Merge branch 'rbd-sysfs' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[pandora-kernel.git] / net / dccp / input.c
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                     after48(ackno, dp->dccps_gar))
244                         dp->dccps_gar = ackno;
245         } else {
246                 unsigned long now = jiffies;
247                 /*
248                  *   Step 6: Check sequence numbers
249                  *      Otherwise,
250                  *         If P.type == Reset,
251                  *            Send Sync packet acknowledging S.GSR
252                  *         Otherwise,
253                  *            Send Sync packet acknowledging P.seqno
254                  *      Drop packet and return
255                  *
256                  *   These Syncs are rate-limited as per RFC 4340, 7.5.4:
257                  *   at most 1 / (dccp_sync_rate_limit * HZ) Syncs per second.
258                  */
259                 if (time_before(now, (dp->dccps_rate_last +
260                                       sysctl_dccp_sync_ratelimit)))
261                         return 0;
262
263                 DCCP_WARN("Step 6 failed for %s packet, "
264                           "(LSWL(%llu) <= P.seqno(%llu) <= S.SWH(%llu)) and "
265                           "(P.ackno %s or LAWL(%llu) <= P.ackno(%llu) <= S.AWH(%llu), "
266                           "sending SYNC...\n",  dccp_packet_name(dh->dccph_type),
267                           (unsigned long long) lswl, (unsigned long long) seqno,
268                           (unsigned long long) dp->dccps_swh,
269                           (ackno == DCCP_PKT_WITHOUT_ACK_SEQ) ? "doesn't exist"
270                                                               : "exists",
271                           (unsigned long long) lawl, (unsigned long long) ackno,
272                           (unsigned long long) dp->dccps_awh);
273
274                 dp->dccps_rate_last = now;
275
276                 if (dh->dccph_type == DCCP_PKT_RESET)
277                         seqno = dp->dccps_gsr;
278                 dccp_send_sync(sk, seqno, DCCP_PKT_SYNC);
279                 return -1;
280         }
281
282         return 0;
283 }
284
285 static int __dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
286                                   const struct dccp_hdr *dh, const unsigned len)
287 {
288         struct dccp_sock *dp = dccp_sk(sk);
289
290         switch (dccp_hdr(skb)->dccph_type) {
291         case DCCP_PKT_DATAACK:
292         case DCCP_PKT_DATA:
293                 /*
294                  * FIXME: schedule DATA_DROPPED (RFC 4340, 11.7.2) if and when
295                  * - sk_shutdown == RCV_SHUTDOWN, use Code 1, "Not Listening"
296                  * - sk_receive_queue is full, use Code 2, "Receive Buffer"
297                  */
298                 dccp_enqueue_skb(sk, skb);
299                 return 0;
300         case DCCP_PKT_ACK:
301                 goto discard;
302         case DCCP_PKT_RESET:
303                 /*
304                  *  Step 9: Process Reset
305                  *      If P.type == Reset,
306                  *              Tear down connection
307                  *              S.state := TIMEWAIT
308                  *              Set TIMEWAIT timer
309                  *              Drop packet and return
310                  */
311                 dccp_rcv_reset(sk, skb);
312                 return 0;
313         case DCCP_PKT_CLOSEREQ:
314                 if (dccp_rcv_closereq(sk, skb))
315                         return 0;
316                 goto discard;
317         case DCCP_PKT_CLOSE:
318                 if (dccp_rcv_close(sk, skb))
319                         return 0;
320                 goto discard;
321         case DCCP_PKT_REQUEST:
322                 /* Step 7
323                  *   or (S.is_server and P.type == Response)
324                  *   or (S.is_client and P.type == Request)
325                  *   or (S.state >= OPEN and P.type == Request
326                  *      and P.seqno >= S.OSR)
327                  *    or (S.state >= OPEN and P.type == Response
328                  *      and P.seqno >= S.OSR)
329                  *    or (S.state == RESPOND and P.type == Data),
330                  *  Send Sync packet acknowledging P.seqno
331                  *  Drop packet and return
332                  */
333                 if (dp->dccps_role != DCCP_ROLE_LISTEN)
334                         goto send_sync;
335                 goto check_seq;
336         case DCCP_PKT_RESPONSE:
337                 if (dp->dccps_role != DCCP_ROLE_CLIENT)
338                         goto send_sync;
339 check_seq:
340                 if (dccp_delta_seqno(dp->dccps_osr,
341                                      DCCP_SKB_CB(skb)->dccpd_seq) >= 0) {
342 send_sync:
343                         dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
344                                        DCCP_PKT_SYNC);
345                 }
346                 break;
347         case DCCP_PKT_SYNC:
348                 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq,
349                                DCCP_PKT_SYNCACK);
350                 /*
351                  * From RFC 4340, sec. 5.7
352                  *
353                  * As with DCCP-Ack packets, DCCP-Sync and DCCP-SyncAck packets
354                  * MAY have non-zero-length application data areas, whose
355                  * contents receivers MUST ignore.
356                  */
357                 goto discard;
358         }
359
360         DCCP_INC_STATS_BH(DCCP_MIB_INERRS);
361 discard:
362         __kfree_skb(skb);
363         return 0;
364 }
365
366 int dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
367                          const struct dccp_hdr *dh, const unsigned len)
368 {
369         struct dccp_sock *dp = dccp_sk(sk);
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         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
378                 dccp_event_ack_recv(sk, skb);
379
380         if (dp->dccps_hc_rx_ackvec != NULL &&
381             dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
382                             DCCP_SKB_CB(skb)->dccpd_seq,
383                             DCCP_ACKVEC_STATE_RECEIVED))
384                 goto discard;
385         dccp_deliver_input_to_ccids(sk, skb);
386
387         return __dccp_rcv_established(sk, skb, dh, len);
388 discard:
389         __kfree_skb(skb);
390         return 0;
391 }
392
393 EXPORT_SYMBOL_GPL(dccp_rcv_established);
394
395 static int dccp_rcv_request_sent_state_process(struct sock *sk,
396                                                struct sk_buff *skb,
397                                                const struct dccp_hdr *dh,
398                                                const unsigned len)
399 {
400         /*
401          *  Step 4: Prepare sequence numbers in REQUEST
402          *     If S.state == REQUEST,
403          *        If (P.type == Response or P.type == Reset)
404          *              and S.AWL <= P.ackno <= S.AWH,
405          *           / * Set sequence number variables corresponding to the
406          *              other endpoint, so P will pass the tests in Step 6 * /
407          *           Set S.GSR, S.ISR, S.SWL, S.SWH
408          *           / * Response processing continues in Step 10; Reset
409          *              processing continues in Step 9 * /
410         */
411         if (dh->dccph_type == DCCP_PKT_RESPONSE) {
412                 const struct inet_connection_sock *icsk = inet_csk(sk);
413                 struct dccp_sock *dp = dccp_sk(sk);
414                 long tstamp = dccp_timestamp();
415
416                 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq,
417                                dp->dccps_awl, dp->dccps_awh)) {
418                         dccp_pr_debug("invalid ackno: S.AWL=%llu, "
419                                       "P.ackno=%llu, S.AWH=%llu\n",
420                                       (unsigned long long)dp->dccps_awl,
421                            (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
422                                       (unsigned long long)dp->dccps_awh);
423                         goto out_invalid_packet;
424                 }
425
426                 /*
427                  * If option processing (Step 8) failed, return 1 here so that
428                  * dccp_v4_do_rcv() sends a Reset. The Reset code depends on
429                  * the option type and is set in dccp_parse_options().
430                  */
431                 if (dccp_parse_options(sk, NULL, skb))
432                         return 1;
433
434                 /* Obtain usec RTT sample from SYN exchange (used by TFRC). */
435                 if (likely(dp->dccps_options_received.dccpor_timestamp_echo))
436                         dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * (tstamp -
437                             dp->dccps_options_received.dccpor_timestamp_echo));
438
439                 /* Stop the REQUEST timer */
440                 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
441                 WARN_ON(sk->sk_send_head == NULL);
442                 kfree_skb(sk->sk_send_head);
443                 sk->sk_send_head = NULL;
444
445                 /*
446                  * Set ISR, GSR from packet. ISS was set in dccp_v{4,6}_connect
447                  * and GSS in dccp_transmit_skb(). Setting AWL/AWH and SWL/SWH
448                  * is done as part of activating the feature values below, since
449                  * these settings depend on the local/remote Sequence Window
450                  * features, which were undefined or not confirmed until now.
451                  */
452                 dp->dccps_gsr = dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq;
453
454                 dccp_sync_mss(sk, icsk->icsk_pmtu_cookie);
455
456                 /*
457                  *    Step 10: Process REQUEST state (second part)
458                  *       If S.state == REQUEST,
459                  *        / * If we get here, P is a valid Response from the
460                  *            server (see Step 4), and we should move to
461                  *            PARTOPEN state. PARTOPEN means send an Ack,
462                  *            don't send Data packets, retransmit Acks
463                  *            periodically, and always include any Init Cookie
464                  *            from the Response * /
465                  *        S.state := PARTOPEN
466                  *        Set PARTOPEN timer
467                  *        Continue with S.state == PARTOPEN
468                  *        / * Step 12 will send the Ack completing the
469                  *            three-way handshake * /
470                  */
471                 dccp_set_state(sk, DCCP_PARTOPEN);
472
473                 /*
474                  * If feature negotiation was successful, activate features now;
475                  * an activation failure means that this host could not activate
476                  * one ore more features (e.g. insufficient memory), which would
477                  * leave at least one feature in an undefined state.
478                  */
479                 if (dccp_feat_activate_values(sk, &dp->dccps_featneg))
480                         goto unable_to_proceed;
481
482                 /* Make sure socket is routed, for correct metrics. */
483                 icsk->icsk_af_ops->rebuild_header(sk);
484
485                 if (!sock_flag(sk, SOCK_DEAD)) {
486                         sk->sk_state_change(sk);
487                         sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
488                 }
489
490                 if (sk->sk_write_pending || icsk->icsk_ack.pingpong ||
491                     icsk->icsk_accept_queue.rskq_defer_accept) {
492                         /* Save one ACK. Data will be ready after
493                          * several ticks, if write_pending is set.
494                          *
495                          * It may be deleted, but with this feature tcpdumps
496                          * look so _wonderfully_ clever, that I was not able
497                          * to stand against the temptation 8)     --ANK
498                          */
499                         /*
500                          * OK, in DCCP we can as well do a similar trick, its
501                          * even in the draft, but there is no need for us to
502                          * schedule an ack here, as dccp_sendmsg does this for
503                          * us, also stated in the draft. -acme
504                          */
505                         __kfree_skb(skb);
506                         return 0;
507                 }
508                 dccp_send_ack(sk);
509                 return -1;
510         }
511
512 out_invalid_packet:
513         /* dccp_v4_do_rcv will send a reset */
514         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR;
515         return 1;
516
517 unable_to_proceed:
518         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_ABORTED;
519         /*
520          * We mark this socket as no longer usable, so that the loop in
521          * dccp_sendmsg() terminates and the application gets notified.
522          */
523         dccp_set_state(sk, DCCP_CLOSED);
524         sk->sk_err = ECOMM;
525         return 1;
526 }
527
528 static int dccp_rcv_respond_partopen_state_process(struct sock *sk,
529                                                    struct sk_buff *skb,
530                                                    const struct dccp_hdr *dh,
531                                                    const unsigned len)
532 {
533         struct dccp_sock *dp = dccp_sk(sk);
534         u32 sample = dp->dccps_options_received.dccpor_timestamp_echo;
535         int queued = 0;
536
537         switch (dh->dccph_type) {
538         case DCCP_PKT_RESET:
539                 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
540                 break;
541         case DCCP_PKT_DATA:
542                 if (sk->sk_state == DCCP_RESPOND)
543                         break;
544         case DCCP_PKT_DATAACK:
545         case DCCP_PKT_ACK:
546                 /*
547                  * FIXME: we should be reseting the PARTOPEN (DELACK) timer
548                  * here but only if we haven't used the DELACK timer for
549                  * something else, like sending a delayed ack for a TIMESTAMP
550                  * echo, etc, for now were not clearing it, sending an extra
551                  * ACK when there is nothing else to do in DELACK is not a big
552                  * deal after all.
553                  */
554
555                 /* Stop the PARTOPEN timer */
556                 if (sk->sk_state == DCCP_PARTOPEN)
557                         inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
558
559                 /* Obtain usec RTT sample from SYN exchange (used by TFRC). */
560                 if (likely(sample)) {
561                         long delta = dccp_timestamp() - sample;
562
563                         dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * delta);
564                 }
565
566                 dp->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq;
567                 dccp_set_state(sk, DCCP_OPEN);
568
569                 if (dh->dccph_type == DCCP_PKT_DATAACK ||
570                     dh->dccph_type == DCCP_PKT_DATA) {
571                         __dccp_rcv_established(sk, skb, dh, len);
572                         queued = 1; /* packet was queued
573                                        (by __dccp_rcv_established) */
574                 }
575                 break;
576         }
577
578         return queued;
579 }
580
581 int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
582                            struct dccp_hdr *dh, unsigned len)
583 {
584         struct dccp_sock *dp = dccp_sk(sk);
585         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
586         const int old_state = sk->sk_state;
587         int queued = 0;
588
589         /*
590          *  Step 3: Process LISTEN state
591          *
592          *     If S.state == LISTEN,
593          *       If P.type == Request or P contains a valid Init Cookie option,
594          *            (* Must scan the packet's options to check for Init
595          *               Cookies.  Only Init Cookies are processed here,
596          *               however; other options are processed in Step 8.  This
597          *               scan need only be performed if the endpoint uses Init
598          *               Cookies *)
599          *            (* Generate a new socket and switch to that socket *)
600          *            Set S := new socket for this port pair
601          *            S.state = RESPOND
602          *            Choose S.ISS (initial seqno) or set from Init Cookies
603          *            Initialize S.GAR := S.ISS
604          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init
605          *            Cookies Continue with S.state == RESPOND
606          *            (* A Response packet will be generated in Step 11 *)
607          *       Otherwise,
608          *            Generate Reset(No Connection) unless P.type == Reset
609          *            Drop packet and return
610          */
611         if (sk->sk_state == DCCP_LISTEN) {
612                 if (dh->dccph_type == DCCP_PKT_REQUEST) {
613                         if (inet_csk(sk)->icsk_af_ops->conn_request(sk,
614                                                                     skb) < 0)
615                                 return 1;
616                         goto discard;
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         }
625
626         if (sk->sk_state != DCCP_REQUESTING && sk->sk_state != DCCP_RESPOND) {
627                 if (dccp_check_seqno(sk, skb))
628                         goto discard;
629
630                 /*
631                  * Step 8: Process options and mark acknowledgeable
632                  */
633                 if (dccp_parse_options(sk, NULL, skb))
634                         return 1;
635
636                 if (dcb->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
637                         dccp_event_ack_recv(sk, skb);
638
639                 if (dp->dccps_hc_rx_ackvec != NULL &&
640                     dccp_ackvec_add(dp->dccps_hc_rx_ackvec, sk,
641                                     DCCP_SKB_CB(skb)->dccpd_seq,
642                                     DCCP_ACKVEC_STATE_RECEIVED))
643                         goto discard;
644
645                 dccp_deliver_input_to_ccids(sk, skb);
646         }
647
648         /*
649          *  Step 9: Process Reset
650          *      If P.type == Reset,
651          *              Tear down connection
652          *              S.state := TIMEWAIT
653          *              Set TIMEWAIT timer
654          *              Drop packet and return
655         */
656         if (dh->dccph_type == DCCP_PKT_RESET) {
657                 dccp_rcv_reset(sk, skb);
658                 return 0;
659                 /*
660                  *   Step 7: Check for unexpected packet types
661                  *      If (S.is_server and P.type == Response)
662                  *          or (S.is_client and P.type == Request)
663                  *          or (S.state == RESPOND and P.type == Data),
664                  *        Send Sync packet acknowledging P.seqno
665                  *        Drop packet and return
666                  */
667         } else if ((dp->dccps_role != DCCP_ROLE_CLIENT &&
668                     dh->dccph_type == DCCP_PKT_RESPONSE) ||
669                     (dp->dccps_role == DCCP_ROLE_CLIENT &&
670                      dh->dccph_type == DCCP_PKT_REQUEST) ||
671                     (sk->sk_state == DCCP_RESPOND &&
672                      dh->dccph_type == DCCP_PKT_DATA)) {
673                 dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNC);
674                 goto discard;
675         } else if (dh->dccph_type == DCCP_PKT_CLOSEREQ) {
676                 if (dccp_rcv_closereq(sk, skb))
677                         return 0;
678                 goto discard;
679         } else if (dh->dccph_type == DCCP_PKT_CLOSE) {
680                 if (dccp_rcv_close(sk, skb))
681                         return 0;
682                 goto discard;
683         }
684
685         switch (sk->sk_state) {
686         case DCCP_CLOSED:
687                 dcb->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
688                 return 1;
689
690         case DCCP_REQUESTING:
691                 queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len);
692                 if (queued >= 0)
693                         return queued;
694
695                 __kfree_skb(skb);
696                 return 0;
697
698         case DCCP_RESPOND:
699         case DCCP_PARTOPEN:
700                 queued = dccp_rcv_respond_partopen_state_process(sk, skb,
701                                                                  dh, len);
702                 break;
703         }
704
705         if (dh->dccph_type == DCCP_PKT_ACK ||
706             dh->dccph_type == DCCP_PKT_DATAACK) {
707                 switch (old_state) {
708                 case DCCP_PARTOPEN:
709                         sk->sk_state_change(sk);
710                         sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
711                         break;
712                 }
713         } else if (unlikely(dh->dccph_type == DCCP_PKT_SYNC)) {
714                 dccp_send_sync(sk, dcb->dccpd_seq, DCCP_PKT_SYNCACK);
715                 goto discard;
716         }
717
718         if (!queued) {
719 discard:
720                 __kfree_skb(skb);
721         }
722         return 0;
723 }
724
725 EXPORT_SYMBOL_GPL(dccp_rcv_state_process);
726
727 /**
728  *  dccp_sample_rtt  -  Validate and finalise computation of RTT sample
729  *  @delta:     number of microseconds between packet and acknowledgment
730  *  The routine is kept generic to work in different contexts. It should be
731  *  called immediately when the ACK used for the RTT sample arrives.
732  */
733 u32 dccp_sample_rtt(struct sock *sk, long delta)
734 {
735         /* dccpor_elapsed_time is either zeroed out or set and > 0 */
736         delta -= dccp_sk(sk)->dccps_options_received.dccpor_elapsed_time * 10;
737
738         if (unlikely(delta <= 0)) {
739                 DCCP_WARN("unusable RTT sample %ld, using min\n", delta);
740                 return DCCP_SANE_RTT_MIN;
741         }
742         if (unlikely(delta > DCCP_SANE_RTT_MAX)) {
743                 DCCP_WARN("RTT sample %ld too large, using max\n", delta);
744                 return DCCP_SANE_RTT_MAX;
745         }
746
747         return delta;
748 }