1 // SPDX-License-Identifier: GPL-2.0-only
5 * An implementation of the DCCP protocol
6 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
9 #include <linux/dccp.h>
10 #include <linux/module.h>
11 #include <linux/types.h>
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/skbuff.h>
15 #include <linux/netdevice.h>
17 #include <linux/if_arp.h>
18 #include <linux/init.h>
19 #include <linux/random.h>
20 #include <linux/slab.h>
21 #include <net/checksum.h>
23 #include <net/inet_sock.h>
24 #include <net/inet_common.h>
28 #include <asm/ioctls.h>
29 #include <linux/spinlock.h>
30 #include <linux/timer.h>
31 #include <linux/delay.h>
32 #include <linux/poll.h>
38 #define CREATE_TRACE_POINTS
41 DEFINE_SNMP_STAT(struct dccp_mib, dccp_statistics) __read_mostly;
43 EXPORT_SYMBOL_GPL(dccp_statistics);
45 DEFINE_PER_CPU(unsigned int, dccp_orphan_count);
46 EXPORT_PER_CPU_SYMBOL_GPL(dccp_orphan_count);
48 struct inet_hashinfo dccp_hashinfo;
49 EXPORT_SYMBOL_GPL(dccp_hashinfo);
51 /* the maximum queue length for tx in packets. 0 is no limit */
52 int sysctl_dccp_tx_qlen __read_mostly = 5;
54 #ifdef CONFIG_IP_DCCP_DEBUG
55 static const char *dccp_state_name(const int state)
57 static const char *const dccp_state_names[] = {
59 [DCCP_REQUESTING] = "REQUESTING",
60 [DCCP_PARTOPEN] = "PARTOPEN",
61 [DCCP_LISTEN] = "LISTEN",
62 [DCCP_RESPOND] = "RESPOND",
63 [DCCP_CLOSING] = "CLOSING",
64 [DCCP_ACTIVE_CLOSEREQ] = "CLOSEREQ",
65 [DCCP_PASSIVE_CLOSE] = "PASSIVE_CLOSE",
66 [DCCP_PASSIVE_CLOSEREQ] = "PASSIVE_CLOSEREQ",
67 [DCCP_TIME_WAIT] = "TIME_WAIT",
68 [DCCP_CLOSED] = "CLOSED",
71 if (state >= DCCP_MAX_STATES)
72 return "INVALID STATE!";
74 return dccp_state_names[state];
78 void dccp_set_state(struct sock *sk, const int state)
80 const int oldstate = sk->sk_state;
82 dccp_pr_debug("%s(%p) %s --> %s\n", dccp_role(sk), sk,
83 dccp_state_name(oldstate), dccp_state_name(state));
84 WARN_ON(state == oldstate);
88 if (oldstate != DCCP_OPEN)
89 DCCP_INC_STATS(DCCP_MIB_CURRESTAB);
90 /* Client retransmits all Confirm options until entering OPEN */
91 if (oldstate == DCCP_PARTOPEN)
92 dccp_feat_list_purge(&dccp_sk(sk)->dccps_featneg);
96 if (oldstate == DCCP_OPEN || oldstate == DCCP_ACTIVE_CLOSEREQ ||
97 oldstate == DCCP_CLOSING)
98 DCCP_INC_STATS(DCCP_MIB_ESTABRESETS);
100 sk->sk_prot->unhash(sk);
101 if (inet_csk(sk)->icsk_bind_hash != NULL &&
102 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
106 if (oldstate == DCCP_OPEN)
107 DCCP_DEC_STATS(DCCP_MIB_CURRESTAB);
110 /* Change state AFTER socket is unhashed to avoid closed
111 * socket sitting in hash tables.
113 inet_sk_set_state(sk, state);
116 EXPORT_SYMBOL_GPL(dccp_set_state);
118 static void dccp_finish_passive_close(struct sock *sk)
120 switch (sk->sk_state) {
121 case DCCP_PASSIVE_CLOSE:
122 /* Node (client or server) has received Close packet. */
123 dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
124 dccp_set_state(sk, DCCP_CLOSED);
126 case DCCP_PASSIVE_CLOSEREQ:
128 * Client received CloseReq. We set the `active' flag so that
129 * dccp_send_close() retransmits the Close as per RFC 4340, 8.3.
131 dccp_send_close(sk, 1);
132 dccp_set_state(sk, DCCP_CLOSING);
136 void dccp_done(struct sock *sk)
138 dccp_set_state(sk, DCCP_CLOSED);
139 dccp_clear_xmit_timers(sk);
141 sk->sk_shutdown = SHUTDOWN_MASK;
143 if (!sock_flag(sk, SOCK_DEAD))
144 sk->sk_state_change(sk);
146 inet_csk_destroy_sock(sk);
149 EXPORT_SYMBOL_GPL(dccp_done);
151 const char *dccp_packet_name(const int type)
153 static const char *const dccp_packet_names[] = {
154 [DCCP_PKT_REQUEST] = "REQUEST",
155 [DCCP_PKT_RESPONSE] = "RESPONSE",
156 [DCCP_PKT_DATA] = "DATA",
157 [DCCP_PKT_ACK] = "ACK",
158 [DCCP_PKT_DATAACK] = "DATAACK",
159 [DCCP_PKT_CLOSEREQ] = "CLOSEREQ",
160 [DCCP_PKT_CLOSE] = "CLOSE",
161 [DCCP_PKT_RESET] = "RESET",
162 [DCCP_PKT_SYNC] = "SYNC",
163 [DCCP_PKT_SYNCACK] = "SYNCACK",
166 if (type >= DCCP_NR_PKT_TYPES)
169 return dccp_packet_names[type];
172 EXPORT_SYMBOL_GPL(dccp_packet_name);
174 void dccp_destruct_common(struct sock *sk)
176 struct dccp_sock *dp = dccp_sk(sk);
178 ccid_hc_tx_delete(dp->dccps_hc_tx_ccid, sk);
179 dp->dccps_hc_tx_ccid = NULL;
181 EXPORT_SYMBOL_GPL(dccp_destruct_common);
183 static void dccp_sk_destruct(struct sock *sk)
185 dccp_destruct_common(sk);
186 inet_sock_destruct(sk);
189 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
191 struct dccp_sock *dp = dccp_sk(sk);
192 struct inet_connection_sock *icsk = inet_csk(sk);
194 icsk->icsk_rto = DCCP_TIMEOUT_INIT;
195 icsk->icsk_syn_retries = sysctl_dccp_request_retries;
196 sk->sk_state = DCCP_CLOSED;
197 sk->sk_write_space = dccp_write_space;
198 sk->sk_destruct = dccp_sk_destruct;
199 icsk->icsk_sync_mss = dccp_sync_mss;
200 dp->dccps_mss_cache = 536;
201 dp->dccps_rate_last = jiffies;
202 dp->dccps_role = DCCP_ROLE_UNDEFINED;
203 dp->dccps_service = DCCP_SERVICE_CODE_IS_ABSENT;
204 dp->dccps_tx_qlen = sysctl_dccp_tx_qlen;
206 dccp_init_xmit_timers(sk);
208 INIT_LIST_HEAD(&dp->dccps_featneg);
209 /* control socket doesn't need feat nego */
210 if (likely(ctl_sock_initialized))
211 return dccp_feat_init(sk);
215 EXPORT_SYMBOL_GPL(dccp_init_sock);
217 void dccp_destroy_sock(struct sock *sk)
219 struct dccp_sock *dp = dccp_sk(sk);
221 __skb_queue_purge(&sk->sk_write_queue);
222 if (sk->sk_send_head != NULL) {
223 kfree_skb(sk->sk_send_head);
224 sk->sk_send_head = NULL;
227 /* Clean up a referenced DCCP bind bucket. */
228 if (inet_csk(sk)->icsk_bind_hash != NULL)
231 kfree(dp->dccps_service_list);
232 dp->dccps_service_list = NULL;
234 if (dp->dccps_hc_rx_ackvec != NULL) {
235 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
236 dp->dccps_hc_rx_ackvec = NULL;
238 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
239 dp->dccps_hc_rx_ccid = NULL;
241 /* clean up feature negotiation state */
242 dccp_feat_list_purge(&dp->dccps_featneg);
245 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
247 static inline int dccp_listen_start(struct sock *sk, int backlog)
249 struct dccp_sock *dp = dccp_sk(sk);
251 dp->dccps_role = DCCP_ROLE_LISTEN;
252 /* do not start to listen if feature negotiation setup fails */
253 if (dccp_feat_finalise_settings(dp))
255 return inet_csk_listen_start(sk, backlog);
258 static inline int dccp_need_reset(int state)
260 return state != DCCP_CLOSED && state != DCCP_LISTEN &&
261 state != DCCP_REQUESTING;
264 int dccp_disconnect(struct sock *sk, int flags)
266 struct inet_connection_sock *icsk = inet_csk(sk);
267 struct inet_sock *inet = inet_sk(sk);
268 struct dccp_sock *dp = dccp_sk(sk);
269 const int old_state = sk->sk_state;
271 if (old_state != DCCP_CLOSED)
272 dccp_set_state(sk, DCCP_CLOSED);
275 * This corresponds to the ABORT function of RFC793, sec. 3.8
276 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
278 if (old_state == DCCP_LISTEN) {
279 inet_csk_listen_stop(sk);
280 } else if (dccp_need_reset(old_state)) {
281 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
282 sk->sk_err = ECONNRESET;
283 } else if (old_state == DCCP_REQUESTING)
284 sk->sk_err = ECONNRESET;
286 dccp_clear_xmit_timers(sk);
287 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
288 dp->dccps_hc_rx_ccid = NULL;
290 __skb_queue_purge(&sk->sk_receive_queue);
291 __skb_queue_purge(&sk->sk_write_queue);
292 if (sk->sk_send_head != NULL) {
293 __kfree_skb(sk->sk_send_head);
294 sk->sk_send_head = NULL;
297 inet->inet_dport = 0;
299 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
300 inet_reset_saddr(sk);
303 sock_reset_flag(sk, SOCK_DONE);
305 icsk->icsk_backoff = 0;
306 inet_csk_delack_init(sk);
309 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
311 sk->sk_error_report(sk);
315 EXPORT_SYMBOL_GPL(dccp_disconnect);
318 * Wait for a DCCP event.
320 * Note that we don't need to lock the socket, as the upper poll layers
321 * take care of normal races (between the test and the event) and we don't
322 * go look at any of the socket buffers directly.
324 __poll_t dccp_poll(struct file *file, struct socket *sock,
327 struct sock *sk = sock->sk;
332 sock_poll_wait(file, sock, wait);
334 state = inet_sk_state_load(sk);
335 if (state == DCCP_LISTEN)
336 return inet_csk_listen_poll(sk);
338 /* Socket is not locked. We are protected from async events
339 by poll logic and correct handling of state changes
340 made by another threads is impossible in any case.
344 if (READ_ONCE(sk->sk_err))
346 shutdown = READ_ONCE(sk->sk_shutdown);
348 if (shutdown == SHUTDOWN_MASK || state == DCCP_CLOSED)
350 if (shutdown & RCV_SHUTDOWN)
351 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
354 if ((1 << state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
355 if (atomic_read(&sk->sk_rmem_alloc) > 0)
356 mask |= EPOLLIN | EPOLLRDNORM;
358 if (!(shutdown & SEND_SHUTDOWN)) {
359 if (sk_stream_is_writeable(sk)) {
360 mask |= EPOLLOUT | EPOLLWRNORM;
361 } else { /* send SIGIO later */
362 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
363 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
365 /* Race breaker. If space is freed after
366 * wspace test but before the flags are set,
367 * IO signal will be lost.
369 if (sk_stream_is_writeable(sk))
370 mask |= EPOLLOUT | EPOLLWRNORM;
376 EXPORT_SYMBOL_GPL(dccp_poll);
378 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
384 if (sk->sk_state == DCCP_LISTEN)
389 int amount = sk_wmem_alloc_get(sk);
390 /* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and
391 * always 0, comparably to UDP.
394 rc = put_user(amount, (int __user *)arg);
399 unsigned long amount = 0;
401 skb = skb_peek(&sk->sk_receive_queue);
404 * We will only return the amount of this packet since
405 * that is all that will be read.
409 rc = put_user(amount, (int __user *)arg);
421 EXPORT_SYMBOL_GPL(dccp_ioctl);
423 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
424 sockptr_t optval, unsigned int optlen)
426 struct dccp_sock *dp = dccp_sk(sk);
427 struct dccp_service_list *sl = NULL;
429 if (service == DCCP_SERVICE_INVALID_VALUE ||
430 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
433 if (optlen > sizeof(service)) {
434 sl = kmalloc(optlen, GFP_KERNEL);
438 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
439 if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
440 sizeof(service), optlen - sizeof(service)) ||
441 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
448 dp->dccps_service = service;
450 kfree(dp->dccps_service_list);
452 dp->dccps_service_list = sl;
457 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
462 if (cscov < 0 || cscov > 15)
465 * Populate a list of permissible values, in the range cscov...15. This
466 * is necessary since feature negotiation of single values only works if
467 * both sides incidentally choose the same value. Since the list starts
468 * lowest-value first, negotiation will pick the smallest shared value.
474 list = kmalloc(len, GFP_KERNEL);
478 for (i = 0; i < len; i++)
481 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
485 dccp_sk(sk)->dccps_pcrlen = cscov;
487 dccp_sk(sk)->dccps_pcslen = cscov;
493 static int dccp_setsockopt_ccid(struct sock *sk, int type,
494 sockptr_t optval, unsigned int optlen)
499 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
502 val = memdup_sockptr(optval, optlen);
507 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
508 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
510 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
511 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
518 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
519 sockptr_t optval, unsigned int optlen)
521 struct dccp_sock *dp = dccp_sk(sk);
525 case DCCP_SOCKOPT_PACKET_SIZE:
526 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
528 case DCCP_SOCKOPT_CHANGE_L:
529 case DCCP_SOCKOPT_CHANGE_R:
530 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
532 case DCCP_SOCKOPT_CCID:
533 case DCCP_SOCKOPT_RX_CCID:
534 case DCCP_SOCKOPT_TX_CCID:
535 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
538 if (optlen < (int)sizeof(int))
541 if (copy_from_sockptr(&val, optval, sizeof(int)))
544 if (optname == DCCP_SOCKOPT_SERVICE)
545 return dccp_setsockopt_service(sk, val, optval, optlen);
549 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
550 if (dp->dccps_role != DCCP_ROLE_SERVER)
553 dp->dccps_server_timewait = (val != 0);
555 case DCCP_SOCKOPT_SEND_CSCOV:
556 err = dccp_setsockopt_cscov(sk, val, false);
558 case DCCP_SOCKOPT_RECV_CSCOV:
559 err = dccp_setsockopt_cscov(sk, val, true);
561 case DCCP_SOCKOPT_QPOLICY_ID:
562 if (sk->sk_state != DCCP_CLOSED)
564 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
567 dp->dccps_qpolicy = val;
569 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
573 dp->dccps_tx_qlen = val;
584 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
587 if (level != SOL_DCCP)
588 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
591 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
594 EXPORT_SYMBOL_GPL(dccp_setsockopt);
596 static int dccp_getsockopt_service(struct sock *sk, int len,
597 __be32 __user *optval,
600 const struct dccp_sock *dp = dccp_sk(sk);
601 const struct dccp_service_list *sl;
602 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
605 if ((sl = dp->dccps_service_list) != NULL) {
606 slen = sl->dccpsl_nr * sizeof(u32);
615 if (put_user(total_len, optlen) ||
616 put_user(dp->dccps_service, optval) ||
617 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
624 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
625 char __user *optval, int __user *optlen)
627 struct dccp_sock *dp;
630 if (get_user(len, optlen))
633 if (len < (int)sizeof(int))
639 case DCCP_SOCKOPT_PACKET_SIZE:
640 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
642 case DCCP_SOCKOPT_SERVICE:
643 return dccp_getsockopt_service(sk, len,
644 (__be32 __user *)optval, optlen);
645 case DCCP_SOCKOPT_GET_CUR_MPS:
646 val = READ_ONCE(dp->dccps_mss_cache);
648 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
649 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
650 case DCCP_SOCKOPT_TX_CCID:
651 val = ccid_get_current_tx_ccid(dp);
655 case DCCP_SOCKOPT_RX_CCID:
656 val = ccid_get_current_rx_ccid(dp);
660 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
661 val = dp->dccps_server_timewait;
663 case DCCP_SOCKOPT_SEND_CSCOV:
664 val = dp->dccps_pcslen;
666 case DCCP_SOCKOPT_RECV_CSCOV:
667 val = dp->dccps_pcrlen;
669 case DCCP_SOCKOPT_QPOLICY_ID:
670 val = dp->dccps_qpolicy;
672 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
673 val = dp->dccps_tx_qlen;
676 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
677 len, (u32 __user *)optval, optlen);
679 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
680 len, (u32 __user *)optval, optlen);
686 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
692 int dccp_getsockopt(struct sock *sk, int level, int optname,
693 char __user *optval, int __user *optlen)
695 if (level != SOL_DCCP)
696 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
699 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
702 EXPORT_SYMBOL_GPL(dccp_getsockopt);
704 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
706 struct cmsghdr *cmsg;
709 * Assign an (opaque) qpolicy priority value to skb->priority.
711 * We are overloading this skb field for use with the qpolicy subystem.
712 * The skb->priority is normally used for the SO_PRIORITY option, which
713 * is initialised from sk_priority. Since the assignment of sk_priority
714 * to skb->priority happens later (on layer 3), we overload this field
715 * for use with queueing priorities as long as the skb is on layer 4.
716 * The default priority value (if nothing is set) is 0.
720 for_each_cmsghdr(cmsg, msg) {
721 if (!CMSG_OK(msg, cmsg))
724 if (cmsg->cmsg_level != SOL_DCCP)
727 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
728 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
731 switch (cmsg->cmsg_type) {
732 case DCCP_SCM_PRIORITY:
733 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
735 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
744 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
746 const struct dccp_sock *dp = dccp_sk(sk);
747 const int flags = msg->msg_flags;
748 const int noblock = flags & MSG_DONTWAIT;
753 trace_dccp_probe(sk, len);
755 if (len > READ_ONCE(dp->dccps_mss_cache))
760 timeo = sock_sndtimeo(sk, noblock);
763 * We have to use sk_stream_wait_connect here to set sk_write_pending,
764 * so that the trick in dccp_rcv_request_sent_state_process.
766 /* Wait for a connection to finish. */
767 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
768 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
771 size = sk->sk_prot->max_header + len;
773 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
778 if (dccp_qpolicy_full(sk)) {
783 if (sk->sk_state == DCCP_CLOSED) {
788 /* We need to check dccps_mss_cache after socket is locked. */
789 if (len > dp->dccps_mss_cache) {
794 skb_reserve(skb, sk->sk_prot->max_header);
795 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
799 rc = dccp_msghdr_parse(msg, skb);
803 dccp_qpolicy_push(sk, skb);
805 * The xmit_timer is set if the TX CCID is rate-based and will expire
806 * when congestion control permits to release further packets into the
807 * network. Window-based CCIDs do not use this timer.
809 if (!timer_pending(&dp->dccps_xmit_timer))
819 EXPORT_SYMBOL_GPL(dccp_sendmsg);
821 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
822 int flags, int *addr_len)
824 const struct dccp_hdr *dh;
829 if (sk->sk_state == DCCP_LISTEN) {
834 timeo = sock_rcvtimeo(sk, nonblock);
837 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
840 goto verify_sock_status;
844 switch (dh->dccph_type) {
846 case DCCP_PKT_DATAACK:
850 case DCCP_PKT_CLOSEREQ:
851 if (!(flags & MSG_PEEK))
852 dccp_finish_passive_close(sk);
855 dccp_pr_debug("found fin (%s) ok!\n",
856 dccp_packet_name(dh->dccph_type));
860 dccp_pr_debug("packet_type=%s\n",
861 dccp_packet_name(dh->dccph_type));
865 if (sock_flag(sk, SOCK_DONE)) {
871 len = sock_error(sk);
875 if (sk->sk_shutdown & RCV_SHUTDOWN) {
880 if (sk->sk_state == DCCP_CLOSED) {
881 if (!sock_flag(sk, SOCK_DONE)) {
882 /* This occurs when user tries to read
883 * from never connected socket.
897 if (signal_pending(current)) {
898 len = sock_intr_errno(timeo);
902 sk_wait_data(sk, &timeo, NULL);
907 else if (len < skb->len)
908 msg->msg_flags |= MSG_TRUNC;
910 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
911 /* Exception. Bailout! */
915 if (flags & MSG_TRUNC)
918 if (!(flags & MSG_PEEK))
927 EXPORT_SYMBOL_GPL(dccp_recvmsg);
929 int inet_dccp_listen(struct socket *sock, int backlog)
931 struct sock *sk = sock->sk;
932 unsigned char old_state;
938 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
941 old_state = sk->sk_state;
942 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
945 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
946 /* Really, if the socket is already in listen state
947 * we can only allow the backlog to be adjusted.
949 if (old_state != DCCP_LISTEN) {
951 * FIXME: here it probably should be sk->sk_prot->listen_start
952 * see tcp_listen_start
954 err = dccp_listen_start(sk, backlog);
965 EXPORT_SYMBOL_GPL(inet_dccp_listen);
967 static void dccp_terminate_connection(struct sock *sk)
969 u8 next_state = DCCP_CLOSED;
971 switch (sk->sk_state) {
972 case DCCP_PASSIVE_CLOSE:
973 case DCCP_PASSIVE_CLOSEREQ:
974 dccp_finish_passive_close(sk);
977 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
978 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
981 dccp_send_close(sk, 1);
983 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
984 !dccp_sk(sk)->dccps_server_timewait)
985 next_state = DCCP_ACTIVE_CLOSEREQ;
987 next_state = DCCP_CLOSING;
990 dccp_set_state(sk, next_state);
994 void dccp_close(struct sock *sk, long timeout)
996 struct dccp_sock *dp = dccp_sk(sk);
998 u32 data_was_unread = 0;
1003 sk->sk_shutdown = SHUTDOWN_MASK;
1005 if (sk->sk_state == DCCP_LISTEN) {
1006 dccp_set_state(sk, DCCP_CLOSED);
1009 inet_csk_listen_stop(sk);
1011 goto adjudge_to_death;
1014 sk_stop_timer(sk, &dp->dccps_xmit_timer);
1017 * We need to flush the recv. buffs. We do this only on the
1018 * descriptor close, not protocol-sourced closes, because the
1019 *reader process may not have drained the data yet!
1021 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1022 data_was_unread += skb->len;
1026 /* If socket has been already reset kill it. */
1027 if (sk->sk_state == DCCP_CLOSED)
1028 goto adjudge_to_death;
1030 if (data_was_unread) {
1031 /* Unread data was tossed, send an appropriate Reset Code */
1032 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1033 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1034 dccp_set_state(sk, DCCP_CLOSED);
1035 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1036 /* Check zero linger _after_ checking for unread data. */
1037 sk->sk_prot->disconnect(sk, 0);
1038 } else if (sk->sk_state != DCCP_CLOSED) {
1040 * Normal connection termination. May need to wait if there are
1041 * still packets in the TX queue that are delayed by the CCID.
1043 dccp_flush_write_queue(sk, &timeout);
1044 dccp_terminate_connection(sk);
1048 * Flush write queue. This may be necessary in several cases:
1049 * - we have been closed by the peer but still have application data;
1050 * - abortive termination (unread data or zero linger time),
1051 * - normal termination but queue could not be flushed within time limit
1053 __skb_queue_purge(&sk->sk_write_queue);
1055 sk_stream_wait_close(sk, timeout);
1058 state = sk->sk_state;
1063 * It is the last release_sock in its life. It will remove backlog.
1067 * Now socket is owned by kernel and we acquire BH lock
1068 * to finish close. No need to check for user refs.
1072 WARN_ON(sock_owned_by_user(sk));
1074 this_cpu_inc(dccp_orphan_count);
1076 /* Have we already been destroyed by a softirq or backlog? */
1077 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1080 if (sk->sk_state == DCCP_CLOSED)
1081 inet_csk_destroy_sock(sk);
1083 /* Otherwise, socket is reprieved until protocol close. */
1091 EXPORT_SYMBOL_GPL(dccp_close);
1093 void dccp_shutdown(struct sock *sk, int how)
1095 dccp_pr_debug("called shutdown(%x)\n", how);
1098 EXPORT_SYMBOL_GPL(dccp_shutdown);
1100 static inline int __init dccp_mib_init(void)
1102 dccp_statistics = alloc_percpu(struct dccp_mib);
1103 if (!dccp_statistics)
1108 static inline void dccp_mib_exit(void)
1110 free_percpu(dccp_statistics);
1113 static int thash_entries;
1114 module_param(thash_entries, int, 0444);
1115 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1117 #ifdef CONFIG_IP_DCCP_DEBUG
1119 module_param(dccp_debug, bool, 0644);
1120 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1122 EXPORT_SYMBOL_GPL(dccp_debug);
1125 static int __init dccp_init(void)
1128 unsigned long nr_pages = totalram_pages();
1129 int ehash_order, bhash_order, i;
1132 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1133 sizeof_field(struct sk_buff, cb));
1134 inet_hashinfo_init(&dccp_hashinfo);
1135 rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
1139 dccp_hashinfo.bind_bucket_cachep =
1140 kmem_cache_create("dccp_bind_bucket",
1141 sizeof(struct inet_bind_bucket), 0,
1142 SLAB_HWCACHE_ALIGN, NULL);
1143 if (!dccp_hashinfo.bind_bucket_cachep)
1144 goto out_free_hashinfo2;
1147 * Size and allocate the main established and bind bucket
1150 * The methodology is similar to that of the buffer cache.
1152 if (nr_pages >= (128 * 1024))
1153 goal = nr_pages >> (21 - PAGE_SHIFT);
1155 goal = nr_pages >> (23 - PAGE_SHIFT);
1158 goal = (thash_entries *
1159 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1160 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1163 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1164 sizeof(struct inet_ehash_bucket);
1166 while (hash_size & (hash_size - 1))
1168 dccp_hashinfo.ehash_mask = hash_size - 1;
1169 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1170 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1171 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1173 if (!dccp_hashinfo.ehash) {
1174 DCCP_CRIT("Failed to allocate DCCP established hash table");
1175 goto out_free_bind_bucket_cachep;
1178 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1179 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1181 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1182 goto out_free_dccp_ehash;
1184 bhash_order = ehash_order;
1187 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1188 sizeof(struct inet_bind_hashbucket);
1189 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1192 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1193 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1194 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1196 if (!dccp_hashinfo.bhash) {
1197 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1198 goto out_free_dccp_locks;
1201 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1202 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1203 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1206 rc = dccp_mib_init();
1208 goto out_free_dccp_bhash;
1210 rc = dccp_ackvec_init();
1212 goto out_free_dccp_mib;
1214 rc = dccp_sysctl_init();
1216 goto out_ackvec_exit;
1218 rc = ccid_initialize_builtins();
1220 goto out_sysctl_exit;
1222 dccp_timestamping_init();
1232 out_free_dccp_bhash:
1233 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1234 out_free_dccp_locks:
1235 inet_ehash_locks_free(&dccp_hashinfo);
1236 out_free_dccp_ehash:
1237 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1238 out_free_bind_bucket_cachep:
1239 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1241 inet_hashinfo2_free_mod(&dccp_hashinfo);
1243 dccp_hashinfo.bhash = NULL;
1244 dccp_hashinfo.ehash = NULL;
1245 dccp_hashinfo.bind_bucket_cachep = NULL;
1249 static void __exit dccp_fini(void)
1251 ccid_cleanup_builtins();
1253 free_pages((unsigned long)dccp_hashinfo.bhash,
1254 get_order(dccp_hashinfo.bhash_size *
1255 sizeof(struct inet_bind_hashbucket)));
1256 free_pages((unsigned long)dccp_hashinfo.ehash,
1257 get_order((dccp_hashinfo.ehash_mask + 1) *
1258 sizeof(struct inet_ehash_bucket)));
1259 inet_ehash_locks_free(&dccp_hashinfo);
1260 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1263 inet_hashinfo2_free_mod(&dccp_hashinfo);
1266 module_init(dccp_init);
1267 module_exit(dccp_fini);
1269 MODULE_LICENSE("GPL");
1270 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1271 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");