xref: /linux/net/dccp/input.c (revision a3a4a816b4b194c45d0217e8b9e08b2639802cda)
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);
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_handle_ackvec_processing(struct sock *sk, struct sk_buff *skb)
164 {
165 	struct dccp_ackvec *av = dccp_sk(sk)->dccps_hc_rx_ackvec;
166 
167 	if (av == NULL)
168 		return;
169 	if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
170 		dccp_ackvec_clear_state(av, DCCP_SKB_CB(skb)->dccpd_ack_seq);
171 	dccp_ackvec_input(av, skb);
172 }
173 
174 static void dccp_deliver_input_to_ccids(struct sock *sk, struct sk_buff *skb)
175 {
176 	const struct dccp_sock *dp = dccp_sk(sk);
177 
178 	/* Don't deliver to RX CCID when node has shut down read end. */
179 	if (!(sk->sk_shutdown & RCV_SHUTDOWN))
180 		ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
181 	/*
182 	 * Until the TX queue has been drained, we can not honour SHUT_WR, since
183 	 * we need received feedback as input to adjust congestion control.
184 	 */
185 	if (sk->sk_write_queue.qlen > 0 || !(sk->sk_shutdown & SEND_SHUTDOWN))
186 		ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
187 }
188 
189 static int dccp_check_seqno(struct sock *sk, struct sk_buff *skb)
190 {
191 	const struct dccp_hdr *dh = dccp_hdr(skb);
192 	struct dccp_sock *dp = dccp_sk(sk);
193 	u64 lswl, lawl, seqno = DCCP_SKB_CB(skb)->dccpd_seq,
194 			ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
195 
196 	/*
197 	 *   Step 5: Prepare sequence numbers for Sync
198 	 *     If P.type == Sync or P.type == SyncAck,
199 	 *	  If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL,
200 	 *	     / * P is valid, so update sequence number variables
201 	 *		 accordingly.  After this update, P will pass the tests
202 	 *		 in Step 6.  A SyncAck is generated if necessary in
203 	 *		 Step 15 * /
204 	 *	     Update S.GSR, S.SWL, S.SWH
205 	 *	  Otherwise,
206 	 *	     Drop packet and return
207 	 */
208 	if (dh->dccph_type == DCCP_PKT_SYNC ||
209 	    dh->dccph_type == DCCP_PKT_SYNCACK) {
210 		if (between48(ackno, dp->dccps_awl, dp->dccps_awh) &&
211 		    dccp_delta_seqno(dp->dccps_swl, seqno) >= 0)
212 			dccp_update_gsr(sk, seqno);
213 		else
214 			return -1;
215 	}
216 
217 	/*
218 	 *   Step 6: Check sequence numbers
219 	 *      Let LSWL = S.SWL and LAWL = S.AWL
220 	 *      If P.type == CloseReq or P.type == Close or P.type == Reset,
221 	 *	  LSWL := S.GSR + 1, LAWL := S.GAR
222 	 *      If LSWL <= P.seqno <= S.SWH
223 	 *	     and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH),
224 	 *	  Update S.GSR, S.SWL, S.SWH
225 	 *	  If P.type != Sync,
226 	 *	     Update S.GAR
227 	 */
228 	lswl = dp->dccps_swl;
229 	lawl = dp->dccps_awl;
230 
231 	if (dh->dccph_type == DCCP_PKT_CLOSEREQ ||
232 	    dh->dccph_type == DCCP_PKT_CLOSE ||
233 	    dh->dccph_type == DCCP_PKT_RESET) {
234 		lswl = ADD48(dp->dccps_gsr, 1);
235 		lawl = dp->dccps_gar;
236 	}
237 
238 	if (between48(seqno, lswl, dp->dccps_swh) &&
239 	    (ackno == DCCP_PKT_WITHOUT_ACK_SEQ ||
240 	     between48(ackno, lawl, dp->dccps_awh))) {
241 		dccp_update_gsr(sk, seqno);
242 
243 		if (dh->dccph_type != DCCP_PKT_SYNC &&
244 		    ackno != DCCP_PKT_WITHOUT_ACK_SEQ &&
245 		    after48(ackno, dp->dccps_gar))
246 			dp->dccps_gar = ackno;
247 	} else {
248 		unsigned long now = jiffies;
249 		/*
250 		 *   Step 6: Check sequence numbers
251 		 *      Otherwise,
252 		 *         If P.type == Reset,
253 		 *            Send Sync packet acknowledging S.GSR
254 		 *         Otherwise,
255 		 *            Send Sync packet acknowledging P.seqno
256 		 *      Drop packet and return
257 		 *
258 		 *   These Syncs are rate-limited as per RFC 4340, 7.5.4:
259 		 *   at most 1 / (dccp_sync_rate_limit * HZ) Syncs per second.
260 		 */
261 		if (time_before(now, (dp->dccps_rate_last +
262 				      sysctl_dccp_sync_ratelimit)))
263 			return -1;
264 
265 		DCCP_WARN("Step 6 failed for %s packet, "
266 			  "(LSWL(%llu) <= P.seqno(%llu) <= S.SWH(%llu)) and "
267 			  "(P.ackno %s or LAWL(%llu) <= P.ackno(%llu) <= S.AWH(%llu), "
268 			  "sending SYNC...\n",  dccp_packet_name(dh->dccph_type),
269 			  (unsigned long long) lswl, (unsigned long long) seqno,
270 			  (unsigned long long) dp->dccps_swh,
271 			  (ackno == DCCP_PKT_WITHOUT_ACK_SEQ) ? "doesn't exist"
272 							      : "exists",
273 			  (unsigned long long) lawl, (unsigned long long) ackno,
274 			  (unsigned long long) dp->dccps_awh);
275 
276 		dp->dccps_rate_last = now;
277 
278 		if (dh->dccph_type == DCCP_PKT_RESET)
279 			seqno = dp->dccps_gsr;
280 		dccp_send_sync(sk, seqno, DCCP_PKT_SYNC);
281 		return -1;
282 	}
283 
284 	return 0;
285 }
286 
287 static int __dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
288 				  const struct dccp_hdr *dh, const unsigned int len)
289 {
290 	struct dccp_sock *dp = dccp_sk(sk);
291 
292 	switch (dccp_hdr(skb)->dccph_type) {
293 	case DCCP_PKT_DATAACK:
294 	case DCCP_PKT_DATA:
295 		/*
296 		 * FIXME: schedule DATA_DROPPED (RFC 4340, 11.7.2) if and when
297 		 * - sk_shutdown == RCV_SHUTDOWN, use Code 1, "Not Listening"
298 		 * - sk_receive_queue is full, use Code 2, "Receive Buffer"
299 		 */
300 		dccp_enqueue_skb(sk, skb);
301 		return 0;
302 	case DCCP_PKT_ACK:
303 		goto discard;
304 	case DCCP_PKT_RESET:
305 		/*
306 		 *  Step 9: Process Reset
307 		 *	If P.type == Reset,
308 		 *		Tear down connection
309 		 *		S.state := TIMEWAIT
310 		 *		Set TIMEWAIT timer
311 		 *		Drop packet and return
312 		 */
313 		dccp_rcv_reset(sk, skb);
314 		return 0;
315 	case DCCP_PKT_CLOSEREQ:
316 		if (dccp_rcv_closereq(sk, skb))
317 			return 0;
318 		goto discard;
319 	case DCCP_PKT_CLOSE:
320 		if (dccp_rcv_close(sk, skb))
321 			return 0;
322 		goto discard;
323 	case DCCP_PKT_REQUEST:
324 		/* Step 7
325 		 *   or (S.is_server and P.type == Response)
326 		 *   or (S.is_client and P.type == Request)
327 		 *   or (S.state >= OPEN and P.type == Request
328 		 *	and P.seqno >= S.OSR)
329 		 *    or (S.state >= OPEN and P.type == Response
330 		 *	and P.seqno >= S.OSR)
331 		 *    or (S.state == RESPOND and P.type == Data),
332 		 *  Send Sync packet acknowledging P.seqno
333 		 *  Drop packet and return
334 		 */
335 		if (dp->dccps_role != DCCP_ROLE_LISTEN)
336 			goto send_sync;
337 		goto check_seq;
338 	case DCCP_PKT_RESPONSE:
339 		if (dp->dccps_role != DCCP_ROLE_CLIENT)
340 			goto send_sync;
341 check_seq:
342 		if (dccp_delta_seqno(dp->dccps_osr,
343 				     DCCP_SKB_CB(skb)->dccpd_seq) >= 0) {
344 send_sync:
345 			dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
346 				       DCCP_PKT_SYNC);
347 		}
348 		break;
349 	case DCCP_PKT_SYNC:
350 		dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
351 			       DCCP_PKT_SYNCACK);
352 		/*
353 		 * From RFC 4340, sec. 5.7
354 		 *
355 		 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets
356 		 * MAY have non-zero-length application data areas, whose
357 		 * contents receivers MUST ignore.
358 		 */
359 		goto discard;
360 	}
361 
362 	DCCP_INC_STATS(DCCP_MIB_INERRS);
363 discard:
364 	__kfree_skb(skb);
365 	return 0;
366 }
367 
368 int dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
369 			 const struct dccp_hdr *dh, const unsigned int len)
370 {
371 	if (dccp_check_seqno(sk, skb))
372 		goto discard;
373 
374 	if (dccp_parse_options(sk, NULL, skb))
375 		return 1;
376 
377 	dccp_handle_ackvec_processing(sk, skb);
378 	dccp_deliver_input_to_ccids(sk, skb);
379 
380 	return __dccp_rcv_established(sk, skb, dh, len);
381 discard:
382 	__kfree_skb(skb);
383 	return 0;
384 }
385 
386 EXPORT_SYMBOL_GPL(dccp_rcv_established);
387 
388 static int dccp_rcv_request_sent_state_process(struct sock *sk,
389 					       struct sk_buff *skb,
390 					       const struct dccp_hdr *dh,
391 					       const unsigned int len)
392 {
393 	/*
394 	 *  Step 4: Prepare sequence numbers in REQUEST
395 	 *     If S.state == REQUEST,
396 	 *	  If (P.type == Response or P.type == Reset)
397 	 *		and S.AWL <= P.ackno <= S.AWH,
398 	 *	     / * Set sequence number variables corresponding to the
399 	 *		other endpoint, so P will pass the tests in Step 6 * /
400 	 *	     Set S.GSR, S.ISR, S.SWL, S.SWH
401 	 *	     / * Response processing continues in Step 10; Reset
402 	 *		processing continues in Step 9 * /
403 	*/
404 	if (dh->dccph_type == DCCP_PKT_RESPONSE) {
405 		const struct inet_connection_sock *icsk = inet_csk(sk);
406 		struct dccp_sock *dp = dccp_sk(sk);
407 		long tstamp = dccp_timestamp();
408 
409 		if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
410 			       dp->dccps_awl, dp->dccps_awh)) {
411 			dccp_pr_debug("invalid ackno: S.AWL=%llu, "
412 				      "P.ackno=%llu, S.AWH=%llu\n",
413 				      (unsigned long long)dp->dccps_awl,
414 			   (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
415 				      (unsigned long long)dp->dccps_awh);
416 			goto out_invalid_packet;
417 		}
418 
419 		/*
420 		 * If option processing (Step 8) failed, return 1 here so that
421 		 * dccp_v4_do_rcv() sends a Reset. The Reset code depends on
422 		 * the option type and is set in dccp_parse_options().
423 		 */
424 		if (dccp_parse_options(sk, NULL, skb))
425 			return 1;
426 
427 		/* Obtain usec RTT sample from SYN exchange (used by TFRC). */
428 		if (likely(dp->dccps_options_received.dccpor_timestamp_echo))
429 			dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * (tstamp -
430 			    dp->dccps_options_received.dccpor_timestamp_echo));
431 
432 		/* Stop the REQUEST timer */
433 		inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
434 		WARN_ON(sk->sk_send_head == NULL);
435 		kfree_skb(sk->sk_send_head);
436 		sk->sk_send_head = NULL;
437 
438 		/*
439 		 * Set ISR, GSR from packet. ISS was set in dccp_v{4,6}_connect
440 		 * and GSS in dccp_transmit_skb(). Setting AWL/AWH and SWL/SWH
441 		 * is done as part of activating the feature values below, since
442 		 * these settings depend on the local/remote Sequence Window
443 		 * features, which were undefined or not confirmed until now.
444 		 */
445 		dp->dccps_gsr = dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq;
446 
447 		dccp_sync_mss(sk, icsk->icsk_pmtu_cookie);
448 
449 		/*
450 		 *    Step 10: Process REQUEST state (second part)
451 		 *       If S.state == REQUEST,
452 		 *	  / * If we get here, P is a valid Response from the
453 		 *	      server (see Step 4), and we should move to
454 		 *	      PARTOPEN state. PARTOPEN means send an Ack,
455 		 *	      don't send Data packets, retransmit Acks
456 		 *	      periodically, and always include any Init Cookie
457 		 *	      from the Response * /
458 		 *	  S.state := PARTOPEN
459 		 *	  Set PARTOPEN timer
460 		 *	  Continue with S.state == PARTOPEN
461 		 *	  / * Step 12 will send the Ack completing the
462 		 *	      three-way handshake * /
463 		 */
464 		dccp_set_state(sk, DCCP_PARTOPEN);
465 
466 		/*
467 		 * If feature negotiation was successful, activate features now;
468 		 * an activation failure means that this host could not activate
469 		 * one ore more features (e.g. insufficient memory), which would
470 		 * leave at least one feature in an undefined state.
471 		 */
472 		if (dccp_feat_activate_values(sk, &dp->dccps_featneg))
473 			goto unable_to_proceed;
474 
475 		/* Make sure socket is routed, for correct metrics. */
476 		icsk->icsk_af_ops->rebuild_header(sk);
477 
478 		if (!sock_flag(sk, SOCK_DEAD)) {
479 			sk->sk_state_change(sk);
480 			sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
481 		}
482 
483 		if (sk->sk_write_pending || icsk->icsk_ack.pingpong ||
484 		    icsk->icsk_accept_queue.rskq_defer_accept) {
485 			/* Save one ACK. Data will be ready after
486 			 * several ticks, if write_pending is set.
487 			 *
488 			 * It may be deleted, but with this feature tcpdumps
489 			 * look so _wonderfully_ clever, that I was not able
490 			 * to stand against the temptation 8)     --ANK
491 			 */
492 			/*
493 			 * OK, in DCCP we can as well do a similar trick, its
494 			 * even in the draft, but there is no need for us to
495 			 * schedule an ack here, as dccp_sendmsg does this for
496 			 * us, also stated in the draft. -acme
497 			 */
498 			__kfree_skb(skb);
499 			return 0;
500 		}
501 		dccp_send_ack(sk);
502 		return -1;
503 	}
504 
505 out_invalid_packet:
506 	/* dccp_v4_do_rcv will send a reset */
507 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
508 	return 1;
509 
510 unable_to_proceed:
511 	DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_ABORTED;
512 	/*
513 	 * We mark this socket as no longer usable, so that the loop in
514 	 * dccp_sendmsg() terminates and the application gets notified.
515 	 */
516 	dccp_set_state(sk, DCCP_CLOSED);
517 	sk->sk_err = ECOMM;
518 	return 1;
519 }
520 
521 static int dccp_rcv_respond_partopen_state_process(struct sock *sk,
522 						   struct sk_buff *skb,
523 						   const struct dccp_hdr *dh,
524 						   const unsigned int len)
525 {
526 	struct dccp_sock *dp = dccp_sk(sk);
527 	u32 sample = dp->dccps_options_received.dccpor_timestamp_echo;
528 	int queued = 0;
529 
530 	switch (dh->dccph_type) {
531 	case DCCP_PKT_RESET:
532 		inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
533 		break;
534 	case DCCP_PKT_DATA:
535 		if (sk->sk_state == DCCP_RESPOND)
536 			break;
537 	case DCCP_PKT_DATAACK:
538 	case DCCP_PKT_ACK:
539 		/*
540 		 * FIXME: we should be resetting the PARTOPEN (DELACK) timer
541 		 * here but only if we haven't used the DELACK timer for
542 		 * something else, like sending a delayed ack for a TIMESTAMP
543 		 * echo, etc, for now were not clearing it, sending an extra
544 		 * ACK when there is nothing else to do in DELACK is not a big
545 		 * deal after all.
546 		 */
547 
548 		/* Stop the PARTOPEN timer */
549 		if (sk->sk_state == DCCP_PARTOPEN)
550 			inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
551 
552 		/* Obtain usec RTT sample from SYN exchange (used by TFRC). */
553 		if (likely(sample)) {
554 			long delta = dccp_timestamp() - sample;
555 
556 			dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * delta);
557 		}
558 
559 		dp->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq;
560 		dccp_set_state(sk, DCCP_OPEN);
561 
562 		if (dh->dccph_type == DCCP_PKT_DATAACK ||
563 		    dh->dccph_type == DCCP_PKT_DATA) {
564 			__dccp_rcv_established(sk, skb, dh, len);
565 			queued = 1; /* packet was queued
566 				       (by __dccp_rcv_established) */
567 		}
568 		break;
569 	}
570 
571 	return queued;
572 }
573 
574 int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
575 			   struct dccp_hdr *dh, unsigned int len)
576 {
577 	struct dccp_sock *dp = dccp_sk(sk);
578 	struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
579 	const int old_state = sk->sk_state;
580 	bool acceptable;
581 	int queued = 0;
582 
583 	/*
584 	 *  Step 3: Process LISTEN state
585 	 *
586 	 *     If S.state == LISTEN,
587 	 *	 If P.type == Request or P contains a valid Init Cookie option,
588 	 *	      (* Must scan the packet's options to check for Init
589 	 *		 Cookies.  Only Init Cookies are processed here,
590 	 *		 however; other options are processed in Step 8.  This
591 	 *		 scan need only be performed if the endpoint uses Init
592 	 *		 Cookies *)
593 	 *	      (* Generate a new socket and switch to that socket *)
594 	 *	      Set S := new socket for this port pair
595 	 *	      S.state = RESPOND
596 	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
597 	 *	      Initialize S.GAR := S.ISS
598 	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
599 	 *	      Cookies Continue with S.state == RESPOND
600 	 *	      (* A Response packet will be generated in Step 11 *)
601 	 *	 Otherwise,
602 	 *	      Generate Reset(No Connection) unless P.type == Reset
603 	 *	      Drop packet and return
604 	 */
605 	if (sk->sk_state == DCCP_LISTEN) {
606 		if (dh->dccph_type == DCCP_PKT_REQUEST) {
607 			/* It is possible that we process SYN packets from backlog,
608 			 * so we need to make sure to disable BH right there.
609 			 */
610 			local_bh_disable();
611 			acceptable = inet_csk(sk)->icsk_af_ops->conn_request(sk, skb) >= 0;
612 			local_bh_enable();
613 			if (!acceptable)
614 				return 1;
615 			consume_skb(skb);
616 			return 0;
617 		}
618 		if (dh->dccph_type == DCCP_PKT_RESET)
619 			goto discard;
620 
621 		/* Caller (dccp_v4_do_rcv) will send Reset */
622 		dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
623 		return 1;
624 	} else if (sk->sk_state == DCCP_CLOSED) {
625 		dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
626 		return 1;
627 	}
628 
629 	/* Step 6: Check sequence numbers (omitted in LISTEN/REQUEST state) */
630 	if (sk->sk_state != DCCP_REQUESTING && dccp_check_seqno(sk, skb))
631 		goto discard;
632 
633 	/*
634 	 *   Step 7: Check for unexpected packet types
635 	 *      If (S.is_server and P.type == Response)
636 	 *	    or (S.is_client and P.type == Request)
637 	 *	    or (S.state == RESPOND and P.type == Data),
638 	 *	  Send Sync packet acknowledging P.seqno
639 	 *	  Drop packet and return
640 	 */
641 	if ((dp->dccps_role != DCCP_ROLE_CLIENT &&
642 	     dh->dccph_type == DCCP_PKT_RESPONSE) ||
643 	    (dp->dccps_role == DCCP_ROLE_CLIENT &&
644 	     dh->dccph_type == DCCP_PKT_REQUEST) ||
645 	    (sk->sk_state == DCCP_RESPOND && dh->dccph_type == DCCP_PKT_DATA)) {
646 		dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNC);
647 		goto discard;
648 	}
649 
650 	/*  Step 8: Process options */
651 	if (dccp_parse_options(sk, NULL, skb))
652 		return 1;
653 
654 	/*
655 	 *  Step 9: Process Reset
656 	 *	If P.type == Reset,
657 	 *		Tear down connection
658 	 *		S.state := TIMEWAIT
659 	 *		Set TIMEWAIT timer
660 	 *		Drop packet and return
661 	 */
662 	if (dh->dccph_type == DCCP_PKT_RESET) {
663 		dccp_rcv_reset(sk, skb);
664 		return 0;
665 	} else if (dh->dccph_type == DCCP_PKT_CLOSEREQ) {	/* Step 13 */
666 		if (dccp_rcv_closereq(sk, skb))
667 			return 0;
668 		goto discard;
669 	} else if (dh->dccph_type == DCCP_PKT_CLOSE) {		/* Step 14 */
670 		if (dccp_rcv_close(sk, skb))
671 			return 0;
672 		goto discard;
673 	}
674 
675 	switch (sk->sk_state) {
676 	case DCCP_REQUESTING:
677 		queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len);
678 		if (queued >= 0)
679 			return queued;
680 
681 		__kfree_skb(skb);
682 		return 0;
683 
684 	case DCCP_PARTOPEN:
685 		/* Step 8: if using Ack Vectors, mark packet acknowledgeable */
686 		dccp_handle_ackvec_processing(sk, skb);
687 		dccp_deliver_input_to_ccids(sk, skb);
688 		/* fall through */
689 	case DCCP_RESPOND:
690 		queued = dccp_rcv_respond_partopen_state_process(sk, skb,
691 								 dh, len);
692 		break;
693 	}
694 
695 	if (dh->dccph_type == DCCP_PKT_ACK ||
696 	    dh->dccph_type == DCCP_PKT_DATAACK) {
697 		switch (old_state) {
698 		case DCCP_PARTOPEN:
699 			sk->sk_state_change(sk);
700 			sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
701 			break;
702 		}
703 	} else if (unlikely(dh->dccph_type == DCCP_PKT_SYNC)) {
704 		dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNCACK);
705 		goto discard;
706 	}
707 
708 	if (!queued) {
709 discard:
710 		__kfree_skb(skb);
711 	}
712 	return 0;
713 }
714 
715 EXPORT_SYMBOL_GPL(dccp_rcv_state_process);
716 
717 /**
718  *  dccp_sample_rtt  -  Validate and finalise computation of RTT sample
719  *  @delta:	number of microseconds between packet and acknowledgment
720  *
721  *  The routine is kept generic to work in different contexts. It should be
722  *  called immediately when the ACK used for the RTT sample arrives.
723  */
724 u32 dccp_sample_rtt(struct sock *sk, long delta)
725 {
726 	/* dccpor_elapsed_time is either zeroed out or set and > 0 */
727 	delta -= dccp_sk(sk)->dccps_options_received.dccpor_elapsed_time * 10;
728 
729 	if (unlikely(delta <= 0)) {
730 		DCCP_WARN("unusable RTT sample %ld, using min\n", delta);
731 		return DCCP_SANE_RTT_MIN;
732 	}
733 	if (unlikely(delta > DCCP_SANE_RTT_MAX)) {
734 		DCCP_WARN("RTT sample %ld too large, using max\n", delta);
735 		return DCCP_SANE_RTT_MAX;
736 	}
737 
738 	return delta;
739 }
740