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 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
289 inet_reset_saddr(sk);
292 sock_reset_flag(sk, SOCK_DONE);
294 icsk->icsk_backoff = 0;
295 inet_csk_delack_init(sk);
298 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
304 EXPORT_SYMBOL_GPL(dccp_disconnect);
307 * Wait for a DCCP event.
309 * Note that we don't need to lock the socket, as the upper poll layers
310 * take care of normal races (between the test and the event) and we don't
311 * go look at any of the socket buffers directly.
313 __poll_t dccp_poll(struct file *file, struct socket *sock,
317 struct sock *sk = sock->sk;
319 sock_poll_wait(file, sock, wait);
320 if (sk->sk_state == DCCP_LISTEN)
321 return inet_csk_listen_poll(sk);
323 /* Socket is not locked. We are protected from async events
324 by poll logic and correct handling of state changes
325 made by another threads is impossible in any case.
332 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == DCCP_CLOSED)
334 if (sk->sk_shutdown & RCV_SHUTDOWN)
335 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
338 if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_RESPOND)) {
339 if (atomic_read(&sk->sk_rmem_alloc) > 0)
340 mask |= EPOLLIN | EPOLLRDNORM;
342 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
343 if (sk_stream_is_writeable(sk)) {
344 mask |= EPOLLOUT | EPOLLWRNORM;
345 } else { /* send SIGIO later */
346 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
347 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
349 /* Race breaker. If space is freed after
350 * wspace test but before the flags are set,
351 * IO signal will be lost.
353 if (sk_stream_is_writeable(sk))
354 mask |= EPOLLOUT | EPOLLWRNORM;
361 EXPORT_SYMBOL_GPL(dccp_poll);
363 int dccp_ioctl(struct sock *sk, int cmd, unsigned long arg)
369 if (sk->sk_state == DCCP_LISTEN)
374 int amount = sk_wmem_alloc_get(sk);
375 /* Using sk_wmem_alloc here because sk_wmem_queued is not used by DCCP and
376 * always 0, comparably to UDP.
379 rc = put_user(amount, (int __user *)arg);
384 unsigned long amount = 0;
386 skb = skb_peek(&sk->sk_receive_queue);
389 * We will only return the amount of this packet since
390 * that is all that will be read.
394 rc = put_user(amount, (int __user *)arg);
406 EXPORT_SYMBOL_GPL(dccp_ioctl);
408 static int dccp_setsockopt_service(struct sock *sk, const __be32 service,
409 sockptr_t optval, unsigned int optlen)
411 struct dccp_sock *dp = dccp_sk(sk);
412 struct dccp_service_list *sl = NULL;
414 if (service == DCCP_SERVICE_INVALID_VALUE ||
415 optlen > DCCP_SERVICE_LIST_MAX_LEN * sizeof(u32))
418 if (optlen > sizeof(service)) {
419 sl = kmalloc(optlen, GFP_KERNEL);
423 sl->dccpsl_nr = optlen / sizeof(u32) - 1;
424 if (copy_from_sockptr_offset(sl->dccpsl_list, optval,
425 sizeof(service), optlen - sizeof(service)) ||
426 dccp_list_has_service(sl, DCCP_SERVICE_INVALID_VALUE)) {
433 dp->dccps_service = service;
435 kfree(dp->dccps_service_list);
437 dp->dccps_service_list = sl;
442 static int dccp_setsockopt_cscov(struct sock *sk, int cscov, bool rx)
447 if (cscov < 0 || cscov > 15)
450 * Populate a list of permissible values, in the range cscov...15. This
451 * is necessary since feature negotiation of single values only works if
452 * both sides incidentally choose the same value. Since the list starts
453 * lowest-value first, negotiation will pick the smallest shared value.
459 list = kmalloc(len, GFP_KERNEL);
463 for (i = 0; i < len; i++)
466 rc = dccp_feat_register_sp(sk, DCCPF_MIN_CSUM_COVER, rx, list, len);
470 dccp_sk(sk)->dccps_pcrlen = cscov;
472 dccp_sk(sk)->dccps_pcslen = cscov;
478 static int dccp_setsockopt_ccid(struct sock *sk, int type,
479 sockptr_t optval, unsigned int optlen)
484 if (optlen < 1 || optlen > DCCP_FEAT_MAX_SP_VALS)
487 val = memdup_sockptr(optval, optlen);
492 if (type == DCCP_SOCKOPT_TX_CCID || type == DCCP_SOCKOPT_CCID)
493 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 1, val, optlen);
495 if (!rc && (type == DCCP_SOCKOPT_RX_CCID || type == DCCP_SOCKOPT_CCID))
496 rc = dccp_feat_register_sp(sk, DCCPF_CCID, 0, val, optlen);
503 static int do_dccp_setsockopt(struct sock *sk, int level, int optname,
504 sockptr_t optval, unsigned int optlen)
506 struct dccp_sock *dp = dccp_sk(sk);
510 case DCCP_SOCKOPT_PACKET_SIZE:
511 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
513 case DCCP_SOCKOPT_CHANGE_L:
514 case DCCP_SOCKOPT_CHANGE_R:
515 DCCP_WARN("sockopt(CHANGE_L/R) is deprecated: fix your app\n");
517 case DCCP_SOCKOPT_CCID:
518 case DCCP_SOCKOPT_RX_CCID:
519 case DCCP_SOCKOPT_TX_CCID:
520 return dccp_setsockopt_ccid(sk, optname, optval, optlen);
523 if (optlen < (int)sizeof(int))
526 if (copy_from_sockptr(&val, optval, sizeof(int)))
529 if (optname == DCCP_SOCKOPT_SERVICE)
530 return dccp_setsockopt_service(sk, val, optval, optlen);
534 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
535 if (dp->dccps_role != DCCP_ROLE_SERVER)
538 dp->dccps_server_timewait = (val != 0);
540 case DCCP_SOCKOPT_SEND_CSCOV:
541 err = dccp_setsockopt_cscov(sk, val, false);
543 case DCCP_SOCKOPT_RECV_CSCOV:
544 err = dccp_setsockopt_cscov(sk, val, true);
546 case DCCP_SOCKOPT_QPOLICY_ID:
547 if (sk->sk_state != DCCP_CLOSED)
549 else if (val < 0 || val >= DCCPQ_POLICY_MAX)
552 dp->dccps_qpolicy = val;
554 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
558 dp->dccps_tx_qlen = val;
569 int dccp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
572 if (level != SOL_DCCP)
573 return inet_csk(sk)->icsk_af_ops->setsockopt(sk, level,
576 return do_dccp_setsockopt(sk, level, optname, optval, optlen);
579 EXPORT_SYMBOL_GPL(dccp_setsockopt);
581 static int dccp_getsockopt_service(struct sock *sk, int len,
582 __be32 __user *optval,
585 const struct dccp_sock *dp = dccp_sk(sk);
586 const struct dccp_service_list *sl;
587 int err = -ENOENT, slen = 0, total_len = sizeof(u32);
590 if ((sl = dp->dccps_service_list) != NULL) {
591 slen = sl->dccpsl_nr * sizeof(u32);
600 if (put_user(total_len, optlen) ||
601 put_user(dp->dccps_service, optval) ||
602 (sl != NULL && copy_to_user(optval + 1, sl->dccpsl_list, slen)))
609 static int do_dccp_getsockopt(struct sock *sk, int level, int optname,
610 char __user *optval, int __user *optlen)
612 struct dccp_sock *dp;
615 if (get_user(len, optlen))
618 if (len < (int)sizeof(int))
624 case DCCP_SOCKOPT_PACKET_SIZE:
625 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
627 case DCCP_SOCKOPT_SERVICE:
628 return dccp_getsockopt_service(sk, len,
629 (__be32 __user *)optval, optlen);
630 case DCCP_SOCKOPT_GET_CUR_MPS:
631 val = dp->dccps_mss_cache;
633 case DCCP_SOCKOPT_AVAILABLE_CCIDS:
634 return ccid_getsockopt_builtin_ccids(sk, len, optval, optlen);
635 case DCCP_SOCKOPT_TX_CCID:
636 val = ccid_get_current_tx_ccid(dp);
640 case DCCP_SOCKOPT_RX_CCID:
641 val = ccid_get_current_rx_ccid(dp);
645 case DCCP_SOCKOPT_SERVER_TIMEWAIT:
646 val = dp->dccps_server_timewait;
648 case DCCP_SOCKOPT_SEND_CSCOV:
649 val = dp->dccps_pcslen;
651 case DCCP_SOCKOPT_RECV_CSCOV:
652 val = dp->dccps_pcrlen;
654 case DCCP_SOCKOPT_QPOLICY_ID:
655 val = dp->dccps_qpolicy;
657 case DCCP_SOCKOPT_QPOLICY_TXQLEN:
658 val = dp->dccps_tx_qlen;
661 return ccid_hc_rx_getsockopt(dp->dccps_hc_rx_ccid, sk, optname,
662 len, (u32 __user *)optval, optlen);
664 return ccid_hc_tx_getsockopt(dp->dccps_hc_tx_ccid, sk, optname,
665 len, (u32 __user *)optval, optlen);
671 if (put_user(len, optlen) || copy_to_user(optval, &val, len))
677 int dccp_getsockopt(struct sock *sk, int level, int optname,
678 char __user *optval, int __user *optlen)
680 if (level != SOL_DCCP)
681 return inet_csk(sk)->icsk_af_ops->getsockopt(sk, level,
684 return do_dccp_getsockopt(sk, level, optname, optval, optlen);
687 EXPORT_SYMBOL_GPL(dccp_getsockopt);
689 static int dccp_msghdr_parse(struct msghdr *msg, struct sk_buff *skb)
691 struct cmsghdr *cmsg;
694 * Assign an (opaque) qpolicy priority value to skb->priority.
696 * We are overloading this skb field for use with the qpolicy subystem.
697 * The skb->priority is normally used for the SO_PRIORITY option, which
698 * is initialised from sk_priority. Since the assignment of sk_priority
699 * to skb->priority happens later (on layer 3), we overload this field
700 * for use with queueing priorities as long as the skb is on layer 4.
701 * The default priority value (if nothing is set) is 0.
705 for_each_cmsghdr(cmsg, msg) {
706 if (!CMSG_OK(msg, cmsg))
709 if (cmsg->cmsg_level != SOL_DCCP)
712 if (cmsg->cmsg_type <= DCCP_SCM_QPOLICY_MAX &&
713 !dccp_qpolicy_param_ok(skb->sk, cmsg->cmsg_type))
716 switch (cmsg->cmsg_type) {
717 case DCCP_SCM_PRIORITY:
718 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u32)))
720 skb->priority = *(__u32 *)CMSG_DATA(cmsg);
729 int dccp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
731 const struct dccp_sock *dp = dccp_sk(sk);
732 const int flags = msg->msg_flags;
733 const int noblock = flags & MSG_DONTWAIT;
738 trace_dccp_probe(sk, len);
740 if (len > dp->dccps_mss_cache)
745 timeo = sock_sndtimeo(sk, noblock);
748 * We have to use sk_stream_wait_connect here to set sk_write_pending,
749 * so that the trick in dccp_rcv_request_sent_state_process.
751 /* Wait for a connection to finish. */
752 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
753 if ((rc = sk_stream_wait_connect(sk, &timeo)) != 0)
756 size = sk->sk_prot->max_header + len;
758 skb = sock_alloc_send_skb(sk, size, noblock, &rc);
763 if (dccp_qpolicy_full(sk)) {
768 if (sk->sk_state == DCCP_CLOSED) {
773 skb_reserve(skb, sk->sk_prot->max_header);
774 rc = memcpy_from_msg(skb_put(skb, len), msg, len);
778 rc = dccp_msghdr_parse(msg, skb);
782 dccp_qpolicy_push(sk, skb);
784 * The xmit_timer is set if the TX CCID is rate-based and will expire
785 * when congestion control permits to release further packets into the
786 * network. Window-based CCIDs do not use this timer.
788 if (!timer_pending(&dp->dccps_xmit_timer))
798 EXPORT_SYMBOL_GPL(dccp_sendmsg);
800 int dccp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
803 const struct dccp_hdr *dh;
808 if (sk->sk_state == DCCP_LISTEN) {
813 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
816 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
819 goto verify_sock_status;
823 switch (dh->dccph_type) {
825 case DCCP_PKT_DATAACK:
829 case DCCP_PKT_CLOSEREQ:
830 if (!(flags & MSG_PEEK))
831 dccp_finish_passive_close(sk);
834 dccp_pr_debug("found fin (%s) ok!\n",
835 dccp_packet_name(dh->dccph_type));
839 dccp_pr_debug("packet_type=%s\n",
840 dccp_packet_name(dh->dccph_type));
844 if (sock_flag(sk, SOCK_DONE)) {
850 len = sock_error(sk);
854 if (sk->sk_shutdown & RCV_SHUTDOWN) {
859 if (sk->sk_state == DCCP_CLOSED) {
860 if (!sock_flag(sk, SOCK_DONE)) {
861 /* This occurs when user tries to read
862 * from never connected socket.
876 if (signal_pending(current)) {
877 len = sock_intr_errno(timeo);
881 sk_wait_data(sk, &timeo, NULL);
886 else if (len < skb->len)
887 msg->msg_flags |= MSG_TRUNC;
889 if (skb_copy_datagram_msg(skb, 0, msg, len)) {
890 /* Exception. Bailout! */
894 if (flags & MSG_TRUNC)
897 if (!(flags & MSG_PEEK))
906 EXPORT_SYMBOL_GPL(dccp_recvmsg);
908 int inet_dccp_listen(struct socket *sock, int backlog)
910 struct sock *sk = sock->sk;
911 unsigned char old_state;
917 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_DCCP)
920 old_state = sk->sk_state;
921 if (!((1 << old_state) & (DCCPF_CLOSED | DCCPF_LISTEN)))
924 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
925 /* Really, if the socket is already in listen state
926 * we can only allow the backlog to be adjusted.
928 if (old_state != DCCP_LISTEN) {
929 struct dccp_sock *dp = dccp_sk(sk);
931 dp->dccps_role = DCCP_ROLE_LISTEN;
933 /* do not start to listen if feature negotiation setup fails */
934 if (dccp_feat_finalise_settings(dp)) {
939 err = inet_csk_listen_start(sk);
950 EXPORT_SYMBOL_GPL(inet_dccp_listen);
952 static void dccp_terminate_connection(struct sock *sk)
954 u8 next_state = DCCP_CLOSED;
956 switch (sk->sk_state) {
957 case DCCP_PASSIVE_CLOSE:
958 case DCCP_PASSIVE_CLOSEREQ:
959 dccp_finish_passive_close(sk);
962 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk);
963 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
966 dccp_send_close(sk, 1);
968 if (dccp_sk(sk)->dccps_role == DCCP_ROLE_SERVER &&
969 !dccp_sk(sk)->dccps_server_timewait)
970 next_state = DCCP_ACTIVE_CLOSEREQ;
972 next_state = DCCP_CLOSING;
975 dccp_set_state(sk, next_state);
979 void dccp_close(struct sock *sk, long timeout)
981 struct dccp_sock *dp = dccp_sk(sk);
983 u32 data_was_unread = 0;
988 sk->sk_shutdown = SHUTDOWN_MASK;
990 if (sk->sk_state == DCCP_LISTEN) {
991 dccp_set_state(sk, DCCP_CLOSED);
994 inet_csk_listen_stop(sk);
996 goto adjudge_to_death;
999 sk_stop_timer(sk, &dp->dccps_xmit_timer);
1002 * We need to flush the recv. buffs. We do this only on the
1003 * descriptor close, not protocol-sourced closes, because the
1004 *reader process may not have drained the data yet!
1006 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1007 data_was_unread += skb->len;
1011 /* If socket has been already reset kill it. */
1012 if (sk->sk_state == DCCP_CLOSED)
1013 goto adjudge_to_death;
1015 if (data_was_unread) {
1016 /* Unread data was tossed, send an appropriate Reset Code */
1017 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread);
1018 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
1019 dccp_set_state(sk, DCCP_CLOSED);
1020 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1021 /* Check zero linger _after_ checking for unread data. */
1022 sk->sk_prot->disconnect(sk, 0);
1023 } else if (sk->sk_state != DCCP_CLOSED) {
1025 * Normal connection termination. May need to wait if there are
1026 * still packets in the TX queue that are delayed by the CCID.
1028 dccp_flush_write_queue(sk, &timeout);
1029 dccp_terminate_connection(sk);
1033 * Flush write queue. This may be necessary in several cases:
1034 * - we have been closed by the peer but still have application data;
1035 * - abortive termination (unread data or zero linger time),
1036 * - normal termination but queue could not be flushed within time limit
1038 __skb_queue_purge(&sk->sk_write_queue);
1040 sk_stream_wait_close(sk, timeout);
1043 state = sk->sk_state;
1048 * It is the last release_sock in its life. It will remove backlog.
1052 * Now socket is owned by kernel and we acquire BH lock
1053 * to finish close. No need to check for user refs.
1057 WARN_ON(sock_owned_by_user(sk));
1059 this_cpu_inc(dccp_orphan_count);
1061 /* Have we already been destroyed by a softirq or backlog? */
1062 if (state != DCCP_CLOSED && sk->sk_state == DCCP_CLOSED)
1065 if (sk->sk_state == DCCP_CLOSED)
1066 inet_csk_destroy_sock(sk);
1068 /* Otherwise, socket is reprieved until protocol close. */
1076 EXPORT_SYMBOL_GPL(dccp_close);
1078 void dccp_shutdown(struct sock *sk, int how)
1080 dccp_pr_debug("called shutdown(%x)\n", how);
1083 EXPORT_SYMBOL_GPL(dccp_shutdown);
1085 static inline int __init dccp_mib_init(void)
1087 dccp_statistics = alloc_percpu(struct dccp_mib);
1088 if (!dccp_statistics)
1093 static inline void dccp_mib_exit(void)
1095 free_percpu(dccp_statistics);
1098 static int thash_entries;
1099 module_param(thash_entries, int, 0444);
1100 MODULE_PARM_DESC(thash_entries, "Number of ehash buckets");
1102 #ifdef CONFIG_IP_DCCP_DEBUG
1104 module_param(dccp_debug, bool, 0644);
1105 MODULE_PARM_DESC(dccp_debug, "Enable debug messages");
1107 EXPORT_SYMBOL_GPL(dccp_debug);
1110 static int __init dccp_init(void)
1113 unsigned long nr_pages = totalram_pages();
1114 int ehash_order, bhash_order, i;
1117 BUILD_BUG_ON(sizeof(struct dccp_skb_cb) >
1118 sizeof_field(struct sk_buff, cb));
1119 rc = inet_hashinfo2_init_mod(&dccp_hashinfo);
1123 dccp_hashinfo.bind_bucket_cachep =
1124 kmem_cache_create("dccp_bind_bucket",
1125 sizeof(struct inet_bind_bucket), 0,
1126 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1127 if (!dccp_hashinfo.bind_bucket_cachep)
1128 goto out_free_hashinfo2;
1129 dccp_hashinfo.bind2_bucket_cachep =
1130 kmem_cache_create("dccp_bind2_bucket",
1131 sizeof(struct inet_bind2_bucket), 0,
1132 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
1133 if (!dccp_hashinfo.bind2_bucket_cachep)
1134 goto out_free_bind_bucket_cachep;
1137 * Size and allocate the main established and bind bucket
1140 * The methodology is similar to that of the buffer cache.
1142 if (nr_pages >= (128 * 1024))
1143 goal = nr_pages >> (21 - PAGE_SHIFT);
1145 goal = nr_pages >> (23 - PAGE_SHIFT);
1148 goal = (thash_entries *
1149 sizeof(struct inet_ehash_bucket)) >> PAGE_SHIFT;
1150 for (ehash_order = 0; (1UL << ehash_order) < goal; ehash_order++)
1153 unsigned long hash_size = (1UL << ehash_order) * PAGE_SIZE /
1154 sizeof(struct inet_ehash_bucket);
1156 while (hash_size & (hash_size - 1))
1158 dccp_hashinfo.ehash_mask = hash_size - 1;
1159 dccp_hashinfo.ehash = (struct inet_ehash_bucket *)
1160 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, ehash_order);
1161 } while (!dccp_hashinfo.ehash && --ehash_order > 0);
1163 if (!dccp_hashinfo.ehash) {
1164 DCCP_CRIT("Failed to allocate DCCP established hash table");
1165 goto out_free_bind2_bucket_cachep;
1168 for (i = 0; i <= dccp_hashinfo.ehash_mask; i++)
1169 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo.ehash[i].chain, i);
1171 if (inet_ehash_locks_alloc(&dccp_hashinfo))
1172 goto out_free_dccp_ehash;
1174 bhash_order = ehash_order;
1177 dccp_hashinfo.bhash_size = (1UL << bhash_order) * PAGE_SIZE /
1178 sizeof(struct inet_bind_hashbucket);
1179 if ((dccp_hashinfo.bhash_size > (64 * 1024)) &&
1182 dccp_hashinfo.bhash = (struct inet_bind_hashbucket *)
1183 __get_free_pages(GFP_ATOMIC|__GFP_NOWARN, bhash_order);
1184 } while (!dccp_hashinfo.bhash && --bhash_order >= 0);
1186 if (!dccp_hashinfo.bhash) {
1187 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1188 goto out_free_dccp_locks;
1191 dccp_hashinfo.bhash2 = (struct inet_bind_hashbucket *)
1192 __get_free_pages(GFP_ATOMIC | __GFP_NOWARN, bhash_order);
1194 if (!dccp_hashinfo.bhash2) {
1195 DCCP_CRIT("Failed to allocate DCCP bind2 hash table");
1196 goto out_free_dccp_bhash;
1199 for (i = 0; i < dccp_hashinfo.bhash_size; i++) {
1200 spin_lock_init(&dccp_hashinfo.bhash[i].lock);
1201 INIT_HLIST_HEAD(&dccp_hashinfo.bhash[i].chain);
1202 spin_lock_init(&dccp_hashinfo.bhash2[i].lock);
1203 INIT_HLIST_HEAD(&dccp_hashinfo.bhash2[i].chain);
1206 dccp_hashinfo.pernet = false;
1208 rc = dccp_mib_init();
1210 goto out_free_dccp_bhash2;
1212 rc = dccp_ackvec_init();
1214 goto out_free_dccp_mib;
1216 rc = dccp_sysctl_init();
1218 goto out_ackvec_exit;
1220 rc = ccid_initialize_builtins();
1222 goto out_sysctl_exit;
1224 dccp_timestamping_init();
1234 out_free_dccp_bhash2:
1235 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1236 out_free_dccp_bhash:
1237 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1238 out_free_dccp_locks:
1239 inet_ehash_locks_free(&dccp_hashinfo);
1240 out_free_dccp_ehash:
1241 free_pages((unsigned long)dccp_hashinfo.ehash, ehash_order);
1242 out_free_bind2_bucket_cachep:
1243 kmem_cache_destroy(dccp_hashinfo.bind2_bucket_cachep);
1244 out_free_bind_bucket_cachep:
1245 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1247 inet_hashinfo2_free_mod(&dccp_hashinfo);
1249 dccp_hashinfo.bhash = NULL;
1250 dccp_hashinfo.bhash2 = NULL;
1251 dccp_hashinfo.ehash = NULL;
1252 dccp_hashinfo.bind_bucket_cachep = NULL;
1253 dccp_hashinfo.bind2_bucket_cachep = NULL;
1257 static void __exit dccp_fini(void)
1259 int bhash_order = get_order(dccp_hashinfo.bhash_size *
1260 sizeof(struct inet_bind_hashbucket));
1262 ccid_cleanup_builtins();
1264 free_pages((unsigned long)dccp_hashinfo.bhash, bhash_order);
1265 free_pages((unsigned long)dccp_hashinfo.bhash2, bhash_order);
1266 free_pages((unsigned long)dccp_hashinfo.ehash,
1267 get_order((dccp_hashinfo.ehash_mask + 1) *
1268 sizeof(struct inet_ehash_bucket)));
1269 inet_ehash_locks_free(&dccp_hashinfo);
1270 kmem_cache_destroy(dccp_hashinfo.bind_bucket_cachep);
1273 inet_hashinfo2_free_mod(&dccp_hashinfo);
1276 module_init(dccp_init);
1277 module_exit(dccp_fini);
1279 MODULE_LICENSE("GPL");
1280 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1281 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");