nfsd: nfsd should drop CAP_MKNOD for non-root
[safe/jmp/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
30 u64 dccp_decode_value_var(const u8 *bf, const u8 len)
31 {
32         u64 value = 0;
33
34         if (len >= DCCP_OPTVAL_MAXLEN)
35                 value += ((u64)*bf++) << 40;
36         if (len > 4)
37                 value += ((u64)*bf++) << 32;
38         if (len > 3)
39                 value += ((u64)*bf++) << 24;
40         if (len > 2)
41                 value += ((u64)*bf++) << 16;
42         if (len > 1)
43                 value += ((u64)*bf++) << 8;
44         if (len > 0)
45                 value += *bf;
46
47         return value;
48 }
49
50 /**
51  * dccp_parse_options  -  Parse DCCP options present in @skb
52  * @sk: client|server|listening dccp socket (when @dreq != NULL)
53  * @dreq: request socket to use during connection setup, or NULL
54  */
55 int dccp_parse_options(struct sock *sk, struct dccp_request_sock *dreq,
56                        struct sk_buff *skb)
57 {
58         struct dccp_sock *dp = dccp_sk(sk);
59         const struct dccp_hdr *dh = dccp_hdr(skb);
60         const u8 pkt_type = DCCP_SKB_CB(skb)->dccpd_type;
61         u64 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
62         unsigned char *options = (unsigned char *)dh + dccp_hdr_len(skb);
63         unsigned char *opt_ptr = options;
64         const unsigned char *opt_end = (unsigned char *)dh +
65                                         (dh->dccph_doff * 4);
66         struct dccp_options_received *opt_recv = &dp->dccps_options_received;
67         unsigned char opt, len;
68         unsigned char *uninitialized_var(value);
69         u32 elapsed_time;
70         __be32 opt_val;
71         int rc;
72         int mandatory = 0;
73
74         memset(opt_recv, 0, sizeof(*opt_recv));
75
76         opt = len = 0;
77         while (opt_ptr != opt_end) {
78                 opt   = *opt_ptr++;
79                 len   = 0;
80                 value = NULL;
81
82                 /* Check if this isn't a single byte option */
83                 if (opt > DCCPO_MAX_RESERVED) {
84                         if (opt_ptr == opt_end)
85                                 goto out_nonsensical_length;
86
87                         len = *opt_ptr++;
88                         if (len < 2)
89                                 goto out_nonsensical_length;
90                         /*
91                          * Remove the type and len fields, leaving
92                          * just the value size
93                          */
94                         len     -= 2;
95                         value   = opt_ptr;
96                         opt_ptr += len;
97
98                         if (opt_ptr > opt_end)
99                                 goto out_nonsensical_length;
100                 }
101
102                 /*
103                  * CCID-Specific Options (from RFC 4340, sec. 10.3):
104                  *
105                  * Option numbers 128 through 191 are for options sent from the
106                  * HC-Sender to the HC-Receiver; option numbers 192 through 255
107                  * are for options sent from the HC-Receiver to the HC-Sender.
108                  *
109                  * CCID-specific options are ignored during connection setup, as
110                  * negotiation may still be in progress (see RFC 4340, 10.3).
111                  * The same applies to Ack Vectors, as these depend on the CCID.
112                  *
113                  */
114                 if (dreq != NULL && (opt >= 128 ||
115                     opt == DCCPO_ACK_VECTOR_0 || opt == DCCPO_ACK_VECTOR_1))
116                         goto ignore_option;
117
118                 switch (opt) {
119                 case DCCPO_PADDING:
120                         break;
121                 case DCCPO_MANDATORY:
122                         if (mandatory)
123                                 goto out_invalid_option;
124                         if (pkt_type != DCCP_PKT_DATA)
125                                 mandatory = 1;
126                         break;
127                 case DCCPO_NDP_COUNT:
128                         if (len > 6)
129                                 goto out_invalid_option;
130
131                         opt_recv->dccpor_ndp = dccp_decode_value_var(value, len);
132                         dccp_pr_debug("%s opt: NDP count=%llu\n", dccp_role(sk),
133                                       (unsigned long long)opt_recv->dccpor_ndp);
134                         break;
135                 case DCCPO_CHANGE_L ... DCCPO_CONFIRM_R:
136                         if (pkt_type == DCCP_PKT_DATA)      /* RFC 4340, 6 */
137                                 break;
138                         rc = dccp_feat_parse_options(sk, dreq, mandatory, opt,
139                                                     *value, value + 1, len - 1);
140                         if (rc)
141                                 goto out_featneg_failed;
142                         break;
143                 case DCCPO_ACK_VECTOR_0:
144                 case DCCPO_ACK_VECTOR_1:
145                         if (dccp_packet_without_ack(skb))   /* RFC 4340, 11.4 */
146                                 break;
147                         if (dp->dccps_hc_rx_ackvec != NULL &&
148                             dccp_ackvec_parse(sk, skb, &ackno, opt, value, len))
149                                 goto out_invalid_option;
150                         break;
151                 case DCCPO_TIMESTAMP:
152                         if (len != 4)
153                                 goto out_invalid_option;
154                         /*
155                          * RFC 4340 13.1: "The precise time corresponding to
156                          * Timestamp Value zero is not specified". We use
157                          * zero to indicate absence of a meaningful timestamp.
158                          */
159                         opt_val = get_unaligned((__be32 *)value);
160                         if (unlikely(opt_val == 0)) {
161                                 DCCP_WARN("Timestamp with zero value\n");
162                                 break;
163                         }
164
165                         if (dreq != NULL) {
166                                 dreq->dreq_timestamp_echo = ntohl(opt_val);
167                                 dreq->dreq_timestamp_time = dccp_timestamp();
168                         } else {
169                                 opt_recv->dccpor_timestamp =
170                                         dp->dccps_timestamp_echo = ntohl(opt_val);
171                                 dp->dccps_timestamp_time = dccp_timestamp();
172                         }
173                         dccp_pr_debug("%s rx opt: TIMESTAMP=%u, ackno=%llu\n",
174                                       dccp_role(sk), ntohl(opt_val),
175                                       (unsigned long long)
176                                       DCCP_SKB_CB(skb)->dccpd_ack_seq);
177                         break;
178                 case DCCPO_TIMESTAMP_ECHO:
179                         if (len != 4 && len != 6 && len != 8)
180                                 goto out_invalid_option;
181
182                         opt_val = get_unaligned((__be32 *)value);
183                         opt_recv->dccpor_timestamp_echo = ntohl(opt_val);
184
185                         dccp_pr_debug("%s rx opt: TIMESTAMP_ECHO=%u, len=%d, "
186                                       "ackno=%llu", dccp_role(sk),
187                                       opt_recv->dccpor_timestamp_echo,
188                                       len + 2,
189                                       (unsigned long long)
190                                       DCCP_SKB_CB(skb)->dccpd_ack_seq);
191
192                         value += 4;
193
194                         if (len == 4) {         /* no elapsed time included */
195                                 dccp_pr_debug_cat("\n");
196                                 break;
197                         }
198
199                         if (len == 6) {         /* 2-byte elapsed time */
200                                 __be16 opt_val2 = get_unaligned((__be16 *)value);
201                                 elapsed_time = ntohs(opt_val2);
202                         } else {                /* 4-byte elapsed time */
203                                 opt_val = get_unaligned((__be32 *)value);
204                                 elapsed_time = ntohl(opt_val);
205                         }
206
207                         dccp_pr_debug_cat(", ELAPSED_TIME=%u\n", elapsed_time);
208
209                         /* Give precedence to the biggest ELAPSED_TIME */
210                         if (elapsed_time > opt_recv->dccpor_elapsed_time)
211                                 opt_recv->dccpor_elapsed_time = elapsed_time;
212                         break;
213                 case DCCPO_ELAPSED_TIME:
214                         if (dccp_packet_without_ack(skb))   /* RFC 4340, 13.2 */
215                                 break;
216
217                         if (len == 2) {
218                                 __be16 opt_val2 = get_unaligned((__be16 *)value);
219                                 elapsed_time = ntohs(opt_val2);
220                         } else if (len == 4) {
221                                 opt_val = get_unaligned((__be32 *)value);
222                                 elapsed_time = ntohl(opt_val);
223                         } else {
224                                 goto out_invalid_option;
225                         }
226
227                         if (elapsed_time > opt_recv->dccpor_elapsed_time)
228                                 opt_recv->dccpor_elapsed_time = elapsed_time;
229
230                         dccp_pr_debug("%s rx opt: ELAPSED_TIME=%d\n",
231                                       dccp_role(sk), elapsed_time);
232                         break;
233                 case 128 ... 191: {
234                         const u16 idx = value - options;
235
236                         if (ccid_hc_rx_parse_options(dp->dccps_hc_rx_ccid, sk,
237                                                      opt, len, idx,
238                                                      value) != 0)
239                                 goto out_invalid_option;
240                 }
241                         break;
242                 case 192 ... 255: {
243                         const u16 idx = value - options;
244
245                         if (ccid_hc_tx_parse_options(dp->dccps_hc_tx_ccid, sk,
246                                                      opt, len, idx,
247                                                      value) != 0)
248                                 goto out_invalid_option;
249                 }
250                         break;
251                 default:
252                         DCCP_CRIT("DCCP(%p): option %d(len=%d) not "
253                                   "implemented, ignoring", sk, opt, len);
254                         break;
255                 }
256 ignore_option:
257                 if (opt != DCCPO_MANDATORY)
258                         mandatory = 0;
259         }
260
261         /* mandatory was the last byte in option list -> reset connection */
262         if (mandatory)
263                 goto out_invalid_option;
264
265 out_nonsensical_length:
266         /* RFC 4340, 5.8: ignore option and all remaining option space */
267         return 0;
268
269 out_invalid_option:
270         DCCP_INC_STATS_BH(DCCP_MIB_INVALIDOPT);
271         rc = DCCP_RESET_CODE_OPTION_ERROR;
272 out_featneg_failed:
273         DCCP_WARN("DCCP(%p): Option %d (len=%d) error=%u\n", sk, opt, len, rc);
274         DCCP_SKB_CB(skb)->dccpd_reset_code = rc;
275         DCCP_SKB_CB(skb)->dccpd_reset_data[0] = opt;
276         DCCP_SKB_CB(skb)->dccpd_reset_data[1] = len > 0 ? value[0] : 0;
277         DCCP_SKB_CB(skb)->dccpd_reset_data[2] = len > 1 ? value[1] : 0;
278         return -1;
279 }
280
281 EXPORT_SYMBOL_GPL(dccp_parse_options);
282
283 void dccp_encode_value_var(const u64 value, u8 *to, const u8 len)
284 {
285         if (len >= DCCP_OPTVAL_MAXLEN)
286                 *to++ = (value & 0xFF0000000000ull) >> 40;
287         if (len > 4)
288                 *to++ = (value & 0xFF00000000ull) >> 32;
289         if (len > 3)
290                 *to++ = (value & 0xFF000000) >> 24;
291         if (len > 2)
292                 *to++ = (value & 0xFF0000) >> 16;
293         if (len > 1)
294                 *to++ = (value & 0xFF00) >> 8;
295         if (len > 0)
296                 *to++ = (value & 0xFF);
297 }
298
299 static inline u8 dccp_ndp_len(const u64 ndp)
300 {
301         if (likely(ndp <= 0xFF))
302                 return 1;
303         return likely(ndp <= USHORT_MAX) ? 2 : (ndp <= UINT_MAX ? 4 : 6);
304 }
305
306 int dccp_insert_option(struct sock *sk, struct sk_buff *skb,
307                         const unsigned char option,
308                         const void *value, const unsigned char len)
309 {
310         unsigned char *to;
311
312         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len + 2 > DCCP_MAX_OPT_LEN)
313                 return -1;
314
315         DCCP_SKB_CB(skb)->dccpd_opt_len += len + 2;
316
317         to    = skb_push(skb, len + 2);
318         *to++ = option;
319         *to++ = len + 2;
320
321         memcpy(to, value, len);
322         return 0;
323 }
324
325 EXPORT_SYMBOL_GPL(dccp_insert_option);
326
327 static int dccp_insert_option_ndp(struct sock *sk, struct sk_buff *skb)
328 {
329         struct dccp_sock *dp = dccp_sk(sk);
330         u64 ndp = dp->dccps_ndp_count;
331
332         if (dccp_non_data_packet(skb))
333                 ++dp->dccps_ndp_count;
334         else
335                 dp->dccps_ndp_count = 0;
336
337         if (ndp > 0) {
338                 unsigned char *ptr;
339                 const int ndp_len = dccp_ndp_len(ndp);
340                 const int len = ndp_len + 2;
341
342                 if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
343                         return -1;
344
345                 DCCP_SKB_CB(skb)->dccpd_opt_len += len;
346
347                 ptr = skb_push(skb, len);
348                 *ptr++ = DCCPO_NDP_COUNT;
349                 *ptr++ = len;
350                 dccp_encode_value_var(ndp, ptr, ndp_len);
351         }
352
353         return 0;
354 }
355
356 static inline int dccp_elapsed_time_len(const u32 elapsed_time)
357 {
358         return elapsed_time == 0 ? 0 : elapsed_time <= 0xFFFF ? 2 : 4;
359 }
360
361 int dccp_insert_option_elapsed_time(struct sock *sk, struct sk_buff *skb,
362                                     u32 elapsed_time)
363 {
364         const int elapsed_time_len = dccp_elapsed_time_len(elapsed_time);
365         const int len = 2 + elapsed_time_len;
366         unsigned char *to;
367
368         if (elapsed_time_len == 0)
369                 return 0;
370
371         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
372                 return -1;
373
374         DCCP_SKB_CB(skb)->dccpd_opt_len += len;
375
376         to    = skb_push(skb, len);
377         *to++ = DCCPO_ELAPSED_TIME;
378         *to++ = len;
379
380         if (elapsed_time_len == 2) {
381                 const __be16 var16 = htons((u16)elapsed_time);
382                 memcpy(to, &var16, 2);
383         } else {
384                 const __be32 var32 = htonl(elapsed_time);
385                 memcpy(to, &var32, 4);
386         }
387
388         return 0;
389 }
390
391 EXPORT_SYMBOL_GPL(dccp_insert_option_elapsed_time);
392
393 int dccp_insert_option_timestamp(struct sock *sk, struct sk_buff *skb)
394 {
395         __be32 now = htonl(dccp_timestamp());
396         /* yes this will overflow but that is the point as we want a
397          * 10 usec 32 bit timer which mean it wraps every 11.9 hours */
398
399         return dccp_insert_option(sk, skb, DCCPO_TIMESTAMP, &now, sizeof(now));
400 }
401
402 EXPORT_SYMBOL_GPL(dccp_insert_option_timestamp);
403
404 static int dccp_insert_option_timestamp_echo(struct dccp_sock *dp,
405                                              struct dccp_request_sock *dreq,
406                                              struct sk_buff *skb)
407 {
408         __be32 tstamp_echo;
409         unsigned char *to;
410         u32 elapsed_time, elapsed_time_len, len;
411
412         if (dreq != NULL) {
413                 elapsed_time = dccp_timestamp() - dreq->dreq_timestamp_time;
414                 tstamp_echo  = htonl(dreq->dreq_timestamp_echo);
415                 dreq->dreq_timestamp_echo = 0;
416         } else {
417                 elapsed_time = dccp_timestamp() - dp->dccps_timestamp_time;
418                 tstamp_echo  = htonl(dp->dccps_timestamp_echo);
419                 dp->dccps_timestamp_echo = 0;
420         }
421
422         elapsed_time_len = dccp_elapsed_time_len(elapsed_time);
423         len = 6 + elapsed_time_len;
424
425         if (DCCP_SKB_CB(skb)->dccpd_opt_len + len > DCCP_MAX_OPT_LEN)
426                 return -1;
427
428         DCCP_SKB_CB(skb)->dccpd_opt_len += len;
429
430         to    = skb_push(skb, len);
431         *to++ = DCCPO_TIMESTAMP_ECHO;
432         *to++ = len;
433
434         memcpy(to, &tstamp_echo, 4);
435         to += 4;
436
437         if (elapsed_time_len == 2) {
438                 const __be16 var16 = htons((u16)elapsed_time);
439                 memcpy(to, &var16, 2);
440         } else if (elapsed_time_len == 4) {
441                 const __be32 var32 = htonl(elapsed_time);
442                 memcpy(to, &var32, 4);
443         }
444
445         return 0;
446 }
447
448 /**
449  * dccp_insert_option_mandatory  -  Mandatory option (5.8.2)
450  * Note that since we are using skb_push, this function needs to be called
451  * _after_ inserting the option it is supposed to influence (stack order).
452  */
453 int dccp_insert_option_mandatory(struct sk_buff *skb)
454 {
455         if (DCCP_SKB_CB(skb)->dccpd_opt_len >= DCCP_MAX_OPT_LEN)
456                 return -1;
457
458         DCCP_SKB_CB(skb)->dccpd_opt_len++;
459         *skb_push(skb, 1) = DCCPO_MANDATORY;
460         return 0;
461 }
462
463 /**
464  * dccp_insert_fn_opt  -  Insert single Feature-Negotiation option into @skb
465  * @type: %DCCPO_CHANGE_L, %DCCPO_CHANGE_R, %DCCPO_CONFIRM_L, %DCCPO_CONFIRM_R
466  * @feat: one out of %dccp_feature_numbers
467  * @val: NN value or SP array (preferred element first) to copy
468  * @len: true length of @val in bytes (excluding first element repetition)
469  * @repeat_first: whether to copy the first element of @val twice
470  * The last argument is used to construct Confirm options, where the preferred
471  * value and the preference list appear separately (RFC 4340, 6.3.1). Preference
472  * lists are kept such that the preferred entry is always first, so we only need
473  * to copy twice, and avoid the overhead of cloning into a bigger array.
474  */
475 int dccp_insert_fn_opt(struct sk_buff *skb, u8 type, u8 feat,
476                        u8 *val, u8 len, bool repeat_first)
477 {
478         u8 tot_len, *to;
479
480         /* take the `Feature' field and possible repetition into account */
481         if (len > (DCCP_SINGLE_OPT_MAXLEN - 2)) {
482                 DCCP_WARN("length %u for feature %u too large\n", len, feat);
483                 return -1;
484         }
485
486         if (unlikely(val == NULL || len == 0))
487                 len = repeat_first = 0;
488         tot_len = 3 + repeat_first + len;
489
490         if (DCCP_SKB_CB(skb)->dccpd_opt_len + tot_len > DCCP_MAX_OPT_LEN) {
491                 DCCP_WARN("packet too small for feature %d option!\n", feat);
492                 return -1;
493         }
494         DCCP_SKB_CB(skb)->dccpd_opt_len += tot_len;
495
496         to    = skb_push(skb, tot_len);
497         *to++ = type;
498         *to++ = tot_len;
499         *to++ = feat;
500
501         if (repeat_first)
502                 *to++ = *val;
503         if (len)
504                 memcpy(to, val, len);
505
506         dccp_pr_debug("%s(%s (%d), ...), length %d\n",
507                       dccp_feat_typename(type),
508                       dccp_feat_name(feat), feat, len);
509         return 0;
510 }
511
512 /* The length of all options needs to be a multiple of 4 (5.8) */
513 static void dccp_insert_option_padding(struct sk_buff *skb)
514 {
515         int padding = DCCP_SKB_CB(skb)->dccpd_opt_len % 4;
516
517         if (padding != 0) {
518                 padding = 4 - padding;
519                 memset(skb_push(skb, padding), 0, padding);
520                 DCCP_SKB_CB(skb)->dccpd_opt_len += padding;
521         }
522 }
523
524 int dccp_insert_options(struct sock *sk, struct sk_buff *skb)
525 {
526         struct dccp_sock *dp = dccp_sk(sk);
527
528         DCCP_SKB_CB(skb)->dccpd_opt_len = 0;
529
530         if (dp->dccps_send_ndp_count && dccp_insert_option_ndp(sk, skb))
531                 return -1;
532
533         if (DCCP_SKB_CB(skb)->dccpd_type != DCCP_PKT_DATA) {
534
535                 /* Feature Negotiation */
536                 if (dccp_feat_insert_opts(dp, NULL, skb))
537                         return -1;
538
539                 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_REQUEST) {
540                         /*
541                          * Obtain RTT sample from Request/Response exchange.
542                          * This is currently used in CCID 3 initialisation.
543                          */
544                         if (dccp_insert_option_timestamp(sk, skb))
545                                 return -1;
546
547                 } else if (dp->dccps_hc_rx_ackvec != NULL &&
548                            dccp_ackvec_pending(dp->dccps_hc_rx_ackvec) &&
549                            dccp_insert_option_ackvec(sk, skb)) {
550                                 return -1;
551                 }
552         }
553
554         if (dp->dccps_hc_rx_insert_options) {
555                 if (ccid_hc_rx_insert_options(dp->dccps_hc_rx_ccid, sk, skb))
556                         return -1;
557                 dp->dccps_hc_rx_insert_options = 0;
558         }
559
560         if (dp->dccps_timestamp_echo != 0 &&
561             dccp_insert_option_timestamp_echo(dp, NULL, skb))
562                 return -1;
563
564         dccp_insert_option_padding(skb);
565         return 0;
566 }
567
568 int dccp_insert_options_rsk(struct dccp_request_sock *dreq, struct sk_buff *skb)
569 {
570         DCCP_SKB_CB(skb)->dccpd_opt_len = 0;
571
572         if (dccp_feat_insert_opts(NULL, dreq, skb))
573                 return -1;
574
575         if (dreq->dreq_timestamp_echo != 0 &&
576             dccp_insert_option_timestamp_echo(NULL, dreq, skb))
577                 return -1;
578
579         dccp_insert_option_padding(skb);
580         return 0;
581 }