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