2 * net/dccp/ccids/ccid3.c
4 * Copyright (c) 2005 The University of Waikato, Hamilton, New Zealand.
5 * Copyright (c) 2005-6 Ian McDonald <ian.mcdonald@jandi.co.nz>
7 * An implementation of the DCCP protocol
9 * This code has been developed by the University of Waikato WAND
10 * research group. For further information please see http://www.wand.net.nz/
12 * This code also uses code from Lulea University, rereleased as GPL by its
14 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
16 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
17 * and to make it work as a loadable module in the DCCP stack written by
18 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
20 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
22 * This program is free software; you can redistribute it and/or modify
23 * it under the terms of the GNU General Public License as published by
24 * the Free Software Foundation; either version 2 of the License, or
25 * (at your option) any later version.
27 * This program is distributed in the hope that it will be useful,
28 * but WITHOUT ANY WARRANTY; without even the implied warranty of
29 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
30 * GNU General Public License for more details.
32 * You should have received a copy of the GNU General Public License
33 * along with this program; if not, write to the Free Software
34 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
39 #include "lib/packet_history.h"
40 #include "lib/loss_interval.h"
45 * Reason for maths here is to avoid 32 bit overflow when a is big.
46 * With this we get close to the limit.
48 static u32 usecs_div(const u32 a, const u32 b)
50 const u32 div = a < (UINT_MAX / (USEC_PER_SEC / 10)) ? 10 :
51 a < (UINT_MAX / (USEC_PER_SEC / 50)) ? 50 :
52 a < (UINT_MAX / (USEC_PER_SEC / 100)) ? 100 :
53 a < (UINT_MAX / (USEC_PER_SEC / 500)) ? 500 :
54 a < (UINT_MAX / (USEC_PER_SEC / 1000)) ? 1000 :
55 a < (UINT_MAX / (USEC_PER_SEC / 5000)) ? 5000 :
56 a < (UINT_MAX / (USEC_PER_SEC / 10000)) ? 10000 :
57 a < (UINT_MAX / (USEC_PER_SEC / 50000)) ? 50000 :
59 const u32 tmp = a * (USEC_PER_SEC / div);
60 return (b >= 2 * div) ? tmp / (b / div) : tmp;
65 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
66 static int ccid3_debug;
67 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
69 #define ccid3_pr_debug(format, a...)
72 static struct dccp_tx_hist *ccid3_tx_hist;
73 static struct dccp_rx_hist *ccid3_rx_hist;
74 static struct dccp_li_hist *ccid3_li_hist;
76 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
77 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
79 static char *ccid3_state_names[] = {
80 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
81 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
82 [TFRC_SSTATE_FBACK] = "FBACK",
83 [TFRC_SSTATE_TERM] = "TERM",
86 return ccid3_state_names[state];
90 static void ccid3_hc_tx_set_state(struct sock *sk,
91 enum ccid3_hc_tx_states state)
93 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
94 enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
96 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
97 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
98 ccid3_tx_state_name(state));
99 WARN_ON(state == oldstate);
100 hctx->ccid3hctx_state = state;
104 * Recalculate scheduled nominal send time t_nom, inter-packet interval
105 * t_ipi, and delta value. Should be called after each change to X.
107 static inline void ccid3_update_send_time(struct ccid3_hc_tx_sock *hctx)
109 timeval_sub_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
111 /* Calculate new t_ipi (inter packet interval) by t_ipi = s / X_inst */
112 hctx->ccid3hctx_t_ipi = usecs_div(hctx->ccid3hctx_s, hctx->ccid3hctx_x);
114 /* Update nominal send time with regard to the new t_ipi */
115 timeval_add_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
117 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
118 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
119 TFRC_OPSYS_HALF_TIME_GRAN);
124 * x_calc = calcX(s, R, p);
125 * X = max(min(X_calc, 2 * X_recv), s / t_mbi);
127 * If (now - tld >= R)
128 * X = max(min(2 * X, 2 * X_recv), s / R);
131 static void ccid3_hc_tx_update_x(struct sock *sk)
133 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
134 const __u32 old_x = hctx->ccid3hctx_x;
136 /* To avoid large error in calcX */
137 if (hctx->ccid3hctx_p >= TFRC_SMALLEST_P) {
138 hctx->ccid3hctx_x_calc = tfrc_calc_x(hctx->ccid3hctx_s,
141 hctx->ccid3hctx_x = max_t(u32, min_t(u32, hctx->ccid3hctx_x_calc,
142 2 * hctx->ccid3hctx_x_recv),
144 TFRC_MAX_BACK_OFF_TIME));
148 dccp_timestamp(sk, &now);
149 if (timeval_delta(&now, &hctx->ccid3hctx_t_ld) >=
150 hctx->ccid3hctx_rtt) {
151 hctx->ccid3hctx_x = max_t(u32, min_t(u32, hctx->ccid3hctx_x_recv,
152 hctx->ccid3hctx_x) * 2,
153 usecs_div(hctx->ccid3hctx_s,
154 hctx->ccid3hctx_rtt));
155 hctx->ccid3hctx_t_ld = now;
158 if (hctx->ccid3hctx_x != old_x)
159 ccid3_update_send_time(hctx);
163 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
164 * @len: DCCP packet payload size in bytes
166 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
168 if (unlikely(len == 0))
169 ccid3_pr_debug("Packet payload length is 0 - not updating\n");
171 hctx->ccid3hctx_s = hctx->ccid3hctx_s == 0 ? len :
172 (9 * hctx->ccid3hctx_s + len) / 10;
174 * Note: We could do a potential optimisation here - when `s' changes,
175 * recalculate sending rate and consequently t_ipi, t_delta, and
176 * t_now. This is however non-standard, and the benefits are not
177 * clear, so it is currently left out.
181 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
183 struct sock *sk = (struct sock *)data;
184 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
185 unsigned long t_nfb = USEC_PER_SEC / 5;
188 if (sock_owned_by_user(sk)) {
189 /* Try again later. */
190 /* XXX: set some sensible MIB */
194 ccid3_pr_debug("%s, sk=%p, state=%s\n", dccp_role(sk), sk,
195 ccid3_tx_state_name(hctx->ccid3hctx_state));
197 switch (hctx->ccid3hctx_state) {
198 case TFRC_SSTATE_NO_FBACK:
199 /* Halve send rate */
200 hctx->ccid3hctx_x /= 2;
201 if (hctx->ccid3hctx_x < (hctx->ccid3hctx_s /
202 TFRC_MAX_BACK_OFF_TIME))
203 hctx->ccid3hctx_x = (hctx->ccid3hctx_s /
204 TFRC_MAX_BACK_OFF_TIME);
206 ccid3_pr_debug("%s, sk=%p, state=%s, updated tx rate to %d "
209 ccid3_tx_state_name(hctx->ccid3hctx_state),
211 /* The value of R is still undefined and so we can not recompute
212 * the timout value. Keep initial value as per [RFC 4342, 5]. */
213 t_nfb = TFRC_INITIAL_TIMEOUT;
214 ccid3_update_send_time(hctx);
216 case TFRC_SSTATE_FBACK:
218 * Check if IDLE since last timeout and recv rate is less than
221 if (!hctx->ccid3hctx_idle ||
222 (hctx->ccid3hctx_x_recv >=
223 4 * usecs_div(hctx->ccid3hctx_s, hctx->ccid3hctx_rtt))) {
224 ccid3_pr_debug("%s, sk=%p, state=%s, not idle\n",
226 ccid3_tx_state_name(hctx->ccid3hctx_state));
227 /* Halve sending rate */
229 /* If (X_calc > 2 * X_recv)
230 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
232 * X_recv = X_calc / 4;
234 BUG_ON(hctx->ccid3hctx_p >= TFRC_SMALLEST_P &&
235 hctx->ccid3hctx_x_calc == 0);
237 /* check also if p is zero -> x_calc is infinity? */
238 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P ||
239 hctx->ccid3hctx_x_calc > 2 * hctx->ccid3hctx_x_recv)
240 hctx->ccid3hctx_x_recv = max_t(u32, hctx->ccid3hctx_x_recv / 2,
241 hctx->ccid3hctx_s / (2 * TFRC_MAX_BACK_OFF_TIME));
243 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc / 4;
245 /* Update sending rate */
246 ccid3_hc_tx_update_x(sk);
249 * Schedule no feedback timer to expire in
250 * max(4 * R, 2 * s/X) = max(4 * R, 2 * t_ipi)
252 t_nfb = max(4 * hctx->ccid3hctx_rtt, 2 * hctx->ccid3hctx_t_ipi);
254 case TFRC_SSTATE_NO_SENT:
255 DCCP_BUG("Illegal %s state NO_SENT, sk=%p", dccp_role(sk), sk);
257 case TFRC_SSTATE_TERM:
261 hctx->ccid3hctx_idle = 1;
264 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
265 jiffies + usecs_to_jiffies(t_nfb));
273 * > 0: delay (in msecs) that should pass before actually sending
274 * = 0: can send immediately
275 * < 0: error condition; do not send packet
277 static int ccid3_hc_tx_send_packet(struct sock *sk,
278 struct sk_buff *skb, int len)
280 struct dccp_sock *dp = dccp_sk(sk);
281 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
282 struct dccp_tx_hist_entry *new_packet;
286 BUG_ON(hctx == NULL);
289 * This function is called only for Data and DataAck packets. Sending
290 * zero-sized Data(Ack)s is theoretically possible, but for congestion
291 * control this case is pathological - ignore it.
293 if (unlikely(len == 0))
296 /* See if last packet allocated was not sent */
297 new_packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
298 if (new_packet == NULL || new_packet->dccphtx_sent) {
299 new_packet = dccp_tx_hist_entry_new(ccid3_tx_hist,
302 if (unlikely(new_packet == NULL)) {
303 DCCP_WARN("%s, sk=%p, not enough mem to add to history,"
304 "send refused\n", dccp_role(sk), sk);
308 dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, new_packet);
311 dccp_timestamp(sk, &now);
313 switch (hctx->ccid3hctx_state) {
314 case TFRC_SSTATE_NO_SENT:
315 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
316 jiffies + usecs_to_jiffies(TFRC_INITIAL_TIMEOUT));
317 hctx->ccid3hctx_last_win_count = 0;
318 hctx->ccid3hctx_t_last_win_count = now;
319 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
321 /* Set initial sending rate to 1 packet per second */
322 ccid3_hc_tx_update_s(hctx, len);
323 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
325 /* First timeout, according to [RFC 3448, 4.2], is 1 second */
326 hctx->ccid3hctx_t_ipi = USEC_PER_SEC;
327 /* Initial delta: minimum of 0.5 sec and t_gran/2 */
328 hctx->ccid3hctx_delta = TFRC_OPSYS_HALF_TIME_GRAN;
330 /* Set t_0 for initial packet */
331 hctx->ccid3hctx_t_nom = now;
333 case TFRC_SSTATE_NO_FBACK:
334 case TFRC_SSTATE_FBACK:
335 delay = timeval_delta(&hctx->ccid3hctx_t_nom, &now);
337 * Scheduling of packet transmissions [RFC 3448, 4.6]
339 * if (t_now > t_nom - delta)
340 * // send the packet now
342 * // send the packet in (t_nom - t_now) milliseconds.
344 if (delay >= hctx->ccid3hctx_delta)
345 return delay / 1000L;
347 case TFRC_SSTATE_TERM:
348 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
352 /* prepare to send now (add options etc.) */
353 dp->dccps_hc_tx_insert_options = 1;
354 new_packet->dccphtx_ccval = DCCP_SKB_CB(skb)->dccpd_ccval =
355 hctx->ccid3hctx_last_win_count;
356 timeval_add_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
361 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more, int len)
363 const struct dccp_sock *dp = dccp_sk(sk);
364 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
367 BUG_ON(hctx == NULL);
369 dccp_timestamp(sk, &now);
371 /* check if we have sent a data packet */
373 unsigned long quarter_rtt;
374 struct dccp_tx_hist_entry *packet;
376 ccid3_hc_tx_update_s(hctx, len);
378 packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
379 if (unlikely(packet == NULL)) {
380 DCCP_WARN("packet doesn't exist in history!\n");
383 if (unlikely(packet->dccphtx_sent)) {
384 DCCP_WARN("no unsent packet in history!\n");
387 packet->dccphtx_tstamp = now;
388 packet->dccphtx_seqno = dp->dccps_gss;
390 * Check if win_count have changed
391 * Algorithm in "8.1. Window Counter Value" in RFC 4342.
393 quarter_rtt = timeval_delta(&now, &hctx->ccid3hctx_t_last_win_count);
394 if (likely(hctx->ccid3hctx_rtt > 8))
395 quarter_rtt /= hctx->ccid3hctx_rtt / 4;
397 if (quarter_rtt > 0) {
398 hctx->ccid3hctx_t_last_win_count = now;
399 hctx->ccid3hctx_last_win_count = (hctx->ccid3hctx_last_win_count +
400 min_t(unsigned long, quarter_rtt, 5)) % 16;
401 ccid3_pr_debug("%s, sk=%p, window changed from "
404 packet->dccphtx_ccval,
405 hctx->ccid3hctx_last_win_count);
408 hctx->ccid3hctx_idle = 0;
409 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
410 packet->dccphtx_sent = 1;
412 ccid3_pr_debug("%s, sk=%p, seqno=%llu NOT inserted!\n",
413 dccp_role(sk), sk, dp->dccps_gss);
416 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
418 const struct dccp_sock *dp = dccp_sk(sk);
419 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
420 struct ccid3_options_received *opt_recv;
421 struct dccp_tx_hist_entry *packet;
429 BUG_ON(hctx == NULL);
431 /* we are only interested in ACKs */
432 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
433 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
436 opt_recv = &hctx->ccid3hctx_options_received;
438 t_elapsed = dp->dccps_options_received.dccpor_elapsed_time * 10;
439 x_recv = opt_recv->ccid3or_receive_rate;
440 pinv = opt_recv->ccid3or_loss_event_rate;
442 switch (hctx->ccid3hctx_state) {
443 case TFRC_SSTATE_NO_FBACK:
444 case TFRC_SSTATE_FBACK:
445 /* Calculate new round trip sample by
446 * R_sample = (now - t_recvdata) - t_delay */
447 /* get t_recvdata from history */
448 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
449 DCCP_SKB_CB(skb)->dccpd_ack_seq);
450 if (unlikely(packet == NULL)) {
451 DCCP_WARN("%s, sk=%p, seqno %llu(%s) does't exist "
452 "in history!\n", dccp_role(sk), sk,
453 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
454 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
459 dccp_timestamp(sk, &now);
460 r_sample = timeval_delta(&now, &packet->dccphtx_tstamp);
461 if (unlikely(r_sample <= t_elapsed))
462 DCCP_WARN("r_sample=%uus,t_elapsed=%uus\n",
463 r_sample, t_elapsed);
465 r_sample -= t_elapsed;
467 /* Update RTT estimate by
468 * If (No feedback recv)
471 * R = q * R + (1 - q) * R_sample;
473 * q is a constant, RFC 3448 recomments 0.9
475 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
476 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
477 hctx->ccid3hctx_rtt = r_sample;
479 hctx->ccid3hctx_rtt = (hctx->ccid3hctx_rtt * 9) / 10 +
482 ccid3_pr_debug("%s, sk=%p, New RTT estimate=%uus, "
483 "r_sample=%us\n", dccp_role(sk), sk,
484 hctx->ccid3hctx_rtt, r_sample);
486 /* Update receive rate */
487 hctx->ccid3hctx_x_recv = x_recv;/* X_recv in bytes per sec */
489 /* Update loss event rate */
490 if (pinv == ~0 || pinv == 0)
491 hctx->ccid3hctx_p = 0;
493 hctx->ccid3hctx_p = 1000000 / pinv;
495 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P) {
496 hctx->ccid3hctx_p = TFRC_SMALLEST_P;
497 ccid3_pr_debug("%s, sk=%p, Smallest p used!\n",
502 /* unschedule no feedback timer */
503 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
505 /* Update sending rate (and likely t_ipi, t_nom, and delta) */
506 ccid3_hc_tx_update_x(sk);
508 /* remove all packets older than the one acked from history */
509 dccp_tx_hist_purge_older(ccid3_tx_hist,
510 &hctx->ccid3hctx_hist, packet);
512 * As we have calculated new ipi, delta, t_nom it is possible that
513 * we now can send a packet, so wake up dccp_wait_for_ccid
515 sk->sk_write_space(sk);
518 /* Update timeout interval. We use the alternative variant of
519 * [RFC 3448, 3.1] which sets the upper bound of t_rto to one
520 * second, as it is suggested for TCP (see RFC 2988, 2.4). */
521 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
524 * Schedule no feedback timer to expire in
525 * max(4 * R, 2 * s/X) = max(4 * R, 2 * t_ipi)
527 t_nfb = max(4 * hctx->ccid3hctx_rtt, 2 * hctx->ccid3hctx_t_ipi);
529 ccid3_pr_debug("%s, sk=%p, Scheduled no feedback timer to "
530 "expire in %lu jiffies (%luus)\n",
532 usecs_to_jiffies(t_nfb), t_nfb);
534 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
535 jiffies + usecs_to_jiffies(t_nfb));
538 hctx->ccid3hctx_idle = 1;
540 case TFRC_SSTATE_NO_SENT:
541 DCCP_WARN("Illegal ACK received - no packet has been sent\n");
543 case TFRC_SSTATE_TERM: /* ignore feedback when closing */
548 static int ccid3_hc_tx_insert_options(struct sock *sk, struct sk_buff *skb)
550 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
552 BUG_ON(hctx == NULL);
554 if (sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)
555 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
559 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
560 unsigned char len, u16 idx,
561 unsigned char *value)
564 const struct dccp_sock *dp = dccp_sk(sk);
565 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
566 struct ccid3_options_received *opt_recv;
568 BUG_ON(hctx == NULL);
570 opt_recv = &hctx->ccid3hctx_options_received;
572 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
573 opt_recv->ccid3or_seqno = dp->dccps_gsr;
574 opt_recv->ccid3or_loss_event_rate = ~0;
575 opt_recv->ccid3or_loss_intervals_idx = 0;
576 opt_recv->ccid3or_loss_intervals_len = 0;
577 opt_recv->ccid3or_receive_rate = 0;
581 case TFRC_OPT_LOSS_EVENT_RATE:
582 if (unlikely(len != 4)) {
583 DCCP_WARN("%s, sk=%p, invalid len %d "
584 "for TFRC_OPT_LOSS_EVENT_RATE\n",
585 dccp_role(sk), sk, len);
588 opt_recv->ccid3or_loss_event_rate = ntohl(*(__be32 *)value);
589 ccid3_pr_debug("%s, sk=%p, LOSS_EVENT_RATE=%u\n",
591 opt_recv->ccid3or_loss_event_rate);
594 case TFRC_OPT_LOSS_INTERVALS:
595 opt_recv->ccid3or_loss_intervals_idx = idx;
596 opt_recv->ccid3or_loss_intervals_len = len;
597 ccid3_pr_debug("%s, sk=%p, LOSS_INTERVALS=(%u, %u)\n",
599 opt_recv->ccid3or_loss_intervals_idx,
600 opt_recv->ccid3or_loss_intervals_len);
602 case TFRC_OPT_RECEIVE_RATE:
603 if (unlikely(len != 4)) {
604 DCCP_WARN("%s, sk=%p, invalid len %d "
605 "for TFRC_OPT_RECEIVE_RATE\n",
606 dccp_role(sk), sk, len);
609 opt_recv->ccid3or_receive_rate = ntohl(*(__be32 *)value);
610 ccid3_pr_debug("%s, sk=%p, RECEIVE_RATE=%u\n",
612 opt_recv->ccid3or_receive_rate);
620 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
622 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
624 hctx->ccid3hctx_s = 0;
625 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
626 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
628 hctx->ccid3hctx_no_feedback_timer.function = ccid3_hc_tx_no_feedback_timer;
629 hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
630 init_timer(&hctx->ccid3hctx_no_feedback_timer);
635 static void ccid3_hc_tx_exit(struct sock *sk)
637 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
639 BUG_ON(hctx == NULL);
641 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
642 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
644 /* Empty packet history */
645 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
649 * RX Half Connection methods
652 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
653 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
655 static char *ccid3_rx_state_names[] = {
656 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
657 [TFRC_RSTATE_DATA] = "DATA",
658 [TFRC_RSTATE_TERM] = "TERM",
661 return ccid3_rx_state_names[state];
665 static void ccid3_hc_rx_set_state(struct sock *sk,
666 enum ccid3_hc_rx_states state)
668 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
669 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
671 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
672 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
673 ccid3_rx_state_name(state));
674 WARN_ON(state == oldstate);
675 hcrx->ccid3hcrx_state = state;
678 static inline void ccid3_hc_rx_update_s(struct ccid3_hc_rx_sock *hcrx, int len)
680 if (unlikely(len == 0)) /* don't update on empty packets (e.g. ACKs) */
681 ccid3_pr_debug("Packet payload length is 0 - not updating\n");
683 hcrx->ccid3hcrx_s = hcrx->ccid3hcrx_s == 0 ? len :
684 (9 * hcrx->ccid3hcrx_s + len) / 10;
687 static void ccid3_hc_rx_send_feedback(struct sock *sk)
689 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
690 struct dccp_sock *dp = dccp_sk(sk);
691 struct dccp_rx_hist_entry *packet;
694 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
696 dccp_timestamp(sk, &now);
698 switch (hcrx->ccid3hcrx_state) {
699 case TFRC_RSTATE_NO_DATA:
700 hcrx->ccid3hcrx_x_recv = 0;
702 case TFRC_RSTATE_DATA: {
703 const u32 delta = timeval_delta(&now,
704 &hcrx->ccid3hcrx_tstamp_last_feedback);
705 hcrx->ccid3hcrx_x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv,
709 case TFRC_RSTATE_TERM:
710 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
714 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
715 if (unlikely(packet == NULL)) {
716 DCCP_WARN("%s, sk=%p, no data packet in history!\n",
721 hcrx->ccid3hcrx_tstamp_last_feedback = now;
722 hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
723 hcrx->ccid3hcrx_bytes_recv = 0;
725 /* Convert to multiples of 10us */
726 hcrx->ccid3hcrx_elapsed_time =
727 timeval_delta(&now, &packet->dccphrx_tstamp) / 10;
728 if (hcrx->ccid3hcrx_p == 0)
729 hcrx->ccid3hcrx_pinv = ~0;
731 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
732 dp->dccps_hc_rx_insert_options = 1;
736 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
738 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
741 BUG_ON(hcrx == NULL);
743 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
746 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
748 if (dccp_packet_without_ack(skb))
751 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
752 pinv = htonl(hcrx->ccid3hcrx_pinv);
754 if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
755 dccp_insert_option_elapsed_time(sk, skb,
756 hcrx->ccid3hcrx_elapsed_time)) ||
757 dccp_insert_option_timestamp(sk, skb) ||
758 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
759 &pinv, sizeof(pinv)) ||
760 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
761 &x_recv, sizeof(x_recv)))
767 /* calculate first loss interval
769 * returns estimated loss interval in usecs */
771 static u32 ccid3_hc_rx_calc_first_li(struct sock *sk)
773 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
774 struct dccp_rx_hist_entry *entry, *next, *tail = NULL;
775 u32 rtt, delta, x_recv, fval, p, tmp2;
776 struct timeval tstamp = { 0, };
782 list_for_each_entry_safe(entry, next, &hcrx->ccid3hcrx_hist,
784 if (dccp_rx_hist_entry_data_packet(entry)) {
789 tstamp = entry->dccphrx_tstamp;
790 win_count = entry->dccphrx_ccval;
794 interval = win_count - entry->dccphrx_ccval;
796 interval += TFRC_WIN_COUNT_LIMIT;
804 if (unlikely(step == 0)) {
805 DCCP_WARN("%s, sk=%p, packet history has no data packets!\n",
810 if (unlikely(interval == 0)) {
811 DCCP_WARN("%s, sk=%p, Could not find a win_count interval > 0."
812 "Defaulting to 1\n", dccp_role(sk), sk);
817 DCCP_CRIT("tail is null\n");
820 rtt = timeval_delta(&tstamp, &tail->dccphrx_tstamp) * 4 / interval;
821 ccid3_pr_debug("%s, sk=%p, approximated RTT to %uus\n",
822 dccp_role(sk), sk, rtt);
825 DCCP_WARN("RTT==0, setting to 1\n");
829 dccp_timestamp(sk, &tstamp);
830 delta = timeval_delta(&tstamp, &hcrx->ccid3hcrx_tstamp_last_feedback);
831 x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv, delta);
834 x_recv = hcrx->ccid3hcrx_x_recv;
836 tmp1 = (u64)x_recv * (u64)rtt;
837 do_div(tmp1,10000000);
841 DCCP_CRIT("tmp2 = 0, x_recv = %u, rtt =%u\n", x_recv, rtt);
845 fval = (hcrx->ccid3hcrx_s * 100000) / tmp2;
846 /* do not alter order above or you will get overflow on 32 bit */
847 p = tfrc_calc_x_reverse_lookup(fval);
848 ccid3_pr_debug("%s, sk=%p, receive rate=%u bytes/s, implied "
849 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
857 static void ccid3_hc_rx_update_li(struct sock *sk, u64 seq_loss, u8 win_loss)
859 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
860 struct dccp_li_hist_entry *head;
863 if (list_empty(&hcrx->ccid3hcrx_li_hist)) {
864 if (!dccp_li_hist_interval_new(ccid3_li_hist,
865 &hcrx->ccid3hcrx_li_hist, seq_loss, win_loss))
868 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
869 struct dccp_li_hist_entry, dccplih_node);
870 head->dccplih_interval = ccid3_hc_rx_calc_first_li(sk);
872 struct dccp_li_hist_entry *entry;
873 struct list_head *tail;
875 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
876 struct dccp_li_hist_entry, dccplih_node);
877 /* FIXME win count check removed as was wrong */
878 /* should make this check with receive history */
879 /* and compare there as per section 10.2 of RFC4342 */
881 /* new loss event detected */
882 /* calculate last interval length */
883 seq_temp = dccp_delta_seqno(head->dccplih_seqno, seq_loss);
884 entry = dccp_li_hist_entry_new(ccid3_li_hist, SLAB_ATOMIC);
887 DCCP_BUG("out of memory - can not allocate entry");
891 list_add(&entry->dccplih_node, &hcrx->ccid3hcrx_li_hist);
893 tail = hcrx->ccid3hcrx_li_hist.prev;
895 kmem_cache_free(ccid3_li_hist->dccplih_slab, tail);
897 /* Create the newest interval */
898 entry->dccplih_seqno = seq_loss;
899 entry->dccplih_interval = seq_temp;
900 entry->dccplih_win_count = win_loss;
904 static int ccid3_hc_rx_detect_loss(struct sock *sk,
905 struct dccp_rx_hist_entry *packet)
907 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
908 struct dccp_rx_hist_entry *rx_hist = dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
909 u64 seqno = packet->dccphrx_seqno;
915 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
918 follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
919 hcrx->ccid3hcrx_seqno_nonloss = seqno;
920 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
925 while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
926 > TFRC_RECV_NUM_LATE_LOSS) {
928 ccid3_hc_rx_update_li(sk, hcrx->ccid3hcrx_seqno_nonloss,
929 hcrx->ccid3hcrx_ccval_nonloss);
930 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
931 dccp_inc_seqno(&tmp_seqno);
932 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
933 dccp_inc_seqno(&tmp_seqno);
934 while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
935 tmp_seqno, &ccval)) {
936 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
937 hcrx->ccid3hcrx_ccval_nonloss = ccval;
938 dccp_inc_seqno(&tmp_seqno);
942 /* FIXME - this code could be simplified with above while */
943 /* but works at moment */
944 if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
945 hcrx->ccid3hcrx_seqno_nonloss = seqno;
946 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
950 dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
951 &hcrx->ccid3hcrx_li_hist, packet,
952 hcrx->ccid3hcrx_seqno_nonloss);
956 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
958 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
959 const struct dccp_options_received *opt_recv;
960 struct dccp_rx_hist_entry *packet;
962 u32 p_prev, rtt_prev, r_sample, t_elapsed;
963 int loss, payload_size;
965 BUG_ON(hcrx == NULL);
967 opt_recv = &dccp_sk(sk)->dccps_options_received;
969 switch (DCCP_SKB_CB(skb)->dccpd_type) {
971 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
973 case DCCP_PKT_DATAACK:
974 if (opt_recv->dccpor_timestamp_echo == 0)
976 rtt_prev = hcrx->ccid3hcrx_rtt;
977 dccp_timestamp(sk, &now);
978 timeval_sub_usecs(&now, opt_recv->dccpor_timestamp_echo * 10);
979 r_sample = timeval_usecs(&now);
980 t_elapsed = opt_recv->dccpor_elapsed_time * 10;
982 if (unlikely(r_sample <= t_elapsed))
983 DCCP_WARN("r_sample=%uus, t_elapsed=%uus\n",
984 r_sample, t_elapsed);
986 r_sample -= t_elapsed;
988 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
989 hcrx->ccid3hcrx_rtt = r_sample;
991 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
994 if (rtt_prev != hcrx->ccid3hcrx_rtt)
995 ccid3_pr_debug("%s, New RTT=%uus, elapsed time=%u\n",
996 dccp_role(sk), hcrx->ccid3hcrx_rtt,
997 opt_recv->dccpor_elapsed_time);
1001 default: /* We're not interested in other packet types, move along */
1005 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, sk, opt_recv->dccpor_ndp,
1007 if (unlikely(packet == NULL)) {
1008 DCCP_WARN("%s, sk=%p, Not enough mem to add rx packet "
1009 "to history, consider it lost!\n", dccp_role(sk), sk);
1013 loss = ccid3_hc_rx_detect_loss(sk, packet);
1015 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
1018 payload_size = skb->len - dccp_hdr(skb)->dccph_doff * 4;
1019 ccid3_hc_rx_update_s(hcrx, payload_size);
1021 switch (hcrx->ccid3hcrx_state) {
1022 case TFRC_RSTATE_NO_DATA:
1023 ccid3_pr_debug("%s, sk=%p(%s), skb=%p, sending initial "
1026 dccp_state_name(sk->sk_state), skb);
1027 ccid3_hc_rx_send_feedback(sk);
1028 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
1030 case TFRC_RSTATE_DATA:
1031 hcrx->ccid3hcrx_bytes_recv += payload_size;
1035 dccp_timestamp(sk, &now);
1036 if (timeval_delta(&now, &hcrx->ccid3hcrx_tstamp_last_ack) >=
1037 hcrx->ccid3hcrx_rtt) {
1038 hcrx->ccid3hcrx_tstamp_last_ack = now;
1039 ccid3_hc_rx_send_feedback(sk);
1042 case TFRC_RSTATE_TERM:
1043 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
1047 /* Dealing with packet loss */
1048 ccid3_pr_debug("%s, sk=%p(%s), data loss! Reacting...\n",
1049 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
1051 p_prev = hcrx->ccid3hcrx_p;
1053 /* Calculate loss event rate */
1054 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
1055 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
1057 /* Scaling up by 1000000 as fixed decimal */
1059 hcrx->ccid3hcrx_p = 1000000 / i_mean;
1061 DCCP_BUG("empty loss history");
1063 if (hcrx->ccid3hcrx_p > p_prev) {
1064 ccid3_hc_rx_send_feedback(sk);
1069 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
1071 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
1073 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
1075 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
1076 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
1077 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
1078 dccp_timestamp(sk, &hcrx->ccid3hcrx_tstamp_last_ack);
1079 hcrx->ccid3hcrx_tstamp_last_feedback = hcrx->ccid3hcrx_tstamp_last_ack;
1080 hcrx->ccid3hcrx_s = 0;
1081 hcrx->ccid3hcrx_rtt = 5000; /* XXX 5ms for now... */
1085 static void ccid3_hc_rx_exit(struct sock *sk)
1087 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1089 BUG_ON(hcrx == NULL);
1091 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
1093 /* Empty packet history */
1094 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
1096 /* Empty loss interval history */
1097 dccp_li_hist_purge(ccid3_li_hist, &hcrx->ccid3hcrx_li_hist);
1100 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
1102 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1104 /* Listen socks doesn't have a private CCID block */
1105 if (sk->sk_state == DCCP_LISTEN)
1108 BUG_ON(hcrx == NULL);
1110 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
1111 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1112 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
1115 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
1117 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1119 /* Listen socks doesn't have a private CCID block */
1120 if (sk->sk_state == DCCP_LISTEN)
1123 BUG_ON(hctx == NULL);
1125 info->tcpi_rto = hctx->ccid3hctx_t_rto;
1126 info->tcpi_rtt = hctx->ccid3hctx_rtt;
1129 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
1130 u32 __user *optval, int __user *optlen)
1132 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1135 /* Listen socks doesn't have a private CCID block */
1136 if (sk->sk_state == DCCP_LISTEN)
1140 case DCCP_SOCKOPT_CCID_RX_INFO:
1141 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
1143 len = sizeof(hcrx->ccid3hcrx_tfrc);
1144 val = &hcrx->ccid3hcrx_tfrc;
1147 return -ENOPROTOOPT;
1150 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1156 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
1157 u32 __user *optval, int __user *optlen)
1159 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1162 /* Listen socks doesn't have a private CCID block */
1163 if (sk->sk_state == DCCP_LISTEN)
1167 case DCCP_SOCKOPT_CCID_TX_INFO:
1168 if (len < sizeof(hctx->ccid3hctx_tfrc))
1170 len = sizeof(hctx->ccid3hctx_tfrc);
1171 val = &hctx->ccid3hctx_tfrc;
1174 return -ENOPROTOOPT;
1177 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1183 static struct ccid_operations ccid3 = {
1184 .ccid_id = DCCPC_CCID3,
1185 .ccid_name = "ccid3",
1186 .ccid_owner = THIS_MODULE,
1187 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
1188 .ccid_hc_tx_init = ccid3_hc_tx_init,
1189 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1190 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1191 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1192 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1193 .ccid_hc_tx_insert_options = ccid3_hc_tx_insert_options,
1194 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1195 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
1196 .ccid_hc_rx_init = ccid3_hc_rx_init,
1197 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1198 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1199 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1200 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1201 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1202 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1203 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1206 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
1207 module_param(ccid3_debug, int, 0444);
1208 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1211 static __init int ccid3_module_init(void)
1215 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1216 if (ccid3_rx_hist == NULL)
1219 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1220 if (ccid3_tx_hist == NULL)
1223 ccid3_li_hist = dccp_li_hist_new("ccid3");
1224 if (ccid3_li_hist == NULL)
1227 rc = ccid_register(&ccid3);
1229 goto out_free_loss_interval_history;
1233 out_free_loss_interval_history:
1234 dccp_li_hist_delete(ccid3_li_hist);
1235 ccid3_li_hist = NULL;
1237 dccp_tx_hist_delete(ccid3_tx_hist);
1238 ccid3_tx_hist = NULL;
1240 dccp_rx_hist_delete(ccid3_rx_hist);
1241 ccid3_rx_hist = NULL;
1244 module_init(ccid3_module_init);
1246 static __exit void ccid3_module_exit(void)
1248 ccid_unregister(&ccid3);
1250 if (ccid3_tx_hist != NULL) {
1251 dccp_tx_hist_delete(ccid3_tx_hist);
1252 ccid3_tx_hist = NULL;
1254 if (ccid3_rx_hist != NULL) {
1255 dccp_rx_hist_delete(ccid3_rx_hist);
1256 ccid3_rx_hist = NULL;
1258 if (ccid3_li_hist != NULL) {
1259 dccp_li_hist_delete(ccid3_li_hist);
1260 ccid3_li_hist = NULL;
1263 module_exit(ccid3_module_exit);
1265 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
1266 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1267 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1268 MODULE_LICENSE("GPL");
1269 MODULE_ALIAS("net-dccp-ccid-3");