?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
break out numa_maps gather_pte_stats() checks
[linux-2.6.git] / net / dccp / proto.c
blob152975d942d9a59a7c26756d0f22419b9e298425
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/sock.h>
28 #include <net/xfrm.h>
29
30 #include <asm/ioctls.h>
31 #include <linux/spinlock.h>
32 #include <linux/timer.h>
33 #include <linux/delay.h>
34 #include <linux/poll.h>
35
36 #include "ccid.h"
37 #include "dccp.h"
38 #include "feat.h"
39
40 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
41
42 EXPORT_SYMBOL_GPL(dccp_statistics);
43
44 struct percpu_counter dccp_orphan_count;
45 EXPORT_SYMBOL_GPL(dccp_orphan_count);
46
47 struct inet_hashinfo dccp_hashinfo;
48 EXPORT_SYMBOL_GPL(dccp_hashinfo);
49
50 /* the maximum queue length for tx in packets. 0 is no limit */
51 int sysctl_dccp_tx_qlen __read_mostly = 5;
52
53 #ifdef CONFIG_IP_DCCP_DEBUG
54 static const char *dccp_state_name(const int state)
55 {
56 static const char *const dccp_state_names[] = {
57 [DCCP_OPEN] = "OPEN",
58 [DCCP_REQUESTING] = "REQUESTING",
59 [DCCP_PARTOPEN] = "PARTOPEN",
60 [DCCP_LISTEN] = "LISTEN",
61 [DCCP_RESPOND] = "RESPOND",
62 [DCCP_CLOSING] = "CLOSING",
63 [DCCP_ACTIVE_CLOSEREQ] = "CLOSEREQ",
64 [DCCP_PASSIVE_CLOSE] = "PASSIVE_CLOSE",
65 [DCCP_PASSIVE_CLOSEREQ] = "PASSIVE_CLOSEREQ",
66 [DCCP_TIME_WAIT] = "TIME_WAIT",
67 [DCCP_CLOSED] = "CLOSED",
68 };
69
70 if (state >= DCCP_MAX_STATES)
71 return "INVALID STATE!";
72 else
73 return dccp_state_names[state];
74 }
75 #endif
76
77 void dccp_set_state(struct sock *sk, const int state)
78 {
79 const int oldstate = sk->sk_state;
80
81 dccp_pr_debug("%s(%p) %s --> %s\n", dccp_role(sk), sk,
82 dccp_state_name(oldstate), dccp_state_name(state));
83 WARN_ON(state == oldstate);
84
85 switch (state) {
86 case DCCP_OPEN:
87 if (oldstate != DCCP_OPEN)
88 DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
89 /* Client retransmits all Confirm options until entering OPEN */
90 if (oldstate == DCCP_PARTOPEN)
91 dccp_feat_list_purge(&dccp_sk(sk)->dccps_featneg);
92 break;
93
94 case DCCP_CLOSED:
95 if (oldstate == DCCP_OPEN || oldstate == DCCP_ACTIVE_CLOSEREQ ||
96 oldstate == DCCP_CLOSING)
97 DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
98
99 sk->sk_prot->unhash(sk);
100 if (inet_csk(sk)->icsk_bind_hash != NULL &&
101 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
102 inet_put_port(sk);
103 /* fall through */
104 default:
105 if (oldstate == DCCP_OPEN)
106 DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
107 }
108
109 /* Change state AFTER socket is unhashed to avoid closed
110 * socket sitting in hash tables.
111 */
112 sk->sk_state = state;
113 }
114
115 EXPORT_SYMBOL_GPL(dccp_set_state);
116
117 static void dccp_finish_passive_close(struct sock *sk)
118 {
119 switch (sk->sk_state) {
120 case DCCP_PASSIVE_CLOSE:
121 /* Node (client or server) has received Close packet. */
122 dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
123 dccp_set_state(sk, DCCP_CLOSED);
124 break;
125 case DCCP_PASSIVE_CLOSEREQ:
126 /*
127 * Client received CloseReq. We set the `active' flag so that
128 * dccp_send_close() retransmits the Close as per RFC 4340, 8.3.
129 */
130 dccp_send_close(sk, 1);
131 dccp_set_state(sk, DCCP_CLOSING);
132 }
133 }
134
135 void dccp_done(struct sock *sk)
136 {
137 dccp_set_state(sk, DCCP_CLOSED);
138 dccp_clear_xmit_timers(sk);
139
140 sk->sk_shutdown = SHUTDOWN_MASK;
141
142 if (!sock_flag(sk, SOCK_DEAD))
143 sk->sk_state_change(sk);
144 else
145 inet_csk_destroy_sock(sk);
146 }
147
148 EXPORT_SYMBOL_GPL(dccp_done);
149
150 const char *dccp_packet_name(const int type)
151 {
152 static const char *const dccp_packet_names[] = {
153 [DCCP_PKT_REQUEST] = "REQUEST",
154 [DCCP_PKT_RESPONSE] = "RESPONSE",
155 [DCCP_PKT_DATA] = "DATA",
156 [DCCP_PKT_ACK] = "ACK",
157 [DCCP_PKT_DATAACK] = "DATAACK",
158 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
159 [DCCP_PKT_CLOSE] = "CLOSE",
160 [DCCP_PKT_RESET] = "RESET",
161 [DCCP_PKT_SYNC] = "SYNC",
162 [DCCP_PKT_SYNCACK] = "SYNCACK",
163 };
164
165 if (type >= DCCP_NR_PKT_TYPES)
166 return "INVALID";
167 else
168 return dccp_packet_names[type];
169 }
170
171 EXPORT_SYMBOL_GPL(dccp_packet_name);
172
173 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
174 {
175 struct dccp_sock *dp = dccp_sk(sk);
176 struct inet_connection_sock *icsk = inet_csk(sk);
177
178 icsk->icsk_rto = DCCP_TIMEOUT_INIT;
179 icsk->icsk_syn_retries = sysctl_dccp_request_retries;
180 sk->sk_state = DCCP_CLOSED;
181 sk->sk_write_space = dccp_write_space;
182 icsk->icsk_sync_mss = dccp_sync_mss;
183 dp->dccps_mss_cache = 536;
184 dp->dccps_rate_last = jiffies;
185 dp->dccps_role = DCCP_ROLE_UNDEFINED;
186 dp->dccps_service = DCCP_SERVICE_CODE_IS_ABSENT;
187 dp->dccps_l_ack_ratio = dp->dccps_r_ack_ratio = 1;
188 dp->dccps_tx_qlen = sysctl_dccp_tx_qlen;
189
190 dccp_init_xmit_timers(sk);
191
192 INIT_LIST_HEAD(&dp->dccps_featneg);
193 /* control socket doesn't need feat nego */
194 if (likely(ctl_sock_initialized))
195 return dccp_feat_init(sk);
196 return 0;
197 }
198
199 EXPORT_SYMBOL_GPL(dccp_init_sock);
200
201 void dccp_destroy_sock(struct sock *sk)
202 {
203 struct dccp_sock *dp = dccp_sk(sk);
204
205 /*
206 * DCCP doesn't use sk_write_queue, just sk_send_head
207 * for retransmissions
208 */
209 if (sk->sk_send_head != NULL) {
210 kfree_skb(sk->sk_send_head);
211 sk->sk_send_head = NULL;
212 }
213
214 /* Clean up a referenced DCCP bind bucket. */
215 if (inet_csk(sk)->icsk_bind_hash != NULL)
216 inet_put_port(sk);
217
218 kfree(dp->dccps_service_list);
219 dp->dccps_service_list = NULL;
220
221 if (dp->dccps_hc_rx_ackvec != NULL) {
222 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
223 dp->dccps_hc_rx_ackvec = NULL;
224 }
225 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
226 ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
227 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
228
229 /* clean up feature negotiation state */
230 dccp_feat_list_purge(&dp->dccps_featneg);
231 }
232
233 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
234
235 static inline int dccp_listen_start(struct sock *sk, int backlog)
236 {
237 struct dccp_sock *dp = dccp_sk(sk);
238
239 dp->dccps_role = DCCP_ROLE_LISTEN;
240 /* do not start to listen if feature negotiation setup fails */
241 if (dccp_feat_finalise_settings(dp))
242 return -EPROTO;
243 return inet_csk_listen_start(sk, backlog);
244 }
245
246 static inline int dccp_need_reset(int state)
247 {
248 return state != DCCP_CLOSED && state != DCCP_LISTEN &&
249 state != DCCP_REQUESTING;
250 }
251
252 int dccp_disconnect(struct sock *sk, int flags)
253 {
254 struct inet_connection_sock *icsk = inet_csk(sk);
255 struct inet_sock *inet = inet_sk(sk);
256 int err = 0;
257 const int old_state = sk->sk_state;
258
259 if (old_state != DCCP_CLOSED)
260 dccp_set_state(sk, DCCP_CLOSED);
261
262 /*
263 * This corresponds to the ABORT function of RFC793, sec. 3.8
264 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
265 */
266 if (old_state == DCCP_LISTEN) {
267 inet_csk_listen_stop(sk);
268 } else if (dccp_need_reset(old_state)) {
269 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
270 sk->sk_err = ECONNRESET;
271 } else if (old_state == DCCP_REQUESTING)
272 sk->sk_err = ECONNRESET;
273
274 dccp_clear_xmit_timers(sk);
275
276 __skb_queue_purge(&sk->sk_receive_queue);
277 __skb_queue_purge(&sk->sk_write_queue);
278 if (sk->sk_send_head != NULL) {
279 __kfree_skb(sk->sk_send_head);
280 sk->sk_send_head = NULL;
281 }
282
283 inet->inet_dport = 0;
284
285 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
286 inet_reset_saddr(sk);
287
288 sk->sk_shutdown = 0;
289 sock_reset_flag(sk, SOCK_DONE);
290
291 icsk->icsk_backoff = 0;
292 inet_csk_delack_init(sk);
293 __sk_dst_reset(sk);
294
295 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
296
297 sk->sk_error_report(sk);
298 return err;
299 }
300
301 EXPORT_SYMBOL_GPL(dccp_disconnect);
302
303 /*
304 * Wait for a DCCP event.
305 *
306 * Note that we don't need to lock the socket, as the upper poll layers
307 * take care of normal races (between the test and the event) and we don't
308 * go look at any of the socket buffers directly.
309 */
310 unsigned int dccp_poll(struct file *file, struct socket *sock,
311 poll_table *wait)
312 {
313 unsigned int mask;
314 struct sock *sk = sock->sk;
315
316 sock_poll_wait(file, sk_sleep(sk), wait);
317 if (sk->sk_state == DCCP_LISTEN)
318 return inet_csk_listen_poll(sk);
319
320 /* Socket is not locked. We are protected from async events
321 by poll logic and correct handling of state changes
322 made by another threads is impossible in any case.
323 */
324
325 mask = 0;
326 if (sk->sk_err)
327 mask = POLLERR;
328
329 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
330 mask |= POLLHUP;
331 if (sk->sk_shutdown & RCV_SHUTDOWN)
332 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
333
334 /* Connected? */
335 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
336 if (atomic_read(&sk->sk_rmem_alloc) > 0)
337 mask |= POLLIN | POLLRDNORM;
338
339 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
340 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
341 mask |= POLLOUT | POLLWRNORM;
342 } else { /* send SIGIO later */
343 set_bit(SOCK_ASYNC_NOSPACE,
344 &sk->sk_socket->flags);
345 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
346
347 /* Race breaker. If space is freed after
348 * wspace test but before the flags are set,
349 * IO signal will be lost.
350 */
351 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
352 mask |= POLLOUT | POLLWRNORM;
353 }
354 }
355 }
356 return mask;
357 }
358
359 EXPORT_SYMBOL_GPL(dccp_poll);
360
361 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
362 {
363 int rc = -ENOTCONN;
364
365 lock_sock(sk);
366
367 if (sk->sk_state == DCCP_LISTEN)
368 goto out;
369
370 switch (cmd) {
371 case SIOCINQ: {
372 struct sk_buff *skb;
373 unsigned long amount = 0;
374
375 skb = skb_peek(&sk->sk_receive_queue);
376 if (skb != NULL) {
377 /*
378 * We will only return the amount of this packet since
379 * that is all that will be read.
380 */
381 amount = skb->len;
382 }
383 rc = put_user(amount, (int __user *)arg);
384 }
385 break;
386 default:
387 rc = -ENOIOCTLCMD;
388 break;
389 }
390 out:
391 release_sock(sk);
392 return rc;
393 }
394
395 EXPORT_SYMBOL_GPL(dccp_ioctl);
396
397 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
398 char __user *optval, unsigned int optlen)
399 {
400 struct dccp_sock *dp = dccp_sk(sk);
401 struct dccp_service_list *sl = NULL;
402
403 if (service == DCCP_SERVICE_INVALID_VALUE ||
404 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
405 return -EINVAL;
406
407 if (optlen > sizeof(service)) {
408 sl = kmalloc(optlen, GFP_KERNEL);
409 if (sl == NULL)
410 return -ENOMEM;
411
412 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
413 if (copy_from_user(sl->dccpsl_list,
414 optval + sizeof(service),
415 optlen - sizeof(service)) ||
416 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
417 kfree(sl);
418 return -EFAULT;
419 }
420 }
421
422 lock_sock(sk);
423 dp->dccps_service = service;
424
425 kfree(dp->dccps_service_list);
426
427 dp->dccps_service_list = sl;
428 release_sock(sk);
429 return 0;
430 }
431
432 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
433 {
434 u8 *list, len;
435 int i, rc;
436
437 if (cscov < 0 || cscov > 15)
438 return -EINVAL;
439 /*
440 * Populate a list of permissible values, in the range cscov...15. This
441 * is necessary since feature negotiation of single values only works if
442 * both sides incidentally choose the same value. Since the list starts
443 * lowest-value first, negotiation will pick the smallest shared value.
444 */
445 if (cscov == 0)
446 return 0;
447 len = 16 - cscov;
448
449 list = kmalloc(len, GFP_KERNEL);
450 if (list == NULL)
451 return -ENOBUFS;
452
453 for (i = 0; i < len; i++)
454 list[i] = cscov++;
455
456 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
457
458 if (rc == 0) {
459 if (rx)
460 dccp_sk(sk)->dccps_pcrlen = cscov;
461 else
462 dccp_sk(sk)->dccps_pcslen = cscov;
463 }
464 kfree(list);
465 return rc;
466 }
467
468 static int dccp_setsockopt_ccid(struct sock *sk, int type,
469 char __user *optval, unsigned int optlen)
470 {
471 u8 *val;
472 int rc = 0;
473
474 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
475 return -EINVAL;
476
477 val = memdup_user(optval, optlen);
478 if (IS_ERR(val))
479 return PTR_ERR(val);
480
481 lock_sock(sk);
482 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
483 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
484
485 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
486 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
487 release_sock(sk);
488
489 kfree(val);
490 return rc;
491 }
492
493 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
494 char __user *optval, unsigned int optlen)
495 {
496 struct dccp_sock *dp = dccp_sk(sk);
497 int val, err = 0;
498
499 switch (optname) {
500 case DCCP_SOCKOPT_PACKET_SIZE:
501 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
502 return 0;
503 case DCCP_SOCKOPT_CHANGE_L:
504 case DCCP_SOCKOPT_CHANGE_R:
505 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
506 return 0;
507 case DCCP_SOCKOPT_CCID:
508 case DCCP_SOCKOPT_RX_CCID:
509 case DCCP_SOCKOPT_TX_CCID:
510 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
511 }
512
513 if (optlen < (int)sizeof(int))
514 return -EINVAL;
515
516 if (get_user(val, (int __user *)optval))
517 return -EFAULT;
518
519 if (optname == DCCP_SOCKOPT_SERVICE)
520 return dccp_setsockopt_service(sk, val, optval, optlen);
521
522 lock_sock(sk);
523 switch (optname) {
524 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
525 if (dp->dccps_role != DCCP_ROLE_SERVER)
526 err = -EOPNOTSUPP;
527 else
528 dp->dccps_server_timewait = (val != 0);
529 break;
530 case DCCP_SOCKOPT_SEND_CSCOV:
531 err = dccp_setsockopt_cscov(sk, val, false);
532 break;
533 case DCCP_SOCKOPT_RECV_CSCOV:
534 err = dccp_setsockopt_cscov(sk, val, true);
535 break;
536 case DCCP_SOCKOPT_QPOLICY_ID:
537 if (sk->sk_state != DCCP_CLOSED)
538 err = -EISCONN;
539 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
540 err = -EINVAL;
541 else
542 dp->dccps_qpolicy = val;
543 break;
544 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
545 if (val < 0)
546 err = -EINVAL;
547 else
548 dp->dccps_tx_qlen = val;
549 break;
550 default:
551 err = -ENOPROTOOPT;
552 break;
553 }
554 release_sock(sk);
555
556 return err;
557 }
558
559 int dccp_setsockopt(struct sock *sk, int level, int optname,
560 char __user *optval, unsigned int optlen)
561 {
562 if (level != SOL_DCCP)
563 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
564 optname, optval,
565 optlen);
566 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
567 }
568
569 EXPORT_SYMBOL_GPL(dccp_setsockopt);
570
571 #ifdef CONFIG_COMPAT
572 int compat_dccp_setsockopt(struct sock *sk, int level, int optname,
573 char __user *optval, unsigned int optlen)
574 {
575 if (level != SOL_DCCP)
576 return inet_csk_compat_setsockopt(sk, level, optname,
577 optval, optlen);
578 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
579 }
580
581 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt);
582 #endif
583
584 static int dccp_getsockopt_service(struct sock *sk, int len,
585 __be32 __user *optval,
586 int __user *optlen)
587 {
588 const struct dccp_sock *dp = dccp_sk(sk);
589 const struct dccp_service_list *sl;
590 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
591
592 lock_sock(sk);
593 if ((sl = dp->dccps_service_list) != NULL) {
594 slen = sl->dccpsl_nr * sizeof(u32);
595 total_len += slen;
596 }
597
598 err = -EINVAL;
599 if (total_len > len)
600 goto out;
601
602 err = 0;
603 if (put_user(total_len, optlen) ||
604 put_user(dp->dccps_service, optval) ||
605 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
606 err = -EFAULT;
607 out:
608 release_sock(sk);
609 return err;
610 }
611
612 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
613 char __user *optval, int __user *optlen)
614 {
615 struct dccp_sock *dp;
616 int val, len;
617
618 if (get_user(len, optlen))
619 return -EFAULT;
620
621 if (len < (int)sizeof(int))
622 return -EINVAL;
623
624 dp = dccp_sk(sk);
625
626 switch (optname) {
627 case DCCP_SOCKOPT_PACKET_SIZE:
628 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
629 return 0;
630 case DCCP_SOCKOPT_SERVICE:
631 return dccp_getsockopt_service(sk, len,
632 (__be32 __user *)optval, optlen);
633 case DCCP_SOCKOPT_GET_CUR_MPS:
634 val = dp->dccps_mss_cache;
635 break;
636 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
637 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
638 case DCCP_SOCKOPT_TX_CCID:
639 val = ccid_get_current_tx_ccid(dp);
640 if (val < 0)
641 return -ENOPROTOOPT;
642 break;
643 case DCCP_SOCKOPT_RX_CCID:
644 val = ccid_get_current_rx_ccid(dp);
645 if (val < 0)
646 return -ENOPROTOOPT;
647 break;
648 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
649 val = dp->dccps_server_timewait;
650 break;
651 case DCCP_SOCKOPT_SEND_CSCOV:
652 val = dp->dccps_pcslen;
653 break;
654 case DCCP_SOCKOPT_RECV_CSCOV:
655 val = dp->dccps_pcrlen;
656 break;
657 case DCCP_SOCKOPT_QPOLICY_ID:
658 val = dp->dccps_qpolicy;
659 break;
660 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
661 val = dp->dccps_tx_qlen;
662 break;
663 case 128 ... 191:
664 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
665 len, (u32 __user *)optval, optlen);
666 case 192 ... 255:
667 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
668 len, (u32 __user *)optval, optlen);
669 default:
670 return -ENOPROTOOPT;
671 }
672
673 len = sizeof(val);
674 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
675 return -EFAULT;
676
677 return 0;
678 }
679
680 int dccp_getsockopt(struct sock *sk, int level, int optname,
681 char __user *optval, int __user *optlen)
682 {
683 if (level != SOL_DCCP)
684 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
685 optname, optval,
686 optlen);
687 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
688 }
689
690 EXPORT_SYMBOL_GPL(dccp_getsockopt);
691
692 #ifdef CONFIG_COMPAT
693 int compat_dccp_getsockopt(struct sock *sk, int level, int optname,
694 char __user *optval, int __user *optlen)
695 {
696 if (level != SOL_DCCP)
697 return inet_csk_compat_getsockopt(sk, level, optname,
698 optval, optlen);
699 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
700 }
701
702 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt);
703 #endif
704
705 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
706 {
707 struct cmsghdr *cmsg = CMSG_FIRSTHDR(msg);
708
709 /*
710 * Assign an (opaque) qpolicy priority value to skb->priority.
711 *
712 * We are overloading this skb field for use with the qpolicy subystem.
713 * The skb->priority is normally used for the SO_PRIORITY option, which
714 * is initialised from sk_priority. Since the assignment of sk_priority
715 * to skb->priority happens later (on layer 3), we overload this field
716 * for use with queueing priorities as long as the skb is on layer 4.
717 * The default priority value (if nothing is set) is 0.
718 */
719 skb->priority = 0;
720
721 for (; cmsg != NULL; cmsg = CMSG_NXTHDR(msg, cmsg)) {
722
723 if (!CMSG_OK(msg, cmsg))
724 return -EINVAL;
725
726 if (cmsg->cmsg_level != SOL_DCCP)
727 continue;
728
729 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
730 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
731 return -EINVAL;
732
733 switch (cmsg->cmsg_type) {
734 case DCCP_SCM_PRIORITY:
735 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
736 return -EINVAL;
737 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
738 break;
739 default:
740 return -EINVAL;
741 }
742 }
743 return 0;
744 }
745
746 int dccp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
747 size_t len)
748 {
749 const struct dccp_sock *dp = dccp_sk(sk);
750 const int flags = msg->msg_flags;
751 const int noblock = flags & MSG_DONTWAIT;
752 struct sk_buff *skb;
753 int rc, size;
754 long timeo;
755
756 if (len > dp->dccps_mss_cache)
757 return -EMSGSIZE;
758
759 lock_sock(sk);
760
761 if (dccp_qpolicy_full(sk)) {
762 rc = -EAGAIN;
763 goto out_release;
764 }
765
766 timeo = sock_sndtimeo(sk, noblock);
767
768 /*
769 * We have to use sk_stream_wait_connect here to set sk_write_pending,
770 * so that the trick in dccp_rcv_request_sent_state_process.
771 */
772 /* Wait for a connection to finish. */
773 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
774 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
775 goto out_release;
776
777 size = sk->sk_prot->max_header + len;
778 release_sock(sk);
779 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
780 lock_sock(sk);
781 if (skb == NULL)
782 goto out_release;
783
784 skb_reserve(skb, sk->sk_prot->max_header);
785 rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
786 if (rc != 0)
787 goto out_discard;
788
789 rc = dccp_msghdr_parse(msg, skb);
790 if (rc != 0)
791 goto out_discard;
792
793 dccp_qpolicy_push(sk, skb);
794 /*
795 * The xmit_timer is set if the TX CCID is rate-based and will expire
796 * when congestion control permits to release further packets into the
797 * network. Window-based CCIDs do not use this timer.
798 */
799 if (!timer_pending(&dp->dccps_xmit_timer))
800 dccp_write_xmit(sk);
801 out_release:
802 release_sock(sk);
803 return rc ? : len;
804 out_discard:
805 kfree_skb(skb);
806 goto out_release;
807 }
808
809 EXPORT_SYMBOL_GPL(dccp_sendmsg);
810
811 int dccp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
812 size_t len, int nonblock, int flags, int *addr_len)
813 {
814 const struct dccp_hdr *dh;
815 long timeo;
816
817 lock_sock(sk);
818
819 if (sk->sk_state == DCCP_LISTEN) {
820 len = -ENOTCONN;
821 goto out;
822 }
823
824 timeo = sock_rcvtimeo(sk, nonblock);
825
826 do {
827 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
828
829 if (skb == NULL)
830 goto verify_sock_status;
831
832 dh = dccp_hdr(skb);
833
834 switch (dh->dccph_type) {
835 case DCCP_PKT_DATA:
836 case DCCP_PKT_DATAACK:
837 goto found_ok_skb;
838
839 case DCCP_PKT_CLOSE:
840 case DCCP_PKT_CLOSEREQ:
841 if (!(flags & MSG_PEEK))
842 dccp_finish_passive_close(sk);
843 /* fall through */
844 case DCCP_PKT_RESET:
845 dccp_pr_debug("found fin (%s) ok!\n",
846 dccp_packet_name(dh->dccph_type));
847 len = 0;
848 goto found_fin_ok;
849 default:
850 dccp_pr_debug("packet_type=%s\n",
851 dccp_packet_name(dh->dccph_type));
852 sk_eat_skb(sk, skb, 0);
853 }
854 verify_sock_status:
855 if (sock_flag(sk, SOCK_DONE)) {
856 len = 0;
857 break;
858 }
859
860 if (sk->sk_err) {
861 len = sock_error(sk);
862 break;
863 }
864
865 if (sk->sk_shutdown & RCV_SHUTDOWN) {
866 len = 0;
867 break;
868 }
869
870 if (sk->sk_state == DCCP_CLOSED) {
871 if (!sock_flag(sk, SOCK_DONE)) {
872 /* This occurs when user tries to read
873 * from never connected socket.
874 */
875 len = -ENOTCONN;
876 break;
877 }
878 len = 0;
879 break;
880 }
881
882 if (!timeo) {
883 len = -EAGAIN;
884 break;
885 }
886
887 if (signal_pending(current)) {
888 len = sock_intr_errno(timeo);
889 break;
890 }
891
892 sk_wait_data(sk, &timeo);
893 continue;
894 found_ok_skb:
895 if (len > skb->len)
896 len = skb->len;
897 else if (len < skb->len)
898 msg->msg_flags |= MSG_TRUNC;
899
900 if (skb_copy_datagram_iovec(skb, 0, msg->msg_iov, len)) {
901 /* Exception. Bailout! */
902 len = -EFAULT;
903 break;
904 }
905 if (flags & MSG_TRUNC)
906 len = skb->len;
907 found_fin_ok:
908 if (!(flags & MSG_PEEK))
909 sk_eat_skb(sk, skb, 0);
910 break;
911 } while (1);
912 out:
913 release_sock(sk);
914 return len;
915 }
916
917 EXPORT_SYMBOL_GPL(dccp_recvmsg);
918
919 int inet_dccp_listen(struct socket *sock, int backlog)
920 {
921 struct sock *sk = sock->sk;
922 unsigned char old_state;
923 int err;
924
925 lock_sock(sk);
926
927 err = -EINVAL;
928 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
929 goto out;
930
931 old_state = sk->sk_state;
932 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
933 goto out;
934
935 /* Really, if the socket is already in listen state
936 * we can only allow the backlog to be adjusted.
937 */
938 if (old_state != DCCP_LISTEN) {
939 /*
940 * FIXME: here it probably should be sk->sk_prot->listen_start
941 * see tcp_listen_start
942 */
943 err = dccp_listen_start(sk, backlog);
944 if (err)
945 goto out;
946 }
947 sk->sk_max_ack_backlog = backlog;
948 err = 0;
949
950 out:
951 release_sock(sk);
952 return err;
953 }
954
955 EXPORT_SYMBOL_GPL(inet_dccp_listen);
956
957 static void dccp_terminate_connection(struct sock *sk)
958 {
959 u8 next_state = DCCP_CLOSED;
960
961 switch (sk->sk_state) {
962 case DCCP_PASSIVE_CLOSE:
963 case DCCP_PASSIVE_CLOSEREQ:
964 dccp_finish_passive_close(sk);
965 break;
966 case DCCP_PARTOPEN:
967 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
968 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
969 /* fall through */
970 case DCCP_OPEN:
971 dccp_send_close(sk, 1);
972
973 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
974 !dccp_sk(sk)->dccps_server_timewait)
975 next_state = DCCP_ACTIVE_CLOSEREQ;
976 else
977 next_state = DCCP_CLOSING;
978 /* fall through */
979 default:
980 dccp_set_state(sk, next_state);
981 }
982 }
983
984 void dccp_close(struct sock *sk, long timeout)
985 {
986 struct dccp_sock *dp = dccp_sk(sk);
987 struct sk_buff *skb;
988 u32 data_was_unread = 0;
989 int state;
990
991 lock_sock(sk);
992
993 sk->sk_shutdown = SHUTDOWN_MASK;
994
995 if (sk->sk_state == DCCP_LISTEN) {
996 dccp_set_state(sk, DCCP_CLOSED);
997
998 /* Special case. */
999 inet_csk_listen_stop(sk);
1000
1001 goto adjudge_to_death;
1002 }
1003
1004 sk_stop_timer(sk, &dp->dccps_xmit_timer);
1005
1006 /*
1007 * We need to flush the recv. buffs. We do this only on the
1008 * descriptor close, not protocol-sourced closes, because the
1009 *reader process may not have drained the data yet!
1010 */
1011 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1012 data_was_unread += skb->len;
1013 __kfree_skb(skb);
1014 }
1015
1016 if (data_was_unread) {
1017 /* Unread data was tossed, send an appropriate Reset Code */
1018 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1019 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1020 dccp_set_state(sk, DCCP_CLOSED);
1021 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1022 /* Check zero linger _after_ checking for unread data. */
1023 sk->sk_prot->disconnect(sk, 0);
1024 } else if (sk->sk_state != DCCP_CLOSED) {
1025 /*
1026 * Normal connection termination. May need to wait if there are
1027 * still packets in the TX queue that are delayed by the CCID.
1028 */
1029 dccp_flush_write_queue(sk, &timeout);
1030 dccp_terminate_connection(sk);
1031 }
1032
1033 /*
1034 * Flush write queue. This may be necessary in several cases:
1035 * - we have been closed by the peer but still have application data;
1036 * - abortive termination (unread data or zero linger time),
1037 * - normal termination but queue could not be flushed within time limit
1038 */
1039 __skb_queue_purge(&sk->sk_write_queue);
1040
1041 sk_stream_wait_close(sk, timeout);
1042
1043 adjudge_to_death:
1044 state = sk->sk_state;
1045 sock_hold(sk);
1046 sock_orphan(sk);
1047
1048 /*
1049 * It is the last release_sock in its life. It will remove backlog.
1050 */
1051 release_sock(sk);
1052 /*
1053 * Now socket is owned by kernel and we acquire BH lock
1054 * to finish close. No need to check for user refs.
1055 */
1056 local_bh_disable();
1057 bh_lock_sock(sk);
1058 WARN_ON(sock_owned_by_user(sk));
1059
1060 percpu_counter_inc(sk->sk_prot->orphan_count);
1061
1062 /* Have we already been destroyed by a softirq or backlog? */
1063 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1064 goto out;
1065
1066 if (sk->sk_state == DCCP_CLOSED)
1067 inet_csk_destroy_sock(sk);
1068
1069 /* Otherwise, socket is reprieved until protocol close. */
1070
1071 out:
1072 bh_unlock_sock(sk);
1073 local_bh_enable();
1074 sock_put(sk);
1075 }
1076
1077 EXPORT_SYMBOL_GPL(dccp_close);
1078
1079 void dccp_shutdown(struct sock *sk, int how)
1080 {
1081 dccp_pr_debug("called shutdown(%x)\n", how);
1082 }
1083
1084 EXPORT_SYMBOL_GPL(dccp_shutdown);
1085
1086 static inline int dccp_mib_init(void)
1087 {
1088 return snmp_mib_init((void __percpu **)dccp_statistics,
1089 sizeof(struct dccp_mib),
1090 __alignof__(struct dccp_mib));
1091 }
1092
1093 static inline void dccp_mib_exit(void)
1094 {
1095 snmp_mib_free((void __percpu **)dccp_statistics);
1096 }
1097
1098 static int thash_entries;
1099 module_param(thash_entries, int, 0444);
1100 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1101
1102 #ifdef CONFIG_IP_DCCP_DEBUG
1103 int dccp_debug;
1104 module_param(dccp_debug, bool, 0644);
1105 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1106
1107 EXPORT_SYMBOL_GPL(dccp_debug);
1108 #endif
1109
1110 static int __init dccp_init(void)
1111 {
1112 unsigned long goal;
1113 int ehash_order, bhash_order, i;
1114 int rc;
1115
1116 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1117 FIELD_SIZEOF(struct sk_buff, cb));
1118 rc = percpu_counter_init(&dccp_orphan_count, 0);
1119 if (rc)
1120 goto out_fail;
1121 rc = -ENOBUFS;
1122 inet_hashinfo_init(&dccp_hashinfo);
1123 dccp_hashinfo.bind_bucket_cachep =
1124 kmem_cache_create("dccp_bind_bucket",
1125 sizeof(struct inet_bind_bucket), 0,
1126 SLAB_HWCACHE_ALIGN, NULL);
1127 if (!dccp_hashinfo.bind_bucket_cachep)
1128 goto out_free_percpu;
1129
1130 /*
1131 * Size and allocate the main established and bind bucket
1132 * hash tables.
1133 *
1134 * The methodology is similar to that of the buffer cache.
1135 */
1136 if (totalram_pages >= (128 * 1024))
1137 goal = totalram_pages >> (21 - PAGE_SHIFT);
1138 else
1139 goal = totalram_pages >> (23 - PAGE_SHIFT);
1140
1141 if (thash_entries)
1142 goal = (thash_entries *
1143 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1144 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1145 ;
1146 do {
1147 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1148 sizeof(struct inet_ehash_bucket);
1149
1150 while (hash_size & (hash_size - 1))
1151 hash_size--;
1152 dccp_hashinfo.ehash_mask = hash_size - 1;
1153 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1154 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1155 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1156
1157 if (!dccp_hashinfo.ehash) {
1158 DCCP_CRIT("Failed to allocate DCCP established hash table");
1159 goto out_free_bind_bucket_cachep;
1160 }
1161
1162 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++) {
1163 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1164 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].twchain, i);
1165 }
1166
1167 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1168 goto out_free_dccp_ehash;
1169
1170 bhash_order = ehash_order;
1171
1172 do {
1173 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1174 sizeof(struct inet_bind_hashbucket);
1175 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1176 bhash_order > 0)
1177 continue;
1178 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1179 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1180 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1181
1182 if (!dccp_hashinfo.bhash) {
1183 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1184 goto out_free_dccp_locks;
1185 }
1186
1187 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1188 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1189 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1190 }
1191
1192 rc = dccp_mib_init();
1193 if (rc)
1194 goto out_free_dccp_bhash;
1195
1196 rc = dccp_ackvec_init();
1197 if (rc)
1198 goto out_free_dccp_mib;
1199
1200 rc = dccp_sysctl_init();
1201 if (rc)
1202 goto out_ackvec_exit;
1203
1204 rc = ccid_initialize_builtins();
1205 if (rc)
1206 goto out_sysctl_exit;
1207
1208 dccp_timestamping_init();
1209
1210 return 0;
1211
1212 out_sysctl_exit:
1213 dccp_sysctl_exit();
1214 out_ackvec_exit:
1215 dccp_ackvec_exit();
1216 out_free_dccp_mib:
1217 dccp_mib_exit();
1218 out_free_dccp_bhash:
1219 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1220 out_free_dccp_locks:
1221 inet_ehash_locks_free(&dccp_hashinfo);
1222 out_free_dccp_ehash:
1223 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1224 out_free_bind_bucket_cachep:
1225 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1226 out_free_percpu:
1227 percpu_counter_destroy(&dccp_orphan_count);
1228 out_fail:
1229 dccp_hashinfo.bhash = NULL;
1230 dccp_hashinfo.ehash = NULL;
1231 dccp_hashinfo.bind_bucket_cachep = NULL;
1232 return rc;
1233 }
1234
1235 static void __exit dccp_fini(void)
1236 {
1237 ccid_cleanup_builtins();
1238 dccp_mib_exit();
1239 free_pages((unsigned long)dccp_hashinfo.bhash,
1240 get_order(dccp_hashinfo.bhash_size *
1241 sizeof(struct inet_bind_hashbucket)));
1242 free_pages((unsigned long)dccp_hashinfo.ehash,
1243 get_order((dccp_hashinfo.ehash_mask + 1) *
1244 sizeof(struct inet_ehash_bucket)));
1245 inet_ehash_locks_free(&dccp_hashinfo);
1246 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1247 dccp_ackvec_exit();
1248 dccp_sysctl_exit();
1249 percpu_counter_destroy(&dccp_orphan_count);
1250 }
1251
1252 module_init(dccp_init);
1253 module_exit(dccp_fini);
1254
1255 MODULE_LICENSE("GPL");
1256 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1257 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
Linus' kernel tree