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, unsigned long arg)
371 if (sk->sk_state == DCCP_LISTEN)
376 int amount = 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.
381 rc = put_user(amount, (int __user *)arg);
386 unsigned long amount = 0;
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.
396 rc = put_user(amount, (int __user *)arg);
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 = 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 > 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 skb_reserve(skb, sk->sk_prot->max_header);
776 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
780 rc = dccp_msghdr_parse(msg, skb);
784 dccp_qpolicy_push(sk, skb);
786 * The xmit_timer is set if the TX CCID is rate-based and will expire
787 * when congestion control permits to release further packets into the
788 * network. Window-based CCIDs do not use this timer.
790 if (!timer_pending(&dp->dccps_xmit_timer))
800 EXPORT_SYMBOL_GPL(dccp_sendmsg);
802 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
805 const struct dccp_hdr *dh;
810 if (sk->sk_state == DCCP_LISTEN) {
815 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
818 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
821 goto verify_sock_status;
825 switch (dh->dccph_type) {
827 case DCCP_PKT_DATAACK:
831 case DCCP_PKT_CLOSEREQ:
832 if (!(flags & MSG_PEEK))
833 dccp_finish_passive_close(sk);
836 dccp_pr_debug("found fin (%s) ok!\n",
837 dccp_packet_name(dh->dccph_type));
841 dccp_pr_debug("packet_type=%s\n",
842 dccp_packet_name(dh->dccph_type));
846 if (sock_flag(sk, SOCK_DONE)) {
852 len = sock_error(sk);
856 if (sk->sk_shutdown & RCV_SHUTDOWN) {
861 if (sk->sk_state == DCCP_CLOSED) {
862 if (!sock_flag(sk, SOCK_DONE)) {
863 /* This occurs when user tries to read
864 * from never connected socket.
878 if (signal_pending(current)) {
879 len = sock_intr_errno(timeo);
883 sk_wait_data(sk, &timeo, NULL);
888 else if (len < skb->len)
889 msg->msg_flags |= MSG_TRUNC;
891 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
892 /* Exception. Bailout! */
896 if (flags & MSG_TRUNC)
899 if (!(flags & MSG_PEEK))
908 EXPORT_SYMBOL_GPL(dccp_recvmsg);
910 int inet_dccp_listen(struct socket *sock, int backlog)
912 struct sock *sk = sock->sk;
913 unsigned char old_state;
919 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
922 old_state = sk->sk_state;
923 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
926 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
927 /* Really, if the socket is already in listen state
928 * we can only allow the backlog to be adjusted.
930 if (old_state != DCCP_LISTEN) {
931 struct dccp_sock *dp = dccp_sk(sk);
933 dp->dccps_role = DCCP_ROLE_LISTEN;
935 /* do not start to listen if feature negotiation setup fails */
936 if (dccp_feat_finalise_settings(dp)) {
941 err = inet_csk_listen_start(sk);
952 EXPORT_SYMBOL_GPL(inet_dccp_listen);
954 static void dccp_terminate_connection(struct sock *sk)
956 u8 next_state = DCCP_CLOSED;
958 switch (sk->sk_state) {
959 case DCCP_PASSIVE_CLOSE:
960 case DCCP_PASSIVE_CLOSEREQ:
961 dccp_finish_passive_close(sk);
964 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
965 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
968 dccp_send_close(sk, 1);
970 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
971 !dccp_sk(sk)->dccps_server_timewait)
972 next_state = DCCP_ACTIVE_CLOSEREQ;
974 next_state = DCCP_CLOSING;
977 dccp_set_state(sk, next_state);
981 void dccp_close(struct sock *sk, long timeout)
983 struct dccp_sock *dp = dccp_sk(sk);
985 u32 data_was_unread = 0;
990 sk->sk_shutdown = SHUTDOWN_MASK;
992 if (sk->sk_state == DCCP_LISTEN) {
993 dccp_set_state(sk, DCCP_CLOSED);
996 inet_csk_listen_stop(sk);
998 goto adjudge_to_death;
1001 sk_stop_timer(sk, &dp->dccps_xmit_timer);
1004 * We need to flush the recv. buffs. We do this only on the
1005 * descriptor close, not protocol-sourced closes, because the
1006 *reader process may not have drained the data yet!
1008 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1009 data_was_unread += skb->len;
1013 /* If socket has been already reset kill it. */
1014 if (sk->sk_state == DCCP_CLOSED)
1015 goto adjudge_to_death;
1017 if (data_was_unread) {
1018 /* Unread data was tossed, send an appropriate Reset Code */
1019 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1020 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1021 dccp_set_state(sk, DCCP_CLOSED);
1022 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1023 /* Check zero linger _after_ checking for unread data. */
1024 sk->sk_prot->disconnect(sk, 0);
1025 } else if (sk->sk_state != DCCP_CLOSED) {
1027 * Normal connection termination. May need to wait if there are
1028 * still packets in the TX queue that are delayed by the CCID.
1030 dccp_flush_write_queue(sk, &timeout);
1031 dccp_terminate_connection(sk);
1035 * Flush write queue. This may be necessary in several cases:
1036 * - we have been closed by the peer but still have application data;
1037 * - abortive termination (unread data or zero linger time),
1038 * - normal termination but queue could not be flushed within time limit
1040 __skb_queue_purge(&sk->sk_write_queue);
1042 sk_stream_wait_close(sk, timeout);
1045 state = sk->sk_state;
1050 * It is the last release_sock in its life. It will remove backlog.
1054 * Now socket is owned by kernel and we acquire BH lock
1055 * to finish close. No need to check for user refs.
1059 WARN_ON(sock_owned_by_user(sk));
1061 this_cpu_inc(dccp_orphan_count);
1063 /* Have we already been destroyed by a softirq or backlog? */
1064 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1067 if (sk->sk_state == DCCP_CLOSED)
1068 inet_csk_destroy_sock(sk);
1070 /* Otherwise, socket is reprieved until protocol close. */
1078 EXPORT_SYMBOL_GPL(dccp_close);
1080 void dccp_shutdown(struct sock *sk, int how)
1082 dccp_pr_debug("called shutdown(%x)\n", how);
1085 EXPORT_SYMBOL_GPL(dccp_shutdown);
1087 static inline int __init dccp_mib_init(void)
1089 dccp_statistics = alloc_percpu(struct dccp_mib);
1090 if (!dccp_statistics)
1095 static inline void dccp_mib_exit(void)
1097 free_percpu(dccp_statistics);
1100 static int thash_entries;
1101 module_param(thash_entries, int, 0444);
1102 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1104 #ifdef CONFIG_IP_DCCP_DEBUG
1106 module_param(dccp_debug, bool, 0644);
1107 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1109 EXPORT_SYMBOL_GPL(dccp_debug);
1112 static int __init dccp_init(void)
1115 unsigned long nr_pages = totalram_pages();
1116 int ehash_order, bhash_order, i;
1119 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1120 sizeof_field(struct sk_buff, cb));
1121 rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
1125 dccp_hashinfo.bind_bucket_cachep =
1126 kmem_cache_create("dccp_bind_bucket",
1127 sizeof(struct inet_bind_bucket), 0,
1128 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1129 if (!dccp_hashinfo.bind_bucket_cachep)
1130 goto out_free_hashinfo2;
1131 dccp_hashinfo.bind2_bucket_cachep =
1132 kmem_cache_create("dccp_bind2_bucket",
1133 sizeof(struct inet_bind2_bucket), 0,
1134 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1135 if (!dccp_hashinfo.bind2_bucket_cachep)
1136 goto out_free_bind_bucket_cachep;
1139 * Size and allocate the main established and bind bucket
1142 * The methodology is similar to that of the buffer cache.
1144 if (nr_pages >= (128 * 1024))
1145 goal = nr_pages >> (21 - PAGE_SHIFT);
1147 goal = nr_pages >> (23 - PAGE_SHIFT);
1150 goal = (thash_entries *
1151 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1152 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1155 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1156 sizeof(struct inet_ehash_bucket);
1158 while (hash_size & (hash_size - 1))
1160 dccp_hashinfo.ehash_mask = hash_size - 1;
1161 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1162 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1163 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1165 if (!dccp_hashinfo.ehash) {
1166 DCCP_CRIT("Failed to allocate DCCP established hash table");
1167 goto out_free_bind2_bucket_cachep;
1170 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1171 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1173 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1174 goto out_free_dccp_ehash;
1176 bhash_order = ehash_order;
1179 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1180 sizeof(struct inet_bind_hashbucket);
1181 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1184 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1185 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1186 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1188 if (!dccp_hashinfo.bhash) {
1189 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1190 goto out_free_dccp_locks;
1193 dccp_hashinfo.bhash2 = (struct inet_bind_hashbucket *)
1194 __get_free_pages(GFP_ATOMIC | __GFP_NOWARN, bhash_order);
1196 if (!dccp_hashinfo.bhash2) {
1197 DCCP_CRIT("Failed to allocate DCCP bind2 hash table");
1198 goto out_free_dccp_bhash;
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);
1204 spin_lock_init(&dccp_hashinfo.bhash2[i].lock);
1205 INIT_HLIST_HEAD(&dccp_hashinfo.bhash2[i].chain);
1208 dccp_hashinfo.pernet = false;
1210 rc = dccp_mib_init();
1212 goto out_free_dccp_bhash2;
1214 rc = dccp_ackvec_init();
1216 goto out_free_dccp_mib;
1218 rc = dccp_sysctl_init();
1220 goto out_ackvec_exit;
1222 rc = ccid_initialize_builtins();
1224 goto out_sysctl_exit;
1226 dccp_timestamping_init();
1236 out_free_dccp_bhash2:
1237 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1238 out_free_dccp_bhash:
1239 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1240 out_free_dccp_locks:
1241 inet_ehash_locks_free(&dccp_hashinfo);
1242 out_free_dccp_ehash:
1243 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1244 out_free_bind2_bucket_cachep:
1245 kmem_cache_destroy(dccp_hashinfo.bind2_bucket_cachep);
1246 out_free_bind_bucket_cachep:
1247 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1249 inet_hashinfo2_free_mod(&dccp_hashinfo);
1251 dccp_hashinfo.bhash = NULL;
1252 dccp_hashinfo.bhash2 = NULL;
1253 dccp_hashinfo.ehash = NULL;
1254 dccp_hashinfo.bind_bucket_cachep = NULL;
1255 dccp_hashinfo.bind2_bucket_cachep = NULL;
1259 static void __exit dccp_fini(void)
1261 int bhash_order = get_order(dccp_hashinfo.bhash_size *
1262 sizeof(struct inet_bind_hashbucket));
1264 ccid_cleanup_builtins();
1266 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1267 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1268 free_pages((unsigned long)dccp_hashinfo.ehash,
1269 get_order((dccp_hashinfo.ehash_mask + 1) *
1270 sizeof(struct inet_ehash_bucket)));
1271 inet_ehash_locks_free(&dccp_hashinfo);
1272 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1275 inet_hashinfo2_free_mod(&dccp_hashinfo);
1278 module_init(dccp_init);
1279 module_exit(dccp_fini);
1281 MODULE_LICENSE("GPL");
1282 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1283 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");