4 * An implementation of the DCCP protocol
5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7 * This program is free software; you can redistribute it and/or
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.
13 #include <linux/dccp.h>
14 #include <linux/kernel.h>
15 #include <linux/skbuff.h>
17 #include <net/inet_sock.h>
24 static inline void dccp_event_ack_sent(struct sock *sk)
26 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
29 static void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
31 skb_set_owner_w(skb, sk);
32 WARN_ON(sk->sk_send_head);
33 sk->sk_send_head = skb;
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
42 static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
44 if (likely(skb != NULL)) {
45 const 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);
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);
55 u64 ackno = dp->dccps_gsr;
57 dccp_inc_seqno(&dp->dccps_gss);
59 switch (dcb->dccpd_type) {
63 case DCCP_PKT_DATAACK:
66 case DCCP_PKT_REQUEST:
71 case DCCP_PKT_SYNCACK:
72 ackno = dcb->dccpd_seq;
76 * Only data packets should come through with skb->sk
80 skb_set_owner_w(skb, sk);
84 dcb->dccpd_seq = dp->dccps_gss;
86 if (dccp_insert_options(sk, skb)) {
92 /* Build DCCP header and checksum it. */
93 dh = dccp_zeroed_hdr(skb, dccp_header_size);
94 dh->dccph_type = dcb->dccpd_type;
95 dh->dccph_sport = inet->sport;
96 dh->dccph_dport = inet->dport;
97 dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4;
98 dh->dccph_ccval = dcb->dccpd_ccval;
99 /* XXX For now we're using only 48 bits sequence numbers */
102 dp->dccps_awh = dp->dccps_gss;
103 dccp_hdr_set_seq(dh, dp->dccps_gss);
105 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
107 switch (dcb->dccpd_type) {
108 case DCCP_PKT_REQUEST:
109 dccp_hdr_request(skb)->dccph_req_service =
113 dccp_hdr_reset(skb)->dccph_reset_code =
114 dcb->dccpd_reset_code;
118 icsk->icsk_af_ops->send_check(sk, skb->len, skb);
121 dccp_event_ack_sent(sk);
123 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
125 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
126 err = icsk->icsk_af_ops->queue_xmit(skb, sk, 0);
130 /* NET_XMIT_CN is special. It does not guarantee,
131 * that this packet is lost. It tells that device
132 * is about to start to drop packets or already
133 * drops some packets of the same priority and
134 * invokes us to send less aggressively.
136 return err == NET_XMIT_CN ? 0 : err;
141 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
143 struct inet_connection_sock *icsk = inet_csk(sk);
144 struct dccp_sock *dp = dccp_sk(sk);
145 int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
146 sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
148 /* Now subtract optional transport overhead */
149 mss_now -= icsk->icsk_ext_hdr_len;
152 * FIXME: this should come from the CCID infrastructure, where, say,
153 * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
154 * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
155 * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
156 * make it a multiple of 4
159 mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
161 /* And store cached results */
162 icsk->icsk_pmtu_cookie = pmtu;
163 dp->dccps_mss_cache = mss_now;
168 EXPORT_SYMBOL_GPL(dccp_sync_mss);
170 void dccp_write_space(struct sock *sk)
172 read_lock(&sk->sk_callback_lock);
174 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
175 wake_up_interruptible(sk->sk_sleep);
176 /* Should agree with poll, otherwise some programs break */
177 if (sock_writeable(sk))
178 sk_wake_async(sk, 2, POLL_OUT);
180 read_unlock(&sk->sk_callback_lock);
184 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
185 * @sk: socket to wait for
186 * @timeo: for how long
188 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
191 struct dccp_sock *dp = dccp_sk(sk);
197 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
203 if (signal_pending(current))
206 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
210 delay = msecs_to_jiffies(rc);
211 if (delay > *timeo || delay < 0)
214 sk->sk_write_pending++;
216 *timeo -= schedule_timeout(delay);
218 sk->sk_write_pending--;
221 finish_wait(sk->sk_sleep, &wait);
231 rc = sock_intr_errno(*timeo);
235 static void dccp_write_xmit_timer(unsigned long data) {
236 struct sock *sk = (struct sock *)data;
237 struct dccp_sock *dp = dccp_sk(sk);
240 if (sock_owned_by_user(sk))
241 sk_reset_timer(sk, &dp->dccps_xmit_timer, jiffies+1);
243 dccp_write_xmit(sk, 0);
248 void dccp_write_xmit(struct sock *sk, int block)
250 struct dccp_sock *dp = dccp_sk(sk);
252 long timeo = 30000; /* If a packet is taking longer than 2 secs
253 we have other issues */
255 while ((skb = skb_peek(&sk->sk_write_queue))) {
256 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
261 sk_reset_timer(sk, &dp->dccps_xmit_timer,
262 msecs_to_jiffies(err)+jiffies);
265 err = dccp_wait_for_ccid(sk, skb, &timeo);
267 printk(KERN_CRIT "%s:err at dccp_wait_for_ccid"
268 " %d\n", __FUNCTION__, err);
273 skb_dequeue(&sk->sk_write_queue);
275 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
276 const int len = skb->len;
278 if (sk->sk_state == DCCP_PARTOPEN) {
279 /* See 8.1.5. Handshake Completion */
280 inet_csk_schedule_ack(sk);
281 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
282 inet_csk(sk)->icsk_rto,
284 dcb->dccpd_type = DCCP_PKT_DATAACK;
285 } else if (dccp_ack_pending(sk))
286 dcb->dccpd_type = DCCP_PKT_DATAACK;
288 dcb->dccpd_type = DCCP_PKT_DATA;
290 err = dccp_transmit_skb(sk, skb);
291 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
293 printk(KERN_CRIT "%s:err from "
294 "ccid_hc_tx_packet_sent %d\n",
303 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
305 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
306 return -EHOSTUNREACH; /* Routing failure or similar. */
308 return dccp_transmit_skb(sk, (skb_cloned(skb) ?
309 pskb_copy(skb, GFP_ATOMIC):
310 skb_clone(skb, GFP_ATOMIC)));
313 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
314 struct request_sock *req)
317 struct dccp_request_sock *dreq;
318 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
319 sizeof(struct dccp_hdr_ext) +
320 sizeof(struct dccp_hdr_response);
321 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
326 /* Reserve space for headers. */
327 skb_reserve(skb, sk->sk_prot->max_header);
329 skb->dst = dst_clone(dst);
332 dreq = dccp_rsk(req);
333 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
334 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
336 if (dccp_insert_options(sk, skb)) {
341 dh = dccp_zeroed_hdr(skb, dccp_header_size);
343 dh->dccph_sport = inet_sk(sk)->sport;
344 dh->dccph_dport = inet_rsk(req)->rmt_port;
345 dh->dccph_doff = (dccp_header_size +
346 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
347 dh->dccph_type = DCCP_PKT_RESPONSE;
349 dccp_hdr_set_seq(dh, dreq->dreq_iss);
350 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
351 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
353 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
357 EXPORT_SYMBOL_GPL(dccp_make_response);
359 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
360 const enum dccp_reset_codes code)
364 struct dccp_sock *dp = dccp_sk(sk);
365 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
366 sizeof(struct dccp_hdr_ext) +
367 sizeof(struct dccp_hdr_reset);
368 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
373 /* Reserve space for headers. */
374 skb_reserve(skb, sk->sk_prot->max_header);
376 skb->dst = dst_clone(dst);
379 dccp_inc_seqno(&dp->dccps_gss);
381 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
382 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
383 DCCP_SKB_CB(skb)->dccpd_seq = dp->dccps_gss;
385 if (dccp_insert_options(sk, skb)) {
390 dh = dccp_zeroed_hdr(skb, dccp_header_size);
392 dh->dccph_sport = inet_sk(sk)->sport;
393 dh->dccph_dport = inet_sk(sk)->dport;
394 dh->dccph_doff = (dccp_header_size +
395 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
396 dh->dccph_type = DCCP_PKT_RESET;
398 dccp_hdr_set_seq(dh, dp->dccps_gss);
399 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
401 dccp_hdr_reset(skb)->dccph_reset_code = code;
402 inet_csk(sk)->icsk_af_ops->send_check(sk, skb->len, skb);
404 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
408 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
411 * FIXME: what if rebuild_header fails?
412 * Should we be doing a rebuild_header here?
414 int err = inet_sk_rebuild_header(sk);
417 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
420 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
421 err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, sk, 0);
422 if (err == NET_XMIT_CN)
431 * Do all connect socket setups that can be done AF independent.
433 static inline void dccp_connect_init(struct sock *sk)
435 struct dccp_sock *dp = dccp_sk(sk);
436 struct dst_entry *dst = __sk_dst_get(sk);
437 struct inet_connection_sock *icsk = inet_csk(sk);
440 sock_reset_flag(sk, SOCK_DONE);
442 dccp_sync_mss(sk, dst_mtu(dst));
444 dccp_update_gss(sk, dp->dccps_iss);
446 * SWL and AWL are initially adjusted so that they are not less than
447 * the initial Sequence Numbers received and sent, respectively:
448 * SWL := max(GSR + 1 - floor(W/4), ISR),
449 * AWL := max(GSS - W' + 1, ISS).
450 * These adjustments MUST be applied only at the beginning of the
453 dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
455 icsk->icsk_retransmits = 0;
456 init_timer(&dp->dccps_xmit_timer);
457 dp->dccps_xmit_timer.data = (unsigned long)sk;
458 dp->dccps_xmit_timer.function = dccp_write_xmit_timer;
461 int dccp_connect(struct sock *sk)
464 struct inet_connection_sock *icsk = inet_csk(sk);
466 dccp_connect_init(sk);
468 skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
469 if (unlikely(skb == NULL))
472 /* Reserve space for headers. */
473 skb_reserve(skb, sk->sk_prot->max_header);
475 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
478 dccp_skb_entail(sk, skb);
479 dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
480 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
482 /* Timer for repeating the REQUEST until an answer. */
483 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
484 icsk->icsk_rto, DCCP_RTO_MAX);
488 EXPORT_SYMBOL_GPL(dccp_connect);
490 void dccp_send_ack(struct sock *sk)
492 /* If we have been reset, we may not send again. */
493 if (sk->sk_state != DCCP_CLOSED) {
494 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
498 inet_csk_schedule_ack(sk);
499 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
500 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
506 /* Reserve space for headers */
507 skb_reserve(skb, sk->sk_prot->max_header);
509 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
510 dccp_transmit_skb(sk, skb);
514 EXPORT_SYMBOL_GPL(dccp_send_ack);
516 void dccp_send_delayed_ack(struct sock *sk)
518 struct inet_connection_sock *icsk = inet_csk(sk);
520 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
521 * with using 2s, and active senders also piggyback the ACK into a
522 * DATAACK packet, so this is really for quiescent senders.
524 unsigned long timeout = jiffies + 2 * HZ;
526 /* Use new timeout only if there wasn't a older one earlier. */
527 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
528 /* If delack timer was blocked or is about to expire,
531 * FIXME: check the "about to expire" part
533 if (icsk->icsk_ack.blocked) {
538 if (!time_before(timeout, icsk->icsk_ack.timeout))
539 timeout = icsk->icsk_ack.timeout;
541 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
542 icsk->icsk_ack.timeout = timeout;
543 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
546 void dccp_send_sync(struct sock *sk, const u64 seq,
547 const enum dccp_pkt_type pkt_type)
550 * We are not putting this on the write queue, so
551 * dccp_transmit_skb() will set the ownership to this
554 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
557 /* FIXME: how to make sure the sync is sent? */
560 /* Reserve space for headers and prepare control bits. */
561 skb_reserve(skb, sk->sk_prot->max_header);
563 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
564 DCCP_SKB_CB(skb)->dccpd_seq = seq;
566 dccp_transmit_skb(sk, skb);
569 EXPORT_SYMBOL_GPL(dccp_send_sync);
572 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
573 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
576 void dccp_send_close(struct sock *sk, const int active)
578 struct dccp_sock *dp = dccp_sk(sk);
580 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
582 skb = alloc_skb(sk->sk_prot->max_header, prio);
586 /* Reserve space for headers and prepare control bits. */
587 skb_reserve(skb, sk->sk_prot->max_header);
589 DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
590 DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
593 dccp_write_xmit(sk, 1);
594 dccp_skb_entail(sk, skb);
595 dccp_transmit_skb(sk, skb_clone(skb, prio));
596 /* FIXME do we need a retransmit timer here? */
598 dccp_transmit_skb(sk, skb);