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