4 * An implementation of the DCCP protocol
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>
27 #include <net/inet_common.h>
31 #include <asm/ioctls.h>
32 #include <linux/spinlock.h>
33 #include <linux/timer.h>
34 #include <linux/delay.h>
35 #include <linux/poll.h>
41 #define CREATE_TRACE_POINTS
44 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
46 EXPORT_SYMBOL_GPL(dccp_statistics);
48 struct percpu_counter dccp_orphan_count;
49 EXPORT_SYMBOL_GPL(dccp_orphan_count);
51 struct inet_hashinfo dccp_hashinfo;
52 EXPORT_SYMBOL_GPL(dccp_hashinfo);
54 /* the maximum queue length for tx in packets. 0 is no limit */
55 int sysctl_dccp_tx_qlen __read_mostly = 5;
57 #ifdef CONFIG_IP_DCCP_DEBUG
58 static const char *dccp_state_name(const int state)
60 static const char *const dccp_state_names[] = {
62 [DCCP_REQUESTING] = "REQUESTING",
63 [DCCP_PARTOPEN] = "PARTOPEN",
64 [DCCP_LISTEN] = "LISTEN",
65 [DCCP_RESPOND] = "RESPOND",
66 [DCCP_CLOSING] = "CLOSING",
67 [DCCP_ACTIVE_CLOSEREQ] = "CLOSEREQ",
68 [DCCP_PASSIVE_CLOSE] = "PASSIVE_CLOSE",
69 [DCCP_PASSIVE_CLOSEREQ] = "PASSIVE_CLOSEREQ",
70 [DCCP_TIME_WAIT] = "TIME_WAIT",
71 [DCCP_CLOSED] = "CLOSED",
74 if (state >= DCCP_MAX_STATES)
75 return "INVALID STATE!";
77 return dccp_state_names[state];
81 void dccp_set_state(struct sock *sk, const int state)
83 const int oldstate = sk->sk_state;
85 dccp_pr_debug("%s(%p) %s --> %s\n", dccp_role(sk), sk,
86 dccp_state_name(oldstate), dccp_state_name(state));
87 WARN_ON(state == oldstate);
91 if (oldstate != DCCP_OPEN)
92 DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
93 /* Client retransmits all Confirm options until entering OPEN */
94 if (oldstate == DCCP_PARTOPEN)
95 dccp_feat_list_purge(&dccp_sk(sk)->dccps_featneg);
99 if (oldstate == DCCP_OPEN || oldstate == DCCP_ACTIVE_CLOSEREQ ||
100 oldstate == DCCP_CLOSING)
101 DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
103 sk->sk_prot->unhash(sk);
104 if (inet_csk(sk)->icsk_bind_hash != NULL &&
105 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
109 if (oldstate == DCCP_OPEN)
110 DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
113 /* Change state AFTER socket is unhashed to avoid closed
114 * socket sitting in hash tables.
116 inet_sk_set_state(sk, state);
119 EXPORT_SYMBOL_GPL(dccp_set_state);
121 static void dccp_finish_passive_close(struct sock *sk)
123 switch (sk->sk_state) {
124 case DCCP_PASSIVE_CLOSE:
125 /* Node (client or server) has received Close packet. */
126 dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
127 dccp_set_state(sk, DCCP_CLOSED);
129 case DCCP_PASSIVE_CLOSEREQ:
131 * Client received CloseReq. We set the `active' flag so that
132 * dccp_send_close() retransmits the Close as per RFC 4340, 8.3.
134 dccp_send_close(sk, 1);
135 dccp_set_state(sk, DCCP_CLOSING);
139 void dccp_done(struct sock *sk)
141 dccp_set_state(sk, DCCP_CLOSED);
142 dccp_clear_xmit_timers(sk);
144 sk->sk_shutdown = SHUTDOWN_MASK;
146 if (!sock_flag(sk, SOCK_DEAD))
147 sk->sk_state_change(sk);
149 inet_csk_destroy_sock(sk);
152 EXPORT_SYMBOL_GPL(dccp_done);
154 const char *dccp_packet_name(const int type)
156 static const char *const dccp_packet_names[] = {
157 [DCCP_PKT_REQUEST] = "REQUEST",
158 [DCCP_PKT_RESPONSE] = "RESPONSE",
159 [DCCP_PKT_DATA] = "DATA",
160 [DCCP_PKT_ACK] = "ACK",
161 [DCCP_PKT_DATAACK] = "DATAACK",
162 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
163 [DCCP_PKT_CLOSE] = "CLOSE",
164 [DCCP_PKT_RESET] = "RESET",
165 [DCCP_PKT_SYNC] = "SYNC",
166 [DCCP_PKT_SYNCACK] = "SYNCACK",
169 if (type >= DCCP_NR_PKT_TYPES)
172 return dccp_packet_names[type];
175 EXPORT_SYMBOL_GPL(dccp_packet_name);
177 static void dccp_sk_destruct(struct sock *sk)
179 struct dccp_sock *dp = dccp_sk(sk);
181 ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
182 dp->dccps_hc_tx_ccid = NULL;
183 inet_sock_destruct(sk);
186 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
188 struct dccp_sock *dp = dccp_sk(sk);
189 struct inet_connection_sock *icsk = inet_csk(sk);
191 icsk->icsk_rto = DCCP_TIMEOUT_INIT;
192 icsk->icsk_syn_retries = sysctl_dccp_request_retries;
193 sk->sk_state = DCCP_CLOSED;
194 sk->sk_write_space = dccp_write_space;
195 sk->sk_destruct = dccp_sk_destruct;
196 icsk->icsk_sync_mss = dccp_sync_mss;
197 dp->dccps_mss_cache = 536;
198 dp->dccps_rate_last = jiffies;
199 dp->dccps_role = DCCP_ROLE_UNDEFINED;
200 dp->dccps_service = DCCP_SERVICE_CODE_IS_ABSENT;
201 dp->dccps_tx_qlen = sysctl_dccp_tx_qlen;
203 dccp_init_xmit_timers(sk);
205 INIT_LIST_HEAD(&dp->dccps_featneg);
206 /* control socket doesn't need feat nego */
207 if (likely(ctl_sock_initialized))
208 return dccp_feat_init(sk);
212 EXPORT_SYMBOL_GPL(dccp_init_sock);
214 void dccp_destroy_sock(struct sock *sk)
216 struct dccp_sock *dp = dccp_sk(sk);
218 __skb_queue_purge(&sk->sk_write_queue);
219 if (sk->sk_send_head != NULL) {
220 kfree_skb(sk->sk_send_head);
221 sk->sk_send_head = NULL;
224 /* Clean up a referenced DCCP bind bucket. */
225 if (inet_csk(sk)->icsk_bind_hash != NULL)
228 kfree(dp->dccps_service_list);
229 dp->dccps_service_list = NULL;
231 if (dp->dccps_hc_rx_ackvec != NULL) {
232 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
233 dp->dccps_hc_rx_ackvec = NULL;
235 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
236 dp->dccps_hc_rx_ccid = NULL;
238 /* clean up feature negotiation state */
239 dccp_feat_list_purge(&dp->dccps_featneg);
242 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
244 static inline int dccp_listen_start(struct sock *sk, int backlog)
246 struct dccp_sock *dp = dccp_sk(sk);
248 dp->dccps_role = DCCP_ROLE_LISTEN;
249 /* do not start to listen if feature negotiation setup fails */
250 if (dccp_feat_finalise_settings(dp))
252 return inet_csk_listen_start(sk, backlog);
255 static inline int dccp_need_reset(int state)
257 return state != DCCP_CLOSED && state != DCCP_LISTEN &&
258 state != DCCP_REQUESTING;
261 int dccp_disconnect(struct sock *sk, int flags)
263 struct inet_connection_sock *icsk = inet_csk(sk);
264 struct inet_sock *inet = inet_sk(sk);
265 struct dccp_sock *dp = dccp_sk(sk);
267 const int old_state = sk->sk_state;
269 if (old_state != DCCP_CLOSED)
270 dccp_set_state(sk, DCCP_CLOSED);
273 * This corresponds to the ABORT function of RFC793, sec. 3.8
274 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
276 if (old_state == DCCP_LISTEN) {
277 inet_csk_listen_stop(sk);
278 } else if (dccp_need_reset(old_state)) {
279 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
280 sk->sk_err = ECONNRESET;
281 } else if (old_state == DCCP_REQUESTING)
282 sk->sk_err = ECONNRESET;
284 dccp_clear_xmit_timers(sk);
285 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
286 dp->dccps_hc_rx_ccid = NULL;
288 __skb_queue_purge(&sk->sk_receive_queue);
289 __skb_queue_purge(&sk->sk_write_queue);
290 if (sk->sk_send_head != NULL) {
291 __kfree_skb(sk->sk_send_head);
292 sk->sk_send_head = NULL;
295 inet->inet_dport = 0;
297 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
298 inet_reset_saddr(sk);
301 sock_reset_flag(sk, SOCK_DONE);
303 icsk->icsk_backoff = 0;
304 inet_csk_delack_init(sk);
307 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
309 sk->sk_error_report(sk);
313 EXPORT_SYMBOL_GPL(dccp_disconnect);
315 __poll_t dccp_poll_mask(struct socket *sock, __poll_t events)
318 struct sock *sk = sock->sk;
320 if (sk->sk_state == DCCP_LISTEN)
321 return inet_csk_listen_poll(sk);
323 /* Socket is not locked. We are protected from async events
324 by poll logic and correct handling of state changes
325 made by another threads is impossible in any case.
332 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
334 if (sk->sk_shutdown & RCV_SHUTDOWN)
335 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
338 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
339 if (atomic_read(&sk->sk_rmem_alloc) > 0)
340 mask |= EPOLLIN | EPOLLRDNORM;
342 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
343 if (sk_stream_is_writeable(sk)) {
344 mask |= EPOLLOUT | EPOLLWRNORM;
345 } else { /* send SIGIO later */
346 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
347 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
349 /* Race breaker. If space is freed after
350 * wspace test but before the flags are set,
351 * IO signal will be lost.
353 if (sk_stream_is_writeable(sk))
354 mask |= EPOLLOUT | EPOLLWRNORM;
361 EXPORT_SYMBOL_GPL(dccp_poll_mask);
363 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
369 if (sk->sk_state == DCCP_LISTEN)
375 unsigned long amount = 0;
377 skb = skb_peek(&sk->sk_receive_queue);
380 * We will only return the amount of this packet since
381 * that is all that will be read.
385 rc = put_user(amount, (int __user *)arg);
397 EXPORT_SYMBOL_GPL(dccp_ioctl);
399 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
400 char __user *optval, unsigned int optlen)
402 struct dccp_sock *dp = dccp_sk(sk);
403 struct dccp_service_list *sl = NULL;
405 if (service == DCCP_SERVICE_INVALID_VALUE ||
406 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
409 if (optlen > sizeof(service)) {
410 sl = kmalloc(optlen, GFP_KERNEL);
414 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
415 if (copy_from_user(sl->dccpsl_list,
416 optval + sizeof(service),
417 optlen - sizeof(service)) ||
418 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
425 dp->dccps_service = service;
427 kfree(dp->dccps_service_list);
429 dp->dccps_service_list = sl;
434 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
439 if (cscov < 0 || cscov > 15)
442 * Populate a list of permissible values, in the range cscov...15. This
443 * is necessary since feature negotiation of single values only works if
444 * both sides incidentally choose the same value. Since the list starts
445 * lowest-value first, negotiation will pick the smallest shared value.
451 list = kmalloc(len, GFP_KERNEL);
455 for (i = 0; i < len; i++)
458 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
462 dccp_sk(sk)->dccps_pcrlen = cscov;
464 dccp_sk(sk)->dccps_pcslen = cscov;
470 static int dccp_setsockopt_ccid(struct sock *sk, int type,
471 char __user *optval, unsigned int optlen)
476 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
479 val = memdup_user(optval, optlen);
484 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
485 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
487 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
488 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
495 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
496 char __user *optval, unsigned int optlen)
498 struct dccp_sock *dp = dccp_sk(sk);
502 case DCCP_SOCKOPT_PACKET_SIZE:
503 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
505 case DCCP_SOCKOPT_CHANGE_L:
506 case DCCP_SOCKOPT_CHANGE_R:
507 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
509 case DCCP_SOCKOPT_CCID:
510 case DCCP_SOCKOPT_RX_CCID:
511 case DCCP_SOCKOPT_TX_CCID:
512 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
515 if (optlen < (int)sizeof(int))
518 if (get_user(val, (int __user *)optval))
521 if (optname == DCCP_SOCKOPT_SERVICE)
522 return dccp_setsockopt_service(sk, val, optval, optlen);
526 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
527 if (dp->dccps_role != DCCP_ROLE_SERVER)
530 dp->dccps_server_timewait = (val != 0);
532 case DCCP_SOCKOPT_SEND_CSCOV:
533 err = dccp_setsockopt_cscov(sk, val, false);
535 case DCCP_SOCKOPT_RECV_CSCOV:
536 err = dccp_setsockopt_cscov(sk, val, true);
538 case DCCP_SOCKOPT_QPOLICY_ID:
539 if (sk->sk_state != DCCP_CLOSED)
541 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
544 dp->dccps_qpolicy = val;
546 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
550 dp->dccps_tx_qlen = val;
561 int dccp_setsockopt(struct sock *sk, int level, int optname,
562 char __user *optval, unsigned int optlen)
564 if (level != SOL_DCCP)
565 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
568 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
571 EXPORT_SYMBOL_GPL(dccp_setsockopt);
574 int compat_dccp_setsockopt(struct sock *sk, int level, int optname,
575 char __user *optval, unsigned int optlen)
577 if (level != SOL_DCCP)
578 return inet_csk_compat_setsockopt(sk, level, optname,
580 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
583 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt);
586 static int dccp_getsockopt_service(struct sock *sk, int len,
587 __be32 __user *optval,
590 const struct dccp_sock *dp = dccp_sk(sk);
591 const struct dccp_service_list *sl;
592 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
595 if ((sl = dp->dccps_service_list) != NULL) {
596 slen = sl->dccpsl_nr * sizeof(u32);
605 if (put_user(total_len, optlen) ||
606 put_user(dp->dccps_service, optval) ||
607 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
614 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
615 char __user *optval, int __user *optlen)
617 struct dccp_sock *dp;
620 if (get_user(len, optlen))
623 if (len < (int)sizeof(int))
629 case DCCP_SOCKOPT_PACKET_SIZE:
630 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
632 case DCCP_SOCKOPT_SERVICE:
633 return dccp_getsockopt_service(sk, len,
634 (__be32 __user *)optval, optlen);
635 case DCCP_SOCKOPT_GET_CUR_MPS:
636 val = dp->dccps_mss_cache;
638 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
639 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
640 case DCCP_SOCKOPT_TX_CCID:
641 val = ccid_get_current_tx_ccid(dp);
645 case DCCP_SOCKOPT_RX_CCID:
646 val = ccid_get_current_rx_ccid(dp);
650 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
651 val = dp->dccps_server_timewait;
653 case DCCP_SOCKOPT_SEND_CSCOV:
654 val = dp->dccps_pcslen;
656 case DCCP_SOCKOPT_RECV_CSCOV:
657 val = dp->dccps_pcrlen;
659 case DCCP_SOCKOPT_QPOLICY_ID:
660 val = dp->dccps_qpolicy;
662 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
663 val = dp->dccps_tx_qlen;
666 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
667 len, (u32 __user *)optval, optlen);
669 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
670 len, (u32 __user *)optval, optlen);
676 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
682 int 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(sk)->icsk_af_ops->getsockopt(sk, level,
689 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
692 EXPORT_SYMBOL_GPL(dccp_getsockopt);
695 int compat_dccp_getsockopt(struct sock *sk, int level, int optname,
696 char __user *optval, int __user *optlen)
698 if (level != SOL_DCCP)
699 return inet_csk_compat_getsockopt(sk, level, optname,
701 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
704 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt);
707 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
709 struct cmsghdr *cmsg;
712 * Assign an (opaque) qpolicy priority value to skb->priority.
714 * We are overloading this skb field for use with the qpolicy subystem.
715 * The skb->priority is normally used for the SO_PRIORITY option, which
716 * is initialised from sk_priority. Since the assignment of sk_priority
717 * to skb->priority happens later (on layer 3), we overload this field
718 * for use with queueing priorities as long as the skb is on layer 4.
719 * The default priority value (if nothing is set) is 0.
723 for_each_cmsghdr(cmsg, msg) {
724 if (!CMSG_OK(msg, cmsg))
727 if (cmsg->cmsg_level != SOL_DCCP)
730 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
731 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
734 switch (cmsg->cmsg_type) {
735 case DCCP_SCM_PRIORITY:
736 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
738 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
747 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
749 const struct dccp_sock *dp = dccp_sk(sk);
750 const int flags = msg->msg_flags;
751 const int noblock = flags & MSG_DONTWAIT;
756 trace_dccp_probe(sk, len);
758 if (len > dp->dccps_mss_cache)
763 if (dccp_qpolicy_full(sk)) {
768 timeo = sock_sndtimeo(sk, noblock);
771 * We have to use sk_stream_wait_connect here to set sk_write_pending,
772 * so that the trick in dccp_rcv_request_sent_state_process.
774 /* Wait for a connection to finish. */
775 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
776 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
779 size = sk->sk_prot->max_header + len;
781 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
786 if (sk->sk_state == DCCP_CLOSED) {
791 skb_reserve(skb, sk->sk_prot->max_header);
792 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
796 rc = dccp_msghdr_parse(msg, skb);
800 dccp_qpolicy_push(sk, skb);
802 * The xmit_timer is set if the TX CCID is rate-based and will expire
803 * when congestion control permits to release further packets into the
804 * network. Window-based CCIDs do not use this timer.
806 if (!timer_pending(&dp->dccps_xmit_timer))
816 EXPORT_SYMBOL_GPL(dccp_sendmsg);
818 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
819 int flags, int *addr_len)
821 const struct dccp_hdr *dh;
826 if (sk->sk_state == DCCP_LISTEN) {
831 timeo = sock_rcvtimeo(sk, nonblock);
834 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
837 goto verify_sock_status;
841 switch (dh->dccph_type) {
843 case DCCP_PKT_DATAACK:
847 case DCCP_PKT_CLOSEREQ:
848 if (!(flags & MSG_PEEK))
849 dccp_finish_passive_close(sk);
852 dccp_pr_debug("found fin (%s) ok!\n",
853 dccp_packet_name(dh->dccph_type));
857 dccp_pr_debug("packet_type=%s\n",
858 dccp_packet_name(dh->dccph_type));
862 if (sock_flag(sk, SOCK_DONE)) {
868 len = sock_error(sk);
872 if (sk->sk_shutdown & RCV_SHUTDOWN) {
877 if (sk->sk_state == DCCP_CLOSED) {
878 if (!sock_flag(sk, SOCK_DONE)) {
879 /* This occurs when user tries to read
880 * from never connected socket.
894 if (signal_pending(current)) {
895 len = sock_intr_errno(timeo);
899 sk_wait_data(sk, &timeo, NULL);
904 else if (len < skb->len)
905 msg->msg_flags |= MSG_TRUNC;
907 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
908 /* Exception. Bailout! */
912 if (flags & MSG_TRUNC)
915 if (!(flags & MSG_PEEK))
924 EXPORT_SYMBOL_GPL(dccp_recvmsg);
926 int inet_dccp_listen(struct socket *sock, int backlog)
928 struct sock *sk = sock->sk;
929 unsigned char old_state;
935 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
938 old_state = sk->sk_state;
939 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
942 /* Really, if the socket is already in listen state
943 * we can only allow the backlog to be adjusted.
945 if (old_state != DCCP_LISTEN) {
947 * FIXME: here it probably should be sk->sk_prot->listen_start
948 * see tcp_listen_start
950 err = dccp_listen_start(sk, backlog);
954 sk->sk_max_ack_backlog = backlog;
962 EXPORT_SYMBOL_GPL(inet_dccp_listen);
964 static void dccp_terminate_connection(struct sock *sk)
966 u8 next_state = DCCP_CLOSED;
968 switch (sk->sk_state) {
969 case DCCP_PASSIVE_CLOSE:
970 case DCCP_PASSIVE_CLOSEREQ:
971 dccp_finish_passive_close(sk);
974 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
975 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
978 dccp_send_close(sk, 1);
980 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
981 !dccp_sk(sk)->dccps_server_timewait)
982 next_state = DCCP_ACTIVE_CLOSEREQ;
984 next_state = DCCP_CLOSING;
987 dccp_set_state(sk, next_state);
991 void dccp_close(struct sock *sk, long timeout)
993 struct dccp_sock *dp = dccp_sk(sk);
995 u32 data_was_unread = 0;
1000 sk->sk_shutdown = SHUTDOWN_MASK;
1002 if (sk->sk_state == DCCP_LISTEN) {
1003 dccp_set_state(sk, DCCP_CLOSED);
1006 inet_csk_listen_stop(sk);
1008 goto adjudge_to_death;
1011 sk_stop_timer(sk, &dp->dccps_xmit_timer);
1014 * We need to flush the recv. buffs. We do this only on the
1015 * descriptor close, not protocol-sourced closes, because the
1016 *reader process may not have drained the data yet!
1018 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1019 data_was_unread += skb->len;
1023 /* If socket has been already reset kill it. */
1024 if (sk->sk_state == DCCP_CLOSED)
1025 goto adjudge_to_death;
1027 if (data_was_unread) {
1028 /* Unread data was tossed, send an appropriate Reset Code */
1029 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1030 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1031 dccp_set_state(sk, DCCP_CLOSED);
1032 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1033 /* Check zero linger _after_ checking for unread data. */
1034 sk->sk_prot->disconnect(sk, 0);
1035 } else if (sk->sk_state != DCCP_CLOSED) {
1037 * Normal connection termination. May need to wait if there are
1038 * still packets in the TX queue that are delayed by the CCID.
1040 dccp_flush_write_queue(sk, &timeout);
1041 dccp_terminate_connection(sk);
1045 * Flush write queue. This may be necessary in several cases:
1046 * - we have been closed by the peer but still have application data;
1047 * - abortive termination (unread data or zero linger time),
1048 * - normal termination but queue could not be flushed within time limit
1050 __skb_queue_purge(&sk->sk_write_queue);
1052 sk_stream_wait_close(sk, timeout);
1055 state = sk->sk_state;
1060 * It is the last release_sock in its life. It will remove backlog.
1064 * Now socket is owned by kernel and we acquire BH lock
1065 * to finish close. No need to check for user refs.
1069 WARN_ON(sock_owned_by_user(sk));
1071 percpu_counter_inc(sk->sk_prot->orphan_count);
1073 /* Have we already been destroyed by a softirq or backlog? */
1074 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1077 if (sk->sk_state == DCCP_CLOSED)
1078 inet_csk_destroy_sock(sk);
1080 /* Otherwise, socket is reprieved until protocol close. */
1088 EXPORT_SYMBOL_GPL(dccp_close);
1090 void dccp_shutdown(struct sock *sk, int how)
1092 dccp_pr_debug("called shutdown(%x)\n", how);
1095 EXPORT_SYMBOL_GPL(dccp_shutdown);
1097 static inline int __init dccp_mib_init(void)
1099 dccp_statistics = alloc_percpu(struct dccp_mib);
1100 if (!dccp_statistics)
1105 static inline void dccp_mib_exit(void)
1107 free_percpu(dccp_statistics);
1110 static int thash_entries;
1111 module_param(thash_entries, int, 0444);
1112 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1114 #ifdef CONFIG_IP_DCCP_DEBUG
1116 module_param(dccp_debug, bool, 0644);
1117 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1119 EXPORT_SYMBOL_GPL(dccp_debug);
1122 static int __init dccp_init(void)
1125 int ehash_order, bhash_order, i;
1128 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1129 FIELD_SIZEOF(struct sk_buff, cb));
1130 rc = percpu_counter_init(&dccp_orphan_count, 0, GFP_KERNEL);
1134 inet_hashinfo_init(&dccp_hashinfo);
1135 dccp_hashinfo.bind_bucket_cachep =
1136 kmem_cache_create("dccp_bind_bucket",
1137 sizeof(struct inet_bind_bucket), 0,
1138 SLAB_HWCACHE_ALIGN, NULL);
1139 if (!dccp_hashinfo.bind_bucket_cachep)
1140 goto out_free_percpu;
1143 * Size and allocate the main established and bind bucket
1146 * The methodology is similar to that of the buffer cache.
1148 if (totalram_pages >= (128 * 1024))
1149 goal = totalram_pages >> (21 - PAGE_SHIFT);
1151 goal = totalram_pages >> (23 - PAGE_SHIFT);
1154 goal = (thash_entries *
1155 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1156 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1159 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1160 sizeof(struct inet_ehash_bucket);
1162 while (hash_size & (hash_size - 1))
1164 dccp_hashinfo.ehash_mask = hash_size - 1;
1165 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1166 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1167 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1169 if (!dccp_hashinfo.ehash) {
1170 DCCP_CRIT("Failed to allocate DCCP established hash table");
1171 goto out_free_bind_bucket_cachep;
1174 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1175 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1177 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1178 goto out_free_dccp_ehash;
1180 bhash_order = ehash_order;
1183 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1184 sizeof(struct inet_bind_hashbucket);
1185 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1188 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1189 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1190 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1192 if (!dccp_hashinfo.bhash) {
1193 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1194 goto out_free_dccp_locks;
1197 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1198 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1199 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1202 rc = dccp_mib_init();
1204 goto out_free_dccp_bhash;
1206 rc = dccp_ackvec_init();
1208 goto out_free_dccp_mib;
1210 rc = dccp_sysctl_init();
1212 goto out_ackvec_exit;
1214 rc = ccid_initialize_builtins();
1216 goto out_sysctl_exit;
1218 dccp_timestamping_init();
1228 out_free_dccp_bhash:
1229 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1230 out_free_dccp_locks:
1231 inet_ehash_locks_free(&dccp_hashinfo);
1232 out_free_dccp_ehash:
1233 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1234 out_free_bind_bucket_cachep:
1235 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1237 percpu_counter_destroy(&dccp_orphan_count);
1239 dccp_hashinfo.bhash = NULL;
1240 dccp_hashinfo.ehash = NULL;
1241 dccp_hashinfo.bind_bucket_cachep = NULL;
1245 static void __exit dccp_fini(void)
1247 ccid_cleanup_builtins();
1249 free_pages((unsigned long)dccp_hashinfo.bhash,
1250 get_order(dccp_hashinfo.bhash_size *
1251 sizeof(struct inet_bind_hashbucket)));
1252 free_pages((unsigned long)dccp_hashinfo.ehash,
1253 get_order((dccp_hashinfo.ehash_mask + 1) *
1254 sizeof(struct inet_ehash_bucket)));
1255 inet_ehash_locks_free(&dccp_hashinfo);
1256 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1259 percpu_counter_destroy(&dccp_orphan_count);
1262 module_init(dccp_init);
1263 module_exit(dccp_fini);
1265 MODULE_LICENSE("GPL");
1267 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");