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
27#ifdef INET_CSK_DEBUG
28const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
29EXPORT_SYMBOL(inet_csk_timer_bug_msg);
30#endif
31
32/*
33 * This struct holds the first and last local port number.
34 */
35struct local_ports sysctl_local_ports __read_mostly = {
36 .lock = __SEQLOCK_UNLOCKED(sysctl_local_ports.lock),
37 .range = { 32768, 61000 },
38};
39
40unsigned long *sysctl_local_reserved_ports;
41EXPORT_SYMBOL(sysctl_local_reserved_ports);
42
43void inet_get_local_port_range(int *low, int *high)
44{
45 unsigned seq;
46 do {
47 seq = read_seqbegin(&sysctl_local_ports.lock);
48
49 *low = sysctl_local_ports.range[0];
50 *high = sysctl_local_ports.range[1];
51 } while (read_seqretry(&sysctl_local_ports.lock, seq));
52}
53EXPORT_SYMBOL(inet_get_local_port_range);
54
55int inet_csk_bind_conflict(const struct sock *sk,
56 const struct inet_bind_bucket *tb)
57{
58 struct sock *sk2;
59 struct hlist_node *node;
60 int reuse = sk->sk_reuse;
61
62 /*
63 * Unlike other sk lookup places we do not check
64 * for sk_net here, since _all_ the socks listed
65 * in tb->owners list belong to the same net - the
66 * one this bucket belongs to.
67 */
68
69 sk_for_each_bound(sk2, node, &tb->owners) {
70 if (sk != sk2 &&
71 !inet_v6_ipv6only(sk2) &&
72 (!sk->sk_bound_dev_if ||
73 !sk2->sk_bound_dev_if ||
74 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
75 if (!reuse || !sk2->sk_reuse ||
76 sk2->sk_state == TCP_LISTEN) {
77 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
78 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
79 sk2_rcv_saddr == sk_rcv_saddr(sk))
80 break;
81 }
82 }
83 }
84 return node != NULL;
85}
86EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
87
88/* Obtain a reference to a local port for the given sock,
89 * if snum is zero it means select any available local port.
90 */
91int inet_csk_get_port(struct sock *sk, unsigned short snum)
92{
93 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
94 struct inet_bind_hashbucket *head;
95 struct hlist_node *node;
96 struct inet_bind_bucket *tb;
97 int ret, attempts = 5;
98 struct net *net = sock_net(sk);
99 int smallest_size = -1, smallest_rover;
100
101 local_bh_disable();
102 if (!snum) {
103 int remaining, rover, low, high;
104
105again:
106 inet_get_local_port_range(&low, &high);
107 remaining = (high - low) + 1;
108 smallest_rover = rover = net_random() % remaining + low;
109
110 smallest_size = -1;
111 do {
112 if (inet_is_reserved_local_port(rover))
113 goto next_nolock;
114 head = &hashinfo->bhash[inet_bhashfn(net, rover,
115 hashinfo->bhash_size)];
116 spin_lock(&head->lock);
117 inet_bind_bucket_for_each(tb, node, &head->chain)
118 if (net_eq(ib_net(tb), net) && tb->port == rover) {
119 if (tb->fastreuse > 0 &&
120 sk->sk_reuse &&
121 sk->sk_state != TCP_LISTEN &&
122 (tb->num_owners < smallest_size || smallest_size == -1)) {
123 smallest_size = tb->num_owners;
124 smallest_rover = rover;
125 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) {
126 spin_unlock(&head->lock);
127 snum = smallest_rover;
128 goto have_snum;
129 }
130 }
131 goto next;
132 }
133 break;
134 next:
135 spin_unlock(&head->lock);
136 next_nolock:
137 if (++rover > high)
138 rover = low;
139 } while (--remaining > 0);
140
141 /* Exhausted local port range during search? It is not
142 * possible for us to be holding one of the bind hash
143 * locks if this test triggers, because if 'remaining'
144 * drops to zero, we broke out of the do/while loop at
145 * the top level, not from the 'break;' statement.
146 */
147 ret = 1;
148 if (remaining <= 0) {
149 if (smallest_size != -1) {
150 snum = smallest_rover;
151 goto have_snum;
152 }
153 goto fail;
154 }
155 /* OK, here is the one we will use. HEAD is
156 * non-NULL and we hold it's mutex.
157 */
158 snum = rover;
159 } else {
160have_snum:
161 head = &hashinfo->bhash[inet_bhashfn(net, snum,
162 hashinfo->bhash_size)];
163 spin_lock(&head->lock);
164 inet_bind_bucket_for_each(tb, node, &head->chain)
165 if (net_eq(ib_net(tb), net) && tb->port == snum)
166 goto tb_found;
167 }
168 tb = NULL;
169 goto tb_not_found;
170tb_found:
171 if (!hlist_empty(&tb->owners)) {
172 if (tb->fastreuse > 0 &&
173 sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
174 smallest_size == -1) {
175 goto success;
176 } else {
177 ret = 1;
178 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
179 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
180 smallest_size != -1 && --attempts >= 0) {
181 spin_unlock(&head->lock);
182 goto again;
183 }
184 goto fail_unlock;
185 }
186 }
187 }
188tb_not_found:
189 ret = 1;
190 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
191 net, head, snum)) == NULL)
192 goto fail_unlock;
193 if (hlist_empty(&tb->owners)) {
194 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
195 tb->fastreuse = 1;
196 else
197 tb->fastreuse = 0;
198 } else if (tb->fastreuse &&
199 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
200 tb->fastreuse = 0;
201success:
202 if (!inet_csk(sk)->icsk_bind_hash)
203 inet_bind_hash(sk, tb, snum);
204 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
205 ret = 0;
206
207fail_unlock:
208 spin_unlock(&head->lock);
209fail:
210 local_bh_enable();
211 return ret;
212}
213EXPORT_SYMBOL_GPL(inet_csk_get_port);
214
215/*
216 * Wait for an incoming connection, avoid race conditions. This must be called
217 * with the socket locked.
218 */
219static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
220{
221 struct inet_connection_sock *icsk = inet_csk(sk);
222 DEFINE_WAIT(wait);
223 int err;
224
225 /*
226 * True wake-one mechanism for incoming connections: only
227 * one process gets woken up, not the 'whole herd'.
228 * Since we do not 'race & poll' for established sockets
229 * anymore, the common case will execute the loop only once.
230 *
231 * Subtle issue: "add_wait_queue_exclusive()" will be added
232 * after any current non-exclusive waiters, and we know that
233 * it will always _stay_ after any new non-exclusive waiters
234 * because all non-exclusive waiters are added at the
235 * beginning of the wait-queue. As such, it's ok to "drop"
236 * our exclusiveness temporarily when we get woken up without
237 * having to remove and re-insert us on the wait queue.
238 */
239 for (;;) {
240 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
241 TASK_INTERRUPTIBLE);
242 release_sock(sk);
243 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
244 timeo = schedule_timeout(timeo);
245 lock_sock(sk);
246 err = 0;
247 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
248 break;
249 err = -EINVAL;
250 if (sk->sk_state != TCP_LISTEN)
251 break;
252 err = sock_intr_errno(timeo);
253 if (signal_pending(current))
254 break;
255 err = -EAGAIN;
256 if (!timeo)
257 break;
258 }
259 finish_wait(sk_sleep(sk), &wait);
260 return err;
261}
262
263/*
264 * This will accept the next outstanding connection.
265 */
266struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
267{
268 struct inet_connection_sock *icsk = inet_csk(sk);
269 struct sock *newsk;
270 int error;
271
272 lock_sock(sk);
273
274 /* We need to make sure that this socket is listening,
275 * and that it has something pending.
276 */
277 error = -EINVAL;
278 if (sk->sk_state != TCP_LISTEN)
279 goto out_err;
280
281 /* Find already established connection */
282 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
283 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
284
285 /* If this is a non blocking socket don't sleep */
286 error = -EAGAIN;
287 if (!timeo)
288 goto out_err;
289
290 error = inet_csk_wait_for_connect(sk, timeo);
291 if (error)
292 goto out_err;
293 }
294
295 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
296 WARN_ON(newsk->sk_state == TCP_SYN_RECV);
297out:
298 release_sock(sk);
299 return newsk;
300out_err:
301 newsk = NULL;
302 *err = error;
303 goto out;
304}
305EXPORT_SYMBOL(inet_csk_accept);
306
307/*
308 * Using different timers for retransmit, delayed acks and probes
309 * We may wish use just one timer maintaining a list of expire jiffies
310 * to optimize.
311 */
312void inet_csk_init_xmit_timers(struct sock *sk,
313 void (*retransmit_handler)(unsigned long),
314 void (*delack_handler)(unsigned long),
315 void (*keepalive_handler)(unsigned long))
316{
317 struct inet_connection_sock *icsk = inet_csk(sk);
318
319 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
320 (unsigned long)sk);
321 setup_timer(&icsk->icsk_delack_timer, delack_handler,
322 (unsigned long)sk);
323 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
324 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
325}
326EXPORT_SYMBOL(inet_csk_init_xmit_timers);
327
328void inet_csk_clear_xmit_timers(struct sock *sk)
329{
330 struct inet_connection_sock *icsk = inet_csk(sk);
331
332 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
333
334 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
335 sk_stop_timer(sk, &icsk->icsk_delack_timer);
336 sk_stop_timer(sk, &sk->sk_timer);
337}
338EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
339
340void inet_csk_delete_keepalive_timer(struct sock *sk)
341{
342 sk_stop_timer(sk, &sk->sk_timer);
343}
344EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
345
346void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
347{
348 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
349}
350EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
351
352struct dst_entry *inet_csk_route_req(struct sock *sk,
353 struct flowi4 *fl4,
354 const struct request_sock *req)
355{
356 struct rtable *rt;
357 const struct inet_request_sock *ireq = inet_rsk(req);
358 struct ip_options_rcu *opt = inet_rsk(req)->opt;
359 struct net *net = sock_net(sk);
360
361 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
362 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
363 sk->sk_protocol, inet_sk_flowi_flags(sk),
364 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
365 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
366 security_req_classify_flow(req, flowi4_to_flowi(fl4));
367 rt = ip_route_output_flow(net, fl4, sk);
368 if (IS_ERR(rt))
369 goto no_route;
370 if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
371 goto route_err;
372 return &rt->dst;
373
374route_err:
375 ip_rt_put(rt);
376no_route:
377 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
378 return NULL;
379}
380EXPORT_SYMBOL_GPL(inet_csk_route_req);
381
382struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
383 struct sock *newsk,
384 const struct request_sock *req)
385{
386 const struct inet_request_sock *ireq = inet_rsk(req);
387 struct inet_sock *newinet = inet_sk(newsk);
388 struct ip_options_rcu *opt = ireq->opt;
389 struct net *net = sock_net(sk);
390 struct flowi4 *fl4;
391 struct rtable *rt;
392
393 fl4 = &newinet->cork.fl.u.ip4;
394 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
395 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
396 sk->sk_protocol, inet_sk_flowi_flags(sk),
397 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
398 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
399 security_req_classify_flow(req, flowi4_to_flowi(fl4));
400 rt = ip_route_output_flow(net, fl4, sk);
401 if (IS_ERR(rt))
402 goto no_route;
403 if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
404 goto route_err;
405 return &rt->dst;
406
407route_err:
408 ip_rt_put(rt);
409no_route:
410 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
411 return NULL;
412}
413EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
414
415static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
416 const u32 rnd, const u32 synq_hsize)
417{
418 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
419}
420
421#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
422#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
423#else
424#define AF_INET_FAMILY(fam) 1
425#endif
426
427struct request_sock *inet_csk_search_req(const struct sock *sk,
428 struct request_sock ***prevp,
429 const __be16 rport, const __be32 raddr,
430 const __be32 laddr)
431{
432 const struct inet_connection_sock *icsk = inet_csk(sk);
433 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
434 struct request_sock *req, **prev;
435
436 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
437 lopt->nr_table_entries)];
438 (req = *prev) != NULL;
439 prev = &req->dl_next) {
440 const struct inet_request_sock *ireq = inet_rsk(req);
441
442 if (ireq->rmt_port == rport &&
443 ireq->rmt_addr == raddr &&
444 ireq->loc_addr == laddr &&
445 AF_INET_FAMILY(req->rsk_ops->family)) {
446 WARN_ON(req->sk);
447 *prevp = prev;
448 break;
449 }
450 }
451
452 return req;
453}
454EXPORT_SYMBOL_GPL(inet_csk_search_req);
455
456void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
457 unsigned long timeout)
458{
459 struct inet_connection_sock *icsk = inet_csk(sk);
460 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
461 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
462 lopt->hash_rnd, lopt->nr_table_entries);
463
464 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
465 inet_csk_reqsk_queue_added(sk, timeout);
466}
467EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
468
469/* Only thing we need from tcp.h */
470extern int sysctl_tcp_synack_retries;
471
472
473/* Decide when to expire the request and when to resend SYN-ACK */
474static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
475 const int max_retries,
476 const u8 rskq_defer_accept,
477 int *expire, int *resend)
478{
479 if (!rskq_defer_accept) {
480 *expire = req->retrans >= thresh;
481 *resend = 1;
482 return;
483 }
484 *expire = req->retrans >= thresh &&
485 (!inet_rsk(req)->acked || req->retrans >= max_retries);
486 /*
487 * Do not resend while waiting for data after ACK,
488 * start to resend on end of deferring period to give
489 * last chance for data or ACK to create established socket.
490 */
491 *resend = !inet_rsk(req)->acked ||
492 req->retrans >= rskq_defer_accept - 1;
493}
494
495void inet_csk_reqsk_queue_prune(struct sock *parent,
496 const unsigned long interval,
497 const unsigned long timeout,
498 const unsigned long max_rto)
499{
500 struct inet_connection_sock *icsk = inet_csk(parent);
501 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
502 struct listen_sock *lopt = queue->listen_opt;
503 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
504 int thresh = max_retries;
505 unsigned long now = jiffies;
506 struct request_sock **reqp, *req;
507 int i, budget;
508
509 if (lopt == NULL || lopt->qlen == 0)
510 return;
511
512 /* Normally all the openreqs are young and become mature
513 * (i.e. converted to established socket) for first timeout.
514 * If synack was not acknowledged for 3 seconds, it means
515 * one of the following things: synack was lost, ack was lost,
516 * rtt is high or nobody planned to ack (i.e. synflood).
517 * When server is a bit loaded, queue is populated with old
518 * open requests, reducing effective size of queue.
519 * When server is well loaded, queue size reduces to zero
520 * after several minutes of work. It is not synflood,
521 * it is normal operation. The solution is pruning
522 * too old entries overriding normal timeout, when
523 * situation becomes dangerous.
524 *
525 * Essentially, we reserve half of room for young
526 * embrions; and abort old ones without pity, if old
527 * ones are about to clog our table.
528 */
529 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
530 int young = (lopt->qlen_young<<1);
531
532 while (thresh > 2) {
533 if (lopt->qlen < young)
534 break;
535 thresh--;
536 young <<= 1;
537 }
538 }
539
540 if (queue->rskq_defer_accept)
541 max_retries = queue->rskq_defer_accept;
542
543 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
544 i = lopt->clock_hand;
545
546 do {
547 reqp=&lopt->syn_table[i];
548 while ((req = *reqp) != NULL) {
549 if (time_after_eq(now, req->expires)) {
550 int expire = 0, resend = 0;
551
552 syn_ack_recalc(req, thresh, max_retries,
553 queue->rskq_defer_accept,
554 &expire, &resend);
555 if (req->rsk_ops->syn_ack_timeout)
556 req->rsk_ops->syn_ack_timeout(parent, req);
557 if (!expire &&
558 (!resend ||
559 !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
560 inet_rsk(req)->acked)) {
561 unsigned long timeo;
562
563 if (req->retrans++ == 0)
564 lopt->qlen_young--;
565 timeo = min((timeout << req->retrans), max_rto);
566 req->expires = now + timeo;
567 reqp = &req->dl_next;
568 continue;
569 }
570
571 /* Drop this request */
572 inet_csk_reqsk_queue_unlink(parent, req, reqp);
573 reqsk_queue_removed(queue, req);
574 reqsk_free(req);
575 continue;
576 }
577 reqp = &req->dl_next;
578 }
579
580 i = (i + 1) & (lopt->nr_table_entries - 1);
581
582 } while (--budget > 0);
583
584 lopt->clock_hand = i;
585
586 if (lopt->qlen)
587 inet_csk_reset_keepalive_timer(parent, interval);
588}
589EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
590
591struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
592 const gfp_t priority)
593{
594 struct sock *newsk = sk_clone(sk, priority);
595
596 if (newsk != NULL) {
597 struct inet_connection_sock *newicsk = inet_csk(newsk);
598
599 newsk->sk_state = TCP_SYN_RECV;
600 newicsk->icsk_bind_hash = NULL;
601
602 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
603 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
604 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
605 newsk->sk_write_space = sk_stream_write_space;
606
607 newicsk->icsk_retransmits = 0;
608 newicsk->icsk_backoff = 0;
609 newicsk->icsk_probes_out = 0;
610
611 /* Deinitialize accept_queue to trap illegal accesses. */
612 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
613
614 security_inet_csk_clone(newsk, req);
615 }
616 return newsk;
617}
618EXPORT_SYMBOL_GPL(inet_csk_clone);
619
620/*
621 * At this point, there should be no process reference to this
622 * socket, and thus no user references at all. Therefore we
623 * can assume the socket waitqueue is inactive and nobody will
624 * try to jump onto it.
625 */
626void inet_csk_destroy_sock(struct sock *sk)
627{
628 WARN_ON(sk->sk_state != TCP_CLOSE);
629 WARN_ON(!sock_flag(sk, SOCK_DEAD));
630
631 /* It cannot be in hash table! */
632 WARN_ON(!sk_unhashed(sk));
633
634 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
635 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
636
637 sk->sk_prot->destroy(sk);
638
639 sk_stream_kill_queues(sk);
640
641 xfrm_sk_free_policy(sk);
642
643 sk_refcnt_debug_release(sk);
644
645 percpu_counter_dec(sk->sk_prot->orphan_count);
646 sock_put(sk);
647}
648EXPORT_SYMBOL(inet_csk_destroy_sock);
649
650int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
651{
652 struct inet_sock *inet = inet_sk(sk);
653 struct inet_connection_sock *icsk = inet_csk(sk);
654 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
655
656 if (rc != 0)
657 return rc;
658
659 sk->sk_max_ack_backlog = 0;
660 sk->sk_ack_backlog = 0;
661 inet_csk_delack_init(sk);
662
663 /* There is race window here: we announce ourselves listening,
664 * but this transition is still not validated by get_port().
665 * It is OK, because this socket enters to hash table only
666 * after validation is complete.
667 */
668 sk->sk_state = TCP_LISTEN;
669 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
670 inet->inet_sport = htons(inet->inet_num);
671
672 sk_dst_reset(sk);
673 sk->sk_prot->hash(sk);
674
675 return 0;
676 }
677
678 sk->sk_state = TCP_CLOSE;
679 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
680 return -EADDRINUSE;
681}
682EXPORT_SYMBOL_GPL(inet_csk_listen_start);
683
684/*
685 * This routine closes sockets which have been at least partially
686 * opened, but not yet accepted.
687 */
688void inet_csk_listen_stop(struct sock *sk)
689{
690 struct inet_connection_sock *icsk = inet_csk(sk);
691 struct request_sock *acc_req;
692 struct request_sock *req;
693
694 inet_csk_delete_keepalive_timer(sk);
695
696 /* make all the listen_opt local to us */
697 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
698
699 /* Following specs, it would be better either to send FIN
700 * (and enter FIN-WAIT-1, it is normal close)
701 * or to send active reset (abort).
702 * Certainly, it is pretty dangerous while synflood, but it is
703 * bad justification for our negligence 8)
704 * To be honest, we are not able to make either
705 * of the variants now. --ANK
706 */
707 reqsk_queue_destroy(&icsk->icsk_accept_queue);
708
709 while ((req = acc_req) != NULL) {
710 struct sock *child = req->sk;
711
712 acc_req = req->dl_next;
713
714 local_bh_disable();
715 bh_lock_sock(child);
716 WARN_ON(sock_owned_by_user(child));
717 sock_hold(child);
718
719 sk->sk_prot->disconnect(child, O_NONBLOCK);
720
721 sock_orphan(child);
722
723 percpu_counter_inc(sk->sk_prot->orphan_count);
724
725 inet_csk_destroy_sock(child);
726
727 bh_unlock_sock(child);
728 local_bh_enable();
729 sock_put(child);
730
731 sk_acceptq_removed(sk);
732 __reqsk_free(req);
733 }
734 WARN_ON(sk->sk_ack_backlog);
735}
736EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
737
738void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
739{
740 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
741 const struct inet_sock *inet = inet_sk(sk);
742
743 sin->sin_family = AF_INET;
744 sin->sin_addr.s_addr = inet->inet_daddr;
745 sin->sin_port = inet->inet_dport;
746}
747EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
748
749#ifdef CONFIG_COMPAT
750int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
751 char __user *optval, int __user *optlen)
752{
753 const struct inet_connection_sock *icsk = inet_csk(sk);
754
755 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
756 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
757 optval, optlen);
758 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
759 optval, optlen);
760}
761EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
762
763int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
764 char __user *optval, unsigned int optlen)
765{
766 const struct inet_connection_sock *icsk = inet_csk(sk);
767
768 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
769 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
770 optval, optlen);
771 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
772 optval, optlen);
773}
774EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
775#endif
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Support for INET connection oriented protocols.
7 *
8 * Authors: See the TCP sources
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or(at your option) any later version.
14 */
15
16#include <linux/module.h>
17#include <linux/jhash.h>
18
19#include <net/inet_connection_sock.h>
20#include <net/inet_hashtables.h>
21#include <net/inet_timewait_sock.h>
22#include <net/ip.h>
23#include <net/route.h>
24#include <net/tcp_states.h>
25#include <net/xfrm.h>
26#include <net/tcp.h>
27#include <net/sock_reuseport.h>
28
29#ifdef INET_CSK_DEBUG
30const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
31EXPORT_SYMBOL(inet_csk_timer_bug_msg);
32#endif
33
34void inet_get_local_port_range(struct net *net, int *low, int *high)
35{
36 unsigned int seq;
37
38 do {
39 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
40
41 *low = net->ipv4.ip_local_ports.range[0];
42 *high = net->ipv4.ip_local_ports.range[1];
43 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
44}
45EXPORT_SYMBOL(inet_get_local_port_range);
46
47int inet_csk_bind_conflict(const struct sock *sk,
48 const struct inet_bind_bucket *tb, bool relax)
49{
50 struct sock *sk2;
51 int reuse = sk->sk_reuse;
52 int reuseport = sk->sk_reuseport;
53 kuid_t uid = sock_i_uid((struct sock *)sk);
54
55 /*
56 * Unlike other sk lookup places we do not check
57 * for sk_net here, since _all_ the socks listed
58 * in tb->owners list belong to the same net - the
59 * one this bucket belongs to.
60 */
61
62 sk_for_each_bound(sk2, &tb->owners) {
63 if (sk != sk2 &&
64 !inet_v6_ipv6only(sk2) &&
65 (!sk->sk_bound_dev_if ||
66 !sk2->sk_bound_dev_if ||
67 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
68 if ((!reuse || !sk2->sk_reuse ||
69 sk2->sk_state == TCP_LISTEN) &&
70 (!reuseport || !sk2->sk_reuseport ||
71 rcu_access_pointer(sk->sk_reuseport_cb) ||
72 (sk2->sk_state != TCP_TIME_WAIT &&
73 !uid_eq(uid, sock_i_uid(sk2))))) {
74
75 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
76 sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
77 break;
78 }
79 if (!relax && reuse && sk2->sk_reuse &&
80 sk2->sk_state != TCP_LISTEN) {
81
82 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
83 sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
84 break;
85 }
86 }
87 }
88 return sk2 != NULL;
89}
90EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
91
92/* Obtain a reference to a local port for the given sock,
93 * if snum is zero it means select any available local port.
94 * We try to allocate an odd port (and leave even ports for connect())
95 */
96int inet_csk_get_port(struct sock *sk, unsigned short snum)
97{
98 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
99 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
100 int ret = 1, attempts = 5, port = snum;
101 int smallest_size = -1, smallest_port;
102 struct inet_bind_hashbucket *head;
103 struct net *net = sock_net(sk);
104 int i, low, high, attempt_half;
105 struct inet_bind_bucket *tb;
106 kuid_t uid = sock_i_uid(sk);
107 u32 remaining, offset;
108
109 if (port) {
110have_port:
111 head = &hinfo->bhash[inet_bhashfn(net, port,
112 hinfo->bhash_size)];
113 spin_lock_bh(&head->lock);
114 inet_bind_bucket_for_each(tb, &head->chain)
115 if (net_eq(ib_net(tb), net) && tb->port == port)
116 goto tb_found;
117
118 goto tb_not_found;
119 }
120again:
121 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
122other_half_scan:
123 inet_get_local_port_range(net, &low, &high);
124 high++; /* [32768, 60999] -> [32768, 61000[ */
125 if (high - low < 4)
126 attempt_half = 0;
127 if (attempt_half) {
128 int half = low + (((high - low) >> 2) << 1);
129
130 if (attempt_half == 1)
131 high = half;
132 else
133 low = half;
134 }
135 remaining = high - low;
136 if (likely(remaining > 1))
137 remaining &= ~1U;
138
139 offset = prandom_u32() % remaining;
140 /* __inet_hash_connect() favors ports having @low parity
141 * We do the opposite to not pollute connect() users.
142 */
143 offset |= 1U;
144 smallest_size = -1;
145 smallest_port = low; /* avoid compiler warning */
146
147other_parity_scan:
148 port = low + offset;
149 for (i = 0; i < remaining; i += 2, port += 2) {
150 if (unlikely(port >= high))
151 port -= remaining;
152 if (inet_is_local_reserved_port(net, port))
153 continue;
154 head = &hinfo->bhash[inet_bhashfn(net, port,
155 hinfo->bhash_size)];
156 spin_lock_bh(&head->lock);
157 inet_bind_bucket_for_each(tb, &head->chain)
158 if (net_eq(ib_net(tb), net) && tb->port == port) {
159 if (((tb->fastreuse > 0 && reuse) ||
160 (tb->fastreuseport > 0 &&
161 sk->sk_reuseport &&
162 !rcu_access_pointer(sk->sk_reuseport_cb) &&
163 uid_eq(tb->fastuid, uid))) &&
164 (tb->num_owners < smallest_size || smallest_size == -1)) {
165 smallest_size = tb->num_owners;
166 smallest_port = port;
167 }
168 if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false))
169 goto tb_found;
170 goto next_port;
171 }
172 goto tb_not_found;
173next_port:
174 spin_unlock_bh(&head->lock);
175 cond_resched();
176 }
177
178 if (smallest_size != -1) {
179 port = smallest_port;
180 goto have_port;
181 }
182 offset--;
183 if (!(offset & 1))
184 goto other_parity_scan;
185
186 if (attempt_half == 1) {
187 /* OK we now try the upper half of the range */
188 attempt_half = 2;
189 goto other_half_scan;
190 }
191 return ret;
192
193tb_not_found:
194 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
195 net, head, port);
196 if (!tb)
197 goto fail_unlock;
198tb_found:
199 if (!hlist_empty(&tb->owners)) {
200 if (sk->sk_reuse == SK_FORCE_REUSE)
201 goto success;
202
203 if (((tb->fastreuse > 0 && reuse) ||
204 (tb->fastreuseport > 0 &&
205 !rcu_access_pointer(sk->sk_reuseport_cb) &&
206 sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
207 smallest_size == -1)
208 goto success;
209 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
210 if ((reuse ||
211 (tb->fastreuseport > 0 &&
212 sk->sk_reuseport &&
213 !rcu_access_pointer(sk->sk_reuseport_cb) &&
214 uid_eq(tb->fastuid, uid))) &&
215 smallest_size != -1 && --attempts >= 0) {
216 spin_unlock_bh(&head->lock);
217 goto again;
218 }
219 goto fail_unlock;
220 }
221 if (!reuse)
222 tb->fastreuse = 0;
223 if (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid))
224 tb->fastreuseport = 0;
225 } else {
226 tb->fastreuse = reuse;
227 if (sk->sk_reuseport) {
228 tb->fastreuseport = 1;
229 tb->fastuid = uid;
230 } else {
231 tb->fastreuseport = 0;
232 }
233 }
234success:
235 if (!inet_csk(sk)->icsk_bind_hash)
236 inet_bind_hash(sk, tb, port);
237 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
238 ret = 0;
239
240fail_unlock:
241 spin_unlock_bh(&head->lock);
242 return ret;
243}
244EXPORT_SYMBOL_GPL(inet_csk_get_port);
245
246/*
247 * Wait for an incoming connection, avoid race conditions. This must be called
248 * with the socket locked.
249 */
250static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
251{
252 struct inet_connection_sock *icsk = inet_csk(sk);
253 DEFINE_WAIT(wait);
254 int err;
255
256 /*
257 * True wake-one mechanism for incoming connections: only
258 * one process gets woken up, not the 'whole herd'.
259 * Since we do not 'race & poll' for established sockets
260 * anymore, the common case will execute the loop only once.
261 *
262 * Subtle issue: "add_wait_queue_exclusive()" will be added
263 * after any current non-exclusive waiters, and we know that
264 * it will always _stay_ after any new non-exclusive waiters
265 * because all non-exclusive waiters are added at the
266 * beginning of the wait-queue. As such, it's ok to "drop"
267 * our exclusiveness temporarily when we get woken up without
268 * having to remove and re-insert us on the wait queue.
269 */
270 for (;;) {
271 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
272 TASK_INTERRUPTIBLE);
273 release_sock(sk);
274 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
275 timeo = schedule_timeout(timeo);
276 sched_annotate_sleep();
277 lock_sock(sk);
278 err = 0;
279 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
280 break;
281 err = -EINVAL;
282 if (sk->sk_state != TCP_LISTEN)
283 break;
284 err = sock_intr_errno(timeo);
285 if (signal_pending(current))
286 break;
287 err = -EAGAIN;
288 if (!timeo)
289 break;
290 }
291 finish_wait(sk_sleep(sk), &wait);
292 return err;
293}
294
295/*
296 * This will accept the next outstanding connection.
297 */
298struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
299{
300 struct inet_connection_sock *icsk = inet_csk(sk);
301 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
302 struct request_sock *req;
303 struct sock *newsk;
304 int error;
305
306 lock_sock(sk);
307
308 /* We need to make sure that this socket is listening,
309 * and that it has something pending.
310 */
311 error = -EINVAL;
312 if (sk->sk_state != TCP_LISTEN)
313 goto out_err;
314
315 /* Find already established connection */
316 if (reqsk_queue_empty(queue)) {
317 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
318
319 /* If this is a non blocking socket don't sleep */
320 error = -EAGAIN;
321 if (!timeo)
322 goto out_err;
323
324 error = inet_csk_wait_for_connect(sk, timeo);
325 if (error)
326 goto out_err;
327 }
328 req = reqsk_queue_remove(queue, sk);
329 newsk = req->sk;
330
331 if (sk->sk_protocol == IPPROTO_TCP &&
332 tcp_rsk(req)->tfo_listener) {
333 spin_lock_bh(&queue->fastopenq.lock);
334 if (tcp_rsk(req)->tfo_listener) {
335 /* We are still waiting for the final ACK from 3WHS
336 * so can't free req now. Instead, we set req->sk to
337 * NULL to signify that the child socket is taken
338 * so reqsk_fastopen_remove() will free the req
339 * when 3WHS finishes (or is aborted).
340 */
341 req->sk = NULL;
342 req = NULL;
343 }
344 spin_unlock_bh(&queue->fastopenq.lock);
345 }
346out:
347 release_sock(sk);
348 if (req)
349 reqsk_put(req);
350 return newsk;
351out_err:
352 newsk = NULL;
353 req = NULL;
354 *err = error;
355 goto out;
356}
357EXPORT_SYMBOL(inet_csk_accept);
358
359/*
360 * Using different timers for retransmit, delayed acks and probes
361 * We may wish use just one timer maintaining a list of expire jiffies
362 * to optimize.
363 */
364void inet_csk_init_xmit_timers(struct sock *sk,
365 void (*retransmit_handler)(unsigned long),
366 void (*delack_handler)(unsigned long),
367 void (*keepalive_handler)(unsigned long))
368{
369 struct inet_connection_sock *icsk = inet_csk(sk);
370
371 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
372 (unsigned long)sk);
373 setup_timer(&icsk->icsk_delack_timer, delack_handler,
374 (unsigned long)sk);
375 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
376 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
377}
378EXPORT_SYMBOL(inet_csk_init_xmit_timers);
379
380void inet_csk_clear_xmit_timers(struct sock *sk)
381{
382 struct inet_connection_sock *icsk = inet_csk(sk);
383
384 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
385
386 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
387 sk_stop_timer(sk, &icsk->icsk_delack_timer);
388 sk_stop_timer(sk, &sk->sk_timer);
389}
390EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
391
392void inet_csk_delete_keepalive_timer(struct sock *sk)
393{
394 sk_stop_timer(sk, &sk->sk_timer);
395}
396EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
397
398void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
399{
400 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
401}
402EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
403
404struct dst_entry *inet_csk_route_req(const struct sock *sk,
405 struct flowi4 *fl4,
406 const struct request_sock *req)
407{
408 const struct inet_request_sock *ireq = inet_rsk(req);
409 struct net *net = read_pnet(&ireq->ireq_net);
410 struct ip_options_rcu *opt = ireq->opt;
411 struct rtable *rt;
412
413 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
414 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
415 sk->sk_protocol, inet_sk_flowi_flags(sk),
416 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
417 ireq->ir_loc_addr, ireq->ir_rmt_port,
418 htons(ireq->ir_num));
419 security_req_classify_flow(req, flowi4_to_flowi(fl4));
420 rt = ip_route_output_flow(net, fl4, sk);
421 if (IS_ERR(rt))
422 goto no_route;
423 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
424 goto route_err;
425 return &rt->dst;
426
427route_err:
428 ip_rt_put(rt);
429no_route:
430 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
431 return NULL;
432}
433EXPORT_SYMBOL_GPL(inet_csk_route_req);
434
435struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
436 struct sock *newsk,
437 const struct request_sock *req)
438{
439 const struct inet_request_sock *ireq = inet_rsk(req);
440 struct net *net = read_pnet(&ireq->ireq_net);
441 struct inet_sock *newinet = inet_sk(newsk);
442 struct ip_options_rcu *opt;
443 struct flowi4 *fl4;
444 struct rtable *rt;
445
446 fl4 = &newinet->cork.fl.u.ip4;
447
448 rcu_read_lock();
449 opt = rcu_dereference(newinet->inet_opt);
450 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
451 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
452 sk->sk_protocol, inet_sk_flowi_flags(sk),
453 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
454 ireq->ir_loc_addr, ireq->ir_rmt_port,
455 htons(ireq->ir_num));
456 security_req_classify_flow(req, flowi4_to_flowi(fl4));
457 rt = ip_route_output_flow(net, fl4, sk);
458 if (IS_ERR(rt))
459 goto no_route;
460 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
461 goto route_err;
462 rcu_read_unlock();
463 return &rt->dst;
464
465route_err:
466 ip_rt_put(rt);
467no_route:
468 rcu_read_unlock();
469 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
470 return NULL;
471}
472EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
473
474#if IS_ENABLED(CONFIG_IPV6)
475#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
476#else
477#define AF_INET_FAMILY(fam) true
478#endif
479
480/* Decide when to expire the request and when to resend SYN-ACK */
481static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
482 const int max_retries,
483 const u8 rskq_defer_accept,
484 int *expire, int *resend)
485{
486 if (!rskq_defer_accept) {
487 *expire = req->num_timeout >= thresh;
488 *resend = 1;
489 return;
490 }
491 *expire = req->num_timeout >= thresh &&
492 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
493 /*
494 * Do not resend while waiting for data after ACK,
495 * start to resend on end of deferring period to give
496 * last chance for data or ACK to create established socket.
497 */
498 *resend = !inet_rsk(req)->acked ||
499 req->num_timeout >= rskq_defer_accept - 1;
500}
501
502int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
503{
504 int err = req->rsk_ops->rtx_syn_ack(parent, req);
505
506 if (!err)
507 req->num_retrans++;
508 return err;
509}
510EXPORT_SYMBOL(inet_rtx_syn_ack);
511
512/* return true if req was found in the ehash table */
513static bool reqsk_queue_unlink(struct request_sock_queue *queue,
514 struct request_sock *req)
515{
516 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
517 bool found = false;
518
519 if (sk_hashed(req_to_sk(req))) {
520 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
521
522 spin_lock(lock);
523 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
524 spin_unlock(lock);
525 }
526 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
527 reqsk_put(req);
528 return found;
529}
530
531void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
532{
533 if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
534 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
535 reqsk_put(req);
536 }
537}
538EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
539
540void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
541{
542 inet_csk_reqsk_queue_drop(sk, req);
543 reqsk_put(req);
544}
545EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
546
547static void reqsk_timer_handler(unsigned long data)
548{
549 struct request_sock *req = (struct request_sock *)data;
550 struct sock *sk_listener = req->rsk_listener;
551 struct net *net = sock_net(sk_listener);
552 struct inet_connection_sock *icsk = inet_csk(sk_listener);
553 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
554 int qlen, expire = 0, resend = 0;
555 int max_retries, thresh;
556 u8 defer_accept;
557
558 if (sk_state_load(sk_listener) != TCP_LISTEN)
559 goto drop;
560
561 max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
562 thresh = max_retries;
563 /* Normally all the openreqs are young and become mature
564 * (i.e. converted to established socket) for first timeout.
565 * If synack was not acknowledged for 1 second, it means
566 * one of the following things: synack was lost, ack was lost,
567 * rtt is high or nobody planned to ack (i.e. synflood).
568 * When server is a bit loaded, queue is populated with old
569 * open requests, reducing effective size of queue.
570 * When server is well loaded, queue size reduces to zero
571 * after several minutes of work. It is not synflood,
572 * it is normal operation. The solution is pruning
573 * too old entries overriding normal timeout, when
574 * situation becomes dangerous.
575 *
576 * Essentially, we reserve half of room for young
577 * embrions; and abort old ones without pity, if old
578 * ones are about to clog our table.
579 */
580 qlen = reqsk_queue_len(queue);
581 if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
582 int young = reqsk_queue_len_young(queue) << 1;
583
584 while (thresh > 2) {
585 if (qlen < young)
586 break;
587 thresh--;
588 young <<= 1;
589 }
590 }
591 defer_accept = READ_ONCE(queue->rskq_defer_accept);
592 if (defer_accept)
593 max_retries = defer_accept;
594 syn_ack_recalc(req, thresh, max_retries, defer_accept,
595 &expire, &resend);
596 req->rsk_ops->syn_ack_timeout(req);
597 if (!expire &&
598 (!resend ||
599 !inet_rtx_syn_ack(sk_listener, req) ||
600 inet_rsk(req)->acked)) {
601 unsigned long timeo;
602
603 if (req->num_timeout++ == 0)
604 atomic_dec(&queue->young);
605 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
606 mod_timer_pinned(&req->rsk_timer, jiffies + timeo);
607 return;
608 }
609drop:
610 inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
611}
612
613static void reqsk_queue_hash_req(struct request_sock *req,
614 unsigned long timeout)
615{
616 req->num_retrans = 0;
617 req->num_timeout = 0;
618 req->sk = NULL;
619
620 setup_timer(&req->rsk_timer, reqsk_timer_handler, (unsigned long)req);
621 mod_timer_pinned(&req->rsk_timer, jiffies + timeout);
622
623 inet_ehash_insert(req_to_sk(req), NULL);
624 /* before letting lookups find us, make sure all req fields
625 * are committed to memory and refcnt initialized.
626 */
627 smp_wmb();
628 atomic_set(&req->rsk_refcnt, 2 + 1);
629}
630
631void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
632 unsigned long timeout)
633{
634 reqsk_queue_hash_req(req, timeout);
635 inet_csk_reqsk_queue_added(sk);
636}
637EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
638
639/**
640 * inet_csk_clone_lock - clone an inet socket, and lock its clone
641 * @sk: the socket to clone
642 * @req: request_sock
643 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
644 *
645 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
646 */
647struct sock *inet_csk_clone_lock(const struct sock *sk,
648 const struct request_sock *req,
649 const gfp_t priority)
650{
651 struct sock *newsk = sk_clone_lock(sk, priority);
652
653 if (newsk) {
654 struct inet_connection_sock *newicsk = inet_csk(newsk);
655
656 newsk->sk_state = TCP_SYN_RECV;
657 newicsk->icsk_bind_hash = NULL;
658
659 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
660 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
661 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
662 newsk->sk_write_space = sk_stream_write_space;
663
664 newsk->sk_mark = inet_rsk(req)->ir_mark;
665 atomic64_set(&newsk->sk_cookie,
666 atomic64_read(&inet_rsk(req)->ir_cookie));
667
668 newicsk->icsk_retransmits = 0;
669 newicsk->icsk_backoff = 0;
670 newicsk->icsk_probes_out = 0;
671
672 /* Deinitialize accept_queue to trap illegal accesses. */
673 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
674
675 security_inet_csk_clone(newsk, req);
676 }
677 return newsk;
678}
679EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
680
681/*
682 * At this point, there should be no process reference to this
683 * socket, and thus no user references at all. Therefore we
684 * can assume the socket waitqueue is inactive and nobody will
685 * try to jump onto it.
686 */
687void inet_csk_destroy_sock(struct sock *sk)
688{
689 WARN_ON(sk->sk_state != TCP_CLOSE);
690 WARN_ON(!sock_flag(sk, SOCK_DEAD));
691
692 /* It cannot be in hash table! */
693 WARN_ON(!sk_unhashed(sk));
694
695 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
696 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
697
698 sk->sk_prot->destroy(sk);
699
700 sk_stream_kill_queues(sk);
701
702 xfrm_sk_free_policy(sk);
703
704 sk_refcnt_debug_release(sk);
705
706 percpu_counter_dec(sk->sk_prot->orphan_count);
707 sock_put(sk);
708}
709EXPORT_SYMBOL(inet_csk_destroy_sock);
710
711/* This function allows to force a closure of a socket after the call to
712 * tcp/dccp_create_openreq_child().
713 */
714void inet_csk_prepare_forced_close(struct sock *sk)
715 __releases(&sk->sk_lock.slock)
716{
717 /* sk_clone_lock locked the socket and set refcnt to 2 */
718 bh_unlock_sock(sk);
719 sock_put(sk);
720
721 /* The below has to be done to allow calling inet_csk_destroy_sock */
722 sock_set_flag(sk, SOCK_DEAD);
723 percpu_counter_inc(sk->sk_prot->orphan_count);
724 inet_sk(sk)->inet_num = 0;
725}
726EXPORT_SYMBOL(inet_csk_prepare_forced_close);
727
728int inet_csk_listen_start(struct sock *sk, int backlog)
729{
730 struct inet_connection_sock *icsk = inet_csk(sk);
731 struct inet_sock *inet = inet_sk(sk);
732 int err = -EADDRINUSE;
733
734 reqsk_queue_alloc(&icsk->icsk_accept_queue);
735
736 sk->sk_max_ack_backlog = backlog;
737 sk->sk_ack_backlog = 0;
738 inet_csk_delack_init(sk);
739
740 /* There is race window here: we announce ourselves listening,
741 * but this transition is still not validated by get_port().
742 * It is OK, because this socket enters to hash table only
743 * after validation is complete.
744 */
745 sk_state_store(sk, TCP_LISTEN);
746 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
747 inet->inet_sport = htons(inet->inet_num);
748
749 sk_dst_reset(sk);
750 err = sk->sk_prot->hash(sk);
751
752 if (likely(!err))
753 return 0;
754 }
755
756 sk->sk_state = TCP_CLOSE;
757 return err;
758}
759EXPORT_SYMBOL_GPL(inet_csk_listen_start);
760
761static void inet_child_forget(struct sock *sk, struct request_sock *req,
762 struct sock *child)
763{
764 sk->sk_prot->disconnect(child, O_NONBLOCK);
765
766 sock_orphan(child);
767
768 percpu_counter_inc(sk->sk_prot->orphan_count);
769
770 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
771 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
772 BUG_ON(sk != req->rsk_listener);
773
774 /* Paranoid, to prevent race condition if
775 * an inbound pkt destined for child is
776 * blocked by sock lock in tcp_v4_rcv().
777 * Also to satisfy an assertion in
778 * tcp_v4_destroy_sock().
779 */
780 tcp_sk(child)->fastopen_rsk = NULL;
781 }
782 inet_csk_destroy_sock(child);
783 reqsk_put(req);
784}
785
786struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
787 struct request_sock *req,
788 struct sock *child)
789{
790 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
791
792 spin_lock(&queue->rskq_lock);
793 if (unlikely(sk->sk_state != TCP_LISTEN)) {
794 inet_child_forget(sk, req, child);
795 child = NULL;
796 } else {
797 req->sk = child;
798 req->dl_next = NULL;
799 if (queue->rskq_accept_head == NULL)
800 queue->rskq_accept_head = req;
801 else
802 queue->rskq_accept_tail->dl_next = req;
803 queue->rskq_accept_tail = req;
804 sk_acceptq_added(sk);
805 }
806 spin_unlock(&queue->rskq_lock);
807 return child;
808}
809EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
810
811struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
812 struct request_sock *req, bool own_req)
813{
814 if (own_req) {
815 inet_csk_reqsk_queue_drop(sk, req);
816 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
817 if (inet_csk_reqsk_queue_add(sk, req, child))
818 return child;
819 }
820 /* Too bad, another child took ownership of the request, undo. */
821 bh_unlock_sock(child);
822 sock_put(child);
823 return NULL;
824}
825EXPORT_SYMBOL(inet_csk_complete_hashdance);
826
827/*
828 * This routine closes sockets which have been at least partially
829 * opened, but not yet accepted.
830 */
831void inet_csk_listen_stop(struct sock *sk)
832{
833 struct inet_connection_sock *icsk = inet_csk(sk);
834 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
835 struct request_sock *next, *req;
836
837 /* Following specs, it would be better either to send FIN
838 * (and enter FIN-WAIT-1, it is normal close)
839 * or to send active reset (abort).
840 * Certainly, it is pretty dangerous while synflood, but it is
841 * bad justification for our negligence 8)
842 * To be honest, we are not able to make either
843 * of the variants now. --ANK
844 */
845 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
846 struct sock *child = req->sk;
847
848 local_bh_disable();
849 bh_lock_sock(child);
850 WARN_ON(sock_owned_by_user(child));
851 sock_hold(child);
852
853 inet_child_forget(sk, req, child);
854 bh_unlock_sock(child);
855 local_bh_enable();
856 sock_put(child);
857
858 cond_resched();
859 }
860 if (queue->fastopenq.rskq_rst_head) {
861 /* Free all the reqs queued in rskq_rst_head. */
862 spin_lock_bh(&queue->fastopenq.lock);
863 req = queue->fastopenq.rskq_rst_head;
864 queue->fastopenq.rskq_rst_head = NULL;
865 spin_unlock_bh(&queue->fastopenq.lock);
866 while (req != NULL) {
867 next = req->dl_next;
868 reqsk_put(req);
869 req = next;
870 }
871 }
872 WARN_ON_ONCE(sk->sk_ack_backlog);
873}
874EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
875
876void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
877{
878 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
879 const struct inet_sock *inet = inet_sk(sk);
880
881 sin->sin_family = AF_INET;
882 sin->sin_addr.s_addr = inet->inet_daddr;
883 sin->sin_port = inet->inet_dport;
884}
885EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
886
887#ifdef CONFIG_COMPAT
888int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
889 char __user *optval, int __user *optlen)
890{
891 const struct inet_connection_sock *icsk = inet_csk(sk);
892
893 if (icsk->icsk_af_ops->compat_getsockopt)
894 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
895 optval, optlen);
896 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
897 optval, optlen);
898}
899EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
900
901int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
902 char __user *optval, unsigned int optlen)
903{
904 const struct inet_connection_sock *icsk = inet_csk(sk);
905
906 if (icsk->icsk_af_ops->compat_setsockopt)
907 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
908 optval, optlen);
909 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
910 optval, optlen);
911}
912EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
913#endif
914
915static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
916{
917 const struct inet_sock *inet = inet_sk(sk);
918 const struct ip_options_rcu *inet_opt;
919 __be32 daddr = inet->inet_daddr;
920 struct flowi4 *fl4;
921 struct rtable *rt;
922
923 rcu_read_lock();
924 inet_opt = rcu_dereference(inet->inet_opt);
925 if (inet_opt && inet_opt->opt.srr)
926 daddr = inet_opt->opt.faddr;
927 fl4 = &fl->u.ip4;
928 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
929 inet->inet_saddr, inet->inet_dport,
930 inet->inet_sport, sk->sk_protocol,
931 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
932 if (IS_ERR(rt))
933 rt = NULL;
934 if (rt)
935 sk_setup_caps(sk, &rt->dst);
936 rcu_read_unlock();
937
938 return &rt->dst;
939}
940
941struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
942{
943 struct dst_entry *dst = __sk_dst_check(sk, 0);
944 struct inet_sock *inet = inet_sk(sk);
945
946 if (!dst) {
947 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
948 if (!dst)
949 goto out;
950 }
951 dst->ops->update_pmtu(dst, sk, NULL, mtu);
952
953 dst = __sk_dst_check(sk, 0);
954 if (!dst)
955 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
956out:
957 return dst;
958}
959EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);