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 inet_bhash2_reset_saddr(sk);
285 sock_reset_flag(sk, SOCK_DONE);
287 icsk->icsk_backoff = 0;
288 inet_csk_delack_init(sk);
291 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
297 EXPORT_SYMBOL_GPL(dccp_disconnect);
300 * Wait for a DCCP event.
302 * Note that we don't need to lock the socket, as the upper poll layers
303 * take care of normal races (between the test and the event) and we don't
304 * go look at any of the socket buffers directly.
306 __poll_t dccp_poll(struct file *file, struct socket *sock,
310 struct sock *sk = sock->sk;
312 sock_poll_wait(file, sock, wait);
313 if (sk->sk_state == DCCP_LISTEN)
314 return inet_csk_listen_poll(sk);
316 /* Socket is not locked. We are protected from async events
317 by poll logic and correct handling of state changes
318 made by another threads is impossible in any case.
325 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
327 if (sk->sk_shutdown & RCV_SHUTDOWN)
328 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
331 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
332 if (atomic_read(&sk->sk_rmem_alloc) > 0)
333 mask |= EPOLLIN | EPOLLRDNORM;
335 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
336 if (sk_stream_is_writeable(sk)) {
337 mask |= EPOLLOUT | EPOLLWRNORM;
338 } else { /* send SIGIO later */
339 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
340 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
342 /* Race breaker. If space is freed after
343 * wspace test but before the flags are set,
344 * IO signal will be lost.
346 if (sk_stream_is_writeable(sk))
347 mask |= EPOLLOUT | EPOLLWRNORM;
354 EXPORT_SYMBOL_GPL(dccp_poll);
356 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
362 if (sk->sk_state == DCCP_LISTEN)
367 int amount = sk_wmem_alloc_get(sk);
368 /* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and
369 * always 0, comparably to UDP.
372 rc = put_user(amount, (int __user *)arg);
377 unsigned long amount = 0;
379 skb = skb_peek(&sk->sk_receive_queue);
382 * We will only return the amount of this packet since
383 * that is all that will be read.
387 rc = put_user(amount, (int __user *)arg);
399 EXPORT_SYMBOL_GPL(dccp_ioctl);
401 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
402 sockptr_t optval, unsigned int optlen)
404 struct dccp_sock *dp = dccp_sk(sk);
405 struct dccp_service_list *sl = NULL;
407 if (service == DCCP_SERVICE_INVALID_VALUE ||
408 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
411 if (optlen > sizeof(service)) {
412 sl = kmalloc(optlen, GFP_KERNEL);
416 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
417 if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
418 sizeof(service), optlen - sizeof(service)) ||
419 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
426 dp->dccps_service = service;
428 kfree(dp->dccps_service_list);
430 dp->dccps_service_list = sl;
435 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
440 if (cscov < 0 || cscov > 15)
443 * Populate a list of permissible values, in the range cscov...15. This
444 * is necessary since feature negotiation of single values only works if
445 * both sides incidentally choose the same value. Since the list starts
446 * lowest-value first, negotiation will pick the smallest shared value.
452 list = kmalloc(len, GFP_KERNEL);
456 for (i = 0; i < len; i++)
459 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
463 dccp_sk(sk)->dccps_pcrlen = cscov;
465 dccp_sk(sk)->dccps_pcslen = cscov;
471 static int dccp_setsockopt_ccid(struct sock *sk, int type,
472 sockptr_t optval, unsigned int optlen)
477 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
480 val = memdup_sockptr(optval, optlen);
485 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
486 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
488 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
489 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
496 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
497 sockptr_t optval, unsigned int optlen)
499 struct dccp_sock *dp = dccp_sk(sk);
503 case DCCP_SOCKOPT_PACKET_SIZE:
504 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
506 case DCCP_SOCKOPT_CHANGE_L:
507 case DCCP_SOCKOPT_CHANGE_R:
508 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
510 case DCCP_SOCKOPT_CCID:
511 case DCCP_SOCKOPT_RX_CCID:
512 case DCCP_SOCKOPT_TX_CCID:
513 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
516 if (optlen < (int)sizeof(int))
519 if (copy_from_sockptr(&val, optval, sizeof(int)))
522 if (optname == DCCP_SOCKOPT_SERVICE)
523 return dccp_setsockopt_service(sk, val, optval, optlen);
527 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
528 if (dp->dccps_role != DCCP_ROLE_SERVER)
531 dp->dccps_server_timewait = (val != 0);
533 case DCCP_SOCKOPT_SEND_CSCOV:
534 err = dccp_setsockopt_cscov(sk, val, false);
536 case DCCP_SOCKOPT_RECV_CSCOV:
537 err = dccp_setsockopt_cscov(sk, val, true);
539 case DCCP_SOCKOPT_QPOLICY_ID:
540 if (sk->sk_state != DCCP_CLOSED)
542 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
545 dp->dccps_qpolicy = val;
547 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
551 dp->dccps_tx_qlen = val;
562 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
565 if (level != SOL_DCCP)
566 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
569 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
572 EXPORT_SYMBOL_GPL(dccp_setsockopt);
574 static int dccp_getsockopt_service(struct sock *sk, int len,
575 __be32 __user *optval,
578 const struct dccp_sock *dp = dccp_sk(sk);
579 const struct dccp_service_list *sl;
580 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
583 if ((sl = dp->dccps_service_list) != NULL) {
584 slen = sl->dccpsl_nr * sizeof(u32);
593 if (put_user(total_len, optlen) ||
594 put_user(dp->dccps_service, optval) ||
595 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
602 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
603 char __user *optval, int __user *optlen)
605 struct dccp_sock *dp;
608 if (get_user(len, optlen))
611 if (len < (int)sizeof(int))
617 case DCCP_SOCKOPT_PACKET_SIZE:
618 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
620 case DCCP_SOCKOPT_SERVICE:
621 return dccp_getsockopt_service(sk, len,
622 (__be32 __user *)optval, optlen);
623 case DCCP_SOCKOPT_GET_CUR_MPS:
624 val = dp->dccps_mss_cache;
626 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
627 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
628 case DCCP_SOCKOPT_TX_CCID:
629 val = ccid_get_current_tx_ccid(dp);
633 case DCCP_SOCKOPT_RX_CCID:
634 val = ccid_get_current_rx_ccid(dp);
638 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
639 val = dp->dccps_server_timewait;
641 case DCCP_SOCKOPT_SEND_CSCOV:
642 val = dp->dccps_pcslen;
644 case DCCP_SOCKOPT_RECV_CSCOV:
645 val = dp->dccps_pcrlen;
647 case DCCP_SOCKOPT_QPOLICY_ID:
648 val = dp->dccps_qpolicy;
650 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
651 val = dp->dccps_tx_qlen;
654 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
655 len, (u32 __user *)optval, optlen);
657 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
658 len, (u32 __user *)optval, optlen);
664 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
670 int dccp_getsockopt(struct sock *sk, int level, int optname,
671 char __user *optval, int __user *optlen)
673 if (level != SOL_DCCP)
674 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
677 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
680 EXPORT_SYMBOL_GPL(dccp_getsockopt);
682 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
684 struct cmsghdr *cmsg;
687 * Assign an (opaque) qpolicy priority value to skb->priority.
689 * We are overloading this skb field for use with the qpolicy subystem.
690 * The skb->priority is normally used for the SO_PRIORITY option, which
691 * is initialised from sk_priority. Since the assignment of sk_priority
692 * to skb->priority happens later (on layer 3), we overload this field
693 * for use with queueing priorities as long as the skb is on layer 4.
694 * The default priority value (if nothing is set) is 0.
698 for_each_cmsghdr(cmsg, msg) {
699 if (!CMSG_OK(msg, cmsg))
702 if (cmsg->cmsg_level != SOL_DCCP)
705 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
706 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
709 switch (cmsg->cmsg_type) {
710 case DCCP_SCM_PRIORITY:
711 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
713 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
722 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
724 const struct dccp_sock *dp = dccp_sk(sk);
725 const int flags = msg->msg_flags;
726 const int noblock = flags & MSG_DONTWAIT;
731 trace_dccp_probe(sk, len);
733 if (len > dp->dccps_mss_cache)
738 timeo = sock_sndtimeo(sk, noblock);
741 * We have to use sk_stream_wait_connect here to set sk_write_pending,
742 * so that the trick in dccp_rcv_request_sent_state_process.
744 /* Wait for a connection to finish. */
745 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
746 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
749 size = sk->sk_prot->max_header + len;
751 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
756 if (dccp_qpolicy_full(sk)) {
761 if (sk->sk_state == DCCP_CLOSED) {
766 skb_reserve(skb, sk->sk_prot->max_header);
767 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
771 rc = dccp_msghdr_parse(msg, skb);
775 dccp_qpolicy_push(sk, skb);
777 * The xmit_timer is set if the TX CCID is rate-based and will expire
778 * when congestion control permits to release further packets into the
779 * network. Window-based CCIDs do not use this timer.
781 if (!timer_pending(&dp->dccps_xmit_timer))
791 EXPORT_SYMBOL_GPL(dccp_sendmsg);
793 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
796 const struct dccp_hdr *dh;
801 if (sk->sk_state == DCCP_LISTEN) {
806 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
809 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
812 goto verify_sock_status;
816 switch (dh->dccph_type) {
818 case DCCP_PKT_DATAACK:
822 case DCCP_PKT_CLOSEREQ:
823 if (!(flags & MSG_PEEK))
824 dccp_finish_passive_close(sk);
827 dccp_pr_debug("found fin (%s) ok!\n",
828 dccp_packet_name(dh->dccph_type));
832 dccp_pr_debug("packet_type=%s\n",
833 dccp_packet_name(dh->dccph_type));
837 if (sock_flag(sk, SOCK_DONE)) {
843 len = sock_error(sk);
847 if (sk->sk_shutdown & RCV_SHUTDOWN) {
852 if (sk->sk_state == DCCP_CLOSED) {
853 if (!sock_flag(sk, SOCK_DONE)) {
854 /* This occurs when user tries to read
855 * from never connected socket.
869 if (signal_pending(current)) {
870 len = sock_intr_errno(timeo);
874 sk_wait_data(sk, &timeo, NULL);
879 else if (len < skb->len)
880 msg->msg_flags |= MSG_TRUNC;
882 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
883 /* Exception. Bailout! */
887 if (flags & MSG_TRUNC)
890 if (!(flags & MSG_PEEK))
899 EXPORT_SYMBOL_GPL(dccp_recvmsg);
901 int inet_dccp_listen(struct socket *sock, int backlog)
903 struct sock *sk = sock->sk;
904 unsigned char old_state;
910 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
913 old_state = sk->sk_state;
914 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
917 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
918 /* Really, if the socket is already in listen state
919 * we can only allow the backlog to be adjusted.
921 if (old_state != DCCP_LISTEN) {
922 struct dccp_sock *dp = dccp_sk(sk);
924 dp->dccps_role = DCCP_ROLE_LISTEN;
926 /* do not start to listen if feature negotiation setup fails */
927 if (dccp_feat_finalise_settings(dp)) {
932 err = inet_csk_listen_start(sk);
943 EXPORT_SYMBOL_GPL(inet_dccp_listen);
945 static void dccp_terminate_connection(struct sock *sk)
947 u8 next_state = DCCP_CLOSED;
949 switch (sk->sk_state) {
950 case DCCP_PASSIVE_CLOSE:
951 case DCCP_PASSIVE_CLOSEREQ:
952 dccp_finish_passive_close(sk);
955 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
956 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
959 dccp_send_close(sk, 1);
961 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
962 !dccp_sk(sk)->dccps_server_timewait)
963 next_state = DCCP_ACTIVE_CLOSEREQ;
965 next_state = DCCP_CLOSING;
968 dccp_set_state(sk, next_state);
972 void dccp_close(struct sock *sk, long timeout)
974 struct dccp_sock *dp = dccp_sk(sk);
976 u32 data_was_unread = 0;
981 sk->sk_shutdown = SHUTDOWN_MASK;
983 if (sk->sk_state == DCCP_LISTEN) {
984 dccp_set_state(sk, DCCP_CLOSED);
987 inet_csk_listen_stop(sk);
989 goto adjudge_to_death;
992 sk_stop_timer(sk, &dp->dccps_xmit_timer);
995 * We need to flush the recv. buffs. We do this only on the
996 * descriptor close, not protocol-sourced closes, because the
997 *reader process may not have drained the data yet!
999 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1000 data_was_unread += skb->len;
1004 /* If socket has been already reset kill it. */
1005 if (sk->sk_state == DCCP_CLOSED)
1006 goto adjudge_to_death;
1008 if (data_was_unread) {
1009 /* Unread data was tossed, send an appropriate Reset Code */
1010 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1011 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1012 dccp_set_state(sk, DCCP_CLOSED);
1013 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1014 /* Check zero linger _after_ checking for unread data. */
1015 sk->sk_prot->disconnect(sk, 0);
1016 } else if (sk->sk_state != DCCP_CLOSED) {
1018 * Normal connection termination. May need to wait if there are
1019 * still packets in the TX queue that are delayed by the CCID.
1021 dccp_flush_write_queue(sk, &timeout);
1022 dccp_terminate_connection(sk);
1026 * Flush write queue. This may be necessary in several cases:
1027 * - we have been closed by the peer but still have application data;
1028 * - abortive termination (unread data or zero linger time),
1029 * - normal termination but queue could not be flushed within time limit
1031 __skb_queue_purge(&sk->sk_write_queue);
1033 sk_stream_wait_close(sk, timeout);
1036 state = sk->sk_state;
1041 * It is the last release_sock in its life. It will remove backlog.
1045 * Now socket is owned by kernel and we acquire BH lock
1046 * to finish close. No need to check for user refs.
1050 WARN_ON(sock_owned_by_user(sk));
1052 this_cpu_inc(dccp_orphan_count);
1054 /* Have we already been destroyed by a softirq or backlog? */
1055 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1058 if (sk->sk_state == DCCP_CLOSED)
1059 inet_csk_destroy_sock(sk);
1061 /* Otherwise, socket is reprieved until protocol close. */
1069 EXPORT_SYMBOL_GPL(dccp_close);
1071 void dccp_shutdown(struct sock *sk, int how)
1073 dccp_pr_debug("called shutdown(%x)\n", how);
1076 EXPORT_SYMBOL_GPL(dccp_shutdown);
1078 static inline int __init dccp_mib_init(void)
1080 dccp_statistics = alloc_percpu(struct dccp_mib);
1081 if (!dccp_statistics)
1086 static inline void dccp_mib_exit(void)
1088 free_percpu(dccp_statistics);
1091 static int thash_entries;
1092 module_param(thash_entries, int, 0444);
1093 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1095 #ifdef CONFIG_IP_DCCP_DEBUG
1097 module_param(dccp_debug, bool, 0644);
1098 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1100 EXPORT_SYMBOL_GPL(dccp_debug);
1103 static int __init dccp_init(void)
1106 unsigned long nr_pages = totalram_pages();
1107 int ehash_order, bhash_order, i;
1110 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1111 sizeof_field(struct sk_buff, cb));
1112 rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
1116 dccp_hashinfo.bind_bucket_cachep =
1117 kmem_cache_create("dccp_bind_bucket",
1118 sizeof(struct inet_bind_bucket), 0,
1119 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1120 if (!dccp_hashinfo.bind_bucket_cachep)
1121 goto out_free_hashinfo2;
1122 dccp_hashinfo.bind2_bucket_cachep =
1123 kmem_cache_create("dccp_bind2_bucket",
1124 sizeof(struct inet_bind2_bucket), 0,
1125 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1126 if (!dccp_hashinfo.bind2_bucket_cachep)
1127 goto out_free_bind_bucket_cachep;
1130 * Size and allocate the main established and bind bucket
1133 * The methodology is similar to that of the buffer cache.
1135 if (nr_pages >= (128 * 1024))
1136 goal = nr_pages >> (21 - PAGE_SHIFT);
1138 goal = nr_pages >> (23 - PAGE_SHIFT);
1141 goal = (thash_entries *
1142 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1143 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1146 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1147 sizeof(struct inet_ehash_bucket);
1149 while (hash_size & (hash_size - 1))
1151 dccp_hashinfo.ehash_mask = hash_size - 1;
1152 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1153 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1154 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1156 if (!dccp_hashinfo.ehash) {
1157 DCCP_CRIT("Failed to allocate DCCP established hash table");
1158 goto out_free_bind2_bucket_cachep;
1161 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1162 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1164 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1165 goto out_free_dccp_ehash;
1167 bhash_order = ehash_order;
1170 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1171 sizeof(struct inet_bind_hashbucket);
1172 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1175 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1176 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1177 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1179 if (!dccp_hashinfo.bhash) {
1180 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1181 goto out_free_dccp_locks;
1184 dccp_hashinfo.bhash2 = (struct inet_bind_hashbucket *)
1185 __get_free_pages(GFP_ATOMIC | __GFP_NOWARN, bhash_order);
1187 if (!dccp_hashinfo.bhash2) {
1188 DCCP_CRIT("Failed to allocate DCCP bind2 hash table");
1189 goto out_free_dccp_bhash;
1192 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1193 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1194 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1195 spin_lock_init(&dccp_hashinfo.bhash2[i].lock);
1196 INIT_HLIST_HEAD(&dccp_hashinfo.bhash2[i].chain);
1199 dccp_hashinfo.pernet = false;
1201 rc = dccp_mib_init();
1203 goto out_free_dccp_bhash2;
1205 rc = dccp_ackvec_init();
1207 goto out_free_dccp_mib;
1209 rc = dccp_sysctl_init();
1211 goto out_ackvec_exit;
1213 rc = ccid_initialize_builtins();
1215 goto out_sysctl_exit;
1217 dccp_timestamping_init();
1227 out_free_dccp_bhash2:
1228 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1229 out_free_dccp_bhash:
1230 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1231 out_free_dccp_locks:
1232 inet_ehash_locks_free(&dccp_hashinfo);
1233 out_free_dccp_ehash:
1234 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1235 out_free_bind2_bucket_cachep:
1236 kmem_cache_destroy(dccp_hashinfo.bind2_bucket_cachep);
1237 out_free_bind_bucket_cachep:
1238 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1240 inet_hashinfo2_free_mod(&dccp_hashinfo);
1242 dccp_hashinfo.bhash = NULL;
1243 dccp_hashinfo.bhash2 = NULL;
1244 dccp_hashinfo.ehash = NULL;
1245 dccp_hashinfo.bind_bucket_cachep = NULL;
1246 dccp_hashinfo.bind2_bucket_cachep = NULL;
1250 static void __exit dccp_fini(void)
1252 int bhash_order = get_order(dccp_hashinfo.bhash_size *
1253 sizeof(struct inet_bind_hashbucket));
1255 ccid_cleanup_builtins();
1257 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1258 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1259 free_pages((unsigned long)dccp_hashinfo.ehash,
1260 get_order((dccp_hashinfo.ehash_mask + 1) *
1261 sizeof(struct inet_ehash_bucket)));
1262 inet_ehash_locks_free(&dccp_hashinfo);
1263 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1266 inet_hashinfo2_free_mod(&dccp_hashinfo);
1269 module_init(dccp_init);
1270 module_exit(dccp_fini);
1272 MODULE_LICENSE("GPL");
1273 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1274 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");