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 * Increment GSS here already in case the option code needs it.
58 * Update GSS for real only if option processing below succeeds.
60 dcb
->dccpd_seq
= ADD48(dp
->dccps_gss
, 1);
62 switch (dcb
->dccpd_type
) {
66 case DCCP_PKT_DATAACK
:
70 case DCCP_PKT_REQUEST
:
72 /* Use ISS on the first (non-retransmitted) Request. */
73 if (icsk
->icsk_retransmits
== 0)
74 dcb
->dccpd_seq
= dp
->dccps_iss
;
78 case DCCP_PKT_SYNCACK
:
79 ackno
= dcb
->dccpd_ack_seq
;
83 * Set owner/destructor: some skbs are allocated via
84 * alloc_skb (e.g. when retransmission may happen).
85 * Only Data, DataAck, and Reset packets should come
86 * through here with skb->sk set.
89 skb_set_owner_w(skb
, sk
);
93 if (dccp_insert_options(sk
, skb
)) {
99 /* Build DCCP header and checksum it. */
100 dh
= dccp_zeroed_hdr(skb
, dccp_header_size
);
101 dh
->dccph_type
= dcb
->dccpd_type
;
102 dh
->dccph_sport
= inet
->inet_sport
;
103 dh
->dccph_dport
= inet
->inet_dport
;
104 dh
->dccph_doff
= (dccp_header_size
+ dcb
->dccpd_opt_len
) / 4;
105 dh
->dccph_ccval
= dcb
->dccpd_ccval
;
106 dh
->dccph_cscov
= dp
->dccps_pcslen
;
107 /* XXX For now we're using only 48 bits sequence numbers */
110 dccp_update_gss(sk
, dcb
->dccpd_seq
);
111 dccp_hdr_set_seq(dh
, dp
->dccps_gss
);
113 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb
), ackno
);
115 switch (dcb
->dccpd_type
) {
116 case DCCP_PKT_REQUEST
:
117 dccp_hdr_request(skb
)->dccph_req_service
=
120 * Limit Ack window to ISS <= P.ackno <= GSS, so that
121 * only Responses to Requests we sent are considered.
123 dp
->dccps_awl
= dp
->dccps_iss
;
126 dccp_hdr_reset(skb
)->dccph_reset_code
=
127 dcb
->dccpd_reset_code
;
131 icsk
->icsk_af_ops
->send_check(sk
, 0, skb
);
134 dccp_event_ack_sent(sk
);
136 DCCP_INC_STATS(DCCP_MIB_OUTSEGS
);
138 err
= icsk
->icsk_af_ops
->queue_xmit(skb
, 0);
139 return net_xmit_eval(err
);
145 * dccp_determine_ccmps - Find out about CCID-specfic packet-size limits
146 * We only consider the HC-sender CCID for setting the CCMPS (RFC 4340, 14.),
147 * since the RX CCID is restricted to feedback packets (Acks), which are small
148 * in comparison with the data traffic. A value of 0 means "no current CCMPS".
150 static u32
dccp_determine_ccmps(const struct dccp_sock
*dp
)
152 const struct ccid
*tx_ccid
= dp
->dccps_hc_tx_ccid
;
154 if (tx_ccid
== NULL
|| tx_ccid
->ccid_ops
== NULL
)
156 return tx_ccid
->ccid_ops
->ccid_ccmps
;
159 unsigned int dccp_sync_mss(struct sock
*sk
, u32 pmtu
)
161 struct inet_connection_sock
*icsk
= inet_csk(sk
);
162 struct dccp_sock
*dp
= dccp_sk(sk
);
163 u32 ccmps
= dccp_determine_ccmps(dp
);
164 u32 cur_mps
= ccmps
? min(pmtu
, ccmps
) : pmtu
;
166 /* Account for header lengths and IPv4/v6 option overhead */
167 cur_mps
-= (icsk
->icsk_af_ops
->net_header_len
+ icsk
->icsk_ext_hdr_len
+
168 sizeof(struct dccp_hdr
) + sizeof(struct dccp_hdr_ext
));
171 * Leave enough headroom for common DCCP header options.
172 * This only considers options which may appear on DCCP-Data packets, as
173 * per table 3 in RFC 4340, 5.8. When running out of space for other
174 * options (eg. Ack Vector which can take up to 255 bytes), it is better
175 * to schedule a separate Ack. Thus we leave headroom for the following:
176 * - 1 byte for Slow Receiver (11.6)
177 * - 6 bytes for Timestamp (13.1)
178 * - 10 bytes for Timestamp Echo (13.3)
179 * - 8 bytes for NDP count (7.7, when activated)
180 * - 6 bytes for Data Checksum (9.3)
181 * - %DCCPAV_MIN_OPTLEN bytes for Ack Vector size (11.4, when enabled)
183 cur_mps
-= roundup(1 + 6 + 10 + dp
->dccps_send_ndp_count
* 8 + 6 +
184 (dp
->dccps_hc_rx_ackvec
? DCCPAV_MIN_OPTLEN
: 0), 4);
186 /* And store cached results */
187 icsk
->icsk_pmtu_cookie
= pmtu
;
188 dp
->dccps_mss_cache
= cur_mps
;
193 EXPORT_SYMBOL_GPL(dccp_sync_mss
);
195 void dccp_write_space(struct sock
*sk
)
197 read_lock(&sk
->sk_callback_lock
);
199 if (sk_has_sleeper(sk
))
200 wake_up_interruptible(sk
->sk_sleep
);
201 /* Should agree with poll, otherwise some programs break */
202 if (sock_writeable(sk
))
203 sk_wake_async(sk
, SOCK_WAKE_SPACE
, POLL_OUT
);
205 read_unlock(&sk
->sk_callback_lock
);
209 * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
210 * @sk: socket to wait for
211 * @skb: current skb to pass on for waiting
212 * @delay: sleep timeout in milliseconds (> 0)
213 * This function is called by default when the socket is closed, and
214 * when a non-zero linger time is set on the socket. For consistency
216 static int dccp_wait_for_ccid(struct sock
*sk
, struct sk_buff
*skb
, int delay
)
218 struct dccp_sock
*dp
= dccp_sk(sk
);
220 unsigned long jiffdelay
;
224 dccp_pr_debug("delayed send by %d msec\n", delay
);
225 jiffdelay
= msecs_to_jiffies(delay
);
227 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
229 sk
->sk_write_pending
++;
231 schedule_timeout(jiffdelay
);
233 sk
->sk_write_pending
--;
237 if (signal_pending(current
))
240 rc
= ccid_hc_tx_send_packet(dp
->dccps_hc_tx_ccid
, sk
, skb
);
241 } while ((delay
= rc
) > 0);
243 finish_wait(sk
->sk_sleep
, &wait
);
254 void dccp_write_xmit(struct sock
*sk
, int block
)
256 struct dccp_sock
*dp
= dccp_sk(sk
);
259 while ((skb
= skb_peek(&sk
->sk_write_queue
))) {
260 int err
= ccid_hc_tx_send_packet(dp
->dccps_hc_tx_ccid
, sk
, skb
);
264 sk_reset_timer(sk
, &dp
->dccps_xmit_timer
,
265 msecs_to_jiffies(err
)+jiffies
);
268 err
= dccp_wait_for_ccid(sk
, skb
, err
);
269 if (err
&& err
!= -EINTR
)
270 DCCP_BUG("err=%d after dccp_wait_for_ccid", err
);
273 skb_dequeue(&sk
->sk_write_queue
);
275 struct dccp_skb_cb
*dcb
= DCCP_SKB_CB(skb
);
276 const int len
= skb
->len
;
278 if (sk
->sk_state
== DCCP_PARTOPEN
) {
279 const u32 cur_mps
= dp
->dccps_mss_cache
- DCCP_FEATNEG_OVERHEAD
;
281 * See 8.1.5 - Handshake Completion.
283 * For robustness we resend Confirm options until the client has
284 * entered OPEN. During the initial feature negotiation, the MPS
285 * is smaller than usual, reduced by the Change/Confirm options.
287 if (!list_empty(&dp
->dccps_featneg
) && len
> cur_mps
) {
288 DCCP_WARN("Payload too large (%d) for featneg.\n", len
);
290 dccp_feat_list_purge(&dp
->dccps_featneg
);
293 inet_csk_schedule_ack(sk
);
294 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
295 inet_csk(sk
)->icsk_rto
,
297 dcb
->dccpd_type
= DCCP_PKT_DATAACK
;
298 } else if (dccp_ack_pending(sk
))
299 dcb
->dccpd_type
= DCCP_PKT_DATAACK
;
301 dcb
->dccpd_type
= DCCP_PKT_DATA
;
303 err
= dccp_transmit_skb(sk
, skb
);
304 ccid_hc_tx_packet_sent(dp
->dccps_hc_tx_ccid
, sk
, 0, len
);
306 DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
309 dccp_pr_debug("packet discarded due to err=%d\n", err
);
316 * dccp_retransmit_skb - Retransmit Request, Close, or CloseReq packets
317 * There are only four retransmittable packet types in DCCP:
318 * - Request in client-REQUEST state (sec. 8.1.1),
319 * - CloseReq in server-CLOSEREQ state (sec. 8.3),
320 * - Close in node-CLOSING state (sec. 8.3),
321 * - Acks in client-PARTOPEN state (sec. 8.1.5, handled by dccp_delack_timer()).
322 * This function expects sk->sk_send_head to contain the original skb.
324 int dccp_retransmit_skb(struct sock
*sk
)
326 WARN_ON(sk
->sk_send_head
== NULL
);
328 if (inet_csk(sk
)->icsk_af_ops
->rebuild_header(sk
) != 0)
329 return -EHOSTUNREACH
; /* Routing failure or similar. */
331 /* this count is used to distinguish original and retransmitted skb */
332 inet_csk(sk
)->icsk_retransmits
++;
334 return dccp_transmit_skb(sk
, skb_clone(sk
->sk_send_head
, GFP_ATOMIC
));
337 struct sk_buff
*dccp_make_response(struct sock
*sk
, struct dst_entry
*dst
,
338 struct request_sock
*req
)
341 struct dccp_request_sock
*dreq
;
342 const u32 dccp_header_size
= sizeof(struct dccp_hdr
) +
343 sizeof(struct dccp_hdr_ext
) +
344 sizeof(struct dccp_hdr_response
);
345 struct sk_buff
*skb
= sock_wmalloc(sk
, sk
->sk_prot
->max_header
, 1,
350 /* Reserve space for headers. */
351 skb_reserve(skb
, sk
->sk_prot
->max_header
);
353 skb_dst_set(skb
, dst_clone(dst
));
355 dreq
= dccp_rsk(req
);
356 if (inet_rsk(req
)->acked
) /* increase ISS upon retransmission */
357 dccp_inc_seqno(&dreq
->dreq_iss
);
358 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_RESPONSE
;
359 DCCP_SKB_CB(skb
)->dccpd_seq
= dreq
->dreq_iss
;
361 /* Resolve feature dependencies resulting from choice of CCID */
362 if (dccp_feat_server_ccid_dependencies(dreq
))
363 goto response_failed
;
365 if (dccp_insert_options_rsk(dreq
, skb
))
366 goto response_failed
;
368 /* Build and checksum header */
369 dh
= dccp_zeroed_hdr(skb
, dccp_header_size
);
371 dh
->dccph_sport
= inet_rsk(req
)->loc_port
;
372 dh
->dccph_dport
= inet_rsk(req
)->rmt_port
;
373 dh
->dccph_doff
= (dccp_header_size
+
374 DCCP_SKB_CB(skb
)->dccpd_opt_len
) / 4;
375 dh
->dccph_type
= DCCP_PKT_RESPONSE
;
377 dccp_hdr_set_seq(dh
, dreq
->dreq_iss
);
378 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb
), dreq
->dreq_isr
);
379 dccp_hdr_response(skb
)->dccph_resp_service
= dreq
->dreq_service
;
381 dccp_csum_outgoing(skb
);
383 /* We use `acked' to remember that a Response was already sent. */
384 inet_rsk(req
)->acked
= 1;
385 DCCP_INC_STATS(DCCP_MIB_OUTSEGS
);
392 EXPORT_SYMBOL_GPL(dccp_make_response
);
394 /* answer offending packet in @rcv_skb with Reset from control socket @ctl */
395 struct sk_buff
*dccp_ctl_make_reset(struct sock
*sk
, struct sk_buff
*rcv_skb
)
397 struct dccp_hdr
*rxdh
= dccp_hdr(rcv_skb
), *dh
;
398 struct dccp_skb_cb
*dcb
= DCCP_SKB_CB(rcv_skb
);
399 const u32 dccp_hdr_reset_len
= sizeof(struct dccp_hdr
) +
400 sizeof(struct dccp_hdr_ext
) +
401 sizeof(struct dccp_hdr_reset
);
402 struct dccp_hdr_reset
*dhr
;
405 skb
= alloc_skb(sk
->sk_prot
->max_header
, GFP_ATOMIC
);
409 skb_reserve(skb
, sk
->sk_prot
->max_header
);
411 /* Swap the send and the receive. */
412 dh
= dccp_zeroed_hdr(skb
, dccp_hdr_reset_len
);
413 dh
->dccph_type
= DCCP_PKT_RESET
;
414 dh
->dccph_sport
= rxdh
->dccph_dport
;
415 dh
->dccph_dport
= rxdh
->dccph_sport
;
416 dh
->dccph_doff
= dccp_hdr_reset_len
/ 4;
419 dhr
= dccp_hdr_reset(skb
);
420 dhr
->dccph_reset_code
= dcb
->dccpd_reset_code
;
422 switch (dcb
->dccpd_reset_code
) {
423 case DCCP_RESET_CODE_PACKET_ERROR
:
424 dhr
->dccph_reset_data
[0] = rxdh
->dccph_type
;
426 case DCCP_RESET_CODE_OPTION_ERROR
: /* fall through */
427 case DCCP_RESET_CODE_MANDATORY_ERROR
:
428 memcpy(dhr
->dccph_reset_data
, dcb
->dccpd_reset_data
, 3);
432 * From RFC 4340, 8.3.1:
433 * If P.ackno exists, set R.seqno := P.ackno + 1.
434 * Else set R.seqno := 0.
436 if (dcb
->dccpd_ack_seq
!= DCCP_PKT_WITHOUT_ACK_SEQ
)
437 dccp_hdr_set_seq(dh
, ADD48(dcb
->dccpd_ack_seq
, 1));
438 dccp_hdr_set_ack(dccp_hdr_ack_bits(skb
), dcb
->dccpd_seq
);
440 dccp_csum_outgoing(skb
);
444 EXPORT_SYMBOL_GPL(dccp_ctl_make_reset
);
446 /* send Reset on established socket, to close or abort the connection */
447 int dccp_send_reset(struct sock
*sk
, enum dccp_reset_codes code
)
451 * FIXME: what if rebuild_header fails?
452 * Should we be doing a rebuild_header here?
454 int err
= inet_csk(sk
)->icsk_af_ops
->rebuild_header(sk
);
459 skb
= sock_wmalloc(sk
, sk
->sk_prot
->max_header
, 1, GFP_ATOMIC
);
463 /* Reserve space for headers and prepare control bits. */
464 skb_reserve(skb
, sk
->sk_prot
->max_header
);
465 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_RESET
;
466 DCCP_SKB_CB(skb
)->dccpd_reset_code
= code
;
468 return dccp_transmit_skb(sk
, skb
);
472 * Do all connect socket setups that can be done AF independent.
474 static inline void dccp_connect_init(struct sock
*sk
)
476 struct dccp_sock
*dp
= dccp_sk(sk
);
477 struct dst_entry
*dst
= __sk_dst_get(sk
);
478 struct inet_connection_sock
*icsk
= inet_csk(sk
);
481 sock_reset_flag(sk
, SOCK_DONE
);
483 dccp_sync_mss(sk
, dst_mtu(dst
));
485 /* Initialise GAR as per 8.5; AWL/AWH are set in dccp_transmit_skb() */
486 dp
->dccps_gar
= dp
->dccps_iss
;
488 icsk
->icsk_retransmits
= 0;
491 int dccp_connect(struct sock
*sk
)
494 struct inet_connection_sock
*icsk
= inet_csk(sk
);
496 /* do not connect if feature negotiation setup fails */
497 if (dccp_feat_finalise_settings(dccp_sk(sk
)))
500 dccp_connect_init(sk
);
502 skb
= alloc_skb(sk
->sk_prot
->max_header
, sk
->sk_allocation
);
503 if (unlikely(skb
== NULL
))
506 /* Reserve space for headers. */
507 skb_reserve(skb
, sk
->sk_prot
->max_header
);
509 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_REQUEST
;
511 dccp_skb_entail(sk
, skb
);
512 dccp_transmit_skb(sk
, skb_clone(skb
, GFP_KERNEL
));
513 DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS
);
515 /* Timer for repeating the REQUEST until an answer. */
516 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
517 icsk
->icsk_rto
, DCCP_RTO_MAX
);
521 EXPORT_SYMBOL_GPL(dccp_connect
);
523 void dccp_send_ack(struct sock
*sk
)
525 /* If we have been reset, we may not send again. */
526 if (sk
->sk_state
!= DCCP_CLOSED
) {
527 struct sk_buff
*skb
= alloc_skb(sk
->sk_prot
->max_header
,
531 inet_csk_schedule_ack(sk
);
532 inet_csk(sk
)->icsk_ack
.ato
= TCP_ATO_MIN
;
533 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
539 /* Reserve space for headers */
540 skb_reserve(skb
, sk
->sk_prot
->max_header
);
541 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_ACK
;
542 dccp_transmit_skb(sk
, skb
);
546 EXPORT_SYMBOL_GPL(dccp_send_ack
);
549 /* FIXME: Is this still necessary (11.3) - currently nowhere used by DCCP. */
550 void dccp_send_delayed_ack(struct sock
*sk
)
552 struct inet_connection_sock
*icsk
= inet_csk(sk
);
554 * FIXME: tune this timer. elapsed time fixes the skew, so no problem
555 * with using 2s, and active senders also piggyback the ACK into a
556 * DATAACK packet, so this is really for quiescent senders.
558 unsigned long timeout
= jiffies
+ 2 * HZ
;
560 /* Use new timeout only if there wasn't a older one earlier. */
561 if (icsk
->icsk_ack
.pending
& ICSK_ACK_TIMER
) {
562 /* If delack timer was blocked or is about to expire,
565 * FIXME: check the "about to expire" part
567 if (icsk
->icsk_ack
.blocked
) {
572 if (!time_before(timeout
, icsk
->icsk_ack
.timeout
))
573 timeout
= icsk
->icsk_ack
.timeout
;
575 icsk
->icsk_ack
.pending
|= ICSK_ACK_SCHED
| ICSK_ACK_TIMER
;
576 icsk
->icsk_ack
.timeout
= timeout
;
577 sk_reset_timer(sk
, &icsk
->icsk_delack_timer
, timeout
);
581 void dccp_send_sync(struct sock
*sk
, const u64 ackno
,
582 const enum dccp_pkt_type pkt_type
)
585 * We are not putting this on the write queue, so
586 * dccp_transmit_skb() will set the ownership to this
589 struct sk_buff
*skb
= alloc_skb(sk
->sk_prot
->max_header
, GFP_ATOMIC
);
592 /* FIXME: how to make sure the sync is sent? */
593 DCCP_CRIT("could not send %s", dccp_packet_name(pkt_type
));
597 /* Reserve space for headers and prepare control bits. */
598 skb_reserve(skb
, sk
->sk_prot
->max_header
);
599 DCCP_SKB_CB(skb
)->dccpd_type
= pkt_type
;
600 DCCP_SKB_CB(skb
)->dccpd_ack_seq
= ackno
;
602 dccp_transmit_skb(sk
, skb
);
605 EXPORT_SYMBOL_GPL(dccp_send_sync
);
608 * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
609 * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
612 void dccp_send_close(struct sock
*sk
, const int active
)
614 struct dccp_sock
*dp
= dccp_sk(sk
);
616 const gfp_t prio
= active
? GFP_KERNEL
: GFP_ATOMIC
;
618 skb
= alloc_skb(sk
->sk_prot
->max_header
, prio
);
622 /* Reserve space for headers and prepare control bits. */
623 skb_reserve(skb
, sk
->sk_prot
->max_header
);
624 if (dp
->dccps_role
== DCCP_ROLE_SERVER
&& !dp
->dccps_server_timewait
)
625 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_CLOSEREQ
;
627 DCCP_SKB_CB(skb
)->dccpd_type
= DCCP_PKT_CLOSE
;
630 dccp_write_xmit(sk
, 1);
631 dccp_skb_entail(sk
, skb
);
632 dccp_transmit_skb(sk
, skb_clone(skb
, prio
));
634 * Retransmission timer for active-close: RFC 4340, 8.3 requires
635 * to retransmit the Close/CloseReq until the CLOSING/CLOSEREQ
636 * state can be left. The initial timeout is 2 RTTs.
637 * Since RTT measurement is done by the CCIDs, there is no easy
638 * way to get an RTT sample. The fallback RTT from RFC 4340, 3.4
639 * is too low (200ms); we use a high value to avoid unnecessary
640 * retransmissions when the link RTT is > 0.2 seconds.
641 * FIXME: Let main module sample RTTs and use that instead.
643 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
644 DCCP_TIMEOUT_INIT
, DCCP_RTO_MAX
);
646 dccp_transmit_skb(sk
, skb
);