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 <linux/slab.h>
24 #include <net/checksum.h>
26 #include <net/inet_sock.h>
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>
40 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
42 EXPORT_SYMBOL_GPL(dccp_statistics);
44 struct percpu_counter dccp_orphan_count;
45 EXPORT_SYMBOL_GPL(dccp_orphan_count);
47 struct inet_hashinfo dccp_hashinfo;
48 EXPORT_SYMBOL_GPL(dccp_hashinfo);
50 /* the maximum queue length for tx in packets. 0 is no limit */
51 int sysctl_dccp_tx_qlen __read_mostly = 5;
53 #ifdef CONFIG_IP_DCCP_DEBUG
54 static const char *dccp_state_name(const int state)
56 static const char *const dccp_state_names[] = {
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",
70 if (state >= DCCP_MAX_STATES)
71 return "INVALID STATE!";
73 return dccp_state_names[state];
77 void dccp_set_state(struct sock *sk, const int state)
79 const int oldstate = sk->sk_state;
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);
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);
95 if (oldstate == DCCP_OPEN || oldstate == DCCP_ACTIVE_CLOSEREQ ||
96 oldstate == DCCP_CLOSING)
97 DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
99 sk->sk_prot->unhash(sk);
100 if (inet_csk(sk)->icsk_bind_hash != NULL &&
101 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
105 if (oldstate == DCCP_OPEN)
106 DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
109 /* Change state AFTER socket is unhashed to avoid closed
110 * socket sitting in hash tables.
112 sk->sk_state = state;
115 EXPORT_SYMBOL_GPL(dccp_set_state);
117 static void dccp_finish_passive_close(struct sock *sk)
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);
125 case DCCP_PASSIVE_CLOSEREQ:
127 * Client received CloseReq. We set the `active' flag so that
128 * dccp_send_close() retransmits the Close as per RFC 4340, 8.3.
130 dccp_send_close(sk, 1);
131 dccp_set_state(sk, DCCP_CLOSING);
135 void dccp_done(struct sock *sk)
137 dccp_set_state(sk, DCCP_CLOSED);
138 dccp_clear_xmit_timers(sk);
140 sk->sk_shutdown = SHUTDOWN_MASK;
142 if (!sock_flag(sk, SOCK_DEAD))
143 sk->sk_state_change(sk);
145 inet_csk_destroy_sock(sk);
148 EXPORT_SYMBOL_GPL(dccp_done);
150 const char *dccp_packet_name(const int type)
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",
165 if (type >= DCCP_NR_PKT_TYPES)
168 return dccp_packet_names[type];
171 EXPORT_SYMBOL_GPL(dccp_packet_name);
173 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
175 struct dccp_sock *dp = dccp_sk(sk);
176 struct inet_connection_sock *icsk = inet_csk(sk);
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_tx_qlen = sysctl_dccp_tx_qlen;
189 dccp_init_xmit_timers(sk);
191 INIT_LIST_HEAD(&dp->dccps_featneg);
192 /* control socket doesn't need feat nego */
193 if (likely(ctl_sock_initialized))
194 return dccp_feat_init(sk);
198 EXPORT_SYMBOL_GPL(dccp_init_sock);
200 void dccp_destroy_sock(struct sock *sk)
202 struct dccp_sock *dp = dccp_sk(sk);
204 __skb_queue_purge(&sk->sk_write_queue);
205 if (sk->sk_send_head != NULL) {
206 kfree_skb(sk->sk_send_head);
207 sk->sk_send_head = NULL;
210 /* Clean up a referenced DCCP bind bucket. */
211 if (inet_csk(sk)->icsk_bind_hash != NULL)
214 kfree(dp->dccps_service_list);
215 dp->dccps_service_list = NULL;
217 if (dp->dccps_hc_rx_ackvec != NULL) {
218 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
219 dp->dccps_hc_rx_ackvec = NULL;
221 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
222 ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
223 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
225 /* clean up feature negotiation state */
226 dccp_feat_list_purge(&dp->dccps_featneg);
229 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
231 static inline int dccp_listen_start(struct sock *sk, int backlog)
233 struct dccp_sock *dp = dccp_sk(sk);
235 dp->dccps_role = DCCP_ROLE_LISTEN;
236 /* do not start to listen if feature negotiation setup fails */
237 if (dccp_feat_finalise_settings(dp))
239 return inet_csk_listen_start(sk, backlog);
242 static inline int dccp_need_reset(int state)
244 return state != DCCP_CLOSED && state != DCCP_LISTEN &&
245 state != DCCP_REQUESTING;
248 int dccp_disconnect(struct sock *sk, int flags)
250 struct inet_connection_sock *icsk = inet_csk(sk);
251 struct inet_sock *inet = inet_sk(sk);
253 const int old_state = sk->sk_state;
255 if (old_state != DCCP_CLOSED)
256 dccp_set_state(sk, DCCP_CLOSED);
259 * This corresponds to the ABORT function of RFC793, sec. 3.8
260 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
262 if (old_state == DCCP_LISTEN) {
263 inet_csk_listen_stop(sk);
264 } else if (dccp_need_reset(old_state)) {
265 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
266 sk->sk_err = ECONNRESET;
267 } else if (old_state == DCCP_REQUESTING)
268 sk->sk_err = ECONNRESET;
270 dccp_clear_xmit_timers(sk);
272 __skb_queue_purge(&sk->sk_receive_queue);
273 __skb_queue_purge(&sk->sk_write_queue);
274 if (sk->sk_send_head != NULL) {
275 __kfree_skb(sk->sk_send_head);
276 sk->sk_send_head = NULL;
279 inet->inet_dport = 0;
281 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
282 inet_reset_saddr(sk);
285 sock_reset_flag(sk, SOCK_DONE);
287 icsk->icsk_backoff = 0;
288 inet_csk_delack_init(sk);
291 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
293 sk->sk_error_report(sk);
297 EXPORT_SYMBOL_GPL(dccp_disconnect);
300 * Wait for a DCCP event.
302 * Note that we don't need to lock the socket, as the upper poll layers
303 * take care of normal races (between the test and the event) and we don't
304 * go look at any of the socket buffers directly.
306 unsigned int dccp_poll(struct file *file, struct socket *sock,
310 struct sock *sk = sock->sk;
312 sock_poll_wait(file, sk_sleep(sk), wait);
313 if (sk->sk_state == DCCP_LISTEN)
314 return inet_csk_listen_poll(sk);
316 /* Socket is not locked. We are protected from async events
317 by poll logic and correct handling of state changes
318 made by another threads is impossible in any case.
325 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
327 if (sk->sk_shutdown & RCV_SHUTDOWN)
328 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
331 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
332 if (atomic_read(&sk->sk_rmem_alloc) > 0)
333 mask |= POLLIN | POLLRDNORM;
335 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
336 if (sk_stream_is_writeable(sk)) {
337 mask |= POLLOUT | POLLWRNORM;
338 } else { /* send SIGIO later */
339 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
340 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
342 /* Race breaker. If space is freed after
343 * wspace test but before the flags are set,
344 * IO signal will be lost.
346 if (sk_stream_is_writeable(sk))
347 mask |= POLLOUT | POLLWRNORM;
354 EXPORT_SYMBOL_GPL(dccp_poll);
356 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
362 if (sk->sk_state == DCCP_LISTEN)
368 unsigned long amount = 0;
370 skb = skb_peek(&sk->sk_receive_queue);
373 * We will only return the amount of this packet since
374 * that is all that will be read.
378 rc = put_user(amount, (int __user *)arg);
390 EXPORT_SYMBOL_GPL(dccp_ioctl);
392 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
393 char __user *optval, unsigned int optlen)
395 struct dccp_sock *dp = dccp_sk(sk);
396 struct dccp_service_list *sl = NULL;
398 if (service == DCCP_SERVICE_INVALID_VALUE ||
399 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
402 if (optlen > sizeof(service)) {
403 sl = kmalloc(optlen, GFP_KERNEL);
407 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
408 if (copy_from_user(sl->dccpsl_list,
409 optval + sizeof(service),
410 optlen - sizeof(service)) ||
411 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
418 dp->dccps_service = service;
420 kfree(dp->dccps_service_list);
422 dp->dccps_service_list = sl;
427 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
432 if (cscov < 0 || cscov > 15)
435 * Populate a list of permissible values, in the range cscov...15. This
436 * is necessary since feature negotiation of single values only works if
437 * both sides incidentally choose the same value. Since the list starts
438 * lowest-value first, negotiation will pick the smallest shared value.
444 list = kmalloc(len, GFP_KERNEL);
448 for (i = 0; i < len; i++)
451 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
455 dccp_sk(sk)->dccps_pcrlen = cscov;
457 dccp_sk(sk)->dccps_pcslen = cscov;
463 static int dccp_setsockopt_ccid(struct sock *sk, int type,
464 char __user *optval, unsigned int optlen)
469 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
472 val = memdup_user(optval, optlen);
477 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
478 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
480 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
481 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
488 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
489 char __user *optval, unsigned int optlen)
491 struct dccp_sock *dp = dccp_sk(sk);
495 case DCCP_SOCKOPT_PACKET_SIZE:
496 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
498 case DCCP_SOCKOPT_CHANGE_L:
499 case DCCP_SOCKOPT_CHANGE_R:
500 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
502 case DCCP_SOCKOPT_CCID:
503 case DCCP_SOCKOPT_RX_CCID:
504 case DCCP_SOCKOPT_TX_CCID:
505 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
508 if (optlen < (int)sizeof(int))
511 if (get_user(val, (int __user *)optval))
514 if (optname == DCCP_SOCKOPT_SERVICE)
515 return dccp_setsockopt_service(sk, val, optval, optlen);
519 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
520 if (dp->dccps_role != DCCP_ROLE_SERVER)
523 dp->dccps_server_timewait = (val != 0);
525 case DCCP_SOCKOPT_SEND_CSCOV:
526 err = dccp_setsockopt_cscov(sk, val, false);
528 case DCCP_SOCKOPT_RECV_CSCOV:
529 err = dccp_setsockopt_cscov(sk, val, true);
531 case DCCP_SOCKOPT_QPOLICY_ID:
532 if (sk->sk_state != DCCP_CLOSED)
534 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
537 dp->dccps_qpolicy = val;
539 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
543 dp->dccps_tx_qlen = val;
554 int dccp_setsockopt(struct sock *sk, int level, int optname,
555 char __user *optval, unsigned int optlen)
557 if (level != SOL_DCCP)
558 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
561 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
564 EXPORT_SYMBOL_GPL(dccp_setsockopt);
567 int compat_dccp_setsockopt(struct sock *sk, int level, int optname,
568 char __user *optval, unsigned int optlen)
570 if (level != SOL_DCCP)
571 return inet_csk_compat_setsockopt(sk, level, optname,
573 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
576 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt);
579 static int dccp_getsockopt_service(struct sock *sk, int len,
580 __be32 __user *optval,
583 const struct dccp_sock *dp = dccp_sk(sk);
584 const struct dccp_service_list *sl;
585 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
588 if ((sl = dp->dccps_service_list) != NULL) {
589 slen = sl->dccpsl_nr * sizeof(u32);
598 if (put_user(total_len, optlen) ||
599 put_user(dp->dccps_service, optval) ||
600 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
607 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
608 char __user *optval, int __user *optlen)
610 struct dccp_sock *dp;
613 if (get_user(len, optlen))
616 if (len < (int)sizeof(int))
622 case DCCP_SOCKOPT_PACKET_SIZE:
623 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
625 case DCCP_SOCKOPT_SERVICE:
626 return dccp_getsockopt_service(sk, len,
627 (__be32 __user *)optval, optlen);
628 case DCCP_SOCKOPT_GET_CUR_MPS:
629 val = dp->dccps_mss_cache;
631 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
632 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
633 case DCCP_SOCKOPT_TX_CCID:
634 val = ccid_get_current_tx_ccid(dp);
638 case DCCP_SOCKOPT_RX_CCID:
639 val = ccid_get_current_rx_ccid(dp);
643 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
644 val = dp->dccps_server_timewait;
646 case DCCP_SOCKOPT_SEND_CSCOV:
647 val = dp->dccps_pcslen;
649 case DCCP_SOCKOPT_RECV_CSCOV:
650 val = dp->dccps_pcrlen;
652 case DCCP_SOCKOPT_QPOLICY_ID:
653 val = dp->dccps_qpolicy;
655 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
656 val = dp->dccps_tx_qlen;
659 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
660 len, (u32 __user *)optval, optlen);
662 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
663 len, (u32 __user *)optval, optlen);
669 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
675 int dccp_getsockopt(struct sock *sk, int level, int optname,
676 char __user *optval, int __user *optlen)
678 if (level != SOL_DCCP)
679 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
682 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
685 EXPORT_SYMBOL_GPL(dccp_getsockopt);
688 int compat_dccp_getsockopt(struct sock *sk, int level, int optname,
689 char __user *optval, int __user *optlen)
691 if (level != SOL_DCCP)
692 return inet_csk_compat_getsockopt(sk, level, optname,
694 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
697 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt);
700 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
702 struct cmsghdr *cmsg;
705 * Assign an (opaque) qpolicy priority value to skb->priority.
707 * We are overloading this skb field for use with the qpolicy subystem.
708 * The skb->priority is normally used for the SO_PRIORITY option, which
709 * is initialised from sk_priority. Since the assignment of sk_priority
710 * to skb->priority happens later (on layer 3), we overload this field
711 * for use with queueing priorities as long as the skb is on layer 4.
712 * The default priority value (if nothing is set) is 0.
716 for_each_cmsghdr(cmsg, msg) {
717 if (!CMSG_OK(msg, cmsg))
720 if (cmsg->cmsg_level != SOL_DCCP)
723 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
724 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
727 switch (cmsg->cmsg_type) {
728 case DCCP_SCM_PRIORITY:
729 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
731 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
740 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
742 const struct dccp_sock *dp = dccp_sk(sk);
743 const int flags = msg->msg_flags;
744 const int noblock = flags & MSG_DONTWAIT;
749 if (len > dp->dccps_mss_cache)
754 if (dccp_qpolicy_full(sk)) {
759 timeo = sock_sndtimeo(sk, noblock);
762 * We have to use sk_stream_wait_connect here to set sk_write_pending,
763 * so that the trick in dccp_rcv_request_sent_state_process.
765 /* Wait for a connection to finish. */
766 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
767 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
770 size = sk->sk_prot->max_header + len;
772 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
777 skb_reserve(skb, sk->sk_prot->max_header);
778 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
782 rc = dccp_msghdr_parse(msg, skb);
786 dccp_qpolicy_push(sk, skb);
788 * The xmit_timer is set if the TX CCID is rate-based and will expire
789 * when congestion control permits to release further packets into the
790 * network. Window-based CCIDs do not use this timer.
792 if (!timer_pending(&dp->dccps_xmit_timer))
802 EXPORT_SYMBOL_GPL(dccp_sendmsg);
804 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
805 int flags, int *addr_len)
807 const struct dccp_hdr *dh;
812 if (sk->sk_state == DCCP_LISTEN) {
817 timeo = sock_rcvtimeo(sk, nonblock);
820 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
823 goto verify_sock_status;
827 switch (dh->dccph_type) {
829 case DCCP_PKT_DATAACK:
833 case DCCP_PKT_CLOSEREQ:
834 if (!(flags & MSG_PEEK))
835 dccp_finish_passive_close(sk);
838 dccp_pr_debug("found fin (%s) ok!\n",
839 dccp_packet_name(dh->dccph_type));
843 dccp_pr_debug("packet_type=%s\n",
844 dccp_packet_name(dh->dccph_type));
848 if (sock_flag(sk, SOCK_DONE)) {
854 len = sock_error(sk);
858 if (sk->sk_shutdown & RCV_SHUTDOWN) {
863 if (sk->sk_state == DCCP_CLOSED) {
864 if (!sock_flag(sk, SOCK_DONE)) {
865 /* This occurs when user tries to read
866 * from never connected socket.
880 if (signal_pending(current)) {
881 len = sock_intr_errno(timeo);
885 sk_wait_data(sk, &timeo, NULL);
890 else if (len < skb->len)
891 msg->msg_flags |= MSG_TRUNC;
893 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
894 /* Exception. Bailout! */
898 if (flags & MSG_TRUNC)
901 if (!(flags & MSG_PEEK))
910 EXPORT_SYMBOL_GPL(dccp_recvmsg);
912 int inet_dccp_listen(struct socket *sock, int backlog)
914 struct sock *sk = sock->sk;
915 unsigned char old_state;
921 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
924 old_state = sk->sk_state;
925 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
928 /* Really, if the socket is already in listen state
929 * we can only allow the backlog to be adjusted.
931 if (old_state != DCCP_LISTEN) {
933 * FIXME: here it probably should be sk->sk_prot->listen_start
934 * see tcp_listen_start
936 err = dccp_listen_start(sk, backlog);
940 sk->sk_max_ack_backlog = backlog;
948 EXPORT_SYMBOL_GPL(inet_dccp_listen);
950 static void dccp_terminate_connection(struct sock *sk)
952 u8 next_state = DCCP_CLOSED;
954 switch (sk->sk_state) {
955 case DCCP_PASSIVE_CLOSE:
956 case DCCP_PASSIVE_CLOSEREQ:
957 dccp_finish_passive_close(sk);
960 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
961 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
964 dccp_send_close(sk, 1);
966 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
967 !dccp_sk(sk)->dccps_server_timewait)
968 next_state = DCCP_ACTIVE_CLOSEREQ;
970 next_state = DCCP_CLOSING;
973 dccp_set_state(sk, next_state);
977 void dccp_close(struct sock *sk, long timeout)
979 struct dccp_sock *dp = dccp_sk(sk);
981 u32 data_was_unread = 0;
986 sk->sk_shutdown = SHUTDOWN_MASK;
988 if (sk->sk_state == DCCP_LISTEN) {
989 dccp_set_state(sk, DCCP_CLOSED);
992 inet_csk_listen_stop(sk);
994 goto adjudge_to_death;
997 sk_stop_timer(sk, &dp->dccps_xmit_timer);
1000 * We need to flush the recv. buffs. We do this only on the
1001 * descriptor close, not protocol-sourced closes, because the
1002 *reader process may not have drained the data yet!
1004 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1005 data_was_unread += skb->len;
1009 /* If socket has been already reset kill it. */
1010 if (sk->sk_state == DCCP_CLOSED)
1011 goto adjudge_to_death;
1013 if (data_was_unread) {
1014 /* Unread data was tossed, send an appropriate Reset Code */
1015 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1016 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1017 dccp_set_state(sk, DCCP_CLOSED);
1018 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1019 /* Check zero linger _after_ checking for unread data. */
1020 sk->sk_prot->disconnect(sk, 0);
1021 } else if (sk->sk_state != DCCP_CLOSED) {
1023 * Normal connection termination. May need to wait if there are
1024 * still packets in the TX queue that are delayed by the CCID.
1026 dccp_flush_write_queue(sk, &timeout);
1027 dccp_terminate_connection(sk);
1031 * Flush write queue. This may be necessary in several cases:
1032 * - we have been closed by the peer but still have application data;
1033 * - abortive termination (unread data or zero linger time),
1034 * - normal termination but queue could not be flushed within time limit
1036 __skb_queue_purge(&sk->sk_write_queue);
1038 sk_stream_wait_close(sk, timeout);
1041 state = sk->sk_state;
1046 * It is the last release_sock in its life. It will remove backlog.
1050 * Now socket is owned by kernel and we acquire BH lock
1051 * to finish close. No need to check for user refs.
1055 WARN_ON(sock_owned_by_user(sk));
1057 percpu_counter_inc(sk->sk_prot->orphan_count);
1059 /* Have we already been destroyed by a softirq or backlog? */
1060 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1063 if (sk->sk_state == DCCP_CLOSED)
1064 inet_csk_destroy_sock(sk);
1066 /* Otherwise, socket is reprieved until protocol close. */
1074 EXPORT_SYMBOL_GPL(dccp_close);
1076 void dccp_shutdown(struct sock *sk, int how)
1078 dccp_pr_debug("called shutdown(%x)\n", how);
1081 EXPORT_SYMBOL_GPL(dccp_shutdown);
1083 static inline int __init dccp_mib_init(void)
1085 dccp_statistics = alloc_percpu(struct dccp_mib);
1086 if (!dccp_statistics)
1091 static inline void dccp_mib_exit(void)
1093 free_percpu(dccp_statistics);
1096 static int thash_entries;
1097 module_param(thash_entries, int, 0444);
1098 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1100 #ifdef CONFIG_IP_DCCP_DEBUG
1102 module_param(dccp_debug, bool, 0644);
1103 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1105 EXPORT_SYMBOL_GPL(dccp_debug);
1108 static int __init dccp_init(void)
1111 int ehash_order, bhash_order, i;
1114 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1115 FIELD_SIZEOF(struct sk_buff, cb));
1116 rc = percpu_counter_init(&dccp_orphan_count, 0, GFP_KERNEL);
1120 inet_hashinfo_init(&dccp_hashinfo);
1121 dccp_hashinfo.bind_bucket_cachep =
1122 kmem_cache_create("dccp_bind_bucket",
1123 sizeof(struct inet_bind_bucket), 0,
1124 SLAB_HWCACHE_ALIGN, NULL);
1125 if (!dccp_hashinfo.bind_bucket_cachep)
1126 goto out_free_percpu;
1129 * Size and allocate the main established and bind bucket
1132 * The methodology is similar to that of the buffer cache.
1134 if (totalram_pages >= (128 * 1024))
1135 goal = totalram_pages >> (21 - PAGE_SHIFT);
1137 goal = totalram_pages >> (23 - PAGE_SHIFT);
1140 goal = (thash_entries *
1141 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1142 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1145 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1146 sizeof(struct inet_ehash_bucket);
1148 while (hash_size & (hash_size - 1))
1150 dccp_hashinfo.ehash_mask = hash_size - 1;
1151 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1152 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1153 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1155 if (!dccp_hashinfo.ehash) {
1156 DCCP_CRIT("Failed to allocate DCCP established hash table");
1157 goto out_free_bind_bucket_cachep;
1160 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1161 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1163 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1164 goto out_free_dccp_ehash;
1166 bhash_order = ehash_order;
1169 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1170 sizeof(struct inet_bind_hashbucket);
1171 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1174 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1175 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1176 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1178 if (!dccp_hashinfo.bhash) {
1179 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1180 goto out_free_dccp_locks;
1183 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1184 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1185 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1188 rc = dccp_mib_init();
1190 goto out_free_dccp_bhash;
1192 rc = dccp_ackvec_init();
1194 goto out_free_dccp_mib;
1196 rc = dccp_sysctl_init();
1198 goto out_ackvec_exit;
1200 rc = ccid_initialize_builtins();
1202 goto out_sysctl_exit;
1204 dccp_timestamping_init();
1214 out_free_dccp_bhash:
1215 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1216 out_free_dccp_locks:
1217 inet_ehash_locks_free(&dccp_hashinfo);
1218 out_free_dccp_ehash:
1219 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1220 out_free_bind_bucket_cachep:
1221 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1223 percpu_counter_destroy(&dccp_orphan_count);
1225 dccp_hashinfo.bhash = NULL;
1226 dccp_hashinfo.ehash = NULL;
1227 dccp_hashinfo.bind_bucket_cachep = NULL;
1231 static void __exit dccp_fini(void)
1233 ccid_cleanup_builtins();
1235 free_pages((unsigned long)dccp_hashinfo.bhash,
1236 get_order(dccp_hashinfo.bhash_size *
1237 sizeof(struct inet_bind_hashbucket)));
1238 free_pages((unsigned long)dccp_hashinfo.ehash,
1239 get_order((dccp_hashinfo.ehash_mask + 1) *
1240 sizeof(struct inet_ehash_bucket)));
1241 inet_ehash_locks_free(&dccp_hashinfo);
1242 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1245 percpu_counter_destroy(&dccp_orphan_count);
1248 module_init(dccp_init);
1249 module_exit(dccp_fini);
1251 MODULE_LICENSE("GPL");
1252 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1253 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");