xref: /linux/net/dccp/input.c (revision cdb138080b78146d1cdadba9f5dadbeb97445b91)
1 /*
2  *  net/dccp/input.c
3  *
4  *  An implementation of the DCCP protocol
5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
6  *
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.
11  */
12 
13 #include <linux/dccp.h>
14 #include <linux/skbuff.h>
15 #include <linux/slab.h>
16 
17 #include <net/sock.h>
18 
19 #include "ackvec.h"
20 #include "ccid.h"
21 #include "dccp.h"
22 
23 /* rate-limit for syncs in reply to sequence-invalid packets; RFC 4340, 7.5.4 */
24 int sysctl_dccp_sync_ratelimit	__read_mostly = HZ / 8;
25 
26 static void dccp_enqueue_skb(struct sock *sk, struct sk_buff *skb)
27 {
28 	__skb_pull(skb, dccp_hdr(skb)->dccph_doff * 4);
29 	__skb_queue_tail(&sk->sk_receive_queue, skb);
30 	skb_set_owner_r(skb, sk);
31 	sk->sk_data_ready(sk, 0);
32 }
33 
34 static void dccp_fin(struct sock *sk, struct sk_buff *skb)
35 {
36 	/*
37 	 * On receiving Close/CloseReq, both RD/WR shutdown are performed.
38 	 * RFC 4340, 8.3 says that we MAY send further Data/DataAcks after
39 	 * receiving the closing segment, but there is no guarantee that such
40 	 * data will be processed at all.
41 	 */
42 	sk->sk_shutdown = SHUTDOWN_MASK;
43 	sock_set_flag(sk, SOCK_DONE);
44 	dccp_enqueue_skb(sk, skb);
45 }
46 
47 static int dccp_rcv_close(struct sock *sk, struct sk_buff *skb)
48 {
49 	int queued = 0;
50 
51 	switch (sk->sk_state) {
52 	/*
53 	 * We ignore Close when received in one of the following states:
54 	 *  - CLOSED		(may be a late or duplicate packet)
55 	 *  - PASSIVE_CLOSEREQ	(the peer has sent a CloseReq earlier)
56 	 *  - RESPOND		(already handled by dccp_check_req)
57 	 */
58 	case DCCP_CLOSING:
59 		/*
60 		 * Simultaneous-close: receiving a Close after sending one. This
61 		 * can happen if both client and server perform active-close and
62 		 * will result in an endless ping-pong of crossing and retrans-
63 		 * mitted Close packets, which only terminates when one of the
64 		 * nodes times out (min. 64 seconds). Quicker convergence can be
65 		 * achieved when one of the nodes acts as tie-breaker.
66 		 * This is ok as both ends are done with data transfer and each
67 		 * end is just waiting for the other to acknowledge termination.
68 		 */
69 		if (dccp_sk(sk)->dccps_role != DCCP_ROLE_CLIENT)
70 			break;
71 		/* fall through */
72 	case DCCP_REQUESTING:
73 	case DCCP_ACTIVE_CLOSEREQ:
74 		dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
75 		dccp_done(sk);
76 		break;
77 	case DCCP_OPEN:
78 	case DCCP_PARTOPEN:
79 		/* Give waiting application a chance to read pending data */
80 		queued = 1;
81 		dccp_fin(sk, skb);
82 		dccp_set_state(sk, DCCP_PASSIVE_CLOSE);
83 		/* fall through */
84 	case DCCP_PASSIVE_CLOSE:
85 		/*
86 		 * Retransmitted Close: we have already enqueued the first one.
87 		 */
88 		sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
89 	}
90 	return queued;
91 }
92 
93 static int dccp_rcv_closereq(struct sock *sk, struct sk_buff *skb)
94 {
95 	int queued = 0;
96 
97 	/*
98 	 *   Step 7: Check for unexpected packet types
99 	 *      If (S.is_server and P.type == CloseReq)
100 	 *	  Send Sync packet acknowledging P.seqno
101 	 *	  Drop packet and return
102 	 */
103 	if (dccp_sk(sk)->dccps_role != DCCP_ROLE_CLIENT) {
104 		dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, DCCP_PKT_SYNC);
105 		return queued;
106 	}
107 
108 	/* Step 13: process relevant Client states < CLOSEREQ */
109 	switch (sk->sk_state) {
110 	case DCCP_REQUESTING:
111 		dccp_send_close(sk, 0);
112 		dccp_set_state(sk, DCCP_CLOSING);
113 		break;
114 	case DCCP_OPEN:
115 	case DCCP_PARTOPEN:
116 		/* Give waiting application a chance to read pending data */
117 		queued = 1;
118 		dccp_fin(sk, skb);
119 		dccp_set_state(sk, DCCP_PASSIVE_CLOSEREQ);
120 		/* fall through */
121 	case DCCP_PASSIVE_CLOSEREQ:
122 		sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
123 	}
124 	return queued;
125 }
126 
127 static u16 dccp_reset_code_convert(const u8 code)
128 {
129 	const u16 error_code[] = {
130 	[DCCP_RESET_CODE_CLOSED]	     = 0,	/* normal termination */
131 	[DCCP_RESET_CODE_UNSPECIFIED]	     = 0,	/* nothing known */
132 	[DCCP_RESET_CODE_ABORTED]	     = ECONNRESET,
133 
134 	[DCCP_RESET_CODE_NO_CONNECTION]	     = ECONNREFUSED,
135 	[DCCP_RESET_CODE_CONNECTION_REFUSED] = ECONNREFUSED,
136 	[DCCP_RESET_CODE_TOO_BUSY]	     = EUSERS,
137 	[DCCP_RESET_CODE_AGGRESSION_PENALTY] = EDQUOT,
138 
139 	[DCCP_RESET_CODE_PACKET_ERROR]	     = ENOMSG,
140 	[DCCP_RESET_CODE_BAD_INIT_COOKIE]    = EBADR,
141 	[DCCP_RESET_CODE_BAD_SERVICE_CODE]   = EBADRQC,
142 	[DCCP_RESET_CODE_OPTION_ERROR]	     = EILSEQ,
143 	[DCCP_RESET_CODE_MANDATORY_ERROR]    = EOPNOTSUPP,
144 	};
145 
146 	return code >= DCCP_MAX_RESET_CODES ? 0 : error_code[code];
147 }
148 
149 static void dccp_rcv_reset(struct sock *sk, struct sk_buff *skb)
150 {
151 	u16 err = dccp_reset_code_convert(dccp_hdr_reset(skb)->dccph_reset_code);
152 
153 	sk->sk_err = err;
154 
155 	/* Queue the equivalent of TCP fin so that dccp_recvmsg exits the loop */
156 	dccp_fin(sk, skb);
157 
158 	if (err && !sock_flag(sk, SOCK_DEAD))
159 		sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
160 	dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
161 }
162 
163 static void dccp_event_ack_recv(struct sock *sk, struct sk_buff *skb)
164 {
165 	struct dccp_sock *dp = dccp_sk(sk);
166 
167 	if (dp->dccps_hc_rx_ackvec != NULL)
168 		dccp_ackvec_check_rcv_ackno(dp->dccps_hc_rx_ackvec, sk,
169 					    DCCP_SKB_CB(skb)->dccpd_ack_seq);
170 }
171 
172 static void dccp_deliver_input_to_ccids(struct sock *sk, struct sk_buff *skb)
173 {
174 	const struct dccp_sock *dp = dccp_sk(sk);
175 
176 	/* Don't deliver to RX CCID when node has shut down read end. */
177 	if (!(sk->sk_shutdown & RCV_SHUTDOWN))
178 		ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
179 	/*
180 	 * Until the TX queue has been drained, we can not honour SHUT_WR, since
181 	 * we need received feedback as input to adjust congestion control.
182 	 */
183 	if (sk->sk_write_queue.qlen > 0 || !(sk->sk_shutdown & SEND_SHUTDOWN))
184 		ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
185 }
186 
187 static int dccp_check_seqno(struct sock *sk, struct sk_buff *skb)
188 {
189 	const struct dccp_hdr *dh = dccp_hdr(skb);
190 	struct dccp_sock *dp = dccp_sk(sk);
191 	u64 lswl, lawl, seqno = DCCP_SKB_CB(skb)->dccpd_seq,
192 			ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
193 
194 	/*
195 	 *   Step 5: Prepare sequence numbers for Sync
196 	 *     If P.type == Sync or P.type == SyncAck,
197 	 *	  If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL,
198 	 *	     / * P is valid, so update sequence number variables
199 	 *		 accordingly.  After this update, P will pass the tests
200 	 *		 in Step 6.  A SyncAck is generated if necessary in
201 	 *		 Step 15 * /
202 	 *	     Update S.GSR, S.SWL, S.SWH
203 	 *	  Otherwise,
204 	 *	     Drop packet and return
205 	 */
206 	if (dh->dccph_type == DCCP_PKT_SYNC ||
207 	    dh->dccph_type == DCCP_PKT_SYNCACK) {
208 		if (between48(ackno, dp->dccps_awl, dp->dccps_awh) &&
209 		    dccp_delta_seqno(dp->dccps_swl, seqno) >= 0)
210 			dccp_update_gsr(sk, seqno);
211 		else
212 			return -1;
213 	}
214 
215 	/*
216 	 *   Step 6: Check sequence numbers
217 	 *      Let LSWL = S.SWL and LAWL = S.AWL
218 	 *      If P.type == CloseReq or P.type == Close or P.type == Reset,
219 	 *	  LSWL := S.GSR + 1, LAWL := S.GAR
220 	 *      If LSWL <= P.seqno <= S.SWH
221 	 *	     and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH),
222 	 *	  Update S.GSR, S.SWL, S.SWH
223 	 *	  If P.type != Sync,
224 	 *	     Update S.GAR
225 	 */
226 	lswl = dp->dccps_swl;
227 	lawl = dp->dccps_awl;
228 
229 	if (dh->dccph_type == DCCP_PKT_CLOSEREQ ||
230 	    dh->dccph_type == DCCP_PKT_CLOSE ||
231 	    dh->dccph_type == DCCP_PKT_RESET) {
232 		lswl = ADD48(dp->dccps_gsr, 1);
233 		lawl = dp->dccps_gar;
234 	}
235 
236 	if (between48(seqno, lswl, dp->dccps_swh) &&
237 	    (ackno == DCCP_PKT_WITHOUT_ACK_SEQ ||
238 	     between48(ackno, lawl, dp->dccps_awh))) {
239 		dccp_update_gsr(sk, seqno);
240 
241 		if (dh->dccph_type != DCCP_PKT_SYNC &&
242 		    (ackno != DCCP_PKT_WITHOUT_ACK_SEQ))
243 			dp->dccps_gar = ackno;
244 	} else {
245 		unsigned long now = jiffies;
246 		/*
247 		 *   Step 6: Check sequence numbers
248 		 *      Otherwise,
249 		 *         If P.type == Reset,
250 		 *            Send Sync packet acknowledging S.GSR
251 		 *         Otherwise,
252 		 *            Send Sync packet acknowledging P.seqno
253 		 *      Drop packet and return
254 		 *
255 		 *   These Syncs are rate-limited as per RFC 4340, 7.5.4:
256 		 *   at most 1 / (dccp_sync_rate_limit * HZ) Syncs per second.
257 		 */
258 		if (time_before(now, (dp->dccps_rate_last +
259 				      sysctl_dccp_sync_ratelimit)))
260 			return 0;
261 
262 		DCCP_WARN("DCCP: Step 6 failed for %s packet, "
263 			  "(LSWL(%llu) <= P.seqno(%llu) <= S.SWH(%llu)) and "
264 			  "(P.ackno %s or LAWL(%llu) <= P.ackno(%llu) <= S.AWH(%llu), "
265 			  "sending SYNC...\n",  dccp_packet_name(dh->dccph_type),
266 			  (unsigned long long) lswl, (unsigned long long) seqno,
267 			  (unsigned long long) dp->dccps_swh,
268 			  (ackno == DCCP_PKT_WITHOUT_ACK_SEQ) ? "doesn't exist"
269 							      : "exists",
270 			  (unsigned long long) lawl, (unsigned long long) ackno,
271 			  (unsigned long long) dp->dccps_awh);
272 
273 		dp->dccps_rate_last = now;
274 
275 		if (dh->dccph_type == DCCP_PKT_RESET)
276 			seqno = dp->dccps_gsr;
277 		dccp_send_sync(sk, seqno, DCCP_PKT_SYNC);
278 		return -1;
279 	}
280 
281 	return 0;
282 }
283 
284 static int __dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
285 				  const struct dccp_hdr *dh, const unsigned len)
286 {
287 	struct dccp_sock *dp = dccp_sk(sk);
288 
289 	switch (dccp_hdr(skb)->dccph_type) {
290 	case DCCP_PKT_DATAACK:
291 	case DCCP_PKT_DATA:
292 		/*
293 		 * FIXME: schedule DATA_DROPPED (RFC 4340, 11.7.2) if and when
294 		 * - sk_shutdown == RCV_SHUTDOWN, use Code 1, "Not Listening"
295 		 * - sk_receive_queue is full, use Code 2, "Receive Buffer"
296 		 */
297 		dccp_enqueue_skb(sk, skb);
298 		return 0;
299 	case DCCP_PKT_ACK:
300 		goto discard;
301 	case DCCP_PKT_RESET:
302 		/*
303 		 *  Step 9: Process Reset
304 		 *	If P.type == Reset,
305 		 *		Tear down connection
306 		 *		S.state := TIMEWAIT
307 		 *		Set TIMEWAIT timer
308 		 *		Drop packet and return
309 		 */
310 		dccp_rcv_reset(sk, skb);
311 		return 0;
312 	case DCCP_PKT_CLOSEREQ:
313 		if (dccp_rcv_closereq(sk, skb))
314 			return 0;
315 		goto discard;
316 	case DCCP_PKT_CLOSE:
317 		if (dccp_rcv_close(sk, skb))
318 			return 0;
319 		goto discard;
320 	case DCCP_PKT_REQUEST:
321 		/* Step 7
322 		 *   or (S.is_server and P.type == Response)
323 		 *   or (S.is_client and P.type == Request)
324 		 *   or (S.state >= OPEN and P.type == Request
325 		 *	and P.seqno >= S.OSR)
326 		 *    or (S.state >= OPEN and P.type == Response
327 		 *	and P.seqno >= S.OSR)
328 		 *    or (S.state == RESPOND and P.type == Data),
329 		 *  Send Sync packet acknowledging P.seqno
330 		 *  Drop packet and return
331 		 */
332 		if (dp->dccps_role != DCCP_ROLE_LISTEN)
333 			goto send_sync;
334 		goto check_seq;
335 	case DCCP_PKT_RESPONSE:
336 		if (dp->dccps_role != DCCP_ROLE_CLIENT)
337 			goto send_sync;
338 check_seq:
339 		if (dccp_delta_seqno(dp->dccps_osr,
340 				     DCCP_SKB_CB(skb)->dccpd_seq) >= 0) {
341 send_sync:
342 			dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
343 				       DCCP_PKT_SYNC);
344 		}
345 		break;
346 	case DCCP_PKT_SYNC:
347 		dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
348 			       DCCP_PKT_SYNCACK);
349 		/*
350 		 * From RFC 4340, sec. 5.7
351 		 *
352 		 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets
353 		 * MAY have non-zero-length application data areas, whose
354 		 * contents receivers MUST ignore.
355 		 */
356 		goto discard;
357 	}
358 
359 	DCCP_INC_STATS_BH(DCCP_MIB_INERRS);
360 discard:
361 	__kfree_skb(skb);
362 	return 0;
363 }
364 
365 int dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
366 			 const struct dccp_hdr *dh, const unsigned len)
367 {
368 	struct dccp_sock *dp = dccp_sk(sk);
369 
370 	if (dccp_check_seqno(sk, skb))
371 		goto discard;
372 
373 	if (dccp_parse_options(sk, NULL, skb))
374 		return 1;
375 
376 	if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
377 		dccp_event_ack_recv(sk, skb);
378 
379 	if (dp->dccps_hc_rx_ackvec != NULL &&
380 	    dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
381 			    DCCP_SKB_CB(skb)->dccpd_seq,
382 			    DCCP_ACKVEC_STATE_RECEIVED))
383 		goto discard;
384 	dccp_deliver_input_to_ccids(sk, skb);
385 
386 	return __dccp_rcv_established(sk, skb, dh, len);
387 discard:
388 	__kfree_skb(skb);
389 	return 0;
390 }
391 
392 EXPORT_SYMBOL_GPL(dccp_rcv_established);
393 
394 static int dccp_rcv_request_sent_state_process(struct sock *sk,
395 					       struct sk_buff *skb,
396 					       const struct dccp_hdr *dh,
397 					       const unsigned len)
398 {
399 	/*
400 	 *  Step 4: Prepare sequence numbers in REQUEST
401 	 *     If S.state == REQUEST,
402 	 *	  If (P.type == Response or P.type == Reset)
403 	 *		and S.AWL <= P.ackno <= S.AWH,
404 	 *	     / * Set sequence number variables corresponding to the
405 	 *		other endpoint, so P will pass the tests in Step 6 * /
406 	 *	     Set S.GSR, S.ISR, S.SWL, S.SWH
407 	 *	     / * Response processing continues in Step 10; Reset
408 	 *		processing continues in Step 9 * /
409 	*/
410 	if (dh->dccph_type == DCCP_PKT_RESPONSE) {
411 		const struct inet_connection_sock *icsk = inet_csk(sk);
412 		struct dccp_sock *dp = dccp_sk(sk);
413 		long tstamp = dccp_timestamp();
414 
415 		if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
416 			       dp->dccps_awl, dp->dccps_awh)) {
417 			dccp_pr_debug("invalid ackno: S.AWL=%llu, "
418 				      "P.ackno=%llu, S.AWH=%llu\n",
419 				      (unsigned long long)dp->dccps_awl,
420 			   (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
421 				      (unsigned long long)dp->dccps_awh);
422 			goto out_invalid_packet;
423 		}
424 
425 		/*
426 		 * If option processing (Step 8) failed, return 1 here so that
427 		 * dccp_v4_do_rcv() sends a Reset. The Reset code depends on
428 		 * the option type and is set in dccp_parse_options().
429 		 */
430 		if (dccp_parse_options(sk, NULL, skb))
431 			return 1;
432 
433 		/* Obtain usec RTT sample from SYN exchange (used by TFRC). */
434 		if (likely(dp->dccps_options_received.dccpor_timestamp_echo))
435 			dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * (tstamp -
436 			    dp->dccps_options_received.dccpor_timestamp_echo));
437 
438 		/* Stop the REQUEST timer */
439 		inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
440 		WARN_ON(sk->sk_send_head == NULL);
441 		kfree_skb(sk->sk_send_head);
442 		sk->sk_send_head = NULL;
443 
444 		dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq;
445 		dccp_update_gsr(sk, dp->dccps_isr);
446 		/*
447 		 * SWL and AWL are initially adjusted so that they are not less than
448 		 * the initial Sequence Numbers received and sent, respectively:
449 		 *	SWL := max(GSR + 1 - floor(W/4), ISR),
450 		 *	AWL := max(GSS - W' + 1, ISS).
451 		 * These adjustments MUST be applied only at the beginning of the
452 		 * connection.
453 		 *
454 		 * AWL was adjusted in dccp_v4_connect -acme
455 		 */
456 		dccp_set_seqno(&dp->dccps_swl,
457 			       max48(dp->dccps_swl, dp->dccps_isr));
458 
459 		dccp_sync_mss(sk, icsk->icsk_pmtu_cookie);
460 
461 		/*
462 		 *    Step 10: Process REQUEST state (second part)
463 		 *       If S.state == REQUEST,
464 		 *	  / * If we get here, P is a valid Response from the
465 		 *	      server (see Step 4), and we should move to
466 		 *	      PARTOPEN state. PARTOPEN means send an Ack,
467 		 *	      don't send Data packets, retransmit Acks
468 		 *	      periodically, and always include any Init Cookie
469 		 *	      from the Response * /
470 		 *	  S.state := PARTOPEN
471 		 *	  Set PARTOPEN timer
472 		 *	  Continue with S.state == PARTOPEN
473 		 *	  / * Step 12 will send the Ack completing the
474 		 *	      three-way handshake * /
475 		 */
476 		dccp_set_state(sk, DCCP_PARTOPEN);
477 
478 		/*
479 		 * If feature negotiation was successful, activate features now;
480 		 * an activation failure means that this host could not activate
481 		 * one ore more features (e.g. insufficient memory), which would
482 		 * leave at least one feature in an undefined state.
483 		 */
484 		if (dccp_feat_activate_values(sk, &dp->dccps_featneg))
485 			goto unable_to_proceed;
486 
487 		/* Make sure socket is routed, for correct metrics. */
488 		icsk->icsk_af_ops->rebuild_header(sk);
489 
490 		if (!sock_flag(sk, SOCK_DEAD)) {
491 			sk->sk_state_change(sk);
492 			sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
493 		}
494 
495 		if (sk->sk_write_pending || icsk->icsk_ack.pingpong ||
496 		    icsk->icsk_accept_queue.rskq_defer_accept) {
497 			/* Save one ACK. Data will be ready after
498 			 * several ticks, if write_pending is set.
499 			 *
500 			 * It may be deleted, but with this feature tcpdumps
501 			 * look so _wonderfully_ clever, that I was not able
502 			 * to stand against the temptation 8)     --ANK
503 			 */
504 			/*
505 			 * OK, in DCCP we can as well do a similar trick, its
506 			 * even in the draft, but there is no need for us to
507 			 * schedule an ack here, as dccp_sendmsg does this for
508 			 * us, also stated in the draft. -acme
509 			 */
510 			__kfree_skb(skb);
511 			return 0;
512 		}
513 		dccp_send_ack(sk);
514 		return -1;
515 	}
516 
517 out_invalid_packet:
518 	/* dccp_v4_do_rcv will send a reset */
519 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
520 	return 1;
521 
522 unable_to_proceed:
523 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_ABORTED;
524 	/*
525 	 * We mark this socket as no longer usable, so that the loop in
526 	 * dccp_sendmsg() terminates and the application gets notified.
527 	 */
528 	dccp_set_state(sk, DCCP_CLOSED);
529 	sk->sk_err = ECOMM;
530 	return 1;
531 }
532 
533 static int dccp_rcv_respond_partopen_state_process(struct sock *sk,
534 						   struct sk_buff *skb,
535 						   const struct dccp_hdr *dh,
536 						   const unsigned len)
537 {
538 	struct dccp_sock *dp = dccp_sk(sk);
539 	u32 sample = dp->dccps_options_received.dccpor_timestamp_echo;
540 	int queued = 0;
541 
542 	switch (dh->dccph_type) {
543 	case DCCP_PKT_RESET:
544 		inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
545 		break;
546 	case DCCP_PKT_DATA:
547 		if (sk->sk_state == DCCP_RESPOND)
548 			break;
549 	case DCCP_PKT_DATAACK:
550 	case DCCP_PKT_ACK:
551 		/*
552 		 * FIXME: we should be reseting the PARTOPEN (DELACK) timer
553 		 * here but only if we haven't used the DELACK timer for
554 		 * something else, like sending a delayed ack for a TIMESTAMP
555 		 * echo, etc, for now were not clearing it, sending an extra
556 		 * ACK when there is nothing else to do in DELACK is not a big
557 		 * deal after all.
558 		 */
559 
560 		/* Stop the PARTOPEN timer */
561 		if (sk->sk_state == DCCP_PARTOPEN)
562 			inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
563 
564 		/* Obtain usec RTT sample from SYN exchange (used by TFRC). */
565 		if (likely(sample)) {
566 			long delta = dccp_timestamp() - sample;
567 
568 			dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * delta);
569 		}
570 
571 		dp->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq;
572 		dccp_set_state(sk, DCCP_OPEN);
573 
574 		if (dh->dccph_type == DCCP_PKT_DATAACK ||
575 		    dh->dccph_type == DCCP_PKT_DATA) {
576 			__dccp_rcv_established(sk, skb, dh, len);
577 			queued = 1; /* packet was queued
578 				       (by __dccp_rcv_established) */
579 		}
580 		break;
581 	}
582 
583 	return queued;
584 }
585 
586 int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
587 			   struct dccp_hdr *dh, unsigned len)
588 {
589 	struct dccp_sock *dp = dccp_sk(sk);
590 	struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
591 	const int old_state = sk->sk_state;
592 	int queued = 0;
593 
594 	/*
595 	 *  Step 3: Process LISTEN state
596 	 *
597 	 *     If S.state == LISTEN,
598 	 *	 If P.type == Request or P contains a valid Init Cookie option,
599 	 *	      (* Must scan the packet's options to check for Init
600 	 *		 Cookies.  Only Init Cookies are processed here,
601 	 *		 however; other options are processed in Step 8.  This
602 	 *		 scan need only be performed if the endpoint uses Init
603 	 *		 Cookies *)
604 	 *	      (* Generate a new socket and switch to that socket *)
605 	 *	      Set S := new socket for this port pair
606 	 *	      S.state = RESPOND
607 	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
608 	 *	      Initialize S.GAR := S.ISS
609 	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
610 	 *	      Cookies Continue with S.state == RESPOND
611 	 *	      (* A Response packet will be generated in Step 11 *)
612 	 *	 Otherwise,
613 	 *	      Generate Reset(No Connection) unless P.type == Reset
614 	 *	      Drop packet and return
615 	 */
616 	if (sk->sk_state == DCCP_LISTEN) {
617 		if (dh->dccph_type == DCCP_PKT_REQUEST) {
618 			if (inet_csk(sk)->icsk_af_ops->conn_request(sk,
619 								    skb) < 0)
620 				return 1;
621 			goto discard;
622 		}
623 		if (dh->dccph_type == DCCP_PKT_RESET)
624 			goto discard;
625 
626 		/* Caller (dccp_v4_do_rcv) will send Reset */
627 		dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
628 		return 1;
629 	}
630 
631 	if (sk->sk_state != DCCP_REQUESTING && sk->sk_state != DCCP_RESPOND) {
632 		if (dccp_check_seqno(sk, skb))
633 			goto discard;
634 
635 		/*
636 		 * Step 8: Process options and mark acknowledgeable
637 		 */
638 		if (dccp_parse_options(sk, NULL, skb))
639 			return 1;
640 
641 		if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
642 			dccp_event_ack_recv(sk, skb);
643 
644 		if (dp->dccps_hc_rx_ackvec != NULL &&
645 		    dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
646 				    DCCP_SKB_CB(skb)->dccpd_seq,
647 				    DCCP_ACKVEC_STATE_RECEIVED))
648 			goto discard;
649 
650 		dccp_deliver_input_to_ccids(sk, skb);
651 	}
652 
653 	/*
654 	 *  Step 9: Process Reset
655 	 *	If P.type == Reset,
656 	 *		Tear down connection
657 	 *		S.state := TIMEWAIT
658 	 *		Set TIMEWAIT timer
659 	 *		Drop packet and return
660 	*/
661 	if (dh->dccph_type == DCCP_PKT_RESET) {
662 		dccp_rcv_reset(sk, skb);
663 		return 0;
664 		/*
665 		 *   Step 7: Check for unexpected packet types
666 		 *      If (S.is_server and P.type == Response)
667 		 *	    or (S.is_client and P.type == Request)
668 		 *	    or (S.state == RESPOND and P.type == Data),
669 		 *	  Send Sync packet acknowledging P.seqno
670 		 *	  Drop packet and return
671 		 */
672 	} else if ((dp->dccps_role != DCCP_ROLE_CLIENT &&
673 		    dh->dccph_type == DCCP_PKT_RESPONSE) ||
674 		    (dp->dccps_role == DCCP_ROLE_CLIENT &&
675 		     dh->dccph_type == DCCP_PKT_REQUEST) ||
676 		    (sk->sk_state == DCCP_RESPOND &&
677 		     dh->dccph_type == DCCP_PKT_DATA)) {
678 		dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNC);
679 		goto discard;
680 	} else if (dh->dccph_type == DCCP_PKT_CLOSEREQ) {
681 		if (dccp_rcv_closereq(sk, skb))
682 			return 0;
683 		goto discard;
684 	} else if (dh->dccph_type == DCCP_PKT_CLOSE) {
685 		if (dccp_rcv_close(sk, skb))
686 			return 0;
687 		goto discard;
688 	}
689 
690 	switch (sk->sk_state) {
691 	case DCCP_CLOSED:
692 		dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
693 		return 1;
694 
695 	case DCCP_REQUESTING:
696 		queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len);
697 		if (queued >= 0)
698 			return queued;
699 
700 		__kfree_skb(skb);
701 		return 0;
702 
703 	case DCCP_RESPOND:
704 	case DCCP_PARTOPEN:
705 		queued = dccp_rcv_respond_partopen_state_process(sk, skb,
706 								 dh, len);
707 		break;
708 	}
709 
710 	if (dh->dccph_type == DCCP_PKT_ACK ||
711 	    dh->dccph_type == DCCP_PKT_DATAACK) {
712 		switch (old_state) {
713 		case DCCP_PARTOPEN:
714 			sk->sk_state_change(sk);
715 			sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
716 			break;
717 		}
718 	} else if (unlikely(dh->dccph_type == DCCP_PKT_SYNC)) {
719 		dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNCACK);
720 		goto discard;
721 	}
722 
723 	if (!queued) {
724 discard:
725 		__kfree_skb(skb);
726 	}
727 	return 0;
728 }
729 
730 EXPORT_SYMBOL_GPL(dccp_rcv_state_process);
731 
732 /**
733  *  dccp_sample_rtt  -  Validate and finalise computation of RTT sample
734  *  @delta:	number of microseconds between packet and acknowledgment
735  *  The routine is kept generic to work in different contexts. It should be
736  *  called immediately when the ACK used for the RTT sample arrives.
737  */
738 u32 dccp_sample_rtt(struct sock *sk, long delta)
739 {
740 	/* dccpor_elapsed_time is either zeroed out or set and > 0 */
741 	delta -= dccp_sk(sk)->dccps_options_received.dccpor_elapsed_time * 10;
742 
743 	if (unlikely(delta <= 0)) {
744 		DCCP_WARN("unusable RTT sample %ld, using min\n", delta);
745 		return DCCP_SANE_RTT_MIN;
746 	}
747 	if (unlikely(delta > DCCP_SANE_RTT_MAX)) {
748 		DCCP_WARN("RTT sample %ld too large, using max\n", delta);
749 		return DCCP_SANE_RTT_MAX;
750 	}
751 
752 	return delta;
753 }
754