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 icsk->icsk_rto = DCCP_TIMEOUT_INIT;
195 icsk->icsk_syn_retries = sysctl_dccp_request_retries;
196 sk->sk_state = DCCP_CLOSED;
197 sk->sk_write_space = dccp_write_space;
198 sk->sk_destruct = dccp_sk_destruct;
199 icsk->icsk_sync_mss = dccp_sync_mss;
200 dp->dccps_mss_cache = 536;
201 dp->dccps_rate_last = jiffies;
202 dp->dccps_role = DCCP_ROLE_UNDEFINED;
203 dp->dccps_service = DCCP_SERVICE_CODE_IS_ABSENT;
204 dp->dccps_tx_qlen = sysctl_dccp_tx_qlen;
206 dccp_init_xmit_timers(sk);
208 INIT_LIST_HEAD(&dp->dccps_featneg);
209 /* control socket doesn't need feat nego */
210 if (likely(ctl_sock_initialized))
211 return dccp_feat_init(sk);
215 EXPORT_SYMBOL_GPL(dccp_init_sock);
217 void dccp_destroy_sock(struct sock *sk)
219 struct dccp_sock *dp = dccp_sk(sk);
221 __skb_queue_purge(&sk->sk_write_queue);
222 if (sk->sk_send_head != NULL) {
223 kfree_skb(sk->sk_send_head);
224 sk->sk_send_head = NULL;
227 /* Clean up a referenced DCCP bind bucket. */
228 if (inet_csk(sk)->icsk_bind_hash != NULL)
231 kfree(dp->dccps_service_list);
232 dp->dccps_service_list = NULL;
234 if (dp->dccps_hc_rx_ackvec != NULL) {
235 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
236 dp->dccps_hc_rx_ackvec = NULL;
238 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
239 dp->dccps_hc_rx_ccid = NULL;
241 /* clean up feature negotiation state */
242 dccp_feat_list_purge(&dp->dccps_featneg);
245 EXPORT_SYMBOL_GPL(dccp_destroy_sock);
247 static inline int dccp_need_reset(int state)
249 return state != DCCP_CLOSED && state != DCCP_LISTEN &&
250 state != DCCP_REQUESTING;
253 int dccp_disconnect(struct sock *sk, int flags)
255 struct inet_connection_sock *icsk = inet_csk(sk);
256 struct inet_sock *inet = inet_sk(sk);
257 struct dccp_sock *dp = dccp_sk(sk);
258 const int old_state = sk->sk_state;
260 if (old_state != DCCP_CLOSED)
261 dccp_set_state(sk, DCCP_CLOSED);
264 * This corresponds to the ABORT function of RFC793, sec. 3.8
265 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
267 if (old_state == DCCP_LISTEN) {
268 inet_csk_listen_stop(sk);
269 } else if (dccp_need_reset(old_state)) {
270 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
271 sk->sk_err = ECONNRESET;
272 } else if (old_state == DCCP_REQUESTING)
273 sk->sk_err = ECONNRESET;
275 dccp_clear_xmit_timers(sk);
276 ccid_hc_rx_delete(dp->dccps_hc_rx_ccid, sk);
277 dp->dccps_hc_rx_ccid = NULL;
279 __skb_queue_purge(&sk->sk_receive_queue);
280 __skb_queue_purge(&sk->sk_write_queue);
281 if (sk->sk_send_head != NULL) {
282 __kfree_skb(sk->sk_send_head);
283 sk->sk_send_head = NULL;
286 inet->inet_dport = 0;
288 inet_bhash2_reset_saddr(sk);
291 sock_reset_flag(sk, SOCK_DONE);
293 icsk->icsk_backoff = 0;
294 inet_csk_delack_init(sk);
297 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
303 EXPORT_SYMBOL_GPL(dccp_disconnect);
306 * Wait for a DCCP event.
308 * Note that we don't need to lock the socket, as the upper poll layers
309 * take care of normal races (between the test and the event) and we don't
310 * go look at any of the socket buffers directly.
312 __poll_t dccp_poll(struct file *file, struct socket *sock,
315 struct sock *sk = sock->sk;
320 sock_poll_wait(file, sock, wait);
322 state = inet_sk_state_load(sk);
323 if (state == DCCP_LISTEN)
324 return inet_csk_listen_poll(sk);
326 /* Socket is not locked. We are protected from async events
327 by poll logic and correct handling of state changes
328 made by another threads is impossible in any case.
332 if (READ_ONCE(sk->sk_err))
334 shutdown = READ_ONCE(sk->sk_shutdown);
336 if (shutdown == SHUTDOWN_MASK || state == DCCP_CLOSED)
338 if (shutdown & RCV_SHUTDOWN)
339 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
342 if ((1 << state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
343 if (atomic_read(&sk->sk_rmem_alloc) > 0)
344 mask |= EPOLLIN | EPOLLRDNORM;
346 if (!(shutdown & SEND_SHUTDOWN)) {
347 if (sk_stream_is_writeable(sk)) {
348 mask |= EPOLLOUT | EPOLLWRNORM;
349 } else { /* send SIGIO later */
350 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
351 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
353 /* Race breaker. If space is freed after
354 * wspace test but before the flags are set,
355 * IO signal will be lost.
357 if (sk_stream_is_writeable(sk))
358 mask |= EPOLLOUT | EPOLLWRNORM;
364 EXPORT_SYMBOL_GPL(dccp_poll);
366 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
372 if (sk->sk_state == DCCP_LISTEN)
377 int amount = sk_wmem_alloc_get(sk);
378 /* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and
379 * always 0, comparably to UDP.
382 rc = put_user(amount, (int __user *)arg);
387 unsigned long amount = 0;
389 skb = skb_peek(&sk->sk_receive_queue);
392 * We will only return the amount of this packet since
393 * that is all that will be read.
397 rc = put_user(amount, (int __user *)arg);
409 EXPORT_SYMBOL_GPL(dccp_ioctl);
411 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
412 sockptr_t optval, unsigned int optlen)
414 struct dccp_sock *dp = dccp_sk(sk);
415 struct dccp_service_list *sl = NULL;
417 if (service == DCCP_SERVICE_INVALID_VALUE ||
418 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
421 if (optlen > sizeof(service)) {
422 sl = kmalloc(optlen, GFP_KERNEL);
426 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
427 if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
428 sizeof(service), optlen - sizeof(service)) ||
429 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
436 dp->dccps_service = service;
438 kfree(dp->dccps_service_list);
440 dp->dccps_service_list = sl;
445 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
450 if (cscov < 0 || cscov > 15)
453 * Populate a list of permissible values, in the range cscov...15. This
454 * is necessary since feature negotiation of single values only works if
455 * both sides incidentally choose the same value. Since the list starts
456 * lowest-value first, negotiation will pick the smallest shared value.
462 list = kmalloc(len, GFP_KERNEL);
466 for (i = 0; i < len; i++)
469 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
473 dccp_sk(sk)->dccps_pcrlen = cscov;
475 dccp_sk(sk)->dccps_pcslen = cscov;
481 static int dccp_setsockopt_ccid(struct sock *sk, int type,
482 sockptr_t optval, unsigned int optlen)
487 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
490 val = memdup_sockptr(optval, optlen);
495 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
496 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
498 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
499 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
506 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
507 sockptr_t optval, unsigned int optlen)
509 struct dccp_sock *dp = dccp_sk(sk);
513 case DCCP_SOCKOPT_PACKET_SIZE:
514 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
516 case DCCP_SOCKOPT_CHANGE_L:
517 case DCCP_SOCKOPT_CHANGE_R:
518 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
520 case DCCP_SOCKOPT_CCID:
521 case DCCP_SOCKOPT_RX_CCID:
522 case DCCP_SOCKOPT_TX_CCID:
523 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
526 if (optlen < (int)sizeof(int))
529 if (copy_from_sockptr(&val, optval, sizeof(int)))
532 if (optname == DCCP_SOCKOPT_SERVICE)
533 return dccp_setsockopt_service(sk, val, optval, optlen);
537 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
538 if (dp->dccps_role != DCCP_ROLE_SERVER)
541 dp->dccps_server_timewait = (val != 0);
543 case DCCP_SOCKOPT_SEND_CSCOV:
544 err = dccp_setsockopt_cscov(sk, val, false);
546 case DCCP_SOCKOPT_RECV_CSCOV:
547 err = dccp_setsockopt_cscov(sk, val, true);
549 case DCCP_SOCKOPT_QPOLICY_ID:
550 if (sk->sk_state != DCCP_CLOSED)
552 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
555 dp->dccps_qpolicy = val;
557 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
561 dp->dccps_tx_qlen = val;
572 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
575 if (level != SOL_DCCP)
576 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
579 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
582 EXPORT_SYMBOL_GPL(dccp_setsockopt);
584 static int dccp_getsockopt_service(struct sock *sk, int len,
585 __be32 __user *optval,
588 const struct dccp_sock *dp = dccp_sk(sk);
589 const struct dccp_service_list *sl;
590 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
593 if ((sl = dp->dccps_service_list) != NULL) {
594 slen = sl->dccpsl_nr * sizeof(u32);
603 if (put_user(total_len, optlen) ||
604 put_user(dp->dccps_service, optval) ||
605 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
612 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
613 char __user *optval, int __user *optlen)
615 struct dccp_sock *dp;
618 if (get_user(len, optlen))
621 if (len < (int)sizeof(int))
627 case DCCP_SOCKOPT_PACKET_SIZE:
628 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
630 case DCCP_SOCKOPT_SERVICE:
631 return dccp_getsockopt_service(sk, len,
632 (__be32 __user *)optval, optlen);
633 case DCCP_SOCKOPT_GET_CUR_MPS:
634 val = READ_ONCE(dp->dccps_mss_cache);
636 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
637 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
638 case DCCP_SOCKOPT_TX_CCID:
639 val = ccid_get_current_tx_ccid(dp);
643 case DCCP_SOCKOPT_RX_CCID:
644 val = ccid_get_current_rx_ccid(dp);
648 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
649 val = dp->dccps_server_timewait;
651 case DCCP_SOCKOPT_SEND_CSCOV:
652 val = dp->dccps_pcslen;
654 case DCCP_SOCKOPT_RECV_CSCOV:
655 val = dp->dccps_pcrlen;
657 case DCCP_SOCKOPT_QPOLICY_ID:
658 val = dp->dccps_qpolicy;
660 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
661 val = dp->dccps_tx_qlen;
664 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
665 len, (u32 __user *)optval, optlen);
667 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
668 len, (u32 __user *)optval, optlen);
674 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
680 int dccp_getsockopt(struct sock *sk, int level, int optname,
681 char __user *optval, int __user *optlen)
683 if (level != SOL_DCCP)
684 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
687 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
690 EXPORT_SYMBOL_GPL(dccp_getsockopt);
692 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
694 struct cmsghdr *cmsg;
697 * Assign an (opaque) qpolicy priority value to skb->priority.
699 * We are overloading this skb field for use with the qpolicy subystem.
700 * The skb->priority is normally used for the SO_PRIORITY option, which
701 * is initialised from sk_priority. Since the assignment of sk_priority
702 * to skb->priority happens later (on layer 3), we overload this field
703 * for use with queueing priorities as long as the skb is on layer 4.
704 * The default priority value (if nothing is set) is 0.
708 for_each_cmsghdr(cmsg, msg) {
709 if (!CMSG_OK(msg, cmsg))
712 if (cmsg->cmsg_level != SOL_DCCP)
715 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
716 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
719 switch (cmsg->cmsg_type) {
720 case DCCP_SCM_PRIORITY:
721 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
723 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
732 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
734 const struct dccp_sock *dp = dccp_sk(sk);
735 const int flags = msg->msg_flags;
736 const int noblock = flags & MSG_DONTWAIT;
741 trace_dccp_probe(sk, len);
743 if (len > READ_ONCE(dp->dccps_mss_cache))
748 timeo = sock_sndtimeo(sk, noblock);
751 * We have to use sk_stream_wait_connect here to set sk_write_pending,
752 * so that the trick in dccp_rcv_request_sent_state_process.
754 /* Wait for a connection to finish. */
755 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
756 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
759 size = sk->sk_prot->max_header + len;
761 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
766 if (dccp_qpolicy_full(sk)) {
771 if (sk->sk_state == DCCP_CLOSED) {
776 /* We need to check dccps_mss_cache after socket is locked. */
777 if (len > dp->dccps_mss_cache) {
782 skb_reserve(skb, sk->sk_prot->max_header);
783 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
787 rc = dccp_msghdr_parse(msg, skb);
791 dccp_qpolicy_push(sk, skb);
793 * The xmit_timer is set if the TX CCID is rate-based and will expire
794 * when congestion control permits to release further packets into the
795 * network. Window-based CCIDs do not use this timer.
797 if (!timer_pending(&dp->dccps_xmit_timer))
807 EXPORT_SYMBOL_GPL(dccp_sendmsg);
809 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
812 const struct dccp_hdr *dh;
817 if (sk->sk_state == DCCP_LISTEN) {
822 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
825 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
828 goto verify_sock_status;
832 switch (dh->dccph_type) {
834 case DCCP_PKT_DATAACK:
838 case DCCP_PKT_CLOSEREQ:
839 if (!(flags & MSG_PEEK))
840 dccp_finish_passive_close(sk);
843 dccp_pr_debug("found fin (%s) ok!\n",
844 dccp_packet_name(dh->dccph_type));
848 dccp_pr_debug("packet_type=%s\n",
849 dccp_packet_name(dh->dccph_type));
853 if (sock_flag(sk, SOCK_DONE)) {
859 len = sock_error(sk);
863 if (sk->sk_shutdown & RCV_SHUTDOWN) {
868 if (sk->sk_state == DCCP_CLOSED) {
869 if (!sock_flag(sk, SOCK_DONE)) {
870 /* This occurs when user tries to read
871 * from never connected socket.
885 if (signal_pending(current)) {
886 len = sock_intr_errno(timeo);
890 sk_wait_data(sk, &timeo, NULL);
895 else if (len < skb->len)
896 msg->msg_flags |= MSG_TRUNC;
898 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
899 /* Exception. Bailout! */
903 if (flags & MSG_TRUNC)
906 if (!(flags & MSG_PEEK))
915 EXPORT_SYMBOL_GPL(dccp_recvmsg);
917 int inet_dccp_listen(struct socket *sock, int backlog)
919 struct sock *sk = sock->sk;
920 unsigned char old_state;
926 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
929 old_state = sk->sk_state;
930 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
933 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
934 /* Really, if the socket is already in listen state
935 * we can only allow the backlog to be adjusted.
937 if (old_state != DCCP_LISTEN) {
938 struct dccp_sock *dp = dccp_sk(sk);
940 dp->dccps_role = DCCP_ROLE_LISTEN;
942 /* do not start to listen if feature negotiation setup fails */
943 if (dccp_feat_finalise_settings(dp)) {
948 err = inet_csk_listen_start(sk);
959 EXPORT_SYMBOL_GPL(inet_dccp_listen);
961 static void dccp_terminate_connection(struct sock *sk)
963 u8 next_state = DCCP_CLOSED;
965 switch (sk->sk_state) {
966 case DCCP_PASSIVE_CLOSE:
967 case DCCP_PASSIVE_CLOSEREQ:
968 dccp_finish_passive_close(sk);
971 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
972 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
975 dccp_send_close(sk, 1);
977 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
978 !dccp_sk(sk)->dccps_server_timewait)
979 next_state = DCCP_ACTIVE_CLOSEREQ;
981 next_state = DCCP_CLOSING;
984 dccp_set_state(sk, next_state);
988 void dccp_close(struct sock *sk, long timeout)
990 struct dccp_sock *dp = dccp_sk(sk);
992 u32 data_was_unread = 0;
997 sk->sk_shutdown = SHUTDOWN_MASK;
999 if (sk->sk_state == DCCP_LISTEN) {
1000 dccp_set_state(sk, DCCP_CLOSED);
1003 inet_csk_listen_stop(sk);
1005 goto adjudge_to_death;
1008 sk_stop_timer(sk, &dp->dccps_xmit_timer);
1011 * We need to flush the recv. buffs. We do this only on the
1012 * descriptor close, not protocol-sourced closes, because the
1013 *reader process may not have drained the data yet!
1015 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1016 data_was_unread += skb->len;
1020 /* If socket has been already reset kill it. */
1021 if (sk->sk_state == DCCP_CLOSED)
1022 goto adjudge_to_death;
1024 if (data_was_unread) {
1025 /* Unread data was tossed, send an appropriate Reset Code */
1026 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1027 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1028 dccp_set_state(sk, DCCP_CLOSED);
1029 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1030 /* Check zero linger _after_ checking for unread data. */
1031 sk->sk_prot->disconnect(sk, 0);
1032 } else if (sk->sk_state != DCCP_CLOSED) {
1034 * Normal connection termination. May need to wait if there are
1035 * still packets in the TX queue that are delayed by the CCID.
1037 dccp_flush_write_queue(sk, &timeout);
1038 dccp_terminate_connection(sk);
1042 * Flush write queue. This may be necessary in several cases:
1043 * - we have been closed by the peer but still have application data;
1044 * - abortive termination (unread data or zero linger time),
1045 * - normal termination but queue could not be flushed within time limit
1047 __skb_queue_purge(&sk->sk_write_queue);
1049 sk_stream_wait_close(sk, timeout);
1052 state = sk->sk_state;
1057 * It is the last release_sock in its life. It will remove backlog.
1061 * Now socket is owned by kernel and we acquire BH lock
1062 * to finish close. No need to check for user refs.
1066 WARN_ON(sock_owned_by_user(sk));
1068 this_cpu_inc(dccp_orphan_count);
1070 /* Have we already been destroyed by a softirq or backlog? */
1071 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1074 if (sk->sk_state == DCCP_CLOSED)
1075 inet_csk_destroy_sock(sk);
1077 /* Otherwise, socket is reprieved until protocol close. */
1085 EXPORT_SYMBOL_GPL(dccp_close);
1087 void dccp_shutdown(struct sock *sk, int how)
1089 dccp_pr_debug("called shutdown(%x)\n", how);
1092 EXPORT_SYMBOL_GPL(dccp_shutdown);
1094 static inline int __init dccp_mib_init(void)
1096 dccp_statistics = alloc_percpu(struct dccp_mib);
1097 if (!dccp_statistics)
1102 static inline void dccp_mib_exit(void)
1104 free_percpu(dccp_statistics);
1107 static int thash_entries;
1108 module_param(thash_entries, int, 0444);
1109 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1111 #ifdef CONFIG_IP_DCCP_DEBUG
1113 module_param(dccp_debug, bool, 0644);
1114 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1116 EXPORT_SYMBOL_GPL(dccp_debug);
1119 static int __init dccp_init(void)
1122 unsigned long nr_pages = totalram_pages();
1123 int ehash_order, bhash_order, i;
1126 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1127 sizeof_field(struct sk_buff, cb));
1128 rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
1132 dccp_hashinfo.bind_bucket_cachep =
1133 kmem_cache_create("dccp_bind_bucket",
1134 sizeof(struct inet_bind_bucket), 0,
1135 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1136 if (!dccp_hashinfo.bind_bucket_cachep)
1137 goto out_free_hashinfo2;
1138 dccp_hashinfo.bind2_bucket_cachep =
1139 kmem_cache_create("dccp_bind2_bucket",
1140 sizeof(struct inet_bind2_bucket), 0,
1141 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1142 if (!dccp_hashinfo.bind2_bucket_cachep)
1143 goto out_free_bind_bucket_cachep;
1146 * Size and allocate the main established and bind bucket
1149 * The methodology is similar to that of the buffer cache.
1151 if (nr_pages >= (128 * 1024))
1152 goal = nr_pages >> (21 - PAGE_SHIFT);
1154 goal = nr_pages >> (23 - PAGE_SHIFT);
1157 goal = (thash_entries *
1158 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1159 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1162 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1163 sizeof(struct inet_ehash_bucket);
1165 while (hash_size & (hash_size - 1))
1167 dccp_hashinfo.ehash_mask = hash_size - 1;
1168 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1169 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1170 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1172 if (!dccp_hashinfo.ehash) {
1173 DCCP_CRIT("Failed to allocate DCCP established hash table");
1174 goto out_free_bind2_bucket_cachep;
1177 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1178 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1180 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1181 goto out_free_dccp_ehash;
1183 bhash_order = ehash_order;
1186 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1187 sizeof(struct inet_bind_hashbucket);
1188 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1191 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1192 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1193 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1195 if (!dccp_hashinfo.bhash) {
1196 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1197 goto out_free_dccp_locks;
1200 dccp_hashinfo.bhash2 = (struct inet_bind_hashbucket *)
1201 __get_free_pages(GFP_ATOMIC | __GFP_NOWARN, bhash_order);
1203 if (!dccp_hashinfo.bhash2) {
1204 DCCP_CRIT("Failed to allocate DCCP bind2 hash table");
1205 goto out_free_dccp_bhash;
1208 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1209 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1210 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1211 spin_lock_init(&dccp_hashinfo.bhash2[i].lock);
1212 INIT_HLIST_HEAD(&dccp_hashinfo.bhash2[i].chain);
1215 dccp_hashinfo.pernet = false;
1217 rc = dccp_mib_init();
1219 goto out_free_dccp_bhash2;
1221 rc = dccp_ackvec_init();
1223 goto out_free_dccp_mib;
1225 rc = dccp_sysctl_init();
1227 goto out_ackvec_exit;
1229 rc = ccid_initialize_builtins();
1231 goto out_sysctl_exit;
1233 dccp_timestamping_init();
1243 out_free_dccp_bhash2:
1244 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1245 out_free_dccp_bhash:
1246 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1247 out_free_dccp_locks:
1248 inet_ehash_locks_free(&dccp_hashinfo);
1249 out_free_dccp_ehash:
1250 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1251 out_free_bind2_bucket_cachep:
1252 kmem_cache_destroy(dccp_hashinfo.bind2_bucket_cachep);
1253 out_free_bind_bucket_cachep:
1254 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1256 inet_hashinfo2_free_mod(&dccp_hashinfo);
1258 dccp_hashinfo.bhash = NULL;
1259 dccp_hashinfo.bhash2 = NULL;
1260 dccp_hashinfo.ehash = NULL;
1261 dccp_hashinfo.bind_bucket_cachep = NULL;
1262 dccp_hashinfo.bind2_bucket_cachep = NULL;
1266 static void __exit dccp_fini(void)
1268 int bhash_order = get_order(dccp_hashinfo.bhash_size *
1269 sizeof(struct inet_bind_hashbucket));
1271 ccid_cleanup_builtins();
1273 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1274 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1275 free_pages((unsigned long)dccp_hashinfo.ehash,
1276 get_order((dccp_hashinfo.ehash_mask + 1) *
1277 sizeof(struct inet_ehash_bucket)));
1278 inet_ehash_locks_free(&dccp_hashinfo);
1279 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1282 inet_hashinfo2_free_mod(&dccp_hashinfo);
1285 module_init(dccp_init);
1286 module_exit(dccp_fini);
1288 MODULE_LICENSE("GPL");
1289 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1290 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");