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);
136 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
138 struct inet_connection_sock *icsk = inet_csk(sk);
139 struct dccp_sock *dp = dccp_sk(sk);
140 int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
141 sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
143 /* Now subtract optional transport overhead */
144 mss_now -= icsk->icsk_ext_hdr_len;
147 * FIXME: this should come from the CCID infrastructure, where, say,
148 * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
149 * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
150 * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
151 * make it a multiple of 4
154 mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
156 /* And store cached results */
157 icsk->icsk_pmtu_cookie = pmtu;
158 dp->dccps_mss_cache = mss_now;
163 EXPORT_SYMBOL_GPL(dccp_sync_mss);
165 void dccp_write_space(struct sock *sk)
167 read_lock(&sk->sk_callback_lock);
169 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
170 wake_up_interruptible(sk->sk_sleep);
171 /* Should agree with poll, otherwise some programs break */
172 if (sock_writeable(sk))
173 sk_wake_async(sk, 2, POLL_OUT);
175 read_unlock(&sk->sk_callback_lock);
179 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
180 * @sk: socket to wait for
182 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb)
184 struct dccp_sock *dp = dccp_sk(sk);
190 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
194 if (signal_pending(current))
197 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
200 dccp_pr_debug("delayed send by %d msec\n", rc);
201 delay = msecs_to_jiffies(rc);
202 sk->sk_write_pending++;
204 schedule_timeout(delay);
206 sk->sk_write_pending--;
209 finish_wait(sk->sk_sleep, &wait);
220 void dccp_write_xmit(struct sock *sk, int block)
222 struct dccp_sock *dp = dccp_sk(sk);
225 while ((skb = skb_peek(&sk->sk_write_queue))) {
226 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
230 sk_reset_timer(sk, &dp->dccps_xmit_timer,
231 msecs_to_jiffies(err)+jiffies);
234 err = dccp_wait_for_ccid(sk, skb);
235 if (err && err != -EINTR)
236 DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
239 skb_dequeue(&sk->sk_write_queue);
241 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
242 const int len = skb->len;
244 if (sk->sk_state == DCCP_PARTOPEN) {
245 /* See 8.1.5. Handshake Completion */
246 inet_csk_schedule_ack(sk);
247 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
248 inet_csk(sk)->icsk_rto,
250 dcb->dccpd_type = DCCP_PKT_DATAACK;
251 } else if (dccp_ack_pending(sk))
252 dcb->dccpd_type = DCCP_PKT_DATAACK;
254 dcb->dccpd_type = DCCP_PKT_DATA;
256 err = dccp_transmit_skb(sk, skb);
257 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
259 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
262 dccp_pr_debug("packet discarded due to err=%d\n", err);
268 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
270 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
271 return -EHOSTUNREACH; /* Routing failure or similar. */
273 return dccp_transmit_skb(sk, (skb_cloned(skb) ?
274 pskb_copy(skb, GFP_ATOMIC):
275 skb_clone(skb, GFP_ATOMIC)));
278 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
279 struct request_sock *req)
282 struct dccp_request_sock *dreq;
283 const u32 dccp_header_size = sizeof(struct dccp_hdr) +
284 sizeof(struct dccp_hdr_ext) +
285 sizeof(struct dccp_hdr_response);
286 struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
291 /* Reserve space for headers. */
292 skb_reserve(skb, sk->sk_prot->max_header);
294 skb->dst = dst_clone(dst);
296 dreq = dccp_rsk(req);
297 if (inet_rsk(req)->acked) /* increase ISS upon retransmission */
298 dccp_inc_seqno(&dreq->dreq_iss);
299 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
300 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
302 if (dccp_insert_options(sk, skb)) {
307 /* Build and checksum header */
308 dh = dccp_zeroed_hdr(skb, dccp_header_size);
310 dh->dccph_sport = inet_sk(sk)->sport;
311 dh->dccph_dport = inet_rsk(req)->rmt_port;
312 dh->dccph_doff = (dccp_header_size +
313 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
314 dh->dccph_type = DCCP_PKT_RESPONSE;
316 dccp_hdr_set_seq(dh, dreq->dreq_iss);
317 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
318 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
320 dccp_csum_outgoing(skb);
322 /* We use `acked' to remember that a Response was already sent. */
323 inet_rsk(req)->acked = 1;
324 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
328 EXPORT_SYMBOL_GPL(dccp_make_response);
330 /* send Reset on established socket, to close or abort the connection */
331 int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
335 * FIXME: what if rebuild_header fails?
336 * Should we be doing a rebuild_header here?
338 int err = inet_sk_rebuild_header(sk);
343 skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1, GFP_ATOMIC);
347 /* Reserve space for headers and prepare control bits. */
348 skb_reserve(skb, sk->sk_prot->max_header);
349 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
350 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
352 return dccp_transmit_skb(sk, skb);
356 * Do all connect socket setups that can be done AF independent.
358 static inline void dccp_connect_init(struct sock *sk)
360 struct dccp_sock *dp = dccp_sk(sk);
361 struct dst_entry *dst = __sk_dst_get(sk);
362 struct inet_connection_sock *icsk = inet_csk(sk);
365 sock_reset_flag(sk, SOCK_DONE);
367 dccp_sync_mss(sk, dst_mtu(dst));
370 * SWL and AWL are initially adjusted so that they are not less than
371 * the initial Sequence Numbers received and sent, respectively:
372 * SWL := max(GSR + 1 - floor(W/4), ISR),
373 * AWL := max(GSS - W' + 1, ISS).
374 * These adjustments MUST be applied only at the beginning of the
377 dccp_update_gss(sk, dp->dccps_iss);
378 dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
380 /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
381 * initialized to S.ISS (sec. 8.5) */
382 dp->dccps_gar = dp->dccps_iss;
384 icsk->icsk_retransmits = 0;
387 int dccp_connect(struct sock *sk)
390 struct inet_connection_sock *icsk = inet_csk(sk);
392 dccp_connect_init(sk);
394 skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
395 if (unlikely(skb == NULL))
398 /* Reserve space for headers. */
399 skb_reserve(skb, sk->sk_prot->max_header);
401 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
403 dccp_skb_entail(sk, skb);
404 dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
405 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
407 /* Timer for repeating the REQUEST until an answer. */
408 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
409 icsk->icsk_rto, DCCP_RTO_MAX);
413 EXPORT_SYMBOL_GPL(dccp_connect);
415 void dccp_send_ack(struct sock *sk)
417 /* If we have been reset, we may not send again. */
418 if (sk->sk_state != DCCP_CLOSED) {
419 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
423 inet_csk_schedule_ack(sk);
424 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
425 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
431 /* Reserve space for headers */
432 skb_reserve(skb, sk->sk_prot->max_header);
433 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
434 dccp_transmit_skb(sk, skb);
438 EXPORT_SYMBOL_GPL(dccp_send_ack);
440 /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
441 void dccp_send_delayed_ack(struct sock *sk)
443 struct inet_connection_sock *icsk = inet_csk(sk);
445 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
446 * with using 2s, and active senders also piggyback the ACK into a
447 * DATAACK packet, so this is really for quiescent senders.
449 unsigned long timeout = jiffies + 2 * HZ;
451 /* Use new timeout only if there wasn't a older one earlier. */
452 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
453 /* If delack timer was blocked or is about to expire,
456 * FIXME: check the "about to expire" part
458 if (icsk->icsk_ack.blocked) {
463 if (!time_before(timeout, icsk->icsk_ack.timeout))
464 timeout = icsk->icsk_ack.timeout;
466 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
467 icsk->icsk_ack.timeout = timeout;
468 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
471 void dccp_send_sync(struct sock *sk, const u64 ackno,
472 const enum dccp_pkt_type pkt_type)
475 * We are not putting this on the write queue, so
476 * dccp_transmit_skb() will set the ownership to this
479 struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
482 /* FIXME: how to make sure the sync is sent? */
483 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type));
487 /* Reserve space for headers and prepare control bits. */
488 skb_reserve(skb, sk->sk_prot->max_header);
489 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
490 DCCP_SKB_CB(skb)->dccpd_ack_seq = ackno;
492 dccp_transmit_skb(sk, skb);
495 EXPORT_SYMBOL_GPL(dccp_send_sync);
498 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
499 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
502 void dccp_send_close(struct sock *sk, const int active)
504 struct dccp_sock *dp = dccp_sk(sk);
506 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
508 skb = alloc_skb(sk->sk_prot->max_header, prio);
512 /* Reserve space for headers and prepare control bits. */
513 skb_reserve(skb, sk->sk_prot->max_header);
514 DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
515 DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
518 dccp_write_xmit(sk, 1);
519 dccp_skb_entail(sk, skb);
520 dccp_transmit_skb(sk, skb_clone(skb, prio));
521 /* FIXME do we need a retransmit timer here? */
523 dccp_transmit_skb(sk, skb);