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 struct percpu_counter dccp_orphan_count;
46 EXPORT_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 static void dccp_sk_destruct(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;
180 inet_sock_destruct(sk);
183 int dccp_init_sock(struct sock *sk, const __u8 ctl_sock_initialized)
185 struct dccp_sock *dp = dccp_sk(sk);
186 struct inet_connection_sock *icsk = inet_csk(sk);
188 icsk->icsk_rto = DCCP_TIMEOUT_INIT;
189 icsk->icsk_syn_retries = sysctl_dccp_request_retries;
190 sk->sk_state = DCCP_CLOSED;
191 sk->sk_write_space = dccp_write_space;
192 sk->sk_destruct = dccp_sk_destruct;
193 icsk->icsk_sync_mss = dccp_sync_mss;
194 dp->dccps_mss_cache = 536;
195 dp->dccps_rate_last = jiffies;
196 dp->dccps_role = DCCP_ROLE_UNDEFINED;
197 dp->dccps_service = DCCP_SERVICE_CODE_IS_ABSENT;
198 dp->dccps_tx_qlen = sysctl_dccp_tx_qlen;
200 dccp_init_xmit_timers(sk);
202 INIT_LIST_HEAD(&dp->dccps_featneg);
203 /* control socket doesn't need feat nego */
204 if (likely(ctl_sock_initialized))
205 return dccp_feat_init(sk);
209 EXPORT_SYMBOL_GPL(dccp_init_sock);
211 void dccp_destroy_sock(struct sock *sk)
213 struct dccp_sock *dp = dccp_sk(sk);
215 __skb_queue_purge(&sk->sk_write_queue);
216 if (sk->sk_send_head != NULL) {
217 kfree_skb(sk->sk_send_head);
218 sk->sk_send_head = NULL;
221 /* Clean up a referenced DCCP bind bucket. */
222 if (inet_csk(sk)->icsk_bind_hash != NULL)
225 kfree(dp->dccps_service_list);
226 dp->dccps_service_list = NULL;
228 if (dp->dccps_hc_rx_ackvec != NULL) {
229 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
230 dp->dccps_hc_rx_ackvec = NULL;
232 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
233 dp->dccps_hc_rx_ccid = NULL;
235 /* clean up feature negotiation state */
236 dccp_feat_list_purge(&dp->dccps_featneg);
239 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
241 static inline int dccp_listen_start(struct sock *sk, int backlog)
243 struct dccp_sock *dp = dccp_sk(sk);
245 dp->dccps_role = DCCP_ROLE_LISTEN;
246 /* do not start to listen if feature negotiation setup fails */
247 if (dccp_feat_finalise_settings(dp))
249 return inet_csk_listen_start(sk, backlog);
252 static inline int dccp_need_reset(int state)
254 return state != DCCP_CLOSED && state != DCCP_LISTEN &&
255 state != DCCP_REQUESTING;
258 int dccp_disconnect(struct sock *sk, int flags)
260 struct inet_connection_sock *icsk = inet_csk(sk);
261 struct inet_sock *inet = inet_sk(sk);
262 struct dccp_sock *dp = dccp_sk(sk);
263 const int old_state = sk->sk_state;
265 if (old_state != DCCP_CLOSED)
266 dccp_set_state(sk, DCCP_CLOSED);
269 * This corresponds to the ABORT function of RFC793, sec. 3.8
270 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
272 if (old_state == DCCP_LISTEN) {
273 inet_csk_listen_stop(sk);
274 } else if (dccp_need_reset(old_state)) {
275 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
276 sk->sk_err = ECONNRESET;
277 } else if (old_state == DCCP_REQUESTING)
278 sk->sk_err = ECONNRESET;
280 dccp_clear_xmit_timers(sk);
281 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
282 dp->dccps_hc_rx_ccid = NULL;
284 __skb_queue_purge(&sk->sk_receive_queue);
285 __skb_queue_purge(&sk->sk_write_queue);
286 if (sk->sk_send_head != NULL) {
287 __kfree_skb(sk->sk_send_head);
288 sk->sk_send_head = NULL;
291 inet->inet_dport = 0;
293 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
294 inet_reset_saddr(sk);
297 sock_reset_flag(sk, SOCK_DONE);
299 icsk->icsk_backoff = 0;
300 inet_csk_delack_init(sk);
303 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
309 EXPORT_SYMBOL_GPL(dccp_disconnect);
312 * Wait for a DCCP event.
314 * Note that we don't need to lock the socket, as the upper poll layers
315 * take care of normal races (between the test and the event) and we don't
316 * go look at any of the socket buffers directly.
318 __poll_t dccp_poll(struct file *file, struct socket *sock,
322 struct sock *sk = sock->sk;
324 sock_poll_wait(file, sock, wait);
325 if (sk->sk_state == DCCP_LISTEN)
326 return inet_csk_listen_poll(sk);
328 /* Socket is not locked. We are protected from async events
329 by poll logic and correct handling of state changes
330 made by another threads is impossible in any case.
337 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
339 if (sk->sk_shutdown & RCV_SHUTDOWN)
340 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
343 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
344 if (atomic_read(&sk->sk_rmem_alloc) > 0)
345 mask |= EPOLLIN | EPOLLRDNORM;
347 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
348 if (sk_stream_is_writeable(sk)) {
349 mask |= EPOLLOUT | EPOLLWRNORM;
350 } else { /* send SIGIO later */
351 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
352 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
354 /* Race breaker. If space is freed after
355 * wspace test but before the flags are set,
356 * IO signal will be lost.
358 if (sk_stream_is_writeable(sk))
359 mask |= EPOLLOUT | EPOLLWRNORM;
366 EXPORT_SYMBOL_GPL(dccp_poll);
368 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
374 if (sk->sk_state == DCCP_LISTEN)
379 int amount = sk_wmem_alloc_get(sk);
380 /* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and
381 * always 0, comparably to UDP.
384 rc = put_user(amount, (int __user *)arg);
389 unsigned long amount = 0;
391 skb = skb_peek(&sk->sk_receive_queue);
394 * We will only return the amount of this packet since
395 * that is all that will be read.
399 rc = put_user(amount, (int __user *)arg);
411 EXPORT_SYMBOL_GPL(dccp_ioctl);
413 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
414 sockptr_t optval, unsigned int optlen)
416 struct dccp_sock *dp = dccp_sk(sk);
417 struct dccp_service_list *sl = NULL;
419 if (service == DCCP_SERVICE_INVALID_VALUE ||
420 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
423 if (optlen > sizeof(service)) {
424 sl = kmalloc(optlen, GFP_KERNEL);
428 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
429 if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
430 sizeof(service), optlen - sizeof(service)) ||
431 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
438 dp->dccps_service = service;
440 kfree(dp->dccps_service_list);
442 dp->dccps_service_list = sl;
447 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
452 if (cscov < 0 || cscov > 15)
455 * Populate a list of permissible values, in the range cscov...15. This
456 * is necessary since feature negotiation of single values only works if
457 * both sides incidentally choose the same value. Since the list starts
458 * lowest-value first, negotiation will pick the smallest shared value.
464 list = kmalloc(len, GFP_KERNEL);
468 for (i = 0; i < len; i++)
471 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
475 dccp_sk(sk)->dccps_pcrlen = cscov;
477 dccp_sk(sk)->dccps_pcslen = cscov;
483 static int dccp_setsockopt_ccid(struct sock *sk, int type,
484 sockptr_t optval, unsigned int optlen)
489 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
492 val = memdup_sockptr(optval, optlen);
497 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
498 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
500 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
501 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
508 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
509 sockptr_t optval, unsigned int optlen)
511 struct dccp_sock *dp = dccp_sk(sk);
515 case DCCP_SOCKOPT_PACKET_SIZE:
516 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
518 case DCCP_SOCKOPT_CHANGE_L:
519 case DCCP_SOCKOPT_CHANGE_R:
520 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
522 case DCCP_SOCKOPT_CCID:
523 case DCCP_SOCKOPT_RX_CCID:
524 case DCCP_SOCKOPT_TX_CCID:
525 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
528 if (optlen < (int)sizeof(int))
531 if (copy_from_sockptr(&val, optval, sizeof(int)))
534 if (optname == DCCP_SOCKOPT_SERVICE)
535 return dccp_setsockopt_service(sk, val, optval, optlen);
539 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
540 if (dp->dccps_role != DCCP_ROLE_SERVER)
543 dp->dccps_server_timewait = (val != 0);
545 case DCCP_SOCKOPT_SEND_CSCOV:
546 err = dccp_setsockopt_cscov(sk, val, false);
548 case DCCP_SOCKOPT_RECV_CSCOV:
549 err = dccp_setsockopt_cscov(sk, val, true);
551 case DCCP_SOCKOPT_QPOLICY_ID:
552 if (sk->sk_state != DCCP_CLOSED)
554 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
557 dp->dccps_qpolicy = val;
559 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
563 dp->dccps_tx_qlen = val;
574 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
577 if (level != SOL_DCCP)
578 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
581 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
584 EXPORT_SYMBOL_GPL(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);
694 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
696 struct cmsghdr *cmsg;
699 * Assign an (opaque) qpolicy priority value to skb->priority.
701 * We are overloading this skb field for use with the qpolicy subystem.
702 * The skb->priority is normally used for the SO_PRIORITY option, which
703 * is initialised from sk_priority. Since the assignment of sk_priority
704 * to skb->priority happens later (on layer 3), we overload this field
705 * for use with queueing priorities as long as the skb is on layer 4.
706 * The default priority value (if nothing is set) is 0.
710 for_each_cmsghdr(cmsg, msg) {
711 if (!CMSG_OK(msg, cmsg))
714 if (cmsg->cmsg_level != SOL_DCCP)
717 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
718 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
721 switch (cmsg->cmsg_type) {
722 case DCCP_SCM_PRIORITY:
723 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
725 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
734 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
736 const struct dccp_sock *dp = dccp_sk(sk);
737 const int flags = msg->msg_flags;
738 const int noblock = flags & MSG_DONTWAIT;
743 trace_dccp_probe(sk, len);
745 if (len > dp->dccps_mss_cache)
750 if (dccp_qpolicy_full(sk)) {
755 timeo = sock_sndtimeo(sk, noblock);
758 * We have to use sk_stream_wait_connect here to set sk_write_pending,
759 * so that the trick in dccp_rcv_request_sent_state_process.
761 /* Wait for a connection to finish. */
762 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
763 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
766 size = sk->sk_prot->max_header + len;
768 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
773 if (sk->sk_state == DCCP_CLOSED) {
778 skb_reserve(skb, sk->sk_prot->max_header);
779 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
783 rc = dccp_msghdr_parse(msg, skb);
787 dccp_qpolicy_push(sk, skb);
789 * The xmit_timer is set if the TX CCID is rate-based and will expire
790 * when congestion control permits to release further packets into the
791 * network. Window-based CCIDs do not use this timer.
793 if (!timer_pending(&dp->dccps_xmit_timer))
803 EXPORT_SYMBOL_GPL(dccp_sendmsg);
805 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
806 int flags, int *addr_len)
808 const struct dccp_hdr *dh;
813 if (sk->sk_state == DCCP_LISTEN) {
818 timeo = sock_rcvtimeo(sk, nonblock);
821 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
824 goto verify_sock_status;
828 switch (dh->dccph_type) {
830 case DCCP_PKT_DATAACK:
834 case DCCP_PKT_CLOSEREQ:
835 if (!(flags & MSG_PEEK))
836 dccp_finish_passive_close(sk);
839 dccp_pr_debug("found fin (%s) ok!\n",
840 dccp_packet_name(dh->dccph_type));
844 dccp_pr_debug("packet_type=%s\n",
845 dccp_packet_name(dh->dccph_type));
849 if (sock_flag(sk, SOCK_DONE)) {
855 len = sock_error(sk);
859 if (sk->sk_shutdown & RCV_SHUTDOWN) {
864 if (sk->sk_state == DCCP_CLOSED) {
865 if (!sock_flag(sk, SOCK_DONE)) {
866 /* This occurs when user tries to read
867 * from never connected socket.
881 if (signal_pending(current)) {
882 len = sock_intr_errno(timeo);
886 sk_wait_data(sk, &timeo, NULL);
891 else if (len < skb->len)
892 msg->msg_flags |= MSG_TRUNC;
894 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
895 /* Exception. Bailout! */
899 if (flags & MSG_TRUNC)
902 if (!(flags & MSG_PEEK))
911 EXPORT_SYMBOL_GPL(dccp_recvmsg);
913 int inet_dccp_listen(struct socket *sock, int backlog)
915 struct sock *sk = sock->sk;
916 unsigned char old_state;
922 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
925 old_state = sk->sk_state;
926 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
929 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
930 /* Really, if the socket is already in listen state
931 * we can only allow the backlog to be adjusted.
933 if (old_state != DCCP_LISTEN) {
935 * FIXME: here it probably should be sk->sk_prot->listen_start
936 * see tcp_listen_start
938 err = dccp_listen_start(sk, backlog);
949 EXPORT_SYMBOL_GPL(inet_dccp_listen);
951 static void dccp_terminate_connection(struct sock *sk)
953 u8 next_state = DCCP_CLOSED;
955 switch (sk->sk_state) {
956 case DCCP_PASSIVE_CLOSE:
957 case DCCP_PASSIVE_CLOSEREQ:
958 dccp_finish_passive_close(sk);
961 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
962 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
965 dccp_send_close(sk, 1);
967 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
968 !dccp_sk(sk)->dccps_server_timewait)
969 next_state = DCCP_ACTIVE_CLOSEREQ;
971 next_state = DCCP_CLOSING;
974 dccp_set_state(sk, next_state);
978 void dccp_close(struct sock *sk, long timeout)
980 struct dccp_sock *dp = dccp_sk(sk);
982 u32 data_was_unread = 0;
987 sk->sk_shutdown = SHUTDOWN_MASK;
989 if (sk->sk_state == DCCP_LISTEN) {
990 dccp_set_state(sk, DCCP_CLOSED);
993 inet_csk_listen_stop(sk);
995 goto adjudge_to_death;
998 sk_stop_timer(sk, &dp->dccps_xmit_timer);
1001 * We need to flush the recv. buffs. We do this only on the
1002 * descriptor close, not protocol-sourced closes, because the
1003 *reader process may not have drained the data yet!
1005 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1006 data_was_unread += skb->len;
1010 /* If socket has been already reset kill it. */
1011 if (sk->sk_state == DCCP_CLOSED)
1012 goto adjudge_to_death;
1014 if (data_was_unread) {
1015 /* Unread data was tossed, send an appropriate Reset Code */
1016 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1017 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1018 dccp_set_state(sk, DCCP_CLOSED);
1019 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1020 /* Check zero linger _after_ checking for unread data. */
1021 sk->sk_prot->disconnect(sk, 0);
1022 } else if (sk->sk_state != DCCP_CLOSED) {
1024 * Normal connection termination. May need to wait if there are
1025 * still packets in the TX queue that are delayed by the CCID.
1027 dccp_flush_write_queue(sk, &timeout);
1028 dccp_terminate_connection(sk);
1032 * Flush write queue. This may be necessary in several cases:
1033 * - we have been closed by the peer but still have application data;
1034 * - abortive termination (unread data or zero linger time),
1035 * - normal termination but queue could not be flushed within time limit
1037 __skb_queue_purge(&sk->sk_write_queue);
1039 sk_stream_wait_close(sk, timeout);
1042 state = sk->sk_state;
1047 * It is the last release_sock in its life. It will remove backlog.
1051 * Now socket is owned by kernel and we acquire BH lock
1052 * to finish close. No need to check for user refs.
1056 WARN_ON(sock_owned_by_user(sk));
1058 percpu_counter_inc(sk->sk_prot->orphan_count);
1060 /* Have we already been destroyed by a softirq or backlog? */
1061 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1064 if (sk->sk_state == DCCP_CLOSED)
1065 inet_csk_destroy_sock(sk);
1067 /* Otherwise, socket is reprieved until protocol close. */
1075 EXPORT_SYMBOL_GPL(dccp_close);
1077 void dccp_shutdown(struct sock *sk, int how)
1079 dccp_pr_debug("called shutdown(%x)\n", how);
1082 EXPORT_SYMBOL_GPL(dccp_shutdown);
1084 static inline int __init dccp_mib_init(void)
1086 dccp_statistics = alloc_percpu(struct dccp_mib);
1087 if (!dccp_statistics)
1092 static inline void dccp_mib_exit(void)
1094 free_percpu(dccp_statistics);
1097 static int thash_entries;
1098 module_param(thash_entries, int, 0444);
1099 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1101 #ifdef CONFIG_IP_DCCP_DEBUG
1103 module_param(dccp_debug, bool, 0644);
1104 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1106 EXPORT_SYMBOL_GPL(dccp_debug);
1109 static int __init dccp_init(void)
1112 unsigned long nr_pages = totalram_pages();
1113 int ehash_order, bhash_order, i;
1116 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1117 sizeof_field(struct sk_buff, cb));
1118 rc = percpu_counter_init(&dccp_orphan_count, 0, GFP_KERNEL);
1121 inet_hashinfo_init(&dccp_hashinfo);
1122 rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
1124 goto out_free_percpu;
1126 dccp_hashinfo.bind_bucket_cachep =
1127 kmem_cache_create("dccp_bind_bucket",
1128 sizeof(struct inet_bind_bucket), 0,
1129 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1130 if (!dccp_hashinfo.bind_bucket_cachep)
1131 goto out_free_hashinfo2;
1134 * Size and allocate the main established and bind bucket
1137 * The methodology is similar to that of the buffer cache.
1139 if (nr_pages >= (128 * 1024))
1140 goal = nr_pages >> (21 - PAGE_SHIFT);
1142 goal = nr_pages >> (23 - PAGE_SHIFT);
1145 goal = (thash_entries *
1146 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1147 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1150 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1151 sizeof(struct inet_ehash_bucket);
1153 while (hash_size & (hash_size - 1))
1155 dccp_hashinfo.ehash_mask = hash_size - 1;
1156 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1157 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1158 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1160 if (!dccp_hashinfo.ehash) {
1161 DCCP_CRIT("Failed to allocate DCCP established hash table");
1162 goto out_free_bind_bucket_cachep;
1165 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1166 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1168 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1169 goto out_free_dccp_ehash;
1171 bhash_order = ehash_order;
1174 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1175 sizeof(struct inet_bind_hashbucket);
1176 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1179 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1180 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1181 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1183 if (!dccp_hashinfo.bhash) {
1184 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1185 goto out_free_dccp_locks;
1188 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1189 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1190 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1193 rc = dccp_mib_init();
1195 goto out_free_dccp_bhash;
1197 rc = dccp_ackvec_init();
1199 goto out_free_dccp_mib;
1201 rc = dccp_sysctl_init();
1203 goto out_ackvec_exit;
1205 rc = ccid_initialize_builtins();
1207 goto out_sysctl_exit;
1209 dccp_timestamping_init();
1219 out_free_dccp_bhash:
1220 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1221 out_free_dccp_locks:
1222 inet_ehash_locks_free(&dccp_hashinfo);
1223 out_free_dccp_ehash:
1224 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1225 out_free_bind_bucket_cachep:
1226 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1228 inet_hashinfo2_free_mod(&dccp_hashinfo);
1230 percpu_counter_destroy(&dccp_orphan_count);
1232 dccp_hashinfo.bhash = NULL;
1233 dccp_hashinfo.ehash = NULL;
1234 dccp_hashinfo.bind_bucket_cachep = NULL;
1238 static void __exit dccp_fini(void)
1240 ccid_cleanup_builtins();
1242 free_pages((unsigned long)dccp_hashinfo.bhash,
1243 get_order(dccp_hashinfo.bhash_size *
1244 sizeof(struct inet_bind_hashbucket)));
1245 free_pages((unsigned long)dccp_hashinfo.ehash,
1246 get_order((dccp_hashinfo.ehash_mask + 1) *
1247 sizeof(struct inet_ehash_bucket)));
1248 inet_ehash_locks_free(&dccp_hashinfo);
1249 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1252 inet_hashinfo2_free_mod(&dccp_hashinfo);
1253 percpu_counter_destroy(&dccp_orphan_count);
1256 module_init(dccp_init);
1257 module_exit(dccp_fini);
1259 MODULE_LICENSE("GPL");
1260 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1261 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");