inotify: fix race
[linux-2.6.22.y-op.git] / net / dccp / input.c
blobda6ec185ed5b5826e6df5b6d58f7870e26330cb9
1 /*
2 * net/dccp/input.c
4 * An implementation of the DCCP protocol
5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
13 #include <linux/dccp.h>
14 #include <linux/skbuff.h>
16 #include <net/sock.h>
18 #include "ackvec.h"
19 #include "ccid.h"
20 #include "dccp.h"
22 static void dccp_fin(struct sock *sk, struct sk_buff *skb)
24 sk->sk_shutdown |= RCV_SHUTDOWN;
25 sock_set_flag(sk, SOCK_DONE);
26 __skb_pull(skb, dccp_hdr(skb)->dccph_doff * 4);
27 __skb_queue_tail(&sk->sk_receive_queue, skb);
28 skb_set_owner_r(skb, sk);
29 sk->sk_data_ready(sk, 0);
32 static void dccp_rcv_close(struct sock *sk, struct sk_buff *skb)
34 dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
35 dccp_fin(sk, skb);
36 dccp_set_state(sk, DCCP_CLOSED);
37 sk_wake_async(sk, 1, POLL_HUP);
40 static void dccp_rcv_closereq(struct sock *sk, struct sk_buff *skb)
43 * Step 7: Check for unexpected packet types
44 * If (S.is_server and P.type == CloseReq)
45 * Send Sync packet acknowledging P.seqno
46 * Drop packet and return
48 if (dccp_sk(sk)->dccps_role != DCCP_ROLE_CLIENT) {
49 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, DCCP_PKT_SYNC);
50 return;
53 if (sk->sk_state != DCCP_CLOSING)
54 dccp_set_state(sk, DCCP_CLOSING);
55 dccp_send_close(sk, 0);
58 static void dccp_event_ack_recv(struct sock *sk, struct sk_buff *skb)
60 struct dccp_sock *dp = dccp_sk(sk);
62 if (dccp_msk(sk)->dccpms_send_ack_vector)
63 dccp_ackvec_check_rcv_ackno(dp->dccps_hc_rx_ackvec, sk,
64 DCCP_SKB_CB(skb)->dccpd_ack_seq);
67 static int dccp_check_seqno(struct sock *sk, struct sk_buff *skb)
69 const struct dccp_hdr *dh = dccp_hdr(skb);
70 struct dccp_sock *dp = dccp_sk(sk);
71 u64 lswl, lawl;
74 * Step 5: Prepare sequence numbers for Sync
75 * If P.type == Sync or P.type == SyncAck,
76 * If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL,
77 * / * P is valid, so update sequence number variables
78 * accordingly. After this update, P will pass the tests
79 * in Step 6. A SyncAck is generated if necessary in
80 * Step 15 * /
81 * Update S.GSR, S.SWL, S.SWH
82 * Otherwise,
83 * Drop packet and return
85 if (dh->dccph_type == DCCP_PKT_SYNC ||
86 dh->dccph_type == DCCP_PKT_SYNCACK) {
87 if (between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
88 dp->dccps_awl, dp->dccps_awh) &&
89 dccp_delta_seqno(dp->dccps_swl,
90 DCCP_SKB_CB(skb)->dccpd_seq) >= 0)
91 dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq);
92 else
93 return -1;
97 * Step 6: Check sequence numbers
98 * Let LSWL = S.SWL and LAWL = S.AWL
99 * If P.type == CloseReq or P.type == Close or P.type == Reset,
100 * LSWL := S.GSR + 1, LAWL := S.GAR
101 * If LSWL <= P.seqno <= S.SWH
102 * and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH),
103 * Update S.GSR, S.SWL, S.SWH
104 * If P.type != Sync,
105 * Update S.GAR
106 * Otherwise,
107 * Send Sync packet acknowledging P.seqno
108 * Drop packet and return
110 lswl = dp->dccps_swl;
111 lawl = dp->dccps_awl;
113 if (dh->dccph_type == DCCP_PKT_CLOSEREQ ||
114 dh->dccph_type == DCCP_PKT_CLOSE ||
115 dh->dccph_type == DCCP_PKT_RESET) {
116 lswl = dp->dccps_gsr;
117 dccp_inc_seqno(&lswl);
118 lawl = dp->dccps_gar;
121 if (between48(DCCP_SKB_CB(skb)->dccpd_seq, lswl, dp->dccps_swh) &&
122 (DCCP_SKB_CB(skb)->dccpd_ack_seq == DCCP_PKT_WITHOUT_ACK_SEQ ||
123 between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
124 lawl, dp->dccps_awh))) {
125 dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq);
127 if (dh->dccph_type != DCCP_PKT_SYNC &&
128 (DCCP_SKB_CB(skb)->dccpd_ack_seq !=
129 DCCP_PKT_WITHOUT_ACK_SEQ))
130 dp->dccps_gar = DCCP_SKB_CB(skb)->dccpd_ack_seq;
131 } else {
132 DCCP_WARN("DCCP: Step 6 failed for %s packet, "
133 "(LSWL(%llu) <= P.seqno(%llu) <= S.SWH(%llu)) and "
134 "(P.ackno %s or LAWL(%llu) <= P.ackno(%llu) <= S.AWH(%llu), "
135 "sending SYNC...\n", dccp_packet_name(dh->dccph_type),
136 (unsigned long long) lswl,
137 (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq,
138 (unsigned long long) dp->dccps_swh,
139 (DCCP_SKB_CB(skb)->dccpd_ack_seq ==
140 DCCP_PKT_WITHOUT_ACK_SEQ) ? "doesn't exist" : "exists",
141 (unsigned long long) lawl,
142 (unsigned long long) DCCP_SKB_CB(skb)->dccpd_ack_seq,
143 (unsigned long long) dp->dccps_awh);
144 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq, DCCP_PKT_SYNC);
145 return -1;
148 return 0;
151 static int __dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
152 const struct dccp_hdr *dh, const unsigned len)
154 struct dccp_sock *dp = dccp_sk(sk);
156 switch (dccp_hdr(skb)->dccph_type) {
157 case DCCP_PKT_DATAACK:
158 case DCCP_PKT_DATA:
160 * FIXME: check if sk_receive_queue is full, schedule DATA_DROPPED
161 * option if it is.
163 __skb_pull(skb, dh->dccph_doff * 4);
164 __skb_queue_tail(&sk->sk_receive_queue, skb);
165 skb_set_owner_r(skb, sk);
166 sk->sk_data_ready(sk, 0);
167 return 0;
168 case DCCP_PKT_ACK:
169 goto discard;
170 case DCCP_PKT_RESET:
172 * Step 9: Process Reset
173 * If P.type == Reset,
174 * Tear down connection
175 * S.state := TIMEWAIT
176 * Set TIMEWAIT timer
177 * Drop packet and return
179 dccp_fin(sk, skb);
180 dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
181 return 0;
182 case DCCP_PKT_CLOSEREQ:
183 dccp_rcv_closereq(sk, skb);
184 goto discard;
185 case DCCP_PKT_CLOSE:
186 dccp_rcv_close(sk, skb);
187 return 0;
188 case DCCP_PKT_REQUEST:
189 /* Step 7
190 * or (S.is_server and P.type == Response)
191 * or (S.is_client and P.type == Request)
192 * or (S.state >= OPEN and P.type == Request
193 * and P.seqno >= S.OSR)
194 * or (S.state >= OPEN and P.type == Response
195 * and P.seqno >= S.OSR)
196 * or (S.state == RESPOND and P.type == Data),
197 * Send Sync packet acknowledging P.seqno
198 * Drop packet and return
200 if (dp->dccps_role != DCCP_ROLE_LISTEN)
201 goto send_sync;
202 goto check_seq;
203 case DCCP_PKT_RESPONSE:
204 if (dp->dccps_role != DCCP_ROLE_CLIENT)
205 goto send_sync;
206 check_seq:
207 if (dccp_delta_seqno(dp->dccps_osr,
208 DCCP_SKB_CB(skb)->dccpd_seq) >= 0) {
209 send_sync:
210 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
211 DCCP_PKT_SYNC);
213 break;
214 case DCCP_PKT_SYNC:
215 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
216 DCCP_PKT_SYNCACK);
218 * From RFC 4340, sec. 5.7
220 * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets
221 * MAY have non-zero-length application data areas, whose
222 * contents receivers MUST ignore.
224 goto discard;
227 DCCP_INC_STATS_BH(DCCP_MIB_INERRS);
228 discard:
229 __kfree_skb(skb);
230 return 0;
233 int dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
234 const struct dccp_hdr *dh, const unsigned len)
236 struct dccp_sock *dp = dccp_sk(sk);
238 if (dccp_check_seqno(sk, skb))
239 goto discard;
241 if (dccp_parse_options(sk, skb))
242 goto discard;
244 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
245 dccp_event_ack_recv(sk, skb);
247 if (dccp_msk(sk)->dccpms_send_ack_vector &&
248 dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
249 DCCP_SKB_CB(skb)->dccpd_seq,
250 DCCP_ACKVEC_STATE_RECEIVED))
251 goto discard;
253 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
254 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
256 return __dccp_rcv_established(sk, skb, dh, len);
257 discard:
258 __kfree_skb(skb);
259 return 0;
262 EXPORT_SYMBOL_GPL(dccp_rcv_established);
264 static int dccp_rcv_request_sent_state_process(struct sock *sk,
265 struct sk_buff *skb,
266 const struct dccp_hdr *dh,
267 const unsigned len)
270 * Step 4: Prepare sequence numbers in REQUEST
271 * If S.state == REQUEST,
272 * If (P.type == Response or P.type == Reset)
273 * and S.AWL <= P.ackno <= S.AWH,
274 * / * Set sequence number variables corresponding to the
275 * other endpoint, so P will pass the tests in Step 6 * /
276 * Set S.GSR, S.ISR, S.SWL, S.SWH
277 * / * Response processing continues in Step 10; Reset
278 * processing continues in Step 9 * /
280 if (dh->dccph_type == DCCP_PKT_RESPONSE) {
281 const struct inet_connection_sock *icsk = inet_csk(sk);
282 struct dccp_sock *dp = dccp_sk(sk);
284 /* Stop the REQUEST timer */
285 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
286 BUG_TRAP(sk->sk_send_head != NULL);
287 __kfree_skb(sk->sk_send_head);
288 sk->sk_send_head = NULL;
290 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
291 dp->dccps_awl, dp->dccps_awh)) {
292 dccp_pr_debug("invalid ackno: S.AWL=%llu, "
293 "P.ackno=%llu, S.AWH=%llu \n",
294 (unsigned long long)dp->dccps_awl,
295 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
296 (unsigned long long)dp->dccps_awh);
297 goto out_invalid_packet;
300 if (dccp_parse_options(sk, skb))
301 goto out_invalid_packet;
303 /* Obtain RTT sample from SYN exchange (used by CCID 3) */
304 if (dp->dccps_options_received.dccpor_timestamp_echo) {
305 struct timeval now;
307 dccp_timestamp(sk, &now);
308 dp->dccps_syn_rtt = dccp_sample_rtt(sk, &now, NULL);
311 if (dccp_msk(sk)->dccpms_send_ack_vector &&
312 dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
313 DCCP_SKB_CB(skb)->dccpd_seq,
314 DCCP_ACKVEC_STATE_RECEIVED))
315 goto out_invalid_packet; /* FIXME: change error code */
317 dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq;
318 dccp_update_gsr(sk, dp->dccps_isr);
320 * SWL and AWL are initially adjusted so that they are not less than
321 * the initial Sequence Numbers received and sent, respectively:
322 * SWL := max(GSR + 1 - floor(W/4), ISR),
323 * AWL := max(GSS - W' + 1, ISS).
324 * These adjustments MUST be applied only at the beginning of the
325 * connection.
327 * AWL was adjusted in dccp_v4_connect -acme
329 dccp_set_seqno(&dp->dccps_swl,
330 max48(dp->dccps_swl, dp->dccps_isr));
332 dccp_sync_mss(sk, icsk->icsk_pmtu_cookie);
335 * Step 10: Process REQUEST state (second part)
336 * If S.state == REQUEST,
337 * / * If we get here, P is a valid Response from the
338 * server (see Step 4), and we should move to
339 * PARTOPEN state. PARTOPEN means send an Ack,
340 * don't send Data packets, retransmit Acks
341 * periodically, and always include any Init Cookie
342 * from the Response * /
343 * S.state := PARTOPEN
344 * Set PARTOPEN timer
345 * Continue with S.state == PARTOPEN
346 * / * Step 12 will send the Ack completing the
347 * three-way handshake * /
349 dccp_set_state(sk, DCCP_PARTOPEN);
351 /* Make sure socket is routed, for correct metrics. */
352 icsk->icsk_af_ops->rebuild_header(sk);
354 if (!sock_flag(sk, SOCK_DEAD)) {
355 sk->sk_state_change(sk);
356 sk_wake_async(sk, 0, POLL_OUT);
359 if (sk->sk_write_pending || icsk->icsk_ack.pingpong ||
360 icsk->icsk_accept_queue.rskq_defer_accept) {
361 /* Save one ACK. Data will be ready after
362 * several ticks, if write_pending is set.
364 * It may be deleted, but with this feature tcpdumps
365 * look so _wonderfully_ clever, that I was not able
366 * to stand against the temptation 8) --ANK
369 * OK, in DCCP we can as well do a similar trick, its
370 * even in the draft, but there is no need for us to
371 * schedule an ack here, as dccp_sendmsg does this for
372 * us, also stated in the draft. -acme
374 __kfree_skb(skb);
375 return 0;
377 dccp_send_ack(sk);
378 return -1;
381 out_invalid_packet:
382 /* dccp_v4_do_rcv will send a reset */
383 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
384 return 1;
387 static int dccp_rcv_respond_partopen_state_process(struct sock *sk,
388 struct sk_buff *skb,
389 const struct dccp_hdr *dh,
390 const unsigned len)
392 int queued = 0;
394 switch (dh->dccph_type) {
395 case DCCP_PKT_RESET:
396 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
397 break;
398 case DCCP_PKT_DATA:
399 if (sk->sk_state == DCCP_RESPOND)
400 break;
401 case DCCP_PKT_DATAACK:
402 case DCCP_PKT_ACK:
404 * FIXME: we should be reseting the PARTOPEN (DELACK) timer
405 * here but only if we haven't used the DELACK timer for
406 * something else, like sending a delayed ack for a TIMESTAMP
407 * echo, etc, for now were not clearing it, sending an extra
408 * ACK when there is nothing else to do in DELACK is not a big
409 * deal after all.
412 /* Stop the PARTOPEN timer */
413 if (sk->sk_state == DCCP_PARTOPEN)
414 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
416 dccp_sk(sk)->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq;
417 dccp_set_state(sk, DCCP_OPEN);
419 if (dh->dccph_type == DCCP_PKT_DATAACK ||
420 dh->dccph_type == DCCP_PKT_DATA) {
421 __dccp_rcv_established(sk, skb, dh, len);
422 queued = 1; /* packet was queued
423 (by __dccp_rcv_established) */
425 break;
428 return queued;
431 int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
432 struct dccp_hdr *dh, unsigned len)
434 struct dccp_sock *dp = dccp_sk(sk);
435 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
436 const int old_state = sk->sk_state;
437 int queued = 0;
440 * Step 3: Process LISTEN state
442 * If S.state == LISTEN,
443 * If P.type == Request or P contains a valid Init Cookie option,
444 * (* Must scan the packet's options to check for Init
445 * Cookies. Only Init Cookies are processed here,
446 * however; other options are processed in Step 8. This
447 * scan need only be performed if the endpoint uses Init
448 * Cookies *)
449 * (* Generate a new socket and switch to that socket *)
450 * Set S := new socket for this port pair
451 * S.state = RESPOND
452 * Choose S.ISS (initial seqno) or set from Init Cookies
453 * Initialize S.GAR := S.ISS
454 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
455 * Cookies Continue with S.state == RESPOND
456 * (* A Response packet will be generated in Step 11 *)
457 * Otherwise,
458 * Generate Reset(No Connection) unless P.type == Reset
459 * Drop packet and return
461 if (sk->sk_state == DCCP_LISTEN) {
462 if (dh->dccph_type == DCCP_PKT_REQUEST) {
463 if (inet_csk(sk)->icsk_af_ops->conn_request(sk,
464 skb) < 0)
465 return 1;
467 /* FIXME: do congestion control initialization */
468 goto discard;
470 if (dh->dccph_type == DCCP_PKT_RESET)
471 goto discard;
473 /* Caller (dccp_v4_do_rcv) will send Reset */
474 dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
475 return 1;
478 if (sk->sk_state != DCCP_REQUESTING) {
479 if (dccp_check_seqno(sk, skb))
480 goto discard;
483 * Step 8: Process options and mark acknowledgeable
485 if (dccp_parse_options(sk, skb))
486 goto discard;
488 if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
489 dccp_event_ack_recv(sk, skb);
491 if (dccp_msk(sk)->dccpms_send_ack_vector &&
492 dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
493 DCCP_SKB_CB(skb)->dccpd_seq,
494 DCCP_ACKVEC_STATE_RECEIVED))
495 goto discard;
497 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
498 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
502 * Step 9: Process Reset
503 * If P.type == Reset,
504 * Tear down connection
505 * S.state := TIMEWAIT
506 * Set TIMEWAIT timer
507 * Drop packet and return
509 if (dh->dccph_type == DCCP_PKT_RESET) {
511 * Queue the equivalent of TCP fin so that dccp_recvmsg
512 * exits the loop
514 dccp_fin(sk, skb);
515 dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
516 return 0;
518 * Step 7: Check for unexpected packet types
519 * If (S.is_server and P.type == CloseReq)
520 * or (S.is_server and P.type == Response)
521 * or (S.is_client and P.type == Request)
522 * or (S.state == RESPOND and P.type == Data),
523 * Send Sync packet acknowledging P.seqno
524 * Drop packet and return
526 } else if ((dp->dccps_role != DCCP_ROLE_CLIENT &&
527 (dh->dccph_type == DCCP_PKT_RESPONSE ||
528 dh->dccph_type == DCCP_PKT_CLOSEREQ)) ||
529 (dp->dccps_role == DCCP_ROLE_CLIENT &&
530 dh->dccph_type == DCCP_PKT_REQUEST) ||
531 (sk->sk_state == DCCP_RESPOND &&
532 dh->dccph_type == DCCP_PKT_DATA)) {
533 dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNC);
534 goto discard;
535 } else if (dh->dccph_type == DCCP_PKT_CLOSEREQ) {
536 dccp_rcv_closereq(sk, skb);
537 goto discard;
538 } else if (dh->dccph_type == DCCP_PKT_CLOSE) {
539 dccp_rcv_close(sk, skb);
540 return 0;
543 if (unlikely(dh->dccph_type == DCCP_PKT_SYNC)) {
544 dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNCACK);
545 goto discard;
548 switch (sk->sk_state) {
549 case DCCP_CLOSED:
550 dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
551 return 1;
553 case DCCP_REQUESTING:
554 /* FIXME: do congestion control initialization */
556 queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len);
557 if (queued >= 0)
558 return queued;
560 __kfree_skb(skb);
561 return 0;
563 case DCCP_RESPOND:
564 case DCCP_PARTOPEN:
565 queued = dccp_rcv_respond_partopen_state_process(sk, skb,
566 dh, len);
567 break;
570 if (dh->dccph_type == DCCP_PKT_ACK ||
571 dh->dccph_type == DCCP_PKT_DATAACK) {
572 switch (old_state) {
573 case DCCP_PARTOPEN:
574 sk->sk_state_change(sk);
575 sk_wake_async(sk, 0, POLL_OUT);
576 break;
580 if (!queued) {
581 discard:
582 __kfree_skb(skb);
584 return 0;
587 EXPORT_SYMBOL_GPL(dccp_rcv_state_process);
590 * dccp_sample_rtt - Sample RTT from packet exchange
592 * @sk: connected dccp_sock
593 * @t_recv: receive timestamp of packet with timestamp echo
594 * @t_hist: packet history timestamp or NULL
596 u32 dccp_sample_rtt(struct sock *sk, struct timeval *t_recv,
597 struct timeval *t_hist)
599 struct dccp_sock *dp = dccp_sk(sk);
600 struct dccp_options_received *or = &dp->dccps_options_received;
601 suseconds_t delta;
603 if (t_hist == NULL) {
604 if (!or->dccpor_timestamp_echo) {
605 DCCP_WARN("packet without timestamp echo\n");
606 return DCCP_SANE_RTT_MAX;
608 timeval_sub_usecs(t_recv, or->dccpor_timestamp_echo * 10);
609 delta = timeval_usecs(t_recv);
610 } else
611 delta = timeval_delta(t_recv, t_hist);
613 delta -= or->dccpor_elapsed_time * 10; /* either set or 0 */
615 if (unlikely(delta <= 0)) {
616 DCCP_WARN("unusable RTT sample %ld, using min\n", (long)delta);
617 return DCCP_SANE_RTT_MIN;
619 if (unlikely(delta - (suseconds_t)DCCP_SANE_RTT_MAX > 0)) {
620 DCCP_WARN("RTT sample %ld too large, using max\n", (long)delta);
621 return DCCP_SANE_RTT_MAX;
624 return delta;
627 EXPORT_SYMBOL_GPL(dccp_sample_rtt);