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