[TCPDIAG]: Introduce inet_diag_{register,unregister}
[safe/jmp/linux-2.6] / 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/config.h>
14 #include <linux/dccp.h>
15 #include <linux/skbuff.h>
16
17 #include <net/sock.h>
18
19 #include "ccid.h"
20 #include "dccp.h"
21
22 static void dccp_fin(struct sock *sk, struct sk_buff *skb)
23 {
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);
30 }
31
32 static void dccp_rcv_close(struct sock *sk, struct sk_buff *skb)
33 {
34         switch (sk->sk_state) {
35         case DCCP_PARTOPEN:
36         case DCCP_OPEN:
37                 dccp_v4_send_reset(sk, DCCP_RESET_CODE_CLOSED);
38                 dccp_fin(sk, skb);
39                 dccp_set_state(sk, DCCP_CLOSED);
40                 break;
41         }
42 }
43
44 static void dccp_rcv_closereq(struct sock *sk, struct sk_buff *skb)
45 {
46         /*
47          *   Step 7: Check for unexpected packet types
48          *      If (S.is_server and P.type == CloseReq)
49          *        Send Sync packet acknowledging P.seqno
50          *        Drop packet and return
51          */
52         if (dccp_sk(sk)->dccps_role != DCCP_ROLE_CLIENT) {
53                 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq);
54                 return;
55         }
56
57         switch (sk->sk_state) {
58         case DCCP_PARTOPEN:
59         case DCCP_OPEN:
60                 dccp_set_state(sk, DCCP_CLOSING);
61                 dccp_send_close(sk);
62                 break;
63         }
64 }
65
66 static inline void dccp_event_ack_recv(struct sock *sk, struct sk_buff *skb)
67 {
68         struct dccp_sock *dp = dccp_sk(sk);
69
70         if (dp->dccps_options.dccpo_send_ack_vector)
71                 dccp_ackpkts_check_rcv_ackno(dp->dccps_hc_rx_ackpkts, sk,
72                                              DCCP_SKB_CB(skb)->dccpd_ack_seq);
73 }
74
75 static int dccp_check_seqno(struct sock *sk, struct sk_buff *skb)
76 {
77         const struct dccp_hdr *dh = dccp_hdr(skb);
78         struct dccp_sock *dp = dccp_sk(sk);
79         u64 lswl = dp->dccps_swl;
80         u64 lawl = dp->dccps_awl;
81
82         /*
83          *   Step 5: Prepare sequence numbers for Sync
84          *     If P.type == Sync or P.type == SyncAck,
85          *        If S.AWL <= P.ackno <= S.AWH and P.seqno >= S.SWL,
86          *           / * P is valid, so update sequence number variables
87          *               accordingly.  After this update, P will pass the tests
88          *               in Step 6.  A SyncAck is generated if necessary in
89          *               Step 15 * /
90          *           Update S.GSR, S.SWL, S.SWH
91          *        Otherwise,
92          *           Drop packet and return
93          */
94         if (dh->dccph_type == DCCP_PKT_SYNC || 
95             dh->dccph_type == DCCP_PKT_SYNCACK) {
96                 if (between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, dp->dccps_awl, dp->dccps_awh) &&
97                     !before48(DCCP_SKB_CB(skb)->dccpd_seq, dp->dccps_swl))
98                         dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq);
99                 else
100                         return -1;
101         /*
102          *   Step 6: Check sequence numbers
103          *      Let LSWL = S.SWL and LAWL = S.AWL
104          *      If P.type == CloseReq or P.type == Close or P.type == Reset,
105          *        LSWL := S.GSR + 1, LAWL := S.GAR
106          *      If LSWL <= P.seqno <= S.SWH
107          *           and (P.ackno does not exist or LAWL <= P.ackno <= S.AWH),
108          *        Update S.GSR, S.SWL, S.SWH
109          *        If P.type != Sync,
110          *           Update S.GAR
111          *      Otherwise,
112          *        Send Sync packet acknowledging P.seqno
113          *        Drop packet and return
114          */
115         } else if (dh->dccph_type == DCCP_PKT_CLOSEREQ ||
116                    dh->dccph_type == DCCP_PKT_CLOSE ||
117                    dh->dccph_type == DCCP_PKT_RESET) {
118                 lswl = dp->dccps_gsr;
119                 dccp_inc_seqno(&lswl);
120                 lawl = dp->dccps_gar;
121         }
122
123         if (between48(DCCP_SKB_CB(skb)->dccpd_seq, lswl, dp->dccps_swh) &&
124             (DCCP_SKB_CB(skb)->dccpd_ack_seq == DCCP_PKT_WITHOUT_ACK_SEQ ||
125              between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, lawl, dp->dccps_awh))) {
126                 dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq);
127
128                 if (dh->dccph_type != DCCP_PKT_SYNC &&
129                     DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
130                         dp->dccps_gar = DCCP_SKB_CB(skb)->dccpd_ack_seq;
131         } else {
132                 dccp_pr_debug("Step 6 failed, sending SYNC...\n");
133                 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq);
134                 return -1;
135         }
136
137         return 0;
138 }
139
140 int dccp_rcv_established(struct sock *sk, struct sk_buff *skb,
141                          const struct dccp_hdr *dh, const unsigned len)
142 {
143         struct dccp_sock *dp = dccp_sk(sk);
144
145         if (dccp_check_seqno(sk, skb))
146                 goto discard;
147
148         if (dccp_parse_options(sk, skb))
149                 goto discard;
150
151         if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
152                 dccp_event_ack_recv(sk, skb);
153
154         /*
155          * FIXME: check ECN to see if we should use
156          * DCCP_ACKPKTS_STATE_ECN_MARKED
157          */
158         if (dp->dccps_options.dccpo_send_ack_vector) {
159                 struct dccp_ackpkts *ap = dp->dccps_hc_rx_ackpkts;
160
161                 if (dccp_ackpkts_add(dp->dccps_hc_rx_ackpkts,
162                                      DCCP_SKB_CB(skb)->dccpd_seq,
163                                      DCCP_ACKPKTS_STATE_RECEIVED)) {
164                         LIMIT_NETDEBUG(KERN_INFO "DCCP: acknowledgeable packets buffer full!\n");
165                         ap->dccpap_ack_seqno = DCCP_MAX_SEQNO + 1;
166                         inet_csk_schedule_ack(sk);
167                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, TCP_DELACK_MIN, TCP_RTO_MAX);
168                         goto discard;
169                 }
170
171                 /*
172                  * FIXME: this activation is probably wrong, have to study more
173                  * TCP delack machinery and how it fits into DCCP draft, but
174                  * for now it kinda "works" 8)
175                  */
176                 if (!inet_csk_ack_scheduled(sk)) {
177                         inet_csk_schedule_ack(sk);
178                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 5 * HZ, TCP_RTO_MAX);
179                 }
180         }
181
182         ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
183         ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
184
185         switch (dccp_hdr(skb)->dccph_type) {
186         case DCCP_PKT_DATAACK:
187         case DCCP_PKT_DATA:
188                 /*
189                  * FIXME: check if sk_receive_queue is full, schedule DATA_DROPPED option
190                  * if it is.
191                  */
192                 __skb_pull(skb, dh->dccph_doff * 4);
193                 __skb_queue_tail(&sk->sk_receive_queue, skb);
194                 skb_set_owner_r(skb, sk);
195                 sk->sk_data_ready(sk, 0);
196                 return 0;
197         case DCCP_PKT_ACK:
198                 goto discard;
199         case DCCP_PKT_RESET:
200                 /*
201                  *  Step 9: Process Reset
202                  *      If P.type == Reset,
203                  *              Tear down connection
204                  *              S.state := TIMEWAIT
205                  *              Set TIMEWAIT timer
206                  *              Drop packet and return
207                 */
208                 dccp_fin(sk, skb);
209                 dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
210                 return 0;
211         case DCCP_PKT_CLOSEREQ:
212                 dccp_rcv_closereq(sk, skb);
213                 goto discard;
214         case DCCP_PKT_CLOSE:
215                 dccp_rcv_close(sk, skb);
216                 return 0;
217         case DCCP_PKT_REQUEST:
218                 /* Step 7 
219                  *   or (S.is_server and P.type == Response)
220                  *   or (S.is_client and P.type == Request)
221                  *   or (S.state >= OPEN and P.type == Request
222                  *      and P.seqno >= S.OSR)
223                  *    or (S.state >= OPEN and P.type == Response
224                  *      and P.seqno >= S.OSR)
225                  *    or (S.state == RESPOND and P.type == Data),
226                  *  Send Sync packet acknowledging P.seqno
227                  *  Drop packet and return
228                  */
229                 if (dp->dccps_role != DCCP_ROLE_LISTEN)
230                         goto send_sync;
231                 goto check_seq;
232         case DCCP_PKT_RESPONSE:
233                 if (dp->dccps_role != DCCP_ROLE_CLIENT)
234                         goto send_sync;
235 check_seq:
236                 if (!before48(DCCP_SKB_CB(skb)->dccpd_seq, dp->dccps_osr)) {
237 send_sync:
238                         dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq);
239                 }
240                 break;
241         }
242
243         DCCP_INC_STATS_BH(DCCP_MIB_INERRS);
244 discard:
245         __kfree_skb(skb);
246         return 0;
247 }
248
249 static int dccp_rcv_request_sent_state_process(struct sock *sk,
250                                                struct sk_buff *skb,
251                                                const struct dccp_hdr *dh,
252                                                const unsigned len)
253 {
254         /* 
255          *  Step 4: Prepare sequence numbers in REQUEST
256          *     If S.state == REQUEST,
257          *        If (P.type == Response or P.type == Reset)
258          *              and S.AWL <= P.ackno <= S.AWH,
259          *           / * Set sequence number variables corresponding to the
260          *              other endpoint, so P will pass the tests in Step 6 * /
261          *           Set S.GSR, S.ISR, S.SWL, S.SWH
262          *           / * Response processing continues in Step 10; Reset
263          *              processing continues in Step 9 * /
264         */
265         if (dh->dccph_type == DCCP_PKT_RESPONSE) {
266                 const struct inet_connection_sock *icsk = inet_csk(sk);
267                 struct dccp_sock *dp = dccp_sk(sk);
268
269                 /* Stop the REQUEST timer */
270                 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
271                 BUG_TRAP(sk->sk_send_head != NULL);
272                 __kfree_skb(sk->sk_send_head);
273                 sk->sk_send_head = NULL;
274
275                 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, dp->dccps_awl, dp->dccps_awh)) {
276                         dccp_pr_debug("invalid ackno: S.AWL=%llu, P.ackno=%llu, S.AWH=%llu \n",
277                                       (unsigned long long) dp->dccps_awl,
278                                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_ack_seq,
279                                       (unsigned long long) dp->dccps_awh);
280                         goto out_invalid_packet;
281                 }
282
283                 dp->dccps_isr = DCCP_SKB_CB(skb)->dccpd_seq;
284                 dccp_update_gsr(sk, DCCP_SKB_CB(skb)->dccpd_seq);
285
286                 if (ccid_hc_rx_init(dp->dccps_hc_rx_ccid, sk) != 0 ||
287                     ccid_hc_tx_init(dp->dccps_hc_tx_ccid, sk) != 0) {
288                         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
289                         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
290                         /* FIXME: send appropriate RESET code */
291                         goto out_invalid_packet;
292                 }
293
294                 dccp_sync_mss(sk, dp->dccps_pmtu_cookie);
295
296                 /*
297                  *    Step 10: Process REQUEST state (second part)
298                  *       If S.state == REQUEST,
299                  *        / * If we get here, P is a valid Response from the server (see
300                  *           Step 4), and we should move to PARTOPEN state.  PARTOPEN
301                  *           means send an Ack, don't send Data packets, retransmit
302                  *           Acks periodically, and always include any Init Cookie from
303                  *           the Response * /
304                  *        S.state := PARTOPEN
305                  *        Set PARTOPEN timer
306                  *        Continue with S.state == PARTOPEN
307                  *        / * Step 12 will send the Ack completing the three-way
308                  *           handshake * /
309                  */
310                 dccp_set_state(sk, DCCP_PARTOPEN);
311
312                 /* Make sure socket is routed, for correct metrics. */
313                 inet_sk_rebuild_header(sk);
314
315                 if (!sock_flag(sk, SOCK_DEAD)) {
316                         sk->sk_state_change(sk);
317                         sk_wake_async(sk, 0, POLL_OUT);
318                 }
319
320                 if (sk->sk_write_pending || icsk->icsk_ack.pingpong ||
321                     icsk->icsk_accept_queue.rskq_defer_accept) {
322                         /* Save one ACK. Data will be ready after
323                          * several ticks, if write_pending is set.
324                          *
325                          * It may be deleted, but with this feature tcpdumps
326                          * look so _wonderfully_ clever, that I was not able
327                          * to stand against the temptation 8)     --ANK
328                          */
329                         /*
330                          * OK, in DCCP we can as well do a similar trick, its
331                          * even in the draft, but there is no need for us to
332                          * schedule an ack here, as dccp_sendmsg does this for
333                          * us, also stated in the draft. -acme
334                          */
335                         __kfree_skb(skb);
336                         return 0;
337                 } 
338                 dccp_send_ack(sk);
339                 return -1;
340         }
341
342 out_invalid_packet:
343         return 1; /* dccp_v4_do_rcv will send a reset, but...
344                      FIXME: the reset code should be DCCP_RESET_CODE_PACKET_ERROR  */
345 }
346
347 static int dccp_rcv_respond_partopen_state_process(struct sock *sk,
348                                                    struct sk_buff *skb,
349                                                    const struct dccp_hdr *dh,
350                                                    const unsigned len)
351 {
352         int queued = 0;
353
354         switch (dh->dccph_type) {
355         case DCCP_PKT_RESET:
356                 inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
357                 break;
358         case DCCP_PKT_DATAACK:
359         case DCCP_PKT_ACK:
360                 /*
361                  * FIXME: we should be reseting the PARTOPEN (DELACK) timer here,
362                  * but only if we haven't used the DELACK timer for something else,
363                  * like sending a delayed ack for a TIMESTAMP echo, etc, for now
364                  * were not clearing it, sending an extra ACK when there is nothing
365                  * else to do in DELACK is not a big deal after all.
366                  */
367
368                 /* Stop the PARTOPEN timer */
369                 if (sk->sk_state == DCCP_PARTOPEN)
370                         inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
371
372                 dccp_sk(sk)->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq;
373                 dccp_set_state(sk, DCCP_OPEN);
374
375                 if (dh->dccph_type == DCCP_PKT_DATAACK) {
376                         dccp_rcv_established(sk, skb, dh, len);
377                         queued = 1; /* packet was queued (by dccp_rcv_established) */
378                 }
379                 break;
380         }
381
382         return queued;
383 }
384
385 int dccp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
386                            struct dccp_hdr *dh, unsigned len)
387 {
388         struct dccp_sock *dp = dccp_sk(sk);
389         const int old_state = sk->sk_state;
390         int queued = 0;
391
392         if (sk->sk_state != DCCP_LISTEN && sk->sk_state != DCCP_REQUESTING) {
393                 if (dccp_check_seqno(sk, skb))
394                         goto discard;
395
396                 /*
397                  * Step 8: Process options and mark acknowledgeable
398                  */
399                 if (dccp_parse_options(sk, skb))
400                         goto discard;
401
402                 if (DCCP_SKB_CB(skb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
403                         dccp_event_ack_recv(sk, skb);
404
405                 ccid_hc_rx_packet_recv(dp->dccps_hc_rx_ccid, sk, skb);
406                 ccid_hc_tx_packet_recv(dp->dccps_hc_tx_ccid, sk, skb);
407
408                 /*
409                  * FIXME: check ECN to see if we should use
410                  * DCCP_ACKPKTS_STATE_ECN_MARKED
411                  */
412                 if (dp->dccps_options.dccpo_send_ack_vector) {
413                         if (dccp_ackpkts_add(dp->dccps_hc_rx_ackpkts,
414                                              DCCP_SKB_CB(skb)->dccpd_seq,
415                                              DCCP_ACKPKTS_STATE_RECEIVED))
416                                 goto discard;
417                         /*
418                          * FIXME: this activation is probably wrong, have to study more
419                          * TCP delack machinery and how it fits into DCCP draft, but
420                          * for now it kinda "works" 8)
421                          */
422                         if (dp->dccps_hc_rx_ackpkts->dccpap_ack_seqno == DCCP_MAX_SEQNO + 1 &&
423                             !inet_csk_ack_scheduled(sk)) {
424                                 inet_csk_schedule_ack(sk);
425                                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, TCP_DELACK_MIN, TCP_RTO_MAX);
426                         }
427                 }
428         }
429
430         /*
431          *  Step 9: Process Reset
432          *      If P.type == Reset,
433          *              Tear down connection
434          *              S.state := TIMEWAIT
435          *              Set TIMEWAIT timer
436          *              Drop packet and return
437         */
438         if (dh->dccph_type == DCCP_PKT_RESET) {
439                 /* Queue the equivalent of TCP fin so that dccp_recvmsg exits the loop */
440                 dccp_fin(sk, skb);
441                 dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
442                 return 0;
443                 /*
444                  *   Step 7: Check for unexpected packet types
445                  *      If (S.is_server and P.type == CloseReq)
446                  *          or (S.is_server and P.type == Response)
447                  *          or (S.is_client and P.type == Request)
448                  *          or (S.state == RESPOND and P.type == Data),
449                  *        Send Sync packet acknowledging P.seqno
450                  *        Drop packet and return
451                  */
452         } else if ((dp->dccps_role != DCCP_ROLE_CLIENT &&
453                     (dh->dccph_type == DCCP_PKT_RESPONSE || dh->dccph_type == DCCP_PKT_CLOSEREQ)) ||
454                     (dp->dccps_role == DCCP_ROLE_CLIENT &&
455                      dh->dccph_type == DCCP_PKT_REQUEST) ||
456                     (sk->sk_state == DCCP_RESPOND && dh->dccph_type == DCCP_PKT_DATA)) {
457                 dccp_send_sync(sk, DCCP_SKB_CB(skb)->dccpd_seq);
458                 goto discard;
459         }
460
461         switch (sk->sk_state) {
462         case DCCP_CLOSED:
463                 return 1;
464
465         case DCCP_LISTEN:
466                 if (dh->dccph_type == DCCP_PKT_ACK ||
467                     dh->dccph_type == DCCP_PKT_DATAACK)
468                         return 1;
469
470                 if (dh->dccph_type == DCCP_PKT_RESET)
471                         goto discard;
472
473                 if (dh->dccph_type == DCCP_PKT_REQUEST) {
474                         if (dccp_v4_conn_request(sk, skb) < 0)
475                                 return 1;
476
477                         /* FIXME: do congestion control initialization */
478                         goto discard;
479                 }
480                 goto discard;
481
482         case DCCP_REQUESTING:
483                 /* FIXME: do congestion control initialization */
484
485                 queued = dccp_rcv_request_sent_state_process(sk, skb, dh, len);
486                 if (queued >= 0)
487                         return queued;
488
489                 __kfree_skb(skb);
490                 return 0;
491
492         case DCCP_RESPOND:
493         case DCCP_PARTOPEN:
494                 queued = dccp_rcv_respond_partopen_state_process(sk, skb, dh, len);
495                 break;
496         }
497
498         if (dh->dccph_type == DCCP_PKT_ACK || dh->dccph_type == DCCP_PKT_DATAACK) {
499                 switch (old_state) {
500                 case DCCP_PARTOPEN:
501                         sk->sk_state_change(sk);
502                         sk_wake_async(sk, 0, POLL_OUT);
503                         break;
504                 }
505         }
506
507         if (!queued) { 
508 discard:
509                 __kfree_skb(skb);
510         }
511         return 0;
512 }