Linux Audio

Check our new training course

Loading...
v3.1
  1/*
  2 *  net/dccp/output.c
  3 *
  4 *  An implementation of the DCCP protocol
  5 *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  6 *
  7 *	This program is free software; you can redistribute it and/or
  8 *	modify it under the terms of the GNU General Public License
  9 *	as published by the Free Software Foundation; either version
 10 *	2 of the License, or (at your option) any later version.
 11 */
 12
 13#include <linux/dccp.h>
 14#include <linux/kernel.h>
 15#include <linux/skbuff.h>
 16#include <linux/slab.h>
 
 17
 18#include <net/inet_sock.h>
 19#include <net/sock.h>
 20
 21#include "ackvec.h"
 22#include "ccid.h"
 23#include "dccp.h"
 24
 25static inline void dccp_event_ack_sent(struct sock *sk)
 26{
 27	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
 28}
 29
 30/* enqueue @skb on sk_send_head for retransmission, return clone to send now */
 31static struct sk_buff *dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
 32{
 33	skb_set_owner_w(skb, sk);
 34	WARN_ON(sk->sk_send_head);
 35	sk->sk_send_head = skb;
 36	return skb_clone(sk->sk_send_head, gfp_any());
 37}
 38
 39/*
 40 * All SKB's seen here are completely headerless. It is our
 41 * job to build the DCCP header, and pass the packet down to
 42 * IP so it can do the same plus pass the packet off to the
 43 * device.
 44 */
 45static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
 46{
 47	if (likely(skb != NULL)) {
 48		struct inet_sock *inet = inet_sk(sk);
 49		const struct inet_connection_sock *icsk = inet_csk(sk);
 50		struct dccp_sock *dp = dccp_sk(sk);
 51		struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 52		struct dccp_hdr *dh;
 53		/* XXX For now we're using only 48 bits sequence numbers */
 54		const u32 dccp_header_size = sizeof(*dh) +
 55					     sizeof(struct dccp_hdr_ext) +
 56					  dccp_packet_hdr_len(dcb->dccpd_type);
 57		int err, set_ack = 1;
 58		u64 ackno = dp->dccps_gsr;
 59		/*
 60		 * Increment GSS here already in case the option code needs it.
 61		 * Update GSS for real only if option processing below succeeds.
 62		 */
 63		dcb->dccpd_seq = ADD48(dp->dccps_gss, 1);
 64
 65		switch (dcb->dccpd_type) {
 66		case DCCP_PKT_DATA:
 67			set_ack = 0;
 68			/* fall through */
 69		case DCCP_PKT_DATAACK:
 70		case DCCP_PKT_RESET:
 71			break;
 72
 73		case DCCP_PKT_REQUEST:
 74			set_ack = 0;
 75			/* Use ISS on the first (non-retransmitted) Request. */
 76			if (icsk->icsk_retransmits == 0)
 77				dcb->dccpd_seq = dp->dccps_iss;
 78			/* fall through */
 79
 80		case DCCP_PKT_SYNC:
 81		case DCCP_PKT_SYNCACK:
 82			ackno = dcb->dccpd_ack_seq;
 83			/* fall through */
 84		default:
 85			/*
 86			 * Set owner/destructor: some skbs are allocated via
 87			 * alloc_skb (e.g. when retransmission may happen).
 88			 * Only Data, DataAck, and Reset packets should come
 89			 * through here with skb->sk set.
 90			 */
 91			WARN_ON(skb->sk);
 92			skb_set_owner_w(skb, sk);
 93			break;
 94		}
 95
 96		if (dccp_insert_options(sk, skb)) {
 97			kfree_skb(skb);
 98			return -EPROTO;
 99		}
100
101
102		/* Build DCCP header and checksum it. */
103		dh = dccp_zeroed_hdr(skb, dccp_header_size);
104		dh->dccph_type	= dcb->dccpd_type;
105		dh->dccph_sport	= inet->inet_sport;
106		dh->dccph_dport	= inet->inet_dport;
107		dh->dccph_doff	= (dccp_header_size + dcb->dccpd_opt_len) / 4;
108		dh->dccph_ccval	= dcb->dccpd_ccval;
109		dh->dccph_cscov = dp->dccps_pcslen;
110		/* XXX For now we're using only 48 bits sequence numbers */
111		dh->dccph_x	= 1;
112
113		dccp_update_gss(sk, dcb->dccpd_seq);
114		dccp_hdr_set_seq(dh, dp->dccps_gss);
115		if (set_ack)
116			dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
117
118		switch (dcb->dccpd_type) {
119		case DCCP_PKT_REQUEST:
120			dccp_hdr_request(skb)->dccph_req_service =
121							dp->dccps_service;
122			/*
123			 * Limit Ack window to ISS <= P.ackno <= GSS, so that
124			 * only Responses to Requests we sent are considered.
125			 */
126			dp->dccps_awl = dp->dccps_iss;
127			break;
128		case DCCP_PKT_RESET:
129			dccp_hdr_reset(skb)->dccph_reset_code =
130							dcb->dccpd_reset_code;
131			break;
132		}
133
134		icsk->icsk_af_ops->send_check(sk, skb);
135
136		if (set_ack)
137			dccp_event_ack_sent(sk);
138
139		DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
140
141		err = icsk->icsk_af_ops->queue_xmit(skb, &inet->cork.fl);
142		return net_xmit_eval(err);
143	}
144	return -ENOBUFS;
145}
146
147/**
148 * dccp_determine_ccmps  -  Find out about CCID-specific packet-size limits
149 * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
150 * since the RX CCID is restricted to feedback packets (Acks), which are small
151 * in comparison with the data traffic. A value of 0 means "no current CCMPS".
152 */
153static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
154{
155	const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
156
157	if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
158		return 0;
159	return tx_ccid->ccid_ops->ccid_ccmps;
160}
161
162unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
163{
164	struct inet_connection_sock *icsk = inet_csk(sk);
165	struct dccp_sock *dp = dccp_sk(sk);
166	u32 ccmps = dccp_determine_ccmps(dp);
167	u32 cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
168
169	/* Account for header lengths and IPv4/v6 option overhead */
170	cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
171		    sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
172
173	/*
174	 * Leave enough headroom for common DCCP header options.
175	 * This only considers options which may appear on DCCP-Data packets, as
176	 * per table 3 in RFC 4340, 5.8. When running out of space for other
177	 * options (eg. Ack Vector which can take up to 255 bytes), it is better
178	 * to schedule a separate Ack. Thus we leave headroom for the following:
179	 *  - 1 byte for Slow Receiver (11.6)
180	 *  - 6 bytes for Timestamp (13.1)
181	 *  - 10 bytes for Timestamp Echo (13.3)
182	 *  - 8 bytes for NDP count (7.7, when activated)
183	 *  - 6 bytes for Data Checksum (9.3)
184	 *  - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
185	 */
186	cur_mps -= roundup(1 + 6 + 10 + dp->dccps_send_ndp_count * 8 + 6 +
187			   (dp->dccps_hc_rx_ackvec ? DCCPAV_MIN_OPTLEN : 0), 4);
188
189	/* And store cached results */
190	icsk->icsk_pmtu_cookie = pmtu;
191	dp->dccps_mss_cache = cur_mps;
192
193	return cur_mps;
194}
195
196EXPORT_SYMBOL_GPL(dccp_sync_mss);
197
198void dccp_write_space(struct sock *sk)
199{
200	struct socket_wq *wq;
201
202	rcu_read_lock();
203	wq = rcu_dereference(sk->sk_wq);
204	if (wq_has_sleeper(wq))
205		wake_up_interruptible(&wq->wait);
206	/* Should agree with poll, otherwise some programs break */
207	if (sock_writeable(sk))
208		sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
209
210	rcu_read_unlock();
211}
212
213/**
214 * dccp_wait_for_ccid  -  Await CCID send permission
215 * @sk:    socket to wait for
216 * @delay: timeout in jiffies
 
217 * This is used by CCIDs which need to delay the send time in process context.
218 */
219static int dccp_wait_for_ccid(struct sock *sk, unsigned long delay)
220{
221	DEFINE_WAIT(wait);
222	long remaining;
223
224	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
225	sk->sk_write_pending++;
226	release_sock(sk);
227
228	remaining = schedule_timeout(delay);
229
230	lock_sock(sk);
231	sk->sk_write_pending--;
232	finish_wait(sk_sleep(sk), &wait);
233
234	if (signal_pending(current) || sk->sk_err)
235		return -1;
236	return remaining;
237}
238
239/**
240 * dccp_xmit_packet  -  Send data packet under control of CCID
241 * Transmits next-queued payload and informs CCID to account for the packet.
242 */
243static void dccp_xmit_packet(struct sock *sk)
244{
245	int err, len;
246	struct dccp_sock *dp = dccp_sk(sk);
247	struct sk_buff *skb = dccp_qpolicy_pop(sk);
248
249	if (unlikely(skb == NULL))
250		return;
251	len = skb->len;
252
253	if (sk->sk_state == DCCP_PARTOPEN) {
254		const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
255		/*
256		 * See 8.1.5 - Handshake Completion.
257		 *
258		 * For robustness we resend Confirm options until the client has
259		 * entered OPEN. During the initial feature negotiation, the MPS
260		 * is smaller than usual, reduced by the Change/Confirm options.
261		 */
262		if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
263			DCCP_WARN("Payload too large (%d) for featneg.\n", len);
264			dccp_send_ack(sk);
265			dccp_feat_list_purge(&dp->dccps_featneg);
266		}
267
268		inet_csk_schedule_ack(sk);
269		inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
270					      inet_csk(sk)->icsk_rto,
271					      DCCP_RTO_MAX);
272		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
273	} else if (dccp_ack_pending(sk)) {
274		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
275	} else {
276		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATA;
277	}
278
279	err = dccp_transmit_skb(sk, skb);
280	if (err)
281		dccp_pr_debug("transmit_skb() returned err=%d\n", err);
282	/*
283	 * Register this one as sent even if an error occurred. To the remote
284	 * end a local packet drop is indistinguishable from network loss, i.e.
285	 * any local drop will eventually be reported via receiver feedback.
286	 */
287	ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, len);
288
289	/*
290	 * If the CCID needs to transfer additional header options out-of-band
291	 * (e.g. Ack Vectors or feature-negotiation options), it activates this
292	 * flag to schedule a Sync. The Sync will automatically incorporate all
293	 * currently pending header options, thus clearing the backlog.
294	 */
295	if (dp->dccps_sync_scheduled)
296		dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
297}
298
299/**
300 * dccp_flush_write_queue  -  Drain queue at end of connection
301 * Since dccp_sendmsg queues packets without waiting for them to be sent, it may
302 * happen that the TX queue is not empty at the end of a connection. We give the
303 * HC-sender CCID a grace period of up to @time_budget jiffies. If this function
304 * returns with a non-empty write queue, it will be purged later.
305 */
306void dccp_flush_write_queue(struct sock *sk, long *time_budget)
307{
308	struct dccp_sock *dp = dccp_sk(sk);
309	struct sk_buff *skb;
310	long delay, rc;
311
312	while (*time_budget > 0 && (skb = skb_peek(&sk->sk_write_queue))) {
313		rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
314
315		switch (ccid_packet_dequeue_eval(rc)) {
316		case CCID_PACKET_WILL_DEQUEUE_LATER:
317			/*
318			 * If the CCID determines when to send, the next sending
319			 * time is unknown or the CCID may not even send again
320			 * (e.g. remote host crashes or lost Ack packets).
321			 */
322			DCCP_WARN("CCID did not manage to send all packets\n");
323			return;
324		case CCID_PACKET_DELAY:
325			delay = msecs_to_jiffies(rc);
326			if (delay > *time_budget)
327				return;
328			rc = dccp_wait_for_ccid(sk, delay);
329			if (rc < 0)
330				return;
331			*time_budget -= (delay - rc);
332			/* check again if we can send now */
333			break;
334		case CCID_PACKET_SEND_AT_ONCE:
335			dccp_xmit_packet(sk);
336			break;
337		case CCID_PACKET_ERR:
338			skb_dequeue(&sk->sk_write_queue);
339			kfree_skb(skb);
340			dccp_pr_debug("packet discarded due to err=%ld\n", rc);
341		}
342	}
343}
344
345void dccp_write_xmit(struct sock *sk)
346{
347	struct dccp_sock *dp = dccp_sk(sk);
348	struct sk_buff *skb;
349
350	while ((skb = dccp_qpolicy_top(sk))) {
351		int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
352
353		switch (ccid_packet_dequeue_eval(rc)) {
354		case CCID_PACKET_WILL_DEQUEUE_LATER:
355			return;
356		case CCID_PACKET_DELAY:
357			sk_reset_timer(sk, &dp->dccps_xmit_timer,
358				       jiffies + msecs_to_jiffies(rc));
359			return;
360		case CCID_PACKET_SEND_AT_ONCE:
361			dccp_xmit_packet(sk);
362			break;
363		case CCID_PACKET_ERR:
364			dccp_qpolicy_drop(sk, skb);
365			dccp_pr_debug("packet discarded due to err=%d\n", rc);
366		}
367	}
368}
369
370/**
371 * dccp_retransmit_skb  -  Retransmit Request, Close, or CloseReq packets
372 * There are only four retransmittable packet types in DCCP:
373 * - Request  in client-REQUEST  state (sec. 8.1.1),
374 * - CloseReq in server-CLOSEREQ state (sec. 8.3),
375 * - Close    in   node-CLOSING  state (sec. 8.3),
376 * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
377 * This function expects sk->sk_send_head to contain the original skb.
378 */
379int dccp_retransmit_skb(struct sock *sk)
380{
381	WARN_ON(sk->sk_send_head == NULL);
382
383	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
384		return -EHOSTUNREACH; /* Routing failure or similar. */
385
386	/* this count is used to distinguish original and retransmitted skb */
387	inet_csk(sk)->icsk_retransmits++;
388
389	return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
390}
391
392struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
393				   struct request_sock *req)
394{
395	struct dccp_hdr *dh;
396	struct dccp_request_sock *dreq;
397	const u32 dccp_header_size = sizeof(struct dccp_hdr) +
398				     sizeof(struct dccp_hdr_ext) +
399				     sizeof(struct dccp_hdr_response);
400	struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
401					   GFP_ATOMIC);
402	if (skb == NULL)
 
 
 
 
 
 
403		return NULL;
404
405	/* Reserve space for headers. */
406	skb_reserve(skb, sk->sk_prot->max_header);
407
408	skb_dst_set(skb, dst_clone(dst));
409
410	dreq = dccp_rsk(req);
411	if (inet_rsk(req)->acked)	/* increase ISS upon retransmission */
412		dccp_inc_seqno(&dreq->dreq_iss);
413	DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
414	DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
415
416	/* Resolve feature dependencies resulting from choice of CCID */
417	if (dccp_feat_server_ccid_dependencies(dreq))
418		goto response_failed;
419
420	if (dccp_insert_options_rsk(dreq, skb))
421		goto response_failed;
422
423	/* Build and checksum header */
424	dh = dccp_zeroed_hdr(skb, dccp_header_size);
425
426	dh->dccph_sport	= inet_rsk(req)->loc_port;
427	dh->dccph_dport	= inet_rsk(req)->rmt_port;
428	dh->dccph_doff	= (dccp_header_size +
429			   DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
430	dh->dccph_type	= DCCP_PKT_RESPONSE;
431	dh->dccph_x	= 1;
432	dccp_hdr_set_seq(dh, dreq->dreq_iss);
433	dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
434	dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
435
436	dccp_csum_outgoing(skb);
437
438	/* We use `acked' to remember that a Response was already sent. */
439	inet_rsk(req)->acked = 1;
440	DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
441	return skb;
442response_failed:
443	kfree_skb(skb);
444	return NULL;
445}
446
447EXPORT_SYMBOL_GPL(dccp_make_response);
448
449/* answer offending packet in @rcv_skb with Reset from control socket @ctl */
450struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
451{
452	struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
453	struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
454	const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
455				       sizeof(struct dccp_hdr_ext) +
456				       sizeof(struct dccp_hdr_reset);
457	struct dccp_hdr_reset *dhr;
458	struct sk_buff *skb;
459
460	skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
461	if (skb == NULL)
462		return NULL;
463
464	skb_reserve(skb, sk->sk_prot->max_header);
465
466	/* Swap the send and the receive. */
467	dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
468	dh->dccph_type	= DCCP_PKT_RESET;
469	dh->dccph_sport	= rxdh->dccph_dport;
470	dh->dccph_dport	= rxdh->dccph_sport;
471	dh->dccph_doff	= dccp_hdr_reset_len / 4;
472	dh->dccph_x	= 1;
473
474	dhr = dccp_hdr_reset(skb);
475	dhr->dccph_reset_code = dcb->dccpd_reset_code;
476
477	switch (dcb->dccpd_reset_code) {
478	case DCCP_RESET_CODE_PACKET_ERROR:
479		dhr->dccph_reset_data[0] = rxdh->dccph_type;
480		break;
481	case DCCP_RESET_CODE_OPTION_ERROR:	/* fall through */
482	case DCCP_RESET_CODE_MANDATORY_ERROR:
483		memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
484		break;
485	}
486	/*
487	 * From RFC 4340, 8.3.1:
488	 *   If P.ackno exists, set R.seqno := P.ackno + 1.
489	 *   Else set R.seqno := 0.
490	 */
491	if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
492		dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
493	dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
494
495	dccp_csum_outgoing(skb);
496	return skb;
497}
498
499EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
500
501/* send Reset on established socket, to close or abort the connection */
502int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
503{
504	struct sk_buff *skb;
505	/*
506	 * FIXME: what if rebuild_header fails?
507	 * Should we be doing a rebuild_header here?
508	 */
509	int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
510
511	if (err != 0)
512		return err;
513
514	skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
515	if (skb == NULL)
516		return -ENOBUFS;
517
518	/* Reserve space for headers and prepare control bits. */
519	skb_reserve(skb, sk->sk_prot->max_header);
520	DCCP_SKB_CB(skb)->dccpd_type	   = DCCP_PKT_RESET;
521	DCCP_SKB_CB(skb)->dccpd_reset_code = code;
522
523	return dccp_transmit_skb(sk, skb);
524}
525
526/*
527 * Do all connect socket setups that can be done AF independent.
528 */
529int dccp_connect(struct sock *sk)
530{
531	struct sk_buff *skb;
532	struct dccp_sock *dp = dccp_sk(sk);
533	struct dst_entry *dst = __sk_dst_get(sk);
534	struct inet_connection_sock *icsk = inet_csk(sk);
535
536	sk->sk_err = 0;
537	sock_reset_flag(sk, SOCK_DONE);
538
539	dccp_sync_mss(sk, dst_mtu(dst));
540
541	/* do not connect if feature negotiation setup fails */
542	if (dccp_feat_finalise_settings(dccp_sk(sk)))
543		return -EPROTO;
544
545	/* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
546	dp->dccps_gar = dp->dccps_iss;
547
548	skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
549	if (unlikely(skb == NULL))
550		return -ENOBUFS;
551
552	/* Reserve space for headers. */
553	skb_reserve(skb, sk->sk_prot->max_header);
554
555	DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
556
557	dccp_transmit_skb(sk, dccp_skb_entail(sk, skb));
558	DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
559
560	/* Timer for repeating the REQUEST until an answer. */
561	icsk->icsk_retransmits = 0;
562	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
563				  icsk->icsk_rto, DCCP_RTO_MAX);
564	return 0;
565}
566
567EXPORT_SYMBOL_GPL(dccp_connect);
568
569void dccp_send_ack(struct sock *sk)
570{
571	/* If we have been reset, we may not send again. */
572	if (sk->sk_state != DCCP_CLOSED) {
573		struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
574						GFP_ATOMIC);
575
576		if (skb == NULL) {
577			inet_csk_schedule_ack(sk);
578			inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
579			inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
580						  TCP_DELACK_MAX,
581						  DCCP_RTO_MAX);
582			return;
583		}
584
585		/* Reserve space for headers */
586		skb_reserve(skb, sk->sk_prot->max_header);
587		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
588		dccp_transmit_skb(sk, skb);
589	}
590}
591
592EXPORT_SYMBOL_GPL(dccp_send_ack);
593
594#if 0
595/* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
596void dccp_send_delayed_ack(struct sock *sk)
597{
598	struct inet_connection_sock *icsk = inet_csk(sk);
599	/*
600	 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
601	 * with using 2s, and active senders also piggyback the ACK into a
602	 * DATAACK packet, so this is really for quiescent senders.
603	 */
604	unsigned long timeout = jiffies + 2 * HZ;
605
606	/* Use new timeout only if there wasn't a older one earlier. */
607	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
608		/* If delack timer was blocked or is about to expire,
609		 * send ACK now.
610		 *
611		 * FIXME: check the "about to expire" part
612		 */
613		if (icsk->icsk_ack.blocked) {
614			dccp_send_ack(sk);
615			return;
616		}
617
618		if (!time_before(timeout, icsk->icsk_ack.timeout))
619			timeout = icsk->icsk_ack.timeout;
620	}
621	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
622	icsk->icsk_ack.timeout = timeout;
623	sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
624}
625#endif
626
627void dccp_send_sync(struct sock *sk, const u64 ackno,
628		    const enum dccp_pkt_type pkt_type)
629{
630	/*
631	 * We are not putting this on the write queue, so
632	 * dccp_transmit_skb() will set the ownership to this
633	 * sock.
634	 */
635	struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
636
637	if (skb == NULL) {
638		/* FIXME: how to make sure the sync is sent? */
639		DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
640		return;
641	}
642
643	/* Reserve space for headers and prepare control bits. */
644	skb_reserve(skb, sk->sk_prot->max_header);
645	DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
646	DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
647
648	/*
649	 * Clear the flag in case the Sync was scheduled for out-of-band data,
650	 * such as carrying a long Ack Vector.
651	 */
652	dccp_sk(sk)->dccps_sync_scheduled = 0;
653
654	dccp_transmit_skb(sk, skb);
655}
656
657EXPORT_SYMBOL_GPL(dccp_send_sync);
658
659/*
660 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
661 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
662 * any circumstances.
663 */
664void dccp_send_close(struct sock *sk, const int active)
665{
666	struct dccp_sock *dp = dccp_sk(sk);
667	struct sk_buff *skb;
668	const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
669
670	skb = alloc_skb(sk->sk_prot->max_header, prio);
671	if (skb == NULL)
672		return;
673
674	/* Reserve space for headers and prepare control bits. */
675	skb_reserve(skb, sk->sk_prot->max_header);
676	if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
677		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
678	else
679		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
680
681	if (active) {
682		skb = dccp_skb_entail(sk, skb);
683		/*
684		 * Retransmission timer for active-close: RFC 4340, 8.3 requires
685		 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
686		 * state can be left. The initial timeout is 2 RTTs.
687		 * Since RTT measurement is done by the CCIDs, there is no easy
688		 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
689		 * is too low (200ms); we use a high value to avoid unnecessary
690		 * retransmissions when the link RTT is > 0.2 seconds.
691		 * FIXME: Let main module sample RTTs and use that instead.
692		 */
693		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
694					  DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
695	}
696	dccp_transmit_skb(sk, skb);
697}
v4.17
  1/*
  2 *  net/dccp/output.c
  3 *
  4 *  An implementation of the DCCP protocol
  5 *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  6 *
  7 *	This program is free software; you can redistribute it and/or
  8 *	modify it under the terms of the GNU General Public License
  9 *	as published by the Free Software Foundation; either version
 10 *	2 of the License, or (at your option) any later version.
 11 */
 12
 13#include <linux/dccp.h>
 14#include <linux/kernel.h>
 15#include <linux/skbuff.h>
 16#include <linux/slab.h>
 17#include <linux/sched/signal.h>
 18
 19#include <net/inet_sock.h>
 20#include <net/sock.h>
 21
 22#include "ackvec.h"
 23#include "ccid.h"
 24#include "dccp.h"
 25
 26static inline void dccp_event_ack_sent(struct sock *sk)
 27{
 28	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
 29}
 30
 31/* enqueue @skb on sk_send_head for retransmission, return clone to send now */
 32static struct sk_buff *dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
 33{
 34	skb_set_owner_w(skb, sk);
 35	WARN_ON(sk->sk_send_head);
 36	sk->sk_send_head = skb;
 37	return skb_clone(sk->sk_send_head, gfp_any());
 38}
 39
 40/*
 41 * All SKB's seen here are completely headerless. It is our
 42 * job to build the DCCP header, and pass the packet down to
 43 * IP so it can do the same plus pass the packet off to the
 44 * device.
 45 */
 46static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
 47{
 48	if (likely(skb != NULL)) {
 49		struct inet_sock *inet = inet_sk(sk);
 50		const struct inet_connection_sock *icsk = inet_csk(sk);
 51		struct dccp_sock *dp = dccp_sk(sk);
 52		struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 53		struct dccp_hdr *dh;
 54		/* XXX For now we're using only 48 bits sequence numbers */
 55		const u32 dccp_header_size = sizeof(*dh) +
 56					     sizeof(struct dccp_hdr_ext) +
 57					  dccp_packet_hdr_len(dcb->dccpd_type);
 58		int err, set_ack = 1;
 59		u64 ackno = dp->dccps_gsr;
 60		/*
 61		 * Increment GSS here already in case the option code needs it.
 62		 * Update GSS for real only if option processing below succeeds.
 63		 */
 64		dcb->dccpd_seq = ADD48(dp->dccps_gss, 1);
 65
 66		switch (dcb->dccpd_type) {
 67		case DCCP_PKT_DATA:
 68			set_ack = 0;
 69			/* fall through */
 70		case DCCP_PKT_DATAACK:
 71		case DCCP_PKT_RESET:
 72			break;
 73
 74		case DCCP_PKT_REQUEST:
 75			set_ack = 0;
 76			/* Use ISS on the first (non-retransmitted) Request. */
 77			if (icsk->icsk_retransmits == 0)
 78				dcb->dccpd_seq = dp->dccps_iss;
 79			/* fall through */
 80
 81		case DCCP_PKT_SYNC:
 82		case DCCP_PKT_SYNCACK:
 83			ackno = dcb->dccpd_ack_seq;
 84			/* fall through */
 85		default:
 86			/*
 87			 * Set owner/destructor: some skbs are allocated via
 88			 * alloc_skb (e.g. when retransmission may happen).
 89			 * Only Data, DataAck, and Reset packets should come
 90			 * through here with skb->sk set.
 91			 */
 92			WARN_ON(skb->sk);
 93			skb_set_owner_w(skb, sk);
 94			break;
 95		}
 96
 97		if (dccp_insert_options(sk, skb)) {
 98			kfree_skb(skb);
 99			return -EPROTO;
100		}
101
102
103		/* Build DCCP header and checksum it. */
104		dh = dccp_zeroed_hdr(skb, dccp_header_size);
105		dh->dccph_type	= dcb->dccpd_type;
106		dh->dccph_sport	= inet->inet_sport;
107		dh->dccph_dport	= inet->inet_dport;
108		dh->dccph_doff	= (dccp_header_size + dcb->dccpd_opt_len) / 4;
109		dh->dccph_ccval	= dcb->dccpd_ccval;
110		dh->dccph_cscov = dp->dccps_pcslen;
111		/* XXX For now we're using only 48 bits sequence numbers */
112		dh->dccph_x	= 1;
113
114		dccp_update_gss(sk, dcb->dccpd_seq);
115		dccp_hdr_set_seq(dh, dp->dccps_gss);
116		if (set_ack)
117			dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
118
119		switch (dcb->dccpd_type) {
120		case DCCP_PKT_REQUEST:
121			dccp_hdr_request(skb)->dccph_req_service =
122							dp->dccps_service;
123			/*
124			 * Limit Ack window to ISS <= P.ackno <= GSS, so that
125			 * only Responses to Requests we sent are considered.
126			 */
127			dp->dccps_awl = dp->dccps_iss;
128			break;
129		case DCCP_PKT_RESET:
130			dccp_hdr_reset(skb)->dccph_reset_code =
131							dcb->dccpd_reset_code;
132			break;
133		}
134
135		icsk->icsk_af_ops->send_check(sk, skb);
136
137		if (set_ack)
138			dccp_event_ack_sent(sk);
139
140		DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
141
142		err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
143		return net_xmit_eval(err);
144	}
145	return -ENOBUFS;
146}
147
148/**
149 * dccp_determine_ccmps  -  Find out about CCID-specific packet-size limits
150 * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
151 * since the RX CCID is restricted to feedback packets (Acks), which are small
152 * in comparison with the data traffic. A value of 0 means "no current CCMPS".
153 */
154static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
155{
156	const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
157
158	if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
159		return 0;
160	return tx_ccid->ccid_ops->ccid_ccmps;
161}
162
163unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
164{
165	struct inet_connection_sock *icsk = inet_csk(sk);
166	struct dccp_sock *dp = dccp_sk(sk);
167	u32 ccmps = dccp_determine_ccmps(dp);
168	u32 cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
169
170	/* Account for header lengths and IPv4/v6 option overhead */
171	cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
172		    sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
173
174	/*
175	 * Leave enough headroom for common DCCP header options.
176	 * This only considers options which may appear on DCCP-Data packets, as
177	 * per table 3 in RFC 4340, 5.8. When running out of space for other
178	 * options (eg. Ack Vector which can take up to 255 bytes), it is better
179	 * to schedule a separate Ack. Thus we leave headroom for the following:
180	 *  - 1 byte for Slow Receiver (11.6)
181	 *  - 6 bytes for Timestamp (13.1)
182	 *  - 10 bytes for Timestamp Echo (13.3)
183	 *  - 8 bytes for NDP count (7.7, when activated)
184	 *  - 6 bytes for Data Checksum (9.3)
185	 *  - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
186	 */
187	cur_mps -= roundup(1 + 6 + 10 + dp->dccps_send_ndp_count * 8 + 6 +
188			   (dp->dccps_hc_rx_ackvec ? DCCPAV_MIN_OPTLEN : 0), 4);
189
190	/* And store cached results */
191	icsk->icsk_pmtu_cookie = pmtu;
192	dp->dccps_mss_cache = cur_mps;
193
194	return cur_mps;
195}
196
197EXPORT_SYMBOL_GPL(dccp_sync_mss);
198
199void dccp_write_space(struct sock *sk)
200{
201	struct socket_wq *wq;
202
203	rcu_read_lock();
204	wq = rcu_dereference(sk->sk_wq);
205	if (skwq_has_sleeper(wq))
206		wake_up_interruptible(&wq->wait);
207	/* Should agree with poll, otherwise some programs break */
208	if (sock_writeable(sk))
209		sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
210
211	rcu_read_unlock();
212}
213
214/**
215 * dccp_wait_for_ccid  -  Await CCID send permission
216 * @sk:    socket to wait for
217 * @delay: timeout in jiffies
218 *
219 * This is used by CCIDs which need to delay the send time in process context.
220 */
221static int dccp_wait_for_ccid(struct sock *sk, unsigned long delay)
222{
223	DEFINE_WAIT(wait);
224	long remaining;
225
226	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
227	sk->sk_write_pending++;
228	release_sock(sk);
229
230	remaining = schedule_timeout(delay);
231
232	lock_sock(sk);
233	sk->sk_write_pending--;
234	finish_wait(sk_sleep(sk), &wait);
235
236	if (signal_pending(current) || sk->sk_err)
237		return -1;
238	return remaining;
239}
240
241/**
242 * dccp_xmit_packet  -  Send data packet under control of CCID
243 * Transmits next-queued payload and informs CCID to account for the packet.
244 */
245static void dccp_xmit_packet(struct sock *sk)
246{
247	int err, len;
248	struct dccp_sock *dp = dccp_sk(sk);
249	struct sk_buff *skb = dccp_qpolicy_pop(sk);
250
251	if (unlikely(skb == NULL))
252		return;
253	len = skb->len;
254
255	if (sk->sk_state == DCCP_PARTOPEN) {
256		const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
257		/*
258		 * See 8.1.5 - Handshake Completion.
259		 *
260		 * For robustness we resend Confirm options until the client has
261		 * entered OPEN. During the initial feature negotiation, the MPS
262		 * is smaller than usual, reduced by the Change/Confirm options.
263		 */
264		if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
265			DCCP_WARN("Payload too large (%d) for featneg.\n", len);
266			dccp_send_ack(sk);
267			dccp_feat_list_purge(&dp->dccps_featneg);
268		}
269
270		inet_csk_schedule_ack(sk);
271		inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
272					      inet_csk(sk)->icsk_rto,
273					      DCCP_RTO_MAX);
274		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
275	} else if (dccp_ack_pending(sk)) {
276		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
277	} else {
278		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATA;
279	}
280
281	err = dccp_transmit_skb(sk, skb);
282	if (err)
283		dccp_pr_debug("transmit_skb() returned err=%d\n", err);
284	/*
285	 * Register this one as sent even if an error occurred. To the remote
286	 * end a local packet drop is indistinguishable from network loss, i.e.
287	 * any local drop will eventually be reported via receiver feedback.
288	 */
289	ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, len);
290
291	/*
292	 * If the CCID needs to transfer additional header options out-of-band
293	 * (e.g. Ack Vectors or feature-negotiation options), it activates this
294	 * flag to schedule a Sync. The Sync will automatically incorporate all
295	 * currently pending header options, thus clearing the backlog.
296	 */
297	if (dp->dccps_sync_scheduled)
298		dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
299}
300
301/**
302 * dccp_flush_write_queue  -  Drain queue at end of connection
303 * Since dccp_sendmsg queues packets without waiting for them to be sent, it may
304 * happen that the TX queue is not empty at the end of a connection. We give the
305 * HC-sender CCID a grace period of up to @time_budget jiffies. If this function
306 * returns with a non-empty write queue, it will be purged later.
307 */
308void dccp_flush_write_queue(struct sock *sk, long *time_budget)
309{
310	struct dccp_sock *dp = dccp_sk(sk);
311	struct sk_buff *skb;
312	long delay, rc;
313
314	while (*time_budget > 0 && (skb = skb_peek(&sk->sk_write_queue))) {
315		rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
316
317		switch (ccid_packet_dequeue_eval(rc)) {
318		case CCID_PACKET_WILL_DEQUEUE_LATER:
319			/*
320			 * If the CCID determines when to send, the next sending
321			 * time is unknown or the CCID may not even send again
322			 * (e.g. remote host crashes or lost Ack packets).
323			 */
324			DCCP_WARN("CCID did not manage to send all packets\n");
325			return;
326		case CCID_PACKET_DELAY:
327			delay = msecs_to_jiffies(rc);
328			if (delay > *time_budget)
329				return;
330			rc = dccp_wait_for_ccid(sk, delay);
331			if (rc < 0)
332				return;
333			*time_budget -= (delay - rc);
334			/* check again if we can send now */
335			break;
336		case CCID_PACKET_SEND_AT_ONCE:
337			dccp_xmit_packet(sk);
338			break;
339		case CCID_PACKET_ERR:
340			skb_dequeue(&sk->sk_write_queue);
341			kfree_skb(skb);
342			dccp_pr_debug("packet discarded due to err=%ld\n", rc);
343		}
344	}
345}
346
347void dccp_write_xmit(struct sock *sk)
348{
349	struct dccp_sock *dp = dccp_sk(sk);
350	struct sk_buff *skb;
351
352	while ((skb = dccp_qpolicy_top(sk))) {
353		int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
354
355		switch (ccid_packet_dequeue_eval(rc)) {
356		case CCID_PACKET_WILL_DEQUEUE_LATER:
357			return;
358		case CCID_PACKET_DELAY:
359			sk_reset_timer(sk, &dp->dccps_xmit_timer,
360				       jiffies + msecs_to_jiffies(rc));
361			return;
362		case CCID_PACKET_SEND_AT_ONCE:
363			dccp_xmit_packet(sk);
364			break;
365		case CCID_PACKET_ERR:
366			dccp_qpolicy_drop(sk, skb);
367			dccp_pr_debug("packet discarded due to err=%d\n", rc);
368		}
369	}
370}
371
372/**
373 * dccp_retransmit_skb  -  Retransmit Request, Close, or CloseReq packets
374 * There are only four retransmittable packet types in DCCP:
375 * - Request  in client-REQUEST  state (sec. 8.1.1),
376 * - CloseReq in server-CLOSEREQ state (sec. 8.3),
377 * - Close    in   node-CLOSING  state (sec. 8.3),
378 * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
379 * This function expects sk->sk_send_head to contain the original skb.
380 */
381int dccp_retransmit_skb(struct sock *sk)
382{
383	WARN_ON(sk->sk_send_head == NULL);
384
385	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
386		return -EHOSTUNREACH; /* Routing failure or similar. */
387
388	/* this count is used to distinguish original and retransmitted skb */
389	inet_csk(sk)->icsk_retransmits++;
390
391	return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
392}
393
394struct sk_buff *dccp_make_response(const struct sock *sk, struct dst_entry *dst,
395				   struct request_sock *req)
396{
397	struct dccp_hdr *dh;
398	struct dccp_request_sock *dreq;
399	const u32 dccp_header_size = sizeof(struct dccp_hdr) +
400				     sizeof(struct dccp_hdr_ext) +
401				     sizeof(struct dccp_hdr_response);
402	struct sk_buff *skb;
403
404	/* sk is marked const to clearly express we dont hold socket lock.
405	 * sock_wmalloc() will atomically change sk->sk_wmem_alloc,
406	 * it is safe to promote sk to non const.
407	 */
408	skb = sock_wmalloc((struct sock *)sk, MAX_DCCP_HEADER, 1,
409			   GFP_ATOMIC);
410	if (!skb)
411		return NULL;
412
413	skb_reserve(skb, MAX_DCCP_HEADER);
 
414
415	skb_dst_set(skb, dst_clone(dst));
416
417	dreq = dccp_rsk(req);
418	if (inet_rsk(req)->acked)	/* increase GSS upon retransmission */
419		dccp_inc_seqno(&dreq->dreq_gss);
420	DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
421	DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_gss;
422
423	/* Resolve feature dependencies resulting from choice of CCID */
424	if (dccp_feat_server_ccid_dependencies(dreq))
425		goto response_failed;
426
427	if (dccp_insert_options_rsk(dreq, skb))
428		goto response_failed;
429
430	/* Build and checksum header */
431	dh = dccp_zeroed_hdr(skb, dccp_header_size);
432
433	dh->dccph_sport	= htons(inet_rsk(req)->ir_num);
434	dh->dccph_dport	= inet_rsk(req)->ir_rmt_port;
435	dh->dccph_doff	= (dccp_header_size +
436			   DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
437	dh->dccph_type	= DCCP_PKT_RESPONSE;
438	dh->dccph_x	= 1;
439	dccp_hdr_set_seq(dh, dreq->dreq_gss);
440	dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_gsr);
441	dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
442
443	dccp_csum_outgoing(skb);
444
445	/* We use `acked' to remember that a Response was already sent. */
446	inet_rsk(req)->acked = 1;
447	DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
448	return skb;
449response_failed:
450	kfree_skb(skb);
451	return NULL;
452}
453
454EXPORT_SYMBOL_GPL(dccp_make_response);
455
456/* answer offending packet in @rcv_skb with Reset from control socket @ctl */
457struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
458{
459	struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
460	struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
461	const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
462				       sizeof(struct dccp_hdr_ext) +
463				       sizeof(struct dccp_hdr_reset);
464	struct dccp_hdr_reset *dhr;
465	struct sk_buff *skb;
466
467	skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
468	if (skb == NULL)
469		return NULL;
470
471	skb_reserve(skb, sk->sk_prot->max_header);
472
473	/* Swap the send and the receive. */
474	dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
475	dh->dccph_type	= DCCP_PKT_RESET;
476	dh->dccph_sport	= rxdh->dccph_dport;
477	dh->dccph_dport	= rxdh->dccph_sport;
478	dh->dccph_doff	= dccp_hdr_reset_len / 4;
479	dh->dccph_x	= 1;
480
481	dhr = dccp_hdr_reset(skb);
482	dhr->dccph_reset_code = dcb->dccpd_reset_code;
483
484	switch (dcb->dccpd_reset_code) {
485	case DCCP_RESET_CODE_PACKET_ERROR:
486		dhr->dccph_reset_data[0] = rxdh->dccph_type;
487		break;
488	case DCCP_RESET_CODE_OPTION_ERROR:	/* fall through */
489	case DCCP_RESET_CODE_MANDATORY_ERROR:
490		memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
491		break;
492	}
493	/*
494	 * From RFC 4340, 8.3.1:
495	 *   If P.ackno exists, set R.seqno := P.ackno + 1.
496	 *   Else set R.seqno := 0.
497	 */
498	if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
499		dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
500	dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
501
502	dccp_csum_outgoing(skb);
503	return skb;
504}
505
506EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
507
508/* send Reset on established socket, to close or abort the connection */
509int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
510{
511	struct sk_buff *skb;
512	/*
513	 * FIXME: what if rebuild_header fails?
514	 * Should we be doing a rebuild_header here?
515	 */
516	int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
517
518	if (err != 0)
519		return err;
520
521	skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
522	if (skb == NULL)
523		return -ENOBUFS;
524
525	/* Reserve space for headers and prepare control bits. */
526	skb_reserve(skb, sk->sk_prot->max_header);
527	DCCP_SKB_CB(skb)->dccpd_type	   = DCCP_PKT_RESET;
528	DCCP_SKB_CB(skb)->dccpd_reset_code = code;
529
530	return dccp_transmit_skb(sk, skb);
531}
532
533/*
534 * Do all connect socket setups that can be done AF independent.
535 */
536int dccp_connect(struct sock *sk)
537{
538	struct sk_buff *skb;
539	struct dccp_sock *dp = dccp_sk(sk);
540	struct dst_entry *dst = __sk_dst_get(sk);
541	struct inet_connection_sock *icsk = inet_csk(sk);
542
543	sk->sk_err = 0;
544	sock_reset_flag(sk, SOCK_DONE);
545
546	dccp_sync_mss(sk, dst_mtu(dst));
547
548	/* do not connect if feature negotiation setup fails */
549	if (dccp_feat_finalise_settings(dccp_sk(sk)))
550		return -EPROTO;
551
552	/* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
553	dp->dccps_gar = dp->dccps_iss;
554
555	skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
556	if (unlikely(skb == NULL))
557		return -ENOBUFS;
558
559	/* Reserve space for headers. */
560	skb_reserve(skb, sk->sk_prot->max_header);
561
562	DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
563
564	dccp_transmit_skb(sk, dccp_skb_entail(sk, skb));
565	DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
566
567	/* Timer for repeating the REQUEST until an answer. */
568	icsk->icsk_retransmits = 0;
569	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
570				  icsk->icsk_rto, DCCP_RTO_MAX);
571	return 0;
572}
573
574EXPORT_SYMBOL_GPL(dccp_connect);
575
576void dccp_send_ack(struct sock *sk)
577{
578	/* If we have been reset, we may not send again. */
579	if (sk->sk_state != DCCP_CLOSED) {
580		struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
581						GFP_ATOMIC);
582
583		if (skb == NULL) {
584			inet_csk_schedule_ack(sk);
585			inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
586			inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
587						  TCP_DELACK_MAX,
588						  DCCP_RTO_MAX);
589			return;
590		}
591
592		/* Reserve space for headers */
593		skb_reserve(skb, sk->sk_prot->max_header);
594		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
595		dccp_transmit_skb(sk, skb);
596	}
597}
598
599EXPORT_SYMBOL_GPL(dccp_send_ack);
600
601#if 0
602/* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
603void dccp_send_delayed_ack(struct sock *sk)
604{
605	struct inet_connection_sock *icsk = inet_csk(sk);
606	/*
607	 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
608	 * with using 2s, and active senders also piggyback the ACK into a
609	 * DATAACK packet, so this is really for quiescent senders.
610	 */
611	unsigned long timeout = jiffies + 2 * HZ;
612
613	/* Use new timeout only if there wasn't a older one earlier. */
614	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
615		/* If delack timer was blocked or is about to expire,
616		 * send ACK now.
617		 *
618		 * FIXME: check the "about to expire" part
619		 */
620		if (icsk->icsk_ack.blocked) {
621			dccp_send_ack(sk);
622			return;
623		}
624
625		if (!time_before(timeout, icsk->icsk_ack.timeout))
626			timeout = icsk->icsk_ack.timeout;
627	}
628	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
629	icsk->icsk_ack.timeout = timeout;
630	sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
631}
632#endif
633
634void dccp_send_sync(struct sock *sk, const u64 ackno,
635		    const enum dccp_pkt_type pkt_type)
636{
637	/*
638	 * We are not putting this on the write queue, so
639	 * dccp_transmit_skb() will set the ownership to this
640	 * sock.
641	 */
642	struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
643
644	if (skb == NULL) {
645		/* FIXME: how to make sure the sync is sent? */
646		DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
647		return;
648	}
649
650	/* Reserve space for headers and prepare control bits. */
651	skb_reserve(skb, sk->sk_prot->max_header);
652	DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
653	DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
654
655	/*
656	 * Clear the flag in case the Sync was scheduled for out-of-band data,
657	 * such as carrying a long Ack Vector.
658	 */
659	dccp_sk(sk)->dccps_sync_scheduled = 0;
660
661	dccp_transmit_skb(sk, skb);
662}
663
664EXPORT_SYMBOL_GPL(dccp_send_sync);
665
666/*
667 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
668 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
669 * any circumstances.
670 */
671void dccp_send_close(struct sock *sk, const int active)
672{
673	struct dccp_sock *dp = dccp_sk(sk);
674	struct sk_buff *skb;
675	const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
676
677	skb = alloc_skb(sk->sk_prot->max_header, prio);
678	if (skb == NULL)
679		return;
680
681	/* Reserve space for headers and prepare control bits. */
682	skb_reserve(skb, sk->sk_prot->max_header);
683	if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
684		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
685	else
686		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
687
688	if (active) {
689		skb = dccp_skb_entail(sk, skb);
690		/*
691		 * Retransmission timer for active-close: RFC 4340, 8.3 requires
692		 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
693		 * state can be left. The initial timeout is 2 RTTs.
694		 * Since RTT measurement is done by the CCIDs, there is no easy
695		 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
696		 * is too low (200ms); we use a high value to avoid unnecessary
697		 * retransmissions when the link RTT is > 0.2 seconds.
698		 * FIXME: Let main module sample RTTs and use that instead.
699		 */
700		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
701					  DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
702	}
703	dccp_transmit_skb(sk, skb);
704}