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/skbuff.h>
22 /* rate-limit for syncs in reply to sequence-invalid packets; RFC 4340, 7.5.4 */
23 int sysctl_dccp_sync_ratelimit __read_mostly
= HZ
/ 8;
25 static void dccp_fin(struct sock
*sk
, struct sk_buff
*skb
)
27 sk
->sk_shutdown
|= RCV_SHUTDOWN
;
28 sock_set_flag(sk
, SOCK_DONE
);
29 __skb_pull(skb
, dccp_hdr(skb
)->dccph_doff
* 4);
30 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
31 skb_set_owner_r(skb
, sk
);
32 sk
->sk_data_ready(sk
, 0);
35 static void dccp_rcv_close(struct sock
*sk
, struct sk_buff
*skb
)
37 dccp_send_reset(sk
, DCCP_RESET_CODE_CLOSED
);
39 dccp_set_state(sk
, DCCP_CLOSED
);
40 sk_wake_async(sk
, 1, POLL_HUP
);
43 static void dccp_rcv_closereq(struct sock
*sk
, struct sk_buff
*skb
)
46 * Step 7: Check for unexpected packet types
47 * If (S.is_server and P.type == CloseReq)
48 * Send Sync packet acknowledging P.seqno
49 * Drop packet and return
51 if (dccp_sk(sk
)->dccps_role
!= DCCP_ROLE_CLIENT
) {
52 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
, DCCP_PKT_SYNC
);
56 if (sk
->sk_state
!= DCCP_CLOSING
)
57 dccp_set_state(sk
, DCCP_CLOSING
);
58 dccp_send_close(sk
, 0);
61 static u8
dccp_reset_code_convert(const u8 code
)
63 const u8 error_code
[] = {
64 [DCCP_RESET_CODE_CLOSED
] = 0, /* normal termination */
65 [DCCP_RESET_CODE_UNSPECIFIED
] = 0, /* nothing known */
66 [DCCP_RESET_CODE_ABORTED
] = ECONNRESET
,
68 [DCCP_RESET_CODE_NO_CONNECTION
] = ECONNREFUSED
,
69 [DCCP_RESET_CODE_CONNECTION_REFUSED
] = ECONNREFUSED
,
70 [DCCP_RESET_CODE_TOO_BUSY
] = EUSERS
,
71 [DCCP_RESET_CODE_AGGRESSION_PENALTY
] = EDQUOT
,
73 [DCCP_RESET_CODE_PACKET_ERROR
] = ENOMSG
,
74 [DCCP_RESET_CODE_BAD_INIT_COOKIE
] = EBADR
,
75 [DCCP_RESET_CODE_BAD_SERVICE_CODE
] = EBADRQC
,
76 [DCCP_RESET_CODE_OPTION_ERROR
] = EILSEQ
,
77 [DCCP_RESET_CODE_MANDATORY_ERROR
] = EOPNOTSUPP
,
80 return code
>= DCCP_MAX_RESET_CODES
? 0 : error_code
[code
];
83 static void dccp_rcv_reset(struct sock
*sk
, struct sk_buff
*skb
)
85 u8 err
= dccp_reset_code_convert(dccp_hdr_reset(skb
)->dccph_reset_code
);
89 /* Queue the equivalent of TCP fin so that dccp_recvmsg exits the loop */
92 if (err
&& !sock_flag(sk
, SOCK_DEAD
))
93 sk_wake_async(sk
, 0, POLL_ERR
);
94 dccp_time_wait(sk
, DCCP_TIME_WAIT
, 0);
97 static void dccp_event_ack_recv(struct sock
*sk
, struct sk_buff
*skb
)
99 struct dccp_sock
*dp
= dccp_sk(sk
);
101 if (dccp_msk(sk
)->dccpms_send_ack_vector
)
102 dccp_ackvec_check_rcv_ackno(dp
->dccps_hc_rx_ackvec
, sk
,
103 DCCP_SKB_CB(skb
)->dccpd_ack_seq
);
106 static int dccp_check_seqno(struct sock
*sk
, struct sk_buff
*skb
)
108 const struct dccp_hdr
*dh
= dccp_hdr(skb
);
109 struct dccp_sock
*dp
= dccp_sk(sk
);
110 u64 lswl
, lawl
, seqno
= DCCP_SKB_CB(skb
)->dccpd_seq
,
111 ackno
= DCCP_SKB_CB(skb
)->dccpd_ack_seq
;
114 * Step 5: Prepare sequence numbers for Sync
115 * If P.type == Sync or P.type == SyncAck,
116 * If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL,
117 * / * P is valid, so update sequence number variables
118 * accordingly. After this update, P will pass the tests
119 * in Step 6. A SyncAck is generated if necessary in
121 * Update S.GSR, S.SWL, S.SWH
123 * Drop packet and return
125 if (dh
->dccph_type
== DCCP_PKT_SYNC
||
126 dh
->dccph_type
== DCCP_PKT_SYNCACK
) {
127 if (between48(ackno
, dp
->dccps_awl
, dp
->dccps_awh
) &&
128 dccp_delta_seqno(dp
->dccps_swl
, seqno
) >= 0)
129 dccp_update_gsr(sk
, seqno
);
135 * Step 6: Check sequence numbers
136 * Let LSWL = S.SWL and LAWL = S.AWL
137 * If P.type == CloseReq or P.type == Close or P.type == Reset,
138 * LSWL := S.GSR + 1, LAWL := S.GAR
139 * If LSWL <= P.seqno <= S.SWH
140 * and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH),
141 * Update S.GSR, S.SWL, S.SWH
145 lswl
= dp
->dccps_swl
;
146 lawl
= dp
->dccps_awl
;
148 if (dh
->dccph_type
== DCCP_PKT_CLOSEREQ
||
149 dh
->dccph_type
== DCCP_PKT_CLOSE
||
150 dh
->dccph_type
== DCCP_PKT_RESET
) {
151 lswl
= ADD48(dp
->dccps_gsr
, 1);
152 lawl
= dp
->dccps_gar
;
155 if (between48(seqno
, lswl
, dp
->dccps_swh
) &&
156 (ackno
== DCCP_PKT_WITHOUT_ACK_SEQ
||
157 between48(ackno
, lawl
, dp
->dccps_awh
))) {
158 dccp_update_gsr(sk
, seqno
);
160 if (dh
->dccph_type
!= DCCP_PKT_SYNC
&&
161 (ackno
!= DCCP_PKT_WITHOUT_ACK_SEQ
))
162 dp
->dccps_gar
= ackno
;
164 unsigned long now
= jiffies
;
166 * Step 6: Check sequence numbers
168 * If P.type == Reset,
169 * Send Sync packet acknowledging S.GSR
171 * Send Sync packet acknowledging P.seqno
172 * Drop packet and return
174 * These Syncs are rate-limited as per RFC 4340, 7.5.4:
175 * at most 1 / (dccp_sync_rate_limit * HZ) Syncs per second.
177 if (time_before(now
, (dp
->dccps_rate_last
+
178 sysctl_dccp_sync_ratelimit
)))
181 DCCP_WARN("DCCP: Step 6 failed for %s packet, "
182 "(LSWL(%llu) <= P.seqno(%llu) <= S.SWH(%llu)) and "
183 "(P.ackno %s or LAWL(%llu) <= P.ackno(%llu) <= S.AWH(%llu), "
184 "sending SYNC...\n", dccp_packet_name(dh
->dccph_type
),
185 (unsigned long long) lswl
, (unsigned long long) seqno
,
186 (unsigned long long) dp
->dccps_swh
,
187 (ackno
== DCCP_PKT_WITHOUT_ACK_SEQ
) ? "doesn't exist"
189 (unsigned long long) lawl
, (unsigned long long) ackno
,
190 (unsigned long long) dp
->dccps_awh
);
192 dp
->dccps_rate_last
= now
;
194 if (dh
->dccph_type
== DCCP_PKT_RESET
)
195 seqno
= dp
->dccps_gsr
;
196 dccp_send_sync(sk
, seqno
, DCCP_PKT_SYNC
);
203 static int __dccp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
204 const struct dccp_hdr
*dh
, const unsigned len
)
206 struct dccp_sock
*dp
= dccp_sk(sk
);
208 switch (dccp_hdr(skb
)->dccph_type
) {
209 case DCCP_PKT_DATAACK
:
212 * FIXME: check if sk_receive_queue is full, schedule DATA_DROPPED
215 __skb_pull(skb
, dh
->dccph_doff
* 4);
216 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
217 skb_set_owner_r(skb
, sk
);
218 sk
->sk_data_ready(sk
, 0);
224 * Step 9: Process Reset
225 * If P.type == Reset,
226 * Tear down connection
227 * S.state := TIMEWAIT
229 * Drop packet and return
231 dccp_rcv_reset(sk
, skb
);
233 case DCCP_PKT_CLOSEREQ
:
234 dccp_rcv_closereq(sk
, skb
);
237 dccp_rcv_close(sk
, skb
);
239 case DCCP_PKT_REQUEST
:
241 * or (S.is_server and P.type == Response)
242 * or (S.is_client and P.type == Request)
243 * or (S.state >= OPEN and P.type == Request
244 * and P.seqno >= S.OSR)
245 * or (S.state >= OPEN and P.type == Response
246 * and P.seqno >= S.OSR)
247 * or (S.state == RESPOND and P.type == Data),
248 * Send Sync packet acknowledging P.seqno
249 * Drop packet and return
251 if (dp
->dccps_role
!= DCCP_ROLE_LISTEN
)
254 case DCCP_PKT_RESPONSE
:
255 if (dp
->dccps_role
!= DCCP_ROLE_CLIENT
)
258 if (dccp_delta_seqno(dp
->dccps_osr
,
259 DCCP_SKB_CB(skb
)->dccpd_seq
) >= 0) {
261 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
,
266 dccp_send_sync(sk
, DCCP_SKB_CB(skb
)->dccpd_seq
,
269 * From RFC 4340, sec. 5.7
271 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets
272 * MAY have non-zero-length application data areas, whose
273 * contents receivers MUST ignore.
278 DCCP_INC_STATS_BH(DCCP_MIB_INERRS
);
284 int dccp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
285 const struct dccp_hdr
*dh
, const unsigned len
)
287 struct dccp_sock
*dp
= dccp_sk(sk
);
289 if (dccp_check_seqno(sk
, skb
))
292 if (dccp_parse_options(sk
, skb
))
295 if (DCCP_SKB_CB(skb
)->dccpd_ack_seq
!= DCCP_PKT_WITHOUT_ACK_SEQ
)
296 dccp_event_ack_recv(sk
, skb
);
298 if (dccp_msk(sk
)->dccpms_send_ack_vector
&&
299 dccp_ackvec_add(dp
->dccps_hc_rx_ackvec
, sk
,
300 DCCP_SKB_CB(skb
)->dccpd_seq
,
301 DCCP_ACKVEC_STATE_RECEIVED
))
304 ccid_hc_rx_packet_recv(dp
->dccps_hc_rx_ccid
, sk
, skb
);
305 ccid_hc_tx_packet_recv(dp
->dccps_hc_tx_ccid
, sk
, skb
);
307 return __dccp_rcv_established(sk
, skb
, dh
, len
);
313 EXPORT_SYMBOL_GPL(dccp_rcv_established
);
315 static int dccp_rcv_request_sent_state_process(struct sock
*sk
,
317 const struct dccp_hdr
*dh
,
321 * Step 4: Prepare sequence numbers in REQUEST
322 * If S.state == REQUEST,
323 * If (P.type == Response or P.type == Reset)
324 * and S.AWL <= P.ackno <= S.AWH,
325 * / * Set sequence number variables corresponding to the
326 * other endpoint, so P will pass the tests in Step 6 * /
327 * Set S.GSR, S.ISR, S.SWL, S.SWH
328 * / * Response processing continues in Step 10; Reset
329 * processing continues in Step 9 * /
331 if (dh
->dccph_type
== DCCP_PKT_RESPONSE
) {
332 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
333 struct dccp_sock
*dp
= dccp_sk(sk
);
334 long tstamp
= dccp_timestamp();
336 /* Stop the REQUEST timer */
337 inet_csk_clear_xmit_timer(sk
, ICSK_TIME_RETRANS
);
338 BUG_TRAP(sk
->sk_send_head
!= NULL
);
339 __kfree_skb(sk
->sk_send_head
);
340 sk
->sk_send_head
= NULL
;
342 if (!between48(DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
343 dp
->dccps_awl
, dp
->dccps_awh
)) {
344 dccp_pr_debug("invalid ackno: S.AWL=%llu, "
345 "P.ackno=%llu, S.AWH=%llu \n",
346 (unsigned long long)dp
->dccps_awl
,
347 (unsigned long long)DCCP_SKB_CB(skb
)->dccpd_ack_seq
,
348 (unsigned long long)dp
->dccps_awh
);
349 goto out_invalid_packet
;
352 if (dccp_parse_options(sk
, skb
))
353 goto out_invalid_packet
;
355 /* Obtain usec RTT sample from SYN exchange (used by CCID 3) */
356 if (likely(dp
->dccps_options_received
.dccpor_timestamp_echo
))
357 dp
->dccps_syn_rtt
= dccp_sample_rtt(sk
, 10 * (tstamp
-
358 dp
->dccps_options_received
.dccpor_timestamp_echo
));
360 if (dccp_msk(sk
)->dccpms_send_ack_vector
&&
361 dccp_ackvec_add(dp
->dccps_hc_rx_ackvec
, sk
,
362 DCCP_SKB_CB(skb
)->dccpd_seq
,
363 DCCP_ACKVEC_STATE_RECEIVED
))
364 goto out_invalid_packet
; /* FIXME: change error code */
366 dp
->dccps_isr
= DCCP_SKB_CB(skb
)->dccpd_seq
;
367 dccp_update_gsr(sk
, dp
->dccps_isr
);
369 * SWL and AWL are initially adjusted so that they are not less than
370 * the initial Sequence Numbers received and sent, respectively:
371 * SWL := max(GSR + 1 - floor(W/4), ISR),
372 * AWL := max(GSS - W' + 1, ISS).
373 * These adjustments MUST be applied only at the beginning of the
376 * AWL was adjusted in dccp_v4_connect -acme
378 dccp_set_seqno(&dp
->dccps_swl
,
379 max48(dp
->dccps_swl
, dp
->dccps_isr
));
381 dccp_sync_mss(sk
, icsk
->icsk_pmtu_cookie
);
384 * Step 10: Process REQUEST state (second part)
385 * If S.state == REQUEST,
386 * / * If we get here, P is a valid Response from the
387 * server (see Step 4), and we should move to
388 * PARTOPEN state. PARTOPEN means send an Ack,
389 * don't send Data packets, retransmit Acks
390 * periodically, and always include any Init Cookie
391 * from the Response * /
392 * S.state := PARTOPEN
394 * Continue with S.state == PARTOPEN
395 * / * Step 12 will send the Ack completing the
396 * three-way handshake * /
398 dccp_set_state(sk
, DCCP_PARTOPEN
);
400 /* Make sure socket is routed, for correct metrics. */
401 icsk
->icsk_af_ops
->rebuild_header(sk
);
403 if (!sock_flag(sk
, SOCK_DEAD
)) {
404 sk
->sk_state_change(sk
);
405 sk_wake_async(sk
, 0, POLL_OUT
);
408 if (sk
->sk_write_pending
|| icsk
->icsk_ack
.pingpong
||
409 icsk
->icsk_accept_queue
.rskq_defer_accept
) {
410 /* Save one ACK. Data will be ready after
411 * several ticks, if write_pending is set.
413 * It may be deleted, but with this feature tcpdumps
414 * look so _wonderfully_ clever, that I was not able
415 * to stand against the temptation 8) --ANK
418 * OK, in DCCP we can as well do a similar trick, its
419 * even in the draft, but there is no need for us to
420 * schedule an ack here, as dccp_sendmsg does this for
421 * us, also stated in the draft. -acme
431 /* dccp_v4_do_rcv will send a reset */
432 DCCP_SKB_CB(skb
)->dccpd_reset_code
= DCCP_RESET_CODE_PACKET_ERROR
;
436 static int dccp_rcv_respond_partopen_state_process(struct sock
*sk
,
438 const struct dccp_hdr
*dh
,
443 switch (dh
->dccph_type
) {
445 inet_csk_clear_xmit_timer(sk
, ICSK_TIME_DACK
);
448 if (sk
->sk_state
== DCCP_RESPOND
)
450 case DCCP_PKT_DATAACK
:
453 * FIXME: we should be reseting the PARTOPEN (DELACK) timer
454 * here but only if we haven't used the DELACK timer for
455 * something else, like sending a delayed ack for a TIMESTAMP
456 * echo, etc, for now were not clearing it, sending an extra
457 * ACK when there is nothing else to do in DELACK is not a big
461 /* Stop the PARTOPEN timer */
462 if (sk
->sk_state
== DCCP_PARTOPEN
)
463 inet_csk_clear_xmit_timer(sk
, ICSK_TIME_DACK
);
465 dccp_sk(sk
)->dccps_osr
= DCCP_SKB_CB(skb
)->dccpd_seq
;
466 dccp_set_state(sk
, DCCP_OPEN
);
468 if (dh
->dccph_type
== DCCP_PKT_DATAACK
||
469 dh
->dccph_type
== DCCP_PKT_DATA
) {
470 __dccp_rcv_established(sk
, skb
, dh
, len
);
471 queued
= 1; /* packet was queued
472 (by __dccp_rcv_established) */
480 int dccp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
481 struct dccp_hdr
*dh
, unsigned len
)
483 struct dccp_sock
*dp
= dccp_sk(sk
);
484 struct dccp_skb_cb
*dcb
= DCCP_SKB_CB(skb
);
485 const int old_state
= sk
->sk_state
;
489 * Step 3: Process LISTEN state
491 * If S.state == LISTEN,
492 * If P.type == Request or P contains a valid Init Cookie option,
493 * (* Must scan the packet's options to check for Init
494 * Cookies. Only Init Cookies are processed here,
495 * however; other options are processed in Step 8. This
496 * scan need only be performed if the endpoint uses Init
498 * (* Generate a new socket and switch to that socket *)
499 * Set S := new socket for this port pair
501 * Choose S.ISS (initial seqno) or set from Init Cookies
502 * Initialize S.GAR := S.ISS
503 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
504 * Cookies Continue with S.state == RESPOND
505 * (* A Response packet will be generated in Step 11 *)
507 * Generate Reset(No Connection) unless P.type == Reset
508 * Drop packet and return
510 if (sk
->sk_state
== DCCP_LISTEN
) {
511 if (dh
->dccph_type
== DCCP_PKT_REQUEST
) {
512 if (inet_csk(sk
)->icsk_af_ops
->conn_request(sk
,
516 /* FIXME: do congestion control initialization */
519 if (dh
->dccph_type
== DCCP_PKT_RESET
)
522 /* Caller (dccp_v4_do_rcv) will send Reset */
523 dcb
->dccpd_reset_code
= DCCP_RESET_CODE_NO_CONNECTION
;
527 if (sk
->sk_state
!= DCCP_REQUESTING
) {
528 if (dccp_check_seqno(sk
, skb
))
532 * Step 8: Process options and mark acknowledgeable
534 if (dccp_parse_options(sk
, skb
))
537 if (dcb
->dccpd_ack_seq
!= DCCP_PKT_WITHOUT_ACK_SEQ
)
538 dccp_event_ack_recv(sk
, skb
);
540 if (dccp_msk(sk
)->dccpms_send_ack_vector
&&
541 dccp_ackvec_add(dp
->dccps_hc_rx_ackvec
, sk
,
542 DCCP_SKB_CB(skb
)->dccpd_seq
,
543 DCCP_ACKVEC_STATE_RECEIVED
))
546 ccid_hc_rx_packet_recv(dp
->dccps_hc_rx_ccid
, sk
, skb
);
547 ccid_hc_tx_packet_recv(dp
->dccps_hc_tx_ccid
, sk
, skb
);
551 * Step 9: Process Reset
552 * If P.type == Reset,
553 * Tear down connection
554 * S.state := TIMEWAIT
556 * Drop packet and return
558 if (dh
->dccph_type
== DCCP_PKT_RESET
) {
559 dccp_rcv_reset(sk
, skb
);
562 * Step 7: Check for unexpected packet types
563 * If (S.is_server and P.type == CloseReq)
564 * or (S.is_server and P.type == Response)
565 * or (S.is_client and P.type == Request)
566 * or (S.state == RESPOND and P.type == Data),
567 * Send Sync packet acknowledging P.seqno
568 * Drop packet and return
570 } else if ((dp
->dccps_role
!= DCCP_ROLE_CLIENT
&&
571 (dh
->dccph_type
== DCCP_PKT_RESPONSE
||
572 dh
->dccph_type
== DCCP_PKT_CLOSEREQ
)) ||
573 (dp
->dccps_role
== DCCP_ROLE_CLIENT
&&
574 dh
->dccph_type
== DCCP_PKT_REQUEST
) ||
575 (sk
->sk_state
== DCCP_RESPOND
&&
576 dh
->dccph_type
== DCCP_PKT_DATA
)) {
577 dccp_send_sync(sk
, dcb
->dccpd_seq
, DCCP_PKT_SYNC
);
579 } else if (dh
->dccph_type
== DCCP_PKT_CLOSEREQ
) {
580 dccp_rcv_closereq(sk
, skb
);
582 } else if (dh
->dccph_type
== DCCP_PKT_CLOSE
) {
583 dccp_rcv_close(sk
, skb
);
587 switch (sk
->sk_state
) {
589 dcb
->dccpd_reset_code
= DCCP_RESET_CODE_NO_CONNECTION
;
592 case DCCP_REQUESTING
:
593 /* FIXME: do congestion control initialization */
595 queued
= dccp_rcv_request_sent_state_process(sk
, skb
, dh
, len
);
604 queued
= dccp_rcv_respond_partopen_state_process(sk
, skb
,
609 if (dh
->dccph_type
== DCCP_PKT_ACK
||
610 dh
->dccph_type
== DCCP_PKT_DATAACK
) {
613 sk
->sk_state_change(sk
);
614 sk_wake_async(sk
, 0, POLL_OUT
);
617 } else if (unlikely(dh
->dccph_type
== DCCP_PKT_SYNC
)) {
618 dccp_send_sync(sk
, dcb
->dccpd_seq
, DCCP_PKT_SYNCACK
);
629 EXPORT_SYMBOL_GPL(dccp_rcv_state_process
);
632 * dccp_sample_rtt - Validate and finalise computation of RTT sample
633 * @delta: number of microseconds between packet and acknowledgment
634 * The routine is kept generic to work in different contexts. It should be
635 * called immediately when the ACK used for the RTT sample arrives.
637 u32
dccp_sample_rtt(struct sock
*sk
, long delta
)
639 /* dccpor_elapsed_time is either zeroed out or set and > 0 */
640 delta
-= dccp_sk(sk
)->dccps_options_received
.dccpor_elapsed_time
* 10;
642 if (unlikely(delta
<= 0)) {
643 DCCP_WARN("unusable RTT sample %ld, using min\n", delta
);
644 return DCCP_SANE_RTT_MIN
;
646 if (unlikely(delta
> DCCP_SANE_RTT_MAX
)) {
647 DCCP_WARN("RTT sample %ld too large, using max\n", delta
);
648 return DCCP_SANE_RTT_MAX
;
654 EXPORT_SYMBOL_GPL(dccp_sample_rtt
);