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 pr_warn_once("DCCP is deprecated and scheduled to be removed in 2025, "
195 "please contact the netdev mailing list\n");
197 icsk->icsk_rto = DCCP_TIMEOUT_INIT;
198 icsk->icsk_syn_retries = sysctl_dccp_request_retries;
199 sk->sk_state = DCCP_CLOSED;
200 sk->sk_write_space = dccp_write_space;
201 sk->sk_destruct = dccp_sk_destruct;
202 icsk->icsk_sync_mss = dccp_sync_mss;
203 dp->dccps_mss_cache = 536;
204 dp->dccps_rate_last = jiffies;
205 dp->dccps_role = DCCP_ROLE_UNDEFINED;
206 dp->dccps_service = DCCP_SERVICE_CODE_IS_ABSENT;
207 dp->dccps_tx_qlen = sysctl_dccp_tx_qlen;
209 dccp_init_xmit_timers(sk);
211 INIT_LIST_HEAD(&dp->dccps_featneg);
212 /* control socket doesn't need feat nego */
213 if (likely(ctl_sock_initialized))
214 return dccp_feat_init(sk);
218 EXPORT_SYMBOL_GPL(dccp_init_sock);
220 void dccp_destroy_sock(struct sock *sk)
222 struct dccp_sock *dp = dccp_sk(sk);
224 __skb_queue_purge(&sk->sk_write_queue);
225 if (sk->sk_send_head != NULL) {
226 kfree_skb(sk->sk_send_head);
227 sk->sk_send_head = NULL;
230 /* Clean up a referenced DCCP bind bucket. */
231 if (inet_csk(sk)->icsk_bind_hash != NULL)
234 kfree(dp->dccps_service_list);
235 dp->dccps_service_list = NULL;
237 if (dp->dccps_hc_rx_ackvec != NULL) {
238 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
239 dp->dccps_hc_rx_ackvec = NULL;
241 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
242 dp->dccps_hc_rx_ccid = NULL;
244 /* clean up feature negotiation state */
245 dccp_feat_list_purge(&dp->dccps_featneg);
248 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
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);
260 struct dccp_sock *dp = dccp_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);
279 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
280 dp->dccps_hc_rx_ccid = NULL;
282 __skb_queue_purge(&sk->sk_receive_queue);
283 __skb_queue_purge(&sk->sk_write_queue);
284 if (sk->sk_send_head != NULL) {
285 __kfree_skb(sk->sk_send_head);
286 sk->sk_send_head = NULL;
289 inet->inet_dport = 0;
291 inet_bhash2_reset_saddr(sk);
294 sock_reset_flag(sk, SOCK_DONE);
296 icsk->icsk_backoff = 0;
297 inet_csk_delack_init(sk);
300 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
306 EXPORT_SYMBOL_GPL(dccp_disconnect);
309 * Wait for a DCCP event.
311 * Note that we don't need to lock the socket, as the upper poll layers
312 * take care of normal races (between the test and the event) and we don't
313 * go look at any of the socket buffers directly.
315 __poll_t dccp_poll(struct file *file, struct socket *sock,
318 struct sock *sk = sock->sk;
323 sock_poll_wait(file, sock, wait);
325 state = inet_sk_state_load(sk);
326 if (state == DCCP_LISTEN)
327 return inet_csk_listen_poll(sk);
329 /* Socket is not locked. We are protected from async events
330 by poll logic and correct handling of state changes
331 made by another threads is impossible in any case.
335 if (READ_ONCE(sk->sk_err))
337 shutdown = READ_ONCE(sk->sk_shutdown);
339 if (shutdown == SHUTDOWN_MASK || state == DCCP_CLOSED)
341 if (shutdown & RCV_SHUTDOWN)
342 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
345 if ((1 << state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
346 if (atomic_read(&sk->sk_rmem_alloc) > 0)
347 mask |= EPOLLIN | EPOLLRDNORM;
349 if (!(shutdown & SEND_SHUTDOWN)) {
350 if (sk_stream_is_writeable(sk)) {
351 mask |= EPOLLOUT | EPOLLWRNORM;
352 } else { /* send SIGIO later */
353 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
354 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
356 /* Race breaker. If space is freed after
357 * wspace test but before the flags are set,
358 * IO signal will be lost.
360 if (sk_stream_is_writeable(sk))
361 mask |= EPOLLOUT | EPOLLWRNORM;
367 EXPORT_SYMBOL_GPL(dccp_poll);
369 int dccp_ioctl(struct sock *sk, int cmd, int *karg)
375 if (sk->sk_state == DCCP_LISTEN)
380 *karg = sk_wmem_alloc_get(sk);
381 /* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and
382 * always 0, comparably to UDP.
392 skb = skb_peek(&sk->sk_receive_queue);
395 * We will only return the amount of this packet since
396 * that is all that will be read.
412 EXPORT_SYMBOL_GPL(dccp_ioctl);
414 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
415 sockptr_t optval, unsigned int optlen)
417 struct dccp_sock *dp = dccp_sk(sk);
418 struct dccp_service_list *sl = NULL;
420 if (service == DCCP_SERVICE_INVALID_VALUE ||
421 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
424 if (optlen > sizeof(service)) {
425 sl = kmalloc(optlen, GFP_KERNEL);
429 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
430 if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
431 sizeof(service), optlen - sizeof(service)) ||
432 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
439 dp->dccps_service = service;
441 kfree(dp->dccps_service_list);
443 dp->dccps_service_list = sl;
448 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
453 if (cscov < 0 || cscov > 15)
456 * Populate a list of permissible values, in the range cscov...15. This
457 * is necessary since feature negotiation of single values only works if
458 * both sides incidentally choose the same value. Since the list starts
459 * lowest-value first, negotiation will pick the smallest shared value.
465 list = kmalloc(len, GFP_KERNEL);
469 for (i = 0; i < len; i++)
472 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
476 dccp_sk(sk)->dccps_pcrlen = cscov;
478 dccp_sk(sk)->dccps_pcslen = cscov;
484 static int dccp_setsockopt_ccid(struct sock *sk, int type,
485 sockptr_t optval, unsigned int optlen)
490 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
493 val = memdup_sockptr(optval, optlen);
498 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
499 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
501 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
502 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
509 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
510 sockptr_t optval, unsigned int optlen)
512 struct dccp_sock *dp = dccp_sk(sk);
516 case DCCP_SOCKOPT_PACKET_SIZE:
517 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
519 case DCCP_SOCKOPT_CHANGE_L:
520 case DCCP_SOCKOPT_CHANGE_R:
521 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
523 case DCCP_SOCKOPT_CCID:
524 case DCCP_SOCKOPT_RX_CCID:
525 case DCCP_SOCKOPT_TX_CCID:
526 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
529 if (optlen < (int)sizeof(int))
532 if (copy_from_sockptr(&val, optval, sizeof(int)))
535 if (optname == DCCP_SOCKOPT_SERVICE)
536 return dccp_setsockopt_service(sk, val, optval, optlen);
540 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
541 if (dp->dccps_role != DCCP_ROLE_SERVER)
544 dp->dccps_server_timewait = (val != 0);
546 case DCCP_SOCKOPT_SEND_CSCOV:
547 err = dccp_setsockopt_cscov(sk, val, false);
549 case DCCP_SOCKOPT_RECV_CSCOV:
550 err = dccp_setsockopt_cscov(sk, val, true);
552 case DCCP_SOCKOPT_QPOLICY_ID:
553 if (sk->sk_state != DCCP_CLOSED)
555 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
558 dp->dccps_qpolicy = val;
560 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
564 dp->dccps_tx_qlen = val;
575 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
578 if (level != SOL_DCCP)
579 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
582 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
585 EXPORT_SYMBOL_GPL(dccp_setsockopt);
587 static int dccp_getsockopt_service(struct sock *sk, int len,
588 __be32 __user *optval,
591 const struct dccp_sock *dp = dccp_sk(sk);
592 const struct dccp_service_list *sl;
593 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
596 if ((sl = dp->dccps_service_list) != NULL) {
597 slen = sl->dccpsl_nr * sizeof(u32);
606 if (put_user(total_len, optlen) ||
607 put_user(dp->dccps_service, optval) ||
608 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
615 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
616 char __user *optval, int __user *optlen)
618 struct dccp_sock *dp;
621 if (get_user(len, optlen))
624 if (len < (int)sizeof(int))
630 case DCCP_SOCKOPT_PACKET_SIZE:
631 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
633 case DCCP_SOCKOPT_SERVICE:
634 return dccp_getsockopt_service(sk, len,
635 (__be32 __user *)optval, optlen);
636 case DCCP_SOCKOPT_GET_CUR_MPS:
637 val = READ_ONCE(dp->dccps_mss_cache);
639 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
640 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
641 case DCCP_SOCKOPT_TX_CCID:
642 val = ccid_get_current_tx_ccid(dp);
646 case DCCP_SOCKOPT_RX_CCID:
647 val = ccid_get_current_rx_ccid(dp);
651 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
652 val = dp->dccps_server_timewait;
654 case DCCP_SOCKOPT_SEND_CSCOV:
655 val = dp->dccps_pcslen;
657 case DCCP_SOCKOPT_RECV_CSCOV:
658 val = dp->dccps_pcrlen;
660 case DCCP_SOCKOPT_QPOLICY_ID:
661 val = dp->dccps_qpolicy;
663 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
664 val = dp->dccps_tx_qlen;
667 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
668 len, (u32 __user *)optval, optlen);
670 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
671 len, (u32 __user *)optval, optlen);
677 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
683 int dccp_getsockopt(struct sock *sk, int level, int optname,
684 char __user *optval, int __user *optlen)
686 if (level != SOL_DCCP)
687 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
690 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
693 EXPORT_SYMBOL_GPL(dccp_getsockopt);
695 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
697 struct cmsghdr *cmsg;
700 * Assign an (opaque) qpolicy priority value to skb->priority.
702 * We are overloading this skb field for use with the qpolicy subystem.
703 * The skb->priority is normally used for the SO_PRIORITY option, which
704 * is initialised from sk_priority. Since the assignment of sk_priority
705 * to skb->priority happens later (on layer 3), we overload this field
706 * for use with queueing priorities as long as the skb is on layer 4.
707 * The default priority value (if nothing is set) is 0.
711 for_each_cmsghdr(cmsg, msg) {
712 if (!CMSG_OK(msg, cmsg))
715 if (cmsg->cmsg_level != SOL_DCCP)
718 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
719 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
722 switch (cmsg->cmsg_type) {
723 case DCCP_SCM_PRIORITY:
724 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
726 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
735 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
737 const struct dccp_sock *dp = dccp_sk(sk);
738 const int flags = msg->msg_flags;
739 const int noblock = flags & MSG_DONTWAIT;
744 trace_dccp_probe(sk, len);
746 if (len > READ_ONCE(dp->dccps_mss_cache))
751 timeo = sock_sndtimeo(sk, noblock);
754 * We have to use sk_stream_wait_connect here to set sk_write_pending,
755 * so that the trick in dccp_rcv_request_sent_state_process.
757 /* Wait for a connection to finish. */
758 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
759 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
762 size = sk->sk_prot->max_header + len;
764 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
769 if (dccp_qpolicy_full(sk)) {
774 if (sk->sk_state == DCCP_CLOSED) {
779 /* We need to check dccps_mss_cache after socket is locked. */
780 if (len > dp->dccps_mss_cache) {
785 skb_reserve(skb, sk->sk_prot->max_header);
786 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
790 rc = dccp_msghdr_parse(msg, skb);
794 dccp_qpolicy_push(sk, skb);
796 * The xmit_timer is set if the TX CCID is rate-based and will expire
797 * when congestion control permits to release further packets into the
798 * network. Window-based CCIDs do not use this timer.
800 if (!timer_pending(&dp->dccps_xmit_timer))
810 EXPORT_SYMBOL_GPL(dccp_sendmsg);
812 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
815 const struct dccp_hdr *dh;
820 if (sk->sk_state == DCCP_LISTEN) {
825 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
828 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
831 goto verify_sock_status;
835 switch (dh->dccph_type) {
837 case DCCP_PKT_DATAACK:
841 case DCCP_PKT_CLOSEREQ:
842 if (!(flags & MSG_PEEK))
843 dccp_finish_passive_close(sk);
846 dccp_pr_debug("found fin (%s) ok!\n",
847 dccp_packet_name(dh->dccph_type));
851 dccp_pr_debug("packet_type=%s\n",
852 dccp_packet_name(dh->dccph_type));
856 if (sock_flag(sk, SOCK_DONE)) {
862 len = sock_error(sk);
866 if (sk->sk_shutdown & RCV_SHUTDOWN) {
871 if (sk->sk_state == DCCP_CLOSED) {
872 if (!sock_flag(sk, SOCK_DONE)) {
873 /* This occurs when user tries to read
874 * from never connected socket.
888 if (signal_pending(current)) {
889 len = sock_intr_errno(timeo);
893 sk_wait_data(sk, &timeo, NULL);
898 else if (len < skb->len)
899 msg->msg_flags |= MSG_TRUNC;
901 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
902 /* Exception. Bailout! */
906 if (flags & MSG_TRUNC)
909 if (!(flags & MSG_PEEK))
918 EXPORT_SYMBOL_GPL(dccp_recvmsg);
920 int inet_dccp_listen(struct socket *sock, int backlog)
922 struct sock *sk = sock->sk;
923 unsigned char old_state;
929 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
932 old_state = sk->sk_state;
933 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
936 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
937 /* Really, if the socket is already in listen state
938 * we can only allow the backlog to be adjusted.
940 if (old_state != DCCP_LISTEN) {
941 struct dccp_sock *dp = dccp_sk(sk);
943 dp->dccps_role = DCCP_ROLE_LISTEN;
945 /* do not start to listen if feature negotiation setup fails */
946 if (dccp_feat_finalise_settings(dp)) {
951 err = inet_csk_listen_start(sk);
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 this_cpu_inc(dccp_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 unsigned long nr_pages = totalram_pages();
1126 int ehash_order, bhash_order, i;
1129 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1130 sizeof_field(struct sk_buff, cb));
1131 rc = inet_hashinfo2_init_mod(&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 | SLAB_ACCOUNT, NULL);
1139 if (!dccp_hashinfo.bind_bucket_cachep)
1140 goto out_free_hashinfo2;
1141 dccp_hashinfo.bind2_bucket_cachep =
1142 kmem_cache_create("dccp_bind2_bucket",
1143 sizeof(struct inet_bind2_bucket), 0,
1144 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1145 if (!dccp_hashinfo.bind2_bucket_cachep)
1146 goto out_free_bind_bucket_cachep;
1149 * Size and allocate the main established and bind bucket
1152 * The methodology is similar to that of the buffer cache.
1154 if (nr_pages >= (128 * 1024))
1155 goal = nr_pages >> (21 - PAGE_SHIFT);
1157 goal = nr_pages >> (23 - PAGE_SHIFT);
1160 goal = (thash_entries *
1161 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1162 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1165 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1166 sizeof(struct inet_ehash_bucket);
1168 while (hash_size & (hash_size - 1))
1170 dccp_hashinfo.ehash_mask = hash_size - 1;
1171 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1172 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1173 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1175 if (!dccp_hashinfo.ehash) {
1176 DCCP_CRIT("Failed to allocate DCCP established hash table");
1177 goto out_free_bind2_bucket_cachep;
1180 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1181 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1183 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1184 goto out_free_dccp_ehash;
1186 bhash_order = ehash_order;
1189 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1190 sizeof(struct inet_bind_hashbucket);
1191 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1194 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1195 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1196 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1198 if (!dccp_hashinfo.bhash) {
1199 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1200 goto out_free_dccp_locks;
1203 dccp_hashinfo.bhash2 = (struct inet_bind_hashbucket *)
1204 __get_free_pages(GFP_ATOMIC | __GFP_NOWARN, bhash_order);
1206 if (!dccp_hashinfo.bhash2) {
1207 DCCP_CRIT("Failed to allocate DCCP bind2 hash table");
1208 goto out_free_dccp_bhash;
1211 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1212 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1213 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1214 spin_lock_init(&dccp_hashinfo.bhash2[i].lock);
1215 INIT_HLIST_HEAD(&dccp_hashinfo.bhash2[i].chain);
1218 dccp_hashinfo.pernet = false;
1220 rc = dccp_mib_init();
1222 goto out_free_dccp_bhash2;
1224 rc = dccp_ackvec_init();
1226 goto out_free_dccp_mib;
1228 rc = dccp_sysctl_init();
1230 goto out_ackvec_exit;
1232 rc = ccid_initialize_builtins();
1234 goto out_sysctl_exit;
1236 dccp_timestamping_init();
1246 out_free_dccp_bhash2:
1247 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1248 out_free_dccp_bhash:
1249 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1250 out_free_dccp_locks:
1251 inet_ehash_locks_free(&dccp_hashinfo);
1252 out_free_dccp_ehash:
1253 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1254 out_free_bind2_bucket_cachep:
1255 kmem_cache_destroy(dccp_hashinfo.bind2_bucket_cachep);
1256 out_free_bind_bucket_cachep:
1257 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1259 inet_hashinfo2_free_mod(&dccp_hashinfo);
1261 dccp_hashinfo.bhash = NULL;
1262 dccp_hashinfo.bhash2 = NULL;
1263 dccp_hashinfo.ehash = NULL;
1264 dccp_hashinfo.bind_bucket_cachep = NULL;
1265 dccp_hashinfo.bind2_bucket_cachep = NULL;
1269 static void __exit dccp_fini(void)
1271 int bhash_order = get_order(dccp_hashinfo.bhash_size *
1272 sizeof(struct inet_bind_hashbucket));
1274 ccid_cleanup_builtins();
1276 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1277 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1278 free_pages((unsigned long)dccp_hashinfo.ehash,
1279 get_order((dccp_hashinfo.ehash_mask + 1) *
1280 sizeof(struct inet_ehash_bucket)));
1281 inet_ehash_locks_free(&dccp_hashinfo);
1282 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1285 inet_hashinfo2_free_mod(&dccp_hashinfo);
1288 module_init(dccp_init);
1289 module_exit(dccp_fini);
1291 MODULE_LICENSE("GPL");
1292 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1293 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");