• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  net/dccp/output.c
4  *
5  *  An implementation of the DCCP protocol
6  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7  */
8 
9 #include <linux/dccp.h>
10 #include <linux/kernel.h>
11 #include <linux/skbuff.h>
12 #include <linux/slab.h>
13 #include <linux/sched/signal.h>
14 
15 #include <net/inet_sock.h>
16 #include <net/sock.h>
17 
18 #include "ackvec.h"
19 #include "ccid.h"
20 #include "dccp.h"
21 
dccp_event_ack_sent(struct sock * sk)22 static inline void dccp_event_ack_sent(struct sock *sk)
23 {
24 	inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
25 }
26 
27 /* enqueue @skb on sk_send_head for retransmission, return clone to send now */
dccp_skb_entail(struct sock * sk,struct sk_buff * skb)28 static struct sk_buff *dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
29 {
30 	skb_set_owner_w(skb, sk);
31 	WARN_ON(sk->sk_send_head);
32 	sk->sk_send_head = skb;
33 	return skb_clone(sk->sk_send_head, gfp_any());
34 }
35 
36 /*
37  * All SKB's seen here are completely headerless. It is our
38  * job to build the DCCP header, and pass the packet down to
39  * IP so it can do the same plus pass the packet off to the
40  * device.
41  */
dccp_transmit_skb(struct sock * sk,struct sk_buff * skb)42 static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
43 {
44 	if (likely(skb != NULL)) {
45 		struct inet_sock *inet = inet_sk(sk);
46 		const struct inet_connection_sock *icsk = inet_csk(sk);
47 		struct dccp_sock *dp = dccp_sk(sk);
48 		struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
49 		struct dccp_hdr *dh;
50 		/* XXX For now we're using only 48 bits sequence numbers */
51 		const u32 dccp_header_size = sizeof(*dh) +
52 					     sizeof(struct dccp_hdr_ext) +
53 					  dccp_packet_hdr_len(dcb->dccpd_type);
54 		int err, set_ack = 1;
55 		u64 ackno = dp->dccps_gsr;
56 		/*
57 		 * Increment GSS here already in case the option code needs it.
58 		 * Update GSS for real only if option processing below succeeds.
59 		 */
60 		dcb->dccpd_seq = ADD48(dp->dccps_gss, 1);
61 
62 		switch (dcb->dccpd_type) {
63 		case DCCP_PKT_DATA:
64 			set_ack = 0;
65 			/* fall through */
66 		case DCCP_PKT_DATAACK:
67 		case DCCP_PKT_RESET:
68 			break;
69 
70 		case DCCP_PKT_REQUEST:
71 			set_ack = 0;
72 			/* Use ISS on the first (non-retransmitted) Request. */
73 			if (icsk->icsk_retransmits == 0)
74 				dcb->dccpd_seq = dp->dccps_iss;
75 			/* fall through */
76 
77 		case DCCP_PKT_SYNC:
78 		case DCCP_PKT_SYNCACK:
79 			ackno = dcb->dccpd_ack_seq;
80 			/* fall through */
81 		default:
82 			/*
83 			 * Set owner/destructor: some skbs are allocated via
84 			 * alloc_skb (e.g. when retransmission may happen).
85 			 * Only Data, DataAck, and Reset packets should come
86 			 * through here with skb->sk set.
87 			 */
88 			WARN_ON(skb->sk);
89 			skb_set_owner_w(skb, sk);
90 			break;
91 		}
92 
93 		if (dccp_insert_options(sk, skb)) {
94 			kfree_skb(skb);
95 			return -EPROTO;
96 		}
97 
98 
99 		/* Build DCCP header and checksum it. */
100 		dh = dccp_zeroed_hdr(skb, dccp_header_size);
101 		dh->dccph_type	= dcb->dccpd_type;
102 		dh->dccph_sport	= inet->inet_sport;
103 		dh->dccph_dport	= inet->inet_dport;
104 		dh->dccph_doff	= (dccp_header_size + dcb->dccpd_opt_len) / 4;
105 		dh->dccph_ccval	= dcb->dccpd_ccval;
106 		dh->dccph_cscov = dp->dccps_pcslen;
107 		/* XXX For now we're using only 48 bits sequence numbers */
108 		dh->dccph_x	= 1;
109 
110 		dccp_update_gss(sk, dcb->dccpd_seq);
111 		dccp_hdr_set_seq(dh, dp->dccps_gss);
112 		if (set_ack)
113 			dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
114 
115 		switch (dcb->dccpd_type) {
116 		case DCCP_PKT_REQUEST:
117 			dccp_hdr_request(skb)->dccph_req_service =
118 							dp->dccps_service;
119 			/*
120 			 * Limit Ack window to ISS <= P.ackno <= GSS, so that
121 			 * only Responses to Requests we sent are considered.
122 			 */
123 			dp->dccps_awl = dp->dccps_iss;
124 			break;
125 		case DCCP_PKT_RESET:
126 			dccp_hdr_reset(skb)->dccph_reset_code =
127 							dcb->dccpd_reset_code;
128 			break;
129 		}
130 
131 		icsk->icsk_af_ops->send_check(sk, skb);
132 
133 		if (set_ack)
134 			dccp_event_ack_sent(sk);
135 
136 		DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
137 
138 		err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
139 		return net_xmit_eval(err);
140 	}
141 	return -ENOBUFS;
142 }
143 
144 /**
145  * dccp_determine_ccmps  -  Find out about CCID-specific packet-size limits
146  * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
147  * since the RX CCID is restricted to feedback packets (Acks), which are small
148  * in comparison with the data traffic. A value of 0 means "no current CCMPS".
149  */
dccp_determine_ccmps(const struct dccp_sock * dp)150 static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
151 {
152 	const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
153 
154 	if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
155 		return 0;
156 	return tx_ccid->ccid_ops->ccid_ccmps;
157 }
158 
dccp_sync_mss(struct sock * sk,u32 pmtu)159 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
160 {
161 	struct inet_connection_sock *icsk = inet_csk(sk);
162 	struct dccp_sock *dp = dccp_sk(sk);
163 	u32 ccmps = dccp_determine_ccmps(dp);
164 	u32 cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
165 
166 	/* Account for header lengths and IPv4/v6 option overhead */
167 	cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
168 		    sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
169 
170 	/*
171 	 * Leave enough headroom for common DCCP header options.
172 	 * This only considers options which may appear on DCCP-Data packets, as
173 	 * per table 3 in RFC 4340, 5.8. When running out of space for other
174 	 * options (eg. Ack Vector which can take up to 255 bytes), it is better
175 	 * to schedule a separate Ack. Thus we leave headroom for the following:
176 	 *  - 1 byte for Slow Receiver (11.6)
177 	 *  - 6 bytes for Timestamp (13.1)
178 	 *  - 10 bytes for Timestamp Echo (13.3)
179 	 *  - 8 bytes for NDP count (7.7, when activated)
180 	 *  - 6 bytes for Data Checksum (9.3)
181 	 *  - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
182 	 */
183 	cur_mps -= roundup(1 + 6 + 10 + dp->dccps_send_ndp_count * 8 + 6 +
184 			   (dp->dccps_hc_rx_ackvec ? DCCPAV_MIN_OPTLEN : 0), 4);
185 
186 	/* And store cached results */
187 	icsk->icsk_pmtu_cookie = pmtu;
188 	dp->dccps_mss_cache = cur_mps;
189 
190 	return cur_mps;
191 }
192 
193 EXPORT_SYMBOL_GPL(dccp_sync_mss);
194 
dccp_write_space(struct sock * sk)195 void dccp_write_space(struct sock *sk)
196 {
197 	struct socket_wq *wq;
198 
199 	rcu_read_lock();
200 	wq = rcu_dereference(sk->sk_wq);
201 	if (skwq_has_sleeper(wq))
202 		wake_up_interruptible(&wq->wait);
203 	/* Should agree with poll, otherwise some programs break */
204 	if (sock_writeable(sk))
205 		sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
206 
207 	rcu_read_unlock();
208 }
209 
210 /**
211  * dccp_wait_for_ccid  -  Await CCID send permission
212  * @sk:    socket to wait for
213  * @delay: timeout in jiffies
214  *
215  * This is used by CCIDs which need to delay the send time in process context.
216  */
dccp_wait_for_ccid(struct sock * sk,unsigned long delay)217 static int dccp_wait_for_ccid(struct sock *sk, unsigned long delay)
218 {
219 	DEFINE_WAIT(wait);
220 	long remaining;
221 
222 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
223 	sk->sk_write_pending++;
224 	release_sock(sk);
225 
226 	remaining = schedule_timeout(delay);
227 
228 	lock_sock(sk);
229 	sk->sk_write_pending--;
230 	finish_wait(sk_sleep(sk), &wait);
231 
232 	if (signal_pending(current) || sk->sk_err)
233 		return -1;
234 	return remaining;
235 }
236 
237 /**
238  * dccp_xmit_packet  -  Send data packet under control of CCID
239  * Transmits next-queued payload and informs CCID to account for the packet.
240  */
dccp_xmit_packet(struct sock * sk)241 static void dccp_xmit_packet(struct sock *sk)
242 {
243 	int err, len;
244 	struct dccp_sock *dp = dccp_sk(sk);
245 	struct sk_buff *skb = dccp_qpolicy_pop(sk);
246 
247 	if (unlikely(skb == NULL))
248 		return;
249 	len = skb->len;
250 
251 	if (sk->sk_state == DCCP_PARTOPEN) {
252 		const u32 cur_mps = dp->dccps_mss_cache - DCCP_FEATNEG_OVERHEAD;
253 		/*
254 		 * See 8.1.5 - Handshake Completion.
255 		 *
256 		 * For robustness we resend Confirm options until the client has
257 		 * entered OPEN. During the initial feature negotiation, the MPS
258 		 * is smaller than usual, reduced by the Change/Confirm options.
259 		 */
260 		if (!list_empty(&dp->dccps_featneg) && len > cur_mps) {
261 			DCCP_WARN("Payload too large (%d) for featneg.\n", len);
262 			dccp_send_ack(sk);
263 			dccp_feat_list_purge(&dp->dccps_featneg);
264 		}
265 
266 		inet_csk_schedule_ack(sk);
267 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
268 					      inet_csk(sk)->icsk_rto,
269 					      DCCP_RTO_MAX);
270 		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
271 	} else if (dccp_ack_pending(sk)) {
272 		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATAACK;
273 	} else {
274 		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_DATA;
275 	}
276 
277 	err = dccp_transmit_skb(sk, skb);
278 	if (err)
279 		dccp_pr_debug("transmit_skb() returned err=%d\n", err);
280 	/*
281 	 * Register this one as sent even if an error occurred. To the remote
282 	 * end a local packet drop is indistinguishable from network loss, i.e.
283 	 * any local drop will eventually be reported via receiver feedback.
284 	 */
285 	ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, len);
286 
287 	/*
288 	 * If the CCID needs to transfer additional header options out-of-band
289 	 * (e.g. Ack Vectors or feature-negotiation options), it activates this
290 	 * flag to schedule a Sync. The Sync will automatically incorporate all
291 	 * currently pending header options, thus clearing the backlog.
292 	 */
293 	if (dp->dccps_sync_scheduled)
294 		dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
295 }
296 
297 /**
298  * dccp_flush_write_queue  -  Drain queue at end of connection
299  * Since dccp_sendmsg queues packets without waiting for them to be sent, it may
300  * happen that the TX queue is not empty at the end of a connection. We give the
301  * HC-sender CCID a grace period of up to @time_budget jiffies. If this function
302  * returns with a non-empty write queue, it will be purged later.
303  */
dccp_flush_write_queue(struct sock * sk,long * time_budget)304 void dccp_flush_write_queue(struct sock *sk, long *time_budget)
305 {
306 	struct dccp_sock *dp = dccp_sk(sk);
307 	struct sk_buff *skb;
308 	long delay, rc;
309 
310 	while (*time_budget > 0 && (skb = skb_peek(&sk->sk_write_queue))) {
311 		rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
312 
313 		switch (ccid_packet_dequeue_eval(rc)) {
314 		case CCID_PACKET_WILL_DEQUEUE_LATER:
315 			/*
316 			 * If the CCID determines when to send, the next sending
317 			 * time is unknown or the CCID may not even send again
318 			 * (e.g. remote host crashes or lost Ack packets).
319 			 */
320 			DCCP_WARN("CCID did not manage to send all packets\n");
321 			return;
322 		case CCID_PACKET_DELAY:
323 			delay = msecs_to_jiffies(rc);
324 			if (delay > *time_budget)
325 				return;
326 			rc = dccp_wait_for_ccid(sk, delay);
327 			if (rc < 0)
328 				return;
329 			*time_budget -= (delay - rc);
330 			/* check again if we can send now */
331 			break;
332 		case CCID_PACKET_SEND_AT_ONCE:
333 			dccp_xmit_packet(sk);
334 			break;
335 		case CCID_PACKET_ERR:
336 			skb_dequeue(&sk->sk_write_queue);
337 			kfree_skb(skb);
338 			dccp_pr_debug("packet discarded due to err=%ld\n", rc);
339 		}
340 	}
341 }
342 
dccp_write_xmit(struct sock * sk)343 void dccp_write_xmit(struct sock *sk)
344 {
345 	struct dccp_sock *dp = dccp_sk(sk);
346 	struct sk_buff *skb;
347 
348 	while ((skb = dccp_qpolicy_top(sk))) {
349 		int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
350 
351 		switch (ccid_packet_dequeue_eval(rc)) {
352 		case CCID_PACKET_WILL_DEQUEUE_LATER:
353 			return;
354 		case CCID_PACKET_DELAY:
355 			sk_reset_timer(sk, &dp->dccps_xmit_timer,
356 				       jiffies + msecs_to_jiffies(rc));
357 			return;
358 		case CCID_PACKET_SEND_AT_ONCE:
359 			dccp_xmit_packet(sk);
360 			break;
361 		case CCID_PACKET_ERR:
362 			dccp_qpolicy_drop(sk, skb);
363 			dccp_pr_debug("packet discarded due to err=%d\n", rc);
364 		}
365 	}
366 }
367 
368 /**
369  * dccp_retransmit_skb  -  Retransmit Request, Close, or CloseReq packets
370  * There are only four retransmittable packet types in DCCP:
371  * - Request  in client-REQUEST  state (sec. 8.1.1),
372  * - CloseReq in server-CLOSEREQ state (sec. 8.3),
373  * - Close    in   node-CLOSING  state (sec. 8.3),
374  * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
375  * This function expects sk->sk_send_head to contain the original skb.
376  */
dccp_retransmit_skb(struct sock * sk)377 int dccp_retransmit_skb(struct sock *sk)
378 {
379 	WARN_ON(sk->sk_send_head == NULL);
380 
381 	if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
382 		return -EHOSTUNREACH; /* Routing failure or similar. */
383 
384 	/* this count is used to distinguish original and retransmitted skb */
385 	inet_csk(sk)->icsk_retransmits++;
386 
387 	return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
388 }
389 
dccp_make_response(const struct sock * sk,struct dst_entry * dst,struct request_sock * req)390 struct sk_buff *dccp_make_response(const struct sock *sk, struct dst_entry *dst,
391 				   struct request_sock *req)
392 {
393 	struct dccp_hdr *dh;
394 	struct dccp_request_sock *dreq;
395 	const u32 dccp_header_size = sizeof(struct dccp_hdr) +
396 				     sizeof(struct dccp_hdr_ext) +
397 				     sizeof(struct dccp_hdr_response);
398 	struct sk_buff *skb;
399 
400 	/* sk is marked const to clearly express we dont hold socket lock.
401 	 * sock_wmalloc() will atomically change sk->sk_wmem_alloc,
402 	 * it is safe to promote sk to non const.
403 	 */
404 	skb = sock_wmalloc((struct sock *)sk, MAX_DCCP_HEADER, 1,
405 			   GFP_ATOMIC);
406 	if (!skb)
407 		return NULL;
408 
409 	skb_reserve(skb, MAX_DCCP_HEADER);
410 
411 	skb_dst_set(skb, dst_clone(dst));
412 
413 	dreq = dccp_rsk(req);
414 	if (inet_rsk(req)->acked)	/* increase GSS upon retransmission */
415 		dccp_inc_seqno(&dreq->dreq_gss);
416 	DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
417 	DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_gss;
418 
419 	/* Resolve feature dependencies resulting from choice of CCID */
420 	if (dccp_feat_server_ccid_dependencies(dreq))
421 		goto response_failed;
422 
423 	if (dccp_insert_options_rsk(dreq, skb))
424 		goto response_failed;
425 
426 	/* Build and checksum header */
427 	dh = dccp_zeroed_hdr(skb, dccp_header_size);
428 
429 	dh->dccph_sport	= htons(inet_rsk(req)->ir_num);
430 	dh->dccph_dport	= inet_rsk(req)->ir_rmt_port;
431 	dh->dccph_doff	= (dccp_header_size +
432 			   DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
433 	dh->dccph_type	= DCCP_PKT_RESPONSE;
434 	dh->dccph_x	= 1;
435 	dccp_hdr_set_seq(dh, dreq->dreq_gss);
436 	dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_gsr);
437 	dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
438 
439 	dccp_csum_outgoing(skb);
440 
441 	/* We use `acked' to remember that a Response was already sent. */
442 	inet_rsk(req)->acked = 1;
443 	DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
444 	return skb;
445 response_failed:
446 	kfree_skb(skb);
447 	return NULL;
448 }
449 
450 EXPORT_SYMBOL_GPL(dccp_make_response);
451 
452 /* answer offending packet in @rcv_skb with Reset from control socket @ctl */
dccp_ctl_make_reset(struct sock * sk,struct sk_buff * rcv_skb)453 struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
454 {
455 	struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
456 	struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
457 	const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
458 				       sizeof(struct dccp_hdr_ext) +
459 				       sizeof(struct dccp_hdr_reset);
460 	struct dccp_hdr_reset *dhr;
461 	struct sk_buff *skb;
462 
463 	skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
464 	if (skb == NULL)
465 		return NULL;
466 
467 	skb_reserve(skb, sk->sk_prot->max_header);
468 
469 	/* Swap the send and the receive. */
470 	dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
471 	dh->dccph_type	= DCCP_PKT_RESET;
472 	dh->dccph_sport	= rxdh->dccph_dport;
473 	dh->dccph_dport	= rxdh->dccph_sport;
474 	dh->dccph_doff	= dccp_hdr_reset_len / 4;
475 	dh->dccph_x	= 1;
476 
477 	dhr = dccp_hdr_reset(skb);
478 	dhr->dccph_reset_code = dcb->dccpd_reset_code;
479 
480 	switch (dcb->dccpd_reset_code) {
481 	case DCCP_RESET_CODE_PACKET_ERROR:
482 		dhr->dccph_reset_data[0] = rxdh->dccph_type;
483 		break;
484 	case DCCP_RESET_CODE_OPTION_ERROR:	/* fall through */
485 	case DCCP_RESET_CODE_MANDATORY_ERROR:
486 		memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
487 		break;
488 	}
489 	/*
490 	 * From RFC 4340, 8.3.1:
491 	 *   If P.ackno exists, set R.seqno := P.ackno + 1.
492 	 *   Else set R.seqno := 0.
493 	 */
494 	if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
495 		dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
496 	dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
497 
498 	dccp_csum_outgoing(skb);
499 	return skb;
500 }
501 
502 EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
503 
504 /* send Reset on established socket, to close or abort the connection */
dccp_send_reset(struct sock * sk,enum dccp_reset_codes code)505 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
506 {
507 	struct sk_buff *skb;
508 	/*
509 	 * FIXME: what if rebuild_header fails?
510 	 * Should we be doing a rebuild_header here?
511 	 */
512 	int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
513 
514 	if (err != 0)
515 		return err;
516 
517 	skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
518 	if (skb == NULL)
519 		return -ENOBUFS;
520 
521 	/* Reserve space for headers and prepare control bits. */
522 	skb_reserve(skb, sk->sk_prot->max_header);
523 	DCCP_SKB_CB(skb)->dccpd_type	   = DCCP_PKT_RESET;
524 	DCCP_SKB_CB(skb)->dccpd_reset_code = code;
525 
526 	return dccp_transmit_skb(sk, skb);
527 }
528 
529 /*
530  * Do all connect socket setups that can be done AF independent.
531  */
dccp_connect(struct sock * sk)532 int dccp_connect(struct sock *sk)
533 {
534 	struct sk_buff *skb;
535 	struct dccp_sock *dp = dccp_sk(sk);
536 	struct dst_entry *dst = __sk_dst_get(sk);
537 	struct inet_connection_sock *icsk = inet_csk(sk);
538 
539 	sk->sk_err = 0;
540 	sock_reset_flag(sk, SOCK_DONE);
541 
542 	dccp_sync_mss(sk, dst_mtu(dst));
543 
544 	/* do not connect if feature negotiation setup fails */
545 	if (dccp_feat_finalise_settings(dccp_sk(sk)))
546 		return -EPROTO;
547 
548 	/* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
549 	dp->dccps_gar = dp->dccps_iss;
550 
551 	skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
552 	if (unlikely(skb == NULL))
553 		return -ENOBUFS;
554 
555 	/* Reserve space for headers. */
556 	skb_reserve(skb, sk->sk_prot->max_header);
557 
558 	DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
559 
560 	dccp_transmit_skb(sk, dccp_skb_entail(sk, skb));
561 	DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
562 
563 	/* Timer for repeating the REQUEST until an answer. */
564 	icsk->icsk_retransmits = 0;
565 	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
566 				  icsk->icsk_rto, DCCP_RTO_MAX);
567 	return 0;
568 }
569 
570 EXPORT_SYMBOL_GPL(dccp_connect);
571 
dccp_send_ack(struct sock * sk)572 void dccp_send_ack(struct sock *sk)
573 {
574 	/* If we have been reset, we may not send again. */
575 	if (sk->sk_state != DCCP_CLOSED) {
576 		struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
577 						GFP_ATOMIC);
578 
579 		if (skb == NULL) {
580 			inet_csk_schedule_ack(sk);
581 			inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
582 			inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
583 						  TCP_DELACK_MAX,
584 						  DCCP_RTO_MAX);
585 			return;
586 		}
587 
588 		/* Reserve space for headers */
589 		skb_reserve(skb, sk->sk_prot->max_header);
590 		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
591 		dccp_transmit_skb(sk, skb);
592 	}
593 }
594 
595 EXPORT_SYMBOL_GPL(dccp_send_ack);
596 
597 #if 0
598 /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
599 void dccp_send_delayed_ack(struct sock *sk)
600 {
601 	struct inet_connection_sock *icsk = inet_csk(sk);
602 	/*
603 	 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
604 	 * with using 2s, and active senders also piggyback the ACK into a
605 	 * DATAACK packet, so this is really for quiescent senders.
606 	 */
607 	unsigned long timeout = jiffies + 2 * HZ;
608 
609 	/* Use new timeout only if there wasn't a older one earlier. */
610 	if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
611 		/* If delack timer was blocked or is about to expire,
612 		 * send ACK now.
613 		 *
614 		 * FIXME: check the "about to expire" part
615 		 */
616 		if (icsk->icsk_ack.blocked) {
617 			dccp_send_ack(sk);
618 			return;
619 		}
620 
621 		if (!time_before(timeout, icsk->icsk_ack.timeout))
622 			timeout = icsk->icsk_ack.timeout;
623 	}
624 	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
625 	icsk->icsk_ack.timeout = timeout;
626 	sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
627 }
628 #endif
629 
dccp_send_sync(struct sock * sk,const u64 ackno,const enum dccp_pkt_type pkt_type)630 void dccp_send_sync(struct sock *sk, const u64 ackno,
631 		    const enum dccp_pkt_type pkt_type)
632 {
633 	/*
634 	 * We are not putting this on the write queue, so
635 	 * dccp_transmit_skb() will set the ownership to this
636 	 * sock.
637 	 */
638 	struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
639 
640 	if (skb == NULL) {
641 		/* FIXME: how to make sure the sync is sent? */
642 		DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
643 		return;
644 	}
645 
646 	/* Reserve space for headers and prepare control bits. */
647 	skb_reserve(skb, sk->sk_prot->max_header);
648 	DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
649 	DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
650 
651 	/*
652 	 * Clear the flag in case the Sync was scheduled for out-of-band data,
653 	 * such as carrying a long Ack Vector.
654 	 */
655 	dccp_sk(sk)->dccps_sync_scheduled = 0;
656 
657 	dccp_transmit_skb(sk, skb);
658 }
659 
660 EXPORT_SYMBOL_GPL(dccp_send_sync);
661 
662 /*
663  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
664  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
665  * any circumstances.
666  */
dccp_send_close(struct sock * sk,const int active)667 void dccp_send_close(struct sock *sk, const int active)
668 {
669 	struct dccp_sock *dp = dccp_sk(sk);
670 	struct sk_buff *skb;
671 	const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
672 
673 	skb = alloc_skb(sk->sk_prot->max_header, prio);
674 	if (skb == NULL)
675 		return;
676 
677 	/* Reserve space for headers and prepare control bits. */
678 	skb_reserve(skb, sk->sk_prot->max_header);
679 	if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
680 		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
681 	else
682 		DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
683 
684 	if (active) {
685 		skb = dccp_skb_entail(sk, skb);
686 		/*
687 		 * Retransmission timer for active-close: RFC 4340, 8.3 requires
688 		 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
689 		 * state can be left. The initial timeout is 2 RTTs.
690 		 * Since RTT measurement is done by the CCIDs, there is no easy
691 		 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
692 		 * is too low (200ms); we use a high value to avoid unnecessary
693 		 * retransmissions when the link RTT is > 0.2 seconds.
694 		 * FIXME: Let main module sample RTTs and use that instead.
695 		 */
696 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
697 					  DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
698 	}
699 	dccp_transmit_skb(sk, skb);
700 }
701