923db06c7e55e2851a84000799e5d3ec23dafe92
[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 struct percpu_counter dccp_orphan_count;
44 EXPORT_SYMBOL_GPL(dccp_orphan_count);
45
46 struct inet_hashinfo dccp_hashinfo;
47 EXPORT_SYMBOL_GPL(dccp_hashinfo);
48
49 /* the maximum queue length for tx in packets. 0 is no limit */
50 int sysctl_dccp_tx_qlen __read_mostly = 5;
51
52 void dccp_set_state(struct sock *sk, const int state)
53 {
54         const int oldstate = sk->sk_state;
55
56         dccp_pr_debug("%s(%p)  %s  -->  %s\n", dccp_role(sk), sk,
57                       dccp_state_name(oldstate), dccp_state_name(state));
58         WARN_ON(state == oldstate);
59
60         switch (state) {
61         case DCCP_OPEN:
62                 if (oldstate != DCCP_OPEN)
63                         DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
64                 /* Client retransmits all Confirm options until entering OPEN */
65                 if (oldstate == DCCP_PARTOPEN)
66                         dccp_feat_list_purge(&dccp_sk(sk)->dccps_featneg);
67                 break;
68
69         case DCCP_CLOSED:
70                 if (oldstate == DCCP_OPEN || oldstate == DCCP_ACTIVE_CLOSEREQ ||
71                     oldstate == DCCP_CLOSING)
72                         DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
73
74                 sk->sk_prot->unhash(sk);
75                 if (inet_csk(sk)->icsk_bind_hash != NULL &&
76                     !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
77                         inet_put_port(sk);
78                 /* fall through */
79         default:
80                 if (oldstate == DCCP_OPEN)
81                         DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
82         }
83
84         /* Change state AFTER socket is unhashed to avoid closed
85          * socket sitting in hash tables.
86          */
87         sk->sk_state = state;
88 }
89
90 EXPORT_SYMBOL_GPL(dccp_set_state);
91
92 static void dccp_finish_passive_close(struct sock *sk)
93 {
94         switch (sk->sk_state) {
95         case DCCP_PASSIVE_CLOSE:
96                 /* Node (client or server) has received Close packet. */
97                 dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
98                 dccp_set_state(sk, DCCP_CLOSED);
99                 break;
100         case DCCP_PASSIVE_CLOSEREQ:
101                 /*
102                  * Client received CloseReq. We set the `active' flag so that
103                  * dccp_send_close() retransmits the Close as per RFC 4340, 8.3.
104                  */
105                 dccp_send_close(sk, 1);
106                 dccp_set_state(sk, DCCP_CLOSING);
107         }
108 }
109
110 void dccp_done(struct sock *sk)
111 {
112         dccp_set_state(sk, DCCP_CLOSED);
113         dccp_clear_xmit_timers(sk);
114
115         sk->sk_shutdown = SHUTDOWN_MASK;
116
117         if (!sock_flag(sk, SOCK_DEAD))
118                 sk->sk_state_change(sk);
119         else
120                 inet_csk_destroy_sock(sk);
121 }
122
123 EXPORT_SYMBOL_GPL(dccp_done);
124
125 const char *dccp_packet_name(const int type)
126 {
127         static const char *const dccp_packet_names[] = {
128                 [DCCP_PKT_REQUEST]  = "REQUEST",
129                 [DCCP_PKT_RESPONSE] = "RESPONSE",
130                 [DCCP_PKT_DATA]     = "DATA",
131                 [DCCP_PKT_ACK]      = "ACK",
132                 [DCCP_PKT_DATAACK]  = "DATAACK",
133                 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
134                 [DCCP_PKT_CLOSE]    = "CLOSE",
135                 [DCCP_PKT_RESET]    = "RESET",
136                 [DCCP_PKT_SYNC]     = "SYNC",
137                 [DCCP_PKT_SYNCACK]  = "SYNCACK",
138         };
139
140         if (type >= DCCP_NR_PKT_TYPES)
141                 return "INVALID";
142         else
143                 return dccp_packet_names[type];
144 }
145
146 EXPORT_SYMBOL_GPL(dccp_packet_name);
147
148 const char *dccp_state_name(const int state)
149 {
150         static const char *const dccp_state_names[] = {
151         [DCCP_OPEN]             = "OPEN",
152         [DCCP_REQUESTING]       = "REQUESTING",
153         [DCCP_PARTOPEN]         = "PARTOPEN",
154         [DCCP_LISTEN]           = "LISTEN",
155         [DCCP_RESPOND]          = "RESPOND",
156         [DCCP_CLOSING]          = "CLOSING",
157         [DCCP_ACTIVE_CLOSEREQ]  = "CLOSEREQ",
158         [DCCP_PASSIVE_CLOSE]    = "PASSIVE_CLOSE",
159         [DCCP_PASSIVE_CLOSEREQ] = "PASSIVE_CLOSEREQ",
160         [DCCP_TIME_WAIT]        = "TIME_WAIT",
161         [DCCP_CLOSED]           = "CLOSED",
162         };
163
164         if (state >= DCCP_MAX_STATES)
165                 return "INVALID STATE!";
166         else
167                 return dccp_state_names[state];
168 }
169
170 EXPORT_SYMBOL_GPL(dccp_state_name);
171
172 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
173 {
174         struct dccp_sock *dp = dccp_sk(sk);
175         struct inet_connection_sock *icsk = inet_csk(sk);
176
177         icsk->icsk_rto          = DCCP_TIMEOUT_INIT;
178         icsk->icsk_syn_retries  = sysctl_dccp_request_retries;
179         sk->sk_state            = DCCP_CLOSED;
180         sk->sk_write_space      = dccp_write_space;
181         icsk->icsk_sync_mss     = dccp_sync_mss;
182         dp->dccps_mss_cache     = 536;
183         dp->dccps_rate_last     = jiffies;
184         dp->dccps_role          = DCCP_ROLE_UNDEFINED;
185         dp->dccps_service       = DCCP_SERVICE_CODE_IS_ABSENT;
186         dp->dccps_l_ack_ratio   = dp->dccps_r_ack_ratio = 1;
187
188         dccp_init_xmit_timers(sk);
189
190         INIT_LIST_HEAD(&dp->dccps_featneg);
191         /* control socket doesn't need feat nego */
192         if (likely(ctl_sock_initialized))
193                 return dccp_feat_init(sk);
194         return 0;
195 }
196
197 EXPORT_SYMBOL_GPL(dccp_init_sock);
198
199 void dccp_destroy_sock(struct sock *sk)
200 {
201         struct dccp_sock *dp = dccp_sk(sk);
202
203         /*
204          * DCCP doesn't use sk_write_queue, just sk_send_head
205          * for retransmissions
206          */
207         if (sk->sk_send_head != NULL) {
208                 kfree_skb(sk->sk_send_head);
209                 sk->sk_send_head = NULL;
210         }
211
212         /* Clean up a referenced DCCP bind bucket. */
213         if (inet_csk(sk)->icsk_bind_hash != NULL)
214                 inet_put_port(sk);
215
216         kfree(dp->dccps_service_list);
217         dp->dccps_service_list = NULL;
218
219         if (dp->dccps_hc_rx_ackvec != NULL) {
220                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
221                 dp->dccps_hc_rx_ackvec = NULL;
222         }
223         ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
224         ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
225         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
226
227         /* clean up feature negotiation state */
228         dccp_feat_list_purge(&dp->dccps_featneg);
229 }
230
231 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
232
233 static inline int dccp_listen_start(struct sock *sk, int backlog)
234 {
235         struct dccp_sock *dp = dccp_sk(sk);
236
237         dp->dccps_role = DCCP_ROLE_LISTEN;
238         /* do not start to listen if feature negotiation setup fails */
239         if (dccp_feat_finalise_settings(dp))
240                 return -EPROTO;
241         return inet_csk_listen_start(sk, backlog);
242 }
243
244 static inline int dccp_need_reset(int state)
245 {
246         return state != DCCP_CLOSED && state != DCCP_LISTEN &&
247                state != DCCP_REQUESTING;
248 }
249
250 int dccp_disconnect(struct sock *sk, int flags)
251 {
252         struct inet_connection_sock *icsk = inet_csk(sk);
253         struct inet_sock *inet = inet_sk(sk);
254         int err = 0;
255         const int old_state = sk->sk_state;
256
257         if (old_state != DCCP_CLOSED)
258                 dccp_set_state(sk, DCCP_CLOSED);
259
260         /*
261          * This corresponds to the ABORT function of RFC793, sec. 3.8
262          * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
263          */
264         if (old_state == DCCP_LISTEN) {
265                 inet_csk_listen_stop(sk);
266         } else if (dccp_need_reset(old_state)) {
267                 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
268                 sk->sk_err = ECONNRESET;
269         } else if (old_state == DCCP_REQUESTING)
270                 sk->sk_err = ECONNRESET;
271
272         dccp_clear_xmit_timers(sk);
273
274         __skb_queue_purge(&sk->sk_receive_queue);
275         __skb_queue_purge(&sk->sk_write_queue);
276         if (sk->sk_send_head != NULL) {
277                 __kfree_skb(sk->sk_send_head);
278                 sk->sk_send_head = NULL;
279         }
280
281         inet->dport = 0;
282
283         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
284                 inet_reset_saddr(sk);
285
286         sk->sk_shutdown = 0;
287         sock_reset_flag(sk, SOCK_DONE);
288
289         icsk->icsk_backoff = 0;
290         inet_csk_delack_init(sk);
291         __sk_dst_reset(sk);
292
293         WARN_ON(inet->num && !icsk->icsk_bind_hash);
294
295         sk->sk_error_report(sk);
296         return err;
297 }
298
299 EXPORT_SYMBOL_GPL(dccp_disconnect);
300
301 /*
302  *      Wait for a DCCP event.
303  *
304  *      Note that we don't need to lock the socket, as the upper poll layers
305  *      take care of normal races (between the test and the event) and we don't
306  *      go look at any of the socket buffers directly.
307  */
308 unsigned int dccp_poll(struct file *file, struct socket *sock,
309                        poll_table *wait)
310 {
311         unsigned int mask;
312         struct sock *sk = sock->sk;
313
314         sock_poll_wait(file, sk->sk_sleep, wait);
315         if (sk->sk_state == DCCP_LISTEN)
316                 return inet_csk_listen_poll(sk);
317
318         /* Socket is not locked. We are protected from async events
319            by poll logic and correct handling of state changes
320            made by another threads is impossible in any case.
321          */
322
323         mask = 0;
324         if (sk->sk_err)
325                 mask = POLLERR;
326
327         if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
328                 mask |= POLLHUP;
329         if (sk->sk_shutdown & RCV_SHUTDOWN)
330                 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
331
332         /* Connected? */
333         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
334                 if (atomic_read(&sk->sk_rmem_alloc) > 0)
335                         mask |= POLLIN | POLLRDNORM;
336
337                 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
338                         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
339                                 mask |= POLLOUT | POLLWRNORM;
340                         } else {  /* send SIGIO later */
341                                 set_bit(SOCK_ASYNC_NOSPACE,
342                                         &sk->sk_socket->flags);
343                                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
344
345                                 /* Race breaker. If space is freed after
346                                  * wspace test but before the flags are set,
347                                  * IO signal will be lost.
348                                  */
349                                 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
350                                         mask |= POLLOUT | POLLWRNORM;
351                         }
352                 }
353         }
354         return mask;
355 }
356
357 EXPORT_SYMBOL_GPL(dccp_poll);
358
359 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
360 {
361         int rc = -ENOTCONN;
362
363         lock_sock(sk);
364
365         if (sk->sk_state == DCCP_LISTEN)
366                 goto out;
367
368         switch (cmd) {
369         case SIOCINQ: {
370                 struct sk_buff *skb;
371                 unsigned long amount = 0;
372
373                 skb = skb_peek(&sk->sk_receive_queue);
374                 if (skb != NULL) {
375                         /*
376                          * We will only return the amount of this packet since
377                          * that is all that will be read.
378                          */
379                         amount = skb->len;
380                 }
381                 rc = put_user(amount, (int __user *)arg);
382         }
383                 break;
384         default:
385                 rc = -ENOIOCTLCMD;
386                 break;
387         }
388 out:
389         release_sock(sk);
390         return rc;
391 }
392
393 EXPORT_SYMBOL_GPL(dccp_ioctl);
394
395 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
396                                    char __user *optval, int optlen)
397 {
398         struct dccp_sock *dp = dccp_sk(sk);
399         struct dccp_service_list *sl = NULL;
400
401         if (service == DCCP_SERVICE_INVALID_VALUE ||
402             optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
403                 return -EINVAL;
404
405         if (optlen > sizeof(service)) {
406                 sl = kmalloc(optlen, GFP_KERNEL);
407                 if (sl == NULL)
408                         return -ENOMEM;
409
410                 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
411                 if (copy_from_user(sl->dccpsl_list,
412                                    optval + sizeof(service),
413                                    optlen - sizeof(service)) ||
414                     dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
415                         kfree(sl);
416                         return -EFAULT;
417                 }
418         }
419
420         lock_sock(sk);
421         dp->dccps_service = service;
422
423         kfree(dp->dccps_service_list);
424
425         dp->dccps_service_list = sl;
426         release_sock(sk);
427         return 0;
428 }
429
430 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
431 {
432         u8 *list, len;
433         int i, rc;
434
435         if (cscov < 0 || cscov > 15)
436                 return -EINVAL;
437         /*
438          * Populate a list of permissible values, in the range cscov...15. This
439          * is necessary since feature negotiation of single values only works if
440          * both sides incidentally choose the same value. Since the list starts
441          * lowest-value first, negotiation will pick the smallest shared value.
442          */
443         if (cscov == 0)
444                 return 0;
445         len = 16 - cscov;
446
447         list = kmalloc(len, GFP_KERNEL);
448         if (list == NULL)
449                 return -ENOBUFS;
450
451         for (i = 0; i < len; i++)
452                 list[i] = cscov++;
453
454         rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
455
456         if (rc == 0) {
457                 if (rx)
458                         dccp_sk(sk)->dccps_pcrlen = cscov;
459                 else
460                         dccp_sk(sk)->dccps_pcslen = cscov;
461         }
462         kfree(list);
463         return rc;
464 }
465
466 static int dccp_setsockopt_ccid(struct sock *sk, int type,
467                                 char __user *optval, int optlen)
468 {
469         u8 *val;
470         int rc = 0;
471
472         if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
473                 return -EINVAL;
474
475         val = kmalloc(optlen, GFP_KERNEL);
476         if (val == NULL)
477                 return -ENOMEM;
478
479         if (copy_from_user(val, optval, optlen)) {
480                 kfree(val);
481                 return -EFAULT;
482         }
483
484         lock_sock(sk);
485         if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
486                 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
487
488         if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
489                 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
490         release_sock(sk);
491
492         kfree(val);
493         return rc;
494 }
495
496 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
497                 char __user *optval, int optlen)
498 {
499         struct dccp_sock *dp = dccp_sk(sk);
500         int val, err = 0;
501
502         switch (optname) {
503         case DCCP_SOCKOPT_PACKET_SIZE:
504                 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
505                 return 0;
506         case DCCP_SOCKOPT_CHANGE_L:
507         case DCCP_SOCKOPT_CHANGE_R:
508                 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
509                 return 0;
510         case DCCP_SOCKOPT_CCID:
511         case DCCP_SOCKOPT_RX_CCID:
512         case DCCP_SOCKOPT_TX_CCID:
513                 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
514         }
515
516         if (optlen < (int)sizeof(int))
517                 return -EINVAL;
518
519         if (get_user(val, (int __user *)optval))
520                 return -EFAULT;
521
522         if (optname == DCCP_SOCKOPT_SERVICE)
523                 return dccp_setsockopt_service(sk, val, optval, optlen);
524
525         lock_sock(sk);
526         switch (optname) {
527         case DCCP_SOCKOPT_SERVER_TIMEWAIT:
528                 if (dp->dccps_role != DCCP_ROLE_SERVER)
529                         err = -EOPNOTSUPP;
530                 else
531                         dp->dccps_server_timewait = (val != 0);
532                 break;
533         case DCCP_SOCKOPT_SEND_CSCOV:
534                 err = dccp_setsockopt_cscov(sk, val, false);
535                 break;
536         case DCCP_SOCKOPT_RECV_CSCOV:
537                 err = dccp_setsockopt_cscov(sk, val, true);
538                 break;
539         default:
540                 err = -ENOPROTOOPT;
541                 break;
542         }
543         release_sock(sk);
544
545         return err;
546 }
547
548 int dccp_setsockopt(struct sock *sk, int level, int optname,
549                     char __user *optval, int optlen)
550 {
551         if (level != SOL_DCCP)
552                 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
553                                                              optname, optval,
554                                                              optlen);
555         return do_dccp_setsockopt(sk, level, optname, optval, optlen);
556 }
557
558 EXPORT_SYMBOL_GPL(dccp_setsockopt);
559
560 #ifdef CONFIG_COMPAT
561 int compat_dccp_setsockopt(struct sock *sk, int level, int optname,
562                            char __user *optval, int optlen)
563 {
564         if (level != SOL_DCCP)
565                 return inet_csk_compat_setsockopt(sk, level, optname,
566                                                   optval, optlen);
567         return do_dccp_setsockopt(sk, level, optname, optval, optlen);
568 }
569
570 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt);
571 #endif
572
573 static int dccp_getsockopt_service(struct sock *sk, int len,
574                                    __be32 __user *optval,
575                                    int __user *optlen)
576 {
577         const struct dccp_sock *dp = dccp_sk(sk);
578         const struct dccp_service_list *sl;
579         int err = -ENOENT, slen = 0, total_len = sizeof(u32);
580
581         lock_sock(sk);
582         if ((sl = dp->dccps_service_list) != NULL) {
583                 slen = sl->dccpsl_nr * sizeof(u32);
584                 total_len += slen;
585         }
586
587         err = -EINVAL;
588         if (total_len > len)
589                 goto out;
590
591         err = 0;
592         if (put_user(total_len, optlen) ||
593             put_user(dp->dccps_service, optval) ||
594             (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
595                 err = -EFAULT;
596 out:
597         release_sock(sk);
598         return err;
599 }
600
601 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
602                     char __user *optval, int __user *optlen)
603 {
604         struct dccp_sock *dp;
605         int val, len;
606
607         if (get_user(len, optlen))
608                 return -EFAULT;
609
610         if (len < (int)sizeof(int))
611                 return -EINVAL;
612
613         dp = dccp_sk(sk);
614
615         switch (optname) {
616         case DCCP_SOCKOPT_PACKET_SIZE:
617                 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
618                 return 0;
619         case DCCP_SOCKOPT_SERVICE:
620                 return dccp_getsockopt_service(sk, len,
621                                                (__be32 __user *)optval, optlen);
622         case DCCP_SOCKOPT_GET_CUR_MPS:
623                 val = dp->dccps_mss_cache;
624                 break;
625         case DCCP_SOCKOPT_AVAILABLE_CCIDS:
626                 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
627         case DCCP_SOCKOPT_TX_CCID:
628                 val = ccid_get_current_tx_ccid(dp);
629                 if (val < 0)
630                         return -ENOPROTOOPT;
631                 break;
632         case DCCP_SOCKOPT_RX_CCID:
633                 val = ccid_get_current_rx_ccid(dp);
634                 if (val < 0)
635                         return -ENOPROTOOPT;
636                 break;
637         case DCCP_SOCKOPT_SERVER_TIMEWAIT:
638                 val = dp->dccps_server_timewait;
639                 break;
640         case DCCP_SOCKOPT_SEND_CSCOV:
641                 val = dp->dccps_pcslen;
642                 break;
643         case DCCP_SOCKOPT_RECV_CSCOV:
644                 val = dp->dccps_pcrlen;
645                 break;
646         case 128 ... 191:
647                 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
648                                              len, (u32 __user *)optval, optlen);
649         case 192 ... 255:
650                 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
651                                              len, (u32 __user *)optval, optlen);
652         default:
653                 return -ENOPROTOOPT;
654         }
655
656         len = sizeof(val);
657         if (put_user(len, optlen) || copy_to_user(optval, &val, len))
658                 return -EFAULT;
659
660         return 0;
661 }
662
663 int dccp_getsockopt(struct sock *sk, int level, int optname,
664                     char __user *optval, int __user *optlen)
665 {
666         if (level != SOL_DCCP)
667                 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
668                                                              optname, optval,
669                                                              optlen);
670         return do_dccp_getsockopt(sk, level, optname, optval, optlen);
671 }
672
673 EXPORT_SYMBOL_GPL(dccp_getsockopt);
674
675 #ifdef CONFIG_COMPAT
676 int compat_dccp_getsockopt(struct sock *sk, int level, int optname,
677                            char __user *optval, int __user *optlen)
678 {
679         if (level != SOL_DCCP)
680                 return inet_csk_compat_getsockopt(sk, level, optname,
681                                                   optval, optlen);
682         return do_dccp_getsockopt(sk, level, optname, optval, optlen);
683 }
684
685 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt);
686 #endif
687
688 int dccp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
689                  size_t len)
690 {
691         const struct dccp_sock *dp = dccp_sk(sk);
692         const int flags = msg->msg_flags;
693         const int noblock = flags & MSG_DONTWAIT;
694         struct sk_buff *skb;
695         int rc, size;
696         long timeo;
697
698         if (len > dp->dccps_mss_cache)
699                 return -EMSGSIZE;
700
701         lock_sock(sk);
702
703         if (sysctl_dccp_tx_qlen &&
704             (sk->sk_write_queue.qlen >= sysctl_dccp_tx_qlen)) {
705                 rc = -EAGAIN;
706                 goto out_release;
707         }
708
709         timeo = sock_sndtimeo(sk, noblock);
710
711         /*
712          * We have to use sk_stream_wait_connect here to set sk_write_pending,
713          * so that the trick in dccp_rcv_request_sent_state_process.
714          */
715         /* Wait for a connection to finish. */
716         if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
717                 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
718                         goto out_release;
719
720         size = sk->sk_prot->max_header + len;
721         release_sock(sk);
722         skb = sock_alloc_send_skb(sk, size, noblock, &rc);
723         lock_sock(sk);
724         if (skb == NULL)
725                 goto out_release;
726
727         skb_reserve(skb, sk->sk_prot->max_header);
728         rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
729         if (rc != 0)
730                 goto out_discard;
731
732         skb_queue_tail(&sk->sk_write_queue, skb);
733         dccp_write_xmit(sk,0);
734 out_release:
735         release_sock(sk);
736         return rc ? : len;
737 out_discard:
738         kfree_skb(skb);
739         goto out_release;
740 }
741
742 EXPORT_SYMBOL_GPL(dccp_sendmsg);
743
744 int dccp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
745                  size_t len, int nonblock, int flags, int *addr_len)
746 {
747         const struct dccp_hdr *dh;
748         long timeo;
749
750         lock_sock(sk);
751
752         if (sk->sk_state == DCCP_LISTEN) {
753                 len = -ENOTCONN;
754                 goto out;
755         }
756
757         timeo = sock_rcvtimeo(sk, nonblock);
758
759         do {
760                 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
761
762                 if (skb == NULL)
763                         goto verify_sock_status;
764
765                 dh = dccp_hdr(skb);
766
767                 switch (dh->dccph_type) {
768                 case DCCP_PKT_DATA:
769                 case DCCP_PKT_DATAACK:
770                         goto found_ok_skb;
771
772                 case DCCP_PKT_CLOSE:
773                 case DCCP_PKT_CLOSEREQ:
774                         if (!(flags & MSG_PEEK))
775                                 dccp_finish_passive_close(sk);
776                         /* fall through */
777                 case DCCP_PKT_RESET:
778                         dccp_pr_debug("found fin (%s) ok!\n",
779                                       dccp_packet_name(dh->dccph_type));
780                         len = 0;
781                         goto found_fin_ok;
782                 default:
783                         dccp_pr_debug("packet_type=%s\n",
784                                       dccp_packet_name(dh->dccph_type));
785                         sk_eat_skb(sk, skb, 0);
786                 }
787 verify_sock_status:
788                 if (sock_flag(sk, SOCK_DONE)) {
789                         len = 0;
790                         break;
791                 }
792
793                 if (sk->sk_err) {
794                         len = sock_error(sk);
795                         break;
796                 }
797
798                 if (sk->sk_shutdown & RCV_SHUTDOWN) {
799                         len = 0;
800                         break;
801                 }
802
803                 if (sk->sk_state == DCCP_CLOSED) {
804                         if (!sock_flag(sk, SOCK_DONE)) {
805                                 /* This occurs when user tries to read
806                                  * from never connected socket.
807                                  */
808                                 len = -ENOTCONN;
809                                 break;
810                         }
811                         len = 0;
812                         break;
813                 }
814
815                 if (!timeo) {
816                         len = -EAGAIN;
817                         break;
818                 }
819
820                 if (signal_pending(current)) {
821                         len = sock_intr_errno(timeo);
822                         break;
823                 }
824
825                 sk_wait_data(sk, &timeo);
826                 continue;
827         found_ok_skb:
828                 if (len > skb->len)
829                         len = skb->len;
830                 else if (len < skb->len)
831                         msg->msg_flags |= MSG_TRUNC;
832
833                 if (skb_copy_datagram_iovec(skb, 0, msg->msg_iov, len)) {
834                         /* Exception. Bailout! */
835                         len = -EFAULT;
836                         break;
837                 }
838         found_fin_ok:
839                 if (!(flags & MSG_PEEK))
840                         sk_eat_skb(sk, skb, 0);
841                 break;
842         } while (1);
843 out:
844         release_sock(sk);
845         return len;
846 }
847
848 EXPORT_SYMBOL_GPL(dccp_recvmsg);
849
850 int inet_dccp_listen(struct socket *sock, int backlog)
851 {
852         struct sock *sk = sock->sk;
853         unsigned char old_state;
854         int err;
855
856         lock_sock(sk);
857
858         err = -EINVAL;
859         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
860                 goto out;
861
862         old_state = sk->sk_state;
863         if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
864                 goto out;
865
866         /* Really, if the socket is already in listen state
867          * we can only allow the backlog to be adjusted.
868          */
869         if (old_state != DCCP_LISTEN) {
870                 /*
871                  * FIXME: here it probably should be sk->sk_prot->listen_start
872                  * see tcp_listen_start
873                  */
874                 err = dccp_listen_start(sk, backlog);
875                 if (err)
876                         goto out;
877         }
878         sk->sk_max_ack_backlog = backlog;
879         err = 0;
880
881 out:
882         release_sock(sk);
883         return err;
884 }
885
886 EXPORT_SYMBOL_GPL(inet_dccp_listen);
887
888 static void dccp_terminate_connection(struct sock *sk)
889 {
890         u8 next_state = DCCP_CLOSED;
891
892         switch (sk->sk_state) {
893         case DCCP_PASSIVE_CLOSE:
894         case DCCP_PASSIVE_CLOSEREQ:
895                 dccp_finish_passive_close(sk);
896                 break;
897         case DCCP_PARTOPEN:
898                 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
899                 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
900                 /* fall through */
901         case DCCP_OPEN:
902                 dccp_send_close(sk, 1);
903
904                 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
905                     !dccp_sk(sk)->dccps_server_timewait)
906                         next_state = DCCP_ACTIVE_CLOSEREQ;
907                 else
908                         next_state = DCCP_CLOSING;
909                 /* fall through */
910         default:
911                 dccp_set_state(sk, next_state);
912         }
913 }
914
915 void dccp_close(struct sock *sk, long timeout)
916 {
917         struct dccp_sock *dp = dccp_sk(sk);
918         struct sk_buff *skb;
919         u32 data_was_unread = 0;
920         int state;
921
922         lock_sock(sk);
923
924         sk->sk_shutdown = SHUTDOWN_MASK;
925
926         if (sk->sk_state == DCCP_LISTEN) {
927                 dccp_set_state(sk, DCCP_CLOSED);
928
929                 /* Special case. */
930                 inet_csk_listen_stop(sk);
931
932                 goto adjudge_to_death;
933         }
934
935         sk_stop_timer(sk, &dp->dccps_xmit_timer);
936
937         /*
938          * We need to flush the recv. buffs.  We do this only on the
939          * descriptor close, not protocol-sourced closes, because the
940           *reader process may not have drained the data yet!
941          */
942         while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
943                 data_was_unread += skb->len;
944                 __kfree_skb(skb);
945         }
946
947         if (data_was_unread) {
948                 /* Unread data was tossed, send an appropriate Reset Code */
949                 DCCP_WARN("DCCP: ABORT -- %u bytes unread\n", data_was_unread);
950                 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
951                 dccp_set_state(sk, DCCP_CLOSED);
952         } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
953                 /* Check zero linger _after_ checking for unread data. */
954                 sk->sk_prot->disconnect(sk, 0);
955         } else if (sk->sk_state != DCCP_CLOSED) {
956                 dccp_terminate_connection(sk);
957         }
958
959         sk_stream_wait_close(sk, timeout);
960
961 adjudge_to_death:
962         state = sk->sk_state;
963         sock_hold(sk);
964         sock_orphan(sk);
965
966         /*
967          * It is the last release_sock in its life. It will remove backlog.
968          */
969         release_sock(sk);
970         /*
971          * Now socket is owned by kernel and we acquire BH lock
972          * to finish close. No need to check for user refs.
973          */
974         local_bh_disable();
975         bh_lock_sock(sk);
976         WARN_ON(sock_owned_by_user(sk));
977
978         percpu_counter_inc(sk->sk_prot->orphan_count);
979
980         /* Have we already been destroyed by a softirq or backlog? */
981         if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
982                 goto out;
983
984         if (sk->sk_state == DCCP_CLOSED)
985                 inet_csk_destroy_sock(sk);
986
987         /* Otherwise, socket is reprieved until protocol close. */
988
989 out:
990         bh_unlock_sock(sk);
991         local_bh_enable();
992         sock_put(sk);
993 }
994
995 EXPORT_SYMBOL_GPL(dccp_close);
996
997 void dccp_shutdown(struct sock *sk, int how)
998 {
999         dccp_pr_debug("called shutdown(%x)\n", how);
1000 }
1001
1002 EXPORT_SYMBOL_GPL(dccp_shutdown);
1003
1004 static inline int dccp_mib_init(void)
1005 {
1006         return snmp_mib_init((void**)dccp_statistics, sizeof(struct dccp_mib));
1007 }
1008
1009 static inline void dccp_mib_exit(void)
1010 {
1011         snmp_mib_free((void**)dccp_statistics);
1012 }
1013
1014 static int thash_entries;
1015 module_param(thash_entries, int, 0444);
1016 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1017
1018 #ifdef CONFIG_IP_DCCP_DEBUG
1019 int dccp_debug;
1020 module_param(dccp_debug, bool, 0644);
1021 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1022
1023 EXPORT_SYMBOL_GPL(dccp_debug);
1024 #endif
1025
1026 static int __init dccp_init(void)
1027 {
1028         unsigned long goal;
1029         int ehash_order, bhash_order, i;
1030         int rc;
1031
1032         BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1033                      FIELD_SIZEOF(struct sk_buff, cb));
1034         rc = percpu_counter_init(&dccp_orphan_count, 0);
1035         if (rc)
1036                 goto out;
1037         rc = -ENOBUFS;
1038         inet_hashinfo_init(&dccp_hashinfo);
1039         dccp_hashinfo.bind_bucket_cachep =
1040                 kmem_cache_create("dccp_bind_bucket",
1041                                   sizeof(struct inet_bind_bucket), 0,
1042                                   SLAB_HWCACHE_ALIGN, NULL);
1043         if (!dccp_hashinfo.bind_bucket_cachep)
1044                 goto out_free_percpu;
1045
1046         /*
1047          * Size and allocate the main established and bind bucket
1048          * hash tables.
1049          *
1050          * The methodology is similar to that of the buffer cache.
1051          */
1052         if (num_physpages >= (128 * 1024))
1053                 goal = num_physpages >> (21 - PAGE_SHIFT);
1054         else
1055                 goal = num_physpages >> (23 - PAGE_SHIFT);
1056
1057         if (thash_entries)
1058                 goal = (thash_entries *
1059                         sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1060         for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1061                 ;
1062         do {
1063                 dccp_hashinfo.ehash_size = (1UL << ehash_order) * PAGE_SIZE /
1064                                         sizeof(struct inet_ehash_bucket);
1065                 while (dccp_hashinfo.ehash_size &
1066                        (dccp_hashinfo.ehash_size - 1))
1067                         dccp_hashinfo.ehash_size--;
1068                 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1069                         __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1070         } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1071
1072         if (!dccp_hashinfo.ehash) {
1073                 DCCP_CRIT("Failed to allocate DCCP established hash table");
1074                 goto out_free_bind_bucket_cachep;
1075         }
1076
1077         for (i = 0; i < dccp_hashinfo.ehash_size; i++) {
1078                 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1079                 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].twchain, i);
1080         }
1081
1082         if (inet_ehash_locks_alloc(&dccp_hashinfo))
1083                         goto out_free_dccp_ehash;
1084
1085         bhash_order = ehash_order;
1086
1087         do {
1088                 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1089                                         sizeof(struct inet_bind_hashbucket);
1090                 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1091                     bhash_order > 0)
1092                         continue;
1093                 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1094                         __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1095         } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1096
1097         if (!dccp_hashinfo.bhash) {
1098                 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1099                 goto out_free_dccp_locks;
1100         }
1101
1102         for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1103                 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1104                 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1105         }
1106
1107         rc = dccp_mib_init();
1108         if (rc)
1109                 goto out_free_dccp_bhash;
1110
1111         rc = dccp_ackvec_init();
1112         if (rc)
1113                 goto out_free_dccp_mib;
1114
1115         rc = dccp_sysctl_init();
1116         if (rc)
1117                 goto out_ackvec_exit;
1118
1119         rc = ccid_initialize_builtins();
1120         if (rc)
1121                 goto out_sysctl_exit;
1122
1123         dccp_timestamping_init();
1124 out:
1125         return rc;
1126 out_sysctl_exit:
1127         dccp_sysctl_exit();
1128 out_ackvec_exit:
1129         dccp_ackvec_exit();
1130 out_free_dccp_mib:
1131         dccp_mib_exit();
1132 out_free_dccp_bhash:
1133         free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1134         dccp_hashinfo.bhash = NULL;
1135 out_free_dccp_locks:
1136         inet_ehash_locks_free(&dccp_hashinfo);
1137 out_free_dccp_ehash:
1138         free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1139         dccp_hashinfo.ehash = NULL;
1140 out_free_bind_bucket_cachep:
1141         kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1142         dccp_hashinfo.bind_bucket_cachep = NULL;
1143 out_free_percpu:
1144         percpu_counter_destroy(&dccp_orphan_count);
1145         goto out;
1146 }
1147
1148 static void __exit dccp_fini(void)
1149 {
1150         ccid_cleanup_builtins();
1151         dccp_mib_exit();
1152         free_pages((unsigned long)dccp_hashinfo.bhash,
1153                    get_order(dccp_hashinfo.bhash_size *
1154                              sizeof(struct inet_bind_hashbucket)));
1155         free_pages((unsigned long)dccp_hashinfo.ehash,
1156                    get_order(dccp_hashinfo.ehash_size *
1157                              sizeof(struct inet_ehash_bucket)));
1158         inet_ehash_locks_free(&dccp_hashinfo);
1159         kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1160         dccp_ackvec_exit();
1161         dccp_sysctl_exit();
1162         percpu_counter_destroy(&dccp_orphan_count);
1163 }
1164
1165 module_init(dccp_init);
1166 module_exit(dccp_fini);
1167
1168 MODULE_LICENSE("GPL");
1169 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1170 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");