ia64/linux-2.6.18-xen.hg

view net/dccp/proto.c @ 673:3161879fdf22

[IA64] xencomm: support XENMEM_add_to_physmap and XENMEM_remove_from_phsymap

support XENMEM_add_to_physmap and XENMEM_remove_from_phsymap.

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
author Isaku Yamahata <yamahata@valinux.co.jp>
date Tue Sep 16 21:26:15 2008 +0900 (2008-09-16)
parents 831230e53067
children
line source
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 */
12 #include <linux/dccp.h>
13 #include <linux/module.h>
14 #include <linux/types.h>
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/skbuff.h>
18 #include <linux/netdevice.h>
19 #include <linux/in.h>
20 #include <linux/if_arp.h>
21 #include <linux/init.h>
22 #include <linux/random.h>
23 #include <net/checksum.h>
25 #include <net/inet_sock.h>
26 #include <net/sock.h>
27 #include <net/xfrm.h>
29 #include <asm/semaphore.h>
30 #include <linux/spinlock.h>
31 #include <linux/timer.h>
32 #include <linux/delay.h>
33 #include <linux/poll.h>
35 #include "ccid.h"
36 #include "dccp.h"
37 #include "feat.h"
39 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
41 EXPORT_SYMBOL_GPL(dccp_statistics);
43 atomic_t dccp_orphan_count = ATOMIC_INIT(0);
45 EXPORT_SYMBOL_GPL(dccp_orphan_count);
47 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
48 .lhash_lock = RW_LOCK_UNLOCKED,
49 .lhash_users = ATOMIC_INIT(0),
50 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
51 };
53 EXPORT_SYMBOL_GPL(dccp_hashinfo);
55 void dccp_set_state(struct sock *sk, const int state)
56 {
57 const int oldstate = sk->sk_state;
59 dccp_pr_debug("%s(%p) %-10.10s -> %s\n",
60 dccp_role(sk), sk,
61 dccp_state_name(oldstate), dccp_state_name(state));
62 WARN_ON(state == oldstate);
64 switch (state) {
65 case DCCP_OPEN:
66 if (oldstate != DCCP_OPEN)
67 DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
68 break;
70 case DCCP_CLOSED:
71 if (oldstate == DCCP_CLOSING || oldstate == DCCP_OPEN)
72 DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
74 sk->sk_prot->unhash(sk);
75 if (inet_csk(sk)->icsk_bind_hash != NULL &&
76 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
77 inet_put_port(&dccp_hashinfo, sk);
78 /* fall through */
79 default:
80 if (oldstate == DCCP_OPEN)
81 DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
82 }
84 /* Change state AFTER socket is unhashed to avoid closed
85 * socket sitting in hash tables.
86 */
87 sk->sk_state = state;
88 }
90 EXPORT_SYMBOL_GPL(dccp_set_state);
92 void dccp_done(struct sock *sk)
93 {
94 dccp_set_state(sk, DCCP_CLOSED);
95 dccp_clear_xmit_timers(sk);
97 sk->sk_shutdown = SHUTDOWN_MASK;
99 if (!sock_flag(sk, SOCK_DEAD))
100 sk->sk_state_change(sk);
101 else
102 inet_csk_destroy_sock(sk);
103 }
105 EXPORT_SYMBOL_GPL(dccp_done);
107 const char *dccp_packet_name(const int type)
108 {
109 static const char *dccp_packet_names[] = {
110 [DCCP_PKT_REQUEST] = "REQUEST",
111 [DCCP_PKT_RESPONSE] = "RESPONSE",
112 [DCCP_PKT_DATA] = "DATA",
113 [DCCP_PKT_ACK] = "ACK",
114 [DCCP_PKT_DATAACK] = "DATAACK",
115 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
116 [DCCP_PKT_CLOSE] = "CLOSE",
117 [DCCP_PKT_RESET] = "RESET",
118 [DCCP_PKT_SYNC] = "SYNC",
119 [DCCP_PKT_SYNCACK] = "SYNCACK",
120 };
122 if (type >= DCCP_NR_PKT_TYPES)
123 return "INVALID";
124 else
125 return dccp_packet_names[type];
126 }
128 EXPORT_SYMBOL_GPL(dccp_packet_name);
130 const char *dccp_state_name(const int state)
131 {
132 static char *dccp_state_names[] = {
133 [DCCP_OPEN] = "OPEN",
134 [DCCP_REQUESTING] = "REQUESTING",
135 [DCCP_PARTOPEN] = "PARTOPEN",
136 [DCCP_LISTEN] = "LISTEN",
137 [DCCP_RESPOND] = "RESPOND",
138 [DCCP_CLOSING] = "CLOSING",
139 [DCCP_TIME_WAIT] = "TIME_WAIT",
140 [DCCP_CLOSED] = "CLOSED",
141 };
143 if (state >= DCCP_MAX_STATES)
144 return "INVALID STATE!";
145 else
146 return dccp_state_names[state];
147 }
149 EXPORT_SYMBOL_GPL(dccp_state_name);
151 void dccp_hash(struct sock *sk)
152 {
153 inet_hash(&dccp_hashinfo, sk);
154 }
156 EXPORT_SYMBOL_GPL(dccp_hash);
158 void dccp_unhash(struct sock *sk)
159 {
160 inet_unhash(&dccp_hashinfo, sk);
161 }
163 EXPORT_SYMBOL_GPL(dccp_unhash);
165 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
166 {
167 struct dccp_sock *dp = dccp_sk(sk);
168 struct dccp_minisock *dmsk = dccp_msk(sk);
169 struct inet_connection_sock *icsk = inet_csk(sk);
171 dccp_minisock_init(&dp->dccps_minisock);
172 do_gettimeofday(&dp->dccps_epoch);
174 /*
175 * FIXME: We're hardcoding the CCID, and doing this at this point makes
176 * the listening (master) sock get CCID control blocks, which is not
177 * necessary, but for now, to not mess with the test userspace apps,
178 * lets leave it here, later the real solution is to do this in a
179 * setsockopt(CCIDs-I-want/accept). -acme
180 */
181 if (likely(ctl_sock_initialized)) {
182 int rc = dccp_feat_init(dmsk);
184 if (rc)
185 return rc;
187 if (dmsk->dccpms_send_ack_vector) {
188 dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(GFP_KERNEL);
189 if (dp->dccps_hc_rx_ackvec == NULL)
190 return -ENOMEM;
191 }
192 dp->dccps_hc_rx_ccid = ccid_hc_rx_new(dmsk->dccpms_rx_ccid,
193 sk, GFP_KERNEL);
194 dp->dccps_hc_tx_ccid = ccid_hc_tx_new(dmsk->dccpms_tx_ccid,
195 sk, GFP_KERNEL);
196 if (unlikely(dp->dccps_hc_rx_ccid == NULL ||
197 dp->dccps_hc_tx_ccid == NULL)) {
198 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
199 ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
200 if (dmsk->dccpms_send_ack_vector) {
201 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
202 dp->dccps_hc_rx_ackvec = NULL;
203 }
204 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
205 return -ENOMEM;
206 }
207 } else {
208 /* control socket doesn't need feat nego */
209 INIT_LIST_HEAD(&dmsk->dccpms_pending);
210 INIT_LIST_HEAD(&dmsk->dccpms_conf);
211 }
213 dccp_init_xmit_timers(sk);
214 icsk->icsk_rto = DCCP_TIMEOUT_INIT;
215 sk->sk_state = DCCP_CLOSED;
216 sk->sk_write_space = dccp_write_space;
217 icsk->icsk_sync_mss = dccp_sync_mss;
218 dp->dccps_mss_cache = 536;
219 dp->dccps_role = DCCP_ROLE_UNDEFINED;
220 dp->dccps_service = DCCP_SERVICE_INVALID_VALUE;
221 dp->dccps_l_ack_ratio = dp->dccps_r_ack_ratio = 1;
223 return 0;
224 }
226 EXPORT_SYMBOL_GPL(dccp_init_sock);
228 int dccp_destroy_sock(struct sock *sk)
229 {
230 struct dccp_sock *dp = dccp_sk(sk);
231 struct dccp_minisock *dmsk = dccp_msk(sk);
233 /*
234 * DCCP doesn't use sk_write_queue, just sk_send_head
235 * for retransmissions
236 */
237 if (sk->sk_send_head != NULL) {
238 kfree_skb(sk->sk_send_head);
239 sk->sk_send_head = NULL;
240 }
242 /* Clean up a referenced DCCP bind bucket. */
243 if (inet_csk(sk)->icsk_bind_hash != NULL)
244 inet_put_port(&dccp_hashinfo, sk);
246 kfree(dp->dccps_service_list);
247 dp->dccps_service_list = NULL;
249 if (dmsk->dccpms_send_ack_vector) {
250 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
251 dp->dccps_hc_rx_ackvec = NULL;
252 }
253 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
254 ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
255 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
257 /* clean up feature negotiation state */
258 dccp_feat_clean(dmsk);
260 return 0;
261 }
263 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
265 static inline int dccp_listen_start(struct sock *sk)
266 {
267 struct dccp_sock *dp = dccp_sk(sk);
269 dp->dccps_role = DCCP_ROLE_LISTEN;
270 /*
271 * Apps need to use setsockopt(DCCP_SOCKOPT_SERVICE)
272 * before calling listen()
273 */
274 if (dccp_service_not_initialized(sk))
275 return -EPROTO;
276 return inet_csk_listen_start(sk, TCP_SYNQ_HSIZE);
277 }
279 int dccp_disconnect(struct sock *sk, int flags)
280 {
281 struct inet_connection_sock *icsk = inet_csk(sk);
282 struct inet_sock *inet = inet_sk(sk);
283 int err = 0;
284 const int old_state = sk->sk_state;
286 if (old_state != DCCP_CLOSED)
287 dccp_set_state(sk, DCCP_CLOSED);
289 /* ABORT function of RFC793 */
290 if (old_state == DCCP_LISTEN) {
291 inet_csk_listen_stop(sk);
292 /* FIXME: do the active reset thing */
293 } else if (old_state == DCCP_REQUESTING)
294 sk->sk_err = ECONNRESET;
296 dccp_clear_xmit_timers(sk);
297 __skb_queue_purge(&sk->sk_receive_queue);
298 if (sk->sk_send_head != NULL) {
299 __kfree_skb(sk->sk_send_head);
300 sk->sk_send_head = NULL;
301 }
303 inet->dport = 0;
305 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
306 inet_reset_saddr(sk);
308 sk->sk_shutdown = 0;
309 sock_reset_flag(sk, SOCK_DONE);
311 icsk->icsk_backoff = 0;
312 inet_csk_delack_init(sk);
313 __sk_dst_reset(sk);
315 BUG_TRAP(!inet->num || icsk->icsk_bind_hash);
317 sk->sk_error_report(sk);
318 return err;
319 }
321 EXPORT_SYMBOL_GPL(dccp_disconnect);
323 /*
324 * Wait for a DCCP event.
325 *
326 * Note that we don't need to lock the socket, as the upper poll layers
327 * take care of normal races (between the test and the event) and we don't
328 * go look at any of the socket buffers directly.
329 */
330 unsigned int dccp_poll(struct file *file, struct socket *sock,
331 poll_table *wait)
332 {
333 unsigned int mask;
334 struct sock *sk = sock->sk;
336 poll_wait(file, sk->sk_sleep, wait);
337 if (sk->sk_state == DCCP_LISTEN)
338 return inet_csk_listen_poll(sk);
340 /* Socket is not locked. We are protected from async events
341 by poll logic and correct handling of state changes
342 made by another threads is impossible in any case.
343 */
345 mask = 0;
346 if (sk->sk_err)
347 mask = POLLERR;
349 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
350 mask |= POLLHUP;
351 if (sk->sk_shutdown & RCV_SHUTDOWN)
352 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
354 /* Connected? */
355 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
356 if (atomic_read(&sk->sk_rmem_alloc) > 0)
357 mask |= POLLIN | POLLRDNORM;
359 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
360 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
361 mask |= POLLOUT | POLLWRNORM;
362 } else { /* send SIGIO later */
363 set_bit(SOCK_ASYNC_NOSPACE,
364 &sk->sk_socket->flags);
365 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
367 /* Race breaker. If space is freed after
368 * wspace test but before the flags are set,
369 * IO signal will be lost.
370 */
371 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
372 mask |= POLLOUT | POLLWRNORM;
373 }
374 }
375 }
376 return mask;
377 }
379 EXPORT_SYMBOL_GPL(dccp_poll);
381 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
382 {
383 dccp_pr_debug("entry\n");
384 return -ENOIOCTLCMD;
385 }
387 EXPORT_SYMBOL_GPL(dccp_ioctl);
389 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
390 char __user *optval, int optlen)
391 {
392 struct dccp_sock *dp = dccp_sk(sk);
393 struct dccp_service_list *sl = NULL;
395 if (service == DCCP_SERVICE_INVALID_VALUE ||
396 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
397 return -EINVAL;
399 if (optlen > sizeof(service)) {
400 sl = kmalloc(optlen, GFP_KERNEL);
401 if (sl == NULL)
402 return -ENOMEM;
404 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
405 if (copy_from_user(sl->dccpsl_list,
406 optval + sizeof(service),
407 optlen - sizeof(service)) ||
408 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
409 kfree(sl);
410 return -EFAULT;
411 }
412 }
414 lock_sock(sk);
415 dp->dccps_service = service;
417 kfree(dp->dccps_service_list);
419 dp->dccps_service_list = sl;
420 release_sock(sk);
421 return 0;
422 }
424 /* byte 1 is feature. the rest is the preference list */
425 static int dccp_setsockopt_change(struct sock *sk, int type,
426 struct dccp_so_feat __user *optval)
427 {
428 struct dccp_so_feat opt;
429 u8 *val;
430 int rc;
432 if (copy_from_user(&opt, optval, sizeof(opt)))
433 return -EFAULT;
435 val = kmalloc(opt.dccpsf_len, GFP_KERNEL);
436 if (!val)
437 return -ENOMEM;
439 if (copy_from_user(val, opt.dccpsf_val, opt.dccpsf_len)) {
440 rc = -EFAULT;
441 goto out_free_val;
442 }
444 rc = dccp_feat_change(dccp_msk(sk), type, opt.dccpsf_feat,
445 val, opt.dccpsf_len, GFP_KERNEL);
446 if (rc)
447 goto out_free_val;
449 out:
450 return rc;
452 out_free_val:
453 kfree(val);
454 goto out;
455 }
457 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
458 char __user *optval, int optlen)
459 {
460 struct dccp_sock *dp;
461 int err;
462 int val;
464 if (optlen < sizeof(int))
465 return -EINVAL;
467 if (get_user(val, (int __user *)optval))
468 return -EFAULT;
470 if (optname == DCCP_SOCKOPT_SERVICE)
471 return dccp_setsockopt_service(sk, val, optval, optlen);
473 lock_sock(sk);
474 dp = dccp_sk(sk);
475 err = 0;
477 switch (optname) {
478 case DCCP_SOCKOPT_PACKET_SIZE:
479 dp->dccps_packet_size = val;
480 break;
482 case DCCP_SOCKOPT_CHANGE_L:
483 if (optlen != sizeof(struct dccp_so_feat))
484 err = -EINVAL;
485 else
486 err = dccp_setsockopt_change(sk, DCCPO_CHANGE_L,
487 (struct dccp_so_feat __user *)
488 optval);
489 break;
491 case DCCP_SOCKOPT_CHANGE_R:
492 if (optlen != sizeof(struct dccp_so_feat))
493 err = -EINVAL;
494 else
495 err = dccp_setsockopt_change(sk, DCCPO_CHANGE_R,
496 (struct dccp_so_feat __user *)
497 optval);
498 break;
500 default:
501 err = -ENOPROTOOPT;
502 break;
503 }
505 release_sock(sk);
506 return err;
507 }
509 int dccp_setsockopt(struct sock *sk, int level, int optname,
510 char __user *optval, int optlen)
511 {
512 if (level != SOL_DCCP)
513 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
514 optname, optval,
515 optlen);
516 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
517 }
519 EXPORT_SYMBOL_GPL(dccp_setsockopt);
521 #ifdef CONFIG_COMPAT
522 int compat_dccp_setsockopt(struct sock *sk, int level, int optname,
523 char __user *optval, int optlen)
524 {
525 if (level != SOL_DCCP)
526 return inet_csk_compat_setsockopt(sk, level, optname,
527 optval, optlen);
528 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
529 }
531 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt);
532 #endif
534 static int dccp_getsockopt_service(struct sock *sk, int len,
535 __be32 __user *optval,
536 int __user *optlen)
537 {
538 const struct dccp_sock *dp = dccp_sk(sk);
539 const struct dccp_service_list *sl;
540 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
542 lock_sock(sk);
543 if (dccp_service_not_initialized(sk))
544 goto out;
546 if ((sl = dp->dccps_service_list) != NULL) {
547 slen = sl->dccpsl_nr * sizeof(u32);
548 total_len += slen;
549 }
551 err = -EINVAL;
552 if (total_len > len)
553 goto out;
555 err = 0;
556 if (put_user(total_len, optlen) ||
557 put_user(dp->dccps_service, optval) ||
558 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
559 err = -EFAULT;
560 out:
561 release_sock(sk);
562 return err;
563 }
565 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
566 char __user *optval, int __user *optlen)
567 {
568 struct dccp_sock *dp;
569 int val, len;
571 if (get_user(len, optlen))
572 return -EFAULT;
574 if (len < sizeof(int))
575 return -EINVAL;
577 dp = dccp_sk(sk);
579 switch (optname) {
580 case DCCP_SOCKOPT_PACKET_SIZE:
581 val = dp->dccps_packet_size;
582 len = sizeof(dp->dccps_packet_size);
583 break;
584 case DCCP_SOCKOPT_SERVICE:
585 return dccp_getsockopt_service(sk, len,
586 (__be32 __user *)optval, optlen);
587 case 128 ... 191:
588 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
589 len, (u32 __user *)optval, optlen);
590 case 192 ... 255:
591 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
592 len, (u32 __user *)optval, optlen);
593 default:
594 return -ENOPROTOOPT;
595 }
597 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
598 return -EFAULT;
600 return 0;
601 }
603 int dccp_getsockopt(struct sock *sk, int level, int optname,
604 char __user *optval, int __user *optlen)
605 {
606 if (level != SOL_DCCP)
607 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
608 optname, optval,
609 optlen);
610 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
611 }
613 EXPORT_SYMBOL_GPL(dccp_getsockopt);
615 #ifdef CONFIG_COMPAT
616 int compat_dccp_getsockopt(struct sock *sk, int level, int optname,
617 char __user *optval, int __user *optlen)
618 {
619 if (level != SOL_DCCP)
620 return inet_csk_compat_getsockopt(sk, level, optname,
621 optval, optlen);
622 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
623 }
625 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt);
626 #endif
628 int dccp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
629 size_t len)
630 {
631 const struct dccp_sock *dp = dccp_sk(sk);
632 const int flags = msg->msg_flags;
633 const int noblock = flags & MSG_DONTWAIT;
634 struct sk_buff *skb;
635 int rc, size;
636 long timeo;
638 if (len > dp->dccps_mss_cache)
639 return -EMSGSIZE;
641 lock_sock(sk);
642 timeo = sock_sndtimeo(sk, noblock);
644 /*
645 * We have to use sk_stream_wait_connect here to set sk_write_pending,
646 * so that the trick in dccp_rcv_request_sent_state_process.
647 */
648 /* Wait for a connection to finish. */
649 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN | DCCPF_CLOSING))
650 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
651 goto out_release;
653 size = sk->sk_prot->max_header + len;
654 release_sock(sk);
655 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
656 lock_sock(sk);
657 if (skb == NULL)
658 goto out_release;
660 skb_reserve(skb, sk->sk_prot->max_header);
661 rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
662 if (rc != 0)
663 goto out_discard;
665 rc = dccp_write_xmit(sk, skb, &timeo);
666 /*
667 * XXX we don't use sk_write_queue, so just discard the packet.
668 * Current plan however is to _use_ sk_write_queue with
669 * an algorith similar to tcp_sendmsg, where the main difference
670 * is that in DCCP we have to respect packet boundaries, so
671 * no coalescing of skbs.
672 *
673 * This bug was _quickly_ found & fixed by just looking at an OSTRA
674 * generated callgraph 8) -acme
675 */
676 out_release:
677 release_sock(sk);
678 return rc ? : len;
679 out_discard:
680 kfree_skb(skb);
681 goto out_release;
682 }
684 EXPORT_SYMBOL_GPL(dccp_sendmsg);
686 int dccp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
687 size_t len, int nonblock, int flags, int *addr_len)
688 {
689 const struct dccp_hdr *dh;
690 long timeo;
692 lock_sock(sk);
694 if (sk->sk_state == DCCP_LISTEN) {
695 len = -ENOTCONN;
696 goto out;
697 }
699 timeo = sock_rcvtimeo(sk, nonblock);
701 do {
702 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
704 if (skb == NULL)
705 goto verify_sock_status;
707 dh = dccp_hdr(skb);
709 if (dh->dccph_type == DCCP_PKT_DATA ||
710 dh->dccph_type == DCCP_PKT_DATAACK)
711 goto found_ok_skb;
713 if (dh->dccph_type == DCCP_PKT_RESET ||
714 dh->dccph_type == DCCP_PKT_CLOSE) {
715 dccp_pr_debug("found fin ok!\n");
716 len = 0;
717 goto found_fin_ok;
718 }
719 dccp_pr_debug("packet_type=%s\n",
720 dccp_packet_name(dh->dccph_type));
721 sk_eat_skb(sk, skb, 0);
722 verify_sock_status:
723 if (sock_flag(sk, SOCK_DONE)) {
724 len = 0;
725 break;
726 }
728 if (sk->sk_err) {
729 len = sock_error(sk);
730 break;
731 }
733 if (sk->sk_shutdown & RCV_SHUTDOWN) {
734 len = 0;
735 break;
736 }
738 if (sk->sk_state == DCCP_CLOSED) {
739 if (!sock_flag(sk, SOCK_DONE)) {
740 /* This occurs when user tries to read
741 * from never connected socket.
742 */
743 len = -ENOTCONN;
744 break;
745 }
746 len = 0;
747 break;
748 }
750 if (!timeo) {
751 len = -EAGAIN;
752 break;
753 }
755 if (signal_pending(current)) {
756 len = sock_intr_errno(timeo);
757 break;
758 }
760 sk_wait_data(sk, &timeo);
761 continue;
762 found_ok_skb:
763 if (len > skb->len)
764 len = skb->len;
765 else if (len < skb->len)
766 msg->msg_flags |= MSG_TRUNC;
768 if (skb_copy_datagram_iovec(skb, 0, msg->msg_iov, len)) {
769 /* Exception. Bailout! */
770 len = -EFAULT;
771 break;
772 }
773 found_fin_ok:
774 if (!(flags & MSG_PEEK))
775 sk_eat_skb(sk, skb, 0);
776 break;
777 } while (1);
778 out:
779 release_sock(sk);
780 return len;
781 }
783 EXPORT_SYMBOL_GPL(dccp_recvmsg);
785 int inet_dccp_listen(struct socket *sock, int backlog)
786 {
787 struct sock *sk = sock->sk;
788 unsigned char old_state;
789 int err;
791 lock_sock(sk);
793 err = -EINVAL;
794 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
795 goto out;
797 old_state = sk->sk_state;
798 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
799 goto out;
801 /* Really, if the socket is already in listen state
802 * we can only allow the backlog to be adjusted.
803 */
804 if (old_state != DCCP_LISTEN) {
805 /*
806 * FIXME: here it probably should be sk->sk_prot->listen_start
807 * see tcp_listen_start
808 */
809 err = dccp_listen_start(sk);
810 if (err)
811 goto out;
812 }
813 sk->sk_max_ack_backlog = backlog;
814 err = 0;
816 out:
817 release_sock(sk);
818 return err;
819 }
821 EXPORT_SYMBOL_GPL(inet_dccp_listen);
823 static const unsigned char dccp_new_state[] = {
824 /* current state: new state: action: */
825 [0] = DCCP_CLOSED,
826 [DCCP_OPEN] = DCCP_CLOSING | DCCP_ACTION_FIN,
827 [DCCP_REQUESTING] = DCCP_CLOSED,
828 [DCCP_PARTOPEN] = DCCP_CLOSING | DCCP_ACTION_FIN,
829 [DCCP_LISTEN] = DCCP_CLOSED,
830 [DCCP_RESPOND] = DCCP_CLOSED,
831 [DCCP_CLOSING] = DCCP_CLOSED,
832 [DCCP_TIME_WAIT] = DCCP_CLOSED,
833 [DCCP_CLOSED] = DCCP_CLOSED,
834 };
836 static int dccp_close_state(struct sock *sk)
837 {
838 const int next = dccp_new_state[sk->sk_state];
839 const int ns = next & DCCP_STATE_MASK;
841 if (ns != sk->sk_state)
842 dccp_set_state(sk, ns);
844 return next & DCCP_ACTION_FIN;
845 }
847 void dccp_close(struct sock *sk, long timeout)
848 {
849 struct sk_buff *skb;
850 int state;
852 lock_sock(sk);
854 sk->sk_shutdown = SHUTDOWN_MASK;
856 if (sk->sk_state == DCCP_LISTEN) {
857 dccp_set_state(sk, DCCP_CLOSED);
859 /* Special case. */
860 inet_csk_listen_stop(sk);
862 goto adjudge_to_death;
863 }
865 /*
866 * We need to flush the recv. buffs. We do this only on the
867 * descriptor close, not protocol-sourced closes, because the
868 *reader process may not have drained the data yet!
869 */
870 /* FIXME: check for unread data */
871 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
872 __kfree_skb(skb);
873 }
875 if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
876 /* Check zero linger _after_ checking for unread data. */
877 sk->sk_prot->disconnect(sk, 0);
878 } else if (dccp_close_state(sk)) {
879 dccp_send_close(sk, 1);
880 }
882 sk_stream_wait_close(sk, timeout);
884 adjudge_to_death:
885 state = sk->sk_state;
886 sock_hold(sk);
887 sock_orphan(sk);
888 atomic_inc(sk->sk_prot->orphan_count);
890 /*
891 * It is the last release_sock in its life. It will remove backlog.
892 */
893 release_sock(sk);
894 /*
895 * Now socket is owned by kernel and we acquire BH lock
896 * to finish close. No need to check for user refs.
897 */
898 local_bh_disable();
899 bh_lock_sock(sk);
900 BUG_TRAP(!sock_owned_by_user(sk));
902 /* Have we already been destroyed by a softirq or backlog? */
903 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
904 goto out;
906 /*
907 * The last release_sock may have processed the CLOSE or RESET
908 * packet moving sock to CLOSED state, if not we have to fire
909 * the CLOSE/CLOSEREQ retransmission timer, see "8.3. Termination"
910 * in draft-ietf-dccp-spec-11. -acme
911 */
912 if (sk->sk_state == DCCP_CLOSING) {
913 /* FIXME: should start at 2 * RTT */
914 /* Timer for repeating the CLOSE/CLOSEREQ until an answer. */
915 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
916 inet_csk(sk)->icsk_rto,
917 DCCP_RTO_MAX);
918 #if 0
919 /* Yeah, we should use sk->sk_prot->orphan_count, etc */
920 dccp_set_state(sk, DCCP_CLOSED);
921 #endif
922 }
924 if (sk->sk_state == DCCP_CLOSED)
925 inet_csk_destroy_sock(sk);
927 /* Otherwise, socket is reprieved until protocol close. */
929 out:
930 bh_unlock_sock(sk);
931 local_bh_enable();
932 sock_put(sk);
933 }
935 EXPORT_SYMBOL_GPL(dccp_close);
937 void dccp_shutdown(struct sock *sk, int how)
938 {
939 dccp_pr_debug("entry\n");
940 }
942 EXPORT_SYMBOL_GPL(dccp_shutdown);
944 static int __init dccp_mib_init(void)
945 {
946 int rc = -ENOMEM;
948 dccp_statistics[0] = alloc_percpu(struct dccp_mib);
949 if (dccp_statistics[0] == NULL)
950 goto out;
952 dccp_statistics[1] = alloc_percpu(struct dccp_mib);
953 if (dccp_statistics[1] == NULL)
954 goto out_free_one;
956 rc = 0;
957 out:
958 return rc;
959 out_free_one:
960 free_percpu(dccp_statistics[0]);
961 dccp_statistics[0] = NULL;
962 goto out;
964 }
966 static void dccp_mib_exit(void)
967 {
968 free_percpu(dccp_statistics[0]);
969 free_percpu(dccp_statistics[1]);
970 dccp_statistics[0] = dccp_statistics[1] = NULL;
971 }
973 static int thash_entries;
974 module_param(thash_entries, int, 0444);
975 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
977 #ifdef CONFIG_IP_DCCP_DEBUG
978 int dccp_debug;
979 module_param(dccp_debug, int, 0444);
980 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
982 EXPORT_SYMBOL_GPL(dccp_debug);
983 #endif
985 static int __init dccp_init(void)
986 {
987 unsigned long goal;
988 int ehash_order, bhash_order, i;
989 int rc = -ENOBUFS;
991 dccp_hashinfo.bind_bucket_cachep =
992 kmem_cache_create("dccp_bind_bucket",
993 sizeof(struct inet_bind_bucket), 0,
994 SLAB_HWCACHE_ALIGN, NULL, NULL);
995 if (!dccp_hashinfo.bind_bucket_cachep)
996 goto out;
998 /*
999 * Size and allocate the main established and bind bucket
1000 * hash tables.
1002 * The methodology is similar to that of the buffer cache.
1003 */
1004 if (num_physpages >= (128 * 1024))
1005 goal = num_physpages >> (21 - PAGE_SHIFT);
1006 else
1007 goal = num_physpages >> (23 - PAGE_SHIFT);
1009 if (thash_entries)
1010 goal = (thash_entries *
1011 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1012 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1014 do {
1015 dccp_hashinfo.ehash_size = (1UL << ehash_order) * PAGE_SIZE /
1016 sizeof(struct inet_ehash_bucket);
1017 dccp_hashinfo.ehash_size >>= 1;
1018 while (dccp_hashinfo.ehash_size &
1019 (dccp_hashinfo.ehash_size - 1))
1020 dccp_hashinfo.ehash_size--;
1021 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1022 __get_free_pages(GFP_ATOMIC, ehash_order);
1023 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1025 if (!dccp_hashinfo.ehash) {
1026 printk(KERN_CRIT "Failed to allocate DCCP "
1027 "established hash table\n");
1028 goto out_free_bind_bucket_cachep;
1031 for (i = 0; i < (dccp_hashinfo.ehash_size << 1); i++) {
1032 rwlock_init(&dccp_hashinfo.ehash[i].lock);
1033 INIT_HLIST_HEAD(&dccp_hashinfo.ehash[i].chain);
1036 bhash_order = ehash_order;
1038 do {
1039 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1040 sizeof(struct inet_bind_hashbucket);
1041 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1042 bhash_order > 0)
1043 continue;
1044 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1045 __get_free_pages(GFP_ATOMIC, bhash_order);
1046 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1048 if (!dccp_hashinfo.bhash) {
1049 printk(KERN_CRIT "Failed to allocate DCCP bind hash table\n");
1050 goto out_free_dccp_ehash;
1053 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1054 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1055 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1058 rc = dccp_mib_init();
1059 if (rc)
1060 goto out_free_dccp_bhash;
1062 rc = dccp_ackvec_init();
1063 if (rc)
1064 goto out_free_dccp_mib;
1066 rc = dccp_sysctl_init();
1067 if (rc)
1068 goto out_ackvec_exit;
1069 out:
1070 return rc;
1071 out_ackvec_exit:
1072 dccp_ackvec_exit();
1073 out_free_dccp_mib:
1074 dccp_mib_exit();
1075 out_free_dccp_bhash:
1076 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1077 dccp_hashinfo.bhash = NULL;
1078 out_free_dccp_ehash:
1079 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1080 dccp_hashinfo.ehash = NULL;
1081 out_free_bind_bucket_cachep:
1082 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1083 dccp_hashinfo.bind_bucket_cachep = NULL;
1084 goto out;
1087 static void __exit dccp_fini(void)
1089 dccp_mib_exit();
1090 free_pages((unsigned long)dccp_hashinfo.bhash,
1091 get_order(dccp_hashinfo.bhash_size *
1092 sizeof(struct inet_bind_hashbucket)));
1093 free_pages((unsigned long)dccp_hashinfo.ehash,
1094 get_order(dccp_hashinfo.ehash_size *
1095 sizeof(struct inet_ehash_bucket)));
1096 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1097 dccp_ackvec_exit();
1098 dccp_sysctl_exit();
1101 module_init(dccp_init);
1102 module_exit(dccp_fini);
1104 MODULE_LICENSE("GPL");
1105 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1106 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");