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 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_need_reset(int state)
243 return state != DCCP_CLOSED && state != DCCP_LISTEN &&
244 state != DCCP_REQUESTING;
247 int dccp_disconnect(struct sock *sk, int flags)
249 struct inet_connection_sock *icsk = inet_csk(sk);
250 struct inet_sock *inet = inet_sk(sk);
251 struct dccp_sock *dp = dccp_sk(sk);
252 const int old_state = sk->sk_state;
254 if (old_state != DCCP_CLOSED)
255 dccp_set_state(sk, DCCP_CLOSED);
258 * This corresponds to the ABORT function of RFC793, sec. 3.8
259 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
261 if (old_state == DCCP_LISTEN) {
262 inet_csk_listen_stop(sk);
263 } else if (dccp_need_reset(old_state)) {
264 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
265 sk->sk_err = ECONNRESET;
266 } else if (old_state == DCCP_REQUESTING)
267 sk->sk_err = ECONNRESET;
269 dccp_clear_xmit_timers(sk);
270 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
271 dp->dccps_hc_rx_ccid = NULL;
273 __skb_queue_purge(&sk->sk_receive_queue);
274 __skb_queue_purge(&sk->sk_write_queue);
275 if (sk->sk_send_head != NULL) {
276 __kfree_skb(sk->sk_send_head);
277 sk->sk_send_head = NULL;
280 inet->inet_dport = 0;
282 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
283 inet_reset_saddr(sk);
286 sock_reset_flag(sk, SOCK_DONE);
288 icsk->icsk_backoff = 0;
289 inet_csk_delack_init(sk);
292 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
298 EXPORT_SYMBOL_GPL(dccp_disconnect);
301 * Wait for a DCCP event.
303 * Note that we don't need to lock the socket, as the upper poll layers
304 * take care of normal races (between the test and the event) and we don't
305 * go look at any of the socket buffers directly.
307 __poll_t dccp_poll(struct file *file, struct socket *sock,
311 struct sock *sk = sock->sk;
313 sock_poll_wait(file, sock, wait);
314 if (sk->sk_state == DCCP_LISTEN)
315 return inet_csk_listen_poll(sk);
317 /* Socket is not locked. We are protected from async events
318 by poll logic and correct handling of state changes
319 made by another threads is impossible in any case.
326 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
328 if (sk->sk_shutdown & RCV_SHUTDOWN)
329 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
332 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
333 if (atomic_read(&sk->sk_rmem_alloc) > 0)
334 mask |= EPOLLIN | EPOLLRDNORM;
336 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
337 if (sk_stream_is_writeable(sk)) {
338 mask |= EPOLLOUT | EPOLLWRNORM;
339 } else { /* send SIGIO later */
340 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
341 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
343 /* Race breaker. If space is freed after
344 * wspace test but before the flags are set,
345 * IO signal will be lost.
347 if (sk_stream_is_writeable(sk))
348 mask |= EPOLLOUT | EPOLLWRNORM;
355 EXPORT_SYMBOL_GPL(dccp_poll);
357 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
363 if (sk->sk_state == DCCP_LISTEN)
368 int amount = sk_wmem_alloc_get(sk);
369 /* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and
370 * always 0, comparably to UDP.
373 rc = put_user(amount, (int __user *)arg);
378 unsigned long amount = 0;
380 skb = skb_peek(&sk->sk_receive_queue);
383 * We will only return the amount of this packet since
384 * that is all that will be read.
388 rc = put_user(amount, (int __user *)arg);
400 EXPORT_SYMBOL_GPL(dccp_ioctl);
402 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
403 sockptr_t optval, unsigned int optlen)
405 struct dccp_sock *dp = dccp_sk(sk);
406 struct dccp_service_list *sl = NULL;
408 if (service == DCCP_SERVICE_INVALID_VALUE ||
409 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
412 if (optlen > sizeof(service)) {
413 sl = kmalloc(optlen, GFP_KERNEL);
417 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
418 if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
419 sizeof(service), optlen - sizeof(service)) ||
420 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
427 dp->dccps_service = service;
429 kfree(dp->dccps_service_list);
431 dp->dccps_service_list = sl;
436 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
441 if (cscov < 0 || cscov > 15)
444 * Populate a list of permissible values, in the range cscov...15. This
445 * is necessary since feature negotiation of single values only works if
446 * both sides incidentally choose the same value. Since the list starts
447 * lowest-value first, negotiation will pick the smallest shared value.
453 list = kmalloc(len, GFP_KERNEL);
457 for (i = 0; i < len; i++)
460 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
464 dccp_sk(sk)->dccps_pcrlen = cscov;
466 dccp_sk(sk)->dccps_pcslen = cscov;
472 static int dccp_setsockopt_ccid(struct sock *sk, int type,
473 sockptr_t optval, unsigned int optlen)
478 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
481 val = memdup_sockptr(optval, optlen);
486 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
487 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
489 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
490 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
497 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
498 sockptr_t optval, unsigned int optlen)
500 struct dccp_sock *dp = dccp_sk(sk);
504 case DCCP_SOCKOPT_PACKET_SIZE:
505 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
507 case DCCP_SOCKOPT_CHANGE_L:
508 case DCCP_SOCKOPT_CHANGE_R:
509 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
511 case DCCP_SOCKOPT_CCID:
512 case DCCP_SOCKOPT_RX_CCID:
513 case DCCP_SOCKOPT_TX_CCID:
514 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
517 if (optlen < (int)sizeof(int))
520 if (copy_from_sockptr(&val, optval, sizeof(int)))
523 if (optname == DCCP_SOCKOPT_SERVICE)
524 return dccp_setsockopt_service(sk, val, optval, optlen);
528 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
529 if (dp->dccps_role != DCCP_ROLE_SERVER)
532 dp->dccps_server_timewait = (val != 0);
534 case DCCP_SOCKOPT_SEND_CSCOV:
535 err = dccp_setsockopt_cscov(sk, val, false);
537 case DCCP_SOCKOPT_RECV_CSCOV:
538 err = dccp_setsockopt_cscov(sk, val, true);
540 case DCCP_SOCKOPT_QPOLICY_ID:
541 if (sk->sk_state != DCCP_CLOSED)
543 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
546 dp->dccps_qpolicy = val;
548 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
552 dp->dccps_tx_qlen = val;
563 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
566 if (level != SOL_DCCP)
567 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
570 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
573 EXPORT_SYMBOL_GPL(dccp_setsockopt);
575 static int dccp_getsockopt_service(struct sock *sk, int len,
576 __be32 __user *optval,
579 const struct dccp_sock *dp = dccp_sk(sk);
580 const struct dccp_service_list *sl;
581 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
584 if ((sl = dp->dccps_service_list) != NULL) {
585 slen = sl->dccpsl_nr * sizeof(u32);
594 if (put_user(total_len, optlen) ||
595 put_user(dp->dccps_service, optval) ||
596 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
603 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
604 char __user *optval, int __user *optlen)
606 struct dccp_sock *dp;
609 if (get_user(len, optlen))
612 if (len < (int)sizeof(int))
618 case DCCP_SOCKOPT_PACKET_SIZE:
619 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
621 case DCCP_SOCKOPT_SERVICE:
622 return dccp_getsockopt_service(sk, len,
623 (__be32 __user *)optval, optlen);
624 case DCCP_SOCKOPT_GET_CUR_MPS:
625 val = dp->dccps_mss_cache;
627 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
628 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
629 case DCCP_SOCKOPT_TX_CCID:
630 val = ccid_get_current_tx_ccid(dp);
634 case DCCP_SOCKOPT_RX_CCID:
635 val = ccid_get_current_rx_ccid(dp);
639 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
640 val = dp->dccps_server_timewait;
642 case DCCP_SOCKOPT_SEND_CSCOV:
643 val = dp->dccps_pcslen;
645 case DCCP_SOCKOPT_RECV_CSCOV:
646 val = dp->dccps_pcrlen;
648 case DCCP_SOCKOPT_QPOLICY_ID:
649 val = dp->dccps_qpolicy;
651 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
652 val = dp->dccps_tx_qlen;
655 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
656 len, (u32 __user *)optval, optlen);
658 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
659 len, (u32 __user *)optval, optlen);
665 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
671 int dccp_getsockopt(struct sock *sk, int level, int optname,
672 char __user *optval, int __user *optlen)
674 if (level != SOL_DCCP)
675 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
678 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
681 EXPORT_SYMBOL_GPL(dccp_getsockopt);
683 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
685 struct cmsghdr *cmsg;
688 * Assign an (opaque) qpolicy priority value to skb->priority.
690 * We are overloading this skb field for use with the qpolicy subystem.
691 * The skb->priority is normally used for the SO_PRIORITY option, which
692 * is initialised from sk_priority. Since the assignment of sk_priority
693 * to skb->priority happens later (on layer 3), we overload this field
694 * for use with queueing priorities as long as the skb is on layer 4.
695 * The default priority value (if nothing is set) is 0.
699 for_each_cmsghdr(cmsg, msg) {
700 if (!CMSG_OK(msg, cmsg))
703 if (cmsg->cmsg_level != SOL_DCCP)
706 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
707 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
710 switch (cmsg->cmsg_type) {
711 case DCCP_SCM_PRIORITY:
712 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
714 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
723 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
725 const struct dccp_sock *dp = dccp_sk(sk);
726 const int flags = msg->msg_flags;
727 const int noblock = flags & MSG_DONTWAIT;
732 trace_dccp_probe(sk, len);
734 if (len > dp->dccps_mss_cache)
739 timeo = sock_sndtimeo(sk, noblock);
742 * We have to use sk_stream_wait_connect here to set sk_write_pending,
743 * so that the trick in dccp_rcv_request_sent_state_process.
745 /* Wait for a connection to finish. */
746 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
747 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
750 size = sk->sk_prot->max_header + len;
752 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
757 if (dccp_qpolicy_full(sk)) {
762 if (sk->sk_state == DCCP_CLOSED) {
767 skb_reserve(skb, sk->sk_prot->max_header);
768 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
772 rc = dccp_msghdr_parse(msg, skb);
776 dccp_qpolicy_push(sk, skb);
778 * The xmit_timer is set if the TX CCID is rate-based and will expire
779 * when congestion control permits to release further packets into the
780 * network. Window-based CCIDs do not use this timer.
782 if (!timer_pending(&dp->dccps_xmit_timer))
792 EXPORT_SYMBOL_GPL(dccp_sendmsg);
794 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
797 const struct dccp_hdr *dh;
802 if (sk->sk_state == DCCP_LISTEN) {
807 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
810 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
813 goto verify_sock_status;
817 switch (dh->dccph_type) {
819 case DCCP_PKT_DATAACK:
823 case DCCP_PKT_CLOSEREQ:
824 if (!(flags & MSG_PEEK))
825 dccp_finish_passive_close(sk);
828 dccp_pr_debug("found fin (%s) ok!\n",
829 dccp_packet_name(dh->dccph_type));
833 dccp_pr_debug("packet_type=%s\n",
834 dccp_packet_name(dh->dccph_type));
838 if (sock_flag(sk, SOCK_DONE)) {
844 len = sock_error(sk);
848 if (sk->sk_shutdown & RCV_SHUTDOWN) {
853 if (sk->sk_state == DCCP_CLOSED) {
854 if (!sock_flag(sk, SOCK_DONE)) {
855 /* This occurs when user tries to read
856 * from never connected socket.
870 if (signal_pending(current)) {
871 len = sock_intr_errno(timeo);
875 sk_wait_data(sk, &timeo, NULL);
880 else if (len < skb->len)
881 msg->msg_flags |= MSG_TRUNC;
883 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
884 /* Exception. Bailout! */
888 if (flags & MSG_TRUNC)
891 if (!(flags & MSG_PEEK))
900 EXPORT_SYMBOL_GPL(dccp_recvmsg);
902 int inet_dccp_listen(struct socket *sock, int backlog)
904 struct sock *sk = sock->sk;
905 unsigned char old_state;
911 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
914 old_state = sk->sk_state;
915 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
918 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
919 /* Really, if the socket is already in listen state
920 * we can only allow the backlog to be adjusted.
922 if (old_state != DCCP_LISTEN) {
923 struct dccp_sock *dp = dccp_sk(sk);
925 dp->dccps_role = DCCP_ROLE_LISTEN;
927 /* do not start to listen if feature negotiation setup fails */
928 if (dccp_feat_finalise_settings(dp)) {
933 err = inet_csk_listen_start(sk);
944 EXPORT_SYMBOL_GPL(inet_dccp_listen);
946 static void dccp_terminate_connection(struct sock *sk)
948 u8 next_state = DCCP_CLOSED;
950 switch (sk->sk_state) {
951 case DCCP_PASSIVE_CLOSE:
952 case DCCP_PASSIVE_CLOSEREQ:
953 dccp_finish_passive_close(sk);
956 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
957 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
960 dccp_send_close(sk, 1);
962 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
963 !dccp_sk(sk)->dccps_server_timewait)
964 next_state = DCCP_ACTIVE_CLOSEREQ;
966 next_state = DCCP_CLOSING;
969 dccp_set_state(sk, next_state);
973 void dccp_close(struct sock *sk, long timeout)
975 struct dccp_sock *dp = dccp_sk(sk);
977 u32 data_was_unread = 0;
982 sk->sk_shutdown = SHUTDOWN_MASK;
984 if (sk->sk_state == DCCP_LISTEN) {
985 dccp_set_state(sk, DCCP_CLOSED);
988 inet_csk_listen_stop(sk);
990 goto adjudge_to_death;
993 sk_stop_timer(sk, &dp->dccps_xmit_timer);
996 * We need to flush the recv. buffs. We do this only on the
997 * descriptor close, not protocol-sourced closes, because the
998 *reader process may not have drained the data yet!
1000 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1001 data_was_unread += skb->len;
1005 /* If socket has been already reset kill it. */
1006 if (sk->sk_state == DCCP_CLOSED)
1007 goto adjudge_to_death;
1009 if (data_was_unread) {
1010 /* Unread data was tossed, send an appropriate Reset Code */
1011 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1012 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1013 dccp_set_state(sk, DCCP_CLOSED);
1014 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1015 /* Check zero linger _after_ checking for unread data. */
1016 sk->sk_prot->disconnect(sk, 0);
1017 } else if (sk->sk_state != DCCP_CLOSED) {
1019 * Normal connection termination. May need to wait if there are
1020 * still packets in the TX queue that are delayed by the CCID.
1022 dccp_flush_write_queue(sk, &timeout);
1023 dccp_terminate_connection(sk);
1027 * Flush write queue. This may be necessary in several cases:
1028 * - we have been closed by the peer but still have application data;
1029 * - abortive termination (unread data or zero linger time),
1030 * - normal termination but queue could not be flushed within time limit
1032 __skb_queue_purge(&sk->sk_write_queue);
1034 sk_stream_wait_close(sk, timeout);
1037 state = sk->sk_state;
1042 * It is the last release_sock in its life. It will remove backlog.
1046 * Now socket is owned by kernel and we acquire BH lock
1047 * to finish close. No need to check for user refs.
1051 WARN_ON(sock_owned_by_user(sk));
1053 this_cpu_inc(dccp_orphan_count);
1055 /* Have we already been destroyed by a softirq or backlog? */
1056 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1059 if (sk->sk_state == DCCP_CLOSED)
1060 inet_csk_destroy_sock(sk);
1062 /* Otherwise, socket is reprieved until protocol close. */
1070 EXPORT_SYMBOL_GPL(dccp_close);
1072 void dccp_shutdown(struct sock *sk, int how)
1074 dccp_pr_debug("called shutdown(%x)\n", how);
1077 EXPORT_SYMBOL_GPL(dccp_shutdown);
1079 static inline int __init dccp_mib_init(void)
1081 dccp_statistics = alloc_percpu(struct dccp_mib);
1082 if (!dccp_statistics)
1087 static inline void dccp_mib_exit(void)
1089 free_percpu(dccp_statistics);
1092 static int thash_entries;
1093 module_param(thash_entries, int, 0444);
1094 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1096 #ifdef CONFIG_IP_DCCP_DEBUG
1098 module_param(dccp_debug, bool, 0644);
1099 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1101 EXPORT_SYMBOL_GPL(dccp_debug);
1104 static int __init dccp_init(void)
1107 unsigned long nr_pages = totalram_pages();
1108 int ehash_order, bhash_order, i;
1111 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1112 sizeof_field(struct sk_buff, cb));
1113 rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
1117 dccp_hashinfo.bind_bucket_cachep =
1118 kmem_cache_create("dccp_bind_bucket",
1119 sizeof(struct inet_bind_bucket), 0,
1120 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1121 if (!dccp_hashinfo.bind_bucket_cachep)
1122 goto out_free_hashinfo2;
1123 dccp_hashinfo.bind2_bucket_cachep =
1124 kmem_cache_create("dccp_bind2_bucket",
1125 sizeof(struct inet_bind2_bucket), 0,
1126 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1127 if (!dccp_hashinfo.bind2_bucket_cachep)
1128 goto out_free_bind_bucket_cachep;
1131 * Size and allocate the main established and bind bucket
1134 * The methodology is similar to that of the buffer cache.
1136 if (nr_pages >= (128 * 1024))
1137 goal = nr_pages >> (21 - PAGE_SHIFT);
1139 goal = nr_pages >> (23 - PAGE_SHIFT);
1142 goal = (thash_entries *
1143 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1144 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1147 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1148 sizeof(struct inet_ehash_bucket);
1150 while (hash_size & (hash_size - 1))
1152 dccp_hashinfo.ehash_mask = hash_size - 1;
1153 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1154 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1155 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1157 if (!dccp_hashinfo.ehash) {
1158 DCCP_CRIT("Failed to allocate DCCP established hash table");
1159 goto out_free_bind2_bucket_cachep;
1162 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1163 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1165 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1166 goto out_free_dccp_ehash;
1168 bhash_order = ehash_order;
1171 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1172 sizeof(struct inet_bind_hashbucket);
1173 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1176 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1177 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1178 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1180 if (!dccp_hashinfo.bhash) {
1181 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1182 goto out_free_dccp_locks;
1185 dccp_hashinfo.bhash2 = (struct inet_bind_hashbucket *)
1186 __get_free_pages(GFP_ATOMIC | __GFP_NOWARN, bhash_order);
1188 if (!dccp_hashinfo.bhash2) {
1189 DCCP_CRIT("Failed to allocate DCCP bind2 hash table");
1190 goto out_free_dccp_bhash;
1193 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1194 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1195 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1196 spin_lock_init(&dccp_hashinfo.bhash2[i].lock);
1197 INIT_HLIST_HEAD(&dccp_hashinfo.bhash2[i].chain);
1200 dccp_hashinfo.pernet = false;
1202 rc = dccp_mib_init();
1204 goto out_free_dccp_bhash2;
1206 rc = dccp_ackvec_init();
1208 goto out_free_dccp_mib;
1210 rc = dccp_sysctl_init();
1212 goto out_ackvec_exit;
1214 rc = ccid_initialize_builtins();
1216 goto out_sysctl_exit;
1218 dccp_timestamping_init();
1228 out_free_dccp_bhash2:
1229 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1230 out_free_dccp_bhash:
1231 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1232 out_free_dccp_locks:
1233 inet_ehash_locks_free(&dccp_hashinfo);
1234 out_free_dccp_ehash:
1235 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1236 out_free_bind2_bucket_cachep:
1237 kmem_cache_destroy(dccp_hashinfo.bind2_bucket_cachep);
1238 out_free_bind_bucket_cachep:
1239 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1241 inet_hashinfo2_free_mod(&dccp_hashinfo);
1243 dccp_hashinfo.bhash = NULL;
1244 dccp_hashinfo.bhash2 = NULL;
1245 dccp_hashinfo.ehash = NULL;
1246 dccp_hashinfo.bind_bucket_cachep = NULL;
1247 dccp_hashinfo.bind2_bucket_cachep = NULL;
1251 static void __exit dccp_fini(void)
1253 int bhash_order = get_order(dccp_hashinfo.bhash_size *
1254 sizeof(struct inet_bind_hashbucket));
1256 ccid_cleanup_builtins();
1258 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1259 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1260 free_pages((unsigned long)dccp_hashinfo.ehash,
1261 get_order((dccp_hashinfo.ehash_mask + 1) *
1262 sizeof(struct inet_ehash_bucket)));
1263 inet_ehash_locks_free(&dccp_hashinfo);
1264 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1267 inet_hashinfo2_free_mod(&dccp_hashinfo);
1270 module_init(dccp_init);
1271 module_exit(dccp_fini);
1273 MODULE_LICENSE("GPL");
1274 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1275 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");