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 err
= icsk
->icsk_af_ops
->queue_xmit(skb
, 0);
130 return net_xmit_eval(err
);
136 * dccp_determine_ccmps - Find out about CCID-specfic packet-size limits
137 * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
138 * since the RX CCID is restricted to feedback packets (Acks), which are small
139 * in comparison with the data traffic. A value of 0 means "no current CCMPS".
141 static u32
dccp_determine_ccmps(const struct dccp_sock
*dp
)
143 const struct ccid
*tx_ccid
= dp
->dccps_hc_tx_ccid
;
145 if (tx_ccid
== NULL
|| tx_ccid
->ccid_ops
== NULL
)
147 return tx_ccid
->ccid_ops
->ccid_ccmps
;
150 unsigned int dccp_sync_mss(struct sock
*sk
, u32 pmtu
)
152 struct inet_connection_sock
*icsk
= inet_csk(sk
);
153 struct dccp_sock
*dp
= dccp_sk(sk
);
154 u32 ccmps
= dccp_determine_ccmps(dp
);
155 int cur_mps
= ccmps
? min(pmtu
, ccmps
) : pmtu
;
157 /* Account for header lengths and IPv4/v6 option overhead */
158 cur_mps
-= (icsk
->icsk_af_ops
->net_header_len
+ icsk
->icsk_ext_hdr_len
+
159 sizeof(struct dccp_hdr
) + sizeof(struct dccp_hdr_ext
));
162 * FIXME: this should come from the CCID infrastructure, where, say,
163 * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
164 * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
165 * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
166 * make it a multiple of 4
169 cur_mps
-= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
171 /* And store cached results */
172 icsk
->icsk_pmtu_cookie
= pmtu
;
173 dp
->dccps_mss_cache
= cur_mps
;
178 EXPORT_SYMBOL_GPL(dccp_sync_mss
);
180 void dccp_write_space(struct sock
*sk
)
182 read_lock(&sk
->sk_callback_lock
);
184 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
185 wake_up_interruptible(sk
->sk_sleep
);
186 /* Should agree with poll, otherwise some programs break */
187 if (sock_writeable(sk
))
188 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
190 read_unlock(&sk
->sk_callback_lock
);
194 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
195 * @sk: socket to wait for
196 * @skb: current skb to pass on for waiting
197 * @delay: sleep timeout in milliseconds (> 0)
198 * This function is called by default when the socket is closed, and
199 * when a non-zero linger time is set on the socket. For consistency
201 static int dccp_wait_for_ccid(struct sock
*sk
, struct sk_buff
*skb
, int delay
)
203 struct dccp_sock
*dp
= dccp_sk(sk
);
205 unsigned long jiffdelay
;
209 dccp_pr_debug("delayed send by %d msec\n", delay
);
210 jiffdelay
= msecs_to_jiffies(delay
);
212 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
214 sk
->sk_write_pending
++;
216 schedule_timeout(jiffdelay
);
218 sk
->sk_write_pending
--;
222 if (signal_pending(current
))
225 rc
= ccid_hc_tx_send_packet(dp
->dccps_hc_tx_ccid
, sk
, skb
);
226 } while ((delay
= rc
) > 0);
228 finish_wait(sk
->sk_sleep
, &wait
);
239 void dccp_write_xmit(struct sock
*sk
, int block
)
241 struct dccp_sock
*dp
= dccp_sk(sk
);
244 while ((skb
= skb_peek(&sk
->sk_write_queue
))) {
245 int err
= ccid_hc_tx_send_packet(dp
->dccps_hc_tx_ccid
, sk
, skb
);
249 sk_reset_timer(sk
, &dp
->dccps_xmit_timer
,
250 msecs_to_jiffies(err
)+jiffies
);
253 err
= dccp_wait_for_ccid(sk
, skb
, err
);
254 if (err
&& err
!= -EINTR
)
255 DCCP_BUG("err=%d after dccp_wait_for_ccid", err
);
258 skb_dequeue(&sk
->sk_write_queue
);
260 struct dccp_skb_cb
*dcb
= DCCP_SKB_CB(skb
);
261 const int len
= skb
->len
;
263 if (sk
->sk_state
== DCCP_PARTOPEN
) {
264 /* See 8.1.5. Handshake Completion */
265 inet_csk_schedule_ack(sk
);
266 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
267 inet_csk(sk
)->icsk_rto
,
269 dcb
->dccpd_type
= DCCP_PKT_DATAACK
;
270 } else if (dccp_ack_pending(sk
))
271 dcb
->dccpd_type
= DCCP_PKT_DATAACK
;
273 dcb
->dccpd_type
= DCCP_PKT_DATA
;
275 err
= dccp_transmit_skb(sk
, skb
);
276 ccid_hc_tx_packet_sent(dp
->dccps_hc_tx_ccid
, sk
, 0, len
);
278 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
281 dccp_pr_debug("packet discarded due to err=%d\n", err
);
287 int dccp_retransmit_skb(struct sock
*sk
, struct sk_buff
*skb
)
289 if (inet_csk(sk
)->icsk_af_ops
->rebuild_header(sk
) != 0)
290 return -EHOSTUNREACH
; /* Routing failure or similar. */
292 return dccp_transmit_skb(sk
, (skb_cloned(skb
) ?
293 pskb_copy(skb
, GFP_ATOMIC
):
294 skb_clone(skb
, GFP_ATOMIC
)));
297 struct sk_buff
*dccp_make_response(struct sock
*sk
, struct dst_entry
*dst
,
298 struct request_sock
*req
)
301 struct dccp_request_sock
*dreq
;
302 const u32 dccp_header_size
= sizeof(struct dccp_hdr
) +
303 sizeof(struct dccp_hdr_ext
) +
304 sizeof(struct dccp_hdr_response
);
305 struct sk_buff
*skb
= sock_wmalloc(sk
, sk
->sk_prot
->max_header
, 1,
310 /* Reserve space for headers. */
311 skb_reserve(skb
, sk
->sk_prot
->max_header
);
313 skb
->dst
= dst_clone(dst
);
315 dreq
= dccp_rsk(req
);
316 if (inet_rsk(req
)->acked
) /* increase ISS upon retransmission */
317 dccp_inc_seqno(&dreq
->dreq_iss
);
318 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_RESPONSE
;
319 DCCP_SKB_CB(skb
)->dccpd_seq
= dreq
->dreq_iss
;
321 if (dccp_insert_options_rsk(dreq
, skb
)) {
326 /* Build and checksum header */
327 dh
= dccp_zeroed_hdr(skb
, dccp_header_size
);
329 dh
->dccph_sport
= inet_sk(sk
)->sport
;
330 dh
->dccph_dport
= inet_rsk(req
)->rmt_port
;
331 dh
->dccph_doff
= (dccp_header_size
+
332 DCCP_SKB_CB(skb
)->dccpd_opt_len
) / 4;
333 dh
->dccph_type
= DCCP_PKT_RESPONSE
;
335 dccp_hdr_set_seq(dh
, dreq
->dreq_iss
);
336 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb
), dreq
->dreq_isr
);
337 dccp_hdr_response(skb
)->dccph_resp_service
= dreq
->dreq_service
;
339 dccp_csum_outgoing(skb
);
341 /* We use `acked' to remember that a Response was already sent. */
342 inet_rsk(req
)->acked
= 1;
343 DCCP_INC_STATS(DCCP_MIB_OUTSEGS
);
347 EXPORT_SYMBOL_GPL(dccp_make_response
);
349 /* answer offending packet in @rcv_skb with Reset from control socket @ctl */
350 struct sk_buff
*dccp_ctl_make_reset(struct socket
*ctl
, struct sk_buff
*rcv_skb
)
352 struct dccp_hdr
*rxdh
= dccp_hdr(rcv_skb
), *dh
;
353 struct dccp_skb_cb
*dcb
= DCCP_SKB_CB(rcv_skb
);
354 const u32 dccp_hdr_reset_len
= sizeof(struct dccp_hdr
) +
355 sizeof(struct dccp_hdr_ext
) +
356 sizeof(struct dccp_hdr_reset
);
357 struct dccp_hdr_reset
*dhr
;
360 skb
= alloc_skb(ctl
->sk
->sk_prot
->max_header
, GFP_ATOMIC
);
364 skb_reserve(skb
, ctl
->sk
->sk_prot
->max_header
);
366 /* Swap the send and the receive. */
367 dh
= dccp_zeroed_hdr(skb
, dccp_hdr_reset_len
);
368 dh
->dccph_type
= DCCP_PKT_RESET
;
369 dh
->dccph_sport
= rxdh
->dccph_dport
;
370 dh
->dccph_dport
= rxdh
->dccph_sport
;
371 dh
->dccph_doff
= dccp_hdr_reset_len
/ 4;
374 dhr
= dccp_hdr_reset(skb
);
375 dhr
->dccph_reset_code
= dcb
->dccpd_reset_code
;
377 switch (dcb
->dccpd_reset_code
) {
378 case DCCP_RESET_CODE_PACKET_ERROR
:
379 dhr
->dccph_reset_data
[0] = rxdh
->dccph_type
;
381 case DCCP_RESET_CODE_OPTION_ERROR
: /* fall through */
382 case DCCP_RESET_CODE_MANDATORY_ERROR
:
383 memcpy(dhr
->dccph_reset_data
, dcb
->dccpd_reset_data
, 3);
387 * From RFC 4340, 8.3.1:
388 * If P.ackno exists, set R.seqno := P.ackno + 1.
389 * Else set R.seqno := 0.
391 if (dcb
->dccpd_ack_seq
!= DCCP_PKT_WITHOUT_ACK_SEQ
)
392 dccp_hdr_set_seq(dh
, ADD48(dcb
->dccpd_ack_seq
, 1));
393 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb
), dcb
->dccpd_seq
);
395 dccp_csum_outgoing(skb
);
399 EXPORT_SYMBOL_GPL(dccp_ctl_make_reset
);
401 /* send Reset on established socket, to close or abort the connection */
402 int dccp_send_reset(struct sock
*sk
, enum dccp_reset_codes code
)
406 * FIXME: what if rebuild_header fails?
407 * Should we be doing a rebuild_header here?
409 int err
= inet_csk(sk
)->icsk_af_ops
->rebuild_header(sk
);
414 skb
= sock_wmalloc(sk
, sk
->sk_prot
->max_header
, 1, GFP_ATOMIC
);
418 /* Reserve space for headers and prepare control bits. */
419 skb_reserve(skb
, sk
->sk_prot
->max_header
);
420 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_RESET
;
421 DCCP_SKB_CB(skb
)->dccpd_reset_code
= code
;
423 return dccp_transmit_skb(sk
, skb
);
427 * Do all connect socket setups that can be done AF independent.
429 static inline void dccp_connect_init(struct sock
*sk
)
431 struct dccp_sock
*dp
= dccp_sk(sk
);
432 struct dst_entry
*dst
= __sk_dst_get(sk
);
433 struct inet_connection_sock
*icsk
= inet_csk(sk
);
436 sock_reset_flag(sk
, SOCK_DONE
);
438 dccp_sync_mss(sk
, dst_mtu(dst
));
441 * SWL and AWL are initially adjusted so that they are not less than
442 * the initial Sequence Numbers received and sent, respectively:
443 * SWL := max(GSR + 1 - floor(W/4), ISR),
444 * AWL := max(GSS - W' + 1, ISS).
445 * These adjustments MUST be applied only at the beginning of the
448 dccp_update_gss(sk
, dp
->dccps_iss
);
449 dccp_set_seqno(&dp
->dccps_awl
, max48(dp
->dccps_awl
, dp
->dccps_iss
));
451 /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
452 * initialized to S.ISS (sec. 8.5) */
453 dp
->dccps_gar
= dp
->dccps_iss
;
455 icsk
->icsk_retransmits
= 0;
458 int dccp_connect(struct sock
*sk
)
461 struct inet_connection_sock
*icsk
= inet_csk(sk
);
463 dccp_connect_init(sk
);
465 skb
= alloc_skb(sk
->sk_prot
->max_header
, sk
->sk_allocation
);
466 if (unlikely(skb
== NULL
))
469 /* Reserve space for headers. */
470 skb_reserve(skb
, sk
->sk_prot
->max_header
);
472 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_REQUEST
;
474 dccp_skb_entail(sk
, skb
);
475 dccp_transmit_skb(sk
, skb_clone(skb
, GFP_KERNEL
));
476 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS
);
478 /* Timer for repeating the REQUEST until an answer. */
479 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
480 icsk
->icsk_rto
, DCCP_RTO_MAX
);
484 EXPORT_SYMBOL_GPL(dccp_connect
);
486 void dccp_send_ack(struct sock
*sk
)
488 /* If we have been reset, we may not send again. */
489 if (sk
->sk_state
!= DCCP_CLOSED
) {
490 struct sk_buff
*skb
= alloc_skb(sk
->sk_prot
->max_header
,
494 inet_csk_schedule_ack(sk
);
495 inet_csk(sk
)->icsk_ack
.ato
= TCP_ATO_MIN
;
496 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
502 /* Reserve space for headers */
503 skb_reserve(skb
, sk
->sk_prot
->max_header
);
504 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_ACK
;
505 dccp_transmit_skb(sk
, skb
);
509 EXPORT_SYMBOL_GPL(dccp_send_ack
);
511 /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
512 void dccp_send_delayed_ack(struct sock
*sk
)
514 struct inet_connection_sock
*icsk
= inet_csk(sk
);
516 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
517 * with using 2s, and active senders also piggyback the ACK into a
518 * DATAACK packet, so this is really for quiescent senders.
520 unsigned long timeout
= jiffies
+ 2 * HZ
;
522 /* Use new timeout only if there wasn't a older one earlier. */
523 if (icsk
->icsk_ack
.pending
& ICSK_ACK_TIMER
) {
524 /* If delack timer was blocked or is about to expire,
527 * FIXME: check the "about to expire" part
529 if (icsk
->icsk_ack
.blocked
) {
534 if (!time_before(timeout
, icsk
->icsk_ack
.timeout
))
535 timeout
= icsk
->icsk_ack
.timeout
;
537 icsk
->icsk_ack
.pending
|= ICSK_ACK_SCHED
| ICSK_ACK_TIMER
;
538 icsk
->icsk_ack
.timeout
= timeout
;
539 sk_reset_timer(sk
, &icsk
->icsk_delack_timer
, timeout
);
542 void dccp_send_sync(struct sock
*sk
, const u64 ackno
,
543 const enum dccp_pkt_type pkt_type
)
546 * We are not putting this on the write queue, so
547 * dccp_transmit_skb() will set the ownership to this
550 struct sk_buff
*skb
= alloc_skb(sk
->sk_prot
->max_header
, GFP_ATOMIC
);
553 /* FIXME: how to make sure the sync is sent? */
554 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type
));
558 /* Reserve space for headers and prepare control bits. */
559 skb_reserve(skb
, sk
->sk_prot
->max_header
);
560 DCCP_SKB_CB(skb
)->dccpd_type
= pkt_type
;
561 DCCP_SKB_CB(skb
)->dccpd_ack_seq
= ackno
;
563 dccp_transmit_skb(sk
, skb
);
566 EXPORT_SYMBOL_GPL(dccp_send_sync
);
569 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
570 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
573 void dccp_send_close(struct sock
*sk
, const int active
)
575 struct dccp_sock
*dp
= dccp_sk(sk
);
577 const gfp_t prio
= active
? GFP_KERNEL
: GFP_ATOMIC
;
579 skb
= alloc_skb(sk
->sk_prot
->max_header
, prio
);
583 /* Reserve space for headers and prepare control bits. */
584 skb_reserve(skb
, sk
->sk_prot
->max_header
);
585 if (dp
->dccps_role
== DCCP_ROLE_SERVER
&& !dp
->dccps_server_timewait
)
586 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_CLOSEREQ
;
588 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_CLOSE
;
591 dccp_write_xmit(sk
, 1);
592 dccp_skb_entail(sk
, skb
);
593 dccp_transmit_skb(sk
, skb_clone(skb
, prio
));
595 * Retransmission timer for active-close: RFC 4340, 8.3 requires
596 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
597 * state can be left. The initial timeout is 2 RTTs.
598 * Since RTT measurement is done by the CCIDs, there is no easy
599 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
600 * is too low (200ms); we use a high value to avoid unnecessary
601 * retransmissions when the link RTT is > 0.2 seconds.
602 * FIXME: Let main module sample RTTs and use that instead.
604 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
605 DCCP_TIMEOUT_INIT
, DCCP_RTO_MAX
);
607 dccp_transmit_skb(sk
, skb
);