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/kernel.h>
16 #include <linux/skbuff.h>
24 static inline void dccp_event_ack_sent(struct sock *sk)
26 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
29 static inline 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 struct dccp_sock *dp = dccp_sk(sk);
47 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
49 /* XXX For now we're using only 48 bits sequence numbers */
50 const int dccp_header_size = sizeof(*dh) +
51 sizeof(struct dccp_hdr_ext) +
52 dccp_packet_hdr_len(dcb->dccpd_type);
54 u64 ackno = dp->dccps_gsr;
56 dccp_inc_seqno(&dp->dccps_gss);
58 switch (dcb->dccpd_type) {
63 case DCCP_PKT_SYNCACK:
64 ackno = dcb->dccpd_seq;
68 dcb->dccpd_seq = dp->dccps_gss;
69 dccp_insert_options(sk, skb);
71 skb->h.raw = skb_push(skb, dccp_header_size);
75 * Only data packets should come through with skb->sk set.
78 skb_set_owner_w(skb, sk);
80 /* Build DCCP header and checksum it. */
81 memset(dh, 0, dccp_header_size);
82 dh->dccph_type = dcb->dccpd_type;
83 dh->dccph_sport = inet->sport;
84 dh->dccph_dport = inet->dport;
85 dh->dccph_doff = (dccp_header_size + dcb->dccpd_opt_len) / 4;
86 dh->dccph_ccval = dcb->dccpd_ccval;
87 /* XXX For now we're using only 48 bits sequence numbers */
90 dp->dccps_awh = dp->dccps_gss;
91 dccp_hdr_set_seq(dh, dp->dccps_gss);
93 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
95 switch (dcb->dccpd_type) {
96 case DCCP_PKT_REQUEST:
97 dccp_hdr_request(skb)->dccph_req_service =
101 dccp_hdr_reset(skb)->dccph_reset_code =
102 dcb->dccpd_reset_code;
106 dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr,
110 dccp_event_ack_sent(sk);
112 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
114 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
115 err = ip_queue_xmit(skb, 0);
119 /* NET_XMIT_CN is special. It does not guarantee,
120 * that this packet is lost. It tells that device
121 * is about to start to drop packets or already
122 * drops some packets of the same priority and
123 * invokes us to send less aggressively.
125 return err == NET_XMIT_CN ? 0 : err;
130 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
132 struct dccp_sock *dp = dccp_sk(sk);
136 * FIXME: we really should be using the af_specific thing to support
138 * mss_now = pmtu - tp->af_specific->net_header_len -
139 * sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
141 mss_now = pmtu - sizeof(struct iphdr) - sizeof(struct dccp_hdr) -
142 sizeof(struct dccp_hdr_ext);
144 /* Now subtract optional transport overhead */
145 mss_now -= dp->dccps_ext_header_len;
148 * FIXME: this should come from the CCID infrastructure, where, say,
149 * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
150 * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
151 * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
152 * make it a multiple of 4
155 mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
157 /* And store cached results */
158 dp->dccps_pmtu_cookie = pmtu;
159 dp->dccps_mss_cache = mss_now;
164 void dccp_write_space(struct sock *sk)
166 read_lock(&sk->sk_callback_lock);
168 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
169 wake_up_interruptible(sk->sk_sleep);
170 /* Should agree with poll, otherwise some programs break */
171 if (sock_writeable(sk))
172 sk_wake_async(sk, 2, POLL_OUT);
174 read_unlock(&sk->sk_callback_lock);
178 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
179 * @sk: socket to wait for
180 * @timeo: for how long
182 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
185 struct dccp_sock *dp = dccp_sk(sk);
191 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
193 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
197 if (signal_pending(current))
200 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
204 delay = msecs_to_jiffies(rc);
205 if (delay > *timeo || delay < 0)
208 sk->sk_write_pending++;
210 *timeo -= schedule_timeout(delay);
212 sk->sk_write_pending--;
215 finish_wait(sk->sk_sleep, &wait);
225 rc = sock_intr_errno(*timeo);
229 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
231 const struct dccp_sock *dp = dccp_sk(sk);
232 int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
236 err = dccp_wait_for_ccid(sk, skb, timeo);
239 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
240 const int len = skb->len;
242 if (sk->sk_state == DCCP_PARTOPEN) {
243 /* See 8.1.5. Handshake Completion */
244 inet_csk_schedule_ack(sk);
245 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
246 inet_csk(sk)->icsk_rto,
248 dcb->dccpd_type = DCCP_PKT_DATAACK;
249 } else if (dccp_ack_pending(sk))
250 dcb->dccpd_type = DCCP_PKT_DATAACK;
252 dcb->dccpd_type = DCCP_PKT_DATA;
254 err = dccp_transmit_skb(sk, skb);
255 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
262 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
264 if (inet_sk_rebuild_header(sk) != 0)
265 return -EHOSTUNREACH; /* Routing failure or similar. */
267 return dccp_transmit_skb(sk, (skb_cloned(skb) ?
268 pskb_copy(skb, GFP_ATOMIC):
269 skb_clone(skb, GFP_ATOMIC)));
272 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
273 struct request_sock *req)
276 struct dccp_request_sock *dreq;
277 const int dccp_header_size = sizeof(struct dccp_hdr) +
278 sizeof(struct dccp_hdr_ext) +
279 sizeof(struct dccp_hdr_response);
280 struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
286 /* Reserve space for headers. */
287 skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
289 skb->dst = dst_clone(dst);
292 dreq = dccp_rsk(req);
293 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
294 DCCP_SKB_CB(skb)->dccpd_seq = dreq->dreq_iss;
295 dccp_insert_options(sk, skb);
297 skb->h.raw = skb_push(skb, dccp_header_size);
300 memset(dh, 0, dccp_header_size);
302 dh->dccph_sport = inet_sk(sk)->sport;
303 dh->dccph_dport = inet_rsk(req)->rmt_port;
304 dh->dccph_doff = (dccp_header_size +
305 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
306 dh->dccph_type = DCCP_PKT_RESPONSE;
308 dccp_hdr_set_seq(dh, dreq->dreq_iss);
309 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
310 dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
312 dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
313 inet_rsk(req)->rmt_addr);
315 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
319 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
320 const enum dccp_reset_codes code)
324 struct dccp_sock *dp = dccp_sk(sk);
325 const int dccp_header_size = sizeof(struct dccp_hdr) +
326 sizeof(struct dccp_hdr_ext) +
327 sizeof(struct dccp_hdr_reset);
328 struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
334 /* Reserve space for headers. */
335 skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
337 skb->dst = dst_clone(dst);
340 dccp_inc_seqno(&dp->dccps_gss);
342 DCCP_SKB_CB(skb)->dccpd_reset_code = code;
343 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESET;
344 DCCP_SKB_CB(skb)->dccpd_seq = dp->dccps_gss;
345 dccp_insert_options(sk, skb);
347 skb->h.raw = skb_push(skb, dccp_header_size);
350 memset(dh, 0, dccp_header_size);
352 dh->dccph_sport = inet_sk(sk)->sport;
353 dh->dccph_dport = inet_sk(sk)->dport;
354 dh->dccph_doff = (dccp_header_size +
355 DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
356 dh->dccph_type = DCCP_PKT_RESET;
358 dccp_hdr_set_seq(dh, dp->dccps_gss);
359 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
361 dccp_hdr_reset(skb)->dccph_reset_code = code;
363 dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
366 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
371 * Do all connect socket setups that can be done AF independent.
373 static inline void dccp_connect_init(struct sock *sk)
375 struct dst_entry *dst = __sk_dst_get(sk);
376 struct inet_connection_sock *icsk = inet_csk(sk);
379 sock_reset_flag(sk, SOCK_DONE);
381 dccp_sync_mss(sk, dst_mtu(dst));
384 * FIXME: set dp->{dccps_swh,dccps_swl}, with
385 * something like dccp_inc_seq
388 icsk->icsk_retransmits = 0;
391 int dccp_connect(struct sock *sk)
394 struct inet_connection_sock *icsk = inet_csk(sk);
396 dccp_connect_init(sk);
398 skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
399 if (unlikely(skb == NULL))
402 /* Reserve space for headers. */
403 skb_reserve(skb, MAX_DCCP_HEADER);
405 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
408 dccp_skb_entail(sk, skb);
409 dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
410 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
412 /* Timer for repeating the REQUEST until an answer. */
413 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
414 icsk->icsk_rto, DCCP_RTO_MAX);
418 void dccp_send_ack(struct sock *sk)
420 /* If we have been reset, we may not send again. */
421 if (sk->sk_state != DCCP_CLOSED) {
422 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
425 inet_csk_schedule_ack(sk);
426 inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
427 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
433 /* Reserve space for headers */
434 skb_reserve(skb, MAX_DCCP_HEADER);
436 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
437 dccp_transmit_skb(sk, skb);
441 EXPORT_SYMBOL_GPL(dccp_send_ack);
443 void dccp_send_delayed_ack(struct sock *sk)
445 struct inet_connection_sock *icsk = inet_csk(sk);
447 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
448 * with using 2s, and active senders also piggyback the ACK into a
449 * DATAACK packet, so this is really for quiescent senders.
451 unsigned long timeout = jiffies + 2 * HZ;
453 /* Use new timeout only if there wasn't a older one earlier. */
454 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
455 /* If delack timer was blocked or is about to expire,
458 * FIXME: check the "about to expire" part
460 if (icsk->icsk_ack.blocked) {
465 if (!time_before(timeout, icsk->icsk_ack.timeout))
466 timeout = icsk->icsk_ack.timeout;
468 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
469 icsk->icsk_ack.timeout = timeout;
470 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
473 void dccp_send_sync(struct sock *sk, const u64 seq,
474 const enum dccp_pkt_type pkt_type)
477 * We are not putting this on the write queue, so
478 * dccp_transmit_skb() will set the ownership to this
481 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
484 /* FIXME: how to make sure the sync is sent? */
487 /* Reserve space for headers and prepare control bits. */
488 skb_reserve(skb, MAX_DCCP_HEADER);
490 DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
491 DCCP_SKB_CB(skb)->dccpd_seq = seq;
493 dccp_transmit_skb(sk, skb);
497 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
498 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
501 void dccp_send_close(struct sock *sk, const int active)
503 struct dccp_sock *dp = dccp_sk(sk);
505 const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
507 skb = alloc_skb(sk->sk_prot->max_header, prio);
511 /* Reserve space for headers and prepare control bits. */
512 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_skb_entail(sk, skb);
519 dccp_transmit_skb(sk, skb_clone(skb, prio));
521 dccp_transmit_skb(sk, skb);