Merge tag 'xfs-4.15-fixes-5' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
[sfrench/cifs-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 <linux/slab.h>
24 #include <net/checksum.h>
25
26 #include <net/inet_sock.h>
27 #include <net/inet_common.h>
28 #include <net/sock.h>
29 #include <net/xfrm.h>
30
31 #include <asm/ioctls.h>
32 #include <linux/spinlock.h>
33 #include <linux/timer.h>
34 #include <linux/delay.h>
35 #include <linux/poll.h>
36
37 #include "ccid.h"
38 #include "dccp.h"
39 #include "feat.h"
40
41 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
42
43 EXPORT_SYMBOL_GPL(dccp_statistics);
44
45 struct percpu_counter dccp_orphan_count;
46 EXPORT_SYMBOL_GPL(dccp_orphan_count);
47
48 struct inet_hashinfo dccp_hashinfo;
49 EXPORT_SYMBOL_GPL(dccp_hashinfo);
50
51 /* the maximum queue length for tx in packets. 0 is no limit */
52 int sysctl_dccp_tx_qlen __read_mostly = 5;
53
54 #ifdef CONFIG_IP_DCCP_DEBUG
55 static const char *dccp_state_name(const int state)
56 {
57         static const char *const dccp_state_names[] = {
58         [DCCP_OPEN]             = "OPEN",
59         [DCCP_REQUESTING]       = "REQUESTING",
60         [DCCP_PARTOPEN]         = "PARTOPEN",
61         [DCCP_LISTEN]           = "LISTEN",
62         [DCCP_RESPOND]          = "RESPOND",
63         [DCCP_CLOSING]          = "CLOSING",
64         [DCCP_ACTIVE_CLOSEREQ]  = "CLOSEREQ",
65         [DCCP_PASSIVE_CLOSE]    = "PASSIVE_CLOSE",
66         [DCCP_PASSIVE_CLOSEREQ] = "PASSIVE_CLOSEREQ",
67         [DCCP_TIME_WAIT]        = "TIME_WAIT",
68         [DCCP_CLOSED]           = "CLOSED",
69         };
70
71         if (state >= DCCP_MAX_STATES)
72                 return "INVALID STATE!";
73         else
74                 return dccp_state_names[state];
75 }
76 #endif
77
78 void dccp_set_state(struct sock *sk, const int state)
79 {
80         const int oldstate = sk->sk_state;
81
82         dccp_pr_debug("%s(%p)  %s  -->  %s\n", dccp_role(sk), sk,
83                       dccp_state_name(oldstate), dccp_state_name(state));
84         WARN_ON(state == oldstate);
85
86         switch (state) {
87         case DCCP_OPEN:
88                 if (oldstate != DCCP_OPEN)
89                         DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
90                 /* Client retransmits all Confirm options until entering OPEN */
91                 if (oldstate == DCCP_PARTOPEN)
92                         dccp_feat_list_purge(&dccp_sk(sk)->dccps_featneg);
93                 break;
94
95         case DCCP_CLOSED:
96                 if (oldstate == DCCP_OPEN || oldstate == DCCP_ACTIVE_CLOSEREQ ||
97                     oldstate == DCCP_CLOSING)
98                         DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
99
100                 sk->sk_prot->unhash(sk);
101                 if (inet_csk(sk)->icsk_bind_hash != NULL &&
102                     !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
103                         inet_put_port(sk);
104                 /* fall through */
105         default:
106                 if (oldstate == DCCP_OPEN)
107                         DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
108         }
109
110         /* Change state AFTER socket is unhashed to avoid closed
111          * socket sitting in hash tables.
112          */
113         sk->sk_state = state;
114 }
115
116 EXPORT_SYMBOL_GPL(dccp_set_state);
117
118 static void dccp_finish_passive_close(struct sock *sk)
119 {
120         switch (sk->sk_state) {
121         case DCCP_PASSIVE_CLOSE:
122                 /* Node (client or server) has received Close packet. */
123                 dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
124                 dccp_set_state(sk, DCCP_CLOSED);
125                 break;
126         case DCCP_PASSIVE_CLOSEREQ:
127                 /*
128                  * Client received CloseReq. We set the `active' flag so that
129                  * dccp_send_close() retransmits the Close as per RFC 4340, 8.3.
130                  */
131                 dccp_send_close(sk, 1);
132                 dccp_set_state(sk, DCCP_CLOSING);
133         }
134 }
135
136 void dccp_done(struct sock *sk)
137 {
138         dccp_set_state(sk, DCCP_CLOSED);
139         dccp_clear_xmit_timers(sk);
140
141         sk->sk_shutdown = SHUTDOWN_MASK;
142
143         if (!sock_flag(sk, SOCK_DEAD))
144                 sk->sk_state_change(sk);
145         else
146                 inet_csk_destroy_sock(sk);
147 }
148
149 EXPORT_SYMBOL_GPL(dccp_done);
150
151 const char *dccp_packet_name(const int type)
152 {
153         static const char *const dccp_packet_names[] = {
154                 [DCCP_PKT_REQUEST]  = "REQUEST",
155                 [DCCP_PKT_RESPONSE] = "RESPONSE",
156                 [DCCP_PKT_DATA]     = "DATA",
157                 [DCCP_PKT_ACK]      = "ACK",
158                 [DCCP_PKT_DATAACK]  = "DATAACK",
159                 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
160                 [DCCP_PKT_CLOSE]    = "CLOSE",
161                 [DCCP_PKT_RESET]    = "RESET",
162                 [DCCP_PKT_SYNC]     = "SYNC",
163                 [DCCP_PKT_SYNCACK]  = "SYNCACK",
164         };
165
166         if (type >= DCCP_NR_PKT_TYPES)
167                 return "INVALID";
168         else
169                 return dccp_packet_names[type];
170 }
171
172 EXPORT_SYMBOL_GPL(dccp_packet_name);
173
174 static void dccp_sk_destruct(struct sock *sk)
175 {
176         struct dccp_sock *dp = dccp_sk(sk);
177
178         ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
179         dp->dccps_hc_tx_ccid = NULL;
180         inet_sock_destruct(sk);
181 }
182
183 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
184 {
185         struct dccp_sock *dp = dccp_sk(sk);
186         struct inet_connection_sock *icsk = inet_csk(sk);
187
188         icsk->icsk_rto          = DCCP_TIMEOUT_INIT;
189         icsk->icsk_syn_retries  = sysctl_dccp_request_retries;
190         sk->sk_state            = DCCP_CLOSED;
191         sk->sk_write_space      = dccp_write_space;
192         sk->sk_destruct         = dccp_sk_destruct;
193         icsk->icsk_sync_mss     = dccp_sync_mss;
194         dp->dccps_mss_cache     = 536;
195         dp->dccps_rate_last     = jiffies;
196         dp->dccps_role          = DCCP_ROLE_UNDEFINED;
197         dp->dccps_service       = DCCP_SERVICE_CODE_IS_ABSENT;
198         dp->dccps_tx_qlen       = sysctl_dccp_tx_qlen;
199
200         dccp_init_xmit_timers(sk);
201
202         INIT_LIST_HEAD(&dp->dccps_featneg);
203         /* control socket doesn't need feat nego */
204         if (likely(ctl_sock_initialized))
205                 return dccp_feat_init(sk);
206         return 0;
207 }
208
209 EXPORT_SYMBOL_GPL(dccp_init_sock);
210
211 void dccp_destroy_sock(struct sock *sk)
212 {
213         struct dccp_sock *dp = dccp_sk(sk);
214
215         __skb_queue_purge(&sk->sk_write_queue);
216         if (sk->sk_send_head != NULL) {
217                 kfree_skb(sk->sk_send_head);
218                 sk->sk_send_head = NULL;
219         }
220
221         /* Clean up a referenced DCCP bind bucket. */
222         if (inet_csk(sk)->icsk_bind_hash != NULL)
223                 inet_put_port(sk);
224
225         kfree(dp->dccps_service_list);
226         dp->dccps_service_list = NULL;
227
228         if (dp->dccps_hc_rx_ackvec != NULL) {
229                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
230                 dp->dccps_hc_rx_ackvec = NULL;
231         }
232         ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
233         dp->dccps_hc_rx_ccid = NULL;
234
235         /* clean up feature negotiation state */
236         dccp_feat_list_purge(&dp->dccps_featneg);
237 }
238
239 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
240
241 static inline int dccp_listen_start(struct sock *sk, int backlog)
242 {
243         struct dccp_sock *dp = dccp_sk(sk);
244
245         dp->dccps_role = DCCP_ROLE_LISTEN;
246         /* do not start to listen if feature negotiation setup fails */
247         if (dccp_feat_finalise_settings(dp))
248                 return -EPROTO;
249         return inet_csk_listen_start(sk, backlog);
250 }
251
252 static inline int dccp_need_reset(int state)
253 {
254         return state != DCCP_CLOSED && state != DCCP_LISTEN &&
255                state != DCCP_REQUESTING;
256 }
257
258 int dccp_disconnect(struct sock *sk, int flags)
259 {
260         struct inet_connection_sock *icsk = inet_csk(sk);
261         struct inet_sock *inet = inet_sk(sk);
262         struct dccp_sock *dp = dccp_sk(sk);
263         int err = 0;
264         const int old_state = sk->sk_state;
265
266         if (old_state != DCCP_CLOSED)
267                 dccp_set_state(sk, DCCP_CLOSED);
268
269         /*
270          * This corresponds to the ABORT function of RFC793, sec. 3.8
271          * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
272          */
273         if (old_state == DCCP_LISTEN) {
274                 inet_csk_listen_stop(sk);
275         } else if (dccp_need_reset(old_state)) {
276                 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
277                 sk->sk_err = ECONNRESET;
278         } else if (old_state == DCCP_REQUESTING)
279                 sk->sk_err = ECONNRESET;
280
281         dccp_clear_xmit_timers(sk);
282         ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
283         ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
284         dp->dccps_hc_rx_ccid = NULL;
285         dp->dccps_hc_tx_ccid = NULL;
286
287         __skb_queue_purge(&sk->sk_receive_queue);
288         __skb_queue_purge(&sk->sk_write_queue);
289         if (sk->sk_send_head != NULL) {
290                 __kfree_skb(sk->sk_send_head);
291                 sk->sk_send_head = NULL;
292         }
293
294         inet->inet_dport = 0;
295
296         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
297                 inet_reset_saddr(sk);
298
299         sk->sk_shutdown = 0;
300         sock_reset_flag(sk, SOCK_DONE);
301
302         icsk->icsk_backoff = 0;
303         inet_csk_delack_init(sk);
304         __sk_dst_reset(sk);
305
306         WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
307
308         sk->sk_error_report(sk);
309         return err;
310 }
311
312 EXPORT_SYMBOL_GPL(dccp_disconnect);
313
314 /*
315  *      Wait for a DCCP event.
316  *
317  *      Note that we don't need to lock the socket, as the upper poll layers
318  *      take care of normal races (between the test and the event) and we don't
319  *      go look at any of the socket buffers directly.
320  */
321 unsigned int dccp_poll(struct file *file, struct socket *sock,
322                        poll_table *wait)
323 {
324         unsigned int mask;
325         struct sock *sk = sock->sk;
326
327         sock_poll_wait(file, sk_sleep(sk), wait);
328         if (sk->sk_state == DCCP_LISTEN)
329                 return inet_csk_listen_poll(sk);
330
331         /* Socket is not locked. We are protected from async events
332            by poll logic and correct handling of state changes
333            made by another threads is impossible in any case.
334          */
335
336         mask = 0;
337         if (sk->sk_err)
338                 mask = POLLERR;
339
340         if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
341                 mask |= POLLHUP;
342         if (sk->sk_shutdown & RCV_SHUTDOWN)
343                 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
344
345         /* Connected? */
346         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
347                 if (atomic_read(&sk->sk_rmem_alloc) > 0)
348                         mask |= POLLIN | POLLRDNORM;
349
350                 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
351                         if (sk_stream_is_writeable(sk)) {
352                                 mask |= POLLOUT | POLLWRNORM;
353                         } else {  /* send SIGIO later */
354                                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
355                                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
356
357                                 /* Race breaker. If space is freed after
358                                  * wspace test but before the flags are set,
359                                  * IO signal will be lost.
360                                  */
361                                 if (sk_stream_is_writeable(sk))
362                                         mask |= POLLOUT | POLLWRNORM;
363                         }
364                 }
365         }
366         return mask;
367 }
368
369 EXPORT_SYMBOL_GPL(dccp_poll);
370
371 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
372 {
373         int rc = -ENOTCONN;
374
375         lock_sock(sk);
376
377         if (sk->sk_state == DCCP_LISTEN)
378                 goto out;
379
380         switch (cmd) {
381         case SIOCINQ: {
382                 struct sk_buff *skb;
383                 unsigned long amount = 0;
384
385                 skb = skb_peek(&sk->sk_receive_queue);
386                 if (skb != NULL) {
387                         /*
388                          * We will only return the amount of this packet since
389                          * that is all that will be read.
390                          */
391                         amount = skb->len;
392                 }
393                 rc = put_user(amount, (int __user *)arg);
394         }
395                 break;
396         default:
397                 rc = -ENOIOCTLCMD;
398                 break;
399         }
400 out:
401         release_sock(sk);
402         return rc;
403 }
404
405 EXPORT_SYMBOL_GPL(dccp_ioctl);
406
407 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
408                                    char __user *optval, unsigned int optlen)
409 {
410         struct dccp_sock *dp = dccp_sk(sk);
411         struct dccp_service_list *sl = NULL;
412
413         if (service == DCCP_SERVICE_INVALID_VALUE ||
414             optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
415                 return -EINVAL;
416
417         if (optlen > sizeof(service)) {
418                 sl = kmalloc(optlen, GFP_KERNEL);
419                 if (sl == NULL)
420                         return -ENOMEM;
421
422                 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
423                 if (copy_from_user(sl->dccpsl_list,
424                                    optval + sizeof(service),
425                                    optlen - sizeof(service)) ||
426                     dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
427                         kfree(sl);
428                         return -EFAULT;
429                 }
430         }
431
432         lock_sock(sk);
433         dp->dccps_service = service;
434
435         kfree(dp->dccps_service_list);
436
437         dp->dccps_service_list = sl;
438         release_sock(sk);
439         return 0;
440 }
441
442 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
443 {
444         u8 *list, len;
445         int i, rc;
446
447         if (cscov < 0 || cscov > 15)
448                 return -EINVAL;
449         /*
450          * Populate a list of permissible values, in the range cscov...15. This
451          * is necessary since feature negotiation of single values only works if
452          * both sides incidentally choose the same value. Since the list starts
453          * lowest-value first, negotiation will pick the smallest shared value.
454          */
455         if (cscov == 0)
456                 return 0;
457         len = 16 - cscov;
458
459         list = kmalloc(len, GFP_KERNEL);
460         if (list == NULL)
461                 return -ENOBUFS;
462
463         for (i = 0; i < len; i++)
464                 list[i] = cscov++;
465
466         rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
467
468         if (rc == 0) {
469                 if (rx)
470                         dccp_sk(sk)->dccps_pcrlen = cscov;
471                 else
472                         dccp_sk(sk)->dccps_pcslen = cscov;
473         }
474         kfree(list);
475         return rc;
476 }
477
478 static int dccp_setsockopt_ccid(struct sock *sk, int type,
479                                 char __user *optval, unsigned int optlen)
480 {
481         u8 *val;
482         int rc = 0;
483
484         if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
485                 return -EINVAL;
486
487         val = memdup_user(optval, optlen);
488         if (IS_ERR(val))
489                 return PTR_ERR(val);
490
491         lock_sock(sk);
492         if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
493                 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
494
495         if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
496                 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
497         release_sock(sk);
498
499         kfree(val);
500         return rc;
501 }
502
503 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
504                 char __user *optval, unsigned int optlen)
505 {
506         struct dccp_sock *dp = dccp_sk(sk);
507         int val, err = 0;
508
509         switch (optname) {
510         case DCCP_SOCKOPT_PACKET_SIZE:
511                 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
512                 return 0;
513         case DCCP_SOCKOPT_CHANGE_L:
514         case DCCP_SOCKOPT_CHANGE_R:
515                 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
516                 return 0;
517         case DCCP_SOCKOPT_CCID:
518         case DCCP_SOCKOPT_RX_CCID:
519         case DCCP_SOCKOPT_TX_CCID:
520                 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
521         }
522
523         if (optlen < (int)sizeof(int))
524                 return -EINVAL;
525
526         if (get_user(val, (int __user *)optval))
527                 return -EFAULT;
528
529         if (optname == DCCP_SOCKOPT_SERVICE)
530                 return dccp_setsockopt_service(sk, val, optval, optlen);
531
532         lock_sock(sk);
533         switch (optname) {
534         case DCCP_SOCKOPT_SERVER_TIMEWAIT:
535                 if (dp->dccps_role != DCCP_ROLE_SERVER)
536                         err = -EOPNOTSUPP;
537                 else
538                         dp->dccps_server_timewait = (val != 0);
539                 break;
540         case DCCP_SOCKOPT_SEND_CSCOV:
541                 err = dccp_setsockopt_cscov(sk, val, false);
542                 break;
543         case DCCP_SOCKOPT_RECV_CSCOV:
544                 err = dccp_setsockopt_cscov(sk, val, true);
545                 break;
546         case DCCP_SOCKOPT_QPOLICY_ID:
547                 if (sk->sk_state != DCCP_CLOSED)
548                         err = -EISCONN;
549                 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
550                         err = -EINVAL;
551                 else
552                         dp->dccps_qpolicy = val;
553                 break;
554         case DCCP_SOCKOPT_QPOLICY_TXQLEN:
555                 if (val < 0)
556                         err = -EINVAL;
557                 else
558                         dp->dccps_tx_qlen = val;
559                 break;
560         default:
561                 err = -ENOPROTOOPT;
562                 break;
563         }
564         release_sock(sk);
565
566         return err;
567 }
568
569 int dccp_setsockopt(struct sock *sk, int level, int optname,
570                     char __user *optval, unsigned int optlen)
571 {
572         if (level != SOL_DCCP)
573                 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
574                                                              optname, optval,
575                                                              optlen);
576         return do_dccp_setsockopt(sk, level, optname, optval, optlen);
577 }
578
579 EXPORT_SYMBOL_GPL(dccp_setsockopt);
580
581 #ifdef CONFIG_COMPAT
582 int compat_dccp_setsockopt(struct sock *sk, int level, int optname,
583                            char __user *optval, unsigned int optlen)
584 {
585         if (level != SOL_DCCP)
586                 return inet_csk_compat_setsockopt(sk, level, optname,
587                                                   optval, optlen);
588         return do_dccp_setsockopt(sk, level, optname, optval, optlen);
589 }
590
591 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt);
592 #endif
593
594 static int dccp_getsockopt_service(struct sock *sk, int len,
595                                    __be32 __user *optval,
596                                    int __user *optlen)
597 {
598         const struct dccp_sock *dp = dccp_sk(sk);
599         const struct dccp_service_list *sl;
600         int err = -ENOENT, slen = 0, total_len = sizeof(u32);
601
602         lock_sock(sk);
603         if ((sl = dp->dccps_service_list) != NULL) {
604                 slen = sl->dccpsl_nr * sizeof(u32);
605                 total_len += slen;
606         }
607
608         err = -EINVAL;
609         if (total_len > len)
610                 goto out;
611
612         err = 0;
613         if (put_user(total_len, optlen) ||
614             put_user(dp->dccps_service, optval) ||
615             (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
616                 err = -EFAULT;
617 out:
618         release_sock(sk);
619         return err;
620 }
621
622 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
623                     char __user *optval, int __user *optlen)
624 {
625         struct dccp_sock *dp;
626         int val, len;
627
628         if (get_user(len, optlen))
629                 return -EFAULT;
630
631         if (len < (int)sizeof(int))
632                 return -EINVAL;
633
634         dp = dccp_sk(sk);
635
636         switch (optname) {
637         case DCCP_SOCKOPT_PACKET_SIZE:
638                 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
639                 return 0;
640         case DCCP_SOCKOPT_SERVICE:
641                 return dccp_getsockopt_service(sk, len,
642                                                (__be32 __user *)optval, optlen);
643         case DCCP_SOCKOPT_GET_CUR_MPS:
644                 val = dp->dccps_mss_cache;
645                 break;
646         case DCCP_SOCKOPT_AVAILABLE_CCIDS:
647                 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
648         case DCCP_SOCKOPT_TX_CCID:
649                 val = ccid_get_current_tx_ccid(dp);
650                 if (val < 0)
651                         return -ENOPROTOOPT;
652                 break;
653         case DCCP_SOCKOPT_RX_CCID:
654                 val = ccid_get_current_rx_ccid(dp);
655                 if (val < 0)
656                         return -ENOPROTOOPT;
657                 break;
658         case DCCP_SOCKOPT_SERVER_TIMEWAIT:
659                 val = dp->dccps_server_timewait;
660                 break;
661         case DCCP_SOCKOPT_SEND_CSCOV:
662                 val = dp->dccps_pcslen;
663                 break;
664         case DCCP_SOCKOPT_RECV_CSCOV:
665                 val = dp->dccps_pcrlen;
666                 break;
667         case DCCP_SOCKOPT_QPOLICY_ID:
668                 val = dp->dccps_qpolicy;
669                 break;
670         case DCCP_SOCKOPT_QPOLICY_TXQLEN:
671                 val = dp->dccps_tx_qlen;
672                 break;
673         case 128 ... 191:
674                 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
675                                              len, (u32 __user *)optval, optlen);
676         case 192 ... 255:
677                 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
678                                              len, (u32 __user *)optval, optlen);
679         default:
680                 return -ENOPROTOOPT;
681         }
682
683         len = sizeof(val);
684         if (put_user(len, optlen) || copy_to_user(optval, &val, len))
685                 return -EFAULT;
686
687         return 0;
688 }
689
690 int dccp_getsockopt(struct sock *sk, int level, int optname,
691                     char __user *optval, int __user *optlen)
692 {
693         if (level != SOL_DCCP)
694                 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
695                                                              optname, optval,
696                                                              optlen);
697         return do_dccp_getsockopt(sk, level, optname, optval, optlen);
698 }
699
700 EXPORT_SYMBOL_GPL(dccp_getsockopt);
701
702 #ifdef CONFIG_COMPAT
703 int compat_dccp_getsockopt(struct sock *sk, int level, int optname,
704                            char __user *optval, int __user *optlen)
705 {
706         if (level != SOL_DCCP)
707                 return inet_csk_compat_getsockopt(sk, level, optname,
708                                                   optval, optlen);
709         return do_dccp_getsockopt(sk, level, optname, optval, optlen);
710 }
711
712 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt);
713 #endif
714
715 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
716 {
717         struct cmsghdr *cmsg;
718
719         /*
720          * Assign an (opaque) qpolicy priority value to skb->priority.
721          *
722          * We are overloading this skb field for use with the qpolicy subystem.
723          * The skb->priority is normally used for the SO_PRIORITY option, which
724          * is initialised from sk_priority. Since the assignment of sk_priority
725          * to skb->priority happens later (on layer 3), we overload this field
726          * for use with queueing priorities as long as the skb is on layer 4.
727          * The default priority value (if nothing is set) is 0.
728          */
729         skb->priority = 0;
730
731         for_each_cmsghdr(cmsg, msg) {
732                 if (!CMSG_OK(msg, cmsg))
733                         return -EINVAL;
734
735                 if (cmsg->cmsg_level != SOL_DCCP)
736                         continue;
737
738                 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
739                     !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
740                         return -EINVAL;
741
742                 switch (cmsg->cmsg_type) {
743                 case DCCP_SCM_PRIORITY:
744                         if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
745                                 return -EINVAL;
746                         skb->priority = *(__u32 *)CMSG_DATA(cmsg);
747                         break;
748                 default:
749                         return -EINVAL;
750                 }
751         }
752         return 0;
753 }
754
755 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
756 {
757         const struct dccp_sock *dp = dccp_sk(sk);
758         const int flags = msg->msg_flags;
759         const int noblock = flags & MSG_DONTWAIT;
760         struct sk_buff *skb;
761         int rc, size;
762         long timeo;
763
764         if (len > dp->dccps_mss_cache)
765                 return -EMSGSIZE;
766
767         lock_sock(sk);
768
769         if (dccp_qpolicy_full(sk)) {
770                 rc = -EAGAIN;
771                 goto out_release;
772         }
773
774         timeo = sock_sndtimeo(sk, noblock);
775
776         /*
777          * We have to use sk_stream_wait_connect here to set sk_write_pending,
778          * so that the trick in dccp_rcv_request_sent_state_process.
779          */
780         /* Wait for a connection to finish. */
781         if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
782                 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
783                         goto out_release;
784
785         size = sk->sk_prot->max_header + len;
786         release_sock(sk);
787         skb = sock_alloc_send_skb(sk, size, noblock, &rc);
788         lock_sock(sk);
789         if (skb == NULL)
790                 goto out_release;
791
792         skb_reserve(skb, sk->sk_prot->max_header);
793         rc = memcpy_from_msg(skb_put(skb, len), msg, len);
794         if (rc != 0)
795                 goto out_discard;
796
797         rc = dccp_msghdr_parse(msg, skb);
798         if (rc != 0)
799                 goto out_discard;
800
801         dccp_qpolicy_push(sk, skb);
802         /*
803          * The xmit_timer is set if the TX CCID is rate-based and will expire
804          * when congestion control permits to release further packets into the
805          * network. Window-based CCIDs do not use this timer.
806          */
807         if (!timer_pending(&dp->dccps_xmit_timer))
808                 dccp_write_xmit(sk);
809 out_release:
810         release_sock(sk);
811         return rc ? : len;
812 out_discard:
813         kfree_skb(skb);
814         goto out_release;
815 }
816
817 EXPORT_SYMBOL_GPL(dccp_sendmsg);
818
819 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
820                  int flags, int *addr_len)
821 {
822         const struct dccp_hdr *dh;
823         long timeo;
824
825         lock_sock(sk);
826
827         if (sk->sk_state == DCCP_LISTEN) {
828                 len = -ENOTCONN;
829                 goto out;
830         }
831
832         timeo = sock_rcvtimeo(sk, nonblock);
833
834         do {
835                 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
836
837                 if (skb == NULL)
838                         goto verify_sock_status;
839
840                 dh = dccp_hdr(skb);
841
842                 switch (dh->dccph_type) {
843                 case DCCP_PKT_DATA:
844                 case DCCP_PKT_DATAACK:
845                         goto found_ok_skb;
846
847                 case DCCP_PKT_CLOSE:
848                 case DCCP_PKT_CLOSEREQ:
849                         if (!(flags & MSG_PEEK))
850                                 dccp_finish_passive_close(sk);
851                         /* fall through */
852                 case DCCP_PKT_RESET:
853                         dccp_pr_debug("found fin (%s) ok!\n",
854                                       dccp_packet_name(dh->dccph_type));
855                         len = 0;
856                         goto found_fin_ok;
857                 default:
858                         dccp_pr_debug("packet_type=%s\n",
859                                       dccp_packet_name(dh->dccph_type));
860                         sk_eat_skb(sk, skb);
861                 }
862 verify_sock_status:
863                 if (sock_flag(sk, SOCK_DONE)) {
864                         len = 0;
865                         break;
866                 }
867
868                 if (sk->sk_err) {
869                         len = sock_error(sk);
870                         break;
871                 }
872
873                 if (sk->sk_shutdown & RCV_SHUTDOWN) {
874                         len = 0;
875                         break;
876                 }
877
878                 if (sk->sk_state == DCCP_CLOSED) {
879                         if (!sock_flag(sk, SOCK_DONE)) {
880                                 /* This occurs when user tries to read
881                                  * from never connected socket.
882                                  */
883                                 len = -ENOTCONN;
884                                 break;
885                         }
886                         len = 0;
887                         break;
888                 }
889
890                 if (!timeo) {
891                         len = -EAGAIN;
892                         break;
893                 }
894
895                 if (signal_pending(current)) {
896                         len = sock_intr_errno(timeo);
897                         break;
898                 }
899
900                 sk_wait_data(sk, &timeo, NULL);
901                 continue;
902         found_ok_skb:
903                 if (len > skb->len)
904                         len = skb->len;
905                 else if (len < skb->len)
906                         msg->msg_flags |= MSG_TRUNC;
907
908                 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
909                         /* Exception. Bailout! */
910                         len = -EFAULT;
911                         break;
912                 }
913                 if (flags & MSG_TRUNC)
914                         len = skb->len;
915         found_fin_ok:
916                 if (!(flags & MSG_PEEK))
917                         sk_eat_skb(sk, skb);
918                 break;
919         } while (1);
920 out:
921         release_sock(sk);
922         return len;
923 }
924
925 EXPORT_SYMBOL_GPL(dccp_recvmsg);
926
927 int inet_dccp_listen(struct socket *sock, int backlog)
928 {
929         struct sock *sk = sock->sk;
930         unsigned char old_state;
931         int err;
932
933         lock_sock(sk);
934
935         err = -EINVAL;
936         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
937                 goto out;
938
939         old_state = sk->sk_state;
940         if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
941                 goto out;
942
943         /* Really, if the socket is already in listen state
944          * we can only allow the backlog to be adjusted.
945          */
946         if (old_state != DCCP_LISTEN) {
947                 /*
948                  * FIXME: here it probably should be sk->sk_prot->listen_start
949                  * see tcp_listen_start
950                  */
951                 err = dccp_listen_start(sk, backlog);
952                 if (err)
953                         goto out;
954         }
955         sk->sk_max_ack_backlog = backlog;
956         err = 0;
957
958 out:
959         release_sock(sk);
960         return err;
961 }
962
963 EXPORT_SYMBOL_GPL(inet_dccp_listen);
964
965 static void dccp_terminate_connection(struct sock *sk)
966 {
967         u8 next_state = DCCP_CLOSED;
968
969         switch (sk->sk_state) {
970         case DCCP_PASSIVE_CLOSE:
971         case DCCP_PASSIVE_CLOSEREQ:
972                 dccp_finish_passive_close(sk);
973                 break;
974         case DCCP_PARTOPEN:
975                 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
976                 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
977                 /* fall through */
978         case DCCP_OPEN:
979                 dccp_send_close(sk, 1);
980
981                 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
982                     !dccp_sk(sk)->dccps_server_timewait)
983                         next_state = DCCP_ACTIVE_CLOSEREQ;
984                 else
985                         next_state = DCCP_CLOSING;
986                 /* fall through */
987         default:
988                 dccp_set_state(sk, next_state);
989         }
990 }
991
992 void dccp_close(struct sock *sk, long timeout)
993 {
994         struct dccp_sock *dp = dccp_sk(sk);
995         struct sk_buff *skb;
996         u32 data_was_unread = 0;
997         int state;
998
999         lock_sock(sk);
1000
1001         sk->sk_shutdown = SHUTDOWN_MASK;
1002
1003         if (sk->sk_state == DCCP_LISTEN) {
1004                 dccp_set_state(sk, DCCP_CLOSED);
1005
1006                 /* Special case. */
1007                 inet_csk_listen_stop(sk);
1008
1009                 goto adjudge_to_death;
1010         }
1011
1012         sk_stop_timer(sk, &dp->dccps_xmit_timer);
1013
1014         /*
1015          * We need to flush the recv. buffs.  We do this only on the
1016          * descriptor close, not protocol-sourced closes, because the
1017           *reader process may not have drained the data yet!
1018          */
1019         while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1020                 data_was_unread += skb->len;
1021                 __kfree_skb(skb);
1022         }
1023
1024         /* If socket has been already reset kill it. */
1025         if (sk->sk_state == DCCP_CLOSED)
1026                 goto adjudge_to_death;
1027
1028         if (data_was_unread) {
1029                 /* Unread data was tossed, send an appropriate Reset Code */
1030                 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1031                 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1032                 dccp_set_state(sk, DCCP_CLOSED);
1033         } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1034                 /* Check zero linger _after_ checking for unread data. */
1035                 sk->sk_prot->disconnect(sk, 0);
1036         } else if (sk->sk_state != DCCP_CLOSED) {
1037                 /*
1038                  * Normal connection termination. May need to wait if there are
1039                  * still packets in the TX queue that are delayed by the CCID.
1040                  */
1041                 dccp_flush_write_queue(sk, &timeout);
1042                 dccp_terminate_connection(sk);
1043         }
1044
1045         /*
1046          * Flush write queue. This may be necessary in several cases:
1047          * - we have been closed by the peer but still have application data;
1048          * - abortive termination (unread data or zero linger time),
1049          * - normal termination but queue could not be flushed within time limit
1050          */
1051         __skb_queue_purge(&sk->sk_write_queue);
1052
1053         sk_stream_wait_close(sk, timeout);
1054
1055 adjudge_to_death:
1056         state = sk->sk_state;
1057         sock_hold(sk);
1058         sock_orphan(sk);
1059
1060         /*
1061          * It is the last release_sock in its life. It will remove backlog.
1062          */
1063         release_sock(sk);
1064         /*
1065          * Now socket is owned by kernel and we acquire BH lock
1066          * to finish close. No need to check for user refs.
1067          */
1068         local_bh_disable();
1069         bh_lock_sock(sk);
1070         WARN_ON(sock_owned_by_user(sk));
1071
1072         percpu_counter_inc(sk->sk_prot->orphan_count);
1073
1074         /* Have we already been destroyed by a softirq or backlog? */
1075         if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1076                 goto out;
1077
1078         if (sk->sk_state == DCCP_CLOSED)
1079                 inet_csk_destroy_sock(sk);
1080
1081         /* Otherwise, socket is reprieved until protocol close. */
1082
1083 out:
1084         bh_unlock_sock(sk);
1085         local_bh_enable();
1086         sock_put(sk);
1087 }
1088
1089 EXPORT_SYMBOL_GPL(dccp_close);
1090
1091 void dccp_shutdown(struct sock *sk, int how)
1092 {
1093         dccp_pr_debug("called shutdown(%x)\n", how);
1094 }
1095
1096 EXPORT_SYMBOL_GPL(dccp_shutdown);
1097
1098 static inline int __init dccp_mib_init(void)
1099 {
1100         dccp_statistics = alloc_percpu(struct dccp_mib);
1101         if (!dccp_statistics)
1102                 return -ENOMEM;
1103         return 0;
1104 }
1105
1106 static inline void dccp_mib_exit(void)
1107 {
1108         free_percpu(dccp_statistics);
1109 }
1110
1111 static int thash_entries;
1112 module_param(thash_entries, int, 0444);
1113 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1114
1115 #ifdef CONFIG_IP_DCCP_DEBUG
1116 bool dccp_debug;
1117 module_param(dccp_debug, bool, 0644);
1118 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1119
1120 EXPORT_SYMBOL_GPL(dccp_debug);
1121 #endif
1122
1123 static int __init dccp_init(void)
1124 {
1125         unsigned long goal;
1126         int ehash_order, bhash_order, i;
1127         int rc;
1128
1129         BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1130                      FIELD_SIZEOF(struct sk_buff, cb));
1131         rc = percpu_counter_init(&dccp_orphan_count, 0, GFP_KERNEL);
1132         if (rc)
1133                 goto out_fail;
1134         rc = -ENOBUFS;
1135         inet_hashinfo_init(&dccp_hashinfo);
1136         dccp_hashinfo.bind_bucket_cachep =
1137                 kmem_cache_create("dccp_bind_bucket",
1138                                   sizeof(struct inet_bind_bucket), 0,
1139                                   SLAB_HWCACHE_ALIGN, NULL);
1140         if (!dccp_hashinfo.bind_bucket_cachep)
1141                 goto out_free_percpu;
1142
1143         /*
1144          * Size and allocate the main established and bind bucket
1145          * hash tables.
1146          *
1147          * The methodology is similar to that of the buffer cache.
1148          */
1149         if (totalram_pages >= (128 * 1024))
1150                 goal = totalram_pages >> (21 - PAGE_SHIFT);
1151         else
1152                 goal = totalram_pages >> (23 - PAGE_SHIFT);
1153
1154         if (thash_entries)
1155                 goal = (thash_entries *
1156                         sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1157         for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1158                 ;
1159         do {
1160                 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1161                                         sizeof(struct inet_ehash_bucket);
1162
1163                 while (hash_size & (hash_size - 1))
1164                         hash_size--;
1165                 dccp_hashinfo.ehash_mask = hash_size - 1;
1166                 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1167                         __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1168         } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1169
1170         if (!dccp_hashinfo.ehash) {
1171                 DCCP_CRIT("Failed to allocate DCCP established hash table");
1172                 goto out_free_bind_bucket_cachep;
1173         }
1174
1175         for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1176                 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1177
1178         if (inet_ehash_locks_alloc(&dccp_hashinfo))
1179                         goto out_free_dccp_ehash;
1180
1181         bhash_order = ehash_order;
1182
1183         do {
1184                 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1185                                         sizeof(struct inet_bind_hashbucket);
1186                 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1187                     bhash_order > 0)
1188                         continue;
1189                 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1190                         __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1191         } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1192
1193         if (!dccp_hashinfo.bhash) {
1194                 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1195                 goto out_free_dccp_locks;
1196         }
1197
1198         for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1199                 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1200                 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1201         }
1202
1203         rc = dccp_mib_init();
1204         if (rc)
1205                 goto out_free_dccp_bhash;
1206
1207         rc = dccp_ackvec_init();
1208         if (rc)
1209                 goto out_free_dccp_mib;
1210
1211         rc = dccp_sysctl_init();
1212         if (rc)
1213                 goto out_ackvec_exit;
1214
1215         rc = ccid_initialize_builtins();
1216         if (rc)
1217                 goto out_sysctl_exit;
1218
1219         dccp_timestamping_init();
1220
1221         return 0;
1222
1223 out_sysctl_exit:
1224         dccp_sysctl_exit();
1225 out_ackvec_exit:
1226         dccp_ackvec_exit();
1227 out_free_dccp_mib:
1228         dccp_mib_exit();
1229 out_free_dccp_bhash:
1230         free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1231 out_free_dccp_locks:
1232         inet_ehash_locks_free(&dccp_hashinfo);
1233 out_free_dccp_ehash:
1234         free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1235 out_free_bind_bucket_cachep:
1236         kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1237 out_free_percpu:
1238         percpu_counter_destroy(&dccp_orphan_count);
1239 out_fail:
1240         dccp_hashinfo.bhash = NULL;
1241         dccp_hashinfo.ehash = NULL;
1242         dccp_hashinfo.bind_bucket_cachep = NULL;
1243         return rc;
1244 }
1245
1246 static void __exit dccp_fini(void)
1247 {
1248         ccid_cleanup_builtins();
1249         dccp_mib_exit();
1250         free_pages((unsigned long)dccp_hashinfo.bhash,
1251                    get_order(dccp_hashinfo.bhash_size *
1252                              sizeof(struct inet_bind_hashbucket)));
1253         free_pages((unsigned long)dccp_hashinfo.ehash,
1254                    get_order((dccp_hashinfo.ehash_mask + 1) *
1255                              sizeof(struct inet_ehash_bucket)));
1256         inet_ehash_locks_free(&dccp_hashinfo);
1257         kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1258         dccp_ackvec_exit();
1259         dccp_sysctl_exit();
1260         percpu_counter_destroy(&dccp_orphan_count);
1261 }
1262
1263 module_init(dccp_init);
1264 module_exit(dccp_fini);
1265
1266 MODULE_LICENSE("GPL");
1267 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1268 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");