4 * An implementation of the DCCP protocol
5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
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.
12 #include <linux/config.h>
13 #include <linux/dccp.h>
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <linux/skbuff.h>
19 #include <linux/netdevice.h>
21 #include <linux/if_arp.h>
22 #include <linux/init.h>
23 #include <linux/random.h>
24 #include <net/checksum.h>
26 #include <net/inet_sock.h>
30 #include <asm/semaphore.h>
31 #include <linux/spinlock.h>
32 #include <linux/timer.h>
33 #include <linux/delay.h>
34 #include <linux/poll.h>
40 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
42 EXPORT_SYMBOL_GPL(dccp_statistics);
44 atomic_t dccp_orphan_count = ATOMIC_INIT(0);
46 EXPORT_SYMBOL_GPL(dccp_orphan_count);
48 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
49 .lhash_lock = RW_LOCK_UNLOCKED,
50 .lhash_users = ATOMIC_INIT(0),
51 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
54 EXPORT_SYMBOL_GPL(dccp_hashinfo);
56 void dccp_set_state(struct sock *sk, const int state)
58 const int oldstate = sk->sk_state;
60 dccp_pr_debug("%s(%p) %-10.10s -> %s\n",
62 dccp_state_name(oldstate), dccp_state_name(state));
63 WARN_ON(state == oldstate);
67 if (oldstate != DCCP_OPEN)
68 DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
72 if (oldstate == DCCP_CLOSING || oldstate == DCCP_OPEN)
73 DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
75 sk->sk_prot->unhash(sk);
76 if (inet_csk(sk)->icsk_bind_hash != NULL &&
77 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
78 inet_put_port(&dccp_hashinfo, sk);
81 if (oldstate == DCCP_OPEN)
82 DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
85 /* Change state AFTER socket is unhashed to avoid closed
86 * socket sitting in hash tables.
91 EXPORT_SYMBOL_GPL(dccp_set_state);
93 void dccp_done(struct sock *sk)
95 dccp_set_state(sk, DCCP_CLOSED);
96 dccp_clear_xmit_timers(sk);
98 sk->sk_shutdown = SHUTDOWN_MASK;
100 if (!sock_flag(sk, SOCK_DEAD))
101 sk->sk_state_change(sk);
103 inet_csk_destroy_sock(sk);
106 EXPORT_SYMBOL_GPL(dccp_done);
108 const char *dccp_packet_name(const int type)
110 static const char *dccp_packet_names[] = {
111 [DCCP_PKT_REQUEST] = "REQUEST",
112 [DCCP_PKT_RESPONSE] = "RESPONSE",
113 [DCCP_PKT_DATA] = "DATA",
114 [DCCP_PKT_ACK] = "ACK",
115 [DCCP_PKT_DATAACK] = "DATAACK",
116 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
117 [DCCP_PKT_CLOSE] = "CLOSE",
118 [DCCP_PKT_RESET] = "RESET",
119 [DCCP_PKT_SYNC] = "SYNC",
120 [DCCP_PKT_SYNCACK] = "SYNCACK",
123 if (type >= DCCP_NR_PKT_TYPES)
126 return dccp_packet_names[type];
129 EXPORT_SYMBOL_GPL(dccp_packet_name);
131 const char *dccp_state_name(const int state)
133 static char *dccp_state_names[] = {
134 [DCCP_OPEN] = "OPEN",
135 [DCCP_REQUESTING] = "REQUESTING",
136 [DCCP_PARTOPEN] = "PARTOPEN",
137 [DCCP_LISTEN] = "LISTEN",
138 [DCCP_RESPOND] = "RESPOND",
139 [DCCP_CLOSING] = "CLOSING",
140 [DCCP_TIME_WAIT] = "TIME_WAIT",
141 [DCCP_CLOSED] = "CLOSED",
144 if (state >= DCCP_MAX_STATES)
145 return "INVALID STATE!";
147 return dccp_state_names[state];
150 EXPORT_SYMBOL_GPL(dccp_state_name);
152 void dccp_hash(struct sock *sk)
154 inet_hash(&dccp_hashinfo, sk);
157 EXPORT_SYMBOL_GPL(dccp_hash);
159 void dccp_unhash(struct sock *sk)
161 inet_unhash(&dccp_hashinfo, sk);
164 EXPORT_SYMBOL_GPL(dccp_unhash);
166 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
168 struct dccp_sock *dp = dccp_sk(sk);
169 struct dccp_minisock *dmsk = dccp_msk(sk);
170 struct inet_connection_sock *icsk = inet_csk(sk);
172 dccp_minisock_init(&dp->dccps_minisock);
173 do_gettimeofday(&dp->dccps_epoch);
176 * FIXME: We're hardcoding the CCID, and doing this at this point makes
177 * the listening (master) sock get CCID control blocks, which is not
178 * necessary, but for now, to not mess with the test userspace apps,
179 * lets leave it here, later the real solution is to do this in a
180 * setsockopt(CCIDs-I-want/accept). -acme
182 if (likely(ctl_sock_initialized)) {
183 int rc = dccp_feat_init(sk);
188 if (dmsk->dccpms_send_ack_vector) {
189 dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(GFP_KERNEL);
190 if (dp->dccps_hc_rx_ackvec == NULL)
193 dp->dccps_hc_rx_ccid = ccid_hc_rx_new(dmsk->dccpms_rx_ccid,
195 dp->dccps_hc_tx_ccid = ccid_hc_tx_new(dmsk->dccpms_tx_ccid,
197 if (unlikely(dp->dccps_hc_rx_ccid == NULL ||
198 dp->dccps_hc_tx_ccid == NULL)) {
199 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
200 ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
201 if (dmsk->dccpms_send_ack_vector) {
202 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
203 dp->dccps_hc_rx_ackvec = NULL;
205 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
209 /* control socket doesn't need feat nego */
210 INIT_LIST_HEAD(&dmsk->dccpms_pending);
211 INIT_LIST_HEAD(&dmsk->dccpms_conf);
214 dccp_init_xmit_timers(sk);
215 icsk->icsk_rto = DCCP_TIMEOUT_INIT;
216 sk->sk_state = DCCP_CLOSED;
217 sk->sk_write_space = dccp_write_space;
218 icsk->icsk_sync_mss = dccp_sync_mss;
219 dp->dccps_mss_cache = 536;
220 dp->dccps_role = DCCP_ROLE_UNDEFINED;
221 dp->dccps_service = DCCP_SERVICE_INVALID_VALUE;
222 dp->dccps_l_ack_ratio = dp->dccps_r_ack_ratio = 1;
227 EXPORT_SYMBOL_GPL(dccp_init_sock);
229 int dccp_destroy_sock(struct sock *sk)
231 struct dccp_sock *dp = dccp_sk(sk);
234 * DCCP doesn't use sk_write_queue, just sk_send_head
235 * for retransmissions
237 if (sk->sk_send_head != NULL) {
238 kfree_skb(sk->sk_send_head);
239 sk->sk_send_head = NULL;
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 (dccp_msk(sk)->dccpms_send_ack_vector) {
250 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
251 dp->dccps_hc_rx_ackvec = NULL;
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 */
263 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
265 static inline int dccp_listen_start(struct sock *sk)
267 struct dccp_sock *dp = dccp_sk(sk);
269 dp->dccps_role = DCCP_ROLE_LISTEN;
271 * Apps need to use setsockopt(DCCP_SOCKOPT_SERVICE)
272 * before calling listen()
274 if (dccp_service_not_initialized(sk))
276 return inet_csk_listen_start(sk, TCP_SYNQ_HSIZE);
279 int dccp_disconnect(struct sock *sk, int flags)
281 struct inet_connection_sock *icsk = inet_csk(sk);
282 struct inet_sock *inet = inet_sk(sk);
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;
305 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
306 inet_reset_saddr(sk);
309 sock_reset_flag(sk, SOCK_DONE);
311 icsk->icsk_backoff = 0;
312 inet_csk_delack_init(sk);
315 BUG_TRAP(!inet->num || icsk->icsk_bind_hash);
317 sk->sk_error_report(sk);
321 EXPORT_SYMBOL_GPL(dccp_disconnect);
324 * Wait for a DCCP event.
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.
330 unsigned int dccp_poll(struct file *file, struct socket *sock,
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.
349 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
351 if (sk->sk_shutdown & RCV_SHUTDOWN)
352 mask |= POLLIN | POLLRDNORM;
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.
371 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
372 mask |= POLLOUT | POLLWRNORM;
379 EXPORT_SYMBOL_GPL(dccp_poll);
381 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
383 dccp_pr_debug("entry\n");
387 EXPORT_SYMBOL_GPL(dccp_ioctl);
389 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
390 char __user *optval, int optlen)
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))
399 if (optlen > sizeof(service)) {
400 sl = kmalloc(optlen, GFP_KERNEL);
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)) {
415 dp->dccps_service = service;
417 kfree(dp->dccps_service_list);
419 dp->dccps_service_list = sl;
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)
428 struct dccp_so_feat opt;
432 if (copy_from_user(&opt, optval, sizeof(opt)))
435 val = kmalloc(opt.dccpsf_len, GFP_KERNEL);
439 if (copy_from_user(val, opt.dccpsf_val, opt.dccpsf_len)) {
444 rc = dccp_feat_change(sk, type, opt.dccpsf_feat, val, opt.dccpsf_len,
457 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
458 char __user *optval, int optlen)
460 struct dccp_sock *dp;
464 if (optlen < sizeof(int))
467 if (get_user(val, (int __user *)optval))
470 if (optname == DCCP_SOCKOPT_SERVICE)
471 return dccp_setsockopt_service(sk, val, optval, optlen);
478 case DCCP_SOCKOPT_PACKET_SIZE:
479 dp->dccps_packet_size = val;
482 case DCCP_SOCKOPT_CHANGE_L:
483 if (optlen != sizeof(struct dccp_so_feat))
486 err = dccp_setsockopt_change(sk, DCCPO_CHANGE_L,
487 (struct dccp_so_feat *)
491 case DCCP_SOCKOPT_CHANGE_R:
492 if (optlen != sizeof(struct dccp_so_feat))
495 err = dccp_setsockopt_change(sk, DCCPO_CHANGE_R,
496 (struct dccp_so_feat *)
509 int dccp_setsockopt(struct sock *sk, int level, int optname,
510 char __user *optval, int optlen)
512 if (level != SOL_DCCP)
513 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
516 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
519 EXPORT_SYMBOL_GPL(dccp_setsockopt);
522 int compat_dccp_setsockopt(struct sock *sk, int level, int optname,
523 char __user *optval, int optlen)
525 if (level != SOL_DCCP)
526 return inet_csk_compat_setsockopt(sk, level, optname,
528 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
531 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt);
534 static int dccp_getsockopt_service(struct sock *sk, int len,
535 __be32 __user *optval,
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);
543 if (dccp_service_not_initialized(sk))
546 if ((sl = dp->dccps_service_list) != NULL) {
547 slen = sl->dccpsl_nr * sizeof(u32);
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)))
565 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
566 char __user *optval, int __user *optlen)
568 struct dccp_sock *dp;
571 if (get_user(len, optlen))
574 if (len < sizeof(int))
580 case DCCP_SOCKOPT_PACKET_SIZE:
581 val = dp->dccps_packet_size;
582 len = sizeof(dp->dccps_packet_size);
584 case DCCP_SOCKOPT_SERVICE:
585 return dccp_getsockopt_service(sk, len,
586 (__be32 __user *)optval, optlen);
588 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
589 len, (u32 __user *)optval, optlen);
591 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
592 len, (u32 __user *)optval, optlen);
597 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
603 int dccp_getsockopt(struct sock *sk, int level, int optname,
604 char __user *optval, int __user *optlen)
606 if (level != SOL_DCCP)
607 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
610 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
613 EXPORT_SYMBOL_GPL(dccp_getsockopt);
616 int compat_dccp_getsockopt(struct sock *sk, int level, int optname,
617 char __user *optval, int __user *optlen)
619 if (level != SOL_DCCP)
620 return inet_csk_compat_getsockopt(sk, level, optname,
622 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
625 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt);
628 int dccp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
631 const struct dccp_sock *dp = dccp_sk(sk);
632 const int flags = msg->msg_flags;
633 const int noblock = flags & MSG_DONTWAIT;
638 if (len > dp->dccps_mss_cache)
642 timeo = sock_sndtimeo(sk, noblock);
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.
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)
653 size = sk->sk_prot->max_header + len;
655 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
660 skb_reserve(skb, sk->sk_prot->max_header);
661 rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
665 rc = dccp_write_xmit(sk, skb, &timeo);
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.
673 * This bug was _quickly_ found & fixed by just looking at an OSTRA
674 * generated callgraph 8) -acme
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)
689 const struct dccp_hdr *dh;
694 if (sk->sk_state == DCCP_LISTEN) {
699 timeo = sock_rcvtimeo(sk, nonblock);
702 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
705 goto verify_sock_status;
709 if (dh->dccph_type == DCCP_PKT_DATA ||
710 dh->dccph_type == DCCP_PKT_DATAACK)
713 if (dh->dccph_type == DCCP_PKT_RESET ||
714 dh->dccph_type == DCCP_PKT_CLOSE) {
715 dccp_pr_debug("found fin ok!\n");
719 dccp_pr_debug("packet_type=%s\n",
720 dccp_packet_name(dh->dccph_type));
723 if (sock_flag(sk, SOCK_DONE)) {
729 len = sock_error(sk);
733 if (sk->sk_shutdown & RCV_SHUTDOWN) {
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.
755 if (signal_pending(current)) {
756 len = sock_intr_errno(timeo);
760 sk_wait_data(sk, &timeo);
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! */
774 if (!(flags & MSG_PEEK))
783 EXPORT_SYMBOL_GPL(dccp_recvmsg);
785 int inet_dccp_listen(struct socket *sock, int backlog)
787 struct sock *sk = sock->sk;
788 unsigned char old_state;
794 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
797 old_state = sk->sk_state;
798 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
801 /* Really, if the socket is already in listen state
802 * we can only allow the backlog to be adjusted.
804 if (old_state != DCCP_LISTEN) {
806 * FIXME: here it probably should be sk->sk_prot->listen_start
807 * see tcp_listen_start
809 err = dccp_listen_start(sk);
813 sk->sk_max_ack_backlog = backlog;
821 EXPORT_SYMBOL_GPL(inet_dccp_listen);
823 static const unsigned char dccp_new_state[] = {
824 /* current state: new state: action: */
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,
836 static int dccp_close_state(struct sock *sk)
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;
847 void dccp_close(struct sock *sk, long timeout)
853 sk->sk_shutdown = SHUTDOWN_MASK;
855 if (sk->sk_state == DCCP_LISTEN) {
856 dccp_set_state(sk, DCCP_CLOSED);
859 inet_csk_listen_stop(sk);
861 goto adjudge_to_death;
865 * We need to flush the recv. buffs. We do this only on the
866 * descriptor close, not protocol-sourced closes, because the
867 *reader process may not have drained the data yet!
869 /* FIXME: check for unread data */
870 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
874 if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
875 /* Check zero linger _after_ checking for unread data. */
876 sk->sk_prot->disconnect(sk, 0);
877 } else if (dccp_close_state(sk)) {
878 dccp_send_close(sk, 1);
881 sk_stream_wait_close(sk, timeout);
885 * It is the last release_sock in its life. It will remove backlog.
889 * Now socket is owned by kernel and we acquire BH lock
890 * to finish close. No need to check for user refs.
894 BUG_TRAP(!sock_owned_by_user(sk));
900 * The last release_sock may have processed the CLOSE or RESET
901 * packet moving sock to CLOSED state, if not we have to fire
902 * the CLOSE/CLOSEREQ retransmission timer, see "8.3. Termination"
903 * in draft-ietf-dccp-spec-11. -acme
905 if (sk->sk_state == DCCP_CLOSING) {
906 /* FIXME: should start at 2 * RTT */
907 /* Timer for repeating the CLOSE/CLOSEREQ until an answer. */
908 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
909 inet_csk(sk)->icsk_rto,
912 /* Yeah, we should use sk->sk_prot->orphan_count, etc */
913 dccp_set_state(sk, DCCP_CLOSED);
917 atomic_inc(sk->sk_prot->orphan_count);
918 if (sk->sk_state == DCCP_CLOSED)
919 inet_csk_destroy_sock(sk);
921 /* Otherwise, socket is reprieved until protocol close. */
928 EXPORT_SYMBOL_GPL(dccp_close);
930 void dccp_shutdown(struct sock *sk, int how)
932 dccp_pr_debug("entry\n");
935 EXPORT_SYMBOL_GPL(dccp_shutdown);
937 static int __init dccp_mib_init(void)
941 dccp_statistics[0] = alloc_percpu(struct dccp_mib);
942 if (dccp_statistics[0] == NULL)
945 dccp_statistics[1] = alloc_percpu(struct dccp_mib);
946 if (dccp_statistics[1] == NULL)
953 free_percpu(dccp_statistics[0]);
954 dccp_statistics[0] = NULL;
959 static void dccp_mib_exit(void)
961 free_percpu(dccp_statistics[0]);
962 free_percpu(dccp_statistics[1]);
963 dccp_statistics[0] = dccp_statistics[1] = NULL;
966 static int thash_entries;
967 module_param(thash_entries, int, 0444);
968 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
970 #ifdef CONFIG_IP_DCCP_DEBUG
972 module_param(dccp_debug, int, 0444);
973 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
975 EXPORT_SYMBOL_GPL(dccp_debug);
978 static int __init dccp_init(void)
981 int ehash_order, bhash_order, i;
984 dccp_hashinfo.bind_bucket_cachep =
985 kmem_cache_create("dccp_bind_bucket",
986 sizeof(struct inet_bind_bucket), 0,
987 SLAB_HWCACHE_ALIGN, NULL, NULL);
988 if (!dccp_hashinfo.bind_bucket_cachep)
992 * Size and allocate the main established and bind bucket
995 * The methodology is similar to that of the buffer cache.
997 if (num_physpages >= (128 * 1024))
998 goal = num_physpages >> (21 - PAGE_SHIFT);
1000 goal = num_physpages >> (23 - PAGE_SHIFT);
1003 goal = (thash_entries *
1004 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1005 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1008 dccp_hashinfo.ehash_size = (1UL << ehash_order) * PAGE_SIZE /
1009 sizeof(struct inet_ehash_bucket);
1010 dccp_hashinfo.ehash_size >>= 1;
1011 while (dccp_hashinfo.ehash_size &
1012 (dccp_hashinfo.ehash_size - 1))
1013 dccp_hashinfo.ehash_size--;
1014 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1015 __get_free_pages(GFP_ATOMIC, ehash_order);
1016 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1018 if (!dccp_hashinfo.ehash) {
1019 printk(KERN_CRIT "Failed to allocate DCCP "
1020 "established hash table\n");
1021 goto out_free_bind_bucket_cachep;
1024 for (i = 0; i < (dccp_hashinfo.ehash_size << 1); i++) {
1025 rwlock_init(&dccp_hashinfo.ehash[i].lock);
1026 INIT_HLIST_HEAD(&dccp_hashinfo.ehash[i].chain);
1029 bhash_order = ehash_order;
1032 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1033 sizeof(struct inet_bind_hashbucket);
1034 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1037 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1038 __get_free_pages(GFP_ATOMIC, bhash_order);
1039 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1041 if (!dccp_hashinfo.bhash) {
1042 printk(KERN_CRIT "Failed to allocate DCCP bind hash table\n");
1043 goto out_free_dccp_ehash;
1046 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1047 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1048 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1051 rc = dccp_mib_init();
1053 goto out_free_dccp_bhash;
1055 rc = dccp_ackvec_init();
1057 goto out_free_dccp_mib;
1059 rc = dccp_sysctl_init();
1061 goto out_ackvec_exit;
1068 out_free_dccp_bhash:
1069 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1070 dccp_hashinfo.bhash = NULL;
1071 out_free_dccp_ehash:
1072 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1073 dccp_hashinfo.ehash = NULL;
1074 out_free_bind_bucket_cachep:
1075 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1076 dccp_hashinfo.bind_bucket_cachep = NULL;
1080 static void __exit dccp_fini(void)
1083 free_pages((unsigned long)dccp_hashinfo.bhash,
1084 get_order(dccp_hashinfo.bhash_size *
1085 sizeof(struct inet_bind_hashbucket)));
1086 free_pages((unsigned long)dccp_hashinfo.ehash,
1087 get_order(dccp_hashinfo.ehash_size *
1088 sizeof(struct inet_ehash_bucket)));
1089 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1094 module_init(dccp_init);
1095 module_exit(dccp_fini);
1097 MODULE_LICENSE("GPL");
1098 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1099 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");