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,
319 struct sock *sk = sock->sk;
321 sock_poll_wait(file, sock, wait);
322 if (sk->sk_state == DCCP_LISTEN)
323 return inet_csk_listen_poll(sk);
325 /* Socket is not locked. We are protected from async events
326 by poll logic and correct handling of state changes
327 made by another threads is impossible in any case.
334 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
336 if (sk->sk_shutdown & RCV_SHUTDOWN)
337 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
340 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
341 if (atomic_read(&sk->sk_rmem_alloc) > 0)
342 mask |= EPOLLIN | EPOLLRDNORM;
344 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
345 if (sk_stream_is_writeable(sk)) {
346 mask |= EPOLLOUT | EPOLLWRNORM;
347 } else { /* send SIGIO later */
348 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
349 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
351 /* Race breaker. If space is freed after
352 * wspace test but before the flags are set,
353 * IO signal will be lost.
355 if (sk_stream_is_writeable(sk))
356 mask |= EPOLLOUT | EPOLLWRNORM;
363 EXPORT_SYMBOL_GPL(dccp_poll);
365 int dccp_ioctl(struct sock *sk, int cmd, int *karg)
371 if (sk->sk_state == DCCP_LISTEN)
376 *karg = sk_wmem_alloc_get(sk);
377 /* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and
378 * always 0, comparably to UDP.
388 skb = skb_peek(&sk->sk_receive_queue);
391 * We will only return the amount of this packet since
392 * that is all that will be read.
408 EXPORT_SYMBOL_GPL(dccp_ioctl);
410 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
411 sockptr_t optval, unsigned int optlen)
413 struct dccp_sock *dp = dccp_sk(sk);
414 struct dccp_service_list *sl = NULL;
416 if (service == DCCP_SERVICE_INVALID_VALUE ||
417 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
420 if (optlen > sizeof(service)) {
421 sl = kmalloc(optlen, GFP_KERNEL);
425 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
426 if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
427 sizeof(service), optlen - sizeof(service)) ||
428 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
435 dp->dccps_service = service;
437 kfree(dp->dccps_service_list);
439 dp->dccps_service_list = sl;
444 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
449 if (cscov < 0 || cscov > 15)
452 * Populate a list of permissible values, in the range cscov...15. This
453 * is necessary since feature negotiation of single values only works if
454 * both sides incidentally choose the same value. Since the list starts
455 * lowest-value first, negotiation will pick the smallest shared value.
461 list = kmalloc(len, GFP_KERNEL);
465 for (i = 0; i < len; i++)
468 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
472 dccp_sk(sk)->dccps_pcrlen = cscov;
474 dccp_sk(sk)->dccps_pcslen = cscov;
480 static int dccp_setsockopt_ccid(struct sock *sk, int type,
481 sockptr_t optval, unsigned int optlen)
486 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
489 val = memdup_sockptr(optval, optlen);
494 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
495 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
497 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
498 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
505 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
506 sockptr_t optval, unsigned int optlen)
508 struct dccp_sock *dp = dccp_sk(sk);
512 case DCCP_SOCKOPT_PACKET_SIZE:
513 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
515 case DCCP_SOCKOPT_CHANGE_L:
516 case DCCP_SOCKOPT_CHANGE_R:
517 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
519 case DCCP_SOCKOPT_CCID:
520 case DCCP_SOCKOPT_RX_CCID:
521 case DCCP_SOCKOPT_TX_CCID:
522 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
525 if (optlen < (int)sizeof(int))
528 if (copy_from_sockptr(&val, optval, sizeof(int)))
531 if (optname == DCCP_SOCKOPT_SERVICE)
532 return dccp_setsockopt_service(sk, val, optval, optlen);
536 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
537 if (dp->dccps_role != DCCP_ROLE_SERVER)
540 dp->dccps_server_timewait = (val != 0);
542 case DCCP_SOCKOPT_SEND_CSCOV:
543 err = dccp_setsockopt_cscov(sk, val, false);
545 case DCCP_SOCKOPT_RECV_CSCOV:
546 err = dccp_setsockopt_cscov(sk, val, true);
548 case DCCP_SOCKOPT_QPOLICY_ID:
549 if (sk->sk_state != DCCP_CLOSED)
551 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
554 dp->dccps_qpolicy = val;
556 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
560 dp->dccps_tx_qlen = val;
571 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
574 if (level != SOL_DCCP)
575 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
578 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
581 EXPORT_SYMBOL_GPL(dccp_setsockopt);
583 static int dccp_getsockopt_service(struct sock *sk, int len,
584 __be32 __user *optval,
587 const struct dccp_sock *dp = dccp_sk(sk);
588 const struct dccp_service_list *sl;
589 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
592 if ((sl = dp->dccps_service_list) != NULL) {
593 slen = sl->dccpsl_nr * sizeof(u32);
602 if (put_user(total_len, optlen) ||
603 put_user(dp->dccps_service, optval) ||
604 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
611 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
612 char __user *optval, int __user *optlen)
614 struct dccp_sock *dp;
617 if (get_user(len, optlen))
620 if (len < (int)sizeof(int))
626 case DCCP_SOCKOPT_PACKET_SIZE:
627 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
629 case DCCP_SOCKOPT_SERVICE:
630 return dccp_getsockopt_service(sk, len,
631 (__be32 __user *)optval, optlen);
632 case DCCP_SOCKOPT_GET_CUR_MPS:
633 val = READ_ONCE(dp->dccps_mss_cache);
635 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
636 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
637 case DCCP_SOCKOPT_TX_CCID:
638 val = ccid_get_current_tx_ccid(dp);
642 case DCCP_SOCKOPT_RX_CCID:
643 val = ccid_get_current_rx_ccid(dp);
647 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
648 val = dp->dccps_server_timewait;
650 case DCCP_SOCKOPT_SEND_CSCOV:
651 val = dp->dccps_pcslen;
653 case DCCP_SOCKOPT_RECV_CSCOV:
654 val = dp->dccps_pcrlen;
656 case DCCP_SOCKOPT_QPOLICY_ID:
657 val = dp->dccps_qpolicy;
659 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
660 val = dp->dccps_tx_qlen;
663 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
664 len, (u32 __user *)optval, optlen);
666 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
667 len, (u32 __user *)optval, optlen);
673 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
679 int dccp_getsockopt(struct sock *sk, int level, int optname,
680 char __user *optval, int __user *optlen)
682 if (level != SOL_DCCP)
683 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
686 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
689 EXPORT_SYMBOL_GPL(dccp_getsockopt);
691 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
693 struct cmsghdr *cmsg;
696 * Assign an (opaque) qpolicy priority value to skb->priority.
698 * We are overloading this skb field for use with the qpolicy subystem.
699 * The skb->priority is normally used for the SO_PRIORITY option, which
700 * is initialised from sk_priority. Since the assignment of sk_priority
701 * to skb->priority happens later (on layer 3), we overload this field
702 * for use with queueing priorities as long as the skb is on layer 4.
703 * The default priority value (if nothing is set) is 0.
707 for_each_cmsghdr(cmsg, msg) {
708 if (!CMSG_OK(msg, cmsg))
711 if (cmsg->cmsg_level != SOL_DCCP)
714 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
715 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
718 switch (cmsg->cmsg_type) {
719 case DCCP_SCM_PRIORITY:
720 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
722 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
731 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
733 const struct dccp_sock *dp = dccp_sk(sk);
734 const int flags = msg->msg_flags;
735 const int noblock = flags & MSG_DONTWAIT;
740 trace_dccp_probe(sk, len);
742 if (len > READ_ONCE(dp->dccps_mss_cache))
747 timeo = sock_sndtimeo(sk, noblock);
750 * We have to use sk_stream_wait_connect here to set sk_write_pending,
751 * so that the trick in dccp_rcv_request_sent_state_process.
753 /* Wait for a connection to finish. */
754 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
755 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
758 size = sk->sk_prot->max_header + len;
760 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
765 if (dccp_qpolicy_full(sk)) {
770 if (sk->sk_state == DCCP_CLOSED) {
775 /* We need to check dccps_mss_cache after socket is locked. */
776 if (len > dp->dccps_mss_cache) {
781 skb_reserve(skb, sk->sk_prot->max_header);
782 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
786 rc = dccp_msghdr_parse(msg, skb);
790 dccp_qpolicy_push(sk, skb);
792 * The xmit_timer is set if the TX CCID is rate-based and will expire
793 * when congestion control permits to release further packets into the
794 * network. Window-based CCIDs do not use this timer.
796 if (!timer_pending(&dp->dccps_xmit_timer))
806 EXPORT_SYMBOL_GPL(dccp_sendmsg);
808 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
811 const struct dccp_hdr *dh;
816 if (sk->sk_state == DCCP_LISTEN) {
821 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
824 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
827 goto verify_sock_status;
831 switch (dh->dccph_type) {
833 case DCCP_PKT_DATAACK:
837 case DCCP_PKT_CLOSEREQ:
838 if (!(flags & MSG_PEEK))
839 dccp_finish_passive_close(sk);
842 dccp_pr_debug("found fin (%s) ok!\n",
843 dccp_packet_name(dh->dccph_type));
847 dccp_pr_debug("packet_type=%s\n",
848 dccp_packet_name(dh->dccph_type));
852 if (sock_flag(sk, SOCK_DONE)) {
858 len = sock_error(sk);
862 if (sk->sk_shutdown & RCV_SHUTDOWN) {
867 if (sk->sk_state == DCCP_CLOSED) {
868 if (!sock_flag(sk, SOCK_DONE)) {
869 /* This occurs when user tries to read
870 * from never connected socket.
884 if (signal_pending(current)) {
885 len = sock_intr_errno(timeo);
889 sk_wait_data(sk, &timeo, NULL);
894 else if (len < skb->len)
895 msg->msg_flags |= MSG_TRUNC;
897 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
898 /* Exception. Bailout! */
902 if (flags & MSG_TRUNC)
905 if (!(flags & MSG_PEEK))
914 EXPORT_SYMBOL_GPL(dccp_recvmsg);
916 int inet_dccp_listen(struct socket *sock, int backlog)
918 struct sock *sk = sock->sk;
919 unsigned char old_state;
925 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
928 old_state = sk->sk_state;
929 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
932 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
933 /* Really, if the socket is already in listen state
934 * we can only allow the backlog to be adjusted.
936 if (old_state != DCCP_LISTEN) {
937 struct dccp_sock *dp = dccp_sk(sk);
939 dp->dccps_role = DCCP_ROLE_LISTEN;
941 /* do not start to listen if feature negotiation setup fails */
942 if (dccp_feat_finalise_settings(dp)) {
947 err = inet_csk_listen_start(sk);
958 EXPORT_SYMBOL_GPL(inet_dccp_listen);
960 static void dccp_terminate_connection(struct sock *sk)
962 u8 next_state = DCCP_CLOSED;
964 switch (sk->sk_state) {
965 case DCCP_PASSIVE_CLOSE:
966 case DCCP_PASSIVE_CLOSEREQ:
967 dccp_finish_passive_close(sk);
970 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
971 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
974 dccp_send_close(sk, 1);
976 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
977 !dccp_sk(sk)->dccps_server_timewait)
978 next_state = DCCP_ACTIVE_CLOSEREQ;
980 next_state = DCCP_CLOSING;
983 dccp_set_state(sk, next_state);
987 void dccp_close(struct sock *sk, long timeout)
989 struct dccp_sock *dp = dccp_sk(sk);
991 u32 data_was_unread = 0;
996 sk->sk_shutdown = SHUTDOWN_MASK;
998 if (sk->sk_state == DCCP_LISTEN) {
999 dccp_set_state(sk, DCCP_CLOSED);
1002 inet_csk_listen_stop(sk);
1004 goto adjudge_to_death;
1007 sk_stop_timer(sk, &dp->dccps_xmit_timer);
1010 * We need to flush the recv. buffs. We do this only on the
1011 * descriptor close, not protocol-sourced closes, because the
1012 *reader process may not have drained the data yet!
1014 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1015 data_was_unread += skb->len;
1019 /* If socket has been already reset kill it. */
1020 if (sk->sk_state == DCCP_CLOSED)
1021 goto adjudge_to_death;
1023 if (data_was_unread) {
1024 /* Unread data was tossed, send an appropriate Reset Code */
1025 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1026 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1027 dccp_set_state(sk, DCCP_CLOSED);
1028 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1029 /* Check zero linger _after_ checking for unread data. */
1030 sk->sk_prot->disconnect(sk, 0);
1031 } else if (sk->sk_state != DCCP_CLOSED) {
1033 * Normal connection termination. May need to wait if there are
1034 * still packets in the TX queue that are delayed by the CCID.
1036 dccp_flush_write_queue(sk, &timeout);
1037 dccp_terminate_connection(sk);
1041 * Flush write queue. This may be necessary in several cases:
1042 * - we have been closed by the peer but still have application data;
1043 * - abortive termination (unread data or zero linger time),
1044 * - normal termination but queue could not be flushed within time limit
1046 __skb_queue_purge(&sk->sk_write_queue);
1048 sk_stream_wait_close(sk, timeout);
1051 state = sk->sk_state;
1056 * It is the last release_sock in its life. It will remove backlog.
1060 * Now socket is owned by kernel and we acquire BH lock
1061 * to finish close. No need to check for user refs.
1065 WARN_ON(sock_owned_by_user(sk));
1067 this_cpu_inc(dccp_orphan_count);
1069 /* Have we already been destroyed by a softirq or backlog? */
1070 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1073 if (sk->sk_state == DCCP_CLOSED)
1074 inet_csk_destroy_sock(sk);
1076 /* Otherwise, socket is reprieved until protocol close. */
1084 EXPORT_SYMBOL_GPL(dccp_close);
1086 void dccp_shutdown(struct sock *sk, int how)
1088 dccp_pr_debug("called shutdown(%x)\n", how);
1091 EXPORT_SYMBOL_GPL(dccp_shutdown);
1093 static inline int __init dccp_mib_init(void)
1095 dccp_statistics = alloc_percpu(struct dccp_mib);
1096 if (!dccp_statistics)
1101 static inline void dccp_mib_exit(void)
1103 free_percpu(dccp_statistics);
1106 static int thash_entries;
1107 module_param(thash_entries, int, 0444);
1108 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1110 #ifdef CONFIG_IP_DCCP_DEBUG
1112 module_param(dccp_debug, bool, 0644);
1113 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1115 EXPORT_SYMBOL_GPL(dccp_debug);
1118 static int __init dccp_init(void)
1121 unsigned long nr_pages = totalram_pages();
1122 int ehash_order, bhash_order, i;
1125 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1126 sizeof_field(struct sk_buff, cb));
1127 rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
1131 dccp_hashinfo.bind_bucket_cachep =
1132 kmem_cache_create("dccp_bind_bucket",
1133 sizeof(struct inet_bind_bucket), 0,
1134 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1135 if (!dccp_hashinfo.bind_bucket_cachep)
1136 goto out_free_hashinfo2;
1137 dccp_hashinfo.bind2_bucket_cachep =
1138 kmem_cache_create("dccp_bind2_bucket",
1139 sizeof(struct inet_bind2_bucket), 0,
1140 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1141 if (!dccp_hashinfo.bind2_bucket_cachep)
1142 goto out_free_bind_bucket_cachep;
1145 * Size and allocate the main established and bind bucket
1148 * The methodology is similar to that of the buffer cache.
1150 if (nr_pages >= (128 * 1024))
1151 goal = nr_pages >> (21 - PAGE_SHIFT);
1153 goal = nr_pages >> (23 - PAGE_SHIFT);
1156 goal = (thash_entries *
1157 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1158 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1161 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1162 sizeof(struct inet_ehash_bucket);
1164 while (hash_size & (hash_size - 1))
1166 dccp_hashinfo.ehash_mask = hash_size - 1;
1167 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1168 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1169 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1171 if (!dccp_hashinfo.ehash) {
1172 DCCP_CRIT("Failed to allocate DCCP established hash table");
1173 goto out_free_bind2_bucket_cachep;
1176 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1177 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1179 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1180 goto out_free_dccp_ehash;
1182 bhash_order = ehash_order;
1185 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1186 sizeof(struct inet_bind_hashbucket);
1187 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1190 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1191 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1192 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1194 if (!dccp_hashinfo.bhash) {
1195 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1196 goto out_free_dccp_locks;
1199 dccp_hashinfo.bhash2 = (struct inet_bind_hashbucket *)
1200 __get_free_pages(GFP_ATOMIC | __GFP_NOWARN, bhash_order);
1202 if (!dccp_hashinfo.bhash2) {
1203 DCCP_CRIT("Failed to allocate DCCP bind2 hash table");
1204 goto out_free_dccp_bhash;
1207 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1208 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1209 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1210 spin_lock_init(&dccp_hashinfo.bhash2[i].lock);
1211 INIT_HLIST_HEAD(&dccp_hashinfo.bhash2[i].chain);
1214 dccp_hashinfo.pernet = false;
1216 rc = dccp_mib_init();
1218 goto out_free_dccp_bhash2;
1220 rc = dccp_ackvec_init();
1222 goto out_free_dccp_mib;
1224 rc = dccp_sysctl_init();
1226 goto out_ackvec_exit;
1228 rc = ccid_initialize_builtins();
1230 goto out_sysctl_exit;
1232 dccp_timestamping_init();
1242 out_free_dccp_bhash2:
1243 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1244 out_free_dccp_bhash:
1245 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1246 out_free_dccp_locks:
1247 inet_ehash_locks_free(&dccp_hashinfo);
1248 out_free_dccp_ehash:
1249 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1250 out_free_bind2_bucket_cachep:
1251 kmem_cache_destroy(dccp_hashinfo.bind2_bucket_cachep);
1252 out_free_bind_bucket_cachep:
1253 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1255 inet_hashinfo2_free_mod(&dccp_hashinfo);
1257 dccp_hashinfo.bhash = NULL;
1258 dccp_hashinfo.bhash2 = NULL;
1259 dccp_hashinfo.ehash = NULL;
1260 dccp_hashinfo.bind_bucket_cachep = NULL;
1261 dccp_hashinfo.bind2_bucket_cachep = NULL;
1265 static void __exit dccp_fini(void)
1267 int bhash_order = get_order(dccp_hashinfo.bhash_size *
1268 sizeof(struct inet_bind_hashbucket));
1270 ccid_cleanup_builtins();
1272 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1273 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1274 free_pages((unsigned long)dccp_hashinfo.ehash,
1275 get_order((dccp_hashinfo.ehash_mask + 1) *
1276 sizeof(struct inet_ehash_bucket)));
1277 inet_ehash_locks_free(&dccp_hashinfo);
1278 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1281 inet_hashinfo2_free_mod(&dccp_hashinfo);
1284 module_init(dccp_init);
1285 module_exit(dccp_fini);
1287 MODULE_LICENSE("GPL");
1288 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1289 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");