->dst;
571
572route_err:
573 ip_rt_put(rt);
574no_route:
575 rcu_read_unlock();
576 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
577 return NULL;
578}
579EXPORT_SYMBOL_GPL(inet_csk_route_req);
580
581struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
582 struct sock *newsk,
583 const struct request_sock *req)
584{
585 const struct inet_request_sock *ireq = inet_rsk(req);
586 struct net *net = read_pnet(&ireq->ireq_net);
587 struct inet_sock *newinet = inet_sk(newsk);
588 struct ip_options_rcu *opt;
589 struct flowi4 *fl4;
590 struct rtable *rt;
591
592 opt = rcu_dereference(ireq->ireq_opt);
593 fl4 = &newinet->cork.fl.u.ip4;
594
595 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
596 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
597 sk->sk_protocol, inet_sk_flowi_flags(sk),
598 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
599 ireq->ir_loc_addr, ireq->ir_rmt_port,
600 htons(ireq->ir_num), sk->sk_uid);
601 security_req_classify_flow(req, flowi4_to_flowi(fl4));
602 rt = ip_route_output_flow(net, fl4, sk);
603 if (IS_ERR(rt))
604 goto no_route;
605 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
606 goto route_err;
607 return &rt->dst;
608
609route_err:
610 ip_rt_put(rt);
611no_route:
612 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
613 return NULL;
614}
615EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
616
617#if IS_ENABLED(CONFIG_IPV6)
618#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
619#else
620#define AF_INET_FAMILY(fam) true
621#endif
622
623/* Decide when to expire the request and when to resend SYN-ACK */
624static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
625 const int max_retries,
626 const u8 rskq_defer_accept,
627 int *expire, int *resend)
628{
629 if (!rskq_defer_accept) {
630 *expire = req->num_timeout >= thresh;
631 *resend = 1;
632 return;
633 }
634 *expire = req->num_timeout >= thresh &&
635 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
636 /*
637 * Do not resend while waiting for data after ACK,
638 * start to resend on end of deferring period to give
639 * last chance for data or ACK to create established socket.
640 */
641 *resend = !inet_rsk(req)->acked ||
642 req->num_timeout >= rskq_defer_accept - 1;
643}
644
645int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
646{
647 int err = req->rsk_ops->rtx_syn_ack(parent, req);
648
649 if (!err)
650 req->num_retrans++;
651 return err;
652}
653EXPORT_SYMBOL(inet_rtx_syn_ack);
654
655/* return true if req was found in the ehash table */
656static bool reqsk_queue_unlink(struct request_sock_queue *queue,
657 struct request_sock *req)
658{
659 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
660 bool found = false;
661
662 if (sk_hashed(req_to_sk(req))) {
663 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
664
665 spin_lock(lock);
666 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
667 spin_unlock(lock);
668 }
669 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
670 reqsk_put(req);
671 return found;
672}
673
674void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
675{
676 if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
677 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
678 reqsk_put(req);
679 }
680}
681EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
682
683void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
684{
685 inet_csk_reqsk_queue_drop(sk, req);
686 reqsk_put(req);
687}
688EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
689
690static void reqsk_timer_handler(struct timer_list *t)
691{
692 struct request_sock *req = from_timer(req, t, rsk_timer);
693 struct sock *sk_listener = req->rsk_listener;
694 struct net *net = sock_net(sk_listener);
695 struct inet_connection_sock *icsk = inet_csk(sk_listener);
696 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
697 int qlen, expire = 0, resend = 0;
698 int max_retries, thresh;
699 u8 defer_accept;
700
701 if (inet_sk_state_load(sk_listener) != TCP_LISTEN)
702 goto drop;
703
704 max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
705 thresh = max_retries;
706 /* Normally all the openreqs are young and become mature
707 * (i.e. converted to established socket) for first timeout.
708 * If synack was not acknowledged for 1 second, it means
709 * one of the following things: synack was lost, ack was lost,
710 * rtt is high or nobody planned to ack (i.e. synflood).
711 * When server is a bit loaded, queue is populated with old
712 * open requests, reducing effective size of queue.
713 * When server is well loaded, queue size reduces to zero
714 * after several minutes of work. It is not synflood,
715 * it is normal operation. The solution is pruning
716 * too old entries overriding normal timeout, when
717 * situation becomes dangerous.
718 *
719 * Essentially, we reserve half of room for young
720 * embrions; and abort old ones without pity, if old
721 * ones are about to clog our table.
722 */
723 qlen = reqsk_queue_len(queue);
724 if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
725 int young = reqsk_queue_len_young(queue) << 1;
726
727 while (thresh > 2) {
728 if (qlen < young)
729 break;
730 thresh--;
731 young <<= 1;
732 }
733 }
734 defer_accept = READ_ONCE(queue->rskq_defer_accept);
735 if (defer_accept)
736 max_retries = defer_accept;
737 syn_ack_recalc(req, thresh, max_retries, defer_accept,
738 &expire, &resend);
739 req->rsk_ops->syn_ack_timeout(req);
740 if (!expire &&
741 (!resend ||
742 !inet_rtx_syn_ack(sk_listener, req) ||
743 inet_rsk(req)->acked)) {
744 unsigned long timeo;
745
746 if (req->num_timeout++ == 0)
747 atomic_dec(&queue->young);
748 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
749 mod_timer(&req->rsk_timer, jiffies + timeo);
750 return;
751 }
752drop:
753 inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
754}
755
756static void reqsk_queue_hash_req(struct request_sock *req,
757 unsigned long timeout)
758{
759 req->num_retrans = 0;
760 req->num_timeout = 0;
761 req->sk = NULL;
762
763 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
764 mod_timer(&req->rsk_timer, jiffies + timeout);
765
766 inet_ehash_insert(req_to_sk(req), NULL);
767 /* before letting lookups find us, make sure all req fields
768 * are committed to memory and refcnt initialized.
769 */
770 smp_wmb();
771 refcount_set(&req->rsk_refcnt, 2 + 1);
772}
773
774void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
775 unsigned long timeout)
776{
777 reqsk_queue_hash_req(req, timeout);
778 inet_csk_reqsk_queue_added(sk);
779}
780EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
781
782/**
783 * inet_csk_clone_lock - clone an inet socket, and lock its clone
784 * @sk: the socket to clone
785 * @req: request_sock
786 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
787 *
788 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
789 */
790struct sock *inet_csk_clone_lock(const struct sock *sk,
791 const struct request_sock *req,
792 const gfp_t priority)
793{
794 struct sock *newsk = sk_clone_lock(sk, priority);
795
796 if (newsk) {
797 struct inet_connection_sock *newicsk = inet_csk(newsk);
798
799 inet_sk_set_state(newsk, TCP_SYN_RECV);
800 newicsk->icsk_bind_hash = NULL;
801
802 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
803 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
804 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
805
806 /* listeners have SOCK_RCU_FREE, not the children */
807 sock_reset_flag(newsk, SOCK_RCU_FREE);
808
809 inet_sk(newsk)->mc_list = NULL;
810
811 newsk->sk_mark = inet_rsk(req)->ir_mark;
812 atomic64_set(&newsk->sk_cookie,
813 atomic64_read(&inet_rsk(req)->ir_cookie));
814
815 newicsk->icsk_retransmits = 0;
816 newicsk->icsk_backoff = 0;
817 newicsk->icsk_probes_out = 0;
818
819 /* Deinitialize accept_queue to trap illegal accesses. */
820 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
821
822 security_inet_csk_clone(newsk, req);
823 }
824 return newsk;
825}
826EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
827
828/*
829 * At this point, there should be no process reference to this
830 * socket, and thus no user references at all. Therefore we
831 * can assume the socket waitqueue is inactive and nobody will
832 * try to jump onto it.
833 */
834void inet_csk_destroy_sock(struct sock *sk)
835{
836 WARN_ON(sk->sk_state != TCP_CLOSE);
837 WARN_ON(!sock_flag(sk, SOCK_DEAD));
838
839 /* It cannot be in hash table! */
840 WARN_ON(!sk_unhashed(sk));
841
842 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
843 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
844
845 sk->sk_prot->destroy(sk);
846
847 sk_stream_kill_queues(sk);
848
849 xfrm_sk_free_policy(sk);
850
851 sk_refcnt_debug_release(sk);
852
853 percpu_counter_dec(sk->sk_prot->orphan_count);
854
855 sock_put(sk);
856}
857EXPORT_SYMBOL(inet_csk_destroy_sock);
858
859/* This function allows to force a closure of a socket after the call to
860 * tcp/dccp_create_openreq_child().
861 */
862void inet_csk_prepare_forced_close(struct sock *sk)
863 __releases(&sk->sk_lock.slock)
864{
865 /* sk_clone_lock locked the socket and set refcnt to 2 */
866 bh_unlock_sock(sk);
867 sock_put(sk);
868
869 /* The below has to be done to allow calling inet_csk_destroy_sock */
870 sock_set_flag(sk, SOCK_DEAD);
871 percpu_counter_inc(sk->sk_prot->orphan_count);
872 inet_sk(sk)->inet_num = 0;
873}
874EXPORT_SYMBOL(inet_csk_prepare_forced_close);
875
876int inet_csk_listen_start(struct sock *sk, int backlog)
877{
878 struct inet_connection_sock *icsk = inet_csk(sk);
879 struct inet_sock *inet = inet_sk(sk);
880 int err = -EADDRINUSE;
881
882 reqsk_queue_alloc(&icsk->icsk_accept_queue);
883
884 sk->sk_ack_backlog = 0;
885 inet_csk_delack_init(sk);
886
887 /* There is race window here: we announce ourselves listening,
888 * but this transition is still not validated by get_port().
889 * It is OK, because this socket enters to hash table only
890 * after validation is complete.
891 */
892 inet_sk_state_store(sk, TCP_LISTEN);
893 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
894 inet->inet_sport = htons(inet->inet_num);
895
896 sk_dst_reset(sk);
897 err = sk->sk_prot->hash(sk);
898
899 if (likely(!err))
900 return 0;
901 }
902
903 inet_sk_set_state(sk, TCP_CLOSE);
904 return err;
905}
906EXPORT_SYMBOL_GPL(inet_csk_listen_start);
907
908static void inet_child_forget(struct sock *sk, struct request_sock *req,
909 struct sock *child)
910{
911 sk->sk_prot->disconnect(child, O_NONBLOCK);
912
913 sock_orphan(child);
914
915 percpu_counter_inc(sk->sk_prot->orphan_count);
916
917 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
918 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
919 BUG_ON(sk != req->rsk_listener);
920
921 /* Paranoid, to prevent race condition if
922 * an inbound pkt destined for child is
923 * blocked by sock lock in tcp_v4_rcv().
924 * Also to satisfy an assertion in
925 * tcp_v4_destroy_sock().
926 */
927 tcp_sk(child)->fastopen_rsk = NULL;
928 }
929 inet_csk_destroy_sock(child);
930}
931
932struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
933 struct request_sock *req,
934 struct sock *child)
935{
936 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
937
938 spin_lock(&queue->rskq_lock);
939 if (unlikely(sk->sk_state != TCP_LISTEN)) {
940 inet_child_forget(sk, req, child);
941 child = NULL;
942 } else {
943 req->sk = child;
944 req->dl_next = NULL;
945 if (queue->rskq_accept_head == NULL)
946 queue->rskq_accept_head = req;
947 else
948 queue->rskq_accept_tail->dl_next = req;
949 queue->rskq_accept_tail = req;
950 sk_acceptq_added(sk);
951 }
952 spin_unlock(&queue->rskq_lock);
953 return child;
954}
955EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
956
957struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
958 struct request_sock *req, bool own_req)
959{
960 if (own_req) {
961 inet_csk_reqsk_queue_drop(sk, req);
962 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
963 if (inet_csk_reqsk_queue_add(sk, req, child))
964 return child;
965 }
966 /* Too bad, another child took ownership of the request, undo. */
967 bh_unlock_sock(child);
968 sock_put(child);
969 return NULL;
970}
971EXPORT_SYMBOL(inet_csk_complete_hashdance);
972
973/*
974 * This routine closes sockets which have been at least partially
975 * opened, but not yet accepted.
976 */
977void inet_csk_listen_stop(struct sock *sk)
978{
979 struct inet_connection_sock *icsk = inet_csk(sk);
980 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
981 struct request_sock *next, *req;
982
983 /* Following specs, it would be better either to send FIN
984 * (and enter FIN-WAIT-1, it is normal close)
985 * or to send active reset (abort).
986 * Certainly, it is pretty dangerous while synflood, but it is
987 * bad justification for our negligence 8)
988 * To be honest, we are not able to make either
989 * of the variants now. --ANK
990 */
991 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
992 struct sock *child = req->sk;
993
994 local_bh_disable();
995 bh_lock_sock(child);
996 WARN_ON(sock_owned_by_user(child));
997 sock_hold(child);
998
999 inet_child_forget(sk, req, child);
1000 reqsk_put(req);
1001 bh_unlock_sock(child);
1002 local_bh_enable();
1003 sock_put(child);
1004
1005 cond_resched();
1006 }
1007 if (queue->fastopenq.rskq_rst_head) {
1008 /* Free all the reqs queued in rskq_rst_head. */
1009 spin_lock_bh(&queue->fastopenq.lock);
1010 req = queue->fastopenq.rskq_rst_head;
1011 queue->fastopenq.rskq_rst_head = NULL;
1012 spin_unlock_bh(&queue->fastopenq.lock);
1013 while (req != NULL) {
1014 next = req->dl_next;
1015 reqsk_put(req);
1016 req = next;
1017 }
1018 }
1019 WARN_ON_ONCE(sk->sk_ack_backlog);
1020}
1021EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1022
1023void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1024{
1025 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1026 const struct inet_sock *inet = inet_sk(sk);
1027
1028 sin->sin_family = AF_INET;
1029 sin->sin_addr.s_addr = inet->inet_daddr;
1030 sin->sin_port = inet->inet_dport;
1031}
1032EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1033
1034#ifdef CONFIG_COMPAT
1035int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
1036 char __user *optval, int __user *optlen)
1037{
1038 const struct inet_connection_sock *icsk = inet_csk(sk);
1039
1040 if (icsk->icsk_af_ops->compat_getsockopt)
1041 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
1042 optval, optlen);
1043 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
1044 optval, optlen);
1045}
1046EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
1047
1048int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
1049 char __user *optval, unsigned int optlen)
1050{
1051 const struct inet_connection_sock *icsk = inet_csk(sk);
1052
1053 if (icsk->icsk_af_ops->compat_setsockopt)
1054 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
1055 optval, optlen);
1056 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
1057 optval, optlen);
1058}
1059EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
1060#endif
1061
1062static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1063{
1064 const struct inet_sock *inet = inet_sk(sk);
1065 const struct ip_options_rcu *inet_opt;
1066 __be32 daddr = inet->inet_daddr;
1067 struct flowi4 *fl4;
1068 struct rtable *rt;
1069
1070 rcu_read_lock();
1071 inet_opt = rcu_dereference(inet->inet_opt);
1072 if (inet_opt && inet_opt->opt.srr)
1073 daddr = inet_opt->opt.faddr;
1074 fl4 = &fl->u.ip4;
1075 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1076 inet->inet_saddr, inet->inet_dport,
1077 inet->inet_sport, sk->sk_protocol,
1078 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1079 if (IS_ERR(rt))
1080 rt = NULL;
1081 if (rt)
1082 sk_setup_caps(sk, &rt->dst);
1083 rcu_read_unlock();
1084
1085 return &rt->dst;
1086}
1087
1088struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1089{
1090 struct dst_entry *dst = __sk_dst_check(sk, 0);
1091 struct inet_sock *inet = inet_sk(sk);
1092
1093 if (!dst) {
1094 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1095 if (!dst)
1096 goto out;
1097 }
1098 dst->ops->update_pmtu(dst, sk, NULL, mtu);
1099
1100 dst = __sk_dst_check(sk, 0);
1101 if (!dst)
1102 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1103out:
1104 return dst;
1105}
1106EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
1107