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/config.h>
14 #include <linux/dccp.h>
15 #include <linux/skbuff.h>
23 static inline void dccp_event_ack_sent(struct sock *sk)
25 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
29 * All SKB's seen here are completely headerless. It is our
30 * job to build the DCCP header, and pass the packet down to
31 * IP so it can do the same plus pass the packet off to the
34 int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
36 if (likely(skb != NULL)) {
37 const struct inet_sock *inet = inet_sk(sk);
38 struct dccp_sock *dp = dccp_sk(sk);
39 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
41 /* XXX For now we're using only 48 bits sequence numbers */
42 const int dccp_header_size = sizeof(*dh) +
43 sizeof(struct dccp_hdr_ext) +
44 dccp_packet_hdr_len(dcb->dccpd_type);
46 u64 ackno = dp->dccps_gsr;
48 dccp_inc_seqno(&dp->dccps_gss);
50 switch (dcb->dccpd_type) {
55 case DCCP_PKT_SYNCACK:
56 ackno = dcb->dccpd_seq;
60 dcb->dccpd_seq = dp->dccps_gss;
61 dccp_insert_options(sk, skb);
63 skb->h.raw = skb_push(skb, dccp_header_size);
67 skb_set_owner_w(skb, sk);
69 /* Build DCCP header and checksum it. */
70 memset(dh, 0, dccp_header_size);
71 dh->dccph_type = dcb->dccpd_type;
72 dh->dccph_sport = inet->sport;
73 dh->dccph_dport = inet->dport;
74 dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4;
75 dh->dccph_ccval = dcb->dccpd_ccval;
76 /* XXX For now we're using only 48 bits sequence numbers */
79 dp->dccps_awh = dp->dccps_gss;
80 dccp_hdr_set_seq(dh, dp->dccps_gss);
82 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
84 switch (dcb->dccpd_type) {
85 case DCCP_PKT_REQUEST:
86 dccp_hdr_request(skb)->dccph_req_service =
90 dccp_hdr_reset(skb)->dccph_reset_code =
91 dcb->dccpd_reset_code;
95 dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr,
99 dccp_event_ack_sent(sk);
101 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
103 err = ip_queue_xmit(skb, 0);
107 /* NET_XMIT_CN is special. It does not guarantee,
108 * that this packet is lost. It tells that device
109 * is about to start to drop packets or already
110 * drops some packets of the same priority and
111 * invokes us to send less aggressively.
113 return err == NET_XMIT_CN ? 0 : err;
118 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
120 struct dccp_sock *dp = dccp_sk(sk);
124 * FIXME: we really should be using the af_specific thing to support
126 * mss_now = pmtu - tp->af_specific->net_header_len -
127 * sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
129 mss_now = pmtu - sizeof(struct iphdr) - sizeof(struct dccp_hdr) -
130 sizeof(struct dccp_hdr_ext);
132 /* Now subtract optional transport overhead */
133 mss_now -= dp->dccps_ext_header_len;
136 * FIXME: this should come from the CCID infrastructure, where, say,
137 * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
138 * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
139 * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
140 * make it a multiple of 4
143 mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
145 /* And store cached results */
146 dp->dccps_pmtu_cookie = pmtu;
147 dp->dccps_mss_cache = mss_now;
152 void dccp_write_space(struct sock *sk)
154 read_lock(&sk->sk_callback_lock);
156 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
157 wake_up_interruptible(sk->sk_sleep);
158 /* Should agree with poll, otherwise some programs break */
159 if (sock_writeable(sk))
160 sk_wake_async(sk, 2, POLL_OUT);
162 read_unlock(&sk->sk_callback_lock);
166 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
167 * @sk: socket to wait for
168 * @timeo: for how long
170 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
173 struct dccp_sock *dp = dccp_sk(sk);
179 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
181 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
185 if (signal_pending(current))
188 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
192 delay = msecs_to_jiffies(rc);
193 if (delay > *timeo || delay < 0)
196 sk->sk_write_pending++;
198 *timeo -= schedule_timeout(delay);
200 sk->sk_write_pending--;
203 finish_wait(sk->sk_sleep, &wait);
213 rc = sock_intr_errno(*timeo);
217 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
219 const struct dccp_sock *dp = dccp_sk(sk);
220 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
224 err = dccp_wait_for_ccid(sk, skb, timeo);
227 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
228 const int len = skb->len;
230 if (sk->sk_state == DCCP_PARTOPEN) {
231 /* See 8.1.5. Handshake Completion */
232 inet_csk_schedule_ack(sk);
233 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
234 inet_csk(sk)->icsk_rto,
236 dcb->dccpd_type = DCCP_PKT_DATAACK;
237 } else if (dccp_ack_pending(sk))
238 dcb->dccpd_type = DCCP_PKT_DATAACK;
240 dcb->dccpd_type = DCCP_PKT_DATA;
242 err = dccp_transmit_skb(sk, skb);
243 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
250 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
252 if (inet_sk_rebuild_header(sk) != 0)
253 return -EHOSTUNREACH; /* Routing failure or similar. */
255 return dccp_transmit_skb(sk, (skb_cloned(skb) ?
256 pskb_copy(skb, GFP_ATOMIC):
257 skb_clone(skb, GFP_ATOMIC)));
260 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
261 struct request_sock *req)
264 struct dccp_request_sock *dreq;
265 const int dccp_header_size = sizeof(struct dccp_hdr) +
266 sizeof(struct dccp_hdr_ext) +
267 sizeof(struct dccp_hdr_response);
268 struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
274 /* Reserve space for headers. */
275 skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
277 skb->dst = dst_clone(dst);
280 dreq = dccp_rsk(req);
281 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
282 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
283 dccp_insert_options(sk, skb);
285 skb->h.raw = skb_push(skb, dccp_header_size);
288 memset(dh, 0, dccp_header_size);
290 dh->dccph_sport = inet_sk(sk)->sport;
291 dh->dccph_dport = inet_rsk(req)->rmt_port;
292 dh->dccph_doff = (dccp_header_size +
293 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
294 dh->dccph_type = DCCP_PKT_RESPONSE;
296 dccp_hdr_set_seq(dh, dreq->dreq_iss);
297 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
298 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
300 dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
301 inet_rsk(req)->rmt_addr);
303 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
307 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
308 const enum dccp_reset_codes code)
312 struct dccp_sock *dp = dccp_sk(sk);
313 const int dccp_header_size = sizeof(struct dccp_hdr) +
314 sizeof(struct dccp_hdr_ext) +
315 sizeof(struct dccp_hdr_reset);
316 struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
322 /* Reserve space for headers. */
323 skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
325 skb->dst = dst_clone(dst);
328 dccp_inc_seqno(&dp->dccps_gss);
330 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
331 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
332 DCCP_SKB_CB(skb)->dccpd_seq = dp->dccps_gss;
333 dccp_insert_options(sk, skb);
335 skb->h.raw = skb_push(skb, dccp_header_size);
338 memset(dh, 0, dccp_header_size);
340 dh->dccph_sport = inet_sk(sk)->sport;
341 dh->dccph_dport = inet_sk(sk)->dport;
342 dh->dccph_doff = (dccp_header_size +
343 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
344 dh->dccph_type = DCCP_PKT_RESET;
346 dccp_hdr_set_seq(dh, dp->dccps_gss);
347 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
349 dccp_hdr_reset(skb)->dccph_reset_code = code;
351 dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
354 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
359 * Do all connect socket setups that can be done AF independent.
361 static inline void dccp_connect_init(struct sock *sk)
363 struct dst_entry *dst = __sk_dst_get(sk);
364 struct inet_connection_sock *icsk = inet_csk(sk);
367 sock_reset_flag(sk, SOCK_DONE);
369 dccp_sync_mss(sk, dst_mtu(dst));
372 * FIXME: set dp->{dccps_swh,dccps_swl}, with
373 * something like dccp_inc_seq
376 icsk->icsk_retransmits = 0;
379 int dccp_connect(struct sock *sk)
382 struct inet_connection_sock *icsk = inet_csk(sk);
384 dccp_connect_init(sk);
386 skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
387 if (unlikely(skb == NULL))
390 /* Reserve space for headers. */
391 skb_reserve(skb, MAX_DCCP_HEADER);
393 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
395 skb_set_owner_w(skb, sk);
397 BUG_TRAP(sk->sk_send_head == NULL);
398 sk->sk_send_head = skb;
399 dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
400 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
402 /* Timer for repeating the REQUEST until an answer. */
403 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
404 icsk->icsk_rto, DCCP_RTO_MAX);
408 void dccp_send_ack(struct sock *sk)
410 /* If we have been reset, we may not send again. */
411 if (sk->sk_state != DCCP_CLOSED) {
412 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
415 inet_csk_schedule_ack(sk);
416 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
417 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
423 /* Reserve space for headers */
424 skb_reserve(skb, MAX_DCCP_HEADER);
426 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
427 skb_set_owner_w(skb, sk);
428 dccp_transmit_skb(sk, skb);
432 EXPORT_SYMBOL_GPL(dccp_send_ack);
434 void dccp_send_delayed_ack(struct sock *sk)
436 struct inet_connection_sock *icsk = inet_csk(sk);
438 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
439 * with using 2s, and active senders also piggyback the ACK into a
440 * DATAACK packet, so this is really for quiescent senders.
442 unsigned long timeout = jiffies + 2 * HZ;
444 /* Use new timeout only if there wasn't a older one earlier. */
445 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
446 /* If delack timer was blocked or is about to expire,
449 * FIXME: check the "about to expire" part
451 if (icsk->icsk_ack.blocked) {
456 if (!time_before(timeout, icsk->icsk_ack.timeout))
457 timeout = icsk->icsk_ack.timeout;
459 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
460 icsk->icsk_ack.timeout = timeout;
461 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
464 void dccp_send_sync(struct sock *sk, const u64 seq,
465 const enum dccp_pkt_type pkt_type)
468 * We are not putting this on the write queue, so
469 * dccp_transmit_skb() will set the ownership to this
472 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
475 /* FIXME: how to make sure the sync is sent? */
478 /* Reserve space for headers and prepare control bits. */
479 skb_reserve(skb, MAX_DCCP_HEADER);
481 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
482 DCCP_SKB_CB(skb)->dccpd_seq = seq;
484 skb_set_owner_w(skb, sk);
485 dccp_transmit_skb(sk, skb);
489 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
490 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
493 void dccp_send_close(struct sock *sk, const int active)
495 struct dccp_sock *dp = dccp_sk(sk);
497 const unsigned int prio = active ? GFP_KERNEL : GFP_ATOMIC;
499 skb = alloc_skb(sk->sk_prot->max_header, prio);
503 /* Reserve space for headers and prepare control bits. */
504 skb_reserve(skb, sk->sk_prot->max_header);
506 DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
507 DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
509 skb_set_owner_w(skb, sk);
511 BUG_TRAP(sk->sk_send_head == NULL);
512 sk->sk_send_head = skb;
513 dccp_transmit_skb(sk, skb_clone(skb, prio));
515 dccp_transmit_skb(sk, skb);