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:
67 case DCCP_PKT_REQUEST:
72 case DCCP_PKT_SYNCACK:
73 ackno = dcb->dccpd_ack_seq;
77 * Set owner/destructor: some skbs are allocated via
78 * alloc_skb (e.g. when retransmission may happen).
79 * Only Data, DataAck, and Reset packets should come
80 * through here with skb->sk set.
83 skb_set_owner_w(skb, sk);
87 dcb->dccpd_seq = dp->dccps_gss;
89 if (dccp_insert_options(sk, skb)) {
95 /* Build DCCP header and checksum it. */
96 dh = dccp_zeroed_hdr(skb, dccp_header_size);
97 dh->dccph_type = dcb->dccpd_type;
98 dh->dccph_sport = inet->sport;
99 dh->dccph_dport = inet->dport;
100 dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4;
101 dh->dccph_ccval = dcb->dccpd_ccval;
102 dh->dccph_cscov = dp->dccps_pcslen;
103 /* XXX For now we're using only 48 bits sequence numbers */
106 dp->dccps_awh = dp->dccps_gss;
107 dccp_hdr_set_seq(dh, dp->dccps_gss);
109 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
111 switch (dcb->dccpd_type) {
112 case DCCP_PKT_REQUEST:
113 dccp_hdr_request(skb)->dccph_req_service =
117 dccp_hdr_reset(skb)->dccph_reset_code =
118 dcb->dccpd_reset_code;
122 icsk->icsk_af_ops->send_check(sk, 0, skb);
125 dccp_event_ack_sent(sk);
127 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
129 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
130 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
131 return net_xmit_eval(err);
137 * dccp_determine_ccmps - Find out about CCID-specfic packet-size limits
138 * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
139 * since the RX CCID is restricted to feedback packets (Acks), which are small
140 * in comparison with the data traffic. A value of 0 means "no current CCMPS".
142 static u32 dccp_determine_ccmps(const struct dccp_sock *dp)
144 const struct ccid *tx_ccid = dp->dccps_hc_tx_ccid;
146 if (tx_ccid == NULL || tx_ccid->ccid_ops == NULL)
148 return tx_ccid->ccid_ops->ccid_ccmps;
151 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
153 struct inet_connection_sock *icsk = inet_csk(sk);
154 struct dccp_sock *dp = dccp_sk(sk);
155 u32 ccmps = dccp_determine_ccmps(dp);
156 int cur_mps = ccmps ? min(pmtu, ccmps) : pmtu;
158 /* Account for header lengths and IPv4/v6 option overhead */
159 cur_mps -= (icsk->icsk_af_ops->net_header_len + icsk->icsk_ext_hdr_len +
160 sizeof(struct dccp_hdr) + sizeof(struct dccp_hdr_ext));
163 * FIXME: this should come from the CCID infrastructure, where, say,
164 * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
165 * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
166 * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
167 * make it a multiple of 4
170 cur_mps -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
172 /* And store cached results */
173 icsk->icsk_pmtu_cookie = pmtu;
174 dp->dccps_mss_cache = cur_mps;
179 EXPORT_SYMBOL_GPL(dccp_sync_mss);
181 void dccp_write_space(struct sock *sk)
183 read_lock(&sk->sk_callback_lock);
185 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
186 wake_up_interruptible(sk->sk_sleep);
187 /* Should agree with poll, otherwise some programs break */
188 if (sock_writeable(sk))
189 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
191 read_unlock(&sk->sk_callback_lock);
195 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
196 * @sk: socket to wait for
197 * @skb: current skb to pass on for waiting
198 * @delay: sleep timeout in milliseconds (> 0)
199 * This function is called by default when the socket is closed, and
200 * when a non-zero linger time is set on the socket. For consistency
202 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb, int delay)
204 struct dccp_sock *dp = dccp_sk(sk);
206 unsigned long jiffdelay;
210 dccp_pr_debug("delayed send by %d msec\n", delay);
211 jiffdelay = msecs_to_jiffies(delay);
213 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
215 sk->sk_write_pending++;
217 schedule_timeout(jiffdelay);
219 sk->sk_write_pending--;
223 if (signal_pending(current))
226 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
227 } while ((delay = rc) > 0);
229 finish_wait(sk->sk_sleep, &wait);
240 void dccp_write_xmit(struct sock *sk, int block)
242 struct dccp_sock *dp = dccp_sk(sk);
245 while ((skb = skb_peek(&sk->sk_write_queue))) {
246 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
250 sk_reset_timer(sk, &dp->dccps_xmit_timer,
251 msecs_to_jiffies(err)+jiffies);
254 err = dccp_wait_for_ccid(sk, skb, err);
255 if (err && err != -EINTR)
256 DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
259 skb_dequeue(&sk->sk_write_queue);
261 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
262 const int len = skb->len;
264 if (sk->sk_state == DCCP_PARTOPEN) {
265 /* See 8.1.5. Handshake Completion */
266 inet_csk_schedule_ack(sk);
267 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
268 inet_csk(sk)->icsk_rto,
270 dcb->dccpd_type = DCCP_PKT_DATAACK;
271 } else if (dccp_ack_pending(sk))
272 dcb->dccpd_type = DCCP_PKT_DATAACK;
274 dcb->dccpd_type = DCCP_PKT_DATA;
276 err = dccp_transmit_skb(sk, skb);
277 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
279 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
282 dccp_pr_debug("packet discarded due to err=%d\n", err);
288 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
290 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
291 return -EHOSTUNREACH; /* Routing failure or similar. */
293 return dccp_transmit_skb(sk, (skb_cloned(skb) ?
294 pskb_copy(skb, GFP_ATOMIC):
295 skb_clone(skb, GFP_ATOMIC)));
298 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
299 struct request_sock *req)
302 struct dccp_request_sock *dreq;
303 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
304 sizeof(struct dccp_hdr_ext) +
305 sizeof(struct dccp_hdr_response);
306 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
311 /* Reserve space for headers. */
312 skb_reserve(skb, sk->sk_prot->max_header);
314 skb->dst = dst_clone(dst);
316 dreq = dccp_rsk(req);
317 if (inet_rsk(req)->acked) /* increase ISS upon retransmission */
318 dccp_inc_seqno(&dreq->dreq_iss);
319 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
320 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
322 if (dccp_insert_options_rsk(dreq, skb)) {
327 /* Build and checksum header */
328 dh = dccp_zeroed_hdr(skb, dccp_header_size);
330 dh->dccph_sport = inet_sk(sk)->sport;
331 dh->dccph_dport = inet_rsk(req)->rmt_port;
332 dh->dccph_doff = (dccp_header_size +
333 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
334 dh->dccph_type = DCCP_PKT_RESPONSE;
336 dccp_hdr_set_seq(dh, dreq->dreq_iss);
337 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
338 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
340 dccp_csum_outgoing(skb);
342 /* We use `acked' to remember that a Response was already sent. */
343 inet_rsk(req)->acked = 1;
344 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
348 EXPORT_SYMBOL_GPL(dccp_make_response);
350 /* answer offending packet in @rcv_skb with Reset from control socket @ctl */
351 struct sk_buff *dccp_ctl_make_reset(struct socket *ctl, struct sk_buff *rcv_skb)
353 struct dccp_hdr *rxdh = dccp_hdr(rcv_skb), *dh;
354 struct dccp_skb_cb *dcb = DCCP_SKB_CB(rcv_skb);
355 const u32 dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
356 sizeof(struct dccp_hdr_ext) +
357 sizeof(struct dccp_hdr_reset);
358 struct dccp_hdr_reset *dhr;
361 skb = alloc_skb(ctl->sk->sk_prot->max_header, GFP_ATOMIC);
365 skb_reserve(skb, ctl->sk->sk_prot->max_header);
367 /* Swap the send and the receive. */
368 dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
369 dh->dccph_type = DCCP_PKT_RESET;
370 dh->dccph_sport = rxdh->dccph_dport;
371 dh->dccph_dport = rxdh->dccph_sport;
372 dh->dccph_doff = dccp_hdr_reset_len / 4;
375 dhr = dccp_hdr_reset(skb);
376 dhr->dccph_reset_code = dcb->dccpd_reset_code;
378 switch (dcb->dccpd_reset_code) {
379 case DCCP_RESET_CODE_PACKET_ERROR:
380 dhr->dccph_reset_data[0] = rxdh->dccph_type;
382 case DCCP_RESET_CODE_OPTION_ERROR: /* fall through */
383 case DCCP_RESET_CODE_MANDATORY_ERROR:
384 memcpy(dhr->dccph_reset_data, dcb->dccpd_reset_data, 3);
388 * From RFC 4340, 8.3.1:
389 * If P.ackno exists, set R.seqno := P.ackno + 1.
390 * Else set R.seqno := 0.
392 if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
393 dccp_hdr_set_seq(dh, ADD48(dcb->dccpd_ack_seq, 1));
394 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dcb->dccpd_seq);
396 dccp_csum_outgoing(skb);
400 EXPORT_SYMBOL_GPL(dccp_ctl_make_reset);
402 /* send Reset on established socket, to close or abort the connection */
403 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
407 * FIXME: what if rebuild_header fails?
408 * Should we be doing a rebuild_header here?
410 int err = inet_csk(sk)->icsk_af_ops->rebuild_header(sk);
415 skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
419 /* Reserve space for headers and prepare control bits. */
420 skb_reserve(skb, sk->sk_prot->max_header);
421 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
422 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
424 return dccp_transmit_skb(sk, skb);
428 * Do all connect socket setups that can be done AF independent.
430 static inline void dccp_connect_init(struct sock *sk)
432 struct dccp_sock *dp = dccp_sk(sk);
433 struct dst_entry *dst = __sk_dst_get(sk);
434 struct inet_connection_sock *icsk = inet_csk(sk);
437 sock_reset_flag(sk, SOCK_DONE);
439 dccp_sync_mss(sk, dst_mtu(dst));
442 * SWL and AWL are initially adjusted so that they are not less than
443 * the initial Sequence Numbers received and sent, respectively:
444 * SWL := max(GSR + 1 - floor(W/4), ISR),
445 * AWL := max(GSS - W' + 1, ISS).
446 * These adjustments MUST be applied only at the beginning of the
449 dccp_update_gss(sk, dp->dccps_iss);
450 dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
452 /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
453 * initialized to S.ISS (sec. 8.5) */
454 dp->dccps_gar = dp->dccps_iss;
456 icsk->icsk_retransmits = 0;
459 int dccp_connect(struct sock *sk)
462 struct inet_connection_sock *icsk = inet_csk(sk);
464 dccp_connect_init(sk);
466 skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
467 if (unlikely(skb == NULL))
470 /* Reserve space for headers. */
471 skb_reserve(skb, sk->sk_prot->max_header);
473 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
475 dccp_skb_entail(sk, skb);
476 dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
477 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
479 /* Timer for repeating the REQUEST until an answer. */
480 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
481 icsk->icsk_rto, DCCP_RTO_MAX);
485 EXPORT_SYMBOL_GPL(dccp_connect);
487 void dccp_send_ack(struct sock *sk)
489 /* If we have been reset, we may not send again. */
490 if (sk->sk_state != DCCP_CLOSED) {
491 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
495 inet_csk_schedule_ack(sk);
496 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
497 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
503 /* Reserve space for headers */
504 skb_reserve(skb, sk->sk_prot->max_header);
505 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
506 dccp_transmit_skb(sk, skb);
510 EXPORT_SYMBOL_GPL(dccp_send_ack);
512 /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
513 void dccp_send_delayed_ack(struct sock *sk)
515 struct inet_connection_sock *icsk = inet_csk(sk);
517 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
518 * with using 2s, and active senders also piggyback the ACK into a
519 * DATAACK packet, so this is really for quiescent senders.
521 unsigned long timeout = jiffies + 2 * HZ;
523 /* Use new timeout only if there wasn't a older one earlier. */
524 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
525 /* If delack timer was blocked or is about to expire,
528 * FIXME: check the "about to expire" part
530 if (icsk->icsk_ack.blocked) {
535 if (!time_before(timeout, icsk->icsk_ack.timeout))
536 timeout = icsk->icsk_ack.timeout;
538 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
539 icsk->icsk_ack.timeout = timeout;
540 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
543 void dccp_send_sync(struct sock *sk, const u64 ackno,
544 const enum dccp_pkt_type pkt_type)
547 * We are not putting this on the write queue, so
548 * dccp_transmit_skb() will set the ownership to this
551 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
554 /* FIXME: how to make sure the sync is sent? */
555 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
559 /* Reserve space for headers and prepare control bits. */
560 skb_reserve(skb, sk->sk_prot->max_header);
561 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
562 DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
564 dccp_transmit_skb(sk, skb);
567 EXPORT_SYMBOL_GPL(dccp_send_sync);
570 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
571 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
574 void dccp_send_close(struct sock *sk, const int active)
576 struct dccp_sock *dp = dccp_sk(sk);
578 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
580 skb = alloc_skb(sk->sk_prot->max_header, prio);
584 /* Reserve space for headers and prepare control bits. */
585 skb_reserve(skb, sk->sk_prot->max_header);
586 if (dp->dccps_role == DCCP_ROLE_SERVER && !dp->dccps_server_timewait)
587 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSEREQ;
589 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_CLOSE;
592 dccp_write_xmit(sk, 1);
593 dccp_skb_entail(sk, skb);
594 dccp_transmit_skb(sk, skb_clone(skb, prio));
596 * Retransmission timer for active-close: RFC 4340, 8.3 requires
597 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
598 * state can be left. The initial timeout is 2 RTTs.
599 * Since RTT measurement is done by the CCIDs, there is no easy
600 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
601 * is too low (200ms); we use a high value to avoid unnecessary
602 * retransmissions when the link RTT is > 0.2 seconds.
603 * FIXME: Let main module sample RTTs and use that instead.
605 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
606 DCCP_TIMEOUT_INIT, DCCP_RTO_MAX);
608 dccp_transmit_skb(sk, skb);