Merge branch 'linus' into x86/irqstats
[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                  * The same applies to Ack Vectors, as these depend on the CCID.
111                  *
112                  */
113                 if (dreq != NULL && (opt >= 128 ||
114                     opt == DCCPO_ACK_VECTOR_0 || opt == DCCPO_ACK_VECTOR_1))
115                         goto ignore_option;
116
117                 switch (opt) {
118                 case DCCPO_PADDING:
119                         break;
120                 case DCCPO_MANDATORY:
121                         if (mandatory)
122                                 goto out_invalid_option;
123                         if (pkt_type != DCCP_PKT_DATA)
124                                 mandatory = 1;
125                         break;
126                 case DCCPO_NDP_COUNT:
127                         if (len > 3)
128                                 goto out_invalid_option;
129
130                         opt_recv->dccpor_ndp = dccp_decode_value_var(value, len);
131                         dccp_pr_debug("%s rx opt: NDP count=%d\n", dccp_role(sk),
132                                       opt_recv->dccpor_ndp);
133                         break;
134                 case DCCPO_CHANGE_L:
135                         /* fall through */
136                 case DCCPO_CHANGE_R:
137                         if (pkt_type == DCCP_PKT_DATA)
138                                 break;
139                         if (len < 2)
140                                 goto out_invalid_option;
141                         rc = dccp_feat_change_recv(sk, opt, *value, value + 1,
142                                                    len - 1);
143                         /*
144                          * When there is a change error, change_recv is
145                          * responsible for dealing with it.  i.e. reply with an
146                          * empty confirm.
147                          * If the change was mandatory, then we need to die.
148                          */
149                         if (rc && mandatory)
150                                 goto out_invalid_option;
151                         break;
152                 case DCCPO_CONFIRM_L:
153                         /* fall through */
154                 case DCCPO_CONFIRM_R:
155                         if (pkt_type == DCCP_PKT_DATA)
156                                 break;
157                         if (len < 2)    /* FIXME this disallows empty confirm */
158                                 goto out_invalid_option;
159                         if (dccp_feat_confirm_recv(sk, opt, *value,
160                                                    value + 1, len - 1))
161                                 goto out_invalid_option;
162                         break;
163                 case DCCPO_ACK_VECTOR_0:
164                 case DCCPO_ACK_VECTOR_1:
165                         if (dccp_packet_without_ack(skb))   /* RFC 4340, 11.4 */
166                                 break;
167
168                         if (dccp_msk(sk)->dccpms_send_ack_vector &&
169                             dccp_ackvec_parse(sk, skb, &ackno, opt, value, len))
170                                 goto out_invalid_option;
171                         break;
172                 case DCCPO_TIMESTAMP:
173                         if (len != 4)
174                                 goto out_invalid_option;
175                         /*
176                          * RFC 4340 13.1: "The precise time corresponding to
177                          * Timestamp Value zero is not specified". We use
178                          * zero to indicate absence of a meaningful timestamp.
179                          */
180                         opt_val = get_unaligned((__be32 *)value);
181                         if (unlikely(opt_val == 0)) {
182                                 DCCP_WARN("Timestamp with zero value\n");
183                                 break;
184                         }
185
186                         if (dreq != NULL) {
187                                 dreq->dreq_timestamp_echo = ntohl(opt_val);
188                                 dreq->dreq_timestamp_time = dccp_timestamp();
189                         } else {
190                                 opt_recv->dccpor_timestamp =
191                                         dp->dccps_timestamp_echo = ntohl(opt_val);
192                                 dp->dccps_timestamp_time = dccp_timestamp();
193                         }
194                         dccp_pr_debug("%s rx opt: TIMESTAMP=%u, ackno=%llu\n",
195                                       dccp_role(sk), ntohl(opt_val),
196                                       (unsigned long long)
197                                       DCCP_SKB_CB(skb)->dccpd_ack_seq);
198                         break;
199                 case DCCPO_TIMESTAMP_ECHO:
200                         if (len != 4 && len != 6 && len != 8)
201                                 goto out_invalid_option;
202
203                         opt_val = get_unaligned((__be32 *)value);
204                         opt_recv->dccpor_timestamp_echo = ntohl(opt_val);
205
206                         dccp_pr_debug("%s rx opt: TIMESTAMP_ECHO=%u, len=%d, "
207                                       "ackno=%llu", dccp_role(sk),
208                                       opt_recv->dccpor_timestamp_echo,
209                                       len + 2,
210                                       (unsigned long long)
211                                       DCCP_SKB_CB(skb)->dccpd_ack_seq);
212
213                         value += 4;
214
215                         if (len == 4) {         /* no elapsed time included */
216                                 dccp_pr_debug_cat("\n");
217                                 break;
218                         }
219
220                         if (len == 6) {         /* 2-byte elapsed time */
221                                 __be16 opt_val2 = get_unaligned((__be16 *)value);
222                                 elapsed_time = ntohs(opt_val2);
223                         } else {                /* 4-byte elapsed time */
224                                 opt_val = get_unaligned((__be32 *)value);
225                                 elapsed_time = ntohl(opt_val);
226                         }
227
228                         dccp_pr_debug_cat(", ELAPSED_TIME=%u\n", elapsed_time);
229
230                         /* Give precedence to the biggest ELAPSED_TIME */
231                         if (elapsed_time > opt_recv->dccpor_elapsed_time)
232                                 opt_recv->dccpor_elapsed_time = elapsed_time;
233                         break;
234                 case DCCPO_ELAPSED_TIME:
235                         if (dccp_packet_without_ack(skb))   /* RFC 4340, 13.2 */
236                                 break;
237
238                         if (len == 2) {
239                                 __be16 opt_val2 = get_unaligned((__be16 *)value);
240                                 elapsed_time = ntohs(opt_val2);
241                         } else if (len == 4) {
242                                 opt_val = get_unaligned((__be32 *)value);
243                                 elapsed_time = ntohl(opt_val);
244                         } else {
245                                 goto out_invalid_option;
246                         }
247
248                         if (elapsed_time > opt_recv->dccpor_elapsed_time)
249                                 opt_recv->dccpor_elapsed_time = elapsed_time;
250
251                         dccp_pr_debug("%s rx opt: ELAPSED_TIME=%d\n",
252                                       dccp_role(sk), elapsed_time);
253                         break;
254                 case 128 ... 191: {
255                         const u16 idx = value - options;
256
257                         if (ccid_hc_rx_parse_options(dp->dccps_hc_rx_ccid, sk,
258                                                      opt, len, idx,
259                                                      value) != 0)
260                                 goto out_invalid_option;
261                 }
262                         break;
263                 case 192 ... 255: {
264                         const u16 idx = value - options;
265
266                         if (ccid_hc_tx_parse_options(dp->dccps_hc_tx_ccid, sk,
267                                                      opt, len, idx,
268                                                      value) != 0)
269                                 goto out_invalid_option;
270                 }
271                         break;
272                 default:
273                         DCCP_CRIT("DCCP(%p): option %d(len=%d) not "
274                                   "implemented, ignoring", sk, opt, len);
275                         break;
276                 }
277 ignore_option:
278                 if (opt != DCCPO_MANDATORY)
279                         mandatory = 0;
280         }
281
282         /* mandatory was the last byte in option list -> reset connection */
283         if (mandatory)
284                 goto out_invalid_option;
285
286         return 0;
287
288 out_invalid_option:
289         DCCP_INC_STATS_BH(DCCP_MIB_INVALIDOPT);
290         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_OPTION_ERROR;
291         DCCP_WARN("DCCP(%p): invalid option %d, len=%d", sk, opt, len);
292         return -1;
293 }
294
295 EXPORT_SYMBOL_GPL(dccp_parse_options);
296
297 static void dccp_encode_value_var(const u32 value, unsigned char *to,
298                                   const unsigned int len)
299 {
300         if (len > 3)
301                 *to++ = (value & 0xFF000000) >> 24;
302         if (len > 2)
303                 *to++ = (value & 0xFF0000) >> 16;
304         if (len > 1)
305                 *to++ = (value & 0xFF00) >> 8;
306         if (len > 0)
307                 *to++ = (value & 0xFF);
308 }
309
310 static inline int dccp_ndp_len(const int ndp)
311 {
312         return likely(ndp <= 0xFF) ? 1 : ndp <= 0xFFFF ? 2 : 3;
313 }
314
315 int dccp_insert_option(struct sock *sk, struct sk_buff *skb,
316                         const unsigned char option,
317                         const void *value, const unsigned char len)
318 {
319         unsigned char *to;
320
321         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len + 2 > DCCP_MAX_OPT_LEN)
322                 return -1;
323
324         DCCP_SKB_CB(skb)->dccpd_opt_len += len + 2;
325
326         to    = skb_push(skb, len + 2);
327         *to++ = option;
328         *to++ = len + 2;
329
330         memcpy(to, value, len);
331         return 0;
332 }
333
334 EXPORT_SYMBOL_GPL(dccp_insert_option);
335
336 static int dccp_insert_option_ndp(struct sock *sk, struct sk_buff *skb)
337 {
338         struct dccp_sock *dp = dccp_sk(sk);
339         int ndp = dp->dccps_ndp_count;
340
341         if (dccp_non_data_packet(skb))
342                 ++dp->dccps_ndp_count;
343         else
344                 dp->dccps_ndp_count = 0;
345
346         if (ndp > 0) {
347                 unsigned char *ptr;
348                 const int ndp_len = dccp_ndp_len(ndp);
349                 const int len = ndp_len + 2;
350
351                 if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
352                         return -1;
353
354                 DCCP_SKB_CB(skb)->dccpd_opt_len += len;
355
356                 ptr = skb_push(skb, len);
357                 *ptr++ = DCCPO_NDP_COUNT;
358                 *ptr++ = len;
359                 dccp_encode_value_var(ndp, ptr, ndp_len);
360         }
361
362         return 0;
363 }
364
365 static inline int dccp_elapsed_time_len(const u32 elapsed_time)
366 {
367         return elapsed_time == 0 ? 0 : elapsed_time <= 0xFFFF ? 2 : 4;
368 }
369
370 int dccp_insert_option_elapsed_time(struct sock *sk, struct sk_buff *skb,
371                                     u32 elapsed_time)
372 {
373         const int elapsed_time_len = dccp_elapsed_time_len(elapsed_time);
374         const int len = 2 + elapsed_time_len;
375         unsigned char *to;
376
377         if (elapsed_time_len == 0)
378                 return 0;
379
380         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
381                 return -1;
382
383         DCCP_SKB_CB(skb)->dccpd_opt_len += len;
384
385         to    = skb_push(skb, len);
386         *to++ = DCCPO_ELAPSED_TIME;
387         *to++ = len;
388
389         if (elapsed_time_len == 2) {
390                 const __be16 var16 = htons((u16)elapsed_time);
391                 memcpy(to, &var16, 2);
392         } else {
393                 const __be32 var32 = htonl(elapsed_time);
394                 memcpy(to, &var32, 4);
395         }
396
397         return 0;
398 }
399
400 EXPORT_SYMBOL_GPL(dccp_insert_option_elapsed_time);
401
402 int dccp_insert_option_timestamp(struct sock *sk, struct sk_buff *skb)
403 {
404         __be32 now = htonl(dccp_timestamp());
405         /* yes this will overflow but that is the point as we want a
406          * 10 usec 32 bit timer which mean it wraps every 11.9 hours */
407
408         return dccp_insert_option(sk, skb, DCCPO_TIMESTAMP, &now, sizeof(now));
409 }
410
411 EXPORT_SYMBOL_GPL(dccp_insert_option_timestamp);
412
413 static int dccp_insert_option_timestamp_echo(struct dccp_sock *dp,
414                                              struct dccp_request_sock *dreq,
415                                              struct sk_buff *skb)
416 {
417         __be32 tstamp_echo;
418         unsigned char *to;
419         u32 elapsed_time, elapsed_time_len, len;
420
421         if (dreq != NULL) {
422                 elapsed_time = dccp_timestamp() - dreq->dreq_timestamp_time;
423                 tstamp_echo  = htonl(dreq->dreq_timestamp_echo);
424                 dreq->dreq_timestamp_echo = 0;
425         } else {
426                 elapsed_time = dccp_timestamp() - dp->dccps_timestamp_time;
427                 tstamp_echo  = htonl(dp->dccps_timestamp_echo);
428                 dp->dccps_timestamp_echo = 0;
429         }
430
431         elapsed_time_len = dccp_elapsed_time_len(elapsed_time);
432         len = 6 + elapsed_time_len;
433
434         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
435                 return -1;
436
437         DCCP_SKB_CB(skb)->dccpd_opt_len += len;
438
439         to    = skb_push(skb, len);
440         *to++ = DCCPO_TIMESTAMP_ECHO;
441         *to++ = len;
442
443         memcpy(to, &tstamp_echo, 4);
444         to += 4;
445
446         if (elapsed_time_len == 2) {
447                 const __be16 var16 = htons((u16)elapsed_time);
448                 memcpy(to, &var16, 2);
449         } else if (elapsed_time_len == 4) {
450                 const __be32 var32 = htonl(elapsed_time);
451                 memcpy(to, &var32, 4);
452         }
453
454         return 0;
455 }
456
457 static int dccp_insert_feat_opt(struct sk_buff *skb, u8 type, u8 feat,
458                                 u8 *val, u8 len)
459 {
460         u8 *to;
461
462         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len + 3 > DCCP_MAX_OPT_LEN) {
463                 DCCP_WARN("packet too small for feature %d option!\n", feat);
464                 return -1;
465         }
466
467         DCCP_SKB_CB(skb)->dccpd_opt_len += len + 3;
468
469         to    = skb_push(skb, len + 3);
470         *to++ = type;
471         *to++ = len + 3;
472         *to++ = feat;
473
474         if (len)
475                 memcpy(to, val, len);
476
477         dccp_pr_debug("%s(%s (%d), ...), length %d\n",
478                       dccp_feat_typename(type),
479                       dccp_feat_name(feat), feat, len);
480         return 0;
481 }
482
483 static int dccp_insert_options_feat(struct sock *sk, struct sk_buff *skb)
484 {
485         struct dccp_sock *dp = dccp_sk(sk);
486         struct dccp_minisock *dmsk = dccp_msk(sk);
487         struct dccp_opt_pend *opt, *next;
488         int change = 0;
489
490         /* confirm any options [NN opts] */
491         list_for_each_entry_safe(opt, next, &dmsk->dccpms_conf, dccpop_node) {
492                 dccp_insert_feat_opt(skb, opt->dccpop_type,
493                                      opt->dccpop_feat, opt->dccpop_val,
494                                      opt->dccpop_len);
495                 /* fear empty confirms */
496                 if (opt->dccpop_val)
497                         kfree(opt->dccpop_val);
498                 kfree(opt);
499         }
500         INIT_LIST_HEAD(&dmsk->dccpms_conf);
501
502         /* see which features we need to send */
503         list_for_each_entry(opt, &dmsk->dccpms_pending, dccpop_node) {
504                 /* see if we need to send any confirm */
505                 if (opt->dccpop_sc) {
506                         dccp_insert_feat_opt(skb, opt->dccpop_type + 1,
507                                              opt->dccpop_feat,
508                                              opt->dccpop_sc->dccpoc_val,
509                                              opt->dccpop_sc->dccpoc_len);
510
511                         BUG_ON(!opt->dccpop_sc->dccpoc_val);
512                         kfree(opt->dccpop_sc->dccpoc_val);
513                         kfree(opt->dccpop_sc);
514                         opt->dccpop_sc = NULL;
515                 }
516
517                 /* any option not confirmed, re-send it */
518                 if (!opt->dccpop_conf) {
519                         dccp_insert_feat_opt(skb, opt->dccpop_type,
520                                              opt->dccpop_feat, opt->dccpop_val,
521                                              opt->dccpop_len);
522                         change++;
523                 }
524         }
525
526         /* Retransmit timer.
527          * If this is the master listening sock, we don't set a timer on it.  It
528          * should be fine because if the dude doesn't receive our RESPONSE
529          * [which will contain the CHANGE] he will send another REQUEST which
530          * will "retrnasmit" the change.
531          */
532         if (change && dp->dccps_role != DCCP_ROLE_LISTEN) {
533                 dccp_pr_debug("reset feat negotiation timer %p\n", sk);
534
535                 /* XXX don't reset the timer on re-transmissions.  I.e. reset it
536                  * only when sending new stuff i guess.  Currently the timer
537                  * never backs off because on re-transmission it just resets it!
538                  */
539                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
540                                           inet_csk(sk)->icsk_rto, DCCP_RTO_MAX);
541         }
542
543         return 0;
544 }
545
546 /* The length of all options needs to be a multiple of 4 (5.8) */
547 static void dccp_insert_option_padding(struct sk_buff *skb)
548 {
549         int padding = DCCP_SKB_CB(skb)->dccpd_opt_len % 4;
550
551         if (padding != 0) {
552                 padding = 4 - padding;
553                 memset(skb_push(skb, padding), 0, padding);
554                 DCCP_SKB_CB(skb)->dccpd_opt_len += padding;
555         }
556 }
557
558 int dccp_insert_options(struct sock *sk, struct sk_buff *skb)
559 {
560         struct dccp_sock *dp = dccp_sk(sk);
561         struct dccp_minisock *dmsk = dccp_msk(sk);
562
563         DCCP_SKB_CB(skb)->dccpd_opt_len = 0;
564
565         if (dmsk->dccpms_send_ndp_count &&
566             dccp_insert_option_ndp(sk, skb))
567                 return -1;
568
569         if (!dccp_packet_without_ack(skb)) {
570                 if (dmsk->dccpms_send_ack_vector &&
571                     dccp_ackvec_pending(dp->dccps_hc_rx_ackvec) &&
572                     dccp_insert_option_ackvec(sk, skb))
573                         return -1;
574         }
575
576         if (dp->dccps_hc_rx_insert_options) {
577                 if (ccid_hc_rx_insert_options(dp->dccps_hc_rx_ccid, sk, skb))
578                         return -1;
579                 dp->dccps_hc_rx_insert_options = 0;
580         }
581
582         /* Feature negotiation */
583         /* Data packets can't do feat negotiation */
584         if (DCCP_SKB_CB(skb)->dccpd_type != DCCP_PKT_DATA &&
585             DCCP_SKB_CB(skb)->dccpd_type != DCCP_PKT_DATAACK &&
586             dccp_insert_options_feat(sk, skb))
587                 return -1;
588
589         /*
590          * Obtain RTT sample from Request/Response exchange.
591          * This is currently used in CCID 3 initialisation.
592          */
593         if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_REQUEST &&
594             dccp_insert_option_timestamp(sk, skb))
595                 return -1;
596
597         if (dp->dccps_timestamp_echo != 0 &&
598             dccp_insert_option_timestamp_echo(dp, NULL, skb))
599                 return -1;
600
601         dccp_insert_option_padding(skb);
602         return 0;
603 }
604
605 int dccp_insert_options_rsk(struct dccp_request_sock *dreq, struct sk_buff *skb)
606 {
607         DCCP_SKB_CB(skb)->dccpd_opt_len = 0;
608
609         if (dreq->dreq_timestamp_echo != 0 &&
610             dccp_insert_option_timestamp_echo(NULL, dreq, skb))
611                 return -1;
612
613         dccp_insert_option_padding(skb);
614         return 0;
615 }