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, struct timeval *now)
134 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
135 const __u32 old_x = hctx->ccid3hctx_x;
137 /* To avoid large error in calcX */
138 if (hctx->ccid3hctx_p >= TFRC_SMALLEST_P) {
139 hctx->ccid3hctx_x_calc = tfrc_calc_x(hctx->ccid3hctx_s,
142 hctx->ccid3hctx_x = max_t(u32, min(hctx->ccid3hctx_x_calc,
143 hctx->ccid3hctx_x_recv * 2),
144 hctx->ccid3hctx_s / TFRC_T_MBI);
146 } else if (timeval_delta(now, &hctx->ccid3hctx_t_ld) >=
147 hctx->ccid3hctx_rtt) {
148 hctx->ccid3hctx_x = max(min(hctx->ccid3hctx_x_recv,
149 hctx->ccid3hctx_x ) * 2,
150 usecs_div(hctx->ccid3hctx_s,
151 hctx->ccid3hctx_rtt) );
152 hctx->ccid3hctx_t_ld = *now;
154 ccid3_pr_debug("Not changing X\n");
156 if (hctx->ccid3hctx_x != old_x)
157 ccid3_update_send_time(hctx);
161 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
162 * @len: DCCP packet payload size in bytes
164 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
166 if (unlikely(len == 0))
167 ccid3_pr_debug("Packet payload length is 0 - not updating\n");
169 hctx->ccid3hctx_s = hctx->ccid3hctx_s == 0 ? len :
170 (9 * hctx->ccid3hctx_s + len) / 10;
172 * Note: We could do a potential optimisation here - when `s' changes,
173 * recalculate sending rate and consequently t_ipi, t_delta, and
174 * t_now. This is however non-standard, and the benefits are not
175 * clear, so it is currently left out.
179 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
181 struct sock *sk = (struct sock *)data;
182 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
183 unsigned long t_nfb = USEC_PER_SEC / 5;
186 if (sock_owned_by_user(sk)) {
187 /* Try again later. */
188 /* XXX: set some sensible MIB */
192 ccid3_pr_debug("%s, sk=%p, state=%s\n", dccp_role(sk), sk,
193 ccid3_tx_state_name(hctx->ccid3hctx_state));
195 switch (hctx->ccid3hctx_state) {
196 case TFRC_SSTATE_NO_FBACK:
197 /* RFC 3448, 4.4: Halve send rate directly */
198 hctx->ccid3hctx_x = min_t(u32, hctx->ccid3hctx_x / 2,
199 hctx->ccid3hctx_s / TFRC_T_MBI);
201 ccid3_pr_debug("%s, sk=%p, state=%s, updated tx rate to %d "
204 ccid3_tx_state_name(hctx->ccid3hctx_state),
206 /* The value of R is still undefined and so we can not recompute
207 * the timout value. Keep initial value as per [RFC 4342, 5]. */
208 t_nfb = TFRC_INITIAL_TIMEOUT;
209 ccid3_update_send_time(hctx);
211 case TFRC_SSTATE_FBACK:
213 * Check if IDLE since last timeout and recv rate is less than
216 if (!hctx->ccid3hctx_idle ||
217 (hctx->ccid3hctx_x_recv >=
218 4 * usecs_div(hctx->ccid3hctx_s, hctx->ccid3hctx_rtt))) {
221 ccid3_pr_debug("%s, sk=%p, state=%s, not idle\n",
223 ccid3_tx_state_name(hctx->ccid3hctx_state));
224 /* Halve sending rate */
226 /* If (X_calc > 2 * X_recv)
227 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
229 * X_recv = X_calc / 4;
231 BUG_ON(hctx->ccid3hctx_p >= TFRC_SMALLEST_P &&
232 hctx->ccid3hctx_x_calc == 0);
234 /* check also if p is zero -> x_calc is infinity? */
235 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P ||
236 hctx->ccid3hctx_x_calc > 2 * hctx->ccid3hctx_x_recv)
237 hctx->ccid3hctx_x_recv = max_t(u32, hctx->ccid3hctx_x_recv / 2,
238 hctx->ccid3hctx_s / (2 * TFRC_T_MBI));
240 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc / 4;
242 /* Update sending rate */
243 dccp_timestamp(sk, &now);
244 ccid3_hc_tx_update_x(sk, &now);
247 * Schedule no feedback timer to expire in
248 * max(4 * R, 2 * s/X) = max(4 * R, 2 * t_ipi)
250 t_nfb = max(4 * hctx->ccid3hctx_rtt, 2 * hctx->ccid3hctx_t_ipi);
252 case TFRC_SSTATE_NO_SENT:
253 DCCP_BUG("Illegal %s state NO_SENT, sk=%p", dccp_role(sk), sk);
255 case TFRC_SSTATE_TERM:
259 hctx->ccid3hctx_idle = 1;
262 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
263 jiffies + usecs_to_jiffies(t_nfb));
271 * > 0: delay (in msecs) that should pass before actually sending
272 * = 0: can send immediately
273 * < 0: error condition; do not send packet
275 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
277 struct dccp_sock *dp = dccp_sk(sk);
278 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
279 struct dccp_tx_hist_entry *new_packet;
283 BUG_ON(hctx == NULL);
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))
293 /* See if last packet allocated was not sent */
294 new_packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
295 if (new_packet == NULL || new_packet->dccphtx_sent) {
296 new_packet = dccp_tx_hist_entry_new(ccid3_tx_hist,
299 if (unlikely(new_packet == NULL)) {
300 DCCP_WARN("%s, sk=%p, not enough mem to add to history,"
301 "send refused\n", dccp_role(sk), sk);
305 dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, new_packet);
308 dccp_timestamp(sk, &now);
310 switch (hctx->ccid3hctx_state) {
311 case TFRC_SSTATE_NO_SENT:
312 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
313 jiffies + usecs_to_jiffies(TFRC_INITIAL_TIMEOUT));
314 hctx->ccid3hctx_last_win_count = 0;
315 hctx->ccid3hctx_t_last_win_count = now;
316 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
318 /* Set initial sending rate to 1 packet per second */
319 ccid3_hc_tx_update_s(hctx, skb->len);
320 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
322 /* First timeout, according to [RFC 3448, 4.2], is 1 second */
323 hctx->ccid3hctx_t_ipi = USEC_PER_SEC;
324 /* Initial delta: minimum of 0.5 sec and t_gran/2 */
325 hctx->ccid3hctx_delta = TFRC_OPSYS_HALF_TIME_GRAN;
327 /* Set t_0 for initial packet */
328 hctx->ccid3hctx_t_nom = now;
330 case TFRC_SSTATE_NO_FBACK:
331 case TFRC_SSTATE_FBACK:
332 delay = timeval_delta(&hctx->ccid3hctx_t_nom, &now);
334 * Scheduling of packet transmissions [RFC 3448, 4.6]
336 * if (t_now > t_nom - delta)
337 * // send the packet now
339 * // send the packet in (t_nom - t_now) milliseconds.
341 if (delay >= hctx->ccid3hctx_delta)
342 return delay / 1000L;
344 case TFRC_SSTATE_TERM:
345 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
349 /* prepare to send now (add options etc.) */
350 dp->dccps_hc_tx_insert_options = 1;
351 new_packet->dccphtx_ccval = DCCP_SKB_CB(skb)->dccpd_ccval =
352 hctx->ccid3hctx_last_win_count;
353 timeval_add_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
358 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more, unsigned int len)
360 const struct dccp_sock *dp = dccp_sk(sk);
361 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
363 unsigned long quarter_rtt;
364 struct dccp_tx_hist_entry *packet;
366 BUG_ON(hctx == NULL);
368 dccp_timestamp(sk, &now);
370 ccid3_hc_tx_update_s(hctx, len);
372 packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
373 if (unlikely(packet == NULL)) {
374 DCCP_WARN("packet doesn't exist in history!\n");
377 if (unlikely(packet->dccphtx_sent)) {
378 DCCP_WARN("no unsent packet in history!\n");
381 packet->dccphtx_tstamp = now;
382 packet->dccphtx_seqno = dp->dccps_gss;
384 * Check if win_count have changed
385 * Algorithm in "8.1. Window Counter Value" in RFC 4342.
387 quarter_rtt = timeval_delta(&now, &hctx->ccid3hctx_t_last_win_count);
388 if (likely(hctx->ccid3hctx_rtt > 8))
389 quarter_rtt /= hctx->ccid3hctx_rtt / 4;
391 if (quarter_rtt > 0) {
392 hctx->ccid3hctx_t_last_win_count = now;
393 hctx->ccid3hctx_last_win_count = (hctx->ccid3hctx_last_win_count +
394 min_t(unsigned long, quarter_rtt, 5)) % 16;
395 ccid3_pr_debug("%s, sk=%p, window changed from "
398 packet->dccphtx_ccval,
399 hctx->ccid3hctx_last_win_count);
402 hctx->ccid3hctx_idle = 0;
403 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
404 packet->dccphtx_sent = 1;
407 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
409 const struct dccp_sock *dp = dccp_sk(sk);
410 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
411 struct ccid3_options_received *opt_recv;
412 struct dccp_tx_hist_entry *packet;
420 BUG_ON(hctx == NULL);
422 /* we are only interested in ACKs */
423 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
424 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
427 opt_recv = &hctx->ccid3hctx_options_received;
429 t_elapsed = dp->dccps_options_received.dccpor_elapsed_time * 10;
430 x_recv = opt_recv->ccid3or_receive_rate;
431 pinv = opt_recv->ccid3or_loss_event_rate;
433 switch (hctx->ccid3hctx_state) {
434 case TFRC_SSTATE_NO_FBACK:
435 case TFRC_SSTATE_FBACK:
436 /* Calculate new round trip sample by
437 * R_sample = (now - t_recvdata) - t_delay */
438 /* get t_recvdata from history */
439 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
440 DCCP_SKB_CB(skb)->dccpd_ack_seq);
441 if (unlikely(packet == NULL)) {
442 DCCP_WARN("%s, sk=%p, seqno %llu(%s) does't exist "
443 "in history!\n", dccp_role(sk), sk,
444 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
445 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
450 dccp_timestamp(sk, &now);
451 r_sample = timeval_delta(&now, &packet->dccphtx_tstamp);
452 if (unlikely(r_sample <= t_elapsed))
453 DCCP_WARN("r_sample=%uus,t_elapsed=%uus\n",
454 r_sample, t_elapsed);
456 r_sample -= t_elapsed;
458 /* Update RTT estimate by
459 * If (No feedback recv)
462 * R = q * R + (1 - q) * R_sample;
464 * q is a constant, RFC 3448 recomments 0.9
466 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
467 /* Use Larger Initial Windows [RFC 4342, sec. 5]
468 * We deviate in that we use `s' instead of `MSS'. */
469 u16 w_init = max( 4 * hctx->ccid3hctx_s,
470 max(2 * hctx->ccid3hctx_s, 4380));
471 hctx->ccid3hctx_rtt = r_sample;
472 hctx->ccid3hctx_x = usecs_div(w_init, r_sample);
473 hctx->ccid3hctx_t_ld = now;
475 ccid3_update_send_time(hctx);
476 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
478 hctx->ccid3hctx_rtt = (hctx->ccid3hctx_rtt * 9) / 10 +
480 ccid3_hc_tx_update_x(sk, &now);
483 ccid3_pr_debug("%s, sk=%p, New RTT estimate=%uus, "
484 "r_sample=%us\n", dccp_role(sk), sk,
485 hctx->ccid3hctx_rtt, r_sample);
487 /* Update receive rate */
488 hctx->ccid3hctx_x_recv = x_recv;/* X_recv in bytes per sec */
490 /* Update loss event rate */
491 if (pinv == ~0 || pinv == 0)
492 hctx->ccid3hctx_p = 0;
494 hctx->ccid3hctx_p = 1000000 / pinv;
496 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P) {
497 hctx->ccid3hctx_p = TFRC_SMALLEST_P;
498 ccid3_pr_debug("%s, sk=%p, Smallest p used!\n",
503 /* unschedule no feedback timer */
504 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
506 /* remove all packets older than the one acked from history */
507 dccp_tx_hist_purge_older(ccid3_tx_hist,
508 &hctx->ccid3hctx_hist, packet);
510 * As we have calculated new ipi, delta, t_nom it is possible that
511 * we now can send a packet, so wake up dccp_wait_for_ccid
513 sk->sk_write_space(sk);
515 /* Update timeout interval. We use the alternative variant of
516 * [RFC 3448, 3.1] which sets the upper bound of t_rto to one
517 * second, as it is suggested for TCP (see RFC 2988, 2.4). */
518 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
521 * Schedule no feedback timer to expire in
522 * max(4 * R, 2 * s/X) = max(4 * R, 2 * t_ipi)
524 t_nfb = max(4 * hctx->ccid3hctx_rtt, 2 * hctx->ccid3hctx_t_ipi);
526 ccid3_pr_debug("%s, sk=%p, Scheduled no feedback timer to "
527 "expire in %lu jiffies (%luus)\n",
529 usecs_to_jiffies(t_nfb), t_nfb);
531 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
532 jiffies + usecs_to_jiffies(t_nfb));
535 hctx->ccid3hctx_idle = 1;
537 case TFRC_SSTATE_NO_SENT:
538 DCCP_WARN("Illegal ACK received - no packet has been sent\n");
540 case TFRC_SSTATE_TERM: /* ignore feedback when closing */
545 static int ccid3_hc_tx_insert_options(struct sock *sk, struct sk_buff *skb)
547 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
549 BUG_ON(hctx == NULL);
551 if (sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)
552 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
556 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
557 unsigned char len, u16 idx,
558 unsigned char *value)
561 const struct dccp_sock *dp = dccp_sk(sk);
562 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
563 struct ccid3_options_received *opt_recv;
565 BUG_ON(hctx == NULL);
567 opt_recv = &hctx->ccid3hctx_options_received;
569 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
570 opt_recv->ccid3or_seqno = dp->dccps_gsr;
571 opt_recv->ccid3or_loss_event_rate = ~0;
572 opt_recv->ccid3or_loss_intervals_idx = 0;
573 opt_recv->ccid3or_loss_intervals_len = 0;
574 opt_recv->ccid3or_receive_rate = 0;
578 case TFRC_OPT_LOSS_EVENT_RATE:
579 if (unlikely(len != 4)) {
580 DCCP_WARN("%s, sk=%p, invalid len %d "
581 "for TFRC_OPT_LOSS_EVENT_RATE\n",
582 dccp_role(sk), sk, len);
585 opt_recv->ccid3or_loss_event_rate = ntohl(*(__be32 *)value);
586 ccid3_pr_debug("%s, sk=%p, LOSS_EVENT_RATE=%u\n",
588 opt_recv->ccid3or_loss_event_rate);
591 case TFRC_OPT_LOSS_INTERVALS:
592 opt_recv->ccid3or_loss_intervals_idx = idx;
593 opt_recv->ccid3or_loss_intervals_len = len;
594 ccid3_pr_debug("%s, sk=%p, LOSS_INTERVALS=(%u, %u)\n",
596 opt_recv->ccid3or_loss_intervals_idx,
597 opt_recv->ccid3or_loss_intervals_len);
599 case TFRC_OPT_RECEIVE_RATE:
600 if (unlikely(len != 4)) {
601 DCCP_WARN("%s, sk=%p, invalid len %d "
602 "for TFRC_OPT_RECEIVE_RATE\n",
603 dccp_role(sk), sk, len);
606 opt_recv->ccid3or_receive_rate = ntohl(*(__be32 *)value);
607 ccid3_pr_debug("%s, sk=%p, RECEIVE_RATE=%u\n",
609 opt_recv->ccid3or_receive_rate);
617 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
619 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
621 hctx->ccid3hctx_s = 0;
622 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
623 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
625 hctx->ccid3hctx_no_feedback_timer.function = ccid3_hc_tx_no_feedback_timer;
626 hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
627 init_timer(&hctx->ccid3hctx_no_feedback_timer);
632 static void ccid3_hc_tx_exit(struct sock *sk)
634 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
636 BUG_ON(hctx == NULL);
638 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
639 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
641 /* Empty packet history */
642 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
646 * RX Half Connection methods
649 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
650 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
652 static char *ccid3_rx_state_names[] = {
653 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
654 [TFRC_RSTATE_DATA] = "DATA",
655 [TFRC_RSTATE_TERM] = "TERM",
658 return ccid3_rx_state_names[state];
662 static void ccid3_hc_rx_set_state(struct sock *sk,
663 enum ccid3_hc_rx_states state)
665 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
666 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
668 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
669 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
670 ccid3_rx_state_name(state));
671 WARN_ON(state == oldstate);
672 hcrx->ccid3hcrx_state = state;
675 static inline void ccid3_hc_rx_update_s(struct ccid3_hc_rx_sock *hcrx, int len)
677 if (unlikely(len == 0)) /* don't update on empty packets (e.g. ACKs) */
678 ccid3_pr_debug("Packet payload length is 0 - not updating\n");
680 hcrx->ccid3hcrx_s = hcrx->ccid3hcrx_s == 0 ? len :
681 (9 * hcrx->ccid3hcrx_s + len) / 10;
684 static void ccid3_hc_rx_send_feedback(struct sock *sk)
686 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
687 struct dccp_sock *dp = dccp_sk(sk);
688 struct dccp_rx_hist_entry *packet;
691 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
693 dccp_timestamp(sk, &now);
695 switch (hcrx->ccid3hcrx_state) {
696 case TFRC_RSTATE_NO_DATA:
697 hcrx->ccid3hcrx_x_recv = 0;
699 case TFRC_RSTATE_DATA: {
700 const u32 delta = timeval_delta(&now,
701 &hcrx->ccid3hcrx_tstamp_last_feedback);
702 hcrx->ccid3hcrx_x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv,
706 case TFRC_RSTATE_TERM:
707 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
711 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
712 if (unlikely(packet == NULL)) {
713 DCCP_WARN("%s, sk=%p, no data packet in history!\n",
718 hcrx->ccid3hcrx_tstamp_last_feedback = now;
719 hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
720 hcrx->ccid3hcrx_bytes_recv = 0;
722 /* Convert to multiples of 10us */
723 hcrx->ccid3hcrx_elapsed_time =
724 timeval_delta(&now, &packet->dccphrx_tstamp) / 10;
725 if (hcrx->ccid3hcrx_p == 0)
726 hcrx->ccid3hcrx_pinv = ~0;
728 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
729 dp->dccps_hc_rx_insert_options = 1;
733 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
735 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
738 BUG_ON(hcrx == NULL);
740 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
743 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
745 if (dccp_packet_without_ack(skb))
748 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
749 pinv = htonl(hcrx->ccid3hcrx_pinv);
751 if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
752 dccp_insert_option_elapsed_time(sk, skb,
753 hcrx->ccid3hcrx_elapsed_time)) ||
754 dccp_insert_option_timestamp(sk, skb) ||
755 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
756 &pinv, sizeof(pinv)) ||
757 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
758 &x_recv, sizeof(x_recv)))
764 /* calculate first loss interval
766 * returns estimated loss interval in usecs */
768 static u32 ccid3_hc_rx_calc_first_li(struct sock *sk)
770 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
771 struct dccp_rx_hist_entry *entry, *next, *tail = NULL;
772 u32 rtt, delta, x_recv, fval, p, tmp2;
773 struct timeval tstamp = { 0, };
779 list_for_each_entry_safe(entry, next, &hcrx->ccid3hcrx_hist,
781 if (dccp_rx_hist_entry_data_packet(entry)) {
786 tstamp = entry->dccphrx_tstamp;
787 win_count = entry->dccphrx_ccval;
791 interval = win_count - entry->dccphrx_ccval;
793 interval += TFRC_WIN_COUNT_LIMIT;
801 if (unlikely(step == 0)) {
802 DCCP_WARN("%s, sk=%p, packet history has no data packets!\n",
807 if (unlikely(interval == 0)) {
808 DCCP_WARN("%s, sk=%p, Could not find a win_count interval > 0."
809 "Defaulting to 1\n", dccp_role(sk), sk);
814 DCCP_CRIT("tail is null\n");
817 rtt = timeval_delta(&tstamp, &tail->dccphrx_tstamp) * 4 / interval;
818 ccid3_pr_debug("%s, sk=%p, approximated RTT to %uus\n",
819 dccp_role(sk), sk, rtt);
822 DCCP_WARN("RTT==0, setting to 1\n");
826 dccp_timestamp(sk, &tstamp);
827 delta = timeval_delta(&tstamp, &hcrx->ccid3hcrx_tstamp_last_feedback);
828 x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv, delta);
831 x_recv = hcrx->ccid3hcrx_x_recv;
833 tmp1 = (u64)x_recv * (u64)rtt;
834 do_div(tmp1,10000000);
838 DCCP_CRIT("tmp2 = 0, x_recv = %u, rtt =%u\n", x_recv, rtt);
842 fval = (hcrx->ccid3hcrx_s * 100000) / tmp2;
843 /* do not alter order above or you will get overflow on 32 bit */
844 p = tfrc_calc_x_reverse_lookup(fval);
845 ccid3_pr_debug("%s, sk=%p, receive rate=%u bytes/s, implied "
846 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
854 static void ccid3_hc_rx_update_li(struct sock *sk, u64 seq_loss, u8 win_loss)
856 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
857 struct dccp_li_hist_entry *head;
860 if (list_empty(&hcrx->ccid3hcrx_li_hist)) {
861 if (!dccp_li_hist_interval_new(ccid3_li_hist,
862 &hcrx->ccid3hcrx_li_hist, seq_loss, win_loss))
865 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
866 struct dccp_li_hist_entry, dccplih_node);
867 head->dccplih_interval = ccid3_hc_rx_calc_first_li(sk);
869 struct dccp_li_hist_entry *entry;
870 struct list_head *tail;
872 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
873 struct dccp_li_hist_entry, dccplih_node);
874 /* FIXME win count check removed as was wrong */
875 /* should make this check with receive history */
876 /* and compare there as per section 10.2 of RFC4342 */
878 /* new loss event detected */
879 /* calculate last interval length */
880 seq_temp = dccp_delta_seqno(head->dccplih_seqno, seq_loss);
881 entry = dccp_li_hist_entry_new(ccid3_li_hist, SLAB_ATOMIC);
884 DCCP_BUG("out of memory - can not allocate entry");
888 list_add(&entry->dccplih_node, &hcrx->ccid3hcrx_li_hist);
890 tail = hcrx->ccid3hcrx_li_hist.prev;
892 kmem_cache_free(ccid3_li_hist->dccplih_slab, tail);
894 /* Create the newest interval */
895 entry->dccplih_seqno = seq_loss;
896 entry->dccplih_interval = seq_temp;
897 entry->dccplih_win_count = win_loss;
901 static int ccid3_hc_rx_detect_loss(struct sock *sk,
902 struct dccp_rx_hist_entry *packet)
904 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
905 struct dccp_rx_hist_entry *rx_hist = dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
906 u64 seqno = packet->dccphrx_seqno;
912 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
915 follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
916 hcrx->ccid3hcrx_seqno_nonloss = seqno;
917 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
922 while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
923 > TFRC_RECV_NUM_LATE_LOSS) {
925 ccid3_hc_rx_update_li(sk, hcrx->ccid3hcrx_seqno_nonloss,
926 hcrx->ccid3hcrx_ccval_nonloss);
927 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
928 dccp_inc_seqno(&tmp_seqno);
929 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
930 dccp_inc_seqno(&tmp_seqno);
931 while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
932 tmp_seqno, &ccval)) {
933 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
934 hcrx->ccid3hcrx_ccval_nonloss = ccval;
935 dccp_inc_seqno(&tmp_seqno);
939 /* FIXME - this code could be simplified with above while */
940 /* but works at moment */
941 if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
942 hcrx->ccid3hcrx_seqno_nonloss = seqno;
943 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
947 dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
948 &hcrx->ccid3hcrx_li_hist, packet,
949 hcrx->ccid3hcrx_seqno_nonloss);
953 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
955 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
956 const struct dccp_options_received *opt_recv;
957 struct dccp_rx_hist_entry *packet;
959 u32 p_prev, rtt_prev, r_sample, t_elapsed;
960 int loss, payload_size;
962 BUG_ON(hcrx == NULL);
964 opt_recv = &dccp_sk(sk)->dccps_options_received;
966 switch (DCCP_SKB_CB(skb)->dccpd_type) {
968 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
970 case DCCP_PKT_DATAACK:
971 if (opt_recv->dccpor_timestamp_echo == 0)
973 rtt_prev = hcrx->ccid3hcrx_rtt;
974 dccp_timestamp(sk, &now);
975 timeval_sub_usecs(&now, opt_recv->dccpor_timestamp_echo * 10);
976 r_sample = timeval_usecs(&now);
977 t_elapsed = opt_recv->dccpor_elapsed_time * 10;
979 if (unlikely(r_sample <= t_elapsed))
980 DCCP_WARN("r_sample=%uus, t_elapsed=%uus\n",
981 r_sample, t_elapsed);
983 r_sample -= t_elapsed;
985 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
986 hcrx->ccid3hcrx_rtt = r_sample;
988 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
991 if (rtt_prev != hcrx->ccid3hcrx_rtt)
992 ccid3_pr_debug("%s, New RTT=%uus, elapsed time=%u\n",
993 dccp_role(sk), hcrx->ccid3hcrx_rtt,
994 opt_recv->dccpor_elapsed_time);
998 default: /* We're not interested in other packet types, move along */
1002 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, sk, opt_recv->dccpor_ndp,
1004 if (unlikely(packet == NULL)) {
1005 DCCP_WARN("%s, sk=%p, Not enough mem to add rx packet "
1006 "to history, consider it lost!\n", dccp_role(sk), sk);
1010 loss = ccid3_hc_rx_detect_loss(sk, packet);
1012 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
1015 payload_size = skb->len - dccp_hdr(skb)->dccph_doff * 4;
1016 ccid3_hc_rx_update_s(hcrx, payload_size);
1018 switch (hcrx->ccid3hcrx_state) {
1019 case TFRC_RSTATE_NO_DATA:
1020 ccid3_pr_debug("%s, sk=%p(%s), skb=%p, sending initial "
1023 dccp_state_name(sk->sk_state), skb);
1024 ccid3_hc_rx_send_feedback(sk);
1025 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
1027 case TFRC_RSTATE_DATA:
1028 hcrx->ccid3hcrx_bytes_recv += payload_size;
1032 dccp_timestamp(sk, &now);
1033 if (timeval_delta(&now, &hcrx->ccid3hcrx_tstamp_last_ack) >=
1034 hcrx->ccid3hcrx_rtt) {
1035 hcrx->ccid3hcrx_tstamp_last_ack = now;
1036 ccid3_hc_rx_send_feedback(sk);
1039 case TFRC_RSTATE_TERM:
1040 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
1044 /* Dealing with packet loss */
1045 ccid3_pr_debug("%s, sk=%p(%s), data loss! Reacting...\n",
1046 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
1048 p_prev = hcrx->ccid3hcrx_p;
1050 /* Calculate loss event rate */
1051 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
1052 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
1054 /* Scaling up by 1000000 as fixed decimal */
1056 hcrx->ccid3hcrx_p = 1000000 / i_mean;
1058 DCCP_BUG("empty loss history");
1060 if (hcrx->ccid3hcrx_p > p_prev) {
1061 ccid3_hc_rx_send_feedback(sk);
1066 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
1068 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
1070 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
1072 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
1073 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
1074 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
1075 dccp_timestamp(sk, &hcrx->ccid3hcrx_tstamp_last_ack);
1076 hcrx->ccid3hcrx_tstamp_last_feedback = hcrx->ccid3hcrx_tstamp_last_ack;
1077 hcrx->ccid3hcrx_s = 0;
1078 hcrx->ccid3hcrx_rtt = 5000; /* XXX 5ms for now... */
1082 static void ccid3_hc_rx_exit(struct sock *sk)
1084 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1086 BUG_ON(hcrx == NULL);
1088 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
1090 /* Empty packet history */
1091 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
1093 /* Empty loss interval history */
1094 dccp_li_hist_purge(ccid3_li_hist, &hcrx->ccid3hcrx_li_hist);
1097 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
1099 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1101 /* Listen socks doesn't have a private CCID block */
1102 if (sk->sk_state == DCCP_LISTEN)
1105 BUG_ON(hcrx == NULL);
1107 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
1108 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1109 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
1112 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
1114 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1116 /* Listen socks doesn't have a private CCID block */
1117 if (sk->sk_state == DCCP_LISTEN)
1120 BUG_ON(hctx == NULL);
1122 info->tcpi_rto = hctx->ccid3hctx_t_rto;
1123 info->tcpi_rtt = hctx->ccid3hctx_rtt;
1126 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
1127 u32 __user *optval, int __user *optlen)
1129 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1132 /* Listen socks doesn't have a private CCID block */
1133 if (sk->sk_state == DCCP_LISTEN)
1137 case DCCP_SOCKOPT_CCID_RX_INFO:
1138 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
1140 len = sizeof(hcrx->ccid3hcrx_tfrc);
1141 val = &hcrx->ccid3hcrx_tfrc;
1144 return -ENOPROTOOPT;
1147 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1153 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
1154 u32 __user *optval, int __user *optlen)
1156 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1159 /* Listen socks doesn't have a private CCID block */
1160 if (sk->sk_state == DCCP_LISTEN)
1164 case DCCP_SOCKOPT_CCID_TX_INFO:
1165 if (len < sizeof(hctx->ccid3hctx_tfrc))
1167 len = sizeof(hctx->ccid3hctx_tfrc);
1168 val = &hctx->ccid3hctx_tfrc;
1171 return -ENOPROTOOPT;
1174 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1180 static struct ccid_operations ccid3 = {
1181 .ccid_id = DCCPC_CCID3,
1182 .ccid_name = "ccid3",
1183 .ccid_owner = THIS_MODULE,
1184 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
1185 .ccid_hc_tx_init = ccid3_hc_tx_init,
1186 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1187 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1188 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1189 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1190 .ccid_hc_tx_insert_options = ccid3_hc_tx_insert_options,
1191 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1192 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
1193 .ccid_hc_rx_init = ccid3_hc_rx_init,
1194 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1195 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1196 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1197 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1198 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1199 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1200 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1203 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
1204 module_param(ccid3_debug, int, 0444);
1205 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1208 static __init int ccid3_module_init(void)
1212 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1213 if (ccid3_rx_hist == NULL)
1216 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1217 if (ccid3_tx_hist == NULL)
1220 ccid3_li_hist = dccp_li_hist_new("ccid3");
1221 if (ccid3_li_hist == NULL)
1224 rc = ccid_register(&ccid3);
1226 goto out_free_loss_interval_history;
1230 out_free_loss_interval_history:
1231 dccp_li_hist_delete(ccid3_li_hist);
1232 ccid3_li_hist = NULL;
1234 dccp_tx_hist_delete(ccid3_tx_hist);
1235 ccid3_tx_hist = NULL;
1237 dccp_rx_hist_delete(ccid3_rx_hist);
1238 ccid3_rx_hist = NULL;
1241 module_init(ccid3_module_init);
1243 static __exit void ccid3_module_exit(void)
1245 ccid_unregister(&ccid3);
1247 if (ccid3_tx_hist != NULL) {
1248 dccp_tx_hist_delete(ccid3_tx_hist);
1249 ccid3_tx_hist = NULL;
1251 if (ccid3_rx_hist != NULL) {
1252 dccp_rx_hist_delete(ccid3_rx_hist);
1253 ccid3_rx_hist = NULL;
1255 if (ccid3_li_hist != NULL) {
1256 dccp_li_hist_delete(ccid3_li_hist);
1257 ccid3_li_hist = NULL;
1260 module_exit(ccid3_module_exit);
1262 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
1263 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1264 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1265 MODULE_LICENSE("GPL");
1266 MODULE_ALIAS("net-dccp-ccid-3");