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/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>
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>
29 #include <asm/ioctls.h>
30 #include <linux/spinlock.h>
31 #include <linux/timer.h>
32 #include <linux/delay.h>
33 #include <linux/poll.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),
53 EXPORT_SYMBOL_GPL(dccp_hashinfo);
55 /* the maximum queue length for tx in packets. 0 is no limit */
56 int sysctl_dccp_tx_qlen __read_mostly = 5;
58 void dccp_set_state(struct sock *sk, const int state)
60 const int oldstate = sk->sk_state;
62 dccp_pr_debug("%s(%p) %s --> %s\n", dccp_role(sk), sk,
63 dccp_state_name(oldstate), dccp_state_name(state));
64 WARN_ON(state == oldstate);
68 if (oldstate != DCCP_OPEN)
69 DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
70 /* Client retransmits all Confirm options until entering OPEN */
71 if (oldstate == DCCP_PARTOPEN)
72 dccp_feat_list_purge(&dccp_sk(sk)->dccps_featneg);
76 if (oldstate == DCCP_OPEN || oldstate == DCCP_ACTIVE_CLOSEREQ ||
77 oldstate == DCCP_CLOSING)
78 DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
80 sk->sk_prot->unhash(sk);
81 if (inet_csk(sk)->icsk_bind_hash != NULL &&
82 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
86 if (oldstate == DCCP_OPEN)
87 DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
90 /* Change state AFTER socket is unhashed to avoid closed
91 * socket sitting in hash tables.
96 EXPORT_SYMBOL_GPL(dccp_set_state);
98 static void dccp_finish_passive_close(struct sock *sk)
100 switch (sk->sk_state) {
101 case DCCP_PASSIVE_CLOSE:
102 /* Node (client or server) has received Close packet. */
103 dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
104 dccp_set_state(sk, DCCP_CLOSED);
106 case DCCP_PASSIVE_CLOSEREQ:
108 * Client received CloseReq. We set the `active' flag so that
109 * dccp_send_close() retransmits the Close as per RFC 4340, 8.3.
111 dccp_send_close(sk, 1);
112 dccp_set_state(sk, DCCP_CLOSING);
116 void dccp_done(struct sock *sk)
118 dccp_set_state(sk, DCCP_CLOSED);
119 dccp_clear_xmit_timers(sk);
121 sk->sk_shutdown = SHUTDOWN_MASK;
123 if (!sock_flag(sk, SOCK_DEAD))
124 sk->sk_state_change(sk);
126 inet_csk_destroy_sock(sk);
129 EXPORT_SYMBOL_GPL(dccp_done);
131 const char *dccp_packet_name(const int type)
133 static const char *dccp_packet_names[] = {
134 [DCCP_PKT_REQUEST] = "REQUEST",
135 [DCCP_PKT_RESPONSE] = "RESPONSE",
136 [DCCP_PKT_DATA] = "DATA",
137 [DCCP_PKT_ACK] = "ACK",
138 [DCCP_PKT_DATAACK] = "DATAACK",
139 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
140 [DCCP_PKT_CLOSE] = "CLOSE",
141 [DCCP_PKT_RESET] = "RESET",
142 [DCCP_PKT_SYNC] = "SYNC",
143 [DCCP_PKT_SYNCACK] = "SYNCACK",
146 if (type >= DCCP_NR_PKT_TYPES)
149 return dccp_packet_names[type];
152 EXPORT_SYMBOL_GPL(dccp_packet_name);
154 const char *dccp_state_name(const int state)
156 static char *dccp_state_names[] = {
157 [DCCP_OPEN] = "OPEN",
158 [DCCP_REQUESTING] = "REQUESTING",
159 [DCCP_PARTOPEN] = "PARTOPEN",
160 [DCCP_LISTEN] = "LISTEN",
161 [DCCP_RESPOND] = "RESPOND",
162 [DCCP_CLOSING] = "CLOSING",
163 [DCCP_ACTIVE_CLOSEREQ] = "CLOSEREQ",
164 [DCCP_PASSIVE_CLOSE] = "PASSIVE_CLOSE",
165 [DCCP_PASSIVE_CLOSEREQ] = "PASSIVE_CLOSEREQ",
166 [DCCP_TIME_WAIT] = "TIME_WAIT",
167 [DCCP_CLOSED] = "CLOSED",
170 if (state >= DCCP_MAX_STATES)
171 return "INVALID STATE!";
173 return dccp_state_names[state];
176 EXPORT_SYMBOL_GPL(dccp_state_name);
178 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
180 struct dccp_sock *dp = dccp_sk(sk);
181 struct inet_connection_sock *icsk = inet_csk(sk);
183 dccp_minisock_init(&dp->dccps_minisock);
185 icsk->icsk_rto = DCCP_TIMEOUT_INIT;
186 icsk->icsk_syn_retries = sysctl_dccp_request_retries;
187 sk->sk_state = DCCP_CLOSED;
188 sk->sk_write_space = dccp_write_space;
189 icsk->icsk_sync_mss = dccp_sync_mss;
190 dp->dccps_mss_cache = 536;
191 dp->dccps_rate_last = jiffies;
192 dp->dccps_role = DCCP_ROLE_UNDEFINED;
193 dp->dccps_service = DCCP_SERVICE_CODE_IS_ABSENT;
194 dp->dccps_l_ack_ratio = dp->dccps_r_ack_ratio = 1;
196 dccp_init_xmit_timers(sk);
198 INIT_LIST_HEAD(&dp->dccps_featneg);
199 /* control socket doesn't need feat nego */
200 if (likely(ctl_sock_initialized))
201 return dccp_feat_init(sk);
205 EXPORT_SYMBOL_GPL(dccp_init_sock);
207 void dccp_destroy_sock(struct sock *sk)
209 struct dccp_sock *dp = dccp_sk(sk);
210 struct dccp_minisock *dmsk = dccp_msk(sk);
213 * DCCP doesn't use sk_write_queue, just sk_send_head
214 * for retransmissions
216 if (sk->sk_send_head != NULL) {
217 kfree_skb(sk->sk_send_head);
218 sk->sk_send_head = NULL;
221 /* Clean up a referenced DCCP bind bucket. */
222 if (inet_csk(sk)->icsk_bind_hash != NULL)
225 kfree(dp->dccps_service_list);
226 dp->dccps_service_list = NULL;
228 if (dmsk->dccpms_send_ack_vector) {
229 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
230 dp->dccps_hc_rx_ackvec = NULL;
232 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
233 ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
234 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
236 /* clean up feature negotiation state */
237 dccp_feat_list_purge(&dp->dccps_featneg);
240 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
242 static inline int dccp_listen_start(struct sock *sk, int backlog)
244 struct dccp_sock *dp = dccp_sk(sk);
246 dp->dccps_role = DCCP_ROLE_LISTEN;
247 /* do not start to listen if feature negotiation setup fails */
248 if (dccp_feat_finalise_settings(dp))
250 return inet_csk_listen_start(sk, backlog);
253 static inline int dccp_need_reset(int state)
255 return state != DCCP_CLOSED && state != DCCP_LISTEN &&
256 state != DCCP_REQUESTING;
259 int dccp_disconnect(struct sock *sk, int flags)
261 struct inet_connection_sock *icsk = inet_csk(sk);
262 struct inet_sock *inet = inet_sk(sk);
264 const int old_state = sk->sk_state;
266 if (old_state != DCCP_CLOSED)
267 dccp_set_state(sk, DCCP_CLOSED);
270 * This corresponds to the ABORT function of RFC793, sec. 3.8
271 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
273 if (old_state == DCCP_LISTEN) {
274 inet_csk_listen_stop(sk);
275 } else if (dccp_need_reset(old_state)) {
276 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
277 sk->sk_err = ECONNRESET;
278 } else if (old_state == DCCP_REQUESTING)
279 sk->sk_err = ECONNRESET;
281 dccp_clear_xmit_timers(sk);
283 __skb_queue_purge(&sk->sk_receive_queue);
284 __skb_queue_purge(&sk->sk_write_queue);
285 if (sk->sk_send_head != NULL) {
286 __kfree_skb(sk->sk_send_head);
287 sk->sk_send_head = NULL;
292 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
293 inet_reset_saddr(sk);
296 sock_reset_flag(sk, SOCK_DONE);
298 icsk->icsk_backoff = 0;
299 inet_csk_delack_init(sk);
302 WARN_ON(inet->num && !icsk->icsk_bind_hash);
304 sk->sk_error_report(sk);
308 EXPORT_SYMBOL_GPL(dccp_disconnect);
311 * Wait for a DCCP event.
313 * Note that we don't need to lock the socket, as the upper poll layers
314 * take care of normal races (between the test and the event) and we don't
315 * go look at any of the socket buffers directly.
317 unsigned int dccp_poll(struct file *file, struct socket *sock,
321 struct sock *sk = sock->sk;
323 poll_wait(file, sk->sk_sleep, wait);
324 if (sk->sk_state == DCCP_LISTEN)
325 return inet_csk_listen_poll(sk);
327 /* Socket is not locked. We are protected from async events
328 by poll logic and correct handling of state changes
329 made by another threads is impossible in any case.
336 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
338 if (sk->sk_shutdown & RCV_SHUTDOWN)
339 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
342 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
343 if (atomic_read(&sk->sk_rmem_alloc) > 0)
344 mask |= POLLIN | POLLRDNORM;
346 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
347 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
348 mask |= POLLOUT | POLLWRNORM;
349 } else { /* send SIGIO later */
350 set_bit(SOCK_ASYNC_NOSPACE,
351 &sk->sk_socket->flags);
352 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
354 /* Race breaker. If space is freed after
355 * wspace test but before the flags are set,
356 * IO signal will be lost.
358 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
359 mask |= POLLOUT | POLLWRNORM;
366 EXPORT_SYMBOL_GPL(dccp_poll);
368 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
374 if (sk->sk_state == DCCP_LISTEN)
380 unsigned long amount = 0;
382 skb = skb_peek(&sk->sk_receive_queue);
385 * We will only return the amount of this packet since
386 * that is all that will be read.
390 rc = put_user(amount, (int __user *)arg);
402 EXPORT_SYMBOL_GPL(dccp_ioctl);
404 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
405 char __user *optval, int optlen)
407 struct dccp_sock *dp = dccp_sk(sk);
408 struct dccp_service_list *sl = NULL;
410 if (service == DCCP_SERVICE_INVALID_VALUE ||
411 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
414 if (optlen > sizeof(service)) {
415 sl = kmalloc(optlen, GFP_KERNEL);
419 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
420 if (copy_from_user(sl->dccpsl_list,
421 optval + sizeof(service),
422 optlen - sizeof(service)) ||
423 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
430 dp->dccps_service = service;
432 kfree(dp->dccps_service_list);
434 dp->dccps_service_list = sl;
439 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
444 if (cscov < 0 || cscov > 15)
447 * Populate a list of permissible values, in the range cscov...15. This
448 * is necessary since feature negotiation of single values only works if
449 * both sides incidentally choose the same value. Since the list starts
450 * lowest-value first, negotiation will pick the smallest shared value.
456 list = kmalloc(len, GFP_KERNEL);
460 for (i = 0; i < len; i++)
463 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
467 dccp_sk(sk)->dccps_pcrlen = cscov;
469 dccp_sk(sk)->dccps_pcslen = cscov;
475 static int dccp_setsockopt_ccid(struct sock *sk, int type,
476 char __user *optval, int optlen)
481 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
484 val = kmalloc(optlen, GFP_KERNEL);
488 if (copy_from_user(val, optval, optlen)) {
494 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
495 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
497 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
498 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
505 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
506 char __user *optval, int optlen)
508 struct dccp_sock *dp = dccp_sk(sk);
512 case DCCP_SOCKOPT_PACKET_SIZE:
513 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
515 case DCCP_SOCKOPT_CHANGE_L:
516 case DCCP_SOCKOPT_CHANGE_R:
517 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
519 case DCCP_SOCKOPT_CCID:
520 case DCCP_SOCKOPT_RX_CCID:
521 case DCCP_SOCKOPT_TX_CCID:
522 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
525 if (optlen < (int)sizeof(int))
528 if (get_user(val, (int __user *)optval))
531 if (optname == DCCP_SOCKOPT_SERVICE)
532 return dccp_setsockopt_service(sk, val, optval, optlen);
536 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
537 if (dp->dccps_role != DCCP_ROLE_SERVER)
540 dp->dccps_server_timewait = (val != 0);
542 case DCCP_SOCKOPT_SEND_CSCOV:
543 err = dccp_setsockopt_cscov(sk, val, false);
545 case DCCP_SOCKOPT_RECV_CSCOV:
546 err = dccp_setsockopt_cscov(sk, val, true);
557 int dccp_setsockopt(struct sock *sk, int level, int optname,
558 char __user *optval, int optlen)
560 if (level != SOL_DCCP)
561 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
564 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
567 EXPORT_SYMBOL_GPL(dccp_setsockopt);
570 int compat_dccp_setsockopt(struct sock *sk, int level, int optname,
571 char __user *optval, int optlen)
573 if (level != SOL_DCCP)
574 return inet_csk_compat_setsockopt(sk, level, optname,
576 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
579 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt);
582 static int dccp_getsockopt_service(struct sock *sk, int len,
583 __be32 __user *optval,
586 const struct dccp_sock *dp = dccp_sk(sk);
587 const struct dccp_service_list *sl;
588 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
591 if ((sl = dp->dccps_service_list) != NULL) {
592 slen = sl->dccpsl_nr * sizeof(u32);
601 if (put_user(total_len, optlen) ||
602 put_user(dp->dccps_service, optval) ||
603 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
610 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
611 char __user *optval, int __user *optlen)
613 struct dccp_sock *dp;
616 if (get_user(len, optlen))
619 if (len < (int)sizeof(int))
625 case DCCP_SOCKOPT_PACKET_SIZE:
626 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
628 case DCCP_SOCKOPT_SERVICE:
629 return dccp_getsockopt_service(sk, len,
630 (__be32 __user *)optval, optlen);
631 case DCCP_SOCKOPT_GET_CUR_MPS:
632 val = dp->dccps_mss_cache;
634 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
635 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
636 case DCCP_SOCKOPT_TX_CCID:
637 val = ccid_get_current_tx_ccid(dp);
641 case DCCP_SOCKOPT_RX_CCID:
642 val = ccid_get_current_rx_ccid(dp);
646 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
647 val = dp->dccps_server_timewait;
649 case DCCP_SOCKOPT_SEND_CSCOV:
650 val = dp->dccps_pcslen;
652 case DCCP_SOCKOPT_RECV_CSCOV:
653 val = dp->dccps_pcrlen;
656 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
657 len, (u32 __user *)optval, optlen);
659 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
660 len, (u32 __user *)optval, optlen);
666 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
672 int dccp_getsockopt(struct sock *sk, int level, int optname,
673 char __user *optval, int __user *optlen)
675 if (level != SOL_DCCP)
676 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
679 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
682 EXPORT_SYMBOL_GPL(dccp_getsockopt);
685 int compat_dccp_getsockopt(struct sock *sk, int level, int optname,
686 char __user *optval, int __user *optlen)
688 if (level != SOL_DCCP)
689 return inet_csk_compat_getsockopt(sk, level, optname,
691 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
694 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt);
697 int dccp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
700 const struct dccp_sock *dp = dccp_sk(sk);
701 const int flags = msg->msg_flags;
702 const int noblock = flags & MSG_DONTWAIT;
707 if (len > dp->dccps_mss_cache)
712 if (sysctl_dccp_tx_qlen &&
713 (sk->sk_write_queue.qlen >= sysctl_dccp_tx_qlen)) {
718 timeo = sock_sndtimeo(sk, noblock);
721 * We have to use sk_stream_wait_connect here to set sk_write_pending,
722 * so that the trick in dccp_rcv_request_sent_state_process.
724 /* Wait for a connection to finish. */
725 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
726 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
729 size = sk->sk_prot->max_header + len;
731 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
736 skb_reserve(skb, sk->sk_prot->max_header);
737 rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
741 skb_queue_tail(&sk->sk_write_queue, skb);
742 dccp_write_xmit(sk,0);
751 EXPORT_SYMBOL_GPL(dccp_sendmsg);
753 int dccp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
754 size_t len, int nonblock, int flags, int *addr_len)
756 const struct dccp_hdr *dh;
761 if (sk->sk_state == DCCP_LISTEN) {
766 timeo = sock_rcvtimeo(sk, nonblock);
769 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
772 goto verify_sock_status;
776 switch (dh->dccph_type) {
778 case DCCP_PKT_DATAACK:
782 case DCCP_PKT_CLOSEREQ:
783 if (!(flags & MSG_PEEK))
784 dccp_finish_passive_close(sk);
787 dccp_pr_debug("found fin (%s) ok!\n",
788 dccp_packet_name(dh->dccph_type));
792 dccp_pr_debug("packet_type=%s\n",
793 dccp_packet_name(dh->dccph_type));
794 sk_eat_skb(sk, skb, 0);
797 if (sock_flag(sk, SOCK_DONE)) {
803 len = sock_error(sk);
807 if (sk->sk_shutdown & RCV_SHUTDOWN) {
812 if (sk->sk_state == DCCP_CLOSED) {
813 if (!sock_flag(sk, SOCK_DONE)) {
814 /* This occurs when user tries to read
815 * from never connected socket.
829 if (signal_pending(current)) {
830 len = sock_intr_errno(timeo);
834 sk_wait_data(sk, &timeo);
839 else if (len < skb->len)
840 msg->msg_flags |= MSG_TRUNC;
842 if (skb_copy_datagram_iovec(skb, 0, msg->msg_iov, len)) {
843 /* Exception. Bailout! */
848 if (!(flags & MSG_PEEK))
849 sk_eat_skb(sk, skb, 0);
857 EXPORT_SYMBOL_GPL(dccp_recvmsg);
859 int inet_dccp_listen(struct socket *sock, int backlog)
861 struct sock *sk = sock->sk;
862 unsigned char old_state;
868 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
871 old_state = sk->sk_state;
872 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
875 /* Really, if the socket is already in listen state
876 * we can only allow the backlog to be adjusted.
878 if (old_state != DCCP_LISTEN) {
880 * FIXME: here it probably should be sk->sk_prot->listen_start
881 * see tcp_listen_start
883 err = dccp_listen_start(sk, backlog);
887 sk->sk_max_ack_backlog = backlog;
895 EXPORT_SYMBOL_GPL(inet_dccp_listen);
897 static void dccp_terminate_connection(struct sock *sk)
899 u8 next_state = DCCP_CLOSED;
901 switch (sk->sk_state) {
902 case DCCP_PASSIVE_CLOSE:
903 case DCCP_PASSIVE_CLOSEREQ:
904 dccp_finish_passive_close(sk);
907 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
908 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
911 dccp_send_close(sk, 1);
913 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
914 !dccp_sk(sk)->dccps_server_timewait)
915 next_state = DCCP_ACTIVE_CLOSEREQ;
917 next_state = DCCP_CLOSING;
920 dccp_set_state(sk, next_state);
924 void dccp_close(struct sock *sk, long timeout)
926 struct dccp_sock *dp = dccp_sk(sk);
928 u32 data_was_unread = 0;
933 sk->sk_shutdown = SHUTDOWN_MASK;
935 if (sk->sk_state == DCCP_LISTEN) {
936 dccp_set_state(sk, DCCP_CLOSED);
939 inet_csk_listen_stop(sk);
941 goto adjudge_to_death;
944 sk_stop_timer(sk, &dp->dccps_xmit_timer);
947 * We need to flush the recv. buffs. We do this only on the
948 * descriptor close, not protocol-sourced closes, because the
949 *reader process may not have drained the data yet!
951 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
952 data_was_unread += skb->len;
956 if (data_was_unread) {
957 /* Unread data was tossed, send an appropriate Reset Code */
958 DCCP_WARN("DCCP: ABORT -- %u bytes unread\n", data_was_unread);
959 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
960 dccp_set_state(sk, DCCP_CLOSED);
961 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
962 /* Check zero linger _after_ checking for unread data. */
963 sk->sk_prot->disconnect(sk, 0);
964 } else if (sk->sk_state != DCCP_CLOSED) {
965 dccp_terminate_connection(sk);
968 sk_stream_wait_close(sk, timeout);
971 state = sk->sk_state;
974 atomic_inc(sk->sk_prot->orphan_count);
977 * It is the last release_sock in its life. It will remove backlog.
981 * Now socket is owned by kernel and we acquire BH lock
982 * to finish close. No need to check for user refs.
986 WARN_ON(sock_owned_by_user(sk));
988 /* Have we already been destroyed by a softirq or backlog? */
989 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
992 if (sk->sk_state == DCCP_CLOSED)
993 inet_csk_destroy_sock(sk);
995 /* Otherwise, socket is reprieved until protocol close. */
1003 EXPORT_SYMBOL_GPL(dccp_close);
1005 void dccp_shutdown(struct sock *sk, int how)
1007 dccp_pr_debug("called shutdown(%x)\n", how);
1010 EXPORT_SYMBOL_GPL(dccp_shutdown);
1012 static inline int dccp_mib_init(void)
1014 return snmp_mib_init((void**)dccp_statistics, sizeof(struct dccp_mib));
1017 static inline void dccp_mib_exit(void)
1019 snmp_mib_free((void**)dccp_statistics);
1022 static int thash_entries;
1023 module_param(thash_entries, int, 0444);
1024 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1026 #ifdef CONFIG_IP_DCCP_DEBUG
1028 module_param(dccp_debug, bool, 0644);
1029 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1031 EXPORT_SYMBOL_GPL(dccp_debug);
1034 static int __init dccp_init(void)
1037 int ehash_order, bhash_order, i;
1040 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1041 FIELD_SIZEOF(struct sk_buff, cb));
1043 dccp_hashinfo.bind_bucket_cachep =
1044 kmem_cache_create("dccp_bind_bucket",
1045 sizeof(struct inet_bind_bucket), 0,
1046 SLAB_HWCACHE_ALIGN, NULL);
1047 if (!dccp_hashinfo.bind_bucket_cachep)
1051 * Size and allocate the main established and bind bucket
1054 * The methodology is similar to that of the buffer cache.
1056 if (num_physpages >= (128 * 1024))
1057 goal = num_physpages >> (21 - PAGE_SHIFT);
1059 goal = num_physpages >> (23 - PAGE_SHIFT);
1062 goal = (thash_entries *
1063 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1064 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1067 dccp_hashinfo.ehash_size = (1UL << ehash_order) * PAGE_SIZE /
1068 sizeof(struct inet_ehash_bucket);
1069 while (dccp_hashinfo.ehash_size &
1070 (dccp_hashinfo.ehash_size - 1))
1071 dccp_hashinfo.ehash_size--;
1072 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1073 __get_free_pages(GFP_ATOMIC, ehash_order);
1074 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1076 if (!dccp_hashinfo.ehash) {
1077 DCCP_CRIT("Failed to allocate DCCP established hash table");
1078 goto out_free_bind_bucket_cachep;
1081 for (i = 0; i < dccp_hashinfo.ehash_size; i++) {
1082 INIT_HLIST_HEAD(&dccp_hashinfo.ehash[i].chain);
1083 INIT_HLIST_HEAD(&dccp_hashinfo.ehash[i].twchain);
1086 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1087 goto out_free_dccp_ehash;
1089 bhash_order = ehash_order;
1092 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1093 sizeof(struct inet_bind_hashbucket);
1094 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1097 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1098 __get_free_pages(GFP_ATOMIC, bhash_order);
1099 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1101 if (!dccp_hashinfo.bhash) {
1102 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1103 goto out_free_dccp_locks;
1106 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1107 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1108 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1111 rc = dccp_mib_init();
1113 goto out_free_dccp_bhash;
1115 rc = dccp_ackvec_init();
1117 goto out_free_dccp_mib;
1119 rc = dccp_sysctl_init();
1121 goto out_ackvec_exit;
1123 dccp_timestamping_init();
1130 out_free_dccp_bhash:
1131 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1132 dccp_hashinfo.bhash = NULL;
1133 out_free_dccp_locks:
1134 inet_ehash_locks_free(&dccp_hashinfo);
1135 out_free_dccp_ehash:
1136 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1137 dccp_hashinfo.ehash = NULL;
1138 out_free_bind_bucket_cachep:
1139 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1140 dccp_hashinfo.bind_bucket_cachep = NULL;
1144 static void __exit dccp_fini(void)
1147 free_pages((unsigned long)dccp_hashinfo.bhash,
1148 get_order(dccp_hashinfo.bhash_size *
1149 sizeof(struct inet_bind_hashbucket)));
1150 free_pages((unsigned long)dccp_hashinfo.ehash,
1151 get_order(dccp_hashinfo.ehash_size *
1152 sizeof(struct inet_ehash_bucket)));
1153 inet_ehash_locks_free(&dccp_hashinfo);
1154 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1159 module_init(dccp_init);
1160 module_exit(dccp_fini);
1162 MODULE_LICENSE("GPL");
1163 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1164 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");