eb1bda08eeb2e734bbc31c59120c9f7aca90e7eb
[deliverable/linux.git] / net / dccp / ccids / ccid3.c
1 /*
2 * net/dccp/ccids/ccid3.c
3 *
4 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
5 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
6 * Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
7 *
8 * An implementation of the DCCP protocol
9 *
10 * This code has been developed by the University of Waikato WAND
11 * research group. For further information please see http://www.wand.net.nz/
12 *
13 * This code also uses code from Lulea University, rereleased as GPL by its
14 * authors:
15 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
16 *
17 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
18 * and to make it work as a loadable module in the DCCP stack written by
19 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
20 *
21 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
22 *
23 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 *
33 * You should have received a copy of the GNU General Public License
34 * along with this program; if not, write to the Free Software
35 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
36 */
37 #include "../dccp.h"
38 #include "ccid3.h"
39
40 #include <asm/unaligned.h>
41
42 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
43 static int ccid3_debug;
44 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
45 #else
46 #define ccid3_pr_debug(format, a...)
47 #endif
48
49 /*
50 * Transmitter Half-Connection Routines
51 */
52 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
53 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
54 {
55 static char *ccid3_state_names[] = {
56 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
57 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
58 [TFRC_SSTATE_FBACK] = "FBACK",
59 [TFRC_SSTATE_TERM] = "TERM",
60 };
61
62 return ccid3_state_names[state];
63 }
64 #endif
65
66 static void ccid3_hc_tx_set_state(struct sock *sk,
67 enum ccid3_hc_tx_states state)
68 {
69 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
70 enum ccid3_hc_tx_states oldstate = hctx->state;
71
72 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
73 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
74 ccid3_tx_state_name(state));
75 WARN_ON(state == oldstate);
76 hctx->state = state;
77 }
78
79 /*
80 * Compute the initial sending rate X_init in the manner of RFC 3390:
81 *
82 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
83 *
84 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
85 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
86 * For consistency with other parts of the code, X_init is scaled by 2^6.
87 */
88 static inline u64 rfc3390_initial_rate(struct sock *sk)
89 {
90 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
91 const __u32 w_init = clamp_t(__u32, 4380U, 2 * hctx->s, 4 * hctx->s);
92
93 return scaled_div(w_init << 6, hctx->rtt);
94 }
95
96 /**
97 * ccid3_update_send_interval - Calculate new t_ipi = s / X_inst
98 * This respects the granularity of X_inst (64 * bytes/second).
99 */
100 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
101 {
102 hctx->t_ipi = scaled_div32(((u64)hctx->s) << 6, hctx->x);
103
104 ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hctx->t_ipi,
105 hctx->s, (unsigned)(hctx->x >> 6));
106 }
107
108 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
109 {
110 u32 delta = ktime_us_delta(now, hctx->t_last_win_count);
111
112 return delta / hctx->rtt;
113 }
114
115 /**
116 * ccid3_hc_tx_update_x - Update allowed sending rate X
117 * @stamp: most recent time if available - can be left NULL.
118 * This function tracks draft rfc3448bis, check there for latest details.
119 *
120 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
121 * fine-grained resolution of sending rates. This requires scaling by 2^6
122 * throughout the code. Only X_calc is unscaled (in bytes/second).
123 *
124 */
125 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
126 {
127 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
128 u64 min_rate = 2 * hctx->x_recv;
129 const u64 old_x = hctx->x;
130 ktime_t now = stamp ? *stamp : ktime_get_real();
131
132 /*
133 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
134 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
135 * a sender is idle if it has not sent anything over a 2-RTT-period.
136 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
137 */
138 if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) {
139 min_rate = rfc3390_initial_rate(sk);
140 min_rate = max(min_rate, 2 * hctx->x_recv);
141 }
142
143 if (hctx->p > 0) {
144
145 hctx->x = min(((u64)hctx->x_calc) << 6, min_rate);
146 hctx->x = max(hctx->x, (((u64)hctx->s) << 6) / TFRC_T_MBI);
147
148 } else if (ktime_us_delta(now, hctx->t_ld) - (s64)hctx->rtt >= 0) {
149
150 hctx->x = min(2 * hctx->x, min_rate);
151 hctx->x = max(hctx->x,
152 scaled_div(((u64)hctx->s) << 6, hctx->rtt));
153 hctx->t_ld = now;
154 }
155
156 if (hctx->x != old_x) {
157 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
158 "X_recv=%u\n", (unsigned)(old_x >> 6),
159 (unsigned)(hctx->x >> 6), hctx->x_calc,
160 (unsigned)(hctx->x_recv >> 6));
161
162 ccid3_update_send_interval(hctx);
163 }
164 }
165
166 /*
167 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
168 * @len: DCCP packet payload size in bytes
169 */
170 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
171 {
172 const u16 old_s = hctx->s;
173
174 hctx->s = tfrc_ewma(hctx->s, len, 9);
175
176 if (hctx->s != old_s)
177 ccid3_update_send_interval(hctx);
178 }
179
180 /*
181 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
182 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
183 */
184 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
185 ktime_t now)
186 {
187 u32 delta = ktime_us_delta(now, hctx->t_last_win_count),
188 quarter_rtts = (4 * delta) / hctx->rtt;
189
190 if (quarter_rtts > 0) {
191 hctx->t_last_win_count = now;
192 hctx->last_win_count += min(quarter_rtts, 5U);
193 hctx->last_win_count &= 0xF; /* mod 16 */
194 }
195 }
196
197 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
198 {
199 struct sock *sk = (struct sock *)data;
200 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
201 unsigned long t_nfb = USEC_PER_SEC / 5;
202
203 bh_lock_sock(sk);
204 if (sock_owned_by_user(sk)) {
205 /* Try again later. */
206 /* XXX: set some sensible MIB */
207 goto restart_timer;
208 }
209
210 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
211 ccid3_tx_state_name(hctx->state));
212
213 if (hctx->state == TFRC_SSTATE_FBACK)
214 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
215 else if (hctx->state != TFRC_SSTATE_NO_FBACK)
216 goto out;
217
218 /*
219 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
220 * RTO is 0 if and only if no feedback has been received yet.
221 */
222 if (hctx->t_rto == 0 || hctx->p == 0) {
223
224 /* halve send rate directly */
225 hctx->x = max(hctx->x / 2, (((u64)hctx->s) << 6) / TFRC_T_MBI);
226 ccid3_update_send_interval(hctx);
227 } else {
228 /*
229 * Modify the cached value of X_recv
230 *
231 * If (X_calc > 2 * X_recv)
232 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
233 * Else
234 * X_recv = X_calc / 4;
235 *
236 * Note that X_recv is scaled by 2^6 while X_calc is not
237 */
238 BUG_ON(hctx->p && !hctx->x_calc);
239
240 if (hctx->x_calc > (hctx->x_recv >> 5))
241 hctx->x_recv =
242 max(hctx->x_recv / 2,
243 (((__u64)hctx->s) << 6) / (2 * TFRC_T_MBI));
244 else {
245 hctx->x_recv = hctx->x_calc;
246 hctx->x_recv <<= 4;
247 }
248 ccid3_hc_tx_update_x(sk, NULL);
249 }
250 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
251 (unsigned long long)hctx->x);
252
253 /*
254 * Set new timeout for the nofeedback timer.
255 * See comments in packet_recv() regarding the value of t_RTO.
256 */
257 if (unlikely(hctx->t_rto == 0)) /* no feedback received yet */
258 t_nfb = TFRC_INITIAL_TIMEOUT;
259 else
260 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
261
262 restart_timer:
263 sk_reset_timer(sk, &hctx->no_feedback_timer,
264 jiffies + usecs_to_jiffies(t_nfb));
265 out:
266 bh_unlock_sock(sk);
267 sock_put(sk);
268 }
269
270 /*
271 * returns
272 * > 0: delay (in msecs) that should pass before actually sending
273 * = 0: can send immediately
274 * < 0: error condition; do not send packet
275 */
276 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
277 {
278 struct dccp_sock *dp = dccp_sk(sk);
279 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
280 ktime_t now = ktime_get_real();
281 s64 delay;
282
283 /*
284 * This function is called only for Data and DataAck packets. Sending
285 * zero-sized Data(Ack)s is theoretically possible, but for congestion
286 * control this case is pathological - ignore it.
287 */
288 if (unlikely(skb->len == 0))
289 return -EBADMSG;
290
291 switch (hctx->state) {
292 case TFRC_SSTATE_NO_SENT:
293 sk_reset_timer(sk, &hctx->no_feedback_timer, (jiffies +
294 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
295 hctx->last_win_count = 0;
296 hctx->t_last_win_count = now;
297
298 /* Set t_0 for initial packet */
299 hctx->t_nom = now;
300
301 hctx->s = skb->len;
302
303 /*
304 * Use initial RTT sample when available: recommended by erratum
305 * to RFC 4342. This implements the initialisation procedure of
306 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
307 */
308 if (dp->dccps_syn_rtt) {
309 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
310 hctx->rtt = dp->dccps_syn_rtt;
311 hctx->x = rfc3390_initial_rate(sk);
312 hctx->t_ld = now;
313 } else {
314 /*
315 * Sender does not have RTT sample:
316 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
317 * is needed in several parts (e.g. window counter);
318 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
319 */
320 hctx->rtt = DCCP_FALLBACK_RTT;
321 hctx->x = hctx->s;
322 hctx->x <<= 6;
323 }
324 ccid3_update_send_interval(hctx);
325
326 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
327 break;
328 case TFRC_SSTATE_NO_FBACK:
329 case TFRC_SSTATE_FBACK:
330 delay = ktime_us_delta(hctx->t_nom, now);
331 ccid3_pr_debug("delay=%ld\n", (long)delay);
332 /*
333 * Scheduling of packet transmissions [RFC 3448, 4.6]
334 *
335 * if (t_now > t_nom - delta)
336 * // send the packet now
337 * else
338 * // send the packet in (t_nom - t_now) milliseconds.
339 */
340 if (delay >= TFRC_T_DELTA)
341 return (u32)delay / USEC_PER_MSEC;
342
343 ccid3_hc_tx_update_win_count(hctx, now);
344 break;
345 case TFRC_SSTATE_TERM:
346 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
347 return -EINVAL;
348 }
349
350 /* prepare to send now (add options etc.) */
351 dp->dccps_hc_tx_insert_options = 1;
352 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->last_win_count;
353
354 /* set the nominal send time for the next following packet */
355 hctx->t_nom = ktime_add_us(hctx->t_nom, hctx->t_ipi);
356 return 0;
357 }
358
359 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
360 unsigned int len)
361 {
362 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
363
364 ccid3_hc_tx_update_s(hctx, len);
365
366 if (tfrc_tx_hist_add(&hctx->hist, dccp_sk(sk)->dccps_gss))
367 DCCP_CRIT("packet history - out of memory!");
368 }
369
370 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
371 {
372 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
373 struct ccid3_options_received *opt_recv = &hctx->options_received;
374 ktime_t now;
375 unsigned long t_nfb;
376 u32 pinv, r_sample;
377
378 /* we are only interested in ACKs */
379 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
380 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
381 return;
382 /* ... and only in the established state */
383 if (hctx->state != TFRC_SSTATE_FBACK &&
384 hctx->state != TFRC_SSTATE_NO_FBACK)
385 return;
386
387 now = ktime_get_real();
388
389 /* Estimate RTT from history if ACK number is valid */
390 r_sample = tfrc_tx_hist_rtt(hctx->hist,
391 DCCP_SKB_CB(skb)->dccpd_ack_seq, now);
392 if (r_sample == 0) {
393 DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk), sk,
394 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type),
395 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq);
396 return;
397 }
398
399 /* Update receive rate in units of 64 * bytes/second */
400 hctx->x_recv = opt_recv->ccid3or_receive_rate;
401 hctx->x_recv <<= 6;
402
403 /* Update loss event rate (which is scaled by 1e6) */
404 pinv = opt_recv->ccid3or_loss_event_rate;
405 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
406 hctx->p = 0;
407 else /* can not exceed 100% */
408 hctx->p = scaled_div(1, pinv);
409 /*
410 * Validate new RTT sample and update moving average
411 */
412 r_sample = dccp_sample_rtt(sk, r_sample);
413 hctx->rtt = tfrc_ewma(hctx->rtt, r_sample, 9);
414 /*
415 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
416 */
417 if (hctx->state == TFRC_SSTATE_NO_FBACK) {
418 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
419
420 if (hctx->t_rto == 0) {
421 /*
422 * Initial feedback packet: Larger Initial Windows (4.2)
423 */
424 hctx->x = rfc3390_initial_rate(sk);
425 hctx->t_ld = now;
426
427 ccid3_update_send_interval(hctx);
428
429 goto done_computing_x;
430 } else if (hctx->p == 0) {
431 /*
432 * First feedback after nofeedback timer expiry (4.3)
433 */
434 goto done_computing_x;
435 }
436 }
437
438 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
439 if (hctx->p > 0)
440 hctx->x_calc = tfrc_calc_x(hctx->s, hctx->rtt, hctx->p);
441 ccid3_hc_tx_update_x(sk, &now);
442
443 done_computing_x:
444 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
445 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
446 dccp_role(sk), sk, hctx->rtt, r_sample,
447 hctx->s, hctx->p, hctx->x_calc,
448 (unsigned)(hctx->x_recv >> 6),
449 (unsigned)(hctx->x >> 6));
450
451 /* unschedule no feedback timer */
452 sk_stop_timer(sk, &hctx->no_feedback_timer);
453
454 /*
455 * As we have calculated new ipi, delta, t_nom it is possible
456 * that we now can send a packet, so wake up dccp_wait_for_ccid
457 */
458 sk->sk_write_space(sk);
459
460 /*
461 * Update timeout interval for the nofeedback timer.
462 * We use a configuration option to increase the lower bound.
463 * This can help avoid triggering the nofeedback timer too
464 * often ('spinning') on LANs with small RTTs.
465 */
466 hctx->t_rto = max_t(u32, 4 * hctx->rtt, (CONFIG_IP_DCCP_CCID3_RTO *
467 (USEC_PER_SEC / 1000)));
468 /*
469 * Schedule no feedback timer to expire in
470 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
471 */
472 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
473
474 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
475 "expire in %lu jiffies (%luus)\n",
476 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
477
478 sk_reset_timer(sk, &hctx->no_feedback_timer,
479 jiffies + usecs_to_jiffies(t_nfb));
480 }
481
482 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
483 unsigned char len, u16 idx,
484 unsigned char *value)
485 {
486 int rc = 0;
487 const struct dccp_sock *dp = dccp_sk(sk);
488 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
489 struct ccid3_options_received *opt_recv = &hctx->options_received;
490 __be32 opt_val;
491
492 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
493 opt_recv->ccid3or_seqno = dp->dccps_gsr;
494 opt_recv->ccid3or_loss_event_rate = ~0;
495 opt_recv->ccid3or_loss_intervals_idx = 0;
496 opt_recv->ccid3or_loss_intervals_len = 0;
497 opt_recv->ccid3or_receive_rate = 0;
498 }
499
500 switch (option) {
501 case TFRC_OPT_LOSS_EVENT_RATE:
502 if (unlikely(len != 4)) {
503 DCCP_WARN("%s(%p), invalid len %d "
504 "for TFRC_OPT_LOSS_EVENT_RATE\n",
505 dccp_role(sk), sk, len);
506 rc = -EINVAL;
507 } else {
508 opt_val = get_unaligned((__be32 *)value);
509 opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
510 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
511 dccp_role(sk), sk,
512 opt_recv->ccid3or_loss_event_rate);
513 }
514 break;
515 case TFRC_OPT_LOSS_INTERVALS:
516 opt_recv->ccid3or_loss_intervals_idx = idx;
517 opt_recv->ccid3or_loss_intervals_len = len;
518 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
519 dccp_role(sk), sk,
520 opt_recv->ccid3or_loss_intervals_idx,
521 opt_recv->ccid3or_loss_intervals_len);
522 break;
523 case TFRC_OPT_RECEIVE_RATE:
524 if (unlikely(len != 4)) {
525 DCCP_WARN("%s(%p), invalid len %d "
526 "for TFRC_OPT_RECEIVE_RATE\n",
527 dccp_role(sk), sk, len);
528 rc = -EINVAL;
529 } else {
530 opt_val = get_unaligned((__be32 *)value);
531 opt_recv->ccid3or_receive_rate = ntohl(opt_val);
532 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
533 dccp_role(sk), sk,
534 opt_recv->ccid3or_receive_rate);
535 }
536 break;
537 }
538
539 return rc;
540 }
541
542 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
543 {
544 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
545
546 hctx->state = TFRC_SSTATE_NO_SENT;
547 hctx->hist = NULL;
548 setup_timer(&hctx->no_feedback_timer,
549 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
550 return 0;
551 }
552
553 static void ccid3_hc_tx_exit(struct sock *sk)
554 {
555 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
556
557 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
558 sk_stop_timer(sk, &hctx->no_feedback_timer);
559
560 tfrc_tx_hist_purge(&hctx->hist);
561 }
562
563 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
564 {
565 info->tcpi_rto = ccid3_hc_tx_sk(sk)->t_rto;
566 info->tcpi_rtt = ccid3_hc_tx_sk(sk)->rtt;
567 }
568
569 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
570 u32 __user *optval, int __user *optlen)
571 {
572 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
573 struct tfrc_tx_info tfrc;
574 const void *val;
575
576 switch (optname) {
577 case DCCP_SOCKOPT_CCID_TX_INFO:
578 if (len < sizeof(tfrc))
579 return -EINVAL;
580 tfrc.tfrctx_x = hctx->x;
581 tfrc.tfrctx_x_recv = hctx->x_recv;
582 tfrc.tfrctx_x_calc = hctx->x_calc;
583 tfrc.tfrctx_rtt = hctx->rtt;
584 tfrc.tfrctx_p = hctx->p;
585 tfrc.tfrctx_rto = hctx->t_rto;
586 tfrc.tfrctx_ipi = hctx->t_ipi;
587 len = sizeof(tfrc);
588 val = &tfrc;
589 break;
590 default:
591 return -ENOPROTOOPT;
592 }
593
594 if (put_user(len, optlen) || copy_to_user(optval, val, len))
595 return -EFAULT;
596
597 return 0;
598 }
599
600 /*
601 * Receiver Half-Connection Routines
602 */
603
604 /* CCID3 feedback types */
605 enum ccid3_fback_type {
606 CCID3_FBACK_NONE = 0,
607 CCID3_FBACK_INITIAL,
608 CCID3_FBACK_PERIODIC,
609 CCID3_FBACK_PARAM_CHANGE
610 };
611
612 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
613 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
614 {
615 static char *ccid3_rx_state_names[] = {
616 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
617 [TFRC_RSTATE_DATA] = "DATA",
618 [TFRC_RSTATE_TERM] = "TERM",
619 };
620
621 return ccid3_rx_state_names[state];
622 }
623 #endif
624
625 static void ccid3_hc_rx_set_state(struct sock *sk,
626 enum ccid3_hc_rx_states state)
627 {
628 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
629 enum ccid3_hc_rx_states oldstate = hcrx->state;
630
631 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
632 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
633 ccid3_rx_state_name(state));
634 WARN_ON(state == oldstate);
635 hcrx->state = state;
636 }
637
638 static void ccid3_hc_rx_send_feedback(struct sock *sk,
639 const struct sk_buff *skb,
640 enum ccid3_fback_type fbtype)
641 {
642 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
643 struct dccp_sock *dp = dccp_sk(sk);
644 ktime_t now;
645 s64 delta = 0;
646
647 if (unlikely(hcrx->state == TFRC_RSTATE_TERM))
648 return;
649
650 now = ktime_get_real();
651
652 switch (fbtype) {
653 case CCID3_FBACK_INITIAL:
654 hcrx->x_recv = 0;
655 hcrx->p_inverse = ~0U; /* see RFC 4342, 8.5 */
656 break;
657 case CCID3_FBACK_PARAM_CHANGE:
658 /*
659 * When parameters change (new loss or p > p_prev), we do not
660 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
661 * need to reuse the previous value of X_recv. However, when
662 * X_recv was 0 (due to early loss), this would kill X down to
663 * s/t_mbi (i.e. one packet in 64 seconds).
664 * To avoid such drastic reduction, we approximate X_recv as
665 * the number of bytes since last feedback.
666 * This is a safe fallback, since X is bounded above by X_calc.
667 */
668 if (hcrx->x_recv > 0)
669 break;
670 /* fall through */
671 case CCID3_FBACK_PERIODIC:
672 delta = ktime_us_delta(now, hcrx->tstamp_last_feedback);
673 if (delta <= 0)
674 DCCP_BUG("delta (%ld) <= 0", (long)delta);
675 else
676 hcrx->x_recv = scaled_div32(hcrx->bytes_recv, delta);
677 break;
678 default:
679 return;
680 }
681
682 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n",
683 (long)delta, hcrx->x_recv, hcrx->p_inverse);
684
685 hcrx->tstamp_last_feedback = now;
686 hcrx->last_counter = dccp_hdr(skb)->dccph_ccval;
687 hcrx->bytes_recv = 0;
688
689 dp->dccps_hc_rx_insert_options = 1;
690 dccp_send_ack(sk);
691 }
692
693 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
694 {
695 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
696 __be32 x_recv, pinv;
697
698 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
699 return 0;
700
701 if (dccp_packet_without_ack(skb))
702 return 0;
703
704 x_recv = htonl(hcrx->x_recv);
705 pinv = htonl(hcrx->p_inverse);
706
707 if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
708 &pinv, sizeof(pinv)) ||
709 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
710 &x_recv, sizeof(x_recv)))
711 return -1;
712
713 return 0;
714 }
715
716 /** ccid3_first_li - Implements [RFC 3448, 6.3.1]
717 *
718 * Determine the length of the first loss interval via inverse lookup.
719 * Assume that X_recv can be computed by the throughput equation
720 * s
721 * X_recv = --------
722 * R * fval
723 * Find some p such that f(p) = fval; return 1/p (scaled).
724 */
725 static u32 ccid3_first_li(struct sock *sk)
726 {
727 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
728 u32 x_recv, p, delta;
729 u64 fval;
730
731 if (hcrx->rtt == 0) {
732 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
733 hcrx->rtt = DCCP_FALLBACK_RTT;
734 }
735
736 delta = ktime_to_us(net_timedelta(hcrx->tstamp_last_feedback));
737 x_recv = scaled_div32(hcrx->bytes_recv, delta);
738 if (x_recv == 0) { /* would also trigger divide-by-zero */
739 DCCP_WARN("X_recv==0\n");
740 if (hcrx->x_recv == 0) {
741 DCCP_BUG("stored value of X_recv is zero");
742 return ~0U;
743 }
744 x_recv = hcrx->x_recv;
745 }
746
747 fval = scaled_div(hcrx->s, hcrx->rtt);
748 fval = scaled_div32(fval, x_recv);
749 p = tfrc_calc_x_reverse_lookup(fval);
750
751 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
752 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
753
754 return p == 0 ? ~0U : scaled_div(1, p);
755 }
756
757 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
758 {
759 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
760 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
761 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
762 const bool is_data_packet = dccp_data_packet(skb);
763
764 if (unlikely(hcrx->state == TFRC_RSTATE_NO_DATA)) {
765 if (is_data_packet) {
766 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
767 do_feedback = CCID3_FBACK_INITIAL;
768 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
769 hcrx->s = payload;
770 /*
771 * Not necessary to update bytes_recv here,
772 * since X_recv = 0 for the first feedback packet (cf.
773 * RFC 3448, 6.3) -- gerrit
774 */
775 }
776 goto update_records;
777 }
778
779 if (tfrc_rx_hist_duplicate(&hcrx->hist, skb))
780 return; /* done receiving */
781
782 if (is_data_packet) {
783 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
784 /*
785 * Update moving-average of s and the sum of received payload bytes
786 */
787 hcrx->s = tfrc_ewma(hcrx->s, payload, 9);
788 hcrx->bytes_recv += payload;
789 }
790
791 /*
792 * Perform loss detection and handle pending losses
793 */
794 if (tfrc_rx_handle_loss(&hcrx->hist, &hcrx->li_hist,
795 skb, ndp, ccid3_first_li, sk)) {
796 do_feedback = CCID3_FBACK_PARAM_CHANGE;
797 goto done_receiving;
798 }
799
800 if (tfrc_rx_hist_loss_pending(&hcrx->hist))
801 return; /* done receiving */
802
803 /*
804 * Handle data packets: RTT sampling and monitoring p
805 */
806 if (unlikely(!is_data_packet))
807 goto update_records;
808
809 if (!tfrc_lh_is_initialised(&hcrx->li_hist)) {
810 const u32 sample = tfrc_rx_hist_sample_rtt(&hcrx->hist, skb);
811 /*
812 * Empty loss history: no loss so far, hence p stays 0.
813 * Sample RTT values, since an RTT estimate is required for the
814 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
815 */
816 if (sample != 0)
817 hcrx->rtt = tfrc_ewma(hcrx->rtt, sample, 9);
818
819 } else if (tfrc_lh_update_i_mean(&hcrx->li_hist, skb)) {
820 /*
821 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
822 * has decreased (resp. p has increased), send feedback now.
823 */
824 do_feedback = CCID3_FBACK_PARAM_CHANGE;
825 }
826
827 /*
828 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
829 */
830 if (SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->last_counter) > 3)
831 do_feedback = CCID3_FBACK_PERIODIC;
832
833 update_records:
834 tfrc_rx_hist_add_packet(&hcrx->hist, skb, ndp);
835
836 done_receiving:
837 if (do_feedback)
838 ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
839 }
840
841 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
842 {
843 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
844
845 hcrx->state = TFRC_RSTATE_NO_DATA;
846 tfrc_lh_init(&hcrx->li_hist);
847 return tfrc_rx_hist_alloc(&hcrx->hist);
848 }
849
850 static void ccid3_hc_rx_exit(struct sock *sk)
851 {
852 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
853
854 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
855
856 tfrc_rx_hist_purge(&hcrx->hist);
857 tfrc_lh_cleanup(&hcrx->li_hist);
858 }
859
860 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
861 {
862 info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->state;
863 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
864 info->tcpi_rcv_rtt = ccid3_hc_rx_sk(sk)->rtt;
865 }
866
867 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
868 u32 __user *optval, int __user *optlen)
869 {
870 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
871 struct tfrc_rx_info rx_info;
872 const void *val;
873
874 switch (optname) {
875 case DCCP_SOCKOPT_CCID_RX_INFO:
876 if (len < sizeof(rx_info))
877 return -EINVAL;
878 rx_info.tfrcrx_x_recv = hcrx->x_recv;
879 rx_info.tfrcrx_rtt = hcrx->rtt;
880 rx_info.tfrcrx_p = hcrx->p_inverse == 0 ? ~0U :
881 scaled_div(1, hcrx->p_inverse);
882 len = sizeof(rx_info);
883 val = &rx_info;
884 break;
885 default:
886 return -ENOPROTOOPT;
887 }
888
889 if (put_user(len, optlen) || copy_to_user(optval, val, len))
890 return -EFAULT;
891
892 return 0;
893 }
894
895 static struct ccid_operations ccid3 = {
896 .ccid_id = DCCPC_CCID3,
897 .ccid_name = "TCP-Friendly Rate Control",
898 .ccid_owner = THIS_MODULE,
899 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
900 .ccid_hc_tx_init = ccid3_hc_tx_init,
901 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
902 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
903 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
904 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
905 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
906 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
907 .ccid_hc_rx_init = ccid3_hc_rx_init,
908 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
909 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
910 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
911 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
912 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
913 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
914 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
915 };
916
917 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
918 module_param(ccid3_debug, bool, 0644);
919 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
920 #endif
921
922 static __init int ccid3_module_init(void)
923 {
924 return ccid_register(&ccid3);
925 }
926 module_init(ccid3_module_init);
927
928 static __exit void ccid3_module_exit(void)
929 {
930 ccid_unregister(&ccid3);
931 }
932 module_exit(ccid3_module_exit);
933
934 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
935 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
936 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
937 MODULE_LICENSE("GPL");
938 MODULE_ALIAS("net-dccp-ccid-3");
This page took 0.065319 seconds and 4 git commands to generate.