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
)) {
91 skb
->h
.raw
= skb_push(skb
, dccp_header_size
);
94 /* Build DCCP header and checksum it. */
95 memset(dh
, 0, dccp_header_size
);
96 dh
->dccph_type
= dcb
->dccpd_type
;
97 dh
->dccph_sport
= inet
->sport
;
98 dh
->dccph_dport
= inet
->dport
;
99 dh
->dccph_doff
= (dccp_header_size
+ dcb
->dccpd_opt_len
) / 4;
100 dh
->dccph_ccval
= dcb
->dccpd_ccval
;
101 /* XXX For now we're using only 48 bits sequence numbers */
104 dp
->dccps_awh
= dp
->dccps_gss
;
105 dccp_hdr_set_seq(dh
, dp
->dccps_gss
);
107 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb
), ackno
);
109 switch (dcb
->dccpd_type
) {
110 case DCCP_PKT_REQUEST
:
111 dccp_hdr_request(skb
)->dccph_req_service
=
115 dccp_hdr_reset(skb
)->dccph_reset_code
=
116 dcb
->dccpd_reset_code
;
120 icsk
->icsk_af_ops
->send_check(sk
, skb
->len
, skb
);
123 dccp_event_ack_sent(sk
);
125 DCCP_INC_STATS(DCCP_MIB_OUTSEGS
);
127 memset(&(IPCB(skb
)->opt
), 0, sizeof(IPCB(skb
)->opt
));
128 err
= icsk
->icsk_af_ops
->queue_xmit(skb
, 0);
132 /* NET_XMIT_CN is special. It does not guarantee,
133 * that this packet is lost. It tells that device
134 * is about to start to drop packets or already
135 * drops some packets of the same priority and
136 * invokes us to send less aggressively.
138 return err
== NET_XMIT_CN
? 0 : err
;
143 unsigned int dccp_sync_mss(struct sock
*sk
, u32 pmtu
)
145 struct inet_connection_sock
*icsk
= inet_csk(sk
);
146 struct dccp_sock
*dp
= dccp_sk(sk
);
147 int mss_now
= (pmtu
- icsk
->icsk_af_ops
->net_header_len
-
148 sizeof(struct dccp_hdr
) - sizeof(struct dccp_hdr_ext
));
150 /* Now subtract optional transport overhead */
151 mss_now
-= icsk
->icsk_ext_hdr_len
;
154 * FIXME: this should come from the CCID infrastructure, where, say,
155 * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
156 * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
157 * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
158 * make it a multiple of 4
161 mss_now
-= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
163 /* And store cached results */
164 icsk
->icsk_pmtu_cookie
= pmtu
;
165 dp
->dccps_mss_cache
= mss_now
;
170 EXPORT_SYMBOL_GPL(dccp_sync_mss
);
172 void dccp_write_space(struct sock
*sk
)
174 read_lock(&sk
->sk_callback_lock
);
176 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
177 wake_up_interruptible(sk
->sk_sleep
);
178 /* Should agree with poll, otherwise some programs break */
179 if (sock_writeable(sk
))
180 sk_wake_async(sk
, 2, POLL_OUT
);
182 read_unlock(&sk
->sk_callback_lock
);
186 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
187 * @sk: socket to wait for
188 * @timeo: for how long
190 static int dccp_wait_for_ccid(struct sock
*sk
, struct sk_buff
*skb
,
193 struct dccp_sock
*dp
= dccp_sk(sk
);
199 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
205 if (signal_pending(current
))
208 rc
= ccid_hc_tx_send_packet(dp
->dccps_hc_tx_ccid
, sk
, skb
,
212 delay
= msecs_to_jiffies(rc
);
213 if (delay
> *timeo
|| delay
< 0)
216 sk
->sk_write_pending
++;
218 *timeo
-= schedule_timeout(delay
);
220 sk
->sk_write_pending
--;
223 finish_wait(sk
->sk_sleep
, &wait
);
233 rc
= sock_intr_errno(*timeo
);
237 static void dccp_write_xmit_timer(unsigned long data
) {
238 struct sock
*sk
= (struct sock
*)data
;
239 struct dccp_sock
*dp
= dccp_sk(sk
);
242 if (sock_owned_by_user(sk
))
243 sk_reset_timer(sk
, &dp
->dccps_xmit_timer
, jiffies
+1);
245 dccp_write_xmit(sk
, 0);
250 void dccp_write_xmit(struct sock
*sk
, int block
)
252 struct dccp_sock
*dp
= dccp_sk(sk
);
254 long timeo
= 30000; /* If a packet is taking longer than 2 secs
255 we have other issues */
257 while ((skb
= skb_peek(&sk
->sk_write_queue
))) {
258 int err
= ccid_hc_tx_send_packet(dp
->dccps_hc_tx_ccid
, sk
, skb
,
263 sk_reset_timer(sk
, &dp
->dccps_xmit_timer
,
264 msecs_to_jiffies(err
)+jiffies
);
267 err
= dccp_wait_for_ccid(sk
, skb
, &timeo
);
269 printk(KERN_CRIT
"%s:err at dccp_wait_for_ccid"
270 " %d\n", __FUNCTION__
, err
);
275 skb_dequeue(&sk
->sk_write_queue
);
277 struct dccp_skb_cb
*dcb
= DCCP_SKB_CB(skb
);
278 const int len
= skb
->len
;
280 if (sk
->sk_state
== DCCP_PARTOPEN
) {
281 /* See 8.1.5. Handshake Completion */
282 inet_csk_schedule_ack(sk
);
283 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
284 inet_csk(sk
)->icsk_rto
,
286 dcb
->dccpd_type
= DCCP_PKT_DATAACK
;
287 } else if (dccp_ack_pending(sk
))
288 dcb
->dccpd_type
= DCCP_PKT_DATAACK
;
290 dcb
->dccpd_type
= DCCP_PKT_DATA
;
292 err
= dccp_transmit_skb(sk
, skb
);
293 ccid_hc_tx_packet_sent(dp
->dccps_hc_tx_ccid
, sk
, 0, len
);
295 printk(KERN_CRIT
"%s:err from "
296 "ccid_hc_tx_packet_sent %d\n",
305 int dccp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
)
307 if (inet_csk(sk
)->icsk_af_ops
->rebuild_header(sk
) != 0)
308 return -EHOSTUNREACH
; /* Routing failure or similar. */
310 return dccp_transmit_skb(sk
, (skb_cloned(skb
) ?
311 pskb_copy(skb
, GFP_ATOMIC
):
312 skb_clone(skb
, GFP_ATOMIC
)));
315 struct sk_buff
*dccp_make_response(struct sock
*sk
, struct dst_entry
*dst
,
316 struct request_sock
*req
)
319 struct dccp_request_sock
*dreq
;
320 const u32 dccp_header_size
= sizeof(struct dccp_hdr
) +
321 sizeof(struct dccp_hdr_ext
) +
322 sizeof(struct dccp_hdr_response
);
323 struct sk_buff
*skb
= sock_wmalloc(sk
, sk
->sk_prot
->max_header
, 1,
328 /* Reserve space for headers. */
329 skb_reserve(skb
, sk
->sk_prot
->max_header
);
331 skb
->dst
= dst_clone(dst
);
334 dreq
= dccp_rsk(req
);
335 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_RESPONSE
;
336 DCCP_SKB_CB(skb
)->dccpd_seq
= dreq
->dreq_iss
;
338 if (dccp_insert_options(sk
, skb
)) {
343 skb
->h
.raw
= skb_push(skb
, dccp_header_size
);
346 memset(dh
, 0, dccp_header_size
);
348 dh
->dccph_sport
= inet_sk(sk
)->sport
;
349 dh
->dccph_dport
= inet_rsk(req
)->rmt_port
;
350 dh
->dccph_doff
= (dccp_header_size
+
351 DCCP_SKB_CB(skb
)->dccpd_opt_len
) / 4;
352 dh
->dccph_type
= DCCP_PKT_RESPONSE
;
354 dccp_hdr_set_seq(dh
, dreq
->dreq_iss
);
355 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb
), dreq
->dreq_isr
);
356 dccp_hdr_response(skb
)->dccph_resp_service
= dreq
->dreq_service
;
358 DCCP_INC_STATS(DCCP_MIB_OUTSEGS
);
362 EXPORT_SYMBOL_GPL(dccp_make_response
);
364 static struct sk_buff
*dccp_make_reset(struct sock
*sk
, struct dst_entry
*dst
,
365 const enum dccp_reset_codes code
)
369 struct dccp_sock
*dp
= dccp_sk(sk
);
370 const u32 dccp_header_size
= sizeof(struct dccp_hdr
) +
371 sizeof(struct dccp_hdr_ext
) +
372 sizeof(struct dccp_hdr_reset
);
373 struct sk_buff
*skb
= sock_wmalloc(sk
, sk
->sk_prot
->max_header
, 1,
378 /* Reserve space for headers. */
379 skb_reserve(skb
, sk
->sk_prot
->max_header
);
381 skb
->dst
= dst_clone(dst
);
384 dccp_inc_seqno(&dp
->dccps_gss
);
386 DCCP_SKB_CB(skb
)->dccpd_reset_code
= code
;
387 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_RESET
;
388 DCCP_SKB_CB(skb
)->dccpd_seq
= dp
->dccps_gss
;
390 if (dccp_insert_options(sk
, skb
)) {
395 skb
->h
.raw
= skb_push(skb
, dccp_header_size
);
398 memset(dh
, 0, dccp_header_size
);
400 dh
->dccph_sport
= inet_sk(sk
)->sport
;
401 dh
->dccph_dport
= inet_sk(sk
)->dport
;
402 dh
->dccph_doff
= (dccp_header_size
+
403 DCCP_SKB_CB(skb
)->dccpd_opt_len
) / 4;
404 dh
->dccph_type
= DCCP_PKT_RESET
;
406 dccp_hdr_set_seq(dh
, dp
->dccps_gss
);
407 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb
), dp
->dccps_gsr
);
409 dccp_hdr_reset(skb
)->dccph_reset_code
= code
;
410 inet_csk(sk
)->icsk_af_ops
->send_check(sk
, skb
->len
, skb
);
412 DCCP_INC_STATS(DCCP_MIB_OUTSEGS
);
416 int dccp_send_reset(struct sock
*sk
, enum dccp_reset_codes code
)
419 * FIXME: what if rebuild_header fails?
420 * Should we be doing a rebuild_header here?
422 int err
= inet_sk_rebuild_header(sk
);
425 struct sk_buff
*skb
= dccp_make_reset(sk
, sk
->sk_dst_cache
,
428 memset(&(IPCB(skb
)->opt
), 0, sizeof(IPCB(skb
)->opt
));
429 err
= inet_csk(sk
)->icsk_af_ops
->queue_xmit(skb
, 0);
430 if (err
== NET_XMIT_CN
)
439 * Do all connect socket setups that can be done AF independent.
441 static inline void dccp_connect_init(struct sock
*sk
)
443 struct dccp_sock
*dp
= dccp_sk(sk
);
444 struct dst_entry
*dst
= __sk_dst_get(sk
);
445 struct inet_connection_sock
*icsk
= inet_csk(sk
);
448 sock_reset_flag(sk
, SOCK_DONE
);
450 dccp_sync_mss(sk
, dst_mtu(dst
));
452 dccp_update_gss(sk
, dp
->dccps_iss
);
454 * SWL and AWL are initially adjusted so that they are not less than
455 * the initial Sequence Numbers received and sent, respectively:
456 * SWL := max(GSR + 1 - floor(W/4), ISR),
457 * AWL := max(GSS - W' + 1, ISS).
458 * These adjustments MUST be applied only at the beginning of the
461 dccp_set_seqno(&dp
->dccps_awl
, max48(dp
->dccps_awl
, dp
->dccps_iss
));
463 icsk
->icsk_retransmits
= 0;
464 init_timer(&dp
->dccps_xmit_timer
);
465 dp
->dccps_xmit_timer
.data
= (unsigned long)sk
;
466 dp
->dccps_xmit_timer
.function
= dccp_write_xmit_timer
;
469 int dccp_connect(struct sock
*sk
)
472 struct inet_connection_sock
*icsk
= inet_csk(sk
);
474 dccp_connect_init(sk
);
476 skb
= alloc_skb(sk
->sk_prot
->max_header
, sk
->sk_allocation
);
477 if (unlikely(skb
== NULL
))
480 /* Reserve space for headers. */
481 skb_reserve(skb
, sk
->sk_prot
->max_header
);
483 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_REQUEST
;
486 dccp_skb_entail(sk
, skb
);
487 dccp_transmit_skb(sk
, skb_clone(skb
, GFP_KERNEL
));
488 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS
);
490 /* Timer for repeating the REQUEST until an answer. */
491 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
492 icsk
->icsk_rto
, DCCP_RTO_MAX
);
496 EXPORT_SYMBOL_GPL(dccp_connect
);
498 void dccp_send_ack(struct sock
*sk
)
500 /* If we have been reset, we may not send again. */
501 if (sk
->sk_state
!= DCCP_CLOSED
) {
502 struct sk_buff
*skb
= alloc_skb(sk
->sk_prot
->max_header
,
506 inet_csk_schedule_ack(sk
);
507 inet_csk(sk
)->icsk_ack
.ato
= TCP_ATO_MIN
;
508 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
514 /* Reserve space for headers */
515 skb_reserve(skb
, sk
->sk_prot
->max_header
);
517 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_ACK
;
518 dccp_transmit_skb(sk
, skb
);
522 EXPORT_SYMBOL_GPL(dccp_send_ack
);
524 void dccp_send_delayed_ack(struct sock
*sk
)
526 struct inet_connection_sock
*icsk
= inet_csk(sk
);
528 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
529 * with using 2s, and active senders also piggyback the ACK into a
530 * DATAACK packet, so this is really for quiescent senders.
532 unsigned long timeout
= jiffies
+ 2 * HZ
;
534 /* Use new timeout only if there wasn't a older one earlier. */
535 if (icsk
->icsk_ack
.pending
& ICSK_ACK_TIMER
) {
536 /* If delack timer was blocked or is about to expire,
539 * FIXME: check the "about to expire" part
541 if (icsk
->icsk_ack
.blocked
) {
546 if (!time_before(timeout
, icsk
->icsk_ack
.timeout
))
547 timeout
= icsk
->icsk_ack
.timeout
;
549 icsk
->icsk_ack
.pending
|= ICSK_ACK_SCHED
| ICSK_ACK_TIMER
;
550 icsk
->icsk_ack
.timeout
= timeout
;
551 sk_reset_timer(sk
, &icsk
->icsk_delack_timer
, timeout
);
554 void dccp_send_sync(struct sock
*sk
, const u64 seq
,
555 const enum dccp_pkt_type pkt_type
)
558 * We are not putting this on the write queue, so
559 * dccp_transmit_skb() will set the ownership to this
562 struct sk_buff
*skb
= alloc_skb(sk
->sk_prot
->max_header
, GFP_ATOMIC
);
565 /* FIXME: how to make sure the sync is sent? */
568 /* Reserve space for headers and prepare control bits. */
569 skb_reserve(skb
, sk
->sk_prot
->max_header
);
571 DCCP_SKB_CB(skb
)->dccpd_type
= pkt_type
;
572 DCCP_SKB_CB(skb
)->dccpd_seq
= seq
;
574 dccp_transmit_skb(sk
, skb
);
577 EXPORT_SYMBOL_GPL(dccp_send_sync
);
580 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
581 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
584 void dccp_send_close(struct sock
*sk
, const int active
)
586 struct dccp_sock
*dp
= dccp_sk(sk
);
588 const gfp_t prio
= active
? GFP_KERNEL
: GFP_ATOMIC
;
590 skb
= alloc_skb(sk
->sk_prot
->max_header
, prio
);
594 /* Reserve space for headers and prepare control bits. */
595 skb_reserve(skb
, sk
->sk_prot
->max_header
);
597 DCCP_SKB_CB(skb
)->dccpd_type
= dp
->dccps_role
== DCCP_ROLE_CLIENT
?
598 DCCP_PKT_CLOSE
: DCCP_PKT_CLOSEREQ
;
601 dccp_write_xmit(sk
, 1);
602 dccp_skb_entail(sk
, skb
);
603 dccp_transmit_skb(sk
, skb_clone(skb
, prio
));
604 /* FIXME do we need a retransmit timer here? */
606 dccp_transmit_skb(sk
, skb
);