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 for %s packet, "
140 "(LSWL(%llu) <= P.seqno(%llu) <= S.SWH(%llu)) and "
141 "(P.ackno %s or LAWL(%llu) <= P.ackno(%llu) <= S.AWH(%llu), "
143 dccp_packet_name(dh
->dccph_type
),
144 (unsigned long long) lswl
,
146 DCCP_SKB_CB(skb
)->dccpd_seq
,
147 (unsigned long long) dp
->dccps_swh
,
148 (DCCP_SKB_CB(skb
)->dccpd_ack_seq
==
149 DCCP_PKT_WITHOUT_ACK_SEQ
) ? "doesn't exist" : "exists",
150 (unsigned long long) lawl
,
152 DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
153 (unsigned long long) dp
->dccps_awh
);
154 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
, DCCP_PKT_SYNC
);
161 int dccp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
162 const struct dccp_hdr
*dh
, const unsigned len
)
164 struct dccp_sock
*dp
= dccp_sk(sk
);
166 if (dccp_check_seqno(sk
, skb
))
169 if (dccp_parse_options(sk
, skb
))
172 if (DCCP_SKB_CB(skb
)->dccpd_ack_seq
!= DCCP_PKT_WITHOUT_ACK_SEQ
)
173 dccp_event_ack_recv(sk
, skb
);
176 * FIXME: check ECN to see if we should use
177 * DCCP_ACKPKTS_STATE_ECN_MARKED
179 if (dp
->dccps_options
.dccpo_send_ack_vector
) {
180 struct dccp_ackpkts
*ap
= dp
->dccps_hc_rx_ackpkts
;
182 if (dccp_ackpkts_add(dp
->dccps_hc_rx_ackpkts
,
183 DCCP_SKB_CB(skb
)->dccpd_seq
,
184 DCCP_ACKPKTS_STATE_RECEIVED
)) {
185 LIMIT_NETDEBUG(KERN_WARNING
"DCCP: acknowledgeable "
186 "packets buffer full!\n");
187 ap
->dccpap_ack_seqno
= DCCP_MAX_SEQNO
+ 1;
188 inet_csk_schedule_ack(sk
);
189 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
196 * FIXME: this activation is probably wrong, have to study more
197 * TCP delack machinery and how it fits into DCCP draft, but
198 * for now it kinda "works" 8)
200 if (!inet_csk_ack_scheduled(sk
)) {
201 inet_csk_schedule_ack(sk
);
202 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
, 5 * HZ
,
207 ccid_hc_rx_packet_recv(dp
->dccps_hc_rx_ccid
, sk
, skb
);
208 ccid_hc_tx_packet_recv(dp
->dccps_hc_tx_ccid
, sk
, skb
);
210 switch (dccp_hdr(skb
)->dccph_type
) {
211 case DCCP_PKT_DATAACK
:
214 * FIXME: check if sk_receive_queue is full, schedule DATA_DROPPED
217 __skb_pull(skb
, dh
->dccph_doff
* 4);
218 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
219 skb_set_owner_r(skb
, sk
);
220 sk
->sk_data_ready(sk
, 0);
226 * Step 9: Process Reset
227 * If P.type == Reset,
228 * Tear down connection
229 * S.state := TIMEWAIT
231 * Drop packet and return
234 dccp_time_wait(sk
, DCCP_TIME_WAIT
, 0);
236 case DCCP_PKT_CLOSEREQ
:
237 dccp_rcv_closereq(sk
, skb
);
240 dccp_rcv_close(sk
, skb
);
242 case DCCP_PKT_REQUEST
:
244 * or (S.is_server and P.type == Response)
245 * or (S.is_client and P.type == Request)
246 * or (S.state >= OPEN and P.type == Request
247 * and P.seqno >= S.OSR)
248 * or (S.state >= OPEN and P.type == Response
249 * and P.seqno >= S.OSR)
250 * or (S.state == RESPOND and P.type == Data),
251 * Send Sync packet acknowledging P.seqno
252 * Drop packet and return
254 if (dp
->dccps_role
!= DCCP_ROLE_LISTEN
)
257 case DCCP_PKT_RESPONSE
:
258 if (dp
->dccps_role
!= DCCP_ROLE_CLIENT
)
261 if (!before48(DCCP_SKB_CB(skb
)->dccpd_seq
, dp
->dccps_osr
)) {
263 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
,
268 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
,
273 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets
274 * MAY have non-zero-length application data areas, whose
275 * contents * receivers MUST ignore.
280 DCCP_INC_STATS_BH(DCCP_MIB_INERRS
);
286 static int dccp_rcv_request_sent_state_process(struct sock
*sk
,
288 const struct dccp_hdr
*dh
,
292 * Step 4: Prepare sequence numbers in REQUEST
293 * If S.state == REQUEST,
294 * If (P.type == Response or P.type == Reset)
295 * and S.AWL <= P.ackno <= S.AWH,
296 * / * Set sequence number variables corresponding to the
297 * other endpoint, so P will pass the tests in Step 6 * /
298 * Set S.GSR, S.ISR, S.SWL, S.SWH
299 * / * Response processing continues in Step 10; Reset
300 * processing continues in Step 9 * /
302 if (dh
->dccph_type
== DCCP_PKT_RESPONSE
) {
303 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
304 struct dccp_sock
*dp
= dccp_sk(sk
);
306 /* Stop the REQUEST timer */
307 inet_csk_clear_xmit_timer(sk
, ICSK_TIME_RETRANS
);
308 BUG_TRAP(sk
->sk_send_head
!= NULL
);
309 __kfree_skb(sk
->sk_send_head
);
310 sk
->sk_send_head
= NULL
;
312 if (!between48(DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
313 dp
->dccps_awl
, dp
->dccps_awh
)) {
314 dccp_pr_debug("invalid ackno: S.AWL=%llu, "
315 "P.ackno=%llu, S.AWH=%llu \n",
316 (unsigned long long)dp
->dccps_awl
,
317 (unsigned long long)DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
318 (unsigned long long)dp
->dccps_awh
);
319 goto out_invalid_packet
;
322 dp
->dccps_isr
= DCCP_SKB_CB(skb
)->dccpd_seq
;
323 dccp_update_gsr(sk
, dp
->dccps_isr
);
325 * SWL and AWL are initially adjusted so that they are not less than
326 * the initial Sequence Numbers received and sent, respectively:
327 * SWL := max(GSR + 1 - floor(W/4), ISR),
328 * AWL := max(GSS - W' + 1, ISS).
329 * These adjustments MUST be applied only at the beginning of the
332 * AWL was adjusted in dccp_v4_connect -acme
334 dccp_set_seqno(&dp
->dccps_swl
,
335 max48(dp
->dccps_swl
, dp
->dccps_isr
));
337 if (ccid_hc_rx_init(dp
->dccps_hc_rx_ccid
, sk
) != 0 ||
338 ccid_hc_tx_init(dp
->dccps_hc_tx_ccid
, sk
) != 0) {
339 ccid_hc_rx_exit(dp
->dccps_hc_rx_ccid
, sk
);
340 ccid_hc_tx_exit(dp
->dccps_hc_tx_ccid
, sk
);
341 /* FIXME: send appropriate RESET code */
342 goto out_invalid_packet
;
345 dccp_sync_mss(sk
, dp
->dccps_pmtu_cookie
);
348 * Step 10: Process REQUEST state (second part)
349 * If S.state == REQUEST,
350 * / * If we get here, P is a valid Response from the
351 * server (see Step 4), and we should move to
352 * PARTOPEN state. PARTOPEN means send an Ack,
353 * don't send Data packets, retransmit Acks
354 * periodically, and always include any Init Cookie
355 * from the Response * /
356 * S.state := PARTOPEN
358 * Continue with S.state == PARTOPEN
359 * / * Step 12 will send the Ack completing the
360 * three-way handshake * /
362 dccp_set_state(sk
, DCCP_PARTOPEN
);
364 /* Make sure socket is routed, for correct metrics. */
365 inet_sk_rebuild_header(sk
);
367 if (!sock_flag(sk
, SOCK_DEAD
)) {
368 sk
->sk_state_change(sk
);
369 sk_wake_async(sk
, 0, POLL_OUT
);
372 if (sk
->sk_write_pending
|| icsk
->icsk_ack
.pingpong
||
373 icsk
->icsk_accept_queue
.rskq_defer_accept
) {
374 /* Save one ACK. Data will be ready after
375 * several ticks, if write_pending is set.
377 * It may be deleted, but with this feature tcpdumps
378 * look so _wonderfully_ clever, that I was not able
379 * to stand against the temptation 8) --ANK
382 * OK, in DCCP we can as well do a similar trick, its
383 * even in the draft, but there is no need for us to
384 * schedule an ack here, as dccp_sendmsg does this for
385 * us, also stated in the draft. -acme
395 return 1; /* dccp_v4_do_rcv will send a reset, but...
396 FIXME: the reset code should be
397 DCCP_RESET_CODE_PACKET_ERROR */
400 static int dccp_rcv_respond_partopen_state_process(struct sock
*sk
,
402 const struct dccp_hdr
*dh
,
407 switch (dh
->dccph_type
) {
409 inet_csk_clear_xmit_timer(sk
, ICSK_TIME_DACK
);
411 case DCCP_PKT_DATAACK
:
414 * FIXME: we should be reseting the PARTOPEN (DELACK) timer
415 * here but only if we haven't used the DELACK timer for
416 * something else, like sending a delayed ack for a TIMESTAMP
417 * echo, etc, for now were not clearing it, sending an extra
418 * ACK when there is nothing else to do in DELACK is not a big
422 /* Stop the PARTOPEN timer */
423 if (sk
->sk_state
== DCCP_PARTOPEN
)
424 inet_csk_clear_xmit_timer(sk
, ICSK_TIME_DACK
);
426 dccp_sk(sk
)->dccps_osr
= DCCP_SKB_CB(skb
)->dccpd_seq
;
427 dccp_set_state(sk
, DCCP_OPEN
);
429 if (dh
->dccph_type
== DCCP_PKT_DATAACK
) {
430 dccp_rcv_established(sk
, skb
, dh
, len
);
431 queued
= 1; /* packet was queued
432 (by dccp_rcv_established) */
440 int dccp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
441 struct dccp_hdr
*dh
, unsigned len
)
443 struct dccp_sock
*dp
= dccp_sk(sk
);
444 const int old_state
= sk
->sk_state
;
448 * Step 3: Process LISTEN state
449 * (Continuing from dccp_v4_do_rcv and dccp_v6_do_rcv)
451 * If S.state == LISTEN,
452 * If P.type == Request or P contains a valid Init Cookie
454 * * Must scan the packet's options to check for an Init
455 * Cookie. Only the Init Cookie is processed here,
456 * however; other options are processed in Step 8. This
457 * scan need only be performed if the endpoint uses Init
459 * * Generate a new socket and switch to that socket *
460 * Set S := new socket for this port pair
462 * Choose S.ISS (initial seqno) or set from Init Cookie
463 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
464 * Continue with S.state == RESPOND
465 * * A Response packet will be generated in Step 11 *
467 * Generate Reset(No Connection) unless P.type == Reset
468 * Drop packet and return
470 * NOTE: the check for the packet types is done in
471 * dccp_rcv_state_process
473 if (sk
->sk_state
== DCCP_LISTEN
) {
474 if (dh
->dccph_type
== DCCP_PKT_REQUEST
) {
475 if (dccp_v4_conn_request(sk
, skb
) < 0)
478 /* FIXME: do congestion control initialization */
481 if (dh
->dccph_type
== DCCP_PKT_RESET
)
484 /* Caller (dccp_v4_do_rcv) will send Reset(No Connection)*/
488 if (sk
->sk_state
!= DCCP_REQUESTING
) {
489 if (dccp_check_seqno(sk
, skb
))
493 * Step 8: Process options and mark acknowledgeable
495 if (dccp_parse_options(sk
, skb
))
498 if (DCCP_SKB_CB(skb
)->dccpd_ack_seq
!=
499 DCCP_PKT_WITHOUT_ACK_SEQ
)
500 dccp_event_ack_recv(sk
, skb
);
502 ccid_hc_rx_packet_recv(dp
->dccps_hc_rx_ccid
, sk
, skb
);
503 ccid_hc_tx_packet_recv(dp
->dccps_hc_tx_ccid
, sk
, skb
);
506 * FIXME: check ECN to see if we should use
507 * DCCP_ACKPKTS_STATE_ECN_MARKED
509 if (dp
->dccps_options
.dccpo_send_ack_vector
) {
510 if (dccp_ackpkts_add(dp
->dccps_hc_rx_ackpkts
,
511 DCCP_SKB_CB(skb
)->dccpd_seq
,
512 DCCP_ACKPKTS_STATE_RECEIVED
))
515 * FIXME: this activation is probably wrong, have to
516 * study more TCP delack machinery and how it fits into
517 * DCCP draft, but for now it kinda "works" 8)
519 if ((dp
->dccps_hc_rx_ackpkts
->dccpap_ack_seqno
==
520 DCCP_MAX_SEQNO
+ 1) &&
521 !inet_csk_ack_scheduled(sk
)) {
522 inet_csk_schedule_ack(sk
);
523 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
531 * Step 9: Process Reset
532 * If P.type == Reset,
533 * Tear down connection
534 * S.state := TIMEWAIT
536 * Drop packet and return
538 if (dh
->dccph_type
== DCCP_PKT_RESET
) {
540 * Queue the equivalent of TCP fin so that dccp_recvmsg
544 dccp_time_wait(sk
, DCCP_TIME_WAIT
, 0);
547 * Step 7: Check for unexpected packet types
548 * If (S.is_server and P.type == CloseReq)
549 * or (S.is_server and P.type == Response)
550 * or (S.is_client and P.type == Request)
551 * or (S.state == RESPOND and P.type == Data),
552 * Send Sync packet acknowledging P.seqno
553 * Drop packet and return
555 } else if ((dp
->dccps_role
!= DCCP_ROLE_CLIENT
&&
556 (dh
->dccph_type
== DCCP_PKT_RESPONSE
||
557 dh
->dccph_type
== DCCP_PKT_CLOSEREQ
)) ||
558 (dp
->dccps_role
== DCCP_ROLE_CLIENT
&&
559 dh
->dccph_type
== DCCP_PKT_REQUEST
) ||
560 (sk
->sk_state
== DCCP_RESPOND
&&
561 dh
->dccph_type
== DCCP_PKT_DATA
)) {
562 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
,
567 switch (sk
->sk_state
) {
571 case DCCP_REQUESTING
:
572 /* FIXME: do congestion control initialization */
574 queued
= dccp_rcv_request_sent_state_process(sk
, skb
, dh
, len
);
583 queued
= dccp_rcv_respond_partopen_state_process(sk
, skb
,
588 if (dh
->dccph_type
== DCCP_PKT_ACK
||
589 dh
->dccph_type
== DCCP_PKT_DATAACK
) {
592 sk
->sk_state_change(sk
);
593 sk_wake_async(sk
, 0, POLL_OUT
);