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