mm-only debug patch...
[mmotm.git] / net / dccp / ccids / ccid3.c
blobbcd7632299f5244ab631f60f84115d211b838d68
1 /*
2 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
3 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
4 * Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
6 * An implementation of the DCCP protocol
8 * This code has been developed by the University of Waikato WAND
9 * research group. For further information please see http://www.wand.net.nz/
11 * This code also uses code from Lulea University, rereleased as GPL by its
12 * authors:
13 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
15 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
16 * and to make it work as a loadable module in the DCCP stack written by
17 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
19 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
21 * This program is free software; you can redistribute it and/or modify
22 * it under the terms of the GNU General Public License as published by
23 * the Free Software Foundation; either version 2 of the License, or
24 * (at your option) any later version.
26 * This program is distributed in the hope that it will be useful,
27 * but WITHOUT ANY WARRANTY; without even the implied warranty of
28 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
29 * GNU General Public License for more details.
31 * You should have received a copy of the GNU General Public License
32 * along with this program; if not, write to the Free Software
33 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include "../dccp.h"
36 #include "ccid3.h"
38 #include <asm/unaligned.h>
40 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
41 static int ccid3_debug;
42 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
43 #else
44 #define ccid3_pr_debug(format, a...)
45 #endif
48 * Transmitter Half-Connection Routines
50 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
51 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
53 static const char *const ccid3_state_names[] = {
54 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
55 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
56 [TFRC_SSTATE_FBACK] = "FBACK",
57 [TFRC_SSTATE_TERM] = "TERM",
60 return ccid3_state_names[state];
62 #endif
64 static void ccid3_hc_tx_set_state(struct sock *sk,
65 enum ccid3_hc_tx_states state)
67 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
68 enum ccid3_hc_tx_states oldstate = hc->tx_state;
70 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
71 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
72 ccid3_tx_state_name(state));
73 WARN_ON(state == oldstate);
74 hc->tx_state = state;
78 * Compute the initial sending rate X_init in the manner of RFC 3390:
80 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
82 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
83 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
84 * For consistency with other parts of the code, X_init is scaled by 2^6.
86 static inline u64 rfc3390_initial_rate(struct sock *sk)
88 const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
89 const __u32 w_init = clamp_t(__u32, 4380U, 2 * hc->tx_s, 4 * hc->tx_s);
91 return scaled_div(w_init << 6, hc->tx_rtt);
95 * Recalculate t_ipi and delta (should be called whenever X changes)
97 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hc)
99 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
100 hc->tx_t_ipi = scaled_div32(((u64)hc->tx_s) << 6, hc->tx_x);
102 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
103 hc->tx_delta = min_t(u32, hc->tx_t_ipi / 2, TFRC_OPSYS_HALF_TIME_GRAN);
105 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n", hc->tx_t_ipi,
106 hc->tx_delta, hc->tx_s, (unsigned)(hc->tx_x >> 6));
109 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hc, ktime_t now)
111 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count);
113 return delta / hc->tx_rtt;
117 * ccid3_hc_tx_update_x - Update allowed sending rate X
118 * @stamp: most recent time if available - can be left NULL.
119 * This function tracks draft rfc3448bis, check there for latest details.
121 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
122 * fine-grained resolution of sending rates. This requires scaling by 2^6
123 * throughout the code. Only X_calc is unscaled (in bytes/second).
126 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
128 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
129 __u64 min_rate = 2 * hc->tx_x_recv;
130 const __u64 old_x = hc->tx_x;
131 ktime_t now = stamp ? *stamp : ktime_get_real();
134 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
135 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
136 * a sender is idle if it has not sent anything over a 2-RTT-period.
137 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
139 if (ccid3_hc_tx_idle_rtt(hc, now) >= 2) {
140 min_rate = rfc3390_initial_rate(sk);
141 min_rate = max(min_rate, 2 * hc->tx_x_recv);
144 if (hc->tx_p > 0) {
146 hc->tx_x = min(((__u64)hc->tx_x_calc) << 6, min_rate);
147 hc->tx_x = max(hc->tx_x, (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
149 } else if (ktime_us_delta(now, hc->tx_t_ld) - (s64)hc->tx_rtt >= 0) {
151 hc->tx_x = min(2 * hc->tx_x, min_rate);
152 hc->tx_x = max(hc->tx_x,
153 scaled_div(((__u64)hc->tx_s) << 6, hc->tx_rtt));
154 hc->tx_t_ld = now;
157 if (hc->tx_x != old_x) {
158 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
159 "X_recv=%u\n", (unsigned)(old_x >> 6),
160 (unsigned)(hc->tx_x >> 6), hc->tx_x_calc,
161 (unsigned)(hc->tx_x_recv >> 6));
163 ccid3_update_send_interval(hc);
168 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
169 * @len: DCCP packet payload size in bytes
171 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hc, int len)
173 const u16 old_s = hc->tx_s;
175 hc->tx_s = tfrc_ewma(hc->tx_s, len, 9);
177 if (hc->tx_s != old_s)
178 ccid3_update_send_interval(hc);
182 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
183 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
185 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hc,
186 ktime_t now)
188 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count),
189 quarter_rtts = (4 * delta) / hc->tx_rtt;
191 if (quarter_rtts > 0) {
192 hc->tx_t_last_win_count = now;
193 hc->tx_last_win_count += min(quarter_rtts, 5U);
194 hc->tx_last_win_count &= 0xF; /* mod 16 */
198 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
200 struct sock *sk = (struct sock *)data;
201 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
202 unsigned long t_nfb = USEC_PER_SEC / 5;
204 bh_lock_sock(sk);
205 if (sock_owned_by_user(sk)) {
206 /* Try again later. */
207 /* XXX: set some sensible MIB */
208 goto restart_timer;
211 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
212 ccid3_tx_state_name(hc->tx_state));
214 if (hc->tx_state == TFRC_SSTATE_FBACK)
215 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
216 else if (hc->tx_state != TFRC_SSTATE_NO_FBACK)
217 goto out;
220 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
222 if (hc->tx_t_rto == 0 || /* no feedback received yet */
223 hc->tx_p == 0) {
225 /* halve send rate directly */
226 hc->tx_x = max(hc->tx_x / 2,
227 (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
228 ccid3_update_send_interval(hc);
229 } else {
231 * Modify the cached value of X_recv
233 * If (X_calc > 2 * X_recv)
234 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
235 * Else
236 * X_recv = X_calc / 4;
238 * Note that X_recv is scaled by 2^6 while X_calc is not
240 BUG_ON(hc->tx_p && !hc->tx_x_calc);
242 if (hc->tx_x_calc > (hc->tx_x_recv >> 5))
243 hc->tx_x_recv =
244 max(hc->tx_x_recv / 2,
245 (((__u64)hc->tx_s) << 6) / (2*TFRC_T_MBI));
246 else {
247 hc->tx_x_recv = hc->tx_x_calc;
248 hc->tx_x_recv <<= 4;
250 ccid3_hc_tx_update_x(sk, NULL);
252 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
253 (unsigned long long)hc->tx_x);
256 * Set new timeout for the nofeedback timer.
257 * See comments in packet_recv() regarding the value of t_RTO.
259 if (unlikely(hc->tx_t_rto == 0)) /* no feedback yet */
260 t_nfb = TFRC_INITIAL_TIMEOUT;
261 else
262 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
264 restart_timer:
265 sk_reset_timer(sk, &hc->tx_no_feedback_timer,
266 jiffies + usecs_to_jiffies(t_nfb));
267 out:
268 bh_unlock_sock(sk);
269 sock_put(sk);
273 * returns
274 * > 0: delay (in msecs) that should pass before actually sending
275 * = 0: can send immediately
276 * < 0: error condition; do not send packet
278 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
280 struct dccp_sock *dp = dccp_sk(sk);
281 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
282 ktime_t now = ktime_get_real();
283 s64 delay;
286 * This function is called only for Data and DataAck packets. Sending
287 * zero-sized Data(Ack)s is theoretically possible, but for congestion
288 * control this case is pathological - ignore it.
290 if (unlikely(skb->len == 0))
291 return -EBADMSG;
293 switch (hc->tx_state) {
294 case TFRC_SSTATE_NO_SENT:
295 sk_reset_timer(sk, &hc->tx_no_feedback_timer, (jiffies +
296 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
297 hc->tx_last_win_count = 0;
298 hc->tx_t_last_win_count = now;
300 /* Set t_0 for initial packet */
301 hc->tx_t_nom = now;
303 hc->tx_s = skb->len;
306 * Use initial RTT sample when available: recommended by erratum
307 * to RFC 4342. This implements the initialisation procedure of
308 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
310 if (dp->dccps_syn_rtt) {
311 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
312 hc->tx_rtt = dp->dccps_syn_rtt;
313 hc->tx_x = rfc3390_initial_rate(sk);
314 hc->tx_t_ld = now;
315 } else {
317 * Sender does not have RTT sample:
318 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
319 * is needed in several parts (e.g. window counter);
320 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
322 hc->tx_rtt = DCCP_FALLBACK_RTT;
323 hc->tx_x = hc->tx_s;
324 hc->tx_x <<= 6;
326 ccid3_update_send_interval(hc);
328 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
329 break;
330 case TFRC_SSTATE_NO_FBACK:
331 case TFRC_SSTATE_FBACK:
332 delay = ktime_us_delta(hc->tx_t_nom, now);
333 ccid3_pr_debug("delay=%ld\n", (long)delay);
335 * Scheduling of packet transmissions [RFC 3448, 4.6]
337 * if (t_now > t_nom - delta)
338 * // send the packet now
339 * else
340 * // send the packet in (t_nom - t_now) milliseconds.
342 if (delay - (s64)hc->tx_delta >= 1000)
343 return (u32)delay / 1000L;
345 ccid3_hc_tx_update_win_count(hc, now);
346 break;
347 case TFRC_SSTATE_TERM:
348 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
349 return -EINVAL;
352 /* prepare to send now (add options etc.) */
353 dp->dccps_hc_tx_insert_options = 1;
354 DCCP_SKB_CB(skb)->dccpd_ccval = hc->tx_last_win_count;
356 /* set the nominal send time for the next following packet */
357 hc->tx_t_nom = ktime_add_us(hc->tx_t_nom, hc->tx_t_ipi);
358 return 0;
361 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
362 unsigned int len)
364 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
366 ccid3_hc_tx_update_s(hc, len);
368 if (tfrc_tx_hist_add(&hc->tx_hist, dccp_sk(sk)->dccps_gss))
369 DCCP_CRIT("packet history - out of memory!");
372 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
374 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
375 struct ccid3_options_received *opt_recv;
376 ktime_t now;
377 unsigned long t_nfb;
378 u32 pinv, r_sample;
380 /* we are only interested in ACKs */
381 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
382 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
383 return;
384 /* ... and only in the established state */
385 if (hc->tx_state != TFRC_SSTATE_FBACK &&
386 hc->tx_state != TFRC_SSTATE_NO_FBACK)
387 return;
389 opt_recv = &hc->tx_options_received;
390 now = ktime_get_real();
392 /* Estimate RTT from history if ACK number is valid */
393 r_sample = tfrc_tx_hist_rtt(hc->tx_hist,
394 DCCP_SKB_CB(skb)->dccpd_ack_seq, now);
395 if (r_sample == 0) {
396 DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk), sk,
397 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type),
398 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq);
399 return;
402 /* Update receive rate in units of 64 * bytes/second */
403 hc->tx_x_recv = opt_recv->ccid3or_receive_rate;
404 hc->tx_x_recv <<= 6;
406 /* Update loss event rate (which is scaled by 1e6) */
407 pinv = opt_recv->ccid3or_loss_event_rate;
408 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
409 hc->tx_p = 0;
410 else /* can not exceed 100% */
411 hc->tx_p = scaled_div(1, pinv);
413 * Validate new RTT sample and update moving average
415 r_sample = dccp_sample_rtt(sk, r_sample);
416 hc->tx_rtt = tfrc_ewma(hc->tx_rtt, r_sample, 9);
418 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
420 if (hc->tx_state == TFRC_SSTATE_NO_FBACK) {
421 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
423 if (hc->tx_t_rto == 0) {
425 * Initial feedback packet: Larger Initial Windows (4.2)
427 hc->tx_x = rfc3390_initial_rate(sk);
428 hc->tx_t_ld = now;
430 ccid3_update_send_interval(hc);
432 goto done_computing_x;
433 } else if (hc->tx_p == 0) {
435 * First feedback after nofeedback timer expiry (4.3)
437 goto done_computing_x;
441 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
442 if (hc->tx_p > 0)
443 hc->tx_x_calc = tfrc_calc_x(hc->tx_s, hc->tx_rtt, hc->tx_p);
444 ccid3_hc_tx_update_x(sk, &now);
446 done_computing_x:
447 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
448 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
449 dccp_role(sk), sk, hc->tx_rtt, r_sample,
450 hc->tx_s, hc->tx_p, hc->tx_x_calc,
451 (unsigned)(hc->tx_x_recv >> 6),
452 (unsigned)(hc->tx_x >> 6));
454 /* unschedule no feedback timer */
455 sk_stop_timer(sk, &hc->tx_no_feedback_timer);
458 * As we have calculated new ipi, delta, t_nom it is possible
459 * that we now can send a packet, so wake up dccp_wait_for_ccid
461 sk->sk_write_space(sk);
464 * Update timeout interval for the nofeedback timer.
465 * We use a configuration option to increase the lower bound.
466 * This can help avoid triggering the nofeedback timer too
467 * often ('spinning') on LANs with small RTTs.
469 hc->tx_t_rto = max_t(u32, 4 * hc->tx_rtt, (CONFIG_IP_DCCP_CCID3_RTO *
470 (USEC_PER_SEC / 1000)));
472 * Schedule no feedback timer to expire in
473 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
475 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
477 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
478 "expire in %lu jiffies (%luus)\n",
479 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
481 sk_reset_timer(sk, &hc->tx_no_feedback_timer,
482 jiffies + usecs_to_jiffies(t_nfb));
485 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
486 unsigned char len, u16 idx,
487 unsigned char *value)
489 int rc = 0;
490 const struct dccp_sock *dp = dccp_sk(sk);
491 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
492 struct ccid3_options_received *opt_recv;
493 __be32 opt_val;
495 opt_recv = &hc->tx_options_received;
497 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
498 opt_recv->ccid3or_seqno = dp->dccps_gsr;
499 opt_recv->ccid3or_loss_event_rate = ~0;
500 opt_recv->ccid3or_loss_intervals_idx = 0;
501 opt_recv->ccid3or_loss_intervals_len = 0;
502 opt_recv->ccid3or_receive_rate = 0;
505 switch (option) {
506 case TFRC_OPT_LOSS_EVENT_RATE:
507 if (unlikely(len != 4)) {
508 DCCP_WARN("%s(%p), invalid len %d "
509 "for TFRC_OPT_LOSS_EVENT_RATE\n",
510 dccp_role(sk), sk, len);
511 rc = -EINVAL;
512 } else {
513 opt_val = get_unaligned((__be32 *)value);
514 opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
515 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
516 dccp_role(sk), sk,
517 opt_recv->ccid3or_loss_event_rate);
519 break;
520 case TFRC_OPT_LOSS_INTERVALS:
521 opt_recv->ccid3or_loss_intervals_idx = idx;
522 opt_recv->ccid3or_loss_intervals_len = len;
523 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
524 dccp_role(sk), sk,
525 opt_recv->ccid3or_loss_intervals_idx,
526 opt_recv->ccid3or_loss_intervals_len);
527 break;
528 case TFRC_OPT_RECEIVE_RATE:
529 if (unlikely(len != 4)) {
530 DCCP_WARN("%s(%p), invalid len %d "
531 "for TFRC_OPT_RECEIVE_RATE\n",
532 dccp_role(sk), sk, len);
533 rc = -EINVAL;
534 } else {
535 opt_val = get_unaligned((__be32 *)value);
536 opt_recv->ccid3or_receive_rate = ntohl(opt_val);
537 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
538 dccp_role(sk), sk,
539 opt_recv->ccid3or_receive_rate);
541 break;
544 return rc;
547 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
549 struct ccid3_hc_tx_sock *hc = ccid_priv(ccid);
551 hc->tx_state = TFRC_SSTATE_NO_SENT;
552 hc->tx_hist = NULL;
553 setup_timer(&hc->tx_no_feedback_timer,
554 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
555 return 0;
558 static void ccid3_hc_tx_exit(struct sock *sk)
560 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
562 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
563 sk_stop_timer(sk, &hc->tx_no_feedback_timer);
565 tfrc_tx_hist_purge(&hc->tx_hist);
568 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
570 struct ccid3_hc_tx_sock *hc;
572 /* Listen socks doesn't have a private CCID block */
573 if (sk->sk_state == DCCP_LISTEN)
574 return;
576 hc = ccid3_hc_tx_sk(sk);
577 info->tcpi_rto = hc->tx_t_rto;
578 info->tcpi_rtt = hc->tx_rtt;
581 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
582 u32 __user *optval, int __user *optlen)
584 const struct ccid3_hc_tx_sock *hc;
585 const void *val;
587 /* Listen socks doesn't have a private CCID block */
588 if (sk->sk_state == DCCP_LISTEN)
589 return -EINVAL;
591 hc = ccid3_hc_tx_sk(sk);
592 switch (optname) {
593 case DCCP_SOCKOPT_CCID_TX_INFO:
594 if (len < sizeof(hc->tx_tfrc))
595 return -EINVAL;
596 len = sizeof(hc->tx_tfrc);
597 val = &hc->tx_tfrc;
598 break;
599 default:
600 return -ENOPROTOOPT;
603 if (put_user(len, optlen) || copy_to_user(optval, val, len))
604 return -EFAULT;
606 return 0;
610 * Receiver Half-Connection Routines
613 /* CCID3 feedback types */
614 enum ccid3_fback_type {
615 CCID3_FBACK_NONE = 0,
616 CCID3_FBACK_INITIAL,
617 CCID3_FBACK_PERIODIC,
618 CCID3_FBACK_PARAM_CHANGE
621 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
622 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
624 static const char *const ccid3_rx_state_names[] = {
625 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
626 [TFRC_RSTATE_DATA] = "DATA",
627 [TFRC_RSTATE_TERM] = "TERM",
630 return ccid3_rx_state_names[state];
632 #endif
634 static void ccid3_hc_rx_set_state(struct sock *sk,
635 enum ccid3_hc_rx_states state)
637 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
638 enum ccid3_hc_rx_states oldstate = hc->rx_state;
640 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
641 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
642 ccid3_rx_state_name(state));
643 WARN_ON(state == oldstate);
644 hc->rx_state = state;
647 static void ccid3_hc_rx_send_feedback(struct sock *sk,
648 const struct sk_buff *skb,
649 enum ccid3_fback_type fbtype)
651 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
652 struct dccp_sock *dp = dccp_sk(sk);
653 ktime_t now;
654 s64 delta = 0;
656 if (unlikely(hc->rx_state == TFRC_RSTATE_TERM))
657 return;
659 now = ktime_get_real();
661 switch (fbtype) {
662 case CCID3_FBACK_INITIAL:
663 hc->rx_x_recv = 0;
664 hc->rx_pinv = ~0U; /* see RFC 4342, 8.5 */
665 break;
666 case CCID3_FBACK_PARAM_CHANGE:
668 * When parameters change (new loss or p > p_prev), we do not
669 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
670 * need to reuse the previous value of X_recv. However, when
671 * X_recv was 0 (due to early loss), this would kill X down to
672 * s/t_mbi (i.e. one packet in 64 seconds).
673 * To avoid such drastic reduction, we approximate X_recv as
674 * the number of bytes since last feedback.
675 * This is a safe fallback, since X is bounded above by X_calc.
677 if (hc->rx_x_recv > 0)
678 break;
679 /* fall through */
680 case CCID3_FBACK_PERIODIC:
681 delta = ktime_us_delta(now, hc->rx_tstamp_last_feedback);
682 if (delta <= 0)
683 DCCP_BUG("delta (%ld) <= 0", (long)delta);
684 else
685 hc->rx_x_recv = scaled_div32(hc->rx_bytes_recv, delta);
686 break;
687 default:
688 return;
691 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta,
692 hc->rx_x_recv, hc->rx_pinv);
694 hc->rx_tstamp_last_feedback = now;
695 hc->rx_last_counter = dccp_hdr(skb)->dccph_ccval;
696 hc->rx_bytes_recv = 0;
698 dp->dccps_hc_rx_insert_options = 1;
699 dccp_send_ack(sk);
702 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
704 const struct ccid3_hc_rx_sock *hc;
705 __be32 x_recv, pinv;
707 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
708 return 0;
710 hc = ccid3_hc_rx_sk(sk);
712 if (dccp_packet_without_ack(skb))
713 return 0;
715 x_recv = htonl(hc->rx_x_recv);
716 pinv = htonl(hc->rx_pinv);
718 if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
719 &pinv, sizeof(pinv)) ||
720 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
721 &x_recv, sizeof(x_recv)))
722 return -1;
724 return 0;
728 * ccid3_first_li - Implements [RFC 5348, 6.3.1]
730 * Determine the length of the first loss interval via inverse lookup.
731 * Assume that X_recv can be computed by the throughput equation
733 * X_recv = --------
734 * R * fval
735 * Find some p such that f(p) = fval; return 1/p (scaled).
737 static u32 ccid3_first_li(struct sock *sk)
739 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
740 u32 x_recv, p, delta;
741 u64 fval;
743 if (hc->rx_rtt == 0) {
744 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
745 hc->rx_rtt = DCCP_FALLBACK_RTT;
748 delta = ktime_to_us(net_timedelta(hc->rx_tstamp_last_feedback));
749 x_recv = scaled_div32(hc->rx_bytes_recv, delta);
750 if (x_recv == 0) { /* would also trigger divide-by-zero */
751 DCCP_WARN("X_recv==0\n");
752 if ((x_recv = hc->rx_x_recv) == 0) {
753 DCCP_BUG("stored value of X_recv is zero");
754 return ~0U;
758 fval = scaled_div(hc->rx_s, hc->rx_rtt);
759 fval = scaled_div32(fval, x_recv);
760 p = tfrc_calc_x_reverse_lookup(fval);
762 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
763 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
765 return p == 0 ? ~0U : scaled_div(1, p);
768 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
770 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
771 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
772 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
773 const bool is_data_packet = dccp_data_packet(skb);
775 if (unlikely(hc->rx_state == TFRC_RSTATE_NO_DATA)) {
776 if (is_data_packet) {
777 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
778 do_feedback = CCID3_FBACK_INITIAL;
779 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
780 hc->rx_s = payload;
782 * Not necessary to update rx_bytes_recv here,
783 * since X_recv = 0 for the first feedback packet (cf.
784 * RFC 3448, 6.3) -- gerrit
787 goto update_records;
790 if (tfrc_rx_hist_duplicate(&hc->rx_hist, skb))
791 return; /* done receiving */
793 if (is_data_packet) {
794 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
796 * Update moving-average of s and the sum of received payload bytes
798 hc->rx_s = tfrc_ewma(hc->rx_s, payload, 9);
799 hc->rx_bytes_recv += payload;
803 * Perform loss detection and handle pending losses
805 if (tfrc_rx_handle_loss(&hc->rx_hist, &hc->rx_li_hist,
806 skb, ndp, ccid3_first_li, sk)) {
807 do_feedback = CCID3_FBACK_PARAM_CHANGE;
808 goto done_receiving;
811 if (tfrc_rx_hist_loss_pending(&hc->rx_hist))
812 return; /* done receiving */
815 * Handle data packets: RTT sampling and monitoring p
817 if (unlikely(!is_data_packet))
818 goto update_records;
820 if (!tfrc_lh_is_initialised(&hc->rx_li_hist)) {
821 const u32 sample = tfrc_rx_hist_sample_rtt(&hc->rx_hist, skb);
823 * Empty loss history: no loss so far, hence p stays 0.
824 * Sample RTT values, since an RTT estimate is required for the
825 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
827 if (sample != 0)
828 hc->rx_rtt = tfrc_ewma(hc->rx_rtt, sample, 9);
830 } else if (tfrc_lh_update_i_mean(&hc->rx_li_hist, skb)) {
832 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
833 * has decreased (resp. p has increased), send feedback now.
835 do_feedback = CCID3_FBACK_PARAM_CHANGE;
839 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
841 if (SUB16(dccp_hdr(skb)->dccph_ccval, hc->rx_last_counter) > 3)
842 do_feedback = CCID3_FBACK_PERIODIC;
844 update_records:
845 tfrc_rx_hist_add_packet(&hc->rx_hist, skb, ndp);
847 done_receiving:
848 if (do_feedback)
849 ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
852 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
854 struct ccid3_hc_rx_sock *hc = ccid_priv(ccid);
856 hc->rx_state = TFRC_RSTATE_NO_DATA;
857 tfrc_lh_init(&hc->rx_li_hist);
858 return tfrc_rx_hist_alloc(&hc->rx_hist);
861 static void ccid3_hc_rx_exit(struct sock *sk)
863 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
865 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
867 tfrc_rx_hist_purge(&hc->rx_hist);
868 tfrc_lh_cleanup(&hc->rx_li_hist);
871 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
873 const struct ccid3_hc_rx_sock *hc;
875 /* Listen socks doesn't have a private CCID block */
876 if (sk->sk_state == DCCP_LISTEN)
877 return;
879 hc = ccid3_hc_rx_sk(sk);
880 info->tcpi_ca_state = hc->rx_state;
881 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
882 info->tcpi_rcv_rtt = hc->rx_rtt;
885 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
886 u32 __user *optval, int __user *optlen)
888 const struct ccid3_hc_rx_sock *hc;
889 struct tfrc_rx_info rx_info;
890 const void *val;
892 /* Listen socks doesn't have a private CCID block */
893 if (sk->sk_state == DCCP_LISTEN)
894 return -EINVAL;
896 hc = ccid3_hc_rx_sk(sk);
897 switch (optname) {
898 case DCCP_SOCKOPT_CCID_RX_INFO:
899 if (len < sizeof(rx_info))
900 return -EINVAL;
901 rx_info.tfrcrx_x_recv = hc->rx_x_recv;
902 rx_info.tfrcrx_rtt = hc->rx_rtt;
903 rx_info.tfrcrx_p = hc->rx_pinv == 0 ? ~0U :
904 scaled_div(1, hc->rx_pinv);
905 len = sizeof(rx_info);
906 val = &rx_info;
907 break;
908 default:
909 return -ENOPROTOOPT;
912 if (put_user(len, optlen) || copy_to_user(optval, val, len))
913 return -EFAULT;
915 return 0;
918 struct ccid_operations ccid3_ops = {
919 .ccid_id = DCCPC_CCID3,
920 .ccid_name = "TCP-Friendly Rate Control",
921 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
922 .ccid_hc_tx_init = ccid3_hc_tx_init,
923 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
924 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
925 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
926 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
927 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
928 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
929 .ccid_hc_rx_init = ccid3_hc_rx_init,
930 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
931 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
932 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
933 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
934 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
935 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
936 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
939 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
940 module_param(ccid3_debug, bool, 0644);
941 MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages");
942 #endif