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 icsk->icsk_rto = DCCP_TIMEOUT_INIT;
184 icsk->icsk_syn_retries = sysctl_dccp_request_retries;
185 sk->sk_state = DCCP_CLOSED;
186 sk->sk_write_space = dccp_write_space;
187 icsk->icsk_sync_mss = dccp_sync_mss;
188 dp->dccps_mss_cache = 536;
189 dp->dccps_rate_last = jiffies;
190 dp->dccps_role = DCCP_ROLE_UNDEFINED;
191 dp->dccps_service = DCCP_SERVICE_CODE_IS_ABSENT;
192 dp->dccps_l_ack_ratio = dp->dccps_r_ack_ratio = 1;
194 dccp_init_xmit_timers(sk);
196 INIT_LIST_HEAD(&dp->dccps_featneg);
197 /* control socket doesn't need feat nego */
198 if (likely(ctl_sock_initialized))
199 return dccp_feat_init(sk);
203 EXPORT_SYMBOL_GPL(dccp_init_sock);
205 void dccp_destroy_sock(struct sock *sk)
207 struct dccp_sock *dp = dccp_sk(sk);
210 * DCCP doesn't use sk_write_queue, just sk_send_head
211 * for retransmissions
213 if (sk->sk_send_head != NULL) {
214 kfree_skb(sk->sk_send_head);
215 sk->sk_send_head = NULL;
218 /* Clean up a referenced DCCP bind bucket. */
219 if (inet_csk(sk)->icsk_bind_hash != NULL)
222 kfree(dp->dccps_service_list);
223 dp->dccps_service_list = NULL;
225 if (dp->dccps_hc_rx_ackvec != NULL) {
226 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
227 dp->dccps_hc_rx_ackvec = NULL;
229 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
230 ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
231 dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
233 /* clean up feature negotiation state */
234 dccp_feat_list_purge(&dp->dccps_featneg);
237 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
239 static inline int dccp_listen_start(struct sock *sk, int backlog)
241 struct dccp_sock *dp = dccp_sk(sk);
243 dp->dccps_role = DCCP_ROLE_LISTEN;
244 /* do not start to listen if feature negotiation setup fails */
245 if (dccp_feat_finalise_settings(dp))
247 return inet_csk_listen_start(sk, backlog);
250 static inline int dccp_need_reset(int state)
252 return state != DCCP_CLOSED && state != DCCP_LISTEN &&
253 state != DCCP_REQUESTING;
256 int dccp_disconnect(struct sock *sk, int flags)
258 struct inet_connection_sock *icsk = inet_csk(sk);
259 struct inet_sock *inet = inet_sk(sk);
261 const int old_state = sk->sk_state;
263 if (old_state != DCCP_CLOSED)
264 dccp_set_state(sk, DCCP_CLOSED);
267 * This corresponds to the ABORT function of RFC793, sec. 3.8
268 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
270 if (old_state == DCCP_LISTEN) {
271 inet_csk_listen_stop(sk);
272 } else if (dccp_need_reset(old_state)) {
273 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
274 sk->sk_err = ECONNRESET;
275 } else if (old_state == DCCP_REQUESTING)
276 sk->sk_err = ECONNRESET;
278 dccp_clear_xmit_timers(sk);
280 __skb_queue_purge(&sk->sk_receive_queue);
281 __skb_queue_purge(&sk->sk_write_queue);
282 if (sk->sk_send_head != NULL) {
283 __kfree_skb(sk->sk_send_head);
284 sk->sk_send_head = NULL;
289 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
290 inet_reset_saddr(sk);
293 sock_reset_flag(sk, SOCK_DONE);
295 icsk->icsk_backoff = 0;
296 inet_csk_delack_init(sk);
299 WARN_ON(inet->num && !icsk->icsk_bind_hash);
301 sk->sk_error_report(sk);
305 EXPORT_SYMBOL_GPL(dccp_disconnect);
308 * Wait for a DCCP event.
310 * Note that we don't need to lock the socket, as the upper poll layers
311 * take care of normal races (between the test and the event) and we don't
312 * go look at any of the socket buffers directly.
314 unsigned int dccp_poll(struct file *file, struct socket *sock,
318 struct sock *sk = sock->sk;
320 poll_wait(file, sk->sk_sleep, wait);
321 if (sk->sk_state == DCCP_LISTEN)
322 return inet_csk_listen_poll(sk);
324 /* Socket is not locked. We are protected from async events
325 by poll logic and correct handling of state changes
326 made by another threads is impossible in any case.
333 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
335 if (sk->sk_shutdown & RCV_SHUTDOWN)
336 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
339 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
340 if (atomic_read(&sk->sk_rmem_alloc) > 0)
341 mask |= POLLIN | POLLRDNORM;
343 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
344 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
345 mask |= POLLOUT | POLLWRNORM;
346 } else { /* send SIGIO later */
347 set_bit(SOCK_ASYNC_NOSPACE,
348 &sk->sk_socket->flags);
349 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
351 /* Race breaker. If space is freed after
352 * wspace test but before the flags are set,
353 * IO signal will be lost.
355 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
356 mask |= POLLOUT | POLLWRNORM;
363 EXPORT_SYMBOL_GPL(dccp_poll);
365 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
371 if (sk->sk_state == DCCP_LISTEN)
377 unsigned long amount = 0;
379 skb = skb_peek(&sk->sk_receive_queue);
382 * We will only return the amount of this packet since
383 * that is all that will be read.
387 rc = put_user(amount, (int __user *)arg);
399 EXPORT_SYMBOL_GPL(dccp_ioctl);
401 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
402 char __user *optval, int optlen)
404 struct dccp_sock *dp = dccp_sk(sk);
405 struct dccp_service_list *sl = NULL;
407 if (service == DCCP_SERVICE_INVALID_VALUE ||
408 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
411 if (optlen > sizeof(service)) {
412 sl = kmalloc(optlen, GFP_KERNEL);
416 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
417 if (copy_from_user(sl->dccpsl_list,
418 optval + sizeof(service),
419 optlen - sizeof(service)) ||
420 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
427 dp->dccps_service = service;
429 kfree(dp->dccps_service_list);
431 dp->dccps_service_list = sl;
436 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
441 if (cscov < 0 || cscov > 15)
444 * Populate a list of permissible values, in the range cscov...15. This
445 * is necessary since feature negotiation of single values only works if
446 * both sides incidentally choose the same value. Since the list starts
447 * lowest-value first, negotiation will pick the smallest shared value.
453 list = kmalloc(len, GFP_KERNEL);
457 for (i = 0; i < len; i++)
460 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
464 dccp_sk(sk)->dccps_pcrlen = cscov;
466 dccp_sk(sk)->dccps_pcslen = cscov;
472 static int dccp_setsockopt_ccid(struct sock *sk, int type,
473 char __user *optval, int optlen)
478 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
481 val = kmalloc(optlen, GFP_KERNEL);
485 if (copy_from_user(val, optval, optlen)) {
491 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
492 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
494 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
495 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
502 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
503 char __user *optval, int optlen)
505 struct dccp_sock *dp = dccp_sk(sk);
509 case DCCP_SOCKOPT_PACKET_SIZE:
510 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
512 case DCCP_SOCKOPT_CHANGE_L:
513 case DCCP_SOCKOPT_CHANGE_R:
514 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
516 case DCCP_SOCKOPT_CCID:
517 case DCCP_SOCKOPT_RX_CCID:
518 case DCCP_SOCKOPT_TX_CCID:
519 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
522 if (optlen < (int)sizeof(int))
525 if (get_user(val, (int __user *)optval))
528 if (optname == DCCP_SOCKOPT_SERVICE)
529 return dccp_setsockopt_service(sk, val, optval, optlen);
533 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
534 if (dp->dccps_role != DCCP_ROLE_SERVER)
537 dp->dccps_server_timewait = (val != 0);
539 case DCCP_SOCKOPT_SEND_CSCOV:
540 err = dccp_setsockopt_cscov(sk, val, false);
542 case DCCP_SOCKOPT_RECV_CSCOV:
543 err = dccp_setsockopt_cscov(sk, val, true);
554 int dccp_setsockopt(struct sock *sk, int level, int optname,
555 char __user *optval, 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, 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;
653 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
654 len, (u32 __user *)optval, optlen);
656 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
657 len, (u32 __user *)optval, optlen);
663 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
669 int dccp_getsockopt(struct sock *sk, int level, int optname,
670 char __user *optval, int __user *optlen)
672 if (level != SOL_DCCP)
673 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
676 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
679 EXPORT_SYMBOL_GPL(dccp_getsockopt);
682 int compat_dccp_getsockopt(struct sock *sk, int level, int optname,
683 char __user *optval, int __user *optlen)
685 if (level != SOL_DCCP)
686 return inet_csk_compat_getsockopt(sk, level, optname,
688 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
691 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt);
694 int dccp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
697 const struct dccp_sock *dp = dccp_sk(sk);
698 const int flags = msg->msg_flags;
699 const int noblock = flags & MSG_DONTWAIT;
704 if (len > dp->dccps_mss_cache)
709 if (sysctl_dccp_tx_qlen &&
710 (sk->sk_write_queue.qlen >= sysctl_dccp_tx_qlen)) {
715 timeo = sock_sndtimeo(sk, noblock);
718 * We have to use sk_stream_wait_connect here to set sk_write_pending,
719 * so that the trick in dccp_rcv_request_sent_state_process.
721 /* Wait for a connection to finish. */
722 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
723 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
726 size = sk->sk_prot->max_header + len;
728 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
733 skb_reserve(skb, sk->sk_prot->max_header);
734 rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
738 skb_queue_tail(&sk->sk_write_queue, skb);
739 dccp_write_xmit(sk,0);
748 EXPORT_SYMBOL_GPL(dccp_sendmsg);
750 int dccp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
751 size_t len, int nonblock, int flags, int *addr_len)
753 const struct dccp_hdr *dh;
758 if (sk->sk_state == DCCP_LISTEN) {
763 timeo = sock_rcvtimeo(sk, nonblock);
766 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
769 goto verify_sock_status;
773 switch (dh->dccph_type) {
775 case DCCP_PKT_DATAACK:
779 case DCCP_PKT_CLOSEREQ:
780 if (!(flags & MSG_PEEK))
781 dccp_finish_passive_close(sk);
784 dccp_pr_debug("found fin (%s) ok!\n",
785 dccp_packet_name(dh->dccph_type));
789 dccp_pr_debug("packet_type=%s\n",
790 dccp_packet_name(dh->dccph_type));
791 sk_eat_skb(sk, skb, 0);
794 if (sock_flag(sk, SOCK_DONE)) {
800 len = sock_error(sk);
804 if (sk->sk_shutdown & RCV_SHUTDOWN) {
809 if (sk->sk_state == DCCP_CLOSED) {
810 if (!sock_flag(sk, SOCK_DONE)) {
811 /* This occurs when user tries to read
812 * from never connected socket.
826 if (signal_pending(current)) {
827 len = sock_intr_errno(timeo);
831 sk_wait_data(sk, &timeo);
836 else if (len < skb->len)
837 msg->msg_flags |= MSG_TRUNC;
839 if (skb_copy_datagram_iovec(skb, 0, msg->msg_iov, len)) {
840 /* Exception. Bailout! */
845 if (!(flags & MSG_PEEK))
846 sk_eat_skb(sk, skb, 0);
854 EXPORT_SYMBOL_GPL(dccp_recvmsg);
856 int inet_dccp_listen(struct socket *sock, int backlog)
858 struct sock *sk = sock->sk;
859 unsigned char old_state;
865 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
868 old_state = sk->sk_state;
869 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
872 /* Really, if the socket is already in listen state
873 * we can only allow the backlog to be adjusted.
875 if (old_state != DCCP_LISTEN) {
877 * FIXME: here it probably should be sk->sk_prot->listen_start
878 * see tcp_listen_start
880 err = dccp_listen_start(sk, backlog);
884 sk->sk_max_ack_backlog = backlog;
892 EXPORT_SYMBOL_GPL(inet_dccp_listen);
894 static void dccp_terminate_connection(struct sock *sk)
896 u8 next_state = DCCP_CLOSED;
898 switch (sk->sk_state) {
899 case DCCP_PASSIVE_CLOSE:
900 case DCCP_PASSIVE_CLOSEREQ:
901 dccp_finish_passive_close(sk);
904 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
905 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
908 dccp_send_close(sk, 1);
910 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
911 !dccp_sk(sk)->dccps_server_timewait)
912 next_state = DCCP_ACTIVE_CLOSEREQ;
914 next_state = DCCP_CLOSING;
917 dccp_set_state(sk, next_state);
921 void dccp_close(struct sock *sk, long timeout)
923 struct dccp_sock *dp = dccp_sk(sk);
925 u32 data_was_unread = 0;
930 sk->sk_shutdown = SHUTDOWN_MASK;
932 if (sk->sk_state == DCCP_LISTEN) {
933 dccp_set_state(sk, DCCP_CLOSED);
936 inet_csk_listen_stop(sk);
938 goto adjudge_to_death;
941 sk_stop_timer(sk, &dp->dccps_xmit_timer);
944 * We need to flush the recv. buffs. We do this only on the
945 * descriptor close, not protocol-sourced closes, because the
946 *reader process may not have drained the data yet!
948 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
949 data_was_unread += skb->len;
953 if (data_was_unread) {
954 /* Unread data was tossed, send an appropriate Reset Code */
955 DCCP_WARN("DCCP: ABORT -- %u bytes unread\n", data_was_unread);
956 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
957 dccp_set_state(sk, DCCP_CLOSED);
958 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
959 /* Check zero linger _after_ checking for unread data. */
960 sk->sk_prot->disconnect(sk, 0);
961 } else if (sk->sk_state != DCCP_CLOSED) {
962 dccp_terminate_connection(sk);
965 sk_stream_wait_close(sk, timeout);
968 state = sk->sk_state;
971 atomic_inc(sk->sk_prot->orphan_count);
974 * It is the last release_sock in its life. It will remove backlog.
978 * Now socket is owned by kernel and we acquire BH lock
979 * to finish close. No need to check for user refs.
983 WARN_ON(sock_owned_by_user(sk));
985 /* Have we already been destroyed by a softirq or backlog? */
986 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
989 if (sk->sk_state == DCCP_CLOSED)
990 inet_csk_destroy_sock(sk);
992 /* Otherwise, socket is reprieved until protocol close. */
1000 EXPORT_SYMBOL_GPL(dccp_close);
1002 void dccp_shutdown(struct sock *sk, int how)
1004 dccp_pr_debug("called shutdown(%x)\n", how);
1007 EXPORT_SYMBOL_GPL(dccp_shutdown);
1009 static inline int dccp_mib_init(void)
1011 return snmp_mib_init((void**)dccp_statistics, sizeof(struct dccp_mib));
1014 static inline void dccp_mib_exit(void)
1016 snmp_mib_free((void**)dccp_statistics);
1019 static int thash_entries;
1020 module_param(thash_entries, int, 0444);
1021 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1023 #ifdef CONFIG_IP_DCCP_DEBUG
1025 module_param(dccp_debug, bool, 0644);
1026 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1028 EXPORT_SYMBOL_GPL(dccp_debug);
1031 static int __init dccp_init(void)
1034 int ehash_order, bhash_order, i;
1037 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1038 FIELD_SIZEOF(struct sk_buff, cb));
1040 dccp_hashinfo.bind_bucket_cachep =
1041 kmem_cache_create("dccp_bind_bucket",
1042 sizeof(struct inet_bind_bucket), 0,
1043 SLAB_HWCACHE_ALIGN, NULL);
1044 if (!dccp_hashinfo.bind_bucket_cachep)
1048 * Size and allocate the main established and bind bucket
1051 * The methodology is similar to that of the buffer cache.
1053 if (num_physpages >= (128 * 1024))
1054 goal = num_physpages >> (21 - PAGE_SHIFT);
1056 goal = num_physpages >> (23 - PAGE_SHIFT);
1059 goal = (thash_entries *
1060 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1061 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1064 dccp_hashinfo.ehash_size = (1UL << ehash_order) * PAGE_SIZE /
1065 sizeof(struct inet_ehash_bucket);
1066 while (dccp_hashinfo.ehash_size &
1067 (dccp_hashinfo.ehash_size - 1))
1068 dccp_hashinfo.ehash_size--;
1069 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1070 __get_free_pages(GFP_ATOMIC, ehash_order);
1071 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1073 if (!dccp_hashinfo.ehash) {
1074 DCCP_CRIT("Failed to allocate DCCP established hash table");
1075 goto out_free_bind_bucket_cachep;
1078 for (i = 0; i < dccp_hashinfo.ehash_size; i++) {
1079 INIT_HLIST_HEAD(&dccp_hashinfo.ehash[i].chain);
1080 INIT_HLIST_HEAD(&dccp_hashinfo.ehash[i].twchain);
1083 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1084 goto out_free_dccp_ehash;
1086 bhash_order = ehash_order;
1089 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1090 sizeof(struct inet_bind_hashbucket);
1091 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1094 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1095 __get_free_pages(GFP_ATOMIC, bhash_order);
1096 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1098 if (!dccp_hashinfo.bhash) {
1099 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1100 goto out_free_dccp_locks;
1103 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1104 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1105 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1108 rc = dccp_mib_init();
1110 goto out_free_dccp_bhash;
1112 rc = dccp_ackvec_init();
1114 goto out_free_dccp_mib;
1116 rc = dccp_sysctl_init();
1118 goto out_ackvec_exit;
1120 dccp_timestamping_init();
1127 out_free_dccp_bhash:
1128 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1129 dccp_hashinfo.bhash = NULL;
1130 out_free_dccp_locks:
1131 inet_ehash_locks_free(&dccp_hashinfo);
1132 out_free_dccp_ehash:
1133 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1134 dccp_hashinfo.ehash = NULL;
1135 out_free_bind_bucket_cachep:
1136 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1137 dccp_hashinfo.bind_bucket_cachep = NULL;
1141 static void __exit dccp_fini(void)
1144 free_pages((unsigned long)dccp_hashinfo.bhash,
1145 get_order(dccp_hashinfo.bhash_size *
1146 sizeof(struct inet_bind_hashbucket)));
1147 free_pages((unsigned long)dccp_hashinfo.ehash,
1148 get_order(dccp_hashinfo.ehash_size *
1149 sizeof(struct inet_ehash_bucket)));
1150 inet_ehash_locks_free(&dccp_hashinfo);
1151 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1156 module_init(dccp_init);
1157 module_exit(dccp_fini);
1159 MODULE_LICENSE("GPL");
1160 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1161 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");