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 lswl, DCCP_SKB_CB(skb)->dccpd_seq, dp->dccps_swh,
145 (DCCP_SKB_CB(skb)->dccpd_ack_seq ==
146 DCCP_PKT_WITHOUT_ACK_SEQ) ? "doesn't exist" : "exists",
147 lawl, DCCP_SKB_CB(skb)->dccpd_ack_seq, dp->dccps_awh);
148 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, DCCP_PKT_SYNC);
155 int dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
156 const struct dccp_hdr *dh, const unsigned len)
158 struct dccp_sock *dp = dccp_sk(sk);
160 if (dccp_check_seqno(sk, skb))
163 if (dccp_parse_options(sk, skb))
166 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
167 dccp_event_ack_recv(sk, skb);
170 * FIXME: check ECN to see if we should use
171 * DCCP_ACKPKTS_STATE_ECN_MARKED
173 if (dp->dccps_options.dccpo_send_ack_vector) {
174 struct dccp_ackpkts *ap = dp->dccps_hc_rx_ackpkts;
176 if (dccp_ackpkts_add(dp->dccps_hc_rx_ackpkts,
177 DCCP_SKB_CB(skb)->dccpd_seq,
178 DCCP_ACKPKTS_STATE_RECEIVED)) {
179 LIMIT_NETDEBUG(KERN_WARNING "DCCP: acknowledgeable "
180 "packets buffer full!\n");
181 ap->dccpap_ack_seqno = DCCP_MAX_SEQNO + 1;
182 inet_csk_schedule_ack(sk);
183 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
190 * FIXME: this activation is probably wrong, have to study more
191 * TCP delack machinery and how it fits into DCCP draft, but
192 * for now it kinda "works" 8)
194 if (!inet_csk_ack_scheduled(sk)) {
195 inet_csk_schedule_ack(sk);
196 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 5 * HZ,
201 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
202 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
204 switch (dccp_hdr(skb)->dccph_type) {
205 case DCCP_PKT_DATAACK:
208 * FIXME: check if sk_receive_queue is full, schedule DATA_DROPPED
211 __skb_pull(skb, dh->dccph_doff * 4);
212 __skb_queue_tail(&sk->sk_receive_queue, skb);
213 skb_set_owner_r(skb, sk);
214 sk->sk_data_ready(sk, 0);
220 * Step 9: Process Reset
221 * If P.type == Reset,
222 * Tear down connection
223 * S.state := TIMEWAIT
225 * Drop packet and return
228 dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
230 case DCCP_PKT_CLOSEREQ:
231 dccp_rcv_closereq(sk, skb);
234 dccp_rcv_close(sk, skb);
236 case DCCP_PKT_REQUEST:
238 * or (S.is_server and P.type == Response)
239 * or (S.is_client and P.type == Request)
240 * or (S.state >= OPEN and P.type == Request
241 * and P.seqno >= S.OSR)
242 * or (S.state >= OPEN and P.type == Response
243 * and P.seqno >= S.OSR)
244 * or (S.state == RESPOND and P.type == Data),
245 * Send Sync packet acknowledging P.seqno
246 * Drop packet and return
248 if (dp->dccps_role != DCCP_ROLE_LISTEN)
251 case DCCP_PKT_RESPONSE:
252 if (dp->dccps_role != DCCP_ROLE_CLIENT)
255 if (!before48(DCCP_SKB_CB(skb)->dccpd_seq, dp->dccps_osr)) {
257 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
262 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
267 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets
268 * MAY have non-zero-length application data areas, whose
269 * contents * receivers MUST ignore.
274 DCCP_INC_STATS_BH(DCCP_MIB_INERRS);
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);
300 /* Stop the REQUEST timer */
301 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
302 BUG_TRAP(sk->sk_send_head != NULL);
303 __kfree_skb(sk->sk_send_head);
304 sk->sk_send_head = NULL;
306 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
307 dp->dccps_awl, dp->dccps_awh)) {
308 dccp_pr_debug("invalid ackno: S.AWL=%llu, "
309 "P.ackno=%llu, S.AWH=%llu \n",
310 (unsigned long long)dp->dccps_awl,
311 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
312 (unsigned long long)dp->dccps_awh);
313 goto out_invalid_packet;
316 dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq;
317 dccp_update_gsr(sk, dp->dccps_isr);
319 * SWL and AWL are initially adjusted so that they are not less than
320 * the initial Sequence Numbers received and sent, respectively:
321 * SWL := max(GSR + 1 - floor(W/4), ISR),
322 * AWL := max(GSS - W' + 1, ISS).
323 * These adjustments MUST be applied only at the beginning of the
326 * AWL was adjusted in dccp_v4_connect -acme
328 dccp_set_seqno(&dp->dccps_swl,
329 max48(dp->dccps_swl, dp->dccps_isr));
331 if (ccid_hc_rx_init(dp->dccps_hc_rx_ccid, sk) != 0 ||
332 ccid_hc_tx_init(dp->dccps_hc_tx_ccid, sk) != 0) {
333 ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
334 ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
335 /* FIXME: send appropriate RESET code */
336 goto out_invalid_packet;
339 dccp_sync_mss(sk, dp->dccps_pmtu_cookie);
342 * Step 10: Process REQUEST state (second part)
343 * If S.state == REQUEST,
344 * / * If we get here, P is a valid Response from the
345 * server (see Step 4), and we should move to
346 * PARTOPEN state. PARTOPEN means send an Ack,
347 * don't send Data packets, retransmit Acks
348 * periodically, and always include any Init Cookie
349 * from the Response * /
350 * S.state := PARTOPEN
352 * Continue with S.state == PARTOPEN
353 * / * Step 12 will send the Ack completing the
354 * three-way handshake * /
356 dccp_set_state(sk, DCCP_PARTOPEN);
358 /* Make sure socket is routed, for correct metrics. */
359 inet_sk_rebuild_header(sk);
361 if (!sock_flag(sk, SOCK_DEAD)) {
362 sk->sk_state_change(sk);
363 sk_wake_async(sk, 0, POLL_OUT);
366 if (sk->sk_write_pending || icsk->icsk_ack.pingpong ||
367 icsk->icsk_accept_queue.rskq_defer_accept) {
368 /* Save one ACK. Data will be ready after
369 * several ticks, if write_pending is set.
371 * It may be deleted, but with this feature tcpdumps
372 * look so _wonderfully_ clever, that I was not able
373 * to stand against the temptation 8) --ANK
376 * OK, in DCCP we can as well do a similar trick, its
377 * even in the draft, but there is no need for us to
378 * schedule an ack here, as dccp_sendmsg does this for
379 * us, also stated in the draft. -acme
389 return 1; /* dccp_v4_do_rcv will send a reset, but...
390 FIXME: the reset code should be
391 DCCP_RESET_CODE_PACKET_ERROR */
394 static int dccp_rcv_respond_partopen_state_process(struct sock *sk,
396 const struct dccp_hdr *dh,
401 switch (dh->dccph_type) {
403 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
405 case DCCP_PKT_DATAACK:
408 * FIXME: we should be reseting the PARTOPEN (DELACK) timer
409 * here but only if we haven't used the DELACK timer for
410 * something else, like sending a delayed ack for a TIMESTAMP
411 * echo, etc, for now were not clearing it, sending an extra
412 * ACK when there is nothing else to do in DELACK is not a big
416 /* Stop the PARTOPEN timer */
417 if (sk->sk_state == DCCP_PARTOPEN)
418 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
420 dccp_sk(sk)->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq;
421 dccp_set_state(sk, DCCP_OPEN);
423 if (dh->dccph_type == DCCP_PKT_DATAACK) {
424 dccp_rcv_established(sk, skb, dh, len);
425 queued = 1; /* packet was queued
426 (by dccp_rcv_established) */
434 int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
435 struct dccp_hdr *dh, unsigned len)
437 struct dccp_sock *dp = dccp_sk(sk);
438 const int old_state = sk->sk_state;
442 * Step 3: Process LISTEN state
443 * (Continuing from dccp_v4_do_rcv and dccp_v6_do_rcv)
445 * If S.state == LISTEN,
446 * If P.type == Request or P contains a valid Init Cookie
448 * * Must scan the packet's options to check for an Init
449 * Cookie. Only the Init Cookie is processed here,
450 * however; other options are processed in Step 8. This
451 * scan need only be performed if the endpoint uses Init
453 * * Generate a new socket and switch to that socket *
454 * Set S := new socket for this port pair
456 * Choose S.ISS (initial seqno) or set from Init Cookie
457 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
458 * Continue with S.state == RESPOND
459 * * A Response packet will be generated in Step 11 *
461 * Generate Reset(No Connection) unless P.type == Reset
462 * Drop packet and return
464 * NOTE: the check for the packet types is done in
465 * dccp_rcv_state_process
467 if (sk->sk_state == DCCP_LISTEN) {
468 if (dh->dccph_type == DCCP_PKT_REQUEST) {
469 if (dccp_v4_conn_request(sk, skb) < 0)
472 /* FIXME: do congestion control initialization */
475 if (dh->dccph_type == DCCP_PKT_RESET)
478 /* Caller (dccp_v4_do_rcv) will send Reset(No Connection)*/
482 if (sk->sk_state != DCCP_REQUESTING) {
483 if (dccp_check_seqno(sk, skb))
487 * Step 8: Process options and mark acknowledgeable
489 if (dccp_parse_options(sk, skb))
492 if (DCCP_SKB_CB(skb)->dccpd_ack_seq !=
493 DCCP_PKT_WITHOUT_ACK_SEQ)
494 dccp_event_ack_recv(sk, skb);
496 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
497 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
500 * FIXME: check ECN to see if we should use
501 * DCCP_ACKPKTS_STATE_ECN_MARKED
503 if (dp->dccps_options.dccpo_send_ack_vector) {
504 if (dccp_ackpkts_add(dp->dccps_hc_rx_ackpkts,
505 DCCP_SKB_CB(skb)->dccpd_seq,
506 DCCP_ACKPKTS_STATE_RECEIVED))
509 * FIXME: this activation is probably wrong, have to
510 * study more TCP delack machinery and how it fits into
511 * DCCP draft, but for now it kinda "works" 8)
513 if ((dp->dccps_hc_rx_ackpkts->dccpap_ack_seqno ==
514 DCCP_MAX_SEQNO + 1) &&
515 !inet_csk_ack_scheduled(sk)) {
516 inet_csk_schedule_ack(sk);
517 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
525 * Step 9: Process Reset
526 * If P.type == Reset,
527 * Tear down connection
528 * S.state := TIMEWAIT
530 * Drop packet and return
532 if (dh->dccph_type == DCCP_PKT_RESET) {
534 * Queue the equivalent of TCP fin so that dccp_recvmsg
538 dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
541 * Step 7: Check for unexpected packet types
542 * If (S.is_server and P.type == CloseReq)
543 * or (S.is_server and P.type == Response)
544 * or (S.is_client and P.type == Request)
545 * or (S.state == RESPOND and P.type == Data),
546 * Send Sync packet acknowledging P.seqno
547 * Drop packet and return
549 } else if ((dp->dccps_role != DCCP_ROLE_CLIENT &&
550 (dh->dccph_type == DCCP_PKT_RESPONSE ||
551 dh->dccph_type == DCCP_PKT_CLOSEREQ)) ||
552 (dp->dccps_role == DCCP_ROLE_CLIENT &&
553 dh->dccph_type == DCCP_PKT_REQUEST) ||
554 (sk->sk_state == DCCP_RESPOND &&
555 dh->dccph_type == DCCP_PKT_DATA)) {
556 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
561 switch (sk->sk_state) {
565 case DCCP_REQUESTING:
566 /* FIXME: do congestion control initialization */
568 queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len);
577 queued = dccp_rcv_respond_partopen_state_process(sk, skb,
582 if (dh->dccph_type == DCCP_PKT_ACK ||
583 dh->dccph_type == DCCP_PKT_DATAACK) {
586 sk->sk_state_change(sk);
587 sk_wake_async(sk, 0, POLL_OUT);