4 * An implementation of the DCCP protocol
5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/dccp.h>
13 #include <linux/module.h>
14 #include <linux/types.h>
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/skbuff.h>
18 #include <linux/netdevice.h>
20 #include <linux/if_arp.h>
21 #include <linux/init.h>
22 #include <linux/random.h>
23 #include <linux/slab.h>
24 #include <net/checksum.h>
26 #include <net/inet_sock.h>
27 #include <net/inet_common.h>
31 #include <asm/ioctls.h>
32 #include <linux/spinlock.h>
33 #include <linux/timer.h>
34 #include <linux/delay.h>
35 #include <linux/poll.h>
41 DEFINE_SNMP_STAT(struct dccp_mib
, dccp_statistics
) __read_mostly
;
43 EXPORT_SYMBOL_GPL(dccp_statistics
);
45 struct percpu_counter dccp_orphan_count
;
46 EXPORT_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 sk
->sk_state
= 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_listen_start(struct sock
*sk
, int backlog
)
243 struct dccp_sock
*dp
= dccp_sk(sk
);
245 dp
->dccps_role
= DCCP_ROLE_LISTEN
;
246 /* do not start to listen if feature negotiation setup fails */
247 if (dccp_feat_finalise_settings(dp
))
249 return inet_csk_listen_start(sk
, backlog
);
252 static inline int dccp_need_reset(int state
)
254 return state
!= DCCP_CLOSED
&& state
!= DCCP_LISTEN
&&
255 state
!= DCCP_REQUESTING
;
258 int dccp_disconnect(struct sock
*sk
, int flags
)
260 struct inet_connection_sock
*icsk
= inet_csk(sk
);
261 struct inet_sock
*inet
= inet_sk(sk
);
262 struct dccp_sock
*dp
= dccp_sk(sk
);
264 const int old_state
= sk
->sk_state
;
266 if (old_state
!= DCCP_CLOSED
)
267 dccp_set_state(sk
, DCCP_CLOSED
);
270 * This corresponds to the ABORT function of RFC793, sec. 3.8
271 * TCP uses a RST segment, DCCP a Reset packet with Code 2, "Aborted".
273 if (old_state
== DCCP_LISTEN
) {
274 inet_csk_listen_stop(sk
);
275 } else if (dccp_need_reset(old_state
)) {
276 dccp_send_reset(sk
, DCCP_RESET_CODE_ABORTED
);
277 sk
->sk_err
= ECONNRESET
;
278 } else if (old_state
== DCCP_REQUESTING
)
279 sk
->sk_err
= ECONNRESET
;
281 dccp_clear_xmit_timers(sk
);
282 ccid_hc_rx_delete(dp
->dccps_hc_rx_ccid
, sk
);
283 ccid_hc_tx_delete(dp
->dccps_hc_tx_ccid
, sk
);
284 dp
->dccps_hc_rx_ccid
= NULL
;
285 dp
->dccps_hc_tx_ccid
= NULL
;
287 __skb_queue_purge(&sk
->sk_receive_queue
);
288 __skb_queue_purge(&sk
->sk_write_queue
);
289 if (sk
->sk_send_head
!= NULL
) {
290 __kfree_skb(sk
->sk_send_head
);
291 sk
->sk_send_head
= NULL
;
294 inet
->inet_dport
= 0;
296 if (!(sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
))
297 inet_reset_saddr(sk
);
300 sock_reset_flag(sk
, SOCK_DONE
);
302 icsk
->icsk_backoff
= 0;
303 inet_csk_delack_init(sk
);
306 WARN_ON(inet
->inet_num
&& !icsk
->icsk_bind_hash
);
308 sk
->sk_error_report(sk
);
312 EXPORT_SYMBOL_GPL(dccp_disconnect
);
315 * Wait for a DCCP event.
317 * Note that we don't need to lock the socket, as the upper poll layers
318 * take care of normal races (between the test and the event) and we don't
319 * go look at any of the socket buffers directly.
321 __poll_t
dccp_poll(struct file
*file
, struct socket
*sock
,
325 struct sock
*sk
= sock
->sk
;
327 sock_poll_wait(file
, sk_sleep(sk
), wait
);
328 if (sk
->sk_state
== DCCP_LISTEN
)
329 return inet_csk_listen_poll(sk
);
331 /* Socket is not locked. We are protected from async events
332 by poll logic and correct handling of state changes
333 made by another threads is impossible in any case.
340 if (sk
->sk_shutdown
== SHUTDOWN_MASK
|| sk
->sk_state
== DCCP_CLOSED
)
342 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
343 mask
|= POLLIN
| POLLRDNORM
| POLLRDHUP
;
346 if ((1 << sk
->sk_state
) & ~(DCCPF_REQUESTING
| DCCPF_RESPOND
)) {
347 if (atomic_read(&sk
->sk_rmem_alloc
) > 0)
348 mask
|= POLLIN
| POLLRDNORM
;
350 if (!(sk
->sk_shutdown
& SEND_SHUTDOWN
)) {
351 if (sk_stream_is_writeable(sk
)) {
352 mask
|= POLLOUT
| POLLWRNORM
;
353 } else { /* send SIGIO later */
354 sk_set_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
355 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
357 /* Race breaker. If space is freed after
358 * wspace test but before the flags are set,
359 * IO signal will be lost.
361 if (sk_stream_is_writeable(sk
))
362 mask
|= POLLOUT
| POLLWRNORM
;
369 EXPORT_SYMBOL_GPL(dccp_poll
);
371 int dccp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
)
377 if (sk
->sk_state
== DCCP_LISTEN
)
383 unsigned long amount
= 0;
385 skb
= skb_peek(&sk
->sk_receive_queue
);
388 * We will only return the amount of this packet since
389 * that is all that will be read.
393 rc
= put_user(amount
, (int __user
*)arg
);
405 EXPORT_SYMBOL_GPL(dccp_ioctl
);
407 static int dccp_setsockopt_service(struct sock
*sk
, const __be32 service
,
408 char __user
*optval
, unsigned int optlen
)
410 struct dccp_sock
*dp
= dccp_sk(sk
);
411 struct dccp_service_list
*sl
= NULL
;
413 if (service
== DCCP_SERVICE_INVALID_VALUE
||
414 optlen
> DCCP_SERVICE_LIST_MAX_LEN
* sizeof(u32
))
417 if (optlen
> sizeof(service
)) {
418 sl
= kmalloc(optlen
, GFP_KERNEL
);
422 sl
->dccpsl_nr
= optlen
/ sizeof(u32
) - 1;
423 if (copy_from_user(sl
->dccpsl_list
,
424 optval
+ sizeof(service
),
425 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 char __user
*optval
, unsigned int optlen
)
484 if (optlen
< 1 || optlen
> DCCP_FEAT_MAX_SP_VALS
)
487 val
= memdup_user(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 char __user
*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 (get_user(val
, (int __user
*)optval
))
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
,
570 char __user
*optval
, unsigned int optlen
)
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
);
582 int compat_dccp_setsockopt(struct sock
*sk
, int level
, int optname
,
583 char __user
*optval
, unsigned int optlen
)
585 if (level
!= SOL_DCCP
)
586 return inet_csk_compat_setsockopt(sk
, level
, optname
,
588 return do_dccp_setsockopt(sk
, level
, optname
, optval
, optlen
);
591 EXPORT_SYMBOL_GPL(compat_dccp_setsockopt
);
594 static int dccp_getsockopt_service(struct sock
*sk
, int len
,
595 __be32 __user
*optval
,
598 const struct dccp_sock
*dp
= dccp_sk(sk
);
599 const struct dccp_service_list
*sl
;
600 int err
= -ENOENT
, slen
= 0, total_len
= sizeof(u32
);
603 if ((sl
= dp
->dccps_service_list
) != NULL
) {
604 slen
= sl
->dccpsl_nr
* sizeof(u32
);
613 if (put_user(total_len
, optlen
) ||
614 put_user(dp
->dccps_service
, optval
) ||
615 (sl
!= NULL
&& copy_to_user(optval
+ 1, sl
->dccpsl_list
, slen
)))
622 static int do_dccp_getsockopt(struct sock
*sk
, int level
, int optname
,
623 char __user
*optval
, int __user
*optlen
)
625 struct dccp_sock
*dp
;
628 if (get_user(len
, optlen
))
631 if (len
< (int)sizeof(int))
637 case DCCP_SOCKOPT_PACKET_SIZE
:
638 DCCP_WARN("sockopt(PACKET_SIZE) is deprecated: fix your app\n");
640 case DCCP_SOCKOPT_SERVICE
:
641 return dccp_getsockopt_service(sk
, len
,
642 (__be32 __user
*)optval
, optlen
);
643 case DCCP_SOCKOPT_GET_CUR_MPS
:
644 val
= dp
->dccps_mss_cache
;
646 case DCCP_SOCKOPT_AVAILABLE_CCIDS
:
647 return ccid_getsockopt_builtin_ccids(sk
, len
, optval
, optlen
);
648 case DCCP_SOCKOPT_TX_CCID
:
649 val
= ccid_get_current_tx_ccid(dp
);
653 case DCCP_SOCKOPT_RX_CCID
:
654 val
= ccid_get_current_rx_ccid(dp
);
658 case DCCP_SOCKOPT_SERVER_TIMEWAIT
:
659 val
= dp
->dccps_server_timewait
;
661 case DCCP_SOCKOPT_SEND_CSCOV
:
662 val
= dp
->dccps_pcslen
;
664 case DCCP_SOCKOPT_RECV_CSCOV
:
665 val
= dp
->dccps_pcrlen
;
667 case DCCP_SOCKOPT_QPOLICY_ID
:
668 val
= dp
->dccps_qpolicy
;
670 case DCCP_SOCKOPT_QPOLICY_TXQLEN
:
671 val
= dp
->dccps_tx_qlen
;
674 return ccid_hc_rx_getsockopt(dp
->dccps_hc_rx_ccid
, sk
, optname
,
675 len
, (u32 __user
*)optval
, optlen
);
677 return ccid_hc_tx_getsockopt(dp
->dccps_hc_tx_ccid
, sk
, optname
,
678 len
, (u32 __user
*)optval
, optlen
);
684 if (put_user(len
, optlen
) || copy_to_user(optval
, &val
, len
))
690 int dccp_getsockopt(struct sock
*sk
, int level
, int optname
,
691 char __user
*optval
, int __user
*optlen
)
693 if (level
!= SOL_DCCP
)
694 return inet_csk(sk
)->icsk_af_ops
->getsockopt(sk
, level
,
697 return do_dccp_getsockopt(sk
, level
, optname
, optval
, optlen
);
700 EXPORT_SYMBOL_GPL(dccp_getsockopt
);
703 int compat_dccp_getsockopt(struct sock
*sk
, int level
, int optname
,
704 char __user
*optval
, int __user
*optlen
)
706 if (level
!= SOL_DCCP
)
707 return inet_csk_compat_getsockopt(sk
, level
, optname
,
709 return do_dccp_getsockopt(sk
, level
, optname
, optval
, optlen
);
712 EXPORT_SYMBOL_GPL(compat_dccp_getsockopt
);
715 static int dccp_msghdr_parse(struct msghdr
*msg
, struct sk_buff
*skb
)
717 struct cmsghdr
*cmsg
;
720 * Assign an (opaque) qpolicy priority value to skb->priority.
722 * We are overloading this skb field for use with the qpolicy subystem.
723 * The skb->priority is normally used for the SO_PRIORITY option, which
724 * is initialised from sk_priority. Since the assignment of sk_priority
725 * to skb->priority happens later (on layer 3), we overload this field
726 * for use with queueing priorities as long as the skb is on layer 4.
727 * The default priority value (if nothing is set) is 0.
731 for_each_cmsghdr(cmsg
, msg
) {
732 if (!CMSG_OK(msg
, cmsg
))
735 if (cmsg
->cmsg_level
!= SOL_DCCP
)
738 if (cmsg
->cmsg_type
<= DCCP_SCM_QPOLICY_MAX
&&
739 !dccp_qpolicy_param_ok(skb
->sk
, cmsg
->cmsg_type
))
742 switch (cmsg
->cmsg_type
) {
743 case DCCP_SCM_PRIORITY
:
744 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(__u32
)))
746 skb
->priority
= *(__u32
*)CMSG_DATA(cmsg
);
755 int dccp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
)
757 const struct dccp_sock
*dp
= dccp_sk(sk
);
758 const int flags
= msg
->msg_flags
;
759 const int noblock
= flags
& MSG_DONTWAIT
;
764 if (len
> dp
->dccps_mss_cache
)
769 if (dccp_qpolicy_full(sk
)) {
774 timeo
= sock_sndtimeo(sk
, noblock
);
777 * We have to use sk_stream_wait_connect here to set sk_write_pending,
778 * so that the trick in dccp_rcv_request_sent_state_process.
780 /* Wait for a connection to finish. */
781 if ((1 << sk
->sk_state
) & ~(DCCPF_OPEN
| DCCPF_PARTOPEN
))
782 if ((rc
= sk_stream_wait_connect(sk
, &timeo
)) != 0)
785 size
= sk
->sk_prot
->max_header
+ len
;
787 skb
= sock_alloc_send_skb(sk
, size
, noblock
, &rc
);
792 skb_reserve(skb
, sk
->sk_prot
->max_header
);
793 rc
= memcpy_from_msg(skb_put(skb
, len
), msg
, len
);
797 rc
= dccp_msghdr_parse(msg
, skb
);
801 dccp_qpolicy_push(sk
, skb
);
803 * The xmit_timer is set if the TX CCID is rate-based and will expire
804 * when congestion control permits to release further packets into the
805 * network. Window-based CCIDs do not use this timer.
807 if (!timer_pending(&dp
->dccps_xmit_timer
))
817 EXPORT_SYMBOL_GPL(dccp_sendmsg
);
819 int dccp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
, int nonblock
,
820 int flags
, int *addr_len
)
822 const struct dccp_hdr
*dh
;
827 if (sk
->sk_state
== DCCP_LISTEN
) {
832 timeo
= sock_rcvtimeo(sk
, nonblock
);
835 struct sk_buff
*skb
= skb_peek(&sk
->sk_receive_queue
);
838 goto verify_sock_status
;
842 switch (dh
->dccph_type
) {
844 case DCCP_PKT_DATAACK
:
848 case DCCP_PKT_CLOSEREQ
:
849 if (!(flags
& MSG_PEEK
))
850 dccp_finish_passive_close(sk
);
853 dccp_pr_debug("found fin (%s) ok!\n",
854 dccp_packet_name(dh
->dccph_type
));
858 dccp_pr_debug("packet_type=%s\n",
859 dccp_packet_name(dh
->dccph_type
));
863 if (sock_flag(sk
, SOCK_DONE
)) {
869 len
= sock_error(sk
);
873 if (sk
->sk_shutdown
& RCV_SHUTDOWN
) {
878 if (sk
->sk_state
== DCCP_CLOSED
) {
879 if (!sock_flag(sk
, SOCK_DONE
)) {
880 /* This occurs when user tries to read
881 * from never connected socket.
895 if (signal_pending(current
)) {
896 len
= sock_intr_errno(timeo
);
900 sk_wait_data(sk
, &timeo
, NULL
);
905 else if (len
< skb
->len
)
906 msg
->msg_flags
|= MSG_TRUNC
;
908 if (skb_copy_datagram_msg(skb
, 0, msg
, len
)) {
909 /* Exception. Bailout! */
913 if (flags
& MSG_TRUNC
)
916 if (!(flags
& MSG_PEEK
))
925 EXPORT_SYMBOL_GPL(dccp_recvmsg
);
927 int inet_dccp_listen(struct socket
*sock
, int backlog
)
929 struct sock
*sk
= sock
->sk
;
930 unsigned char old_state
;
936 if (sock
->state
!= SS_UNCONNECTED
|| sock
->type
!= SOCK_DCCP
)
939 old_state
= sk
->sk_state
;
940 if (!((1 << old_state
) & (DCCPF_CLOSED
| DCCPF_LISTEN
)))
943 /* Really, if the socket is already in listen state
944 * we can only allow the backlog to be adjusted.
946 if (old_state
!= DCCP_LISTEN
) {
948 * FIXME: here it probably should be sk->sk_prot->listen_start
949 * see tcp_listen_start
951 err
= dccp_listen_start(sk
, backlog
);
955 sk
->sk_max_ack_backlog
= backlog
;
963 EXPORT_SYMBOL_GPL(inet_dccp_listen
);
965 static void dccp_terminate_connection(struct sock
*sk
)
967 u8 next_state
= DCCP_CLOSED
;
969 switch (sk
->sk_state
) {
970 case DCCP_PASSIVE_CLOSE
:
971 case DCCP_PASSIVE_CLOSEREQ
:
972 dccp_finish_passive_close(sk
);
975 dccp_pr_debug("Stop PARTOPEN timer (%p)\n", sk
);
976 inet_csk_clear_xmit_timer(sk
, ICSK_TIME_DACK
);
979 dccp_send_close(sk
, 1);
981 if (dccp_sk(sk
)->dccps_role
== DCCP_ROLE_SERVER
&&
982 !dccp_sk(sk
)->dccps_server_timewait
)
983 next_state
= DCCP_ACTIVE_CLOSEREQ
;
985 next_state
= DCCP_CLOSING
;
988 dccp_set_state(sk
, next_state
);
992 void dccp_close(struct sock
*sk
, long timeout
)
994 struct dccp_sock
*dp
= dccp_sk(sk
);
996 u32 data_was_unread
= 0;
1001 sk
->sk_shutdown
= SHUTDOWN_MASK
;
1003 if (sk
->sk_state
== DCCP_LISTEN
) {
1004 dccp_set_state(sk
, DCCP_CLOSED
);
1007 inet_csk_listen_stop(sk
);
1009 goto adjudge_to_death
;
1012 sk_stop_timer(sk
, &dp
->dccps_xmit_timer
);
1015 * We need to flush the recv. buffs. We do this only on the
1016 * descriptor close, not protocol-sourced closes, because the
1017 *reader process may not have drained the data yet!
1019 while ((skb
= __skb_dequeue(&sk
->sk_receive_queue
)) != NULL
) {
1020 data_was_unread
+= skb
->len
;
1024 /* If socket has been already reset kill it. */
1025 if (sk
->sk_state
== DCCP_CLOSED
)
1026 goto adjudge_to_death
;
1028 if (data_was_unread
) {
1029 /* Unread data was tossed, send an appropriate Reset Code */
1030 DCCP_WARN("ABORT with %u bytes unread\n", data_was_unread
);
1031 dccp_send_reset(sk
, DCCP_RESET_CODE_ABORTED
);
1032 dccp_set_state(sk
, DCCP_CLOSED
);
1033 } else if (sock_flag(sk
, SOCK_LINGER
) && !sk
->sk_lingertime
) {
1034 /* Check zero linger _after_ checking for unread data. */
1035 sk
->sk_prot
->disconnect(sk
, 0);
1036 } else if (sk
->sk_state
!= DCCP_CLOSED
) {
1038 * Normal connection termination. May need to wait if there are
1039 * still packets in the TX queue that are delayed by the CCID.
1041 dccp_flush_write_queue(sk
, &timeout
);
1042 dccp_terminate_connection(sk
);
1046 * Flush write queue. This may be necessary in several cases:
1047 * - we have been closed by the peer but still have application data;
1048 * - abortive termination (unread data or zero linger time),
1049 * - normal termination but queue could not be flushed within time limit
1051 __skb_queue_purge(&sk
->sk_write_queue
);
1053 sk_stream_wait_close(sk
, timeout
);
1056 state
= sk
->sk_state
;
1061 * It is the last release_sock in its life. It will remove backlog.
1065 * Now socket is owned by kernel and we acquire BH lock
1066 * to finish close. No need to check for user refs.
1070 WARN_ON(sock_owned_by_user(sk
));
1072 percpu_counter_inc(sk
->sk_prot
->orphan_count
);
1074 /* Have we already been destroyed by a softirq or backlog? */
1075 if (state
!= DCCP_CLOSED
&& sk
->sk_state
== DCCP_CLOSED
)
1078 if (sk
->sk_state
== DCCP_CLOSED
)
1079 inet_csk_destroy_sock(sk
);
1081 /* Otherwise, socket is reprieved until protocol close. */
1089 EXPORT_SYMBOL_GPL(dccp_close
);
1091 void dccp_shutdown(struct sock
*sk
, int how
)
1093 dccp_pr_debug("called shutdown(%x)\n", how
);
1096 EXPORT_SYMBOL_GPL(dccp_shutdown
);
1098 static inline int __init
dccp_mib_init(void)
1100 dccp_statistics
= alloc_percpu(struct dccp_mib
);
1101 if (!dccp_statistics
)
1106 static inline void dccp_mib_exit(void)
1108 free_percpu(dccp_statistics
);
1111 static int thash_entries
;
1112 module_param(thash_entries
, int, 0444);
1113 MODULE_PARM_DESC(thash_entries
, "Number of ehash buckets");
1115 #ifdef CONFIG_IP_DCCP_DEBUG
1117 module_param(dccp_debug
, bool, 0644);
1118 MODULE_PARM_DESC(dccp_debug
, "Enable debug messages");
1120 EXPORT_SYMBOL_GPL(dccp_debug
);
1123 static int __init
dccp_init(void)
1126 int ehash_order
, bhash_order
, i
;
1129 BUILD_BUG_ON(sizeof(struct dccp_skb_cb
) >
1130 FIELD_SIZEOF(struct sk_buff
, cb
));
1131 rc
= percpu_counter_init(&dccp_orphan_count
, 0, GFP_KERNEL
);
1135 inet_hashinfo_init(&dccp_hashinfo
);
1136 dccp_hashinfo
.bind_bucket_cachep
=
1137 kmem_cache_create("dccp_bind_bucket",
1138 sizeof(struct inet_bind_bucket
), 0,
1139 SLAB_HWCACHE_ALIGN
, NULL
);
1140 if (!dccp_hashinfo
.bind_bucket_cachep
)
1141 goto out_free_percpu
;
1144 * Size and allocate the main established and bind bucket
1147 * The methodology is similar to that of the buffer cache.
1149 if (totalram_pages
>= (128 * 1024))
1150 goal
= totalram_pages
>> (21 - PAGE_SHIFT
);
1152 goal
= totalram_pages
>> (23 - PAGE_SHIFT
);
1155 goal
= (thash_entries
*
1156 sizeof(struct inet_ehash_bucket
)) >> PAGE_SHIFT
;
1157 for (ehash_order
= 0; (1UL << ehash_order
) < goal
; ehash_order
++)
1160 unsigned long hash_size
= (1UL << ehash_order
) * PAGE_SIZE
/
1161 sizeof(struct inet_ehash_bucket
);
1163 while (hash_size
& (hash_size
- 1))
1165 dccp_hashinfo
.ehash_mask
= hash_size
- 1;
1166 dccp_hashinfo
.ehash
= (struct inet_ehash_bucket
*)
1167 __get_free_pages(GFP_ATOMIC
|__GFP_NOWARN
, ehash_order
);
1168 } while (!dccp_hashinfo
.ehash
&& --ehash_order
> 0);
1170 if (!dccp_hashinfo
.ehash
) {
1171 DCCP_CRIT("Failed to allocate DCCP established hash table");
1172 goto out_free_bind_bucket_cachep
;
1175 for (i
= 0; i
<= dccp_hashinfo
.ehash_mask
; i
++)
1176 INIT_HLIST_NULLS_HEAD(&dccp_hashinfo
.ehash
[i
].chain
, i
);
1178 if (inet_ehash_locks_alloc(&dccp_hashinfo
))
1179 goto out_free_dccp_ehash
;
1181 bhash_order
= ehash_order
;
1184 dccp_hashinfo
.bhash_size
= (1UL << bhash_order
) * PAGE_SIZE
/
1185 sizeof(struct inet_bind_hashbucket
);
1186 if ((dccp_hashinfo
.bhash_size
> (64 * 1024)) &&
1189 dccp_hashinfo
.bhash
= (struct inet_bind_hashbucket
*)
1190 __get_free_pages(GFP_ATOMIC
|__GFP_NOWARN
, bhash_order
);
1191 } while (!dccp_hashinfo
.bhash
&& --bhash_order
>= 0);
1193 if (!dccp_hashinfo
.bhash
) {
1194 DCCP_CRIT("Failed to allocate DCCP bind hash table");
1195 goto out_free_dccp_locks
;
1198 for (i
= 0; i
< dccp_hashinfo
.bhash_size
; i
++) {
1199 spin_lock_init(&dccp_hashinfo
.bhash
[i
].lock
);
1200 INIT_HLIST_HEAD(&dccp_hashinfo
.bhash
[i
].chain
);
1203 rc
= dccp_mib_init();
1205 goto out_free_dccp_bhash
;
1207 rc
= dccp_ackvec_init();
1209 goto out_free_dccp_mib
;
1211 rc
= dccp_sysctl_init();
1213 goto out_ackvec_exit
;
1215 rc
= ccid_initialize_builtins();
1217 goto out_sysctl_exit
;
1219 dccp_timestamping_init();
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_bind_bucket_cachep
:
1236 kmem_cache_destroy(dccp_hashinfo
.bind_bucket_cachep
);
1238 percpu_counter_destroy(&dccp_orphan_count
);
1240 dccp_hashinfo
.bhash
= NULL
;
1241 dccp_hashinfo
.ehash
= NULL
;
1242 dccp_hashinfo
.bind_bucket_cachep
= NULL
;
1246 static void __exit
dccp_fini(void)
1248 ccid_cleanup_builtins();
1250 free_pages((unsigned long)dccp_hashinfo
.bhash
,
1251 get_order(dccp_hashinfo
.bhash_size
*
1252 sizeof(struct inet_bind_hashbucket
)));
1253 free_pages((unsigned long)dccp_hashinfo
.ehash
,
1254 get_order((dccp_hashinfo
.ehash_mask
+ 1) *
1255 sizeof(struct inet_ehash_bucket
)));
1256 inet_ehash_locks_free(&dccp_hashinfo
);
1257 kmem_cache_destroy(dccp_hashinfo
.bind_bucket_cachep
);
1260 percpu_counter_destroy(&dccp_orphan_count
);
1263 module_init(dccp_init
);
1264 module_exit(dccp_fini
);
1266 MODULE_LICENSE("GPL");
1267 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@conectiva.com.br>");
1268 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");