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 dh->dccph_cscov = dp->dccps_pcslen;
100 /* XXX For now we're using only 48 bits sequence numbers */
103 dp->dccps_awh = dp->dccps_gss;
104 dccp_hdr_set_seq(dh, dp->dccps_gss);
106 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
108 switch (dcb->dccpd_type) {
109 case DCCP_PKT_REQUEST:
110 dccp_hdr_request(skb)->dccph_req_service =
114 dccp_hdr_reset(skb)->dccph_reset_code =
115 dcb->dccpd_reset_code;
119 icsk->icsk_af_ops->send_check(sk, 0, skb);
122 dccp_event_ack_sent(sk);
124 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
126 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
127 err = icsk->icsk_af_ops->queue_xmit(skb, sk, 0);
128 return net_xmit_eval(err);
133 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
135 struct inet_connection_sock *icsk = inet_csk(sk);
136 struct dccp_sock *dp = dccp_sk(sk);
137 int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
138 sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
140 /* Now subtract optional transport overhead */
141 mss_now -= icsk->icsk_ext_hdr_len;
144 * FIXME: this should come from the CCID infrastructure, where, say,
145 * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
146 * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
147 * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
148 * make it a multiple of 4
151 mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
153 /* And store cached results */
154 icsk->icsk_pmtu_cookie = pmtu;
155 dp->dccps_mss_cache = mss_now;
160 EXPORT_SYMBOL_GPL(dccp_sync_mss);
162 void dccp_write_space(struct sock *sk)
164 read_lock(&sk->sk_callback_lock);
166 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
167 wake_up_interruptible(sk->sk_sleep);
168 /* Should agree with poll, otherwise some programs break */
169 if (sock_writeable(sk))
170 sk_wake_async(sk, 2, POLL_OUT);
172 read_unlock(&sk->sk_callback_lock);
176 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
177 * @sk: socket to wait for
178 * @timeo: for how long
180 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
183 struct dccp_sock *dp = dccp_sk(sk);
189 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
195 if (signal_pending(current))
198 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
202 delay = msecs_to_jiffies(rc);
203 if (delay > *timeo || delay < 0)
206 sk->sk_write_pending++;
208 *timeo -= schedule_timeout(delay);
210 sk->sk_write_pending--;
213 finish_wait(sk->sk_sleep, &wait);
223 rc = sock_intr_errno(*timeo);
227 static void dccp_write_xmit_timer(unsigned long data) {
228 struct sock *sk = (struct sock *)data;
229 struct dccp_sock *dp = dccp_sk(sk);
232 if (sock_owned_by_user(sk))
233 sk_reset_timer(sk, &dp->dccps_xmit_timer, jiffies+1);
235 dccp_write_xmit(sk, 0);
240 void dccp_write_xmit(struct sock *sk, int block)
242 struct dccp_sock *dp = dccp_sk(sk);
244 long timeo = DCCP_XMIT_TIMEO; /* If a packet is taking longer than
245 this we have other issues */
247 while ((skb = skb_peek(&sk->sk_write_queue))) {
248 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
253 sk_reset_timer(sk, &dp->dccps_xmit_timer,
254 msecs_to_jiffies(err)+jiffies);
257 err = dccp_wait_for_ccid(sk, skb, &timeo);
258 timeo = DCCP_XMIT_TIMEO;
261 DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
264 skb_dequeue(&sk->sk_write_queue);
266 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
267 const int len = skb->len;
269 if (sk->sk_state == DCCP_PARTOPEN) {
270 /* See 8.1.5. Handshake Completion */
271 inet_csk_schedule_ack(sk);
272 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
273 inet_csk(sk)->icsk_rto,
275 dcb->dccpd_type = DCCP_PKT_DATAACK;
276 } else if (dccp_ack_pending(sk))
277 dcb->dccpd_type = DCCP_PKT_DATAACK;
279 dcb->dccpd_type = DCCP_PKT_DATA;
281 err = dccp_transmit_skb(sk, skb);
282 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
284 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
291 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
293 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
294 return -EHOSTUNREACH; /* Routing failure or similar. */
296 return dccp_transmit_skb(sk, (skb_cloned(skb) ?
297 pskb_copy(skb, GFP_ATOMIC):
298 skb_clone(skb, GFP_ATOMIC)));
301 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
302 struct request_sock *req)
305 struct dccp_request_sock *dreq;
306 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
307 sizeof(struct dccp_hdr_ext) +
308 sizeof(struct dccp_hdr_response);
309 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
314 /* Reserve space for headers. */
315 skb_reserve(skb, sk->sk_prot->max_header);
317 skb->dst = dst_clone(dst);
319 dreq = dccp_rsk(req);
320 if (inet_rsk(req)->acked) /* increase ISS upon retransmission */
321 dccp_inc_seqno(&dreq->dreq_iss);
322 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
323 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
325 if (dccp_insert_options(sk, skb)) {
330 /* Build and checksum header */
331 dh = dccp_zeroed_hdr(skb, dccp_header_size);
333 dh->dccph_sport = inet_sk(sk)->sport;
334 dh->dccph_dport = inet_rsk(req)->rmt_port;
335 dh->dccph_doff = (dccp_header_size +
336 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
337 dh->dccph_type = DCCP_PKT_RESPONSE;
339 dccp_hdr_set_seq(dh, dreq->dreq_iss);
340 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
341 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
343 dccp_csum_outgoing(skb);
345 /* We use `acked' to remember that a Response was already sent. */
346 inet_rsk(req)->acked = 1;
347 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
351 EXPORT_SYMBOL_GPL(dccp_make_response);
353 static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
354 const enum dccp_reset_codes code)
358 struct dccp_sock *dp = dccp_sk(sk);
359 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
360 sizeof(struct dccp_hdr_ext) +
361 sizeof(struct dccp_hdr_reset);
362 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
367 /* Reserve space for headers. */
368 skb_reserve(skb, sk->sk_prot->max_header);
370 skb->dst = dst_clone(dst);
372 dccp_inc_seqno(&dp->dccps_gss);
374 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
375 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
376 DCCP_SKB_CB(skb)->dccpd_seq = dp->dccps_gss;
378 if (dccp_insert_options(sk, skb)) {
383 dh = dccp_zeroed_hdr(skb, dccp_header_size);
385 dh->dccph_sport = inet_sk(sk)->sport;
386 dh->dccph_dport = inet_sk(sk)->dport;
387 dh->dccph_doff = (dccp_header_size +
388 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
389 dh->dccph_type = DCCP_PKT_RESET;
391 dccp_hdr_set_seq(dh, dp->dccps_gss);
392 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
394 dccp_hdr_reset(skb)->dccph_reset_code = code;
395 inet_csk(sk)->icsk_af_ops->send_check(sk, 0, skb);
397 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
401 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
404 * FIXME: what if rebuild_header fails?
405 * Should we be doing a rebuild_header here?
407 int err = inet_sk_rebuild_header(sk);
410 struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
413 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
414 err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, sk, 0);
415 return net_xmit_eval(err);
423 * Do all connect socket setups that can be done AF independent.
425 static inline void dccp_connect_init(struct sock *sk)
427 struct dccp_sock *dp = dccp_sk(sk);
428 struct dst_entry *dst = __sk_dst_get(sk);
429 struct inet_connection_sock *icsk = inet_csk(sk);
432 sock_reset_flag(sk, SOCK_DONE);
434 dccp_sync_mss(sk, dst_mtu(dst));
437 * SWL and AWL are initially adjusted so that they are not less than
438 * the initial Sequence Numbers received and sent, respectively:
439 * SWL := max(GSR + 1 - floor(W/4), ISR),
440 * AWL := max(GSS - W' + 1, ISS).
441 * These adjustments MUST be applied only at the beginning of the
444 dccp_update_gss(sk, dp->dccps_iss);
445 dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
447 /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
448 * initialized to S.ISS (sec. 8.5) */
449 dp->dccps_gar = dp->dccps_iss;
451 icsk->icsk_retransmits = 0;
452 init_timer(&dp->dccps_xmit_timer);
453 dp->dccps_xmit_timer.data = (unsigned long)sk;
454 dp->dccps_xmit_timer.function = dccp_write_xmit_timer;
457 int dccp_connect(struct sock *sk)
460 struct inet_connection_sock *icsk = inet_csk(sk);
462 dccp_connect_init(sk);
464 skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
465 if (unlikely(skb == NULL))
468 /* Reserve space for headers. */
469 skb_reserve(skb, sk->sk_prot->max_header);
471 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
473 dccp_skb_entail(sk, skb);
474 dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
475 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
477 /* Timer for repeating the REQUEST until an answer. */
478 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
479 icsk->icsk_rto, DCCP_RTO_MAX);
483 EXPORT_SYMBOL_GPL(dccp_connect);
485 void dccp_send_ack(struct sock *sk)
487 /* If we have been reset, we may not send again. */
488 if (sk->sk_state != DCCP_CLOSED) {
489 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
493 inet_csk_schedule_ack(sk);
494 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
495 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
501 /* Reserve space for headers */
502 skb_reserve(skb, sk->sk_prot->max_header);
503 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
504 dccp_transmit_skb(sk, skb);
508 EXPORT_SYMBOL_GPL(dccp_send_ack);
510 void dccp_send_delayed_ack(struct sock *sk)
512 struct inet_connection_sock *icsk = inet_csk(sk);
514 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
515 * with using 2s, and active senders also piggyback the ACK into a
516 * DATAACK packet, so this is really for quiescent senders.
518 unsigned long timeout = jiffies + 2 * HZ;
520 /* Use new timeout only if there wasn't a older one earlier. */
521 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
522 /* If delack timer was blocked or is about to expire,
525 * FIXME: check the "about to expire" part
527 if (icsk->icsk_ack.blocked) {
532 if (!time_before(timeout, icsk->icsk_ack.timeout))
533 timeout = icsk->icsk_ack.timeout;
535 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
536 icsk->icsk_ack.timeout = timeout;
537 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
540 void dccp_send_sync(struct sock *sk, const u64 seq,
541 const enum dccp_pkt_type pkt_type)
544 * We are not putting this on the write queue, so
545 * dccp_transmit_skb() will set the ownership to this
548 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
551 /* FIXME: how to make sure the sync is sent? */
554 /* Reserve space for headers and prepare control bits. */
555 skb_reserve(skb, sk->sk_prot->max_header);
556 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
557 DCCP_SKB_CB(skb)->dccpd_seq = seq;
559 dccp_transmit_skb(sk, skb);
562 EXPORT_SYMBOL_GPL(dccp_send_sync);
565 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
566 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
569 void dccp_send_close(struct sock *sk, const int active)
571 struct dccp_sock *dp = dccp_sk(sk);
573 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
575 skb = alloc_skb(sk->sk_prot->max_header, prio);
579 /* Reserve space for headers and prepare control bits. */
580 skb_reserve(skb, sk->sk_prot->max_header);
581 DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
582 DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
585 dccp_write_xmit(sk, 1);
586 dccp_skb_entail(sk, skb);
587 dccp_transmit_skb(sk, skb_clone(skb, prio));
588 /* FIXME do we need a retransmit timer here? */
590 dccp_transmit_skb(sk, skb);