Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/dccp/input.c
4 *
5 * An implementation of the DCCP protocol
6 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7 */
8
9#include <linux/dccp.h>
10#include <linux/skbuff.h>
11#include <linux/slab.h>
12
13#include <net/sock.h>
14
15#include "ackvec.h"
16#include "ccid.h"
17#include "dccp.h"
18
19/* rate-limit for syncs in reply to sequence-invalid packets; RFC 4340, 7.5.4 */
20int sysctl_dccp_sync_ratelimit __read_mostly = HZ / 8;
21
22static void dccp_enqueue_skb(struct sock *sk, struct sk_buff *skb)
23{
24 __skb_pull(skb, dccp_hdr(skb)->dccph_doff * 4);
25 __skb_queue_tail(&sk->sk_receive_queue, skb);
26 skb_set_owner_r(skb, sk);
27 sk->sk_data_ready(sk);
28}
29
30static void dccp_fin(struct sock *sk, struct sk_buff *skb)
31{
32 /*
33 * On receiving Close/CloseReq, both RD/WR shutdown are performed.
34 * RFC 4340, 8.3 says that we MAY send further Data/DataAcks after
35 * receiving the closing segment, but there is no guarantee that such
36 * data will be processed at all.
37 */
38 sk->sk_shutdown = SHUTDOWN_MASK;
39 sock_set_flag(sk, SOCK_DONE);
40 dccp_enqueue_skb(sk, skb);
41}
42
43static int dccp_rcv_close(struct sock *sk, struct sk_buff *skb)
44{
45 int queued = 0;
46
47 switch (sk->sk_state) {
48 /*
49 * We ignore Close when received in one of the following states:
50 * - CLOSED (may be a late or duplicate packet)
51 * - PASSIVE_CLOSEREQ (the peer has sent a CloseReq earlier)
52 * - RESPOND (already handled by dccp_check_req)
53 */
54 case DCCP_CLOSING:
55 /*
56 * Simultaneous-close: receiving a Close after sending one. This
57 * can happen if both client and server perform active-close and
58 * will result in an endless ping-pong of crossing and retrans-
59 * mitted Close packets, which only terminates when one of the
60 * nodes times out (min. 64 seconds). Quicker convergence can be
61 * achieved when one of the nodes acts as tie-breaker.
62 * This is ok as both ends are done with data transfer and each
63 * end is just waiting for the other to acknowledge termination.
64 */
65 if (dccp_sk(sk)->dccps_role != DCCP_ROLE_CLIENT)
66 break;
67 fallthrough;
68 case DCCP_REQUESTING:
69 case DCCP_ACTIVE_CLOSEREQ:
70 dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
71 dccp_done(sk);
72 break;
73 case DCCP_OPEN:
74 case DCCP_PARTOPEN:
75 /* Give waiting application a chance to read pending data */
76 queued = 1;
77 dccp_fin(sk, skb);
78 dccp_set_state(sk, DCCP_PASSIVE_CLOSE);
79 fallthrough;
80 case DCCP_PASSIVE_CLOSE:
81 /*
82 * Retransmitted Close: we have already enqueued the first one.
83 */
84 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
85 }
86 return queued;
87}
88
89static int dccp_rcv_closereq(struct sock *sk, struct sk_buff *skb)
90{
91 int queued = 0;
92
93 /*
94 * Step 7: Check for unexpected packet types
95 * If (S.is_server and P.type == CloseReq)
96 * Send Sync packet acknowledging P.seqno
97 * Drop packet and return
98 */
99 if (dccp_sk(sk)->dccps_role != DCCP_ROLE_CLIENT) {
100 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, DCCP_PKT_SYNC);
101 return queued;
102 }
103
104 /* Step 13: process relevant Client states < CLOSEREQ */
105 switch (sk->sk_state) {
106 case DCCP_REQUESTING:
107 dccp_send_close(sk, 0);
108 dccp_set_state(sk, DCCP_CLOSING);
109 break;
110 case DCCP_OPEN:
111 case DCCP_PARTOPEN:
112 /* Give waiting application a chance to read pending data */
113 queued = 1;
114 dccp_fin(sk, skb);
115 dccp_set_state(sk, DCCP_PASSIVE_CLOSEREQ);
116 fallthrough;
117 case DCCP_PASSIVE_CLOSEREQ:
118 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
119 }
120 return queued;
121}
122
123static u16 dccp_reset_code_convert(const u8 code)
124{
125 static const u16 error_code[] = {
126 [DCCP_RESET_CODE_CLOSED] = 0, /* normal termination */
127 [DCCP_RESET_CODE_UNSPECIFIED] = 0, /* nothing known */
128 [DCCP_RESET_CODE_ABORTED] = ECONNRESET,
129
130 [DCCP_RESET_CODE_NO_CONNECTION] = ECONNREFUSED,
131 [DCCP_RESET_CODE_CONNECTION_REFUSED] = ECONNREFUSED,
132 [DCCP_RESET_CODE_TOO_BUSY] = EUSERS,
133 [DCCP_RESET_CODE_AGGRESSION_PENALTY] = EDQUOT,
134
135 [DCCP_RESET_CODE_PACKET_ERROR] = ENOMSG,
136 [DCCP_RESET_CODE_BAD_INIT_COOKIE] = EBADR,
137 [DCCP_RESET_CODE_BAD_SERVICE_CODE] = EBADRQC,
138 [DCCP_RESET_CODE_OPTION_ERROR] = EILSEQ,
139 [DCCP_RESET_CODE_MANDATORY_ERROR] = EOPNOTSUPP,
140 };
141
142 return code >= DCCP_MAX_RESET_CODES ? 0 : error_code[code];
143}
144
145static void dccp_rcv_reset(struct sock *sk, struct sk_buff *skb)
146{
147 u16 err = dccp_reset_code_convert(dccp_hdr_reset(skb)->dccph_reset_code);
148
149 sk->sk_err = err;
150
151 /* Queue the equivalent of TCP fin so that dccp_recvmsg exits the loop */
152 dccp_fin(sk, skb);
153
154 if (err && !sock_flag(sk, SOCK_DEAD))
155 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
156 dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
157}
158
159static void dccp_handle_ackvec_processing(struct sock *sk, struct sk_buff *skb)
160{
161 struct dccp_ackvec *av = dccp_sk(sk)->dccps_hc_rx_ackvec;
162
163 if (av == NULL)
164 return;
165 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
166 dccp_ackvec_clear_state(av, DCCP_SKB_CB(skb)->dccpd_ack_seq);
167 dccp_ackvec_input(av, skb);
168}
169
170static void dccp_deliver_input_to_ccids(struct sock *sk, struct sk_buff *skb)
171{
172 const struct dccp_sock *dp = dccp_sk(sk);
173
174 /* Don't deliver to RX CCID when node has shut down read end. */
175 if (!(sk->sk_shutdown & RCV_SHUTDOWN))
176 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
177 /*
178 * Until the TX queue has been drained, we can not honour SHUT_WR, since
179 * we need received feedback as input to adjust congestion control.
180 */
181 if (sk->sk_write_queue.qlen > 0 || !(sk->sk_shutdown & SEND_SHUTDOWN))
182 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
183}
184
185static int dccp_check_seqno(struct sock *sk, struct sk_buff *skb)
186{
187 const struct dccp_hdr *dh = dccp_hdr(skb);
188 struct dccp_sock *dp = dccp_sk(sk);
189 u64 lswl, lawl, seqno = DCCP_SKB_CB(skb)->dccpd_seq,
190 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
191
192 /*
193 * Step 5: Prepare sequence numbers for Sync
194 * If P.type == Sync or P.type == SyncAck,
195 * If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL,
196 * / * P is valid, so update sequence number variables
197 * accordingly. After this update, P will pass the tests
198 * in Step 6. A SyncAck is generated if necessary in
199 * Step 15 * /
200 * Update S.GSR, S.SWL, S.SWH
201 * Otherwise,
202 * Drop packet and return
203 */
204 if (dh->dccph_type == DCCP_PKT_SYNC ||
205 dh->dccph_type == DCCP_PKT_SYNCACK) {
206 if (between48(ackno, dp->dccps_awl, dp->dccps_awh) &&
207 dccp_delta_seqno(dp->dccps_swl, seqno) >= 0)
208 dccp_update_gsr(sk, seqno);
209 else
210 return -1;
211 }
212
213 /*
214 * Step 6: Check sequence numbers
215 * Let LSWL = S.SWL and LAWL = S.AWL
216 * If P.type == CloseReq or P.type == Close or P.type == Reset,
217 * LSWL := S.GSR + 1, LAWL := S.GAR
218 * If LSWL <= P.seqno <= S.SWH
219 * and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH),
220 * Update S.GSR, S.SWL, S.SWH
221 * If P.type != Sync,
222 * Update S.GAR
223 */
224 lswl = dp->dccps_swl;
225 lawl = dp->dccps_awl;
226
227 if (dh->dccph_type == DCCP_PKT_CLOSEREQ ||
228 dh->dccph_type == DCCP_PKT_CLOSE ||
229 dh->dccph_type == DCCP_PKT_RESET) {
230 lswl = ADD48(dp->dccps_gsr, 1);
231 lawl = dp->dccps_gar;
232 }
233
234 if (between48(seqno, lswl, dp->dccps_swh) &&
235 (ackno == DCCP_PKT_WITHOUT_ACK_SEQ ||
236 between48(ackno, lawl, dp->dccps_awh))) {
237 dccp_update_gsr(sk, seqno);
238
239 if (dh->dccph_type != DCCP_PKT_SYNC &&
240 ackno != DCCP_PKT_WITHOUT_ACK_SEQ &&
241 after48(ackno, dp->dccps_gar))
242 dp->dccps_gar = ackno;
243 } else {
244 unsigned long now = jiffies;
245 /*
246 * Step 6: Check sequence numbers
247 * Otherwise,
248 * If P.type == Reset,
249 * Send Sync packet acknowledging S.GSR
250 * Otherwise,
251 * Send Sync packet acknowledging P.seqno
252 * Drop packet and return
253 *
254 * These Syncs are rate-limited as per RFC 4340, 7.5.4:
255 * at most 1 / (dccp_sync_rate_limit * HZ) Syncs per second.
256 */
257 if (time_before(now, (dp->dccps_rate_last +
258 sysctl_dccp_sync_ratelimit)))
259 return -1;
260
261 DCCP_WARN("Step 6 failed for %s packet, "
262 "(LSWL(%llu) <= P.seqno(%llu) <= S.SWH(%llu)) and "
263 "(P.ackno %s or LAWL(%llu) <= P.ackno(%llu) <= S.AWH(%llu), "
264 "sending SYNC...\n", dccp_packet_name(dh->dccph_type),
265 (unsigned long long) lswl, (unsigned long long) seqno,
266 (unsigned long long) dp->dccps_swh,
267 (ackno == DCCP_PKT_WITHOUT_ACK_SEQ) ? "doesn't exist"
268 : "exists",
269 (unsigned long long) lawl, (unsigned long long) ackno,
270 (unsigned long long) dp->dccps_awh);
271
272 dp->dccps_rate_last = now;
273
274 if (dh->dccph_type == DCCP_PKT_RESET)
275 seqno = dp->dccps_gsr;
276 dccp_send_sync(sk, seqno, DCCP_PKT_SYNC);
277 return -1;
278 }
279
280 return 0;
281}
282
283static int __dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
284 const struct dccp_hdr *dh, const unsigned int len)
285{
286 struct dccp_sock *dp = dccp_sk(sk);
287
288 switch (dccp_hdr(skb)->dccph_type) {
289 case DCCP_PKT_DATAACK:
290 case DCCP_PKT_DATA:
291 /*
292 * FIXME: schedule DATA_DROPPED (RFC 4340, 11.7.2) if and when
293 * - sk_shutdown == RCV_SHUTDOWN, use Code 1, "Not Listening"
294 * - sk_receive_queue is full, use Code 2, "Receive Buffer"
295 */
296 dccp_enqueue_skb(sk, skb);
297 return 0;
298 case DCCP_PKT_ACK:
299 goto discard;
300 case DCCP_PKT_RESET:
301 /*
302 * Step 9: Process Reset
303 * If P.type == Reset,
304 * Tear down connection
305 * S.state := TIMEWAIT
306 * Set TIMEWAIT timer
307 * Drop packet and return
308 */
309 dccp_rcv_reset(sk, skb);
310 return 0;
311 case DCCP_PKT_CLOSEREQ:
312 if (dccp_rcv_closereq(sk, skb))
313 return 0;
314 goto discard;
315 case DCCP_PKT_CLOSE:
316 if (dccp_rcv_close(sk, skb))
317 return 0;
318 goto discard;
319 case DCCP_PKT_REQUEST:
320 /* Step 7
321 * or (S.is_server and P.type == Response)
322 * or (S.is_client and P.type == Request)
323 * or (S.state >= OPEN and P.type == Request
324 * and P.seqno >= S.OSR)
325 * or (S.state >= OPEN and P.type == Response
326 * and P.seqno >= S.OSR)
327 * or (S.state == RESPOND and P.type == Data),
328 * Send Sync packet acknowledging P.seqno
329 * Drop packet and return
330 */
331 if (dp->dccps_role != DCCP_ROLE_LISTEN)
332 goto send_sync;
333 goto check_seq;
334 case DCCP_PKT_RESPONSE:
335 if (dp->dccps_role != DCCP_ROLE_CLIENT)
336 goto send_sync;
337check_seq:
338 if (dccp_delta_seqno(dp->dccps_osr,
339 DCCP_SKB_CB(skb)->dccpd_seq) >= 0) {
340send_sync:
341 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
342 DCCP_PKT_SYNC);
343 }
344 break;
345 case DCCP_PKT_SYNC:
346 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
347 DCCP_PKT_SYNCACK);
348 /*
349 * From RFC 4340, sec. 5.7
350 *
351 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets
352 * MAY have non-zero-length application data areas, whose
353 * contents receivers MUST ignore.
354 */
355 goto discard;
356 }
357
358 DCCP_INC_STATS(DCCP_MIB_INERRS);
359discard:
360 __kfree_skb(skb);
361 return 0;
362}
363
364int dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
365 const struct dccp_hdr *dh, const unsigned int len)
366{
367 if (dccp_check_seqno(sk, skb))
368 goto discard;
369
370 if (dccp_parse_options(sk, NULL, skb))
371 return 1;
372
373 dccp_handle_ackvec_processing(sk, skb);
374 dccp_deliver_input_to_ccids(sk, skb);
375
376 return __dccp_rcv_established(sk, skb, dh, len);
377discard:
378 __kfree_skb(skb);
379 return 0;
380}
381
382EXPORT_SYMBOL_GPL(dccp_rcv_established);
383
384static int dccp_rcv_request_sent_state_process(struct sock *sk,
385 struct sk_buff *skb,
386 const struct dccp_hdr *dh,
387 const unsigned int len)
388{
389 /*
390 * Step 4: Prepare sequence numbers in REQUEST
391 * If S.state == REQUEST,
392 * If (P.type == Response or P.type == Reset)
393 * and S.AWL <= P.ackno <= S.AWH,
394 * / * Set sequence number variables corresponding to the
395 * other endpoint, so P will pass the tests in Step 6 * /
396 * Set S.GSR, S.ISR, S.SWL, S.SWH
397 * / * Response processing continues in Step 10; Reset
398 * processing continues in Step 9 * /
399 */
400 if (dh->dccph_type == DCCP_PKT_RESPONSE) {
401 const struct inet_connection_sock *icsk = inet_csk(sk);
402 struct dccp_sock *dp = dccp_sk(sk);
403 long tstamp = dccp_timestamp();
404
405 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
406 dp->dccps_awl, dp->dccps_awh)) {
407 dccp_pr_debug("invalid ackno: S.AWL=%llu, "
408 "P.ackno=%llu, S.AWH=%llu\n",
409 (unsigned long long)dp->dccps_awl,
410 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
411 (unsigned long long)dp->dccps_awh);
412 goto out_invalid_packet;
413 }
414
415 /*
416 * If option processing (Step 8) failed, return 1 here so that
417 * dccp_v4_do_rcv() sends a Reset. The Reset code depends on
418 * the option type and is set in dccp_parse_options().
419 */
420 if (dccp_parse_options(sk, NULL, skb))
421 return 1;
422
423 /* Obtain usec RTT sample from SYN exchange (used by TFRC). */
424 if (likely(dp->dccps_options_received.dccpor_timestamp_echo))
425 dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * (tstamp -
426 dp->dccps_options_received.dccpor_timestamp_echo));
427
428 /* Stop the REQUEST timer */
429 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
430 WARN_ON(sk->sk_send_head == NULL);
431 kfree_skb(sk->sk_send_head);
432 sk->sk_send_head = NULL;
433
434 /*
435 * Set ISR, GSR from packet. ISS was set in dccp_v{4,6}_connect
436 * and GSS in dccp_transmit_skb(). Setting AWL/AWH and SWL/SWH
437 * is done as part of activating the feature values below, since
438 * these settings depend on the local/remote Sequence Window
439 * features, which were undefined or not confirmed until now.
440 */
441 dp->dccps_gsr = dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq;
442
443 dccp_sync_mss(sk, icsk->icsk_pmtu_cookie);
444
445 /*
446 * Step 10: Process REQUEST state (second part)
447 * If S.state == REQUEST,
448 * / * If we get here, P is a valid Response from the
449 * server (see Step 4), and we should move to
450 * PARTOPEN state. PARTOPEN means send an Ack,
451 * don't send Data packets, retransmit Acks
452 * periodically, and always include any Init Cookie
453 * from the Response * /
454 * S.state := PARTOPEN
455 * Set PARTOPEN timer
456 * Continue with S.state == PARTOPEN
457 * / * Step 12 will send the Ack completing the
458 * three-way handshake * /
459 */
460 dccp_set_state(sk, DCCP_PARTOPEN);
461
462 /*
463 * If feature negotiation was successful, activate features now;
464 * an activation failure means that this host could not activate
465 * one ore more features (e.g. insufficient memory), which would
466 * leave at least one feature in an undefined state.
467 */
468 if (dccp_feat_activate_values(sk, &dp->dccps_featneg))
469 goto unable_to_proceed;
470
471 /* Make sure socket is routed, for correct metrics. */
472 icsk->icsk_af_ops->rebuild_header(sk);
473
474 if (!sock_flag(sk, SOCK_DEAD)) {
475 sk->sk_state_change(sk);
476 sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
477 }
478
479 if (sk->sk_write_pending || inet_csk_in_pingpong_mode(sk) ||
480 icsk->icsk_accept_queue.rskq_defer_accept) {
481 /* Save one ACK. Data will be ready after
482 * several ticks, if write_pending is set.
483 *
484 * It may be deleted, but with this feature tcpdumps
485 * look so _wonderfully_ clever, that I was not able
486 * to stand against the temptation 8) --ANK
487 */
488 /*
489 * OK, in DCCP we can as well do a similar trick, its
490 * even in the draft, but there is no need for us to
491 * schedule an ack here, as dccp_sendmsg does this for
492 * us, also stated in the draft. -acme
493 */
494 __kfree_skb(skb);
495 return 0;
496 }
497 dccp_send_ack(sk);
498 return -1;
499 }
500
501out_invalid_packet:
502 /* dccp_v4_do_rcv will send a reset */
503 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
504 return 1;
505
506unable_to_proceed:
507 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_ABORTED;
508 /*
509 * We mark this socket as no longer usable, so that the loop in
510 * dccp_sendmsg() terminates and the application gets notified.
511 */
512 dccp_set_state(sk, DCCP_CLOSED);
513 sk->sk_err = ECOMM;
514 return 1;
515}
516
517static int dccp_rcv_respond_partopen_state_process(struct sock *sk,
518 struct sk_buff *skb,
519 const struct dccp_hdr *dh,
520 const unsigned int len)
521{
522 struct dccp_sock *dp = dccp_sk(sk);
523 u32 sample = dp->dccps_options_received.dccpor_timestamp_echo;
524 int queued = 0;
525
526 switch (dh->dccph_type) {
527 case DCCP_PKT_RESET:
528 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
529 break;
530 case DCCP_PKT_DATA:
531 if (sk->sk_state == DCCP_RESPOND)
532 break;
533 fallthrough;
534 case DCCP_PKT_DATAACK:
535 case DCCP_PKT_ACK:
536 /*
537 * FIXME: we should be resetting the PARTOPEN (DELACK) timer
538 * here but only if we haven't used the DELACK timer for
539 * something else, like sending a delayed ack for a TIMESTAMP
540 * echo, etc, for now were not clearing it, sending an extra
541 * ACK when there is nothing else to do in DELACK is not a big
542 * deal after all.
543 */
544
545 /* Stop the PARTOPEN timer */
546 if (sk->sk_state == DCCP_PARTOPEN)
547 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
548
549 /* Obtain usec RTT sample from SYN exchange (used by TFRC). */
550 if (likely(sample)) {
551 long delta = dccp_timestamp() - sample;
552
553 dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * delta);
554 }
555
556 dp->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq;
557 dccp_set_state(sk, DCCP_OPEN);
558
559 if (dh->dccph_type == DCCP_PKT_DATAACK ||
560 dh->dccph_type == DCCP_PKT_DATA) {
561 __dccp_rcv_established(sk, skb, dh, len);
562 queued = 1; /* packet was queued
563 (by __dccp_rcv_established) */
564 }
565 break;
566 }
567
568 return queued;
569}
570
571int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
572 struct dccp_hdr *dh, unsigned int len)
573{
574 struct dccp_sock *dp = dccp_sk(sk);
575 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
576 const int old_state = sk->sk_state;
577 bool acceptable;
578 int queued = 0;
579
580 /*
581 * Step 3: Process LISTEN state
582 *
583 * If S.state == LISTEN,
584 * If P.type == Request or P contains a valid Init Cookie option,
585 * (* Must scan the packet's options to check for Init
586 * Cookies. Only Init Cookies are processed here,
587 * however; other options are processed in Step 8. This
588 * scan need only be performed if the endpoint uses Init
589 * Cookies *)
590 * (* Generate a new socket and switch to that socket *)
591 * Set S := new socket for this port pair
592 * S.state = RESPOND
593 * Choose S.ISS (initial seqno) or set from Init Cookies
594 * Initialize S.GAR := S.ISS
595 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
596 * Cookies Continue with S.state == RESPOND
597 * (* A Response packet will be generated in Step 11 *)
598 * Otherwise,
599 * Generate Reset(No Connection) unless P.type == Reset
600 * Drop packet and return
601 */
602 if (sk->sk_state == DCCP_LISTEN) {
603 if (dh->dccph_type == DCCP_PKT_REQUEST) {
604 /* It is possible that we process SYN packets from backlog,
605 * so we need to make sure to disable BH and RCU right there.
606 */
607 rcu_read_lock();
608 local_bh_disable();
609 acceptable = inet_csk(sk)->icsk_af_ops->conn_request(sk, skb) >= 0;
610 local_bh_enable();
611 rcu_read_unlock();
612 if (!acceptable)
613 return 1;
614 consume_skb(skb);
615 return 0;
616 }
617 if (dh->dccph_type == DCCP_PKT_RESET)
618 goto discard;
619
620 /* Caller (dccp_v4_do_rcv) will send Reset */
621 dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
622 return 1;
623 } else if (sk->sk_state == DCCP_CLOSED) {
624 dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
625 return 1;
626 }
627
628 /* Step 6: Check sequence numbers (omitted in LISTEN/REQUEST state) */
629 if (sk->sk_state != DCCP_REQUESTING && dccp_check_seqno(sk, skb))
630 goto discard;
631
632 /*
633 * Step 7: Check for unexpected packet types
634 * If (S.is_server and P.type == Response)
635 * or (S.is_client and P.type == Request)
636 * or (S.state == RESPOND and P.type == Data),
637 * Send Sync packet acknowledging P.seqno
638 * Drop packet and return
639 */
640 if ((dp->dccps_role != DCCP_ROLE_CLIENT &&
641 dh->dccph_type == DCCP_PKT_RESPONSE) ||
642 (dp->dccps_role == DCCP_ROLE_CLIENT &&
643 dh->dccph_type == DCCP_PKT_REQUEST) ||
644 (sk->sk_state == DCCP_RESPOND && dh->dccph_type == DCCP_PKT_DATA)) {
645 dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNC);
646 goto discard;
647 }
648
649 /* Step 8: Process options */
650 if (dccp_parse_options(sk, NULL, skb))
651 return 1;
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 } else if (dh->dccph_type == DCCP_PKT_CLOSEREQ) { /* Step 13 */
665 if (dccp_rcv_closereq(sk, skb))
666 return 0;
667 goto discard;
668 } else if (dh->dccph_type == DCCP_PKT_CLOSE) { /* Step 14 */
669 if (dccp_rcv_close(sk, skb))
670 return 0;
671 goto discard;
672 }
673
674 switch (sk->sk_state) {
675 case DCCP_REQUESTING:
676 queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len);
677 if (queued >= 0)
678 return queued;
679
680 __kfree_skb(skb);
681 return 0;
682
683 case DCCP_PARTOPEN:
684 /* Step 8: if using Ack Vectors, mark packet acknowledgeable */
685 dccp_handle_ackvec_processing(sk, skb);
686 dccp_deliver_input_to_ccids(sk, skb);
687 fallthrough;
688 case DCCP_RESPOND:
689 queued = dccp_rcv_respond_partopen_state_process(sk, skb,
690 dh, len);
691 break;
692 }
693
694 if (dh->dccph_type == DCCP_PKT_ACK ||
695 dh->dccph_type == DCCP_PKT_DATAACK) {
696 switch (old_state) {
697 case DCCP_PARTOPEN:
698 sk->sk_state_change(sk);
699 sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
700 break;
701 }
702 } else if (unlikely(dh->dccph_type == DCCP_PKT_SYNC)) {
703 dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNCACK);
704 goto discard;
705 }
706
707 if (!queued) {
708discard:
709 __kfree_skb(skb);
710 }
711 return 0;
712}
713
714EXPORT_SYMBOL_GPL(dccp_rcv_state_process);
715
716/**
717 * dccp_sample_rtt - Validate and finalise computation of RTT sample
718 * @sk: socket structure
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 */
724u32 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}
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 */
24int sysctl_dccp_sync_ratelimit __read_mostly = HZ / 8;
25
26static 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
34static 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
47static 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
93static 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
127static 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
149static 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
163static 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
174static 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
189static 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
287static 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;
341check_seq:
342 if (dccp_delta_seqno(dp->dccps_osr,
343 DCCP_SKB_CB(skb)->dccpd_seq) >= 0) {
344send_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_BH(DCCP_MIB_INERRS);
363discard:
364 __kfree_skb(skb);
365 return 0;
366}
367
368int 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);
381discard:
382 __kfree_skb(skb);
383 return 0;
384}
385
386EXPORT_SYMBOL_GPL(dccp_rcv_established);
387
388static 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
505out_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
510unable_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
521static 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
574int 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 int queued = 0;
581
582 /*
583 * Step 3: Process LISTEN state
584 *
585 * If S.state == LISTEN,
586 * If P.type == Request or P contains a valid Init Cookie option,
587 * (* Must scan the packet's options to check for Init
588 * Cookies. Only Init Cookies are processed here,
589 * however; other options are processed in Step 8. This
590 * scan need only be performed if the endpoint uses Init
591 * Cookies *)
592 * (* Generate a new socket and switch to that socket *)
593 * Set S := new socket for this port pair
594 * S.state = RESPOND
595 * Choose S.ISS (initial seqno) or set from Init Cookies
596 * Initialize S.GAR := S.ISS
597 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
598 * Cookies Continue with S.state == RESPOND
599 * (* A Response packet will be generated in Step 11 *)
600 * Otherwise,
601 * Generate Reset(No Connection) unless P.type == Reset
602 * Drop packet and return
603 */
604 if (sk->sk_state == DCCP_LISTEN) {
605 if (dh->dccph_type == DCCP_PKT_REQUEST) {
606 if (inet_csk(sk)->icsk_af_ops->conn_request(sk,
607 skb) < 0)
608 return 1;
609 goto discard;
610 }
611 if (dh->dccph_type == DCCP_PKT_RESET)
612 goto discard;
613
614 /* Caller (dccp_v4_do_rcv) will send Reset */
615 dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
616 return 1;
617 } else if (sk->sk_state == DCCP_CLOSED) {
618 dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
619 return 1;
620 }
621
622 /* Step 6: Check sequence numbers (omitted in LISTEN/REQUEST state) */
623 if (sk->sk_state != DCCP_REQUESTING && dccp_check_seqno(sk, skb))
624 goto discard;
625
626 /*
627 * Step 7: Check for unexpected packet types
628 * If (S.is_server and P.type == Response)
629 * or (S.is_client and P.type == Request)
630 * or (S.state == RESPOND and P.type == Data),
631 * Send Sync packet acknowledging P.seqno
632 * Drop packet and return
633 */
634 if ((dp->dccps_role != DCCP_ROLE_CLIENT &&
635 dh->dccph_type == DCCP_PKT_RESPONSE) ||
636 (dp->dccps_role == DCCP_ROLE_CLIENT &&
637 dh->dccph_type == DCCP_PKT_REQUEST) ||
638 (sk->sk_state == DCCP_RESPOND && dh->dccph_type == DCCP_PKT_DATA)) {
639 dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNC);
640 goto discard;
641 }
642
643 /* Step 8: Process options */
644 if (dccp_parse_options(sk, NULL, skb))
645 return 1;
646
647 /*
648 * Step 9: Process Reset
649 * If P.type == Reset,
650 * Tear down connection
651 * S.state := TIMEWAIT
652 * Set TIMEWAIT timer
653 * Drop packet and return
654 */
655 if (dh->dccph_type == DCCP_PKT_RESET) {
656 dccp_rcv_reset(sk, skb);
657 return 0;
658 } else if (dh->dccph_type == DCCP_PKT_CLOSEREQ) { /* Step 13 */
659 if (dccp_rcv_closereq(sk, skb))
660 return 0;
661 goto discard;
662 } else if (dh->dccph_type == DCCP_PKT_CLOSE) { /* Step 14 */
663 if (dccp_rcv_close(sk, skb))
664 return 0;
665 goto discard;
666 }
667
668 switch (sk->sk_state) {
669 case DCCP_REQUESTING:
670 queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len);
671 if (queued >= 0)
672 return queued;
673
674 __kfree_skb(skb);
675 return 0;
676
677 case DCCP_PARTOPEN:
678 /* Step 8: if using Ack Vectors, mark packet acknowledgeable */
679 dccp_handle_ackvec_processing(sk, skb);
680 dccp_deliver_input_to_ccids(sk, skb);
681 /* fall through */
682 case DCCP_RESPOND:
683 queued = dccp_rcv_respond_partopen_state_process(sk, skb,
684 dh, len);
685 break;
686 }
687
688 if (dh->dccph_type == DCCP_PKT_ACK ||
689 dh->dccph_type == DCCP_PKT_DATAACK) {
690 switch (old_state) {
691 case DCCP_PARTOPEN:
692 sk->sk_state_change(sk);
693 sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
694 break;
695 }
696 } else if (unlikely(dh->dccph_type == DCCP_PKT_SYNC)) {
697 dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNCACK);
698 goto discard;
699 }
700
701 if (!queued) {
702discard:
703 __kfree_skb(skb);
704 }
705 return 0;
706}
707
708EXPORT_SYMBOL_GPL(dccp_rcv_state_process);
709
710/**
711 * dccp_sample_rtt - Validate and finalise computation of RTT sample
712 * @delta: number of microseconds between packet and acknowledgment
713 *
714 * The routine is kept generic to work in different contexts. It should be
715 * called immediately when the ACK used for the RTT sample arrives.
716 */
717u32 dccp_sample_rtt(struct sock *sk, long delta)
718{
719 /* dccpor_elapsed_time is either zeroed out or set and > 0 */
720 delta -= dccp_sk(sk)->dccps_options_received.dccpor_elapsed_time * 10;
721
722 if (unlikely(delta <= 0)) {
723 DCCP_WARN("unusable RTT sample %ld, using min\n", delta);
724 return DCCP_SANE_RTT_MIN;
725 }
726 if (unlikely(delta > DCCP_SANE_RTT_MAX)) {
727 DCCP_WARN("RTT sample %ld too large, using max\n", delta);
728 return DCCP_SANE_RTT_MAX;
729 }
730
731 return delta;
732}