projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net ipv4: Convert ipv4.ip_local_port_range to be per netns v3
[deliverable/linux.git] / net / ipv4 / inet_connection_sock.c
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
28 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
29 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
30 #endif
31
32 unsigned long *sysctl_local_reserved_ports;
33 EXPORT_SYMBOL(sysctl_local_reserved_ports);
34
35 void inet_get_local_port_range(struct net *net, int *low, int *high)
36 {
37 unsigned int seq;
38
39 do {
40 seq = read_seqbegin(&net->ipv4.sysctl_local_ports.lock);
41
42 *low = net->ipv4.sysctl_local_ports.range[0];
43 *high = net->ipv4.sysctl_local_ports.range[1];
44 } while (read_seqretry(&net->ipv4.sysctl_local_ports.lock, seq));
45 }
46 EXPORT_SYMBOL(inet_get_local_port_range);
47
48 int inet_csk_bind_conflict(const struct sock *sk,
49 const struct inet_bind_bucket *tb, bool relax)
50 {
51 struct sock *sk2;
52 int reuse = sk->sk_reuse;
53 int reuseport = sk->sk_reuseport;
54 kuid_t uid = sock_i_uid((struct sock *)sk);
55
56 /*
57 * Unlike other sk lookup places we do not check
58 * for sk_net here, since _all_ the socks listed
59 * in tb->owners list belong to the same net - the
60 * one this bucket belongs to.
61 */
62
63 sk_for_each_bound(sk2, &tb->owners) {
64 if (sk != sk2 &&
65 !inet_v6_ipv6only(sk2) &&
66 (!sk->sk_bound_dev_if ||
67 !sk2->sk_bound_dev_if ||
68 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
69 if ((!reuse || !sk2->sk_reuse ||
70 sk2->sk_state == TCP_LISTEN) &&
71 (!reuseport || !sk2->sk_reuseport ||
72 (sk2->sk_state != TCP_TIME_WAIT &&
73 !uid_eq(uid, sock_i_uid(sk2))))) {
74 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
75 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
76 sk2_rcv_saddr == sk_rcv_saddr(sk))
77 break;
78 }
79 if (!relax && reuse && sk2->sk_reuse &&
80 sk2->sk_state != TCP_LISTEN) {
81 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
82
83 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
84 sk2_rcv_saddr == sk_rcv_saddr(sk))
85 break;
86 }
87 }
88 }
89 return sk2 != NULL;
90 }
91 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
92
93 /* Obtain a reference to a local port for the given sock,
94 * if snum is zero it means select any available local port.
95 */
96 int inet_csk_get_port(struct sock *sk, unsigned short snum)
97 {
98 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
99 struct inet_bind_hashbucket *head;
100 struct inet_bind_bucket *tb;
101 int ret, attempts = 5;
102 struct net *net = sock_net(sk);
103 int smallest_size = -1, smallest_rover;
104 kuid_t uid = sock_i_uid(sk);
105
106 local_bh_disable();
107 if (!snum) {
108 int remaining, rover, low, high;
109
110 again:
111 inet_get_local_port_range(net, &low, &high);
112 remaining = (high - low) + 1;
113 smallest_rover = rover = net_random() % remaining + low;
114
115 smallest_size = -1;
116 do {
117 if (inet_is_reserved_local_port(rover))
118 goto next_nolock;
119 head = &hashinfo->bhash[inet_bhashfn(net, rover,
120 hashinfo->bhash_size)];
121 spin_lock(&head->lock);
122 inet_bind_bucket_for_each(tb, &head->chain)
123 if (net_eq(ib_net(tb), net) && tb->port == rover) {
124 if (((tb->fastreuse > 0 &&
125 sk->sk_reuse &&
126 sk->sk_state != TCP_LISTEN) ||
127 (tb->fastreuseport > 0 &&
128 sk->sk_reuseport &&
129 uid_eq(tb->fastuid, uid))) &&
130 (tb->num_owners < smallest_size || smallest_size == -1)) {
131 smallest_size = tb->num_owners;
132 smallest_rover = rover;
133 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&
134 !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
135 snum = smallest_rover;
136 goto tb_found;
137 }
138 }
139 if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
140 snum = rover;
141 goto tb_found;
142 }
143 goto next;
144 }
145 break;
146 next:
147 spin_unlock(&head->lock);
148 next_nolock:
149 if (++rover > high)
150 rover = low;
151 } while (--remaining > 0);
152
153 /* Exhausted local port range during search? It is not
154 * possible for us to be holding one of the bind hash
155 * locks if this test triggers, because if 'remaining'
156 * drops to zero, we broke out of the do/while loop at
157 * the top level, not from the 'break;' statement.
158 */
159 ret = 1;
160 if (remaining <= 0) {
161 if (smallest_size != -1) {
162 snum = smallest_rover;
163 goto have_snum;
164 }
165 goto fail;
166 }
167 /* OK, here is the one we will use. HEAD is
168 * non-NULL and we hold it's mutex.
169 */
170 snum = rover;
171 } else {
172 have_snum:
173 head = &hashinfo->bhash[inet_bhashfn(net, snum,
174 hashinfo->bhash_size)];
175 spin_lock(&head->lock);
176 inet_bind_bucket_for_each(tb, &head->chain)
177 if (net_eq(ib_net(tb), net) && tb->port == snum)
178 goto tb_found;
179 }
180 tb = NULL;
181 goto tb_not_found;
182 tb_found:
183 if (!hlist_empty(&tb->owners)) {
184 if (sk->sk_reuse == SK_FORCE_REUSE)
185 goto success;
186
187 if (((tb->fastreuse > 0 &&
188 sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
189 (tb->fastreuseport > 0 &&
190 sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
191 smallest_size == -1) {
192 goto success;
193 } else {
194 ret = 1;
195 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
196 if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
197 (tb->fastreuseport > 0 &&
198 sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
199 smallest_size != -1 && --attempts >= 0) {
200 spin_unlock(&head->lock);
201 goto again;
202 }
203
204 goto fail_unlock;
205 }
206 }
207 }
208 tb_not_found:
209 ret = 1;
210 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
211 net, head, snum)) == NULL)
212 goto fail_unlock;
213 if (hlist_empty(&tb->owners)) {
214 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
215 tb->fastreuse = 1;
216 else
217 tb->fastreuse = 0;
218 if (sk->sk_reuseport) {
219 tb->fastreuseport = 1;
220 tb->fastuid = uid;
221 } else
222 tb->fastreuseport = 0;
223 } else {
224 if (tb->fastreuse &&
225 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
226 tb->fastreuse = 0;
227 if (tb->fastreuseport &&
228 (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid)))
229 tb->fastreuseport = 0;
230 }
231 success:
232 if (!inet_csk(sk)->icsk_bind_hash)
233 inet_bind_hash(sk, tb, snum);
234 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
235 ret = 0;
236
237 fail_unlock:
238 spin_unlock(&head->lock);
239 fail:
240 local_bh_enable();
241 return ret;
242 }
243 EXPORT_SYMBOL_GPL(inet_csk_get_port);
244
245 /*
246 * Wait for an incoming connection, avoid race conditions. This must be called
247 * with the socket locked.
248 */
249 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
250 {
251 struct inet_connection_sock *icsk = inet_csk(sk);
252 DEFINE_WAIT(wait);
253 int err;
254
255 /*
256 * True wake-one mechanism for incoming connections: only
257 * one process gets woken up, not the 'whole herd'.
258 * Since we do not 'race & poll' for established sockets
259 * anymore, the common case will execute the loop only once.
260 *
261 * Subtle issue: "add_wait_queue_exclusive()" will be added
262 * after any current non-exclusive waiters, and we know that
263 * it will always _stay_ after any new non-exclusive waiters
264 * because all non-exclusive waiters are added at the
265 * beginning of the wait-queue. As such, it's ok to "drop"
266 * our exclusiveness temporarily when we get woken up without
267 * having to remove and re-insert us on the wait queue.
268 */
269 for (;;) {
270 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
271 TASK_INTERRUPTIBLE);
272 release_sock(sk);
273 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
274 timeo = schedule_timeout(timeo);
275 lock_sock(sk);
276 err = 0;
277 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
278 break;
279 err = -EINVAL;
280 if (sk->sk_state != TCP_LISTEN)
281 break;
282 err = sock_intr_errno(timeo);
283 if (signal_pending(current))
284 break;
285 err = -EAGAIN;
286 if (!timeo)
287 break;
288 }
289 finish_wait(sk_sleep(sk), &wait);
290 return err;
291 }
292
293 /*
294 * This will accept the next outstanding connection.
295 */
296 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
297 {
298 struct inet_connection_sock *icsk = inet_csk(sk);
299 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
300 struct sock *newsk;
301 struct request_sock *req;
302 int error;
303
304 lock_sock(sk);
305
306 /* We need to make sure that this socket is listening,
307 * and that it has something pending.
308 */
309 error = -EINVAL;
310 if (sk->sk_state != TCP_LISTEN)
311 goto out_err;
312
313 /* Find already established connection */
314 if (reqsk_queue_empty(queue)) {
315 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
316
317 /* If this is a non blocking socket don't sleep */
318 error = -EAGAIN;
319 if (!timeo)
320 goto out_err;
321
322 error = inet_csk_wait_for_connect(sk, timeo);
323 if (error)
324 goto out_err;
325 }
326 req = reqsk_queue_remove(queue);
327 newsk = req->sk;
328
329 sk_acceptq_removed(sk);
330 if (sk->sk_protocol == IPPROTO_TCP && queue->fastopenq != NULL) {
331 spin_lock_bh(&queue->fastopenq->lock);
332 if (tcp_rsk(req)->listener) {
333 /* We are still waiting for the final ACK from 3WHS
334 * so can't free req now. Instead, we set req->sk to
335 * NULL to signify that the child socket is taken
336 * so reqsk_fastopen_remove() will free the req
337 * when 3WHS finishes (or is aborted).
338 */
339 req->sk = NULL;
340 req = NULL;
341 }
342 spin_unlock_bh(&queue->fastopenq->lock);
343 }
344 out:
345 release_sock(sk);
346 if (req)
347 __reqsk_free(req);
348 return newsk;
349 out_err:
350 newsk = NULL;
351 req = NULL;
352 *err = error;
353 goto out;
354 }
355 EXPORT_SYMBOL(inet_csk_accept);
356
357 /*
358 * Using different timers for retransmit, delayed acks and probes
359 * We may wish use just one timer maintaining a list of expire jiffies
360 * to optimize.
361 */
362 void inet_csk_init_xmit_timers(struct sock *sk,
363 void (*retransmit_handler)(unsigned long),
364 void (*delack_handler)(unsigned long),
365 void (*keepalive_handler)(unsigned long))
366 {
367 struct inet_connection_sock *icsk = inet_csk(sk);
368
369 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
370 (unsigned long)sk);
371 setup_timer(&icsk->icsk_delack_timer, delack_handler,
372 (unsigned long)sk);
373 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
374 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
375 }
376 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
377
378 void inet_csk_clear_xmit_timers(struct sock *sk)
379 {
380 struct inet_connection_sock *icsk = inet_csk(sk);
381
382 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
383
384 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
385 sk_stop_timer(sk, &icsk->icsk_delack_timer);
386 sk_stop_timer(sk, &sk->sk_timer);
387 }
388 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
389
390 void inet_csk_delete_keepalive_timer(struct sock *sk)
391 {
392 sk_stop_timer(sk, &sk->sk_timer);
393 }
394 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
395
396 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
397 {
398 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
399 }
400 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
401
402 struct dst_entry *inet_csk_route_req(struct sock *sk,
403 struct flowi4 *fl4,
404 const struct request_sock *req)
405 {
406 struct rtable *rt;
407 const struct inet_request_sock *ireq = inet_rsk(req);
408 struct ip_options_rcu *opt = inet_rsk(req)->opt;
409 struct net *net = sock_net(sk);
410 int flags = inet_sk_flowi_flags(sk);
411
412 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
413 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
414 sk->sk_protocol,
415 flags,
416 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
417 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
418 security_req_classify_flow(req, flowi4_to_flowi(fl4));
419 rt = ip_route_output_flow(net, fl4, sk);
420 if (IS_ERR(rt))
421 goto no_route;
422 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
423 goto route_err;
424 return &rt->dst;
425
426 route_err:
427 ip_rt_put(rt);
428 no_route:
429 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
430 return NULL;
431 }
432 EXPORT_SYMBOL_GPL(inet_csk_route_req);
433
434 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
435 struct sock *newsk,
436 const struct request_sock *req)
437 {
438 const struct inet_request_sock *ireq = inet_rsk(req);
439 struct inet_sock *newinet = inet_sk(newsk);
440 struct ip_options_rcu *opt;
441 struct net *net = sock_net(sk);
442 struct flowi4 *fl4;
443 struct rtable *rt;
444
445 fl4 = &newinet->cork.fl.u.ip4;
446
447 rcu_read_lock();
448 opt = rcu_dereference(newinet->inet_opt);
449 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
450 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
451 sk->sk_protocol, inet_sk_flowi_flags(sk),
452 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
453 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
454 security_req_classify_flow(req, flowi4_to_flowi(fl4));
455 rt = ip_route_output_flow(net, fl4, sk);
456 if (IS_ERR(rt))
457 goto no_route;
458 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
459 goto route_err;
460 rcu_read_unlock();
461 return &rt->dst;
462
463 route_err:
464 ip_rt_put(rt);
465 no_route:
466 rcu_read_unlock();
467 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
468 return NULL;
469 }
470 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
471
472 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
473 const u32 rnd, const u32 synq_hsize)
474 {
475 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
476 }
477
478 #if IS_ENABLED(CONFIG_IPV6)
479 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
480 #else
481 #define AF_INET_FAMILY(fam) 1
482 #endif
483
484 struct request_sock *inet_csk_search_req(const struct sock *sk,
485 struct request_sock ***prevp,
486 const __be16 rport, const __be32 raddr,
487 const __be32 laddr)
488 {
489 const struct inet_connection_sock *icsk = inet_csk(sk);
490 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
491 struct request_sock *req, **prev;
492
493 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
494 lopt->nr_table_entries)];
495 (req = *prev) != NULL;
496 prev = &req->dl_next) {
497 const struct inet_request_sock *ireq = inet_rsk(req);
498
499 if (ireq->rmt_port == rport &&
500 ireq->rmt_addr == raddr &&
501 ireq->loc_addr == laddr &&
502 AF_INET_FAMILY(req->rsk_ops->family)) {
503 WARN_ON(req->sk);
504 *prevp = prev;
505 break;
506 }
507 }
508
509 return req;
510 }
511 EXPORT_SYMBOL_GPL(inet_csk_search_req);
512
513 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
514 unsigned long timeout)
515 {
516 struct inet_connection_sock *icsk = inet_csk(sk);
517 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
518 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
519 lopt->hash_rnd, lopt->nr_table_entries);
520
521 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
522 inet_csk_reqsk_queue_added(sk, timeout);
523 }
524 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
525
526 /* Only thing we need from tcp.h */
527 extern int sysctl_tcp_synack_retries;
528
529
530 /* Decide when to expire the request and when to resend SYN-ACK */
531 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
532 const int max_retries,
533 const u8 rskq_defer_accept,
534 int *expire, int *resend)
535 {
536 if (!rskq_defer_accept) {
537 *expire = req->num_timeout >= thresh;
538 *resend = 1;
539 return;
540 }
541 *expire = req->num_timeout >= thresh &&
542 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
543 /*
544 * Do not resend while waiting for data after ACK,
545 * start to resend on end of deferring period to give
546 * last chance for data or ACK to create established socket.
547 */
548 *resend = !inet_rsk(req)->acked ||
549 req->num_timeout >= rskq_defer_accept - 1;
550 }
551
552 int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req)
553 {
554 int err = req->rsk_ops->rtx_syn_ack(parent, req);
555
556 if (!err)
557 req->num_retrans++;
558 return err;
559 }
560 EXPORT_SYMBOL(inet_rtx_syn_ack);
561
562 void inet_csk_reqsk_queue_prune(struct sock *parent,
563 const unsigned long interval,
564 const unsigned long timeout,
565 const unsigned long max_rto)
566 {
567 struct inet_connection_sock *icsk = inet_csk(parent);
568 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
569 struct listen_sock *lopt = queue->listen_opt;
570 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
571 int thresh = max_retries;
572 unsigned long now = jiffies;
573 struct request_sock **reqp, *req;
574 int i, budget;
575
576 if (lopt == NULL || lopt->qlen == 0)
577 return;
578
579 /* Normally all the openreqs are young and become mature
580 * (i.e. converted to established socket) for first timeout.
581 * If synack was not acknowledged for 1 second, it means
582 * one of the following things: synack was lost, ack was lost,
583 * rtt is high or nobody planned to ack (i.e. synflood).
584 * When server is a bit loaded, queue is populated with old
585 * open requests, reducing effective size of queue.
586 * When server is well loaded, queue size reduces to zero
587 * after several minutes of work. It is not synflood,
588 * it is normal operation. The solution is pruning
589 * too old entries overriding normal timeout, when
590 * situation becomes dangerous.
591 *
592 * Essentially, we reserve half of room for young
593 * embrions; and abort old ones without pity, if old
594 * ones are about to clog our table.
595 */
596 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
597 int young = (lopt->qlen_young<<1);
598
599 while (thresh > 2) {
600 if (lopt->qlen < young)
601 break;
602 thresh--;
603 young <<= 1;
604 }
605 }
606
607 if (queue->rskq_defer_accept)
608 max_retries = queue->rskq_defer_accept;
609
610 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
611 i = lopt->clock_hand;
612
613 do {
614 reqp=&lopt->syn_table[i];
615 while ((req = *reqp) != NULL) {
616 if (time_after_eq(now, req->expires)) {
617 int expire = 0, resend = 0;
618
619 syn_ack_recalc(req, thresh, max_retries,
620 queue->rskq_defer_accept,
621 &expire, &resend);
622 req->rsk_ops->syn_ack_timeout(parent, req);
623 if (!expire &&
624 (!resend ||
625 !inet_rtx_syn_ack(parent, req) ||
626 inet_rsk(req)->acked)) {
627 unsigned long timeo;
628
629 if (req->num_timeout++ == 0)
630 lopt->qlen_young--;
631 timeo = min(timeout << req->num_timeout,
632 max_rto);
633 req->expires = now + timeo;
634 reqp = &req->dl_next;
635 continue;
636 }
637
638 /* Drop this request */
639 inet_csk_reqsk_queue_unlink(parent, req, reqp);
640 reqsk_queue_removed(queue, req);
641 reqsk_free(req);
642 continue;
643 }
644 reqp = &req->dl_next;
645 }
646
647 i = (i + 1) & (lopt->nr_table_entries - 1);
648
649 } while (--budget > 0);
650
651 lopt->clock_hand = i;
652
653 if (lopt->qlen)
654 inet_csk_reset_keepalive_timer(parent, interval);
655 }
656 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
657
658 /**
659 * inet_csk_clone_lock - clone an inet socket, and lock its clone
660 * @sk: the socket to clone
661 * @req: request_sock
662 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
663 *
664 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
665 */
666 struct sock *inet_csk_clone_lock(const struct sock *sk,
667 const struct request_sock *req,
668 const gfp_t priority)
669 {
670 struct sock *newsk = sk_clone_lock(sk, priority);
671
672 if (newsk != NULL) {
673 struct inet_connection_sock *newicsk = inet_csk(newsk);
674
675 newsk->sk_state = TCP_SYN_RECV;
676 newicsk->icsk_bind_hash = NULL;
677
678 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
679 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
680 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
681 newsk->sk_write_space = sk_stream_write_space;
682
683 newicsk->icsk_retransmits = 0;
684 newicsk->icsk_backoff = 0;
685 newicsk->icsk_probes_out = 0;
686
687 /* Deinitialize accept_queue to trap illegal accesses. */
688 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
689
690 security_inet_csk_clone(newsk, req);
691 }
692 return newsk;
693 }
694 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
695
696 /*
697 * At this point, there should be no process reference to this
698 * socket, and thus no user references at all. Therefore we
699 * can assume the socket waitqueue is inactive and nobody will
700 * try to jump onto it.
701 */
702 void inet_csk_destroy_sock(struct sock *sk)
703 {
704 WARN_ON(sk->sk_state != TCP_CLOSE);
705 WARN_ON(!sock_flag(sk, SOCK_DEAD));
706
707 /* It cannot be in hash table! */
708 WARN_ON(!sk_unhashed(sk));
709
710 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
711 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
712
713 sk->sk_prot->destroy(sk);
714
715 sk_stream_kill_queues(sk);
716
717 xfrm_sk_free_policy(sk);
718
719 sk_refcnt_debug_release(sk);
720
721 percpu_counter_dec(sk->sk_prot->orphan_count);
722 sock_put(sk);
723 }
724 EXPORT_SYMBOL(inet_csk_destroy_sock);
725
726 /* This function allows to force a closure of a socket after the call to
727 * tcp/dccp_create_openreq_child().
728 */
729 void inet_csk_prepare_forced_close(struct sock *sk)
730 __releases(&sk->sk_lock.slock)
731 {
732 /* sk_clone_lock locked the socket and set refcnt to 2 */
733 bh_unlock_sock(sk);
734 sock_put(sk);
735
736 /* The below has to be done to allow calling inet_csk_destroy_sock */
737 sock_set_flag(sk, SOCK_DEAD);
738 percpu_counter_inc(sk->sk_prot->orphan_count);
739 inet_sk(sk)->inet_num = 0;
740 }
741 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
742
743 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
744 {
745 struct inet_sock *inet = inet_sk(sk);
746 struct inet_connection_sock *icsk = inet_csk(sk);
747 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
748
749 if (rc != 0)
750 return rc;
751
752 sk->sk_max_ack_backlog = 0;
753 sk->sk_ack_backlog = 0;
754 inet_csk_delack_init(sk);
755
756 /* There is race window here: we announce ourselves listening,
757 * but this transition is still not validated by get_port().
758 * It is OK, because this socket enters to hash table only
759 * after validation is complete.
760 */
761 sk->sk_state = TCP_LISTEN;
762 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
763 inet->inet_sport = htons(inet->inet_num);
764
765 sk_dst_reset(sk);
766 sk->sk_prot->hash(sk);
767
768 return 0;
769 }
770
771 sk->sk_state = TCP_CLOSE;
772 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
773 return -EADDRINUSE;
774 }
775 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
776
777 /*
778 * This routine closes sockets which have been at least partially
779 * opened, but not yet accepted.
780 */
781 void inet_csk_listen_stop(struct sock *sk)
782 {
783 struct inet_connection_sock *icsk = inet_csk(sk);
784 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
785 struct request_sock *acc_req;
786 struct request_sock *req;
787
788 inet_csk_delete_keepalive_timer(sk);
789
790 /* make all the listen_opt local to us */
791 acc_req = reqsk_queue_yank_acceptq(queue);
792
793 /* Following specs, it would be better either to send FIN
794 * (and enter FIN-WAIT-1, it is normal close)
795 * or to send active reset (abort).
796 * Certainly, it is pretty dangerous while synflood, but it is
797 * bad justification for our negligence 8)
798 * To be honest, we are not able to make either
799 * of the variants now. --ANK
800 */
801 reqsk_queue_destroy(queue);
802
803 while ((req = acc_req) != NULL) {
804 struct sock *child = req->sk;
805
806 acc_req = req->dl_next;
807
808 local_bh_disable();
809 bh_lock_sock(child);
810 WARN_ON(sock_owned_by_user(child));
811 sock_hold(child);
812
813 sk->sk_prot->disconnect(child, O_NONBLOCK);
814
815 sock_orphan(child);
816
817 percpu_counter_inc(sk->sk_prot->orphan_count);
818
819 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->listener) {
820 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
821 BUG_ON(sk != tcp_rsk(req)->listener);
822
823 /* Paranoid, to prevent race condition if
824 * an inbound pkt destined for child is
825 * blocked by sock lock in tcp_v4_rcv().
826 * Also to satisfy an assertion in
827 * tcp_v4_destroy_sock().
828 */
829 tcp_sk(child)->fastopen_rsk = NULL;
830 sock_put(sk);
831 }
832 inet_csk_destroy_sock(child);
833
834 bh_unlock_sock(child);
835 local_bh_enable();
836 sock_put(child);
837
838 sk_acceptq_removed(sk);
839 __reqsk_free(req);
840 }
841 if (queue->fastopenq != NULL) {
842 /* Free all the reqs queued in rskq_rst_head. */
843 spin_lock_bh(&queue->fastopenq->lock);
844 acc_req = queue->fastopenq->rskq_rst_head;
845 queue->fastopenq->rskq_rst_head = NULL;
846 spin_unlock_bh(&queue->fastopenq->lock);
847 while ((req = acc_req) != NULL) {
848 acc_req = req->dl_next;
849 __reqsk_free(req);
850 }
851 }
852 WARN_ON(sk->sk_ack_backlog);
853 }
854 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
855
856 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
857 {
858 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
859 const struct inet_sock *inet = inet_sk(sk);
860
861 sin->sin_family = AF_INET;
862 sin->sin_addr.s_addr = inet->inet_daddr;
863 sin->sin_port = inet->inet_dport;
864 }
865 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
866
867 #ifdef CONFIG_COMPAT
868 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
869 char __user *optval, int __user *optlen)
870 {
871 const struct inet_connection_sock *icsk = inet_csk(sk);
872
873 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
874 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
875 optval, optlen);
876 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
877 optval, optlen);
878 }
879 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
880
881 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
882 char __user *optval, unsigned int optlen)
883 {
884 const struct inet_connection_sock *icsk = inet_csk(sk);
885
886 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
887 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
888 optval, optlen);
889 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
890 optval, optlen);
891 }
892 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
893 #endif
894
895 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
896 {
897 const struct inet_sock *inet = inet_sk(sk);
898 const struct ip_options_rcu *inet_opt;
899 __be32 daddr = inet->inet_daddr;
900 struct flowi4 *fl4;
901 struct rtable *rt;
902
903 rcu_read_lock();
904 inet_opt = rcu_dereference(inet->inet_opt);
905 if (inet_opt && inet_opt->opt.srr)
906 daddr = inet_opt->opt.faddr;
907 fl4 = &fl->u.ip4;
908 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
909 inet->inet_saddr, inet->inet_dport,
910 inet->inet_sport, sk->sk_protocol,
911 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
912 if (IS_ERR(rt))
913 rt = NULL;
914 if (rt)
915 sk_setup_caps(sk, &rt->dst);
916 rcu_read_unlock();
917
918 return &rt->dst;
919 }
920
921 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
922 {
923 struct dst_entry *dst = __sk_dst_check(sk, 0);
924 struct inet_sock *inet = inet_sk(sk);
925
926 if (!dst) {
927 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
928 if (!dst)
929 goto out;
930 }
931 dst->ops->update_pmtu(dst, sk, NULL, mtu);
932
933 dst = __sk_dst_check(sk, 0);
934 if (!dst)
935 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
936 out:
937 return dst;
938 }
939 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
Linux kernel - Deliverables
This page took 0.079425 seconds and 6 git commands to generate.