Loading...
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}
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
22static 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 */
28static 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 */
42static 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 fallthrough;
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 fallthrough;
76
77 case DCCP_PKT_SYNC:
78 case DCCP_PKT_SYNCACK:
79 ackno = dcb->dccpd_ack_seq;
80 fallthrough;
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 * @dp: socket to find packet size limits of
147 *
148 * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
149 * since the RX CCID is restricted to feedback packets (Acks), which are small
150 * in comparison with the data traffic. A value of 0 means "no current CCMPS".
151 */
152static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
153{
154 const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
155
156 if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
157 return 0;
158 return tx_ccid->ccid_ops->ccid_ccmps;
159}
160
161unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
162{
163 struct inet_connection_sock *icsk = inet_csk(sk);
164 struct dccp_sock *dp = dccp_sk(sk);
165 u32 ccmps = dccp_determine_ccmps(dp);
166 u32 cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
167
168 /* Account for header lengths and IPv4/v6 option overhead */
169 cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
170 sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
171
172 /*
173 * Leave enough headroom for common DCCP header options.
174 * This only considers options which may appear on DCCP-Data packets, as
175 * per table 3 in RFC 4340, 5.8. When running out of space for other
176 * options (eg. Ack Vector which can take up to 255 bytes), it is better
177 * to schedule a separate Ack. Thus we leave headroom for the following:
178 * - 1 byte for Slow Receiver (11.6)
179 * - 6 bytes for Timestamp (13.1)
180 * - 10 bytes for Timestamp Echo (13.3)
181 * - 8 bytes for NDP count (7.7, when activated)
182 * - 6 bytes for Data Checksum (9.3)
183 * - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
184 */
185 cur_mps -= roundup(1 + 6 + 10 + dp->dccps_send_ndp_count * 8 + 6 +
186 (dp->dccps_hc_rx_ackvec ? DCCPAV_MIN_OPTLEN : 0), 4);
187
188 /* And store cached results */
189 icsk->icsk_pmtu_cookie = pmtu;
190 WRITE_ONCE(dp->dccps_mss_cache, cur_mps);
191
192 return cur_mps;
193}
194
195EXPORT_SYMBOL_GPL(dccp_sync_mss);
196
197void dccp_write_space(struct sock *sk)
198{
199 struct socket_wq *wq;
200
201 rcu_read_lock();
202 wq = rcu_dereference(sk->sk_wq);
203 if (skwq_has_sleeper(wq))
204 wake_up_interruptible(&wq->wait);
205 /* Should agree with poll, otherwise some programs break */
206 if (sock_writeable(sk))
207 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
208
209 rcu_read_unlock();
210}
211
212/**
213 * dccp_wait_for_ccid - Await CCID send permission
214 * @sk: socket to wait for
215 * @delay: timeout in jiffies
216 *
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 * @sk: socket to send data packet on
242 *
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 * @sk: socket to be drained
304 * @time_budget: time allowed to drain the queue
305 *
306 * Since dccp_sendmsg queues packets without waiting for them to be sent, it may
307 * happen that the TX queue is not empty at the end of a connection. We give the
308 * HC-sender CCID a grace period of up to @time_budget jiffies. If this function
309 * returns with a non-empty write queue, it will be purged later.
310 */
311void dccp_flush_write_queue(struct sock *sk, long *time_budget)
312{
313 struct dccp_sock *dp = dccp_sk(sk);
314 struct sk_buff *skb;
315 long delay, rc;
316
317 while (*time_budget > 0 && (skb = skb_peek(&sk->sk_write_queue))) {
318 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
319
320 switch (ccid_packet_dequeue_eval(rc)) {
321 case CCID_PACKET_WILL_DEQUEUE_LATER:
322 /*
323 * If the CCID determines when to send, the next sending
324 * time is unknown or the CCID may not even send again
325 * (e.g. remote host crashes or lost Ack packets).
326 */
327 DCCP_WARN("CCID did not manage to send all packets\n");
328 return;
329 case CCID_PACKET_DELAY:
330 delay = msecs_to_jiffies(rc);
331 if (delay > *time_budget)
332 return;
333 rc = dccp_wait_for_ccid(sk, delay);
334 if (rc < 0)
335 return;
336 *time_budget -= (delay - rc);
337 /* check again if we can send now */
338 break;
339 case CCID_PACKET_SEND_AT_ONCE:
340 dccp_xmit_packet(sk);
341 break;
342 case CCID_PACKET_ERR:
343 skb_dequeue(&sk->sk_write_queue);
344 kfree_skb(skb);
345 dccp_pr_debug("packet discarded due to err=%ld\n", rc);
346 }
347 }
348}
349
350void dccp_write_xmit(struct sock *sk)
351{
352 struct dccp_sock *dp = dccp_sk(sk);
353 struct sk_buff *skb;
354
355 while ((skb = dccp_qpolicy_top(sk))) {
356 int rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
357
358 switch (ccid_packet_dequeue_eval(rc)) {
359 case CCID_PACKET_WILL_DEQUEUE_LATER:
360 return;
361 case CCID_PACKET_DELAY:
362 sk_reset_timer(sk, &dp->dccps_xmit_timer,
363 jiffies + msecs_to_jiffies(rc));
364 return;
365 case CCID_PACKET_SEND_AT_ONCE:
366 dccp_xmit_packet(sk);
367 break;
368 case CCID_PACKET_ERR:
369 dccp_qpolicy_drop(sk, skb);
370 dccp_pr_debug("packet discarded due to err=%d\n", rc);
371 }
372 }
373}
374
375/**
376 * dccp_retransmit_skb - Retransmit Request, Close, or CloseReq packets
377 * @sk: socket to perform retransmit on
378 *
379 * There are only four retransmittable packet types in DCCP:
380 * - Request in client-REQUEST state (sec. 8.1.1),
381 * - CloseReq in server-CLOSEREQ state (sec. 8.3),
382 * - Close in node-CLOSING state (sec. 8.3),
383 * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
384 * This function expects sk->sk_send_head to contain the original skb.
385 */
386int dccp_retransmit_skb(struct sock *sk)
387{
388 WARN_ON(sk->sk_send_head == NULL);
389
390 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
391 return -EHOSTUNREACH; /* Routing failure or similar. */
392
393 /* this count is used to distinguish original and retransmitted skb */
394 inet_csk(sk)->icsk_retransmits++;
395
396 return dccp_transmit_skb(sk, skb_clone(sk->sk_send_head, GFP_ATOMIC));
397}
398
399struct sk_buff *dccp_make_response(const struct sock *sk, struct dst_entry *dst,
400 struct request_sock *req)
401{
402 struct dccp_hdr *dh;
403 struct dccp_request_sock *dreq;
404 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
405 sizeof(struct dccp_hdr_ext) +
406 sizeof(struct dccp_hdr_response);
407 struct sk_buff *skb;
408
409 /* sk is marked const to clearly express we dont hold socket lock.
410 * sock_wmalloc() will atomically change sk->sk_wmem_alloc,
411 * it is safe to promote sk to non const.
412 */
413 skb = sock_wmalloc((struct sock *)sk, MAX_DCCP_HEADER, 1,
414 GFP_ATOMIC);
415 if (!skb)
416 return NULL;
417
418 skb_reserve(skb, MAX_DCCP_HEADER);
419
420 skb_dst_set(skb, dst_clone(dst));
421
422 dreq = dccp_rsk(req);
423 if (inet_rsk(req)->acked) /* increase GSS upon retransmission */
424 dccp_inc_seqno(&dreq->dreq_gss);
425 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
426 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_gss;
427
428 /* Resolve feature dependencies resulting from choice of CCID */
429 if (dccp_feat_server_ccid_dependencies(dreq))
430 goto response_failed;
431
432 if (dccp_insert_options_rsk(dreq, skb))
433 goto response_failed;
434
435 /* Build and checksum header */
436 dh = dccp_zeroed_hdr(skb, dccp_header_size);
437
438 dh->dccph_sport = htons(inet_rsk(req)->ir_num);
439 dh->dccph_dport = inet_rsk(req)->ir_rmt_port;
440 dh->dccph_doff = (dccp_header_size +
441 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
442 dh->dccph_type = DCCP_PKT_RESPONSE;
443 dh->dccph_x = 1;
444 dccp_hdr_set_seq(dh, dreq->dreq_gss);
445 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_gsr);
446 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
447
448 dccp_csum_outgoing(skb);
449
450 /* We use `acked' to remember that a Response was already sent. */
451 inet_rsk(req)->acked = 1;
452 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
453 return skb;
454response_failed:
455 kfree_skb(skb);
456 return NULL;
457}
458
459EXPORT_SYMBOL_GPL(dccp_make_response);
460
461/* answer offending packet in @rcv_skb with Reset from control socket @ctl */
462struct sk_buff *dccp_ctl_make_reset(struct sock *sk, struct sk_buff *rcv_skb)
463{
464 struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
465 struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
466 const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
467 sizeof(struct dccp_hdr_ext) +
468 sizeof(struct dccp_hdr_reset);
469 struct dccp_hdr_reset *dhr;
470 struct sk_buff *skb;
471
472 skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
473 if (skb == NULL)
474 return NULL;
475
476 skb_reserve(skb, sk->sk_prot->max_header);
477
478 /* Swap the send and the receive. */
479 dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
480 dh->dccph_type = DCCP_PKT_RESET;
481 dh->dccph_sport = rxdh->dccph_dport;
482 dh->dccph_dport = rxdh->dccph_sport;
483 dh->dccph_doff = dccp_hdr_reset_len / 4;
484 dh->dccph_x = 1;
485
486 dhr = dccp_hdr_reset(skb);
487 dhr->dccph_reset_code = dcb->dccpd_reset_code;
488
489 switch (dcb->dccpd_reset_code) {
490 case DCCP_RESET_CODE_PACKET_ERROR:
491 dhr->dccph_reset_data[0] = rxdh->dccph_type;
492 break;
493 case DCCP_RESET_CODE_OPTION_ERROR:
494 case DCCP_RESET_CODE_MANDATORY_ERROR:
495 memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
496 break;
497 }
498 /*
499 * From RFC 4340, 8.3.1:
500 * If P.ackno exists, set R.seqno := P.ackno + 1.
501 * Else set R.seqno := 0.
502 */
503 if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
504 dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
505 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
506
507 dccp_csum_outgoing(skb);
508 return skb;
509}
510
511EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
512
513/* send Reset on established socket, to close or abort the connection */
514int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
515{
516 struct sk_buff *skb;
517 /*
518 * FIXME: what if rebuild_header fails?
519 * Should we be doing a rebuild_header here?
520 */
521 int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
522
523 if (err != 0)
524 return err;
525
526 skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
527 if (skb == NULL)
528 return -ENOBUFS;
529
530 /* Reserve space for headers and prepare control bits. */
531 skb_reserve(skb, sk->sk_prot->max_header);
532 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
533 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
534
535 return dccp_transmit_skb(sk, skb);
536}
537
538/*
539 * Do all connect socket setups that can be done AF independent.
540 */
541int dccp_connect(struct sock *sk)
542{
543 struct sk_buff *skb;
544 struct dccp_sock *dp = dccp_sk(sk);
545 struct dst_entry *dst = __sk_dst_get(sk);
546 struct inet_connection_sock *icsk = inet_csk(sk);
547
548 sk->sk_err = 0;
549 sock_reset_flag(sk, SOCK_DONE);
550
551 dccp_sync_mss(sk, dst_mtu(dst));
552
553 /* do not connect if feature negotiation setup fails */
554 if (dccp_feat_finalise_settings(dccp_sk(sk)))
555 return -EPROTO;
556
557 /* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
558 dp->dccps_gar = dp->dccps_iss;
559
560 skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
561 if (unlikely(skb == NULL))
562 return -ENOBUFS;
563
564 /* Reserve space for headers. */
565 skb_reserve(skb, sk->sk_prot->max_header);
566
567 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
568
569 dccp_transmit_skb(sk, dccp_skb_entail(sk, skb));
570 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
571
572 /* Timer for repeating the REQUEST until an answer. */
573 icsk->icsk_retransmits = 0;
574 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
575 icsk->icsk_rto, DCCP_RTO_MAX);
576 return 0;
577}
578
579EXPORT_SYMBOL_GPL(dccp_connect);
580
581void dccp_send_ack(struct sock *sk)
582{
583 /* If we have been reset, we may not send again. */
584 if (sk->sk_state != DCCP_CLOSED) {
585 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
586 GFP_ATOMIC);
587
588 if (skb == NULL) {
589 inet_csk_schedule_ack(sk);
590 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
591 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
592 TCP_DELACK_MAX,
593 DCCP_RTO_MAX);
594 return;
595 }
596
597 /* Reserve space for headers */
598 skb_reserve(skb, sk->sk_prot->max_header);
599 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
600 dccp_transmit_skb(sk, skb);
601 }
602}
603
604EXPORT_SYMBOL_GPL(dccp_send_ack);
605
606#if 0
607/* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
608void dccp_send_delayed_ack(struct sock *sk)
609{
610 struct inet_connection_sock *icsk = inet_csk(sk);
611 /*
612 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
613 * with using 2s, and active senders also piggyback the ACK into a
614 * DATAACK packet, so this is really for quiescent senders.
615 */
616 unsigned long timeout = jiffies + 2 * HZ;
617
618 /* Use new timeout only if there wasn't a older one earlier. */
619 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
620 /* If delack timer was blocked or is about to expire,
621 * send ACK now.
622 *
623 * FIXME: check the "about to expire" part
624 */
625 if (icsk->icsk_ack.blocked) {
626 dccp_send_ack(sk);
627 return;
628 }
629
630 if (!time_before(timeout, icsk->icsk_ack.timeout))
631 timeout = icsk->icsk_ack.timeout;
632 }
633 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
634 icsk->icsk_ack.timeout = timeout;
635 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
636}
637#endif
638
639void dccp_send_sync(struct sock *sk, const u64 ackno,
640 const enum dccp_pkt_type pkt_type)
641{
642 /*
643 * We are not putting this on the write queue, so
644 * dccp_transmit_skb() will set the ownership to this
645 * sock.
646 */
647 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
648
649 if (skb == NULL) {
650 /* FIXME: how to make sure the sync is sent? */
651 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
652 return;
653 }
654
655 /* Reserve space for headers and prepare control bits. */
656 skb_reserve(skb, sk->sk_prot->max_header);
657 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
658 DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
659
660 /*
661 * Clear the flag in case the Sync was scheduled for out-of-band data,
662 * such as carrying a long Ack Vector.
663 */
664 dccp_sk(sk)->dccps_sync_scheduled = 0;
665
666 dccp_transmit_skb(sk, skb);
667}
668
669EXPORT_SYMBOL_GPL(dccp_send_sync);
670
671/*
672 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
673 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
674 * any circumstances.
675 */
676void dccp_send_close(struct sock *sk, const int active)
677{
678 struct dccp_sock *dp = dccp_sk(sk);
679 struct sk_buff *skb;
680 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
681
682 skb = alloc_skb(sk->sk_prot->max_header, prio);
683 if (skb == NULL)
684 return;
685
686 /* Reserve space for headers and prepare control bits. */
687 skb_reserve(skb, sk->sk_prot->max_header);
688 if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
689 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
690 else
691 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
692
693 if (active) {
694 skb = dccp_skb_entail(sk, skb);
695 /*
696 * Retransmission timer for active-close: RFC 4340, 8.3 requires
697 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
698 * state can be left. The initial timeout is 2 RTTs.
699 * Since RTT measurement is done by the CCIDs, there is no easy
700 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
701 * is too low (200ms); we use a high value to avoid unnecessary
702 * retransmissions when the link RTT is > 0.2 seconds.
703 * FIXME: Let main module sample RTTs and use that instead.
704 */
705 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
706 DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
707 }
708 dccp_transmit_skb(sk, skb);
709}