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>
22 static void dccp_fin(struct sock
*sk
, struct sk_buff
*skb
)
24 sk
->sk_shutdown
|= RCV_SHUTDOWN
;
25 sock_set_flag(sk
, SOCK_DONE
);
26 __skb_pull(skb
, dccp_hdr(skb
)->dccph_doff
* 4);
27 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
28 skb_set_owner_r(skb
, sk
);
29 sk
->sk_data_ready(sk
, 0);
32 static void dccp_rcv_close(struct sock
*sk
, struct sk_buff
*skb
)
34 switch (sk
->sk_state
) {
37 dccp_v4_send_reset(sk
, DCCP_RESET_CODE_CLOSED
);
39 dccp_set_state(sk
, DCCP_CLOSED
);
44 static void dccp_rcv_closereq(struct sock
*sk
, struct sk_buff
*skb
)
47 * Step 7: Check for unexpected packet types
48 * If (S.is_server and P.type == CloseReq)
49 * Send Sync packet acknowledging P.seqno
50 * Drop packet and return
52 if (dccp_sk(sk
)->dccps_role
!= DCCP_ROLE_CLIENT
) {
53 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
, DCCP_PKT_SYNC
);
57 switch (sk
->sk_state
) {
60 dccp_set_state(sk
, DCCP_CLOSING
);
66 static inline void dccp_event_ack_recv(struct sock
*sk
, struct sk_buff
*skb
)
68 struct dccp_sock
*dp
= dccp_sk(sk
);
70 if (dp
->dccps_options
.dccpo_send_ack_vector
)
71 dccp_ackpkts_check_rcv_ackno(dp
->dccps_hc_rx_ackpkts
, sk
,
72 DCCP_SKB_CB(skb
)->dccpd_ack_seq
);
75 static int dccp_check_seqno(struct sock
*sk
, struct sk_buff
*skb
)
77 const struct dccp_hdr
*dh
= dccp_hdr(skb
);
78 struct dccp_sock
*dp
= dccp_sk(sk
);
82 * Step 5: Prepare sequence numbers for Sync
83 * If P.type == Sync or P.type == SyncAck,
84 * If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL,
85 * / * P is valid, so update sequence number variables
86 * accordingly. After this update, P will pass the tests
87 * in Step 6. A SyncAck is generated if necessary in
89 * Update S.GSR, S.SWL, S.SWH
91 * Drop packet and return
93 if (dh
->dccph_type
== DCCP_PKT_SYNC
||
94 dh
->dccph_type
== DCCP_PKT_SYNCACK
) {
95 if (between48(DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
96 dp
->dccps_awl
, dp
->dccps_awh
) &&
97 !before48(DCCP_SKB_CB(skb
)->dccpd_seq
, dp
->dccps_swl
))
98 dccp_update_gsr(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
);
104 * Step 6: Check sequence numbers
105 * Let LSWL = S.SWL and LAWL = S.AWL
106 * If P.type == CloseReq or P.type == Close or P.type == Reset,
107 * LSWL := S.GSR + 1, LAWL := S.GAR
108 * If LSWL <= P.seqno <= S.SWH
109 * and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH),
110 * Update S.GSR, S.SWL, S.SWH
114 * Send Sync packet acknowledging P.seqno
115 * Drop packet and return
117 lswl
= dp
->dccps_swl
;
118 lawl
= dp
->dccps_awl
;
120 if (dh
->dccph_type
== DCCP_PKT_CLOSEREQ
||
121 dh
->dccph_type
== DCCP_PKT_CLOSE
||
122 dh
->dccph_type
== DCCP_PKT_RESET
) {
123 lswl
= dp
->dccps_gsr
;
124 dccp_inc_seqno(&lswl
);
125 lawl
= dp
->dccps_gar
;
128 if (between48(DCCP_SKB_CB(skb
)->dccpd_seq
, lswl
, dp
->dccps_swh
) &&
129 (DCCP_SKB_CB(skb
)->dccpd_ack_seq
== DCCP_PKT_WITHOUT_ACK_SEQ
||
130 between48(DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
131 lawl
, dp
->dccps_awh
))) {
132 dccp_update_gsr(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
);
134 if (dh
->dccph_type
!= DCCP_PKT_SYNC
&&
135 (DCCP_SKB_CB(skb
)->dccpd_ack_seq
!=
136 DCCP_PKT_WITHOUT_ACK_SEQ
))
137 dp
->dccps_gar
= DCCP_SKB_CB(skb
)->dccpd_ack_seq
;
139 LIMIT_NETDEBUG(KERN_WARNING
"DCCP: Step 6 failed, "
140 "sending SYNC...\n");
141 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
, DCCP_PKT_SYNC
);
148 int dccp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
149 const struct dccp_hdr
*dh
, const unsigned len
)
151 struct dccp_sock
*dp
= dccp_sk(sk
);
153 if (dccp_check_seqno(sk
, skb
))
156 if (dccp_parse_options(sk
, skb
))
159 if (DCCP_SKB_CB(skb
)->dccpd_ack_seq
!= DCCP_PKT_WITHOUT_ACK_SEQ
)
160 dccp_event_ack_recv(sk
, skb
);
163 * FIXME: check ECN to see if we should use
164 * DCCP_ACKPKTS_STATE_ECN_MARKED
166 if (dp
->dccps_options
.dccpo_send_ack_vector
) {
167 struct dccp_ackpkts
*ap
= dp
->dccps_hc_rx_ackpkts
;
169 if (dccp_ackpkts_add(dp
->dccps_hc_rx_ackpkts
,
170 DCCP_SKB_CB(skb
)->dccpd_seq
,
171 DCCP_ACKPKTS_STATE_RECEIVED
)) {
172 LIMIT_NETDEBUG(KERN_WARNING
"DCCP: acknowledgeable "
173 "packets buffer full!\n");
174 ap
->dccpap_ack_seqno
= DCCP_MAX_SEQNO
+ 1;
175 inet_csk_schedule_ack(sk
);
176 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
183 * FIXME: this activation is probably wrong, have to study more
184 * TCP delack machinery and how it fits into DCCP draft, but
185 * for now it kinda "works" 8)
187 if (!inet_csk_ack_scheduled(sk
)) {
188 inet_csk_schedule_ack(sk
);
189 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
, 5 * HZ
,
194 ccid_hc_rx_packet_recv(dp
->dccps_hc_rx_ccid
, sk
, skb
);
195 ccid_hc_tx_packet_recv(dp
->dccps_hc_tx_ccid
, sk
, skb
);
197 switch (dccp_hdr(skb
)->dccph_type
) {
198 case DCCP_PKT_DATAACK
:
201 * FIXME: check if sk_receive_queue is full, schedule DATA_DROPPED
204 __skb_pull(skb
, dh
->dccph_doff
* 4);
205 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
206 skb_set_owner_r(skb
, sk
);
207 sk
->sk_data_ready(sk
, 0);
213 * Step 9: Process Reset
214 * If P.type == Reset,
215 * Tear down connection
216 * S.state := TIMEWAIT
218 * Drop packet and return
221 dccp_time_wait(sk
, DCCP_TIME_WAIT
, 0);
223 case DCCP_PKT_CLOSEREQ
:
224 dccp_rcv_closereq(sk
, skb
);
227 dccp_rcv_close(sk
, skb
);
229 case DCCP_PKT_REQUEST
:
231 * or (S.is_server and P.type == Response)
232 * or (S.is_client and P.type == Request)
233 * or (S.state >= OPEN and P.type == Request
234 * and P.seqno >= S.OSR)
235 * or (S.state >= OPEN and P.type == Response
236 * and P.seqno >= S.OSR)
237 * or (S.state == RESPOND and P.type == Data),
238 * Send Sync packet acknowledging P.seqno
239 * Drop packet and return
241 if (dp
->dccps_role
!= DCCP_ROLE_LISTEN
)
244 case DCCP_PKT_RESPONSE
:
245 if (dp
->dccps_role
!= DCCP_ROLE_CLIENT
)
248 if (!before48(DCCP_SKB_CB(skb
)->dccpd_seq
, dp
->dccps_osr
)) {
250 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
,
255 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
,
260 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets
261 * MAY have non-zero-length application data areas, whose
262 * contents * receivers MUST ignore.
267 DCCP_INC_STATS_BH(DCCP_MIB_INERRS
);
273 static int dccp_rcv_request_sent_state_process(struct sock
*sk
,
275 const struct dccp_hdr
*dh
,
279 * Step 4: Prepare sequence numbers in REQUEST
280 * If S.state == REQUEST,
281 * If (P.type == Response or P.type == Reset)
282 * and S.AWL <= P.ackno <= S.AWH,
283 * / * Set sequence number variables corresponding to the
284 * other endpoint, so P will pass the tests in Step 6 * /
285 * Set S.GSR, S.ISR, S.SWL, S.SWH
286 * / * Response processing continues in Step 10; Reset
287 * processing continues in Step 9 * /
289 if (dh
->dccph_type
== DCCP_PKT_RESPONSE
) {
290 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
291 struct dccp_sock
*dp
= dccp_sk(sk
);
293 /* Stop the REQUEST timer */
294 inet_csk_clear_xmit_timer(sk
, ICSK_TIME_RETRANS
);
295 BUG_TRAP(sk
->sk_send_head
!= NULL
);
296 __kfree_skb(sk
->sk_send_head
);
297 sk
->sk_send_head
= NULL
;
299 if (!between48(DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
300 dp
->dccps_awl
, dp
->dccps_awh
)) {
301 dccp_pr_debug("invalid ackno: S.AWL=%llu, "
302 "P.ackno=%llu, S.AWH=%llu \n",
303 (unsigned long long)dp
->dccps_awl
,
304 (unsigned long long)DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
305 (unsigned long long)dp
->dccps_awh
);
306 goto out_invalid_packet
;
309 dp
->dccps_isr
= DCCP_SKB_CB(skb
)->dccpd_seq
;
310 dccp_update_gsr(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
);
312 if (ccid_hc_rx_init(dp
->dccps_hc_rx_ccid
, sk
) != 0 ||
313 ccid_hc_tx_init(dp
->dccps_hc_tx_ccid
, sk
) != 0) {
314 ccid_hc_rx_exit(dp
->dccps_hc_rx_ccid
, sk
);
315 ccid_hc_tx_exit(dp
->dccps_hc_tx_ccid
, sk
);
316 /* FIXME: send appropriate RESET code */
317 goto out_invalid_packet
;
320 dccp_sync_mss(sk
, dp
->dccps_pmtu_cookie
);
323 * Step 10: Process REQUEST state (second part)
324 * If S.state == REQUEST,
325 * / * If we get here, P is a valid Response from the
326 * server (see Step 4), and we should move to
327 * PARTOPEN state. PARTOPEN means send an Ack,
328 * don't send Data packets, retransmit Acks
329 * periodically, and always include any Init Cookie
330 * from the Response * /
331 * S.state := PARTOPEN
333 * Continue with S.state == PARTOPEN
334 * / * Step 12 will send the Ack completing the
335 * three-way handshake * /
337 dccp_set_state(sk
, DCCP_PARTOPEN
);
339 /* Make sure socket is routed, for correct metrics. */
340 inet_sk_rebuild_header(sk
);
342 if (!sock_flag(sk
, SOCK_DEAD
)) {
343 sk
->sk_state_change(sk
);
344 sk_wake_async(sk
, 0, POLL_OUT
);
347 if (sk
->sk_write_pending
|| icsk
->icsk_ack
.pingpong
||
348 icsk
->icsk_accept_queue
.rskq_defer_accept
) {
349 /* Save one ACK. Data will be ready after
350 * several ticks, if write_pending is set.
352 * It may be deleted, but with this feature tcpdumps
353 * look so _wonderfully_ clever, that I was not able
354 * to stand against the temptation 8) --ANK
357 * OK, in DCCP we can as well do a similar trick, its
358 * even in the draft, but there is no need for us to
359 * schedule an ack here, as dccp_sendmsg does this for
360 * us, also stated in the draft. -acme
370 return 1; /* dccp_v4_do_rcv will send a reset, but...
371 FIXME: the reset code should be
372 DCCP_RESET_CODE_PACKET_ERROR */
375 static int dccp_rcv_respond_partopen_state_process(struct sock
*sk
,
377 const struct dccp_hdr
*dh
,
382 switch (dh
->dccph_type
) {
384 inet_csk_clear_xmit_timer(sk
, ICSK_TIME_DACK
);
386 case DCCP_PKT_DATAACK
:
389 * FIXME: we should be reseting the PARTOPEN (DELACK) timer
390 * here but only if we haven't used the DELACK timer for
391 * something else, like sending a delayed ack for a TIMESTAMP
392 * echo, etc, for now were not clearing it, sending an extra
393 * ACK when there is nothing else to do in DELACK is not a big
397 /* Stop the PARTOPEN timer */
398 if (sk
->sk_state
== DCCP_PARTOPEN
)
399 inet_csk_clear_xmit_timer(sk
, ICSK_TIME_DACK
);
401 dccp_sk(sk
)->dccps_osr
= DCCP_SKB_CB(skb
)->dccpd_seq
;
402 dccp_set_state(sk
, DCCP_OPEN
);
404 if (dh
->dccph_type
== DCCP_PKT_DATAACK
) {
405 dccp_rcv_established(sk
, skb
, dh
, len
);
406 queued
= 1; /* packet was queued
407 (by dccp_rcv_established) */
415 int dccp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
416 struct dccp_hdr
*dh
, unsigned len
)
418 struct dccp_sock
*dp
= dccp_sk(sk
);
419 const int old_state
= sk
->sk_state
;
423 * Step 3: Process LISTEN state
424 * (Continuing from dccp_v4_do_rcv and dccp_v6_do_rcv)
426 * If S.state == LISTEN,
427 * If P.type == Request or P contains a valid Init Cookie
429 * * Must scan the packet's options to check for an Init
430 * Cookie. Only the Init Cookie is processed here,
431 * however; other options are processed in Step 8. This
432 * scan need only be performed if the endpoint uses Init
434 * * Generate a new socket and switch to that socket *
435 * Set S := new socket for this port pair
437 * Choose S.ISS (initial seqno) or set from Init Cookie
438 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
439 * Continue with S.state == RESPOND
440 * * A Response packet will be generated in Step 11 *
442 * Generate Reset(No Connection) unless P.type == Reset
443 * Drop packet and return
445 * NOTE: the check for the packet types is done in
446 * dccp_rcv_state_process
448 if (sk
->sk_state
== DCCP_LISTEN
) {
449 if (dh
->dccph_type
== DCCP_PKT_REQUEST
) {
450 if (dccp_v4_conn_request(sk
, skb
) < 0)
453 /* FIXME: do congestion control initialization */
456 if (dh
->dccph_type
== DCCP_PKT_RESET
)
459 /* Caller (dccp_v4_do_rcv) will send Reset(No Connection)*/
463 if (sk
->sk_state
!= DCCP_REQUESTING
) {
464 if (dccp_check_seqno(sk
, skb
))
468 * Step 8: Process options and mark acknowledgeable
470 if (dccp_parse_options(sk
, skb
))
473 if (DCCP_SKB_CB(skb
)->dccpd_ack_seq
!=
474 DCCP_PKT_WITHOUT_ACK_SEQ
)
475 dccp_event_ack_recv(sk
, skb
);
477 ccid_hc_rx_packet_recv(dp
->dccps_hc_rx_ccid
, sk
, skb
);
478 ccid_hc_tx_packet_recv(dp
->dccps_hc_tx_ccid
, sk
, skb
);
481 * FIXME: check ECN to see if we should use
482 * DCCP_ACKPKTS_STATE_ECN_MARKED
484 if (dp
->dccps_options
.dccpo_send_ack_vector
) {
485 if (dccp_ackpkts_add(dp
->dccps_hc_rx_ackpkts
,
486 DCCP_SKB_CB(skb
)->dccpd_seq
,
487 DCCP_ACKPKTS_STATE_RECEIVED
))
490 * FIXME: this activation is probably wrong, have to
491 * study more TCP delack machinery and how it fits into
492 * DCCP draft, but for now it kinda "works" 8)
494 if ((dp
->dccps_hc_rx_ackpkts
->dccpap_ack_seqno
==
495 DCCP_MAX_SEQNO
+ 1) &&
496 !inet_csk_ack_scheduled(sk
)) {
497 inet_csk_schedule_ack(sk
);
498 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
506 * Step 9: Process Reset
507 * If P.type == Reset,
508 * Tear down connection
509 * S.state := TIMEWAIT
511 * Drop packet and return
513 if (dh
->dccph_type
== DCCP_PKT_RESET
) {
515 * Queue the equivalent of TCP fin so that dccp_recvmsg
519 dccp_time_wait(sk
, DCCP_TIME_WAIT
, 0);
522 * Step 7: Check for unexpected packet types
523 * If (S.is_server and P.type == CloseReq)
524 * or (S.is_server and P.type == Response)
525 * or (S.is_client and P.type == Request)
526 * or (S.state == RESPOND and P.type == Data),
527 * Send Sync packet acknowledging P.seqno
528 * Drop packet and return
530 } else if ((dp
->dccps_role
!= DCCP_ROLE_CLIENT
&&
531 (dh
->dccph_type
== DCCP_PKT_RESPONSE
||
532 dh
->dccph_type
== DCCP_PKT_CLOSEREQ
)) ||
533 (dp
->dccps_role
== DCCP_ROLE_CLIENT
&&
534 dh
->dccph_type
== DCCP_PKT_REQUEST
) ||
535 (sk
->sk_state
== DCCP_RESPOND
&&
536 dh
->dccph_type
== DCCP_PKT_DATA
)) {
537 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
,
542 switch (sk
->sk_state
) {
546 case DCCP_REQUESTING
:
547 /* FIXME: do congestion control initialization */
549 queued
= dccp_rcv_request_sent_state_process(sk
, skb
, dh
, len
);
558 queued
= dccp_rcv_respond_partopen_state_process(sk
, skb
,
563 if (dh
->dccph_type
== DCCP_PKT_ACK
||
564 dh
->dccph_type
== DCCP_PKT_DATAACK
) {
567 sk
->sk_state_change(sk
);
568 sk_wake_async(sk
, 0, POLL_OUT
);