445884cf1c29c3a3c30914b81a013dc5e48e0737
[linux-2.6.git] / net / dccp / proto.c
1 /*
2  *  net/dccp/proto.c
3  *
4  *  An implementation of the DCCP protocol
5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
6  *
7  *      This program is free software; you can redistribute it and/or modify it
8  *      under the terms of the GNU General Public License version 2 as
9  *      published by the Free Software Foundation.
10  */
11
12 #include <linux/dccp.h>
13 #include <linux/module.h>
14 #include <linux/types.h>
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/skbuff.h>
18 #include <linux/netdevice.h>
19 #include <linux/in.h>
20 #include <linux/if_arp.h>
21 #include <linux/init.h>
22 #include <linux/random.h>
23 #include <net/checksum.h>
24
25 #include <net/inet_sock.h>
26 #include <net/sock.h>
27 #include <net/xfrm.h>
28
29 #include <asm/ioctls.h>
30 #include <linux/spinlock.h>
31 #include <linux/timer.h>
32 #include <linux/delay.h>
33 #include <linux/poll.h>
34
35 #include "ccid.h"
36 #include "dccp.h"
37 #include "feat.h"
38
39 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
40
41 EXPORT_SYMBOL_GPL(dccp_statistics);
42
43 atomic_t dccp_orphan_count = ATOMIC_INIT(0);
44
45 EXPORT_SYMBOL_GPL(dccp_orphan_count);
46
47 struct inet_hashinfo dccp_hashinfo;
48 EXPORT_SYMBOL_GPL(dccp_hashinfo);
49
50 /* the maximum queue length for tx in packets. 0 is no limit */
51 int sysctl_dccp_tx_qlen __read_mostly = 5;
52
53 void dccp_set_state(struct sock *sk, const int state)
54 {
55         const int oldstate = sk->sk_state;
56
57         dccp_pr_debug("%s(%p)  %s  -->  %s\n", dccp_role(sk), sk,
58                       dccp_state_name(oldstate), dccp_state_name(state));
59         WARN_ON(state == oldstate);
60
61         switch (state) {
62         case DCCP_OPEN:
63                 if (oldstate != DCCP_OPEN)
64                         DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
65                 break;
66
67         case DCCP_CLOSED:
68                 if (oldstate == DCCP_OPEN || oldstate == DCCP_ACTIVE_CLOSEREQ ||
69                     oldstate == DCCP_CLOSING)
70                         DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
71
72                 sk->sk_prot->unhash(sk);
73                 if (inet_csk(sk)->icsk_bind_hash != NULL &&
74                     !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
75                         inet_put_port(sk);
76                 /* fall through */
77         default:
78                 if (oldstate == DCCP_OPEN)
79                         DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
80         }
81
82         /* Change state AFTER socket is unhashed to avoid closed
83          * socket sitting in hash tables.
84          */
85         sk->sk_state = state;
86 }
87
88 EXPORT_SYMBOL_GPL(dccp_set_state);
89
90 static void dccp_finish_passive_close(struct sock *sk)
91 {
92         switch (sk->sk_state) {
93         case DCCP_PASSIVE_CLOSE:
94                 /* Node (client or server) has received Close packet. */
95                 dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
96                 dccp_set_state(sk, DCCP_CLOSED);
97                 break;
98         case DCCP_PASSIVE_CLOSEREQ:
99                 /*
100                  * Client received CloseReq. We set the `active' flag so that
101                  * dccp_send_close() retransmits the Close as per RFC 4340, 8.3.
102                  */
103                 dccp_send_close(sk, 1);
104                 dccp_set_state(sk, DCCP_CLOSING);
105         }
106 }
107
108 void dccp_done(struct sock *sk)
109 {
110         dccp_set_state(sk, DCCP_CLOSED);
111         dccp_clear_xmit_timers(sk);
112
113         sk->sk_shutdown = SHUTDOWN_MASK;
114
115         if (!sock_flag(sk, SOCK_DEAD))
116                 sk->sk_state_change(sk);
117         else
118                 inet_csk_destroy_sock(sk);
119 }
120
121 EXPORT_SYMBOL_GPL(dccp_done);
122
123 const char *dccp_packet_name(const int type)
124 {
125         static const char *dccp_packet_names[] = {
126                 [DCCP_PKT_REQUEST]  = "REQUEST",
127                 [DCCP_PKT_RESPONSE] = "RESPONSE",
128                 [DCCP_PKT_DATA]     = "DATA",
129                 [DCCP_PKT_ACK]      = "ACK",
130                 [DCCP_PKT_DATAACK]  = "DATAACK",
131                 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
132                 [DCCP_PKT_CLOSE]    = "CLOSE",
133                 [DCCP_PKT_RESET]    = "RESET",
134                 [DCCP_PKT_SYNC]     = "SYNC",
135                 [DCCP_PKT_SYNCACK]  = "SYNCACK",
136         };
137
138         if (type >= DCCP_NR_PKT_TYPES)
139                 return "INVALID";
140         else
141                 return dccp_packet_names[type];
142 }
143
144 EXPORT_SYMBOL_GPL(dccp_packet_name);
145
146 const char *dccp_state_name(const int state)
147 {
148         static char *dccp_state_names[] = {
149         [DCCP_OPEN]             = "OPEN",
150         [DCCP_REQUESTING]       = "REQUESTING",
151         [DCCP_PARTOPEN]         = "PARTOPEN",
152         [DCCP_LISTEN]           = "LISTEN",
153         [DCCP_RESPOND]          = "RESPOND",
154         [DCCP_CLOSING]          = "CLOSING",
155         [DCCP_ACTIVE_CLOSEREQ]  = "CLOSEREQ",
156         [DCCP_PASSIVE_CLOSE]    = "PASSIVE_CLOSE",
157         [DCCP_PASSIVE_CLOSEREQ] = "PASSIVE_CLOSEREQ",
158         [DCCP_TIME_WAIT]        = "TIME_WAIT",
159         [DCCP_CLOSED]           = "CLOSED",
160         };
161
162         if (state >= DCCP_MAX_STATES)
163                 return "INVALID STATE!";
164         else
165                 return dccp_state_names[state];
166 }
167
168 EXPORT_SYMBOL_GPL(dccp_state_name);
169
170 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
171 {
172         struct dccp_sock *dp = dccp_sk(sk);
173         struct dccp_minisock *dmsk = dccp_msk(sk);
174         struct inet_connection_sock *icsk = inet_csk(sk);
175
176         dccp_minisock_init(&dp->dccps_minisock);
177
178         icsk->icsk_rto          = DCCP_TIMEOUT_INIT;
179         icsk->icsk_syn_retries  = sysctl_dccp_request_retries;
180         sk->sk_state            = DCCP_CLOSED;
181         sk->sk_write_space      = dccp_write_space;
182         icsk->icsk_sync_mss     = dccp_sync_mss;
183         dp->dccps_mss_cache     = 536;
184         dp->dccps_rate_last     = jiffies;
185         dp->dccps_role          = DCCP_ROLE_UNDEFINED;
186         dp->dccps_service       = DCCP_SERVICE_CODE_IS_ABSENT;
187         dp->dccps_l_ack_ratio   = dp->dccps_r_ack_ratio = 1;
188
189         dccp_init_xmit_timers(sk);
190
191         INIT_LIST_HEAD(&dp->dccps_featneg);
192         /*
193          * FIXME: We're hardcoding the CCID, and doing this at this point makes
194          * the listening (master) sock get CCID control blocks, which is not
195          * necessary, but for now, to not mess with the test userspace apps,
196          * lets leave it here, later the real solution is to do this in a
197          * setsockopt(CCIDs-I-want/accept). -acme
198          */
199         if (likely(ctl_sock_initialized)) {
200                 int rc = dccp_feat_init(sk);
201
202                 if (rc)
203                         return rc;
204
205                 if (dmsk->dccpms_send_ack_vector) {
206                         dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(GFP_KERNEL);
207                         if (dp->dccps_hc_rx_ackvec == NULL)
208                                 return -ENOMEM;
209                 }
210                 dp->dccps_hc_rx_ccid = ccid_hc_rx_new(dmsk->dccpms_rx_ccid,
211                                                       sk, GFP_KERNEL);
212                 dp->dccps_hc_tx_ccid = ccid_hc_tx_new(dmsk->dccpms_tx_ccid,
213                                                       sk, GFP_KERNEL);
214                 if (unlikely(dp->dccps_hc_rx_ccid == NULL ||
215                              dp->dccps_hc_tx_ccid == NULL)) {
216                         ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
217                         ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
218                         if (dmsk->dccpms_send_ack_vector) {
219                                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
220                                 dp->dccps_hc_rx_ackvec = NULL;
221                         }
222                         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
223                         return -ENOMEM;
224                 }
225         } else {
226                 /* control socket doesn't need feat nego */
227                 INIT_LIST_HEAD(&dmsk->dccpms_pending);
228                 INIT_LIST_HEAD(&dmsk->dccpms_conf);
229         }
230
231         return 0;
232 }
233
234 EXPORT_SYMBOL_GPL(dccp_init_sock);
235
236 void dccp_destroy_sock(struct sock *sk)
237 {
238         struct dccp_sock *dp = dccp_sk(sk);
239         struct dccp_minisock *dmsk = dccp_msk(sk);
240
241         /*
242          * DCCP doesn't use sk_write_queue, just sk_send_head
243          * for retransmissions
244          */
245         if (sk->sk_send_head != NULL) {
246                 kfree_skb(sk->sk_send_head);
247                 sk->sk_send_head = NULL;
248         }
249
250         /* Clean up a referenced DCCP bind bucket. */
251         if (inet_csk(sk)->icsk_bind_hash != NULL)
252                 inet_put_port(sk);
253
254         kfree(dp->dccps_service_list);
255         dp->dccps_service_list = NULL;
256
257         if (dmsk->dccpms_send_ack_vector) {
258                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
259                 dp->dccps_hc_rx_ackvec = NULL;
260         }
261         ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
262         ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
263         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
264
265         /* clean up feature negotiation state */
266         dccp_feat_list_purge(&dp->dccps_featneg);
267 }
268
269 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
270
271 static inline int dccp_listen_start(struct sock *sk, int backlog)
272 {
273         struct dccp_sock *dp = dccp_sk(sk);
274
275         dp->dccps_role = DCCP_ROLE_LISTEN;
276         /* do not start to listen if feature negotiation setup fails */
277         if (dccp_feat_finalise_settings(dp))
278                 return -EPROTO;
279         return inet_csk_listen_start(sk, backlog);
280 }
281
282 static inline int dccp_need_reset(int state)
283 {
284         return state != DCCP_CLOSED && state != DCCP_LISTEN &&
285                state != DCCP_REQUESTING;
286 }
287
288 int dccp_disconnect(struct sock *sk, int flags)
289 {
290         struct inet_connection_sock *icsk = inet_csk(sk);
291         struct inet_sock *inet = inet_sk(sk);
292         int err = 0;
293         const int old_state = sk->sk_state;
294
295         if (old_state != DCCP_CLOSED)
296                 dccp_set_state(sk, DCCP_CLOSED);
297
298         /*
299          * This corresponds to the ABORT function of RFC793, sec. 3.8
300          * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
301          */
302         if (old_state == DCCP_LISTEN) {
303                 inet_csk_listen_stop(sk);
304         } else if (dccp_need_reset(old_state)) {
305                 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
306                 sk->sk_err = ECONNRESET;
307         } else if (old_state == DCCP_REQUESTING)
308                 sk->sk_err = ECONNRESET;
309
310         dccp_clear_xmit_timers(sk);
311
312         __skb_queue_purge(&sk->sk_receive_queue);
313         __skb_queue_purge(&sk->sk_write_queue);
314         if (sk->sk_send_head != NULL) {
315                 __kfree_skb(sk->sk_send_head);
316                 sk->sk_send_head = NULL;
317         }
318
319         inet->dport = 0;
320
321         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
322                 inet_reset_saddr(sk);
323
324         sk->sk_shutdown = 0;
325         sock_reset_flag(sk, SOCK_DONE);
326
327         icsk->icsk_backoff = 0;
328         inet_csk_delack_init(sk);
329         __sk_dst_reset(sk);
330
331         WARN_ON(inet->num && !icsk->icsk_bind_hash);
332
333         sk->sk_error_report(sk);
334         return err;
335 }
336
337 EXPORT_SYMBOL_GPL(dccp_disconnect);
338
339 /*
340  *      Wait for a DCCP event.
341  *
342  *      Note that we don't need to lock the socket, as the upper poll layers
343  *      take care of normal races (between the test and the event) and we don't
344  *      go look at any of the socket buffers directly.
345  */
346 unsigned int dccp_poll(struct file *file, struct socket *sock,
347                        poll_table *wait)
348 {
349         unsigned int mask;
350         struct sock *sk = sock->sk;
351
352         poll_wait(file, sk->sk_sleep, wait);
353         if (sk->sk_state == DCCP_LISTEN)
354                 return inet_csk_listen_poll(sk);
355
356         /* Socket is not locked. We are protected from async events
357            by poll logic and correct handling of state changes
358            made by another threads is impossible in any case.
359          */
360
361         mask = 0;
362         if (sk->sk_err)
363                 mask = POLLERR;
364
365         if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
366                 mask |= POLLHUP;
367         if (sk->sk_shutdown & RCV_SHUTDOWN)
368                 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
369
370         /* Connected? */
371         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
372                 if (atomic_read(&sk->sk_rmem_alloc) > 0)
373                         mask |= POLLIN | POLLRDNORM;
374
375                 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
376                         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
377                                 mask |= POLLOUT | POLLWRNORM;
378                         } else {  /* send SIGIO later */
379                                 set_bit(SOCK_ASYNC_NOSPACE,
380                                         &sk->sk_socket->flags);
381                                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
382
383                                 /* Race breaker. If space is freed after
384                                  * wspace test but before the flags are set,
385                                  * IO signal will be lost.
386                                  */
387                                 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
388                                         mask |= POLLOUT | POLLWRNORM;
389                         }
390                 }
391         }
392         return mask;
393 }
394
395 EXPORT_SYMBOL_GPL(dccp_poll);
396
397 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
398 {
399         int rc = -ENOTCONN;
400
401         lock_sock(sk);
402
403         if (sk->sk_state == DCCP_LISTEN)
404                 goto out;
405
406         switch (cmd) {
407         case SIOCINQ: {
408                 struct sk_buff *skb;
409                 unsigned long amount = 0;
410
411                 skb = skb_peek(&sk->sk_receive_queue);
412                 if (skb != NULL) {
413                         /*
414                          * We will only return the amount of this packet since
415                          * that is all that will be read.
416                          */
417                         amount = skb->len;
418                 }
419                 rc = put_user(amount, (int __user *)arg);
420         }
421                 break;
422         default:
423                 rc = -ENOIOCTLCMD;
424                 break;
425         }
426 out:
427         release_sock(sk);
428         return rc;
429 }
430
431 EXPORT_SYMBOL_GPL(dccp_ioctl);
432
433 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
434                                    char __user *optval, int optlen)
435 {
436         struct dccp_sock *dp = dccp_sk(sk);
437         struct dccp_service_list *sl = NULL;
438
439         if (service == DCCP_SERVICE_INVALID_VALUE ||
440             optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
441                 return -EINVAL;
442
443         if (optlen > sizeof(service)) {
444                 sl = kmalloc(optlen, GFP_KERNEL);
445                 if (sl == NULL)
446                         return -ENOMEM;
447
448                 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
449                 if (copy_from_user(sl->dccpsl_list,
450                                    optval + sizeof(service),
451                                    optlen - sizeof(service)) ||
452                     dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
453                         kfree(sl);
454                         return -EFAULT;
455                 }
456         }
457
458         lock_sock(sk);
459         dp->dccps_service = service;
460
461         kfree(dp->dccps_service_list);
462
463         dp->dccps_service_list = sl;
464         release_sock(sk);
465         return 0;
466 }
467
468 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
469 {
470         u8 *list, len;
471         int i, rc;
472
473         if (cscov < 0 || cscov > 15)
474                 return -EINVAL;
475         /*
476          * Populate a list of permissible values, in the range cscov...15. This
477          * is necessary since feature negotiation of single values only works if
478          * both sides incidentally choose the same value. Since the list starts
479          * lowest-value first, negotiation will pick the smallest shared value.
480          */
481         if (cscov == 0)
482                 return 0;
483         len = 16 - cscov;
484
485         list = kmalloc(len, GFP_KERNEL);
486         if (list == NULL)
487                 return -ENOBUFS;
488
489         for (i = 0; i < len; i++)
490                 list[i] = cscov++;
491
492         rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
493
494         if (rc == 0) {
495                 if (rx)
496                         dccp_sk(sk)->dccps_pcrlen = cscov;
497                 else
498                         dccp_sk(sk)->dccps_pcslen = cscov;
499         }
500         kfree(list);
501         return rc;
502 }
503
504 static int dccp_setsockopt_ccid(struct sock *sk, int type,
505                                 char __user *optval, int optlen)
506 {
507         u8 *val;
508         int rc = 0;
509
510         if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
511                 return -EINVAL;
512
513         val = kmalloc(optlen, GFP_KERNEL);
514         if (val == NULL)
515                 return -ENOMEM;
516
517         if (copy_from_user(val, optval, optlen)) {
518                 kfree(val);
519                 return -EFAULT;
520         }
521
522         lock_sock(sk);
523         if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
524                 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
525
526         if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
527                 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
528         release_sock(sk);
529
530         kfree(val);
531         return rc;
532 }
533
534 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
535                 char __user *optval, int optlen)
536 {
537         struct dccp_sock *dp = dccp_sk(sk);
538         int val, err = 0;
539
540         switch (optname) {
541         case DCCP_SOCKOPT_PACKET_SIZE:
542                 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
543                 return 0;
544         case DCCP_SOCKOPT_CHANGE_L:
545         case DCCP_SOCKOPT_CHANGE_R:
546                 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
547                 return 0;
548         case DCCP_SOCKOPT_CCID:
549         case DCCP_SOCKOPT_RX_CCID:
550         case DCCP_SOCKOPT_TX_CCID:
551                 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
552         }
553
554         if (optlen < (int)sizeof(int))
555                 return -EINVAL;
556
557         if (get_user(val, (int __user *)optval))
558                 return -EFAULT;
559
560         if (optname == DCCP_SOCKOPT_SERVICE)
561                 return dccp_setsockopt_service(sk, val, optval, optlen);
562
563         lock_sock(sk);
564         switch (optname) {
565         case DCCP_SOCKOPT_SERVER_TIMEWAIT:
566                 if (dp->dccps_role != DCCP_ROLE_SERVER)
567                         err = -EOPNOTSUPP;
568                 else
569                         dp->dccps_server_timewait = (val != 0);
570                 break;
571         case DCCP_SOCKOPT_SEND_CSCOV:
572                 err = dccp_setsockopt_cscov(sk, val, false);
573                 break;
574         case DCCP_SOCKOPT_RECV_CSCOV:
575                 err = dccp_setsockopt_cscov(sk, val, true);
576                 break;
577         default:
578                 err = -ENOPROTOOPT;
579                 break;
580         }
581         release_sock(sk);
582
583         return err;
584 }
585
586 int dccp_setsockopt(struct sock *sk, int level, int optname,
587                     char __user *optval, int optlen)
588 {
589         if (level != SOL_DCCP)
590                 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
591                                                              optname, optval,
592                                                              optlen);
593         return do_dccp_setsockopt(sk, level, optname, optval, optlen);
594 }
595
596 EXPORT_SYMBOL_GPL(dccp_setsockopt);
597
598 #ifdef CONFIG_COMPAT
599 int compat_dccp_setsockopt(struct sock *sk, int level, int optname,
600                            char __user *optval, int optlen)
601 {
602         if (level != SOL_DCCP)
603                 return inet_csk_compat_setsockopt(sk, level, optname,
604                                                   optval, optlen);
605         return do_dccp_setsockopt(sk, level, optname, optval, optlen);
606 }
607
608 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt);
609 #endif
610
611 static int dccp_getsockopt_service(struct sock *sk, int len,
612                                    __be32 __user *optval,
613                                    int __user *optlen)
614 {
615         const struct dccp_sock *dp = dccp_sk(sk);
616         const struct dccp_service_list *sl;
617         int err = -ENOENT, slen = 0, total_len = sizeof(u32);
618
619         lock_sock(sk);
620         if ((sl = dp->dccps_service_list) != NULL) {
621                 slen = sl->dccpsl_nr * sizeof(u32);
622                 total_len += slen;
623         }
624
625         err = -EINVAL;
626         if (total_len > len)
627                 goto out;
628
629         err = 0;
630         if (put_user(total_len, optlen) ||
631             put_user(dp->dccps_service, optval) ||
632             (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
633                 err = -EFAULT;
634 out:
635         release_sock(sk);
636         return err;
637 }
638
639 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
640                     char __user *optval, int __user *optlen)
641 {
642         struct dccp_sock *dp;
643         int val, len;
644
645         if (get_user(len, optlen))
646                 return -EFAULT;
647
648         if (len < (int)sizeof(int))
649                 return -EINVAL;
650
651         dp = dccp_sk(sk);
652
653         switch (optname) {
654         case DCCP_SOCKOPT_PACKET_SIZE:
655                 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
656                 return 0;
657         case DCCP_SOCKOPT_SERVICE:
658                 return dccp_getsockopt_service(sk, len,
659                                                (__be32 __user *)optval, optlen);
660         case DCCP_SOCKOPT_GET_CUR_MPS:
661                 val = dp->dccps_mss_cache;
662                 break;
663         case DCCP_SOCKOPT_AVAILABLE_CCIDS:
664                 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
665         case DCCP_SOCKOPT_SERVER_TIMEWAIT:
666                 val = dp->dccps_server_timewait;
667                 break;
668         case DCCP_SOCKOPT_SEND_CSCOV:
669                 val = dp->dccps_pcslen;
670                 break;
671         case DCCP_SOCKOPT_RECV_CSCOV:
672                 val = dp->dccps_pcrlen;
673                 break;
674         case 128 ... 191:
675                 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
676                                              len, (u32 __user *)optval, optlen);
677         case 192 ... 255:
678                 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
679                                              len, (u32 __user *)optval, optlen);
680         default:
681                 return -ENOPROTOOPT;
682         }
683
684         len = sizeof(val);
685         if (put_user(len, optlen) || copy_to_user(optval, &val, len))
686                 return -EFAULT;
687
688         return 0;
689 }
690
691 int dccp_getsockopt(struct sock *sk, int level, int optname,
692                     char __user *optval, int __user *optlen)
693 {
694         if (level != SOL_DCCP)
695                 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
696                                                              optname, optval,
697                                                              optlen);
698         return do_dccp_getsockopt(sk, level, optname, optval, optlen);
699 }
700
701 EXPORT_SYMBOL_GPL(dccp_getsockopt);
702
703 #ifdef CONFIG_COMPAT
704 int compat_dccp_getsockopt(struct sock *sk, int level, int optname,
705                            char __user *optval, int __user *optlen)
706 {
707         if (level != SOL_DCCP)
708                 return inet_csk_compat_getsockopt(sk, level, optname,
709                                                   optval, optlen);
710         return do_dccp_getsockopt(sk, level, optname, optval, optlen);
711 }
712
713 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt);
714 #endif
715
716 int dccp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
717                  size_t len)
718 {
719         const struct dccp_sock *dp = dccp_sk(sk);
720         const int flags = msg->msg_flags;
721         const int noblock = flags & MSG_DONTWAIT;
722         struct sk_buff *skb;
723         int rc, size;
724         long timeo;
725
726         if (len > dp->dccps_mss_cache)
727                 return -EMSGSIZE;
728
729         lock_sock(sk);
730
731         if (sysctl_dccp_tx_qlen &&
732             (sk->sk_write_queue.qlen >= sysctl_dccp_tx_qlen)) {
733                 rc = -EAGAIN;
734                 goto out_release;
735         }
736
737         timeo = sock_sndtimeo(sk, noblock);
738
739         /*
740          * We have to use sk_stream_wait_connect here to set sk_write_pending,
741          * so that the trick in dccp_rcv_request_sent_state_process.
742          */
743         /* Wait for a connection to finish. */
744         if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
745                 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
746                         goto out_release;
747
748         size = sk->sk_prot->max_header + len;
749         release_sock(sk);
750         skb = sock_alloc_send_skb(sk, size, noblock, &rc);
751         lock_sock(sk);
752         if (skb == NULL)
753                 goto out_release;
754
755         skb_reserve(skb, sk->sk_prot->max_header);
756         rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
757         if (rc != 0)
758                 goto out_discard;
759
760         skb_queue_tail(&sk->sk_write_queue, skb);
761         dccp_write_xmit(sk,0);
762 out_release:
763         release_sock(sk);
764         return rc ? : len;
765 out_discard:
766         kfree_skb(skb);
767         goto out_release;
768 }
769
770 EXPORT_SYMBOL_GPL(dccp_sendmsg);
771
772 int dccp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
773                  size_t len, int nonblock, int flags, int *addr_len)
774 {
775         const struct dccp_hdr *dh;
776         long timeo;
777
778         lock_sock(sk);
779
780         if (sk->sk_state == DCCP_LISTEN) {
781                 len = -ENOTCONN;
782                 goto out;
783         }
784
785         timeo = sock_rcvtimeo(sk, nonblock);
786
787         do {
788                 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
789
790                 if (skb == NULL)
791                         goto verify_sock_status;
792
793                 dh = dccp_hdr(skb);
794
795                 switch (dh->dccph_type) {
796                 case DCCP_PKT_DATA:
797                 case DCCP_PKT_DATAACK:
798                         goto found_ok_skb;
799
800                 case DCCP_PKT_CLOSE:
801                 case DCCP_PKT_CLOSEREQ:
802                         if (!(flags & MSG_PEEK))
803                                 dccp_finish_passive_close(sk);
804                         /* fall through */
805                 case DCCP_PKT_RESET:
806                         dccp_pr_debug("found fin (%s) ok!\n",
807                                       dccp_packet_name(dh->dccph_type));
808                         len = 0;
809                         goto found_fin_ok;
810                 default:
811                         dccp_pr_debug("packet_type=%s\n",
812                                       dccp_packet_name(dh->dccph_type));
813                         sk_eat_skb(sk, skb, 0);
814                 }
815 verify_sock_status:
816                 if (sock_flag(sk, SOCK_DONE)) {
817                         len = 0;
818                         break;
819                 }
820
821                 if (sk->sk_err) {
822                         len = sock_error(sk);
823                         break;
824                 }
825
826                 if (sk->sk_shutdown & RCV_SHUTDOWN) {
827                         len = 0;
828                         break;
829                 }
830
831                 if (sk->sk_state == DCCP_CLOSED) {
832                         if (!sock_flag(sk, SOCK_DONE)) {
833                                 /* This occurs when user tries to read
834                                  * from never connected socket.
835                                  */
836                                 len = -ENOTCONN;
837                                 break;
838                         }
839                         len = 0;
840                         break;
841                 }
842
843                 if (!timeo) {
844                         len = -EAGAIN;
845                         break;
846                 }
847
848                 if (signal_pending(current)) {
849                         len = sock_intr_errno(timeo);
850                         break;
851                 }
852
853                 sk_wait_data(sk, &timeo);
854                 continue;
855         found_ok_skb:
856                 if (len > skb->len)
857                         len = skb->len;
858                 else if (len < skb->len)
859                         msg->msg_flags |= MSG_TRUNC;
860
861                 if (skb_copy_datagram_iovec(skb, 0, msg->msg_iov, len)) {
862                         /* Exception. Bailout! */
863                         len = -EFAULT;
864                         break;
865                 }
866         found_fin_ok:
867                 if (!(flags & MSG_PEEK))
868                         sk_eat_skb(sk, skb, 0);
869                 break;
870         } while (1);
871 out:
872         release_sock(sk);
873         return len;
874 }
875
876 EXPORT_SYMBOL_GPL(dccp_recvmsg);
877
878 int inet_dccp_listen(struct socket *sock, int backlog)
879 {
880         struct sock *sk = sock->sk;
881         unsigned char old_state;
882         int err;
883
884         lock_sock(sk);
885
886         err = -EINVAL;
887         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
888                 goto out;
889
890         old_state = sk->sk_state;
891         if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
892                 goto out;
893
894         /* Really, if the socket is already in listen state
895          * we can only allow the backlog to be adjusted.
896          */
897         if (old_state != DCCP_LISTEN) {
898                 /*
899                  * FIXME: here it probably should be sk->sk_prot->listen_start
900                  * see tcp_listen_start
901                  */
902                 err = dccp_listen_start(sk, backlog);
903                 if (err)
904                         goto out;
905         }
906         sk->sk_max_ack_backlog = backlog;
907         err = 0;
908
909 out:
910         release_sock(sk);
911         return err;
912 }
913
914 EXPORT_SYMBOL_GPL(inet_dccp_listen);
915
916 static void dccp_terminate_connection(struct sock *sk)
917 {
918         u8 next_state = DCCP_CLOSED;
919
920         switch (sk->sk_state) {
921         case DCCP_PASSIVE_CLOSE:
922         case DCCP_PASSIVE_CLOSEREQ:
923                 dccp_finish_passive_close(sk);
924                 break;
925         case DCCP_PARTOPEN:
926                 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
927                 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
928                 /* fall through */
929         case DCCP_OPEN:
930                 dccp_send_close(sk, 1);
931
932                 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
933                     !dccp_sk(sk)->dccps_server_timewait)
934                         next_state = DCCP_ACTIVE_CLOSEREQ;
935                 else
936                         next_state = DCCP_CLOSING;
937                 /* fall through */
938         default:
939                 dccp_set_state(sk, next_state);
940         }
941 }
942
943 void dccp_close(struct sock *sk, long timeout)
944 {
945         struct dccp_sock *dp = dccp_sk(sk);
946         struct sk_buff *skb;
947         u32 data_was_unread = 0;
948         int state;
949
950         lock_sock(sk);
951
952         sk->sk_shutdown = SHUTDOWN_MASK;
953
954         if (sk->sk_state == DCCP_LISTEN) {
955                 dccp_set_state(sk, DCCP_CLOSED);
956
957                 /* Special case. */
958                 inet_csk_listen_stop(sk);
959
960                 goto adjudge_to_death;
961         }
962
963         sk_stop_timer(sk, &dp->dccps_xmit_timer);
964
965         /*
966          * We need to flush the recv. buffs.  We do this only on the
967          * descriptor close, not protocol-sourced closes, because the
968           *reader process may not have drained the data yet!
969          */
970         while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
971                 data_was_unread += skb->len;
972                 __kfree_skb(skb);
973         }
974
975         if (data_was_unread) {
976                 /* Unread data was tossed, send an appropriate Reset Code */
977                 DCCP_WARN("DCCP: ABORT -- %u bytes unread\n", data_was_unread);
978                 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
979                 dccp_set_state(sk, DCCP_CLOSED);
980         } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
981                 /* Check zero linger _after_ checking for unread data. */
982                 sk->sk_prot->disconnect(sk, 0);
983         } else if (sk->sk_state != DCCP_CLOSED) {
984                 dccp_terminate_connection(sk);
985         }
986
987         sk_stream_wait_close(sk, timeout);
988
989 adjudge_to_death:
990         state = sk->sk_state;
991         sock_hold(sk);
992         sock_orphan(sk);
993         atomic_inc(sk->sk_prot->orphan_count);
994
995         /*
996          * It is the last release_sock in its life. It will remove backlog.
997          */
998         release_sock(sk);
999         /*
1000          * Now socket is owned by kernel and we acquire BH lock
1001          * to finish close. No need to check for user refs.
1002          */
1003         local_bh_disable();
1004         bh_lock_sock(sk);
1005         WARN_ON(sock_owned_by_user(sk));
1006
1007         /* Have we already been destroyed by a softirq or backlog? */
1008         if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1009                 goto out;
1010
1011         if (sk->sk_state == DCCP_CLOSED)
1012                 inet_csk_destroy_sock(sk);
1013
1014         /* Otherwise, socket is reprieved until protocol close. */
1015
1016 out:
1017         bh_unlock_sock(sk);
1018         local_bh_enable();
1019         sock_put(sk);
1020 }
1021
1022 EXPORT_SYMBOL_GPL(dccp_close);
1023
1024 void dccp_shutdown(struct sock *sk, int how)
1025 {
1026         dccp_pr_debug("called shutdown(%x)\n", how);
1027 }
1028
1029 EXPORT_SYMBOL_GPL(dccp_shutdown);
1030
1031 static inline int dccp_mib_init(void)
1032 {
1033         return snmp_mib_init((void**)dccp_statistics, sizeof(struct dccp_mib));
1034 }
1035
1036 static inline void dccp_mib_exit(void)
1037 {
1038         snmp_mib_free((void**)dccp_statistics);
1039 }
1040
1041 static int thash_entries;
1042 module_param(thash_entries, int, 0444);
1043 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1044
1045 #ifdef CONFIG_IP_DCCP_DEBUG
1046 int dccp_debug;
1047 module_param(dccp_debug, bool, 0644);
1048 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1049
1050 EXPORT_SYMBOL_GPL(dccp_debug);
1051 #endif
1052
1053 static int __init dccp_init(void)
1054 {
1055         unsigned long goal;
1056         int ehash_order, bhash_order, i;
1057         int rc = -ENOBUFS;
1058
1059         BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1060                      FIELD_SIZEOF(struct sk_buff, cb));
1061
1062         inet_hashinfo_init(&dccp_hashinfo);
1063         dccp_hashinfo.bind_bucket_cachep =
1064                 kmem_cache_create("dccp_bind_bucket",
1065                                   sizeof(struct inet_bind_bucket), 0,
1066                                   SLAB_HWCACHE_ALIGN, NULL);
1067         if (!dccp_hashinfo.bind_bucket_cachep)
1068                 goto out;
1069
1070         /*
1071          * Size and allocate the main established and bind bucket
1072          * hash tables.
1073          *
1074          * The methodology is similar to that of the buffer cache.
1075          */
1076         if (num_physpages >= (128 * 1024))
1077                 goal = num_physpages >> (21 - PAGE_SHIFT);
1078         else
1079                 goal = num_physpages >> (23 - PAGE_SHIFT);
1080
1081         if (thash_entries)
1082                 goal = (thash_entries *
1083                         sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1084         for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1085                 ;
1086         do {
1087                 dccp_hashinfo.ehash_size = (1UL << ehash_order) * PAGE_SIZE /
1088                                         sizeof(struct inet_ehash_bucket);
1089                 while (dccp_hashinfo.ehash_size &
1090                        (dccp_hashinfo.ehash_size - 1))
1091                         dccp_hashinfo.ehash_size--;
1092                 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1093                         __get_free_pages(GFP_ATOMIC, ehash_order);
1094         } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1095
1096         if (!dccp_hashinfo.ehash) {
1097                 DCCP_CRIT("Failed to allocate DCCP established hash table");
1098                 goto out_free_bind_bucket_cachep;
1099         }
1100
1101         for (i = 0; i < dccp_hashinfo.ehash_size; i++) {
1102                 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1103                 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].twchain, i);
1104         }
1105
1106         if (inet_ehash_locks_alloc(&dccp_hashinfo))
1107                         goto out_free_dccp_ehash;
1108
1109         bhash_order = ehash_order;
1110
1111         do {
1112                 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1113                                         sizeof(struct inet_bind_hashbucket);
1114                 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1115                     bhash_order > 0)
1116                         continue;
1117                 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1118                         __get_free_pages(GFP_ATOMIC, bhash_order);
1119         } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1120
1121         if (!dccp_hashinfo.bhash) {
1122                 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1123                 goto out_free_dccp_locks;
1124         }
1125
1126         for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1127                 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1128                 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1129         }
1130
1131         rc = dccp_mib_init();
1132         if (rc)
1133                 goto out_free_dccp_bhash;
1134
1135         rc = dccp_ackvec_init();
1136         if (rc)
1137                 goto out_free_dccp_mib;
1138
1139         rc = dccp_sysctl_init();
1140         if (rc)
1141                 goto out_ackvec_exit;
1142
1143         dccp_timestamping_init();
1144 out:
1145         return rc;
1146 out_ackvec_exit:
1147         dccp_ackvec_exit();
1148 out_free_dccp_mib:
1149         dccp_mib_exit();
1150 out_free_dccp_bhash:
1151         free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1152         dccp_hashinfo.bhash = NULL;
1153 out_free_dccp_locks:
1154         inet_ehash_locks_free(&dccp_hashinfo);
1155 out_free_dccp_ehash:
1156         free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1157         dccp_hashinfo.ehash = NULL;
1158 out_free_bind_bucket_cachep:
1159         kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1160         dccp_hashinfo.bind_bucket_cachep = NULL;
1161         goto out;
1162 }
1163
1164 static void __exit dccp_fini(void)
1165 {
1166         dccp_mib_exit();
1167         free_pages((unsigned long)dccp_hashinfo.bhash,
1168                    get_order(dccp_hashinfo.bhash_size *
1169                              sizeof(struct inet_bind_hashbucket)));
1170         free_pages((unsigned long)dccp_hashinfo.ehash,
1171                    get_order(dccp_hashinfo.ehash_size *
1172                              sizeof(struct inet_ehash_bucket)));
1173         inet_ehash_locks_free(&dccp_hashinfo);
1174         kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1175         dccp_ackvec_exit();
1176         dccp_sysctl_exit();
1177 }
1178
1179 module_init(dccp_init);
1180 module_exit(dccp_fini);
1181
1182 MODULE_LICENSE("GPL");
1183 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1184 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");