Merge git://git.infradead.org/~dedekind/ubi-2.6
[linux-2.6] / net / dccp / options.c
1 /*
2  *  net/dccp/options.c
3  *
4  *  An implementation of the DCCP protocol
5  *  Copyright (c) 2005 Aristeu Sergio Rozanski Filho <aris@cathedrallabs.org>
6  *  Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
7  *  Copyright (c) 2005 Ian McDonald <ian.mcdonald@jandi.co.nz>
8  *
9  *      This program is free software; you can redistribute it and/or
10  *      modify it under the terms of the GNU General Public License
11  *      as published by the Free Software Foundation; either version
12  *      2 of the License, or (at your option) any later version.
13  */
14 #include <linux/dccp.h>
15 #include <linux/module.h>
16 #include <linux/types.h>
17 #include <asm/unaligned.h>
18 #include <linux/kernel.h>
19 #include <linux/skbuff.h>
20
21 #include "ackvec.h"
22 #include "ccid.h"
23 #include "dccp.h"
24 #include "feat.h"
25
26 int sysctl_dccp_feat_sequence_window = DCCPF_INITIAL_SEQUENCE_WINDOW;
27 int sysctl_dccp_feat_rx_ccid          = DCCPF_INITIAL_CCID;
28 int sysctl_dccp_feat_tx_ccid          = DCCPF_INITIAL_CCID;
29 int sysctl_dccp_feat_ack_ratio        = DCCPF_INITIAL_ACK_RATIO;
30 int sysctl_dccp_feat_send_ack_vector = DCCPF_INITIAL_SEND_ACK_VECTOR;
31 int sysctl_dccp_feat_send_ndp_count  = DCCPF_INITIAL_SEND_NDP_COUNT;
32
33 static u32 dccp_decode_value_var(const unsigned char *bf, const u8 len)
34 {
35         u32 value = 0;
36
37         if (len > 3)
38                 value += *bf++ << 24;
39         if (len > 2)
40                 value += *bf++ << 16;
41         if (len > 1)
42                 value += *bf++ << 8;
43         if (len > 0)
44                 value += *bf;
45
46         return value;
47 }
48
49 /**
50  * dccp_parse_options  -  Parse DCCP options present in @skb
51  * @sk: client|server|listening dccp socket (when @dreq != NULL)
52  * @dreq: request socket to use during connection setup, or NULL
53  */
54 int dccp_parse_options(struct sock *sk, struct dccp_request_sock *dreq,
55                        struct sk_buff *skb)
56 {
57         struct dccp_sock *dp = dccp_sk(sk);
58         const struct dccp_hdr *dh = dccp_hdr(skb);
59         const u8 pkt_type = DCCP_SKB_CB(skb)->dccpd_type;
60         u64 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
61         unsigned char *options = (unsigned char *)dh + dccp_hdr_len(skb);
62         unsigned char *opt_ptr = options;
63         const unsigned char *opt_end = (unsigned char *)dh +
64                                         (dh->dccph_doff * 4);
65         struct dccp_options_received *opt_recv = &dp->dccps_options_received;
66         unsigned char opt, len;
67         unsigned char *value;
68         u32 elapsed_time;
69         __be32 opt_val;
70         int rc;
71         int mandatory = 0;
72
73         memset(opt_recv, 0, sizeof(*opt_recv));
74
75         opt = len = 0;
76         while (opt_ptr != opt_end) {
77                 opt   = *opt_ptr++;
78                 len   = 0;
79                 value = NULL;
80
81                 /* Check if this isn't a single byte option */
82                 if (opt > DCCPO_MAX_RESERVED) {
83                         if (opt_ptr == opt_end)
84                                 goto out_invalid_option;
85
86                         len = *opt_ptr++;
87                         if (len < 3)
88                                 goto out_invalid_option;
89                         /*
90                          * Remove the type and len fields, leaving
91                          * just the value size
92                          */
93                         len     -= 2;
94                         value   = opt_ptr;
95                         opt_ptr += len;
96
97                         if (opt_ptr > opt_end)
98                                 goto out_invalid_option;
99                 }
100
101                 /*
102                  * CCID-Specific Options (from RFC 4340, sec. 10.3):
103                  *
104                  * Option numbers 128 through 191 are for options sent from the
105                  * HC-Sender to the HC-Receiver; option numbers 192 through 255
106                  * are for options sent from the HC-Receiver to the HC-Sender.
107                  *
108                  * CCID-specific options are ignored during connection setup, as
109                  * negotiation may still be in progress (see RFC 4340, 10.3).
110                  *
111                  */
112                 if (dreq != NULL && opt >= 128)
113                         goto ignore_option;
114
115                 switch (opt) {
116                 case DCCPO_PADDING:
117                         break;
118                 case DCCPO_MANDATORY:
119                         if (mandatory)
120                                 goto out_invalid_option;
121                         if (pkt_type != DCCP_PKT_DATA)
122                                 mandatory = 1;
123                         break;
124                 case DCCPO_NDP_COUNT:
125                         if (len > 3)
126                                 goto out_invalid_option;
127
128                         opt_recv->dccpor_ndp = dccp_decode_value_var(value, len);
129                         dccp_pr_debug("%s rx opt: NDP count=%d\n", dccp_role(sk),
130                                       opt_recv->dccpor_ndp);
131                         break;
132                 case DCCPO_CHANGE_L:
133                         /* fall through */
134                 case DCCPO_CHANGE_R:
135                         if (pkt_type == DCCP_PKT_DATA)
136                                 break;
137                         if (len < 2)
138                                 goto out_invalid_option;
139                         rc = dccp_feat_change_recv(sk, opt, *value, value + 1,
140                                                    len - 1);
141                         /*
142                          * When there is a change error, change_recv is
143                          * responsible for dealing with it.  i.e. reply with an
144                          * empty confirm.
145                          * If the change was mandatory, then we need to die.
146                          */
147                         if (rc && mandatory)
148                                 goto out_invalid_option;
149                         break;
150                 case DCCPO_CONFIRM_L:
151                         /* fall through */
152                 case DCCPO_CONFIRM_R:
153                         if (pkt_type == DCCP_PKT_DATA)
154                                 break;
155                         if (len < 2)    /* FIXME this disallows empty confirm */
156                                 goto out_invalid_option;
157                         if (dccp_feat_confirm_recv(sk, opt, *value,
158                                                    value + 1, len - 1))
159                                 goto out_invalid_option;
160                         break;
161                 case DCCPO_ACK_VECTOR_0:
162                 case DCCPO_ACK_VECTOR_1:
163                         if (dccp_packet_without_ack(skb))   /* RFC 4340, 11.4 */
164                                 break;
165
166                         if (dccp_msk(sk)->dccpms_send_ack_vector &&
167                             dccp_ackvec_parse(sk, skb, &ackno, opt, value, len))
168                                 goto out_invalid_option;
169                         break;
170                 case DCCPO_TIMESTAMP:
171                         if (len != 4)
172                                 goto out_invalid_option;
173                         /*
174                          * RFC 4340 13.1: "The precise time corresponding to
175                          * Timestamp Value zero is not specified". We use
176                          * zero to indicate absence of a meaningful timestamp.
177                          */
178                         opt_val = get_unaligned((__be32 *)value);
179                         if (unlikely(opt_val == 0)) {
180                                 DCCP_WARN("Timestamp with zero value\n");
181                                 break;
182                         }
183
184                         if (dreq != NULL) {
185                                 dreq->dreq_timestamp_echo = ntohl(opt_val);
186                                 dreq->dreq_timestamp_time = dccp_timestamp();
187                         } else {
188                                 opt_recv->dccpor_timestamp =
189                                         dp->dccps_timestamp_echo = ntohl(opt_val);
190                                 dp->dccps_timestamp_time = dccp_timestamp();
191                         }
192                         dccp_pr_debug("%s rx opt: TIMESTAMP=%u, ackno=%llu\n",
193                                       dccp_role(sk), ntohl(opt_val),
194                                       (unsigned long long)
195                                       DCCP_SKB_CB(skb)->dccpd_ack_seq);
196                         break;
197                 case DCCPO_TIMESTAMP_ECHO:
198                         if (len != 4 && len != 6 && len != 8)
199                                 goto out_invalid_option;
200
201                         opt_val = get_unaligned((__be32 *)value);
202                         opt_recv->dccpor_timestamp_echo = ntohl(opt_val);
203
204                         dccp_pr_debug("%s rx opt: TIMESTAMP_ECHO=%u, len=%d, "
205                                       "ackno=%llu", dccp_role(sk),
206                                       opt_recv->dccpor_timestamp_echo,
207                                       len + 2,
208                                       (unsigned long long)
209                                       DCCP_SKB_CB(skb)->dccpd_ack_seq);
210
211                         value += 4;
212
213                         if (len == 4) {         /* no elapsed time included */
214                                 dccp_pr_debug_cat("\n");
215                                 break;
216                         }
217
218                         if (len == 6) {         /* 2-byte elapsed time */
219                                 __be16 opt_val2 = get_unaligned((__be16 *)value);
220                                 elapsed_time = ntohs(opt_val2);
221                         } else {                /* 4-byte elapsed time */
222                                 opt_val = get_unaligned((__be32 *)value);
223                                 elapsed_time = ntohl(opt_val);
224                         }
225
226                         dccp_pr_debug_cat(", ELAPSED_TIME=%u\n", elapsed_time);
227
228                         /* Give precedence to the biggest ELAPSED_TIME */
229                         if (elapsed_time > opt_recv->dccpor_elapsed_time)
230                                 opt_recv->dccpor_elapsed_time = elapsed_time;
231                         break;
232                 case DCCPO_ELAPSED_TIME:
233                         if (dccp_packet_without_ack(skb))   /* RFC 4340, 13.2 */
234                                 break;
235
236                         if (len == 2) {
237                                 __be16 opt_val2 = get_unaligned((__be16 *)value);
238                                 elapsed_time = ntohs(opt_val2);
239                         } else if (len == 4) {
240                                 opt_val = get_unaligned((__be32 *)value);
241                                 elapsed_time = ntohl(opt_val);
242                         } else {
243                                 goto out_invalid_option;
244                         }
245
246                         if (elapsed_time > opt_recv->dccpor_elapsed_time)
247                                 opt_recv->dccpor_elapsed_time = elapsed_time;
248
249                         dccp_pr_debug("%s rx opt: ELAPSED_TIME=%d\n",
250                                       dccp_role(sk), elapsed_time);
251                         break;
252                 case 128 ... 191: {
253                         const u16 idx = value - options;
254
255                         if (ccid_hc_rx_parse_options(dp->dccps_hc_rx_ccid, sk,
256                                                      opt, len, idx,
257                                                      value) != 0)
258                                 goto out_invalid_option;
259                 }
260                         break;
261                 case 192 ... 255: {
262                         const u16 idx = value - options;
263
264                         if (ccid_hc_tx_parse_options(dp->dccps_hc_tx_ccid, sk,
265                                                      opt, len, idx,
266                                                      value) != 0)
267                                 goto out_invalid_option;
268                 }
269                         break;
270                 default:
271                         DCCP_CRIT("DCCP(%p): option %d(len=%d) not "
272                                   "implemented, ignoring", sk, opt, len);
273                         break;
274                 }
275 ignore_option:
276                 if (opt != DCCPO_MANDATORY)
277                         mandatory = 0;
278         }
279
280         /* mandatory was the last byte in option list -> reset connection */
281         if (mandatory)
282                 goto out_invalid_option;
283
284         return 0;
285
286 out_invalid_option:
287         DCCP_INC_STATS_BH(DCCP_MIB_INVALIDOPT);
288         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_OPTION_ERROR;
289         DCCP_WARN("DCCP(%p): invalid option %d, len=%d", sk, opt, len);
290         return -1;
291 }
292
293 EXPORT_SYMBOL_GPL(dccp_parse_options);
294
295 static void dccp_encode_value_var(const u32 value, unsigned char *to,
296                                   const unsigned int len)
297 {
298         if (len > 3)
299                 *to++ = (value & 0xFF000000) >> 24;
300         if (len > 2)
301                 *to++ = (value & 0xFF0000) >> 16;
302         if (len > 1)
303                 *to++ = (value & 0xFF00) >> 8;
304         if (len > 0)
305                 *to++ = (value & 0xFF);
306 }
307
308 static inline int dccp_ndp_len(const int ndp)
309 {
310         return likely(ndp <= 0xFF) ? 1 : ndp <= 0xFFFF ? 2 : 3;
311 }
312
313 int dccp_insert_option(struct sock *sk, struct sk_buff *skb,
314                         const unsigned char option,
315                         const void *value, const unsigned char len)
316 {
317         unsigned char *to;
318
319         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len + 2 > DCCP_MAX_OPT_LEN)
320                 return -1;
321
322         DCCP_SKB_CB(skb)->dccpd_opt_len += len + 2;
323
324         to    = skb_push(skb, len + 2);
325         *to++ = option;
326         *to++ = len + 2;
327
328         memcpy(to, value, len);
329         return 0;
330 }
331
332 EXPORT_SYMBOL_GPL(dccp_insert_option);
333
334 static int dccp_insert_option_ndp(struct sock *sk, struct sk_buff *skb)
335 {
336         struct dccp_sock *dp = dccp_sk(sk);
337         int ndp = dp->dccps_ndp_count;
338
339         if (dccp_non_data_packet(skb))
340                 ++dp->dccps_ndp_count;
341         else
342                 dp->dccps_ndp_count = 0;
343
344         if (ndp > 0) {
345                 unsigned char *ptr;
346                 const int ndp_len = dccp_ndp_len(ndp);
347                 const int len = ndp_len + 2;
348
349                 if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
350                         return -1;
351
352                 DCCP_SKB_CB(skb)->dccpd_opt_len += len;
353
354                 ptr = skb_push(skb, len);
355                 *ptr++ = DCCPO_NDP_COUNT;
356                 *ptr++ = len;
357                 dccp_encode_value_var(ndp, ptr, ndp_len);
358         }
359
360         return 0;
361 }
362
363 static inline int dccp_elapsed_time_len(const u32 elapsed_time)
364 {
365         return elapsed_time == 0 ? 0 : elapsed_time <= 0xFFFF ? 2 : 4;
366 }
367
368 int dccp_insert_option_elapsed_time(struct sock *sk, struct sk_buff *skb,
369                                     u32 elapsed_time)
370 {
371         const int elapsed_time_len = dccp_elapsed_time_len(elapsed_time);
372         const int len = 2 + elapsed_time_len;
373         unsigned char *to;
374
375         if (elapsed_time_len == 0)
376                 return 0;
377
378         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
379                 return -1;
380
381         DCCP_SKB_CB(skb)->dccpd_opt_len += len;
382
383         to    = skb_push(skb, len);
384         *to++ = DCCPO_ELAPSED_TIME;
385         *to++ = len;
386
387         if (elapsed_time_len == 2) {
388                 const __be16 var16 = htons((u16)elapsed_time);
389                 memcpy(to, &var16, 2);
390         } else {
391                 const __be32 var32 = htonl(elapsed_time);
392                 memcpy(to, &var32, 4);
393         }
394
395         return 0;
396 }
397
398 EXPORT_SYMBOL_GPL(dccp_insert_option_elapsed_time);
399
400 int dccp_insert_option_timestamp(struct sock *sk, struct sk_buff *skb)
401 {
402         __be32 now = htonl(dccp_timestamp());
403         /* yes this will overflow but that is the point as we want a
404          * 10 usec 32 bit timer which mean it wraps every 11.9 hours */
405
406         return dccp_insert_option(sk, skb, DCCPO_TIMESTAMP, &now, sizeof(now));
407 }
408
409 EXPORT_SYMBOL_GPL(dccp_insert_option_timestamp);
410
411 static int dccp_insert_option_timestamp_echo(struct dccp_sock *dp,
412                                              struct dccp_request_sock *dreq,
413                                              struct sk_buff *skb)
414 {
415         __be32 tstamp_echo;
416         unsigned char *to;
417         u32 elapsed_time, elapsed_time_len, len;
418
419         if (dreq != NULL) {
420                 elapsed_time = dccp_timestamp() - dreq->dreq_timestamp_time;
421                 tstamp_echo  = htonl(dreq->dreq_timestamp_echo);
422                 dreq->dreq_timestamp_echo = 0;
423         } else {
424                 elapsed_time = dccp_timestamp() - dp->dccps_timestamp_time;
425                 tstamp_echo  = htonl(dp->dccps_timestamp_echo);
426                 dp->dccps_timestamp_echo = 0;
427         }
428
429         elapsed_time_len = dccp_elapsed_time_len(elapsed_time);
430         len = 6 + elapsed_time_len;
431
432         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
433                 return -1;
434
435         DCCP_SKB_CB(skb)->dccpd_opt_len += len;
436
437         to    = skb_push(skb, len);
438         *to++ = DCCPO_TIMESTAMP_ECHO;
439         *to++ = len;
440
441         memcpy(to, &tstamp_echo, 4);
442         to += 4;
443
444         if (elapsed_time_len == 2) {
445                 const __be16 var16 = htons((u16)elapsed_time);
446                 memcpy(to, &var16, 2);
447         } else if (elapsed_time_len == 4) {
448                 const __be32 var32 = htonl(elapsed_time);
449                 memcpy(to, &var32, 4);
450         }
451
452         return 0;
453 }
454
455 static int dccp_insert_feat_opt(struct sk_buff *skb, u8 type, u8 feat,
456                                 u8 *val, u8 len)
457 {
458         u8 *to;
459
460         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len + 3 > DCCP_MAX_OPT_LEN) {
461                 DCCP_WARN("packet too small for feature %d option!\n", feat);
462                 return -1;
463         }
464
465         DCCP_SKB_CB(skb)->dccpd_opt_len += len + 3;
466
467         to    = skb_push(skb, len + 3);
468         *to++ = type;
469         *to++ = len + 3;
470         *to++ = feat;
471
472         if (len)
473                 memcpy(to, val, len);
474
475         dccp_pr_debug("%s(%s (%d), ...), length %d\n",
476                       dccp_feat_typename(type),
477                       dccp_feat_name(feat), feat, len);
478         return 0;
479 }
480
481 static int dccp_insert_options_feat(struct sock *sk, struct sk_buff *skb)
482 {
483         struct dccp_sock *dp = dccp_sk(sk);
484         struct dccp_minisock *dmsk = dccp_msk(sk);
485         struct dccp_opt_pend *opt, *next;
486         int change = 0;
487
488         /* confirm any options [NN opts] */
489         list_for_each_entry_safe(opt, next, &dmsk->dccpms_conf, dccpop_node) {
490                 dccp_insert_feat_opt(skb, opt->dccpop_type,
491                                      opt->dccpop_feat, opt->dccpop_val,
492                                      opt->dccpop_len);
493                 /* fear empty confirms */
494                 if (opt->dccpop_val)
495                         kfree(opt->dccpop_val);
496                 kfree(opt);
497         }
498         INIT_LIST_HEAD(&dmsk->dccpms_conf);
499
500         /* see which features we need to send */
501         list_for_each_entry(opt, &dmsk->dccpms_pending, dccpop_node) {
502                 /* see if we need to send any confirm */
503                 if (opt->dccpop_sc) {
504                         dccp_insert_feat_opt(skb, opt->dccpop_type + 1,
505                                              opt->dccpop_feat,
506                                              opt->dccpop_sc->dccpoc_val,
507                                              opt->dccpop_sc->dccpoc_len);
508
509                         BUG_ON(!opt->dccpop_sc->dccpoc_val);
510                         kfree(opt->dccpop_sc->dccpoc_val);
511                         kfree(opt->dccpop_sc);
512                         opt->dccpop_sc = NULL;
513                 }
514
515                 /* any option not confirmed, re-send it */
516                 if (!opt->dccpop_conf) {
517                         dccp_insert_feat_opt(skb, opt->dccpop_type,
518                                              opt->dccpop_feat, opt->dccpop_val,
519                                              opt->dccpop_len);
520                         change++;
521                 }
522         }
523
524         /* Retransmit timer.
525          * If this is the master listening sock, we don't set a timer on it.  It
526          * should be fine because if the dude doesn't receive our RESPONSE
527          * [which will contain the CHANGE] he will send another REQUEST which
528          * will "retrnasmit" the change.
529          */
530         if (change && dp->dccps_role != DCCP_ROLE_LISTEN) {
531                 dccp_pr_debug("reset feat negotiation timer %p\n", sk);
532
533                 /* XXX don't reset the timer on re-transmissions.  I.e. reset it
534                  * only when sending new stuff i guess.  Currently the timer
535                  * never backs off because on re-transmission it just resets it!
536                  */
537                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
538                                           inet_csk(sk)->icsk_rto, DCCP_RTO_MAX);
539         }
540
541         return 0;
542 }
543
544 /* The length of all options needs to be a multiple of 4 (5.8) */
545 static void dccp_insert_option_padding(struct sk_buff *skb)
546 {
547         int padding = DCCP_SKB_CB(skb)->dccpd_opt_len % 4;
548
549         if (padding != 0) {
550                 padding = 4 - padding;
551                 memset(skb_push(skb, padding), 0, padding);
552                 DCCP_SKB_CB(skb)->dccpd_opt_len += padding;
553         }
554 }
555
556 int dccp_insert_options(struct sock *sk, struct sk_buff *skb)
557 {
558         struct dccp_sock *dp = dccp_sk(sk);
559         struct dccp_minisock *dmsk = dccp_msk(sk);
560
561         DCCP_SKB_CB(skb)->dccpd_opt_len = 0;
562
563         if (dmsk->dccpms_send_ndp_count &&
564             dccp_insert_option_ndp(sk, skb))
565                 return -1;
566
567         if (!dccp_packet_without_ack(skb)) {
568                 if (dmsk->dccpms_send_ack_vector &&
569                     dccp_ackvec_pending(dp->dccps_hc_rx_ackvec) &&
570                     dccp_insert_option_ackvec(sk, skb))
571                         return -1;
572         }
573
574         if (dp->dccps_hc_rx_insert_options) {
575                 if (ccid_hc_rx_insert_options(dp->dccps_hc_rx_ccid, sk, skb))
576                         return -1;
577                 dp->dccps_hc_rx_insert_options = 0;
578         }
579
580         /* Feature negotiation */
581         /* Data packets can't do feat negotiation */
582         if (DCCP_SKB_CB(skb)->dccpd_type != DCCP_PKT_DATA &&
583             DCCP_SKB_CB(skb)->dccpd_type != DCCP_PKT_DATAACK &&
584             dccp_insert_options_feat(sk, skb))
585                 return -1;
586
587         /*
588          * Obtain RTT sample from Request/Response exchange.
589          * This is currently used in CCID 3 initialisation.
590          */
591         if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_REQUEST &&
592             dccp_insert_option_timestamp(sk, skb))
593                 return -1;
594
595         if (dp->dccps_timestamp_echo != 0 &&
596             dccp_insert_option_timestamp_echo(dp, NULL, skb))
597                 return -1;
598
599         dccp_insert_option_padding(skb);
600         return 0;
601 }
602
603 int dccp_insert_options_rsk(struct dccp_request_sock *dreq, struct sk_buff *skb)
604 {
605         DCCP_SKB_CB(skb)->dccpd_opt_len = 0;
606
607         if (dreq->dreq_timestamp_echo != 0 &&
608             dccp_insert_option_timestamp_echo(NULL, dreq, skb))
609                 return -1;
610
611         dccp_insert_option_padding(skb);
612         return 0;
613 }