Merge branch 'x86-setup-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6] / 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 /*
33  * This struct holds the first and last local port number.
34  */
35 struct local_ports sysctl_local_ports __read_mostly = {
36         .lock = SEQLOCK_UNLOCKED,
37         .range = { 32768, 61000 },
38 };
39
40 void inet_get_local_port_range(int *low, int *high)
41 {
42         unsigned seq;
43         do {
44                 seq = read_seqbegin(&sysctl_local_ports.lock);
45
46                 *low = sysctl_local_ports.range[0];
47                 *high = sysctl_local_ports.range[1];
48         } while (read_seqretry(&sysctl_local_ports.lock, seq));
49 }
50 EXPORT_SYMBOL(inet_get_local_port_range);
51
52 int inet_csk_bind_conflict(const struct sock *sk,
53                            const struct inet_bind_bucket *tb)
54 {
55         const __be32 sk_rcv_saddr = inet_rcv_saddr(sk);
56         struct sock *sk2;
57         struct hlist_node *node;
58         int reuse = sk->sk_reuse;
59
60         /*
61          * Unlike other sk lookup places we do not check
62          * for sk_net here, since _all_ the socks listed
63          * in tb->owners list belong to the same net - the
64          * one this bucket belongs to.
65          */
66
67         sk_for_each_bound(sk2, node, &tb->owners) {
68                 if (sk != sk2 &&
69                     !inet_v6_ipv6only(sk2) &&
70                     (!sk->sk_bound_dev_if ||
71                      !sk2->sk_bound_dev_if ||
72                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
73                         if (!reuse || !sk2->sk_reuse ||
74                             sk2->sk_state == TCP_LISTEN) {
75                                 const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
76                                 if (!sk2_rcv_saddr || !sk_rcv_saddr ||
77                                     sk2_rcv_saddr == sk_rcv_saddr)
78                                         break;
79                         }
80                 }
81         }
82         return node != NULL;
83 }
84
85 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
86
87 /* Obtain a reference to a local port for the given sock,
88  * if snum is zero it means select any available local port.
89  */
90 int inet_csk_get_port(struct sock *sk, unsigned short snum)
91 {
92         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
93         struct inet_bind_hashbucket *head;
94         struct hlist_node *node;
95         struct inet_bind_bucket *tb;
96         int ret, attempts = 5;
97         struct net *net = sock_net(sk);
98         int smallest_size = -1, smallest_rover;
99
100         local_bh_disable();
101         if (!snum) {
102                 int remaining, rover, low, high;
103
104 again:
105                 inet_get_local_port_range(&low, &high);
106                 remaining = (high - low) + 1;
107                 smallest_rover = rover = net_random() % remaining + low;
108
109                 smallest_size = -1;
110                 do {
111                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
112                                         hashinfo->bhash_size)];
113                         spin_lock(&head->lock);
114                         inet_bind_bucket_for_each(tb, node, &head->chain)
115                                 if (ib_net(tb) == net && tb->port == rover) {
116                                         if (tb->fastreuse > 0 &&
117                                             sk->sk_reuse &&
118                                             sk->sk_state != TCP_LISTEN &&
119                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
120                                                 smallest_size = tb->num_owners;
121                                                 smallest_rover = rover;
122                                                 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) {
123                                                         spin_unlock(&head->lock);
124                                                         snum = smallest_rover;
125                                                         goto have_snum;
126                                                 }
127                                         }
128                                         goto next;
129                                 }
130                         break;
131                 next:
132                         spin_unlock(&head->lock);
133                         if (++rover > high)
134                                 rover = low;
135                 } while (--remaining > 0);
136
137                 /* Exhausted local port range during search?  It is not
138                  * possible for us to be holding one of the bind hash
139                  * locks if this test triggers, because if 'remaining'
140                  * drops to zero, we broke out of the do/while loop at
141                  * the top level, not from the 'break;' statement.
142                  */
143                 ret = 1;
144                 if (remaining <= 0) {
145                         if (smallest_size != -1) {
146                                 snum = smallest_rover;
147                                 goto have_snum;
148                         }
149                         goto fail;
150                 }
151                 /* OK, here is the one we will use.  HEAD is
152                  * non-NULL and we hold it's mutex.
153                  */
154                 snum = rover;
155         } else {
156 have_snum:
157                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
158                                 hashinfo->bhash_size)];
159                 spin_lock(&head->lock);
160                 inet_bind_bucket_for_each(tb, node, &head->chain)
161                         if (ib_net(tb) == net && tb->port == snum)
162                                 goto tb_found;
163         }
164         tb = NULL;
165         goto tb_not_found;
166 tb_found:
167         if (!hlist_empty(&tb->owners)) {
168                 if (tb->fastreuse > 0 &&
169                     sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
170                     smallest_size == -1) {
171                         goto success;
172                 } else {
173                         ret = 1;
174                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
175                                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
176                                     smallest_size != -1 && --attempts >= 0) {
177                                         spin_unlock(&head->lock);
178                                         goto again;
179                                 }
180                                 goto fail_unlock;
181                         }
182                 }
183         }
184 tb_not_found:
185         ret = 1;
186         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
187                                         net, head, snum)) == NULL)
188                 goto fail_unlock;
189         if (hlist_empty(&tb->owners)) {
190                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
191                         tb->fastreuse = 1;
192                 else
193                         tb->fastreuse = 0;
194         } else if (tb->fastreuse &&
195                    (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
196                 tb->fastreuse = 0;
197 success:
198         if (!inet_csk(sk)->icsk_bind_hash)
199                 inet_bind_hash(sk, tb, snum);
200         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
201         ret = 0;
202
203 fail_unlock:
204         spin_unlock(&head->lock);
205 fail:
206         local_bh_enable();
207         return ret;
208 }
209
210 EXPORT_SYMBOL_GPL(inet_csk_get_port);
211
212 /*
213  * Wait for an incoming connection, avoid race conditions. This must be called
214  * with the socket locked.
215  */
216 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
217 {
218         struct inet_connection_sock *icsk = inet_csk(sk);
219         DEFINE_WAIT(wait);
220         int err;
221
222         /*
223          * True wake-one mechanism for incoming connections: only
224          * one process gets woken up, not the 'whole herd'.
225          * Since we do not 'race & poll' for established sockets
226          * anymore, the common case will execute the loop only once.
227          *
228          * Subtle issue: "add_wait_queue_exclusive()" will be added
229          * after any current non-exclusive waiters, and we know that
230          * it will always _stay_ after any new non-exclusive waiters
231          * because all non-exclusive waiters are added at the
232          * beginning of the wait-queue. As such, it's ok to "drop"
233          * our exclusiveness temporarily when we get woken up without
234          * having to remove and re-insert us on the wait queue.
235          */
236         for (;;) {
237                 prepare_to_wait_exclusive(sk->sk_sleep, &wait,
238                                           TASK_INTERRUPTIBLE);
239                 release_sock(sk);
240                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
241                         timeo = schedule_timeout(timeo);
242                 lock_sock(sk);
243                 err = 0;
244                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
245                         break;
246                 err = -EINVAL;
247                 if (sk->sk_state != TCP_LISTEN)
248                         break;
249                 err = sock_intr_errno(timeo);
250                 if (signal_pending(current))
251                         break;
252                 err = -EAGAIN;
253                 if (!timeo)
254                         break;
255         }
256         finish_wait(sk->sk_sleep, &wait);
257         return err;
258 }
259
260 /*
261  * This will accept the next outstanding connection.
262  */
263 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
264 {
265         struct inet_connection_sock *icsk = inet_csk(sk);
266         struct sock *newsk;
267         int error;
268
269         lock_sock(sk);
270
271         /* We need to make sure that this socket is listening,
272          * and that it has something pending.
273          */
274         error = -EINVAL;
275         if (sk->sk_state != TCP_LISTEN)
276                 goto out_err;
277
278         /* Find already established connection */
279         if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
280                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
281
282                 /* If this is a non blocking socket don't sleep */
283                 error = -EAGAIN;
284                 if (!timeo)
285                         goto out_err;
286
287                 error = inet_csk_wait_for_connect(sk, timeo);
288                 if (error)
289                         goto out_err;
290         }
291
292         newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
293         WARN_ON(newsk->sk_state == TCP_SYN_RECV);
294 out:
295         release_sock(sk);
296         return newsk;
297 out_err:
298         newsk = NULL;
299         *err = error;
300         goto out;
301 }
302
303 EXPORT_SYMBOL(inet_csk_accept);
304
305 /*
306  * Using different timers for retransmit, delayed acks and probes
307  * We may wish use just one timer maintaining a list of expire jiffies
308  * to optimize.
309  */
310 void inet_csk_init_xmit_timers(struct sock *sk,
311                                void (*retransmit_handler)(unsigned long),
312                                void (*delack_handler)(unsigned long),
313                                void (*keepalive_handler)(unsigned long))
314 {
315         struct inet_connection_sock *icsk = inet_csk(sk);
316
317         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
318                         (unsigned long)sk);
319         setup_timer(&icsk->icsk_delack_timer, delack_handler,
320                         (unsigned long)sk);
321         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
322         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
323 }
324
325 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
326
327 void inet_csk_clear_xmit_timers(struct sock *sk)
328 {
329         struct inet_connection_sock *icsk = inet_csk(sk);
330
331         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
332
333         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
334         sk_stop_timer(sk, &icsk->icsk_delack_timer);
335         sk_stop_timer(sk, &sk->sk_timer);
336 }
337
338 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
339
340 void inet_csk_delete_keepalive_timer(struct sock *sk)
341 {
342         sk_stop_timer(sk, &sk->sk_timer);
343 }
344
345 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
346
347 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
348 {
349         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
350 }
351
352 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
353
354 struct dst_entry *inet_csk_route_req(struct sock *sk,
355                                      const struct request_sock *req)
356 {
357         struct rtable *rt;
358         const struct inet_request_sock *ireq = inet_rsk(req);
359         struct ip_options *opt = inet_rsk(req)->opt;
360         struct flowi fl = { .oif = sk->sk_bound_dev_if,
361                             .nl_u = { .ip4_u =
362                                       { .daddr = ((opt && opt->srr) ?
363                                                   opt->faddr :
364                                                   ireq->rmt_addr),
365                                         .saddr = ireq->loc_addr,
366                                         .tos = RT_CONN_FLAGS(sk) } },
367                             .proto = sk->sk_protocol,
368                             .flags = inet_sk_flowi_flags(sk),
369                             .uli_u = { .ports =
370                                        { .sport = inet_sk(sk)->sport,
371                                          .dport = ireq->rmt_port } } };
372         struct net *net = sock_net(sk);
373
374         security_req_classify_flow(req, &fl);
375         if (ip_route_output_flow(net, &rt, &fl, sk, 0))
376                 goto no_route;
377         if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
378                 goto route_err;
379         return &rt->u.dst;
380
381 route_err:
382         ip_rt_put(rt);
383 no_route:
384         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
385         return NULL;
386 }
387
388 EXPORT_SYMBOL_GPL(inet_csk_route_req);
389
390 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
391                                  const u32 rnd, const u32 synq_hsize)
392 {
393         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
394 }
395
396 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
397 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
398 #else
399 #define AF_INET_FAMILY(fam) 1
400 #endif
401
402 struct request_sock *inet_csk_search_req(const struct sock *sk,
403                                          struct request_sock ***prevp,
404                                          const __be16 rport, const __be32 raddr,
405                                          const __be32 laddr)
406 {
407         const struct inet_connection_sock *icsk = inet_csk(sk);
408         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
409         struct request_sock *req, **prev;
410
411         for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
412                                                     lopt->nr_table_entries)];
413              (req = *prev) != NULL;
414              prev = &req->dl_next) {
415                 const struct inet_request_sock *ireq = inet_rsk(req);
416
417                 if (ireq->rmt_port == rport &&
418                     ireq->rmt_addr == raddr &&
419                     ireq->loc_addr == laddr &&
420                     AF_INET_FAMILY(req->rsk_ops->family)) {
421                         WARN_ON(req->sk);
422                         *prevp = prev;
423                         break;
424                 }
425         }
426
427         return req;
428 }
429
430 EXPORT_SYMBOL_GPL(inet_csk_search_req);
431
432 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
433                                    unsigned long timeout)
434 {
435         struct inet_connection_sock *icsk = inet_csk(sk);
436         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
437         const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
438                                      lopt->hash_rnd, lopt->nr_table_entries);
439
440         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
441         inet_csk_reqsk_queue_added(sk, timeout);
442 }
443
444 /* Only thing we need from tcp.h */
445 extern int sysctl_tcp_synack_retries;
446
447 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
448
449 void inet_csk_reqsk_queue_prune(struct sock *parent,
450                                 const unsigned long interval,
451                                 const unsigned long timeout,
452                                 const unsigned long max_rto)
453 {
454         struct inet_connection_sock *icsk = inet_csk(parent);
455         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
456         struct listen_sock *lopt = queue->listen_opt;
457         int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
458         int thresh = max_retries;
459         unsigned long now = jiffies;
460         struct request_sock **reqp, *req;
461         int i, budget;
462
463         if (lopt == NULL || lopt->qlen == 0)
464                 return;
465
466         /* Normally all the openreqs are young and become mature
467          * (i.e. converted to established socket) for first timeout.
468          * If synack was not acknowledged for 3 seconds, it means
469          * one of the following things: synack was lost, ack was lost,
470          * rtt is high or nobody planned to ack (i.e. synflood).
471          * When server is a bit loaded, queue is populated with old
472          * open requests, reducing effective size of queue.
473          * When server is well loaded, queue size reduces to zero
474          * after several minutes of work. It is not synflood,
475          * it is normal operation. The solution is pruning
476          * too old entries overriding normal timeout, when
477          * situation becomes dangerous.
478          *
479          * Essentially, we reserve half of room for young
480          * embrions; and abort old ones without pity, if old
481          * ones are about to clog our table.
482          */
483         if (lopt->qlen>>(lopt->max_qlen_log-1)) {
484                 int young = (lopt->qlen_young<<1);
485
486                 while (thresh > 2) {
487                         if (lopt->qlen < young)
488                                 break;
489                         thresh--;
490                         young <<= 1;
491                 }
492         }
493
494         if (queue->rskq_defer_accept)
495                 max_retries = queue->rskq_defer_accept;
496
497         budget = 2 * (lopt->nr_table_entries / (timeout / interval));
498         i = lopt->clock_hand;
499
500         do {
501                 reqp=&lopt->syn_table[i];
502                 while ((req = *reqp) != NULL) {
503                         if (time_after_eq(now, req->expires)) {
504                                 if ((req->retrans < thresh ||
505                                      (inet_rsk(req)->acked && req->retrans < max_retries))
506                                     && !req->rsk_ops->rtx_syn_ack(parent, req)) {
507                                         unsigned long timeo;
508
509                                         if (req->retrans++ == 0)
510                                                 lopt->qlen_young--;
511                                         timeo = min((timeout << req->retrans), max_rto);
512                                         req->expires = now + timeo;
513                                         reqp = &req->dl_next;
514                                         continue;
515                                 }
516
517                                 /* Drop this request */
518                                 inet_csk_reqsk_queue_unlink(parent, req, reqp);
519                                 reqsk_queue_removed(queue, req);
520                                 reqsk_free(req);
521                                 continue;
522                         }
523                         reqp = &req->dl_next;
524                 }
525
526                 i = (i + 1) & (lopt->nr_table_entries - 1);
527
528         } while (--budget > 0);
529
530         lopt->clock_hand = i;
531
532         if (lopt->qlen)
533                 inet_csk_reset_keepalive_timer(parent, interval);
534 }
535
536 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
537
538 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
539                             const gfp_t priority)
540 {
541         struct sock *newsk = sk_clone(sk, priority);
542
543         if (newsk != NULL) {
544                 struct inet_connection_sock *newicsk = inet_csk(newsk);
545
546                 newsk->sk_state = TCP_SYN_RECV;
547                 newicsk->icsk_bind_hash = NULL;
548
549                 inet_sk(newsk)->dport = inet_rsk(req)->rmt_port;
550                 inet_sk(newsk)->num = ntohs(inet_rsk(req)->loc_port);
551                 inet_sk(newsk)->sport = inet_rsk(req)->loc_port;
552                 newsk->sk_write_space = sk_stream_write_space;
553
554                 newicsk->icsk_retransmits = 0;
555                 newicsk->icsk_backoff     = 0;
556                 newicsk->icsk_probes_out  = 0;
557
558                 /* Deinitialize accept_queue to trap illegal accesses. */
559                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
560
561                 security_inet_csk_clone(newsk, req);
562         }
563         return newsk;
564 }
565
566 EXPORT_SYMBOL_GPL(inet_csk_clone);
567
568 /*
569  * At this point, there should be no process reference to this
570  * socket, and thus no user references at all.  Therefore we
571  * can assume the socket waitqueue is inactive and nobody will
572  * try to jump onto it.
573  */
574 void inet_csk_destroy_sock(struct sock *sk)
575 {
576         WARN_ON(sk->sk_state != TCP_CLOSE);
577         WARN_ON(!sock_flag(sk, SOCK_DEAD));
578
579         /* It cannot be in hash table! */
580         WARN_ON(!sk_unhashed(sk));
581
582         /* If it has not 0 inet_sk(sk)->num, it must be bound */
583         WARN_ON(inet_sk(sk)->num && !inet_csk(sk)->icsk_bind_hash);
584
585         sk->sk_prot->destroy(sk);
586
587         sk_stream_kill_queues(sk);
588
589         xfrm_sk_free_policy(sk);
590
591         sk_refcnt_debug_release(sk);
592
593         percpu_counter_dec(sk->sk_prot->orphan_count);
594         sock_put(sk);
595 }
596
597 EXPORT_SYMBOL(inet_csk_destroy_sock);
598
599 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
600 {
601         struct inet_sock *inet = inet_sk(sk);
602         struct inet_connection_sock *icsk = inet_csk(sk);
603         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
604
605         if (rc != 0)
606                 return rc;
607
608         sk->sk_max_ack_backlog = 0;
609         sk->sk_ack_backlog = 0;
610         inet_csk_delack_init(sk);
611
612         /* There is race window here: we announce ourselves listening,
613          * but this transition is still not validated by get_port().
614          * It is OK, because this socket enters to hash table only
615          * after validation is complete.
616          */
617         sk->sk_state = TCP_LISTEN;
618         if (!sk->sk_prot->get_port(sk, inet->num)) {
619                 inet->sport = htons(inet->num);
620
621                 sk_dst_reset(sk);
622                 sk->sk_prot->hash(sk);
623
624                 return 0;
625         }
626
627         sk->sk_state = TCP_CLOSE;
628         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
629         return -EADDRINUSE;
630 }
631
632 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
633
634 /*
635  *      This routine closes sockets which have been at least partially
636  *      opened, but not yet accepted.
637  */
638 void inet_csk_listen_stop(struct sock *sk)
639 {
640         struct inet_connection_sock *icsk = inet_csk(sk);
641         struct request_sock *acc_req;
642         struct request_sock *req;
643
644         inet_csk_delete_keepalive_timer(sk);
645
646         /* make all the listen_opt local to us */
647         acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
648
649         /* Following specs, it would be better either to send FIN
650          * (and enter FIN-WAIT-1, it is normal close)
651          * or to send active reset (abort).
652          * Certainly, it is pretty dangerous while synflood, but it is
653          * bad justification for our negligence 8)
654          * To be honest, we are not able to make either
655          * of the variants now.                 --ANK
656          */
657         reqsk_queue_destroy(&icsk->icsk_accept_queue);
658
659         while ((req = acc_req) != NULL) {
660                 struct sock *child = req->sk;
661
662                 acc_req = req->dl_next;
663
664                 local_bh_disable();
665                 bh_lock_sock(child);
666                 WARN_ON(sock_owned_by_user(child));
667                 sock_hold(child);
668
669                 sk->sk_prot->disconnect(child, O_NONBLOCK);
670
671                 sock_orphan(child);
672
673                 percpu_counter_inc(sk->sk_prot->orphan_count);
674
675                 inet_csk_destroy_sock(child);
676
677                 bh_unlock_sock(child);
678                 local_bh_enable();
679                 sock_put(child);
680
681                 sk_acceptq_removed(sk);
682                 __reqsk_free(req);
683         }
684         WARN_ON(sk->sk_ack_backlog);
685 }
686
687 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
688
689 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
690 {
691         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
692         const struct inet_sock *inet = inet_sk(sk);
693
694         sin->sin_family         = AF_INET;
695         sin->sin_addr.s_addr    = inet->daddr;
696         sin->sin_port           = inet->dport;
697 }
698
699 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
700
701 #ifdef CONFIG_COMPAT
702 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
703                                char __user *optval, int __user *optlen)
704 {
705         const struct inet_connection_sock *icsk = inet_csk(sk);
706
707         if (icsk->icsk_af_ops->compat_getsockopt != NULL)
708                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
709                                                             optval, optlen);
710         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
711                                              optval, optlen);
712 }
713
714 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
715
716 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
717                                char __user *optval, int optlen)
718 {
719         const struct inet_connection_sock *icsk = inet_csk(sk);
720
721         if (icsk->icsk_af_ops->compat_setsockopt != NULL)
722                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
723                                                             optval, optlen);
724         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
725                                              optval, optlen);
726 }
727
728 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
729 #endif