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 void dccp_event_ack_recv(struct sock *sk, struct sk_buff *skb)
63 struct dccp_sock *dp = dccp_sk(sk);
65 if (dccp_msk(sk)->dccpms_send_ack_vector)
66 dccp_ackvec_check_rcv_ackno(dp->dccps_hc_rx_ackvec, sk,
67 DCCP_SKB_CB(skb)->dccpd_ack_seq);
70 static int dccp_check_seqno(struct sock *sk, struct sk_buff *skb)
72 const struct dccp_hdr *dh = dccp_hdr(skb);
73 struct dccp_sock *dp = dccp_sk(sk);
74 u64 lswl, lawl, seqno = DCCP_SKB_CB(skb)->dccpd_seq,
75 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
78 * Step 5: Prepare sequence numbers for Sync
79 * If P.type == Sync or P.type == SyncAck,
80 * If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL,
81 * / * P is valid, so update sequence number variables
82 * accordingly. After this update, P will pass the tests
83 * in Step 6. A SyncAck is generated if necessary in
85 * Update S.GSR, S.SWL, S.SWH
87 * Drop packet and return
89 if (dh->dccph_type == DCCP_PKT_SYNC ||
90 dh->dccph_type == DCCP_PKT_SYNCACK) {
91 if (between48(ackno, dp->dccps_awl, dp->dccps_awh) &&
92 dccp_delta_seqno(dp->dccps_swl, seqno) >= 0)
93 dccp_update_gsr(sk, seqno);
99 * Step 6: Check sequence numbers
100 * Let LSWL = S.SWL and LAWL = S.AWL
101 * If P.type == CloseReq or P.type == Close or P.type == Reset,
102 * LSWL := S.GSR + 1, LAWL := S.GAR
103 * If LSWL <= P.seqno <= S.SWH
104 * and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH),
105 * Update S.GSR, S.SWL, S.SWH
109 lswl = dp->dccps_swl;
110 lawl = dp->dccps_awl;
112 if (dh->dccph_type == DCCP_PKT_CLOSEREQ ||
113 dh->dccph_type == DCCP_PKT_CLOSE ||
114 dh->dccph_type == DCCP_PKT_RESET) {
115 lswl = ADD48(dp->dccps_gsr, 1);
116 lawl = dp->dccps_gar;
119 if (between48(seqno, lswl, dp->dccps_swh) &&
120 (ackno == DCCP_PKT_WITHOUT_ACK_SEQ ||
121 between48(ackno, lawl, dp->dccps_awh))) {
122 dccp_update_gsr(sk, seqno);
124 if (dh->dccph_type != DCCP_PKT_SYNC &&
125 (ackno != DCCP_PKT_WITHOUT_ACK_SEQ))
126 dp->dccps_gar = ackno;
128 unsigned long now = jiffies;
130 * Step 6: Check sequence numbers
132 * If P.type == Reset,
133 * Send Sync packet acknowledging S.GSR
135 * Send Sync packet acknowledging P.seqno
136 * Drop packet and return
138 * These Syncs are rate-limited as per RFC 4340, 7.5.4:
139 * at most 1 / (dccp_sync_rate_limit * HZ) Syncs per second.
141 if (time_before(now, (dp->dccps_rate_last +
142 sysctl_dccp_sync_ratelimit)))
145 DCCP_WARN("DCCP: Step 6 failed for %s packet, "
146 "(LSWL(%llu) <= P.seqno(%llu) <= S.SWH(%llu)) and "
147 "(P.ackno %s or LAWL(%llu) <= P.ackno(%llu) <= S.AWH(%llu), "
148 "sending SYNC...\n", dccp_packet_name(dh->dccph_type),
149 (unsigned long long) lswl, (unsigned long long) seqno,
150 (unsigned long long) dp->dccps_swh,
151 (ackno == DCCP_PKT_WITHOUT_ACK_SEQ) ? "doesn't exist"
153 (unsigned long long) lawl, (unsigned long long) ackno,
154 (unsigned long long) dp->dccps_awh);
156 dp->dccps_rate_last = now;
158 if (dh->dccph_type == DCCP_PKT_RESET)
159 seqno = dp->dccps_gsr;
160 dccp_send_sync(sk, seqno, DCCP_PKT_SYNC);
167 static int __dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
168 const struct dccp_hdr *dh, const unsigned len)
170 struct dccp_sock *dp = dccp_sk(sk);
172 switch (dccp_hdr(skb)->dccph_type) {
173 case DCCP_PKT_DATAACK:
176 * FIXME: check if sk_receive_queue is full, schedule DATA_DROPPED
179 __skb_pull(skb, dh->dccph_doff * 4);
180 __skb_queue_tail(&sk->sk_receive_queue, skb);
181 skb_set_owner_r(skb, sk);
182 sk->sk_data_ready(sk, 0);
188 * Step 9: Process Reset
189 * If P.type == Reset,
190 * Tear down connection
191 * S.state := TIMEWAIT
193 * Drop packet and return
196 dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
198 case DCCP_PKT_CLOSEREQ:
199 dccp_rcv_closereq(sk, skb);
202 dccp_rcv_close(sk, skb);
204 case DCCP_PKT_REQUEST:
206 * or (S.is_server and P.type == Response)
207 * or (S.is_client and P.type == Request)
208 * or (S.state >= OPEN and P.type == Request
209 * and P.seqno >= S.OSR)
210 * or (S.state >= OPEN and P.type == Response
211 * and P.seqno >= S.OSR)
212 * or (S.state == RESPOND and P.type == Data),
213 * Send Sync packet acknowledging P.seqno
214 * Drop packet and return
216 if (dp->dccps_role != DCCP_ROLE_LISTEN)
219 case DCCP_PKT_RESPONSE:
220 if (dp->dccps_role != DCCP_ROLE_CLIENT)
223 if (dccp_delta_seqno(dp->dccps_osr,
224 DCCP_SKB_CB(skb)->dccpd_seq) >= 0) {
226 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
231 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
234 * From RFC 4340, sec. 5.7
236 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets
237 * MAY have non-zero-length application data areas, whose
238 * contents receivers MUST ignore.
243 DCCP_INC_STATS_BH(DCCP_MIB_INERRS);
249 int dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
250 const struct dccp_hdr *dh, const unsigned len)
252 struct dccp_sock *dp = dccp_sk(sk);
254 if (dccp_check_seqno(sk, skb))
257 if (dccp_parse_options(sk, skb))
260 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
261 dccp_event_ack_recv(sk, skb);
263 if (dccp_msk(sk)->dccpms_send_ack_vector &&
264 dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
265 DCCP_SKB_CB(skb)->dccpd_seq,
266 DCCP_ACKVEC_STATE_RECEIVED))
269 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
270 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
272 return __dccp_rcv_established(sk, skb, dh, len);
278 EXPORT_SYMBOL_GPL(dccp_rcv_established);
280 static int dccp_rcv_request_sent_state_process(struct sock *sk,
282 const struct dccp_hdr *dh,
286 * Step 4: Prepare sequence numbers in REQUEST
287 * If S.state == REQUEST,
288 * If (P.type == Response or P.type == Reset)
289 * and S.AWL <= P.ackno <= S.AWH,
290 * / * Set sequence number variables corresponding to the
291 * other endpoint, so P will pass the tests in Step 6 * /
292 * Set S.GSR, S.ISR, S.SWL, S.SWH
293 * / * Response processing continues in Step 10; Reset
294 * processing continues in Step 9 * /
296 if (dh->dccph_type == DCCP_PKT_RESPONSE) {
297 const struct inet_connection_sock *icsk = inet_csk(sk);
298 struct dccp_sock *dp = dccp_sk(sk);
299 long tstamp = dccp_timestamp();
301 /* Stop the REQUEST timer */
302 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
303 BUG_TRAP(sk->sk_send_head != NULL);
304 __kfree_skb(sk->sk_send_head);
305 sk->sk_send_head = NULL;
307 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
308 dp->dccps_awl, dp->dccps_awh)) {
309 dccp_pr_debug("invalid ackno: S.AWL=%llu, "
310 "P.ackno=%llu, S.AWH=%llu \n",
311 (unsigned long long)dp->dccps_awl,
312 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
313 (unsigned long long)dp->dccps_awh);
314 goto out_invalid_packet;
317 if (dccp_parse_options(sk, skb))
318 goto out_invalid_packet;
320 /* Obtain usec RTT sample from SYN exchange (used by CCID 3) */
321 if (likely(dp->dccps_options_received.dccpor_timestamp_echo))
322 dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * (tstamp -
323 dp->dccps_options_received.dccpor_timestamp_echo));
325 if (dccp_msk(sk)->dccpms_send_ack_vector &&
326 dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
327 DCCP_SKB_CB(skb)->dccpd_seq,
328 DCCP_ACKVEC_STATE_RECEIVED))
329 goto out_invalid_packet; /* FIXME: change error code */
331 dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq;
332 dccp_update_gsr(sk, dp->dccps_isr);
334 * SWL and AWL are initially adjusted so that they are not less than
335 * the initial Sequence Numbers received and sent, respectively:
336 * SWL := max(GSR + 1 - floor(W/4), ISR),
337 * AWL := max(GSS - W' + 1, ISS).
338 * These adjustments MUST be applied only at the beginning of the
341 * AWL was adjusted in dccp_v4_connect -acme
343 dccp_set_seqno(&dp->dccps_swl,
344 max48(dp->dccps_swl, dp->dccps_isr));
346 dccp_sync_mss(sk, icsk->icsk_pmtu_cookie);
349 * Step 10: Process REQUEST state (second part)
350 * If S.state == REQUEST,
351 * / * If we get here, P is a valid Response from the
352 * server (see Step 4), and we should move to
353 * PARTOPEN state. PARTOPEN means send an Ack,
354 * don't send Data packets, retransmit Acks
355 * periodically, and always include any Init Cookie
356 * from the Response * /
357 * S.state := PARTOPEN
359 * Continue with S.state == PARTOPEN
360 * / * Step 12 will send the Ack completing the
361 * three-way handshake * /
363 dccp_set_state(sk, DCCP_PARTOPEN);
365 /* Make sure socket is routed, for correct metrics. */
366 icsk->icsk_af_ops->rebuild_header(sk);
368 if (!sock_flag(sk, SOCK_DEAD)) {
369 sk->sk_state_change(sk);
370 sk_wake_async(sk, 0, POLL_OUT);
373 if (sk->sk_write_pending || icsk->icsk_ack.pingpong ||
374 icsk->icsk_accept_queue.rskq_defer_accept) {
375 /* Save one ACK. Data will be ready after
376 * several ticks, if write_pending is set.
378 * It may be deleted, but with this feature tcpdumps
379 * look so _wonderfully_ clever, that I was not able
380 * to stand against the temptation 8) --ANK
383 * OK, in DCCP we can as well do a similar trick, its
384 * even in the draft, but there is no need for us to
385 * schedule an ack here, as dccp_sendmsg does this for
386 * us, also stated in the draft. -acme
396 /* dccp_v4_do_rcv will send a reset */
397 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
401 static int dccp_rcv_respond_partopen_state_process(struct sock *sk,
403 const struct dccp_hdr *dh,
408 switch (dh->dccph_type) {
410 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
413 if (sk->sk_state == DCCP_RESPOND)
415 case DCCP_PKT_DATAACK:
418 * FIXME: we should be reseting the PARTOPEN (DELACK) timer
419 * here but only if we haven't used the DELACK timer for
420 * something else, like sending a delayed ack for a TIMESTAMP
421 * echo, etc, for now were not clearing it, sending an extra
422 * ACK when there is nothing else to do in DELACK is not a big
426 /* Stop the PARTOPEN timer */
427 if (sk->sk_state == DCCP_PARTOPEN)
428 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
430 dccp_sk(sk)->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq;
431 dccp_set_state(sk, DCCP_OPEN);
433 if (dh->dccph_type == DCCP_PKT_DATAACK ||
434 dh->dccph_type == DCCP_PKT_DATA) {
435 __dccp_rcv_established(sk, skb, dh, len);
436 queued = 1; /* packet was queued
437 (by __dccp_rcv_established) */
445 int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
446 struct dccp_hdr *dh, unsigned len)
448 struct dccp_sock *dp = dccp_sk(sk);
449 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
450 const int old_state = sk->sk_state;
454 * Step 3: Process LISTEN state
456 * If S.state == LISTEN,
457 * If P.type == Request or P contains a valid Init Cookie option,
458 * (* Must scan the packet's options to check for Init
459 * Cookies. Only Init Cookies are processed here,
460 * however; other options are processed in Step 8. This
461 * scan need only be performed if the endpoint uses Init
463 * (* Generate a new socket and switch to that socket *)
464 * Set S := new socket for this port pair
466 * Choose S.ISS (initial seqno) or set from Init Cookies
467 * Initialize S.GAR := S.ISS
468 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
469 * Cookies Continue with S.state == RESPOND
470 * (* A Response packet will be generated in Step 11 *)
472 * Generate Reset(No Connection) unless P.type == Reset
473 * Drop packet and return
475 if (sk->sk_state == DCCP_LISTEN) {
476 if (dh->dccph_type == DCCP_PKT_REQUEST) {
477 if (inet_csk(sk)->icsk_af_ops->conn_request(sk,
481 /* FIXME: do congestion control initialization */
484 if (dh->dccph_type == DCCP_PKT_RESET)
487 /* Caller (dccp_v4_do_rcv) will send Reset */
488 dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
492 if (sk->sk_state != DCCP_REQUESTING) {
493 if (dccp_check_seqno(sk, skb))
497 * Step 8: Process options and mark acknowledgeable
499 if (dccp_parse_options(sk, skb))
502 if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
503 dccp_event_ack_recv(sk, skb);
505 if (dccp_msk(sk)->dccpms_send_ack_vector &&
506 dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
507 DCCP_SKB_CB(skb)->dccpd_seq,
508 DCCP_ACKVEC_STATE_RECEIVED))
511 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
512 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
516 * Step 9: Process Reset
517 * If P.type == Reset,
518 * Tear down connection
519 * S.state := TIMEWAIT
521 * Drop packet and return
523 if (dh->dccph_type == DCCP_PKT_RESET) {
525 * Queue the equivalent of TCP fin so that dccp_recvmsg
529 dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
532 * Step 7: Check for unexpected packet types
533 * If (S.is_server and P.type == CloseReq)
534 * or (S.is_server and P.type == Response)
535 * or (S.is_client and P.type == Request)
536 * or (S.state == RESPOND and P.type == Data),
537 * Send Sync packet acknowledging P.seqno
538 * Drop packet and return
540 } else if ((dp->dccps_role != DCCP_ROLE_CLIENT &&
541 (dh->dccph_type == DCCP_PKT_RESPONSE ||
542 dh->dccph_type == DCCP_PKT_CLOSEREQ)) ||
543 (dp->dccps_role == DCCP_ROLE_CLIENT &&
544 dh->dccph_type == DCCP_PKT_REQUEST) ||
545 (sk->sk_state == DCCP_RESPOND &&
546 dh->dccph_type == DCCP_PKT_DATA)) {
547 dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNC);
549 } else if (dh->dccph_type == DCCP_PKT_CLOSEREQ) {
550 dccp_rcv_closereq(sk, skb);
552 } else if (dh->dccph_type == DCCP_PKT_CLOSE) {
553 dccp_rcv_close(sk, skb);
557 switch (sk->sk_state) {
559 dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
562 case DCCP_REQUESTING:
563 /* FIXME: do congestion control initialization */
565 queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len);
574 queued = dccp_rcv_respond_partopen_state_process(sk, skb,
579 if (dh->dccph_type == DCCP_PKT_ACK ||
580 dh->dccph_type == DCCP_PKT_DATAACK) {
583 sk->sk_state_change(sk);
584 sk_wake_async(sk, 0, POLL_OUT);
587 } else if (unlikely(dh->dccph_type == DCCP_PKT_SYNC)) {
588 dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNCACK);
599 EXPORT_SYMBOL_GPL(dccp_rcv_state_process);
602 * dccp_sample_rtt - Validate and finalise computation of RTT sample
603 * @delta: number of microseconds between packet and acknowledgment
604 * The routine is kept generic to work in different contexts. It should be
605 * called immediately when the ACK used for the RTT sample arrives.
607 u32 dccp_sample_rtt(struct sock *sk, long delta)
609 /* dccpor_elapsed_time is either zeroed out or set and > 0 */
610 delta -= dccp_sk(sk)->dccps_options_received.dccpor_elapsed_time * 10;
612 if (unlikely(delta <= 0)) {
613 DCCP_WARN("unusable RTT sample %ld, using min\n", delta);
614 return DCCP_SANE_RTT_MIN;
616 if (unlikely(delta > DCCP_SANE_RTT_MAX)) {
617 DCCP_WARN("RTT sample %ld too large, using max\n", delta);
618 return DCCP_SANE_RTT_MAX;
624 EXPORT_SYMBOL_GPL(dccp_sample_rtt);