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,
276 struct sk_buff *skb, int len)
278 struct dccp_sock *dp = dccp_sk(sk);
279 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
280 struct dccp_tx_hist_entry *new_packet;
284 BUG_ON(hctx == NULL);
287 * This function is called only for Data and DataAck packets. Sending
288 * zero-sized Data(Ack)s is theoretically possible, but for congestion
289 * control this case is pathological - ignore it.
291 if (unlikely(len == 0))
294 /* See if last packet allocated was not sent */
295 new_packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
296 if (new_packet == NULL || new_packet->dccphtx_sent) {
297 new_packet = dccp_tx_hist_entry_new(ccid3_tx_hist,
300 if (unlikely(new_packet == NULL)) {
301 DCCP_WARN("%s, sk=%p, not enough mem to add to history,"
302 "send refused\n", dccp_role(sk), sk);
306 dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, new_packet);
309 dccp_timestamp(sk, &now);
311 switch (hctx->ccid3hctx_state) {
312 case TFRC_SSTATE_NO_SENT:
313 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
314 jiffies + usecs_to_jiffies(TFRC_INITIAL_TIMEOUT));
315 hctx->ccid3hctx_last_win_count = 0;
316 hctx->ccid3hctx_t_last_win_count = now;
317 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
319 /* Set initial sending rate to 1 packet per second */
320 ccid3_hc_tx_update_s(hctx, len);
321 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
323 /* First timeout, according to [RFC 3448, 4.2], is 1 second */
324 hctx->ccid3hctx_t_ipi = USEC_PER_SEC;
325 /* Initial delta: minimum of 0.5 sec and t_gran/2 */
326 hctx->ccid3hctx_delta = TFRC_OPSYS_HALF_TIME_GRAN;
328 /* Set t_0 for initial packet */
329 hctx->ccid3hctx_t_nom = now;
331 case TFRC_SSTATE_NO_FBACK:
332 case TFRC_SSTATE_FBACK:
333 delay = timeval_delta(&hctx->ccid3hctx_t_nom, &now);
335 * Scheduling of packet transmissions [RFC 3448, 4.6]
337 * if (t_now > t_nom - delta)
338 * // send the packet now
340 * // send the packet in (t_nom - t_now) milliseconds.
342 if (delay >= hctx->ccid3hctx_delta)
343 return delay / 1000L;
345 case TFRC_SSTATE_TERM:
346 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
350 /* prepare to send now (add options etc.) */
351 dp->dccps_hc_tx_insert_options = 1;
352 new_packet->dccphtx_ccval = DCCP_SKB_CB(skb)->dccpd_ccval =
353 hctx->ccid3hctx_last_win_count;
354 timeval_add_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
359 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more, int len)
361 const struct dccp_sock *dp = dccp_sk(sk);
362 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
365 BUG_ON(hctx == NULL);
367 dccp_timestamp(sk, &now);
369 /* check if we have sent a data packet */
371 unsigned long quarter_rtt;
372 struct dccp_tx_hist_entry *packet;
374 ccid3_hc_tx_update_s(hctx, len);
376 packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
377 if (unlikely(packet == NULL)) {
378 DCCP_WARN("packet doesn't exist in history!\n");
381 if (unlikely(packet->dccphtx_sent)) {
382 DCCP_WARN("no unsent packet in history!\n");
385 packet->dccphtx_tstamp = now;
386 packet->dccphtx_seqno = dp->dccps_gss;
388 * Check if win_count have changed
389 * Algorithm in "8.1. Window Counter Value" in RFC 4342.
391 quarter_rtt = timeval_delta(&now, &hctx->ccid3hctx_t_last_win_count);
392 if (likely(hctx->ccid3hctx_rtt > 8))
393 quarter_rtt /= hctx->ccid3hctx_rtt / 4;
395 if (quarter_rtt > 0) {
396 hctx->ccid3hctx_t_last_win_count = now;
397 hctx->ccid3hctx_last_win_count = (hctx->ccid3hctx_last_win_count +
398 min_t(unsigned long, quarter_rtt, 5)) % 16;
399 ccid3_pr_debug("%s, sk=%p, window changed from "
402 packet->dccphtx_ccval,
403 hctx->ccid3hctx_last_win_count);
406 hctx->ccid3hctx_idle = 0;
407 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
408 packet->dccphtx_sent = 1;
410 ccid3_pr_debug("%s, sk=%p, seqno=%llu NOT inserted!\n",
411 dccp_role(sk), sk, dp->dccps_gss);
414 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
416 const struct dccp_sock *dp = dccp_sk(sk);
417 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
418 struct ccid3_options_received *opt_recv;
419 struct dccp_tx_hist_entry *packet;
427 BUG_ON(hctx == NULL);
429 /* we are only interested in ACKs */
430 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
431 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
434 opt_recv = &hctx->ccid3hctx_options_received;
436 t_elapsed = dp->dccps_options_received.dccpor_elapsed_time * 10;
437 x_recv = opt_recv->ccid3or_receive_rate;
438 pinv = opt_recv->ccid3or_loss_event_rate;
440 switch (hctx->ccid3hctx_state) {
441 case TFRC_SSTATE_NO_FBACK:
442 case TFRC_SSTATE_FBACK:
443 /* Calculate new round trip sample by
444 * R_sample = (now - t_recvdata) - t_delay */
445 /* get t_recvdata from history */
446 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
447 DCCP_SKB_CB(skb)->dccpd_ack_seq);
448 if (unlikely(packet == NULL)) {
449 DCCP_WARN("%s, sk=%p, seqno %llu(%s) does't exist "
450 "in history!\n", dccp_role(sk), sk,
451 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
452 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
457 dccp_timestamp(sk, &now);
458 r_sample = timeval_delta(&now, &packet->dccphtx_tstamp);
459 if (unlikely(r_sample <= t_elapsed))
460 DCCP_WARN("r_sample=%uus,t_elapsed=%uus\n",
461 r_sample, t_elapsed);
463 r_sample -= t_elapsed;
465 /* Update RTT estimate by
466 * If (No feedback recv)
469 * R = q * R + (1 - q) * R_sample;
471 * q is a constant, RFC 3448 recomments 0.9
473 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
474 /* Use Larger Initial Windows [RFC 4342, sec. 5]
475 * We deviate in that we use `s' instead of `MSS'. */
476 u16 w_init = max( 4 * hctx->ccid3hctx_s,
477 max(2 * hctx->ccid3hctx_s, 4380));
478 hctx->ccid3hctx_rtt = r_sample;
479 hctx->ccid3hctx_x = usecs_div(w_init, r_sample);
480 hctx->ccid3hctx_t_ld = now;
482 ccid3_update_send_time(hctx);
483 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
485 hctx->ccid3hctx_rtt = (hctx->ccid3hctx_rtt * 9) / 10 +
487 ccid3_hc_tx_update_x(sk, &now);
490 ccid3_pr_debug("%s, sk=%p, New RTT estimate=%uus, "
491 "r_sample=%us\n", dccp_role(sk), sk,
492 hctx->ccid3hctx_rtt, r_sample);
494 /* Update receive rate */
495 hctx->ccid3hctx_x_recv = x_recv;/* X_recv in bytes per sec */
497 /* Update loss event rate */
498 if (pinv == ~0 || pinv == 0)
499 hctx->ccid3hctx_p = 0;
501 hctx->ccid3hctx_p = 1000000 / pinv;
503 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P) {
504 hctx->ccid3hctx_p = TFRC_SMALLEST_P;
505 ccid3_pr_debug("%s, sk=%p, Smallest p used!\n",
510 /* unschedule no feedback timer */
511 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
513 /* remove all packets older than the one acked from history */
514 dccp_tx_hist_purge_older(ccid3_tx_hist,
515 &hctx->ccid3hctx_hist, packet);
517 * As we have calculated new ipi, delta, t_nom it is possible that
518 * we now can send a packet, so wake up dccp_wait_for_ccid
520 sk->sk_write_space(sk);
522 /* Update timeout interval. We use the alternative variant of
523 * [RFC 3448, 3.1] which sets the upper bound of t_rto to one
524 * second, as it is suggested for TCP (see RFC 2988, 2.4). */
525 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
528 * Schedule no feedback timer to expire in
529 * max(4 * R, 2 * s/X) = max(4 * R, 2 * t_ipi)
531 t_nfb = max(4 * hctx->ccid3hctx_rtt, 2 * hctx->ccid3hctx_t_ipi);
533 ccid3_pr_debug("%s, sk=%p, Scheduled no feedback timer to "
534 "expire in %lu jiffies (%luus)\n",
536 usecs_to_jiffies(t_nfb), t_nfb);
538 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
539 jiffies + usecs_to_jiffies(t_nfb));
542 hctx->ccid3hctx_idle = 1;
544 case TFRC_SSTATE_NO_SENT:
545 DCCP_WARN("Illegal ACK received - no packet has been sent\n");
547 case TFRC_SSTATE_TERM: /* ignore feedback when closing */
552 static int ccid3_hc_tx_insert_options(struct sock *sk, struct sk_buff *skb)
554 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
556 BUG_ON(hctx == NULL);
558 if (sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)
559 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
563 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
564 unsigned char len, u16 idx,
565 unsigned char *value)
568 const struct dccp_sock *dp = dccp_sk(sk);
569 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
570 struct ccid3_options_received *opt_recv;
572 BUG_ON(hctx == NULL);
574 opt_recv = &hctx->ccid3hctx_options_received;
576 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
577 opt_recv->ccid3or_seqno = dp->dccps_gsr;
578 opt_recv->ccid3or_loss_event_rate = ~0;
579 opt_recv->ccid3or_loss_intervals_idx = 0;
580 opt_recv->ccid3or_loss_intervals_len = 0;
581 opt_recv->ccid3or_receive_rate = 0;
585 case TFRC_OPT_LOSS_EVENT_RATE:
586 if (unlikely(len != 4)) {
587 DCCP_WARN("%s, sk=%p, invalid len %d "
588 "for TFRC_OPT_LOSS_EVENT_RATE\n",
589 dccp_role(sk), sk, len);
592 opt_recv->ccid3or_loss_event_rate = ntohl(*(__be32 *)value);
593 ccid3_pr_debug("%s, sk=%p, LOSS_EVENT_RATE=%u\n",
595 opt_recv->ccid3or_loss_event_rate);
598 case TFRC_OPT_LOSS_INTERVALS:
599 opt_recv->ccid3or_loss_intervals_idx = idx;
600 opt_recv->ccid3or_loss_intervals_len = len;
601 ccid3_pr_debug("%s, sk=%p, LOSS_INTERVALS=(%u, %u)\n",
603 opt_recv->ccid3or_loss_intervals_idx,
604 opt_recv->ccid3or_loss_intervals_len);
606 case TFRC_OPT_RECEIVE_RATE:
607 if (unlikely(len != 4)) {
608 DCCP_WARN("%s, sk=%p, invalid len %d "
609 "for TFRC_OPT_RECEIVE_RATE\n",
610 dccp_role(sk), sk, len);
613 opt_recv->ccid3or_receive_rate = ntohl(*(__be32 *)value);
614 ccid3_pr_debug("%s, sk=%p, RECEIVE_RATE=%u\n",
616 opt_recv->ccid3or_receive_rate);
624 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
626 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
628 hctx->ccid3hctx_s = 0;
629 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
630 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
632 hctx->ccid3hctx_no_feedback_timer.function = ccid3_hc_tx_no_feedback_timer;
633 hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
634 init_timer(&hctx->ccid3hctx_no_feedback_timer);
639 static void ccid3_hc_tx_exit(struct sock *sk)
641 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
643 BUG_ON(hctx == NULL);
645 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
646 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
648 /* Empty packet history */
649 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
653 * RX Half Connection methods
656 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
657 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
659 static char *ccid3_rx_state_names[] = {
660 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
661 [TFRC_RSTATE_DATA] = "DATA",
662 [TFRC_RSTATE_TERM] = "TERM",
665 return ccid3_rx_state_names[state];
669 static void ccid3_hc_rx_set_state(struct sock *sk,
670 enum ccid3_hc_rx_states state)
672 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
673 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
675 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
676 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
677 ccid3_rx_state_name(state));
678 WARN_ON(state == oldstate);
679 hcrx->ccid3hcrx_state = state;
682 static inline void ccid3_hc_rx_update_s(struct ccid3_hc_rx_sock *hcrx, int len)
684 if (unlikely(len == 0)) /* don't update on empty packets (e.g. ACKs) */
685 ccid3_pr_debug("Packet payload length is 0 - not updating\n");
687 hcrx->ccid3hcrx_s = hcrx->ccid3hcrx_s == 0 ? len :
688 (9 * hcrx->ccid3hcrx_s + len) / 10;
691 static void ccid3_hc_rx_send_feedback(struct sock *sk)
693 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
694 struct dccp_sock *dp = dccp_sk(sk);
695 struct dccp_rx_hist_entry *packet;
698 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
700 dccp_timestamp(sk, &now);
702 switch (hcrx->ccid3hcrx_state) {
703 case TFRC_RSTATE_NO_DATA:
704 hcrx->ccid3hcrx_x_recv = 0;
706 case TFRC_RSTATE_DATA: {
707 const u32 delta = timeval_delta(&now,
708 &hcrx->ccid3hcrx_tstamp_last_feedback);
709 hcrx->ccid3hcrx_x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv,
713 case TFRC_RSTATE_TERM:
714 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
718 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
719 if (unlikely(packet == NULL)) {
720 DCCP_WARN("%s, sk=%p, no data packet in history!\n",
725 hcrx->ccid3hcrx_tstamp_last_feedback = now;
726 hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
727 hcrx->ccid3hcrx_bytes_recv = 0;
729 /* Convert to multiples of 10us */
730 hcrx->ccid3hcrx_elapsed_time =
731 timeval_delta(&now, &packet->dccphrx_tstamp) / 10;
732 if (hcrx->ccid3hcrx_p == 0)
733 hcrx->ccid3hcrx_pinv = ~0;
735 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
736 dp->dccps_hc_rx_insert_options = 1;
740 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
742 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
745 BUG_ON(hcrx == NULL);
747 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
750 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
752 if (dccp_packet_without_ack(skb))
755 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
756 pinv = htonl(hcrx->ccid3hcrx_pinv);
758 if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
759 dccp_insert_option_elapsed_time(sk, skb,
760 hcrx->ccid3hcrx_elapsed_time)) ||
761 dccp_insert_option_timestamp(sk, skb) ||
762 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
763 &pinv, sizeof(pinv)) ||
764 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
765 &x_recv, sizeof(x_recv)))
771 /* calculate first loss interval
773 * returns estimated loss interval in usecs */
775 static u32 ccid3_hc_rx_calc_first_li(struct sock *sk)
777 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
778 struct dccp_rx_hist_entry *entry, *next, *tail = NULL;
779 u32 rtt, delta, x_recv, fval, p, tmp2;
780 struct timeval tstamp = { 0, };
786 list_for_each_entry_safe(entry, next, &hcrx->ccid3hcrx_hist,
788 if (dccp_rx_hist_entry_data_packet(entry)) {
793 tstamp = entry->dccphrx_tstamp;
794 win_count = entry->dccphrx_ccval;
798 interval = win_count - entry->dccphrx_ccval;
800 interval += TFRC_WIN_COUNT_LIMIT;
808 if (unlikely(step == 0)) {
809 DCCP_WARN("%s, sk=%p, packet history has no data packets!\n",
814 if (unlikely(interval == 0)) {
815 DCCP_WARN("%s, sk=%p, Could not find a win_count interval > 0."
816 "Defaulting to 1\n", dccp_role(sk), sk);
821 DCCP_CRIT("tail is null\n");
824 rtt = timeval_delta(&tstamp, &tail->dccphrx_tstamp) * 4 / interval;
825 ccid3_pr_debug("%s, sk=%p, approximated RTT to %uus\n",
826 dccp_role(sk), sk, rtt);
829 DCCP_WARN("RTT==0, setting to 1\n");
833 dccp_timestamp(sk, &tstamp);
834 delta = timeval_delta(&tstamp, &hcrx->ccid3hcrx_tstamp_last_feedback);
835 x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv, delta);
838 x_recv = hcrx->ccid3hcrx_x_recv;
840 tmp1 = (u64)x_recv * (u64)rtt;
841 do_div(tmp1,10000000);
845 DCCP_CRIT("tmp2 = 0, x_recv = %u, rtt =%u\n", x_recv, rtt);
849 fval = (hcrx->ccid3hcrx_s * 100000) / tmp2;
850 /* do not alter order above or you will get overflow on 32 bit */
851 p = tfrc_calc_x_reverse_lookup(fval);
852 ccid3_pr_debug("%s, sk=%p, receive rate=%u bytes/s, implied "
853 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
861 static void ccid3_hc_rx_update_li(struct sock *sk, u64 seq_loss, u8 win_loss)
863 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
864 struct dccp_li_hist_entry *head;
867 if (list_empty(&hcrx->ccid3hcrx_li_hist)) {
868 if (!dccp_li_hist_interval_new(ccid3_li_hist,
869 &hcrx->ccid3hcrx_li_hist, seq_loss, win_loss))
872 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
873 struct dccp_li_hist_entry, dccplih_node);
874 head->dccplih_interval = ccid3_hc_rx_calc_first_li(sk);
876 struct dccp_li_hist_entry *entry;
877 struct list_head *tail;
879 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
880 struct dccp_li_hist_entry, dccplih_node);
881 /* FIXME win count check removed as was wrong */
882 /* should make this check with receive history */
883 /* and compare there as per section 10.2 of RFC4342 */
885 /* new loss event detected */
886 /* calculate last interval length */
887 seq_temp = dccp_delta_seqno(head->dccplih_seqno, seq_loss);
888 entry = dccp_li_hist_entry_new(ccid3_li_hist, SLAB_ATOMIC);
891 DCCP_BUG("out of memory - can not allocate entry");
895 list_add(&entry->dccplih_node, &hcrx->ccid3hcrx_li_hist);
897 tail = hcrx->ccid3hcrx_li_hist.prev;
899 kmem_cache_free(ccid3_li_hist->dccplih_slab, tail);
901 /* Create the newest interval */
902 entry->dccplih_seqno = seq_loss;
903 entry->dccplih_interval = seq_temp;
904 entry->dccplih_win_count = win_loss;
908 static int ccid3_hc_rx_detect_loss(struct sock *sk,
909 struct dccp_rx_hist_entry *packet)
911 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
912 struct dccp_rx_hist_entry *rx_hist = dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
913 u64 seqno = packet->dccphrx_seqno;
919 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
922 follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
923 hcrx->ccid3hcrx_seqno_nonloss = seqno;
924 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
929 while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
930 > TFRC_RECV_NUM_LATE_LOSS) {
932 ccid3_hc_rx_update_li(sk, hcrx->ccid3hcrx_seqno_nonloss,
933 hcrx->ccid3hcrx_ccval_nonloss);
934 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
935 dccp_inc_seqno(&tmp_seqno);
936 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
937 dccp_inc_seqno(&tmp_seqno);
938 while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
939 tmp_seqno, &ccval)) {
940 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
941 hcrx->ccid3hcrx_ccval_nonloss = ccval;
942 dccp_inc_seqno(&tmp_seqno);
946 /* FIXME - this code could be simplified with above while */
947 /* but works at moment */
948 if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
949 hcrx->ccid3hcrx_seqno_nonloss = seqno;
950 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
954 dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
955 &hcrx->ccid3hcrx_li_hist, packet,
956 hcrx->ccid3hcrx_seqno_nonloss);
960 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
962 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
963 const struct dccp_options_received *opt_recv;
964 struct dccp_rx_hist_entry *packet;
966 u32 p_prev, rtt_prev, r_sample, t_elapsed;
967 int loss, payload_size;
969 BUG_ON(hcrx == NULL);
971 opt_recv = &dccp_sk(sk)->dccps_options_received;
973 switch (DCCP_SKB_CB(skb)->dccpd_type) {
975 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
977 case DCCP_PKT_DATAACK:
978 if (opt_recv->dccpor_timestamp_echo == 0)
980 rtt_prev = hcrx->ccid3hcrx_rtt;
981 dccp_timestamp(sk, &now);
982 timeval_sub_usecs(&now, opt_recv->dccpor_timestamp_echo * 10);
983 r_sample = timeval_usecs(&now);
984 t_elapsed = opt_recv->dccpor_elapsed_time * 10;
986 if (unlikely(r_sample <= t_elapsed))
987 DCCP_WARN("r_sample=%uus, t_elapsed=%uus\n",
988 r_sample, t_elapsed);
990 r_sample -= t_elapsed;
992 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
993 hcrx->ccid3hcrx_rtt = r_sample;
995 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
998 if (rtt_prev != hcrx->ccid3hcrx_rtt)
999 ccid3_pr_debug("%s, New RTT=%uus, elapsed time=%u\n",
1000 dccp_role(sk), hcrx->ccid3hcrx_rtt,
1001 opt_recv->dccpor_elapsed_time);
1005 default: /* We're not interested in other packet types, move along */
1009 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, sk, opt_recv->dccpor_ndp,
1011 if (unlikely(packet == NULL)) {
1012 DCCP_WARN("%s, sk=%p, Not enough mem to add rx packet "
1013 "to history, consider it lost!\n", dccp_role(sk), sk);
1017 loss = ccid3_hc_rx_detect_loss(sk, packet);
1019 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
1022 payload_size = skb->len - dccp_hdr(skb)->dccph_doff * 4;
1023 ccid3_hc_rx_update_s(hcrx, payload_size);
1025 switch (hcrx->ccid3hcrx_state) {
1026 case TFRC_RSTATE_NO_DATA:
1027 ccid3_pr_debug("%s, sk=%p(%s), skb=%p, sending initial "
1030 dccp_state_name(sk->sk_state), skb);
1031 ccid3_hc_rx_send_feedback(sk);
1032 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
1034 case TFRC_RSTATE_DATA:
1035 hcrx->ccid3hcrx_bytes_recv += payload_size;
1039 dccp_timestamp(sk, &now);
1040 if (timeval_delta(&now, &hcrx->ccid3hcrx_tstamp_last_ack) >=
1041 hcrx->ccid3hcrx_rtt) {
1042 hcrx->ccid3hcrx_tstamp_last_ack = now;
1043 ccid3_hc_rx_send_feedback(sk);
1046 case TFRC_RSTATE_TERM:
1047 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
1051 /* Dealing with packet loss */
1052 ccid3_pr_debug("%s, sk=%p(%s), data loss! Reacting...\n",
1053 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
1055 p_prev = hcrx->ccid3hcrx_p;
1057 /* Calculate loss event rate */
1058 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
1059 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
1061 /* Scaling up by 1000000 as fixed decimal */
1063 hcrx->ccid3hcrx_p = 1000000 / i_mean;
1065 DCCP_BUG("empty loss history");
1067 if (hcrx->ccid3hcrx_p > p_prev) {
1068 ccid3_hc_rx_send_feedback(sk);
1073 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
1075 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
1077 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
1079 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
1080 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
1081 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
1082 dccp_timestamp(sk, &hcrx->ccid3hcrx_tstamp_last_ack);
1083 hcrx->ccid3hcrx_tstamp_last_feedback = hcrx->ccid3hcrx_tstamp_last_ack;
1084 hcrx->ccid3hcrx_s = 0;
1085 hcrx->ccid3hcrx_rtt = 5000; /* XXX 5ms for now... */
1089 static void ccid3_hc_rx_exit(struct sock *sk)
1091 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1093 BUG_ON(hcrx == NULL);
1095 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
1097 /* Empty packet history */
1098 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
1100 /* Empty loss interval history */
1101 dccp_li_hist_purge(ccid3_li_hist, &hcrx->ccid3hcrx_li_hist);
1104 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
1106 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1108 /* Listen socks doesn't have a private CCID block */
1109 if (sk->sk_state == DCCP_LISTEN)
1112 BUG_ON(hcrx == NULL);
1114 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
1115 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1116 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
1119 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
1121 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1123 /* Listen socks doesn't have a private CCID block */
1124 if (sk->sk_state == DCCP_LISTEN)
1127 BUG_ON(hctx == NULL);
1129 info->tcpi_rto = hctx->ccid3hctx_t_rto;
1130 info->tcpi_rtt = hctx->ccid3hctx_rtt;
1133 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
1134 u32 __user *optval, int __user *optlen)
1136 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1139 /* Listen socks doesn't have a private CCID block */
1140 if (sk->sk_state == DCCP_LISTEN)
1144 case DCCP_SOCKOPT_CCID_RX_INFO:
1145 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
1147 len = sizeof(hcrx->ccid3hcrx_tfrc);
1148 val = &hcrx->ccid3hcrx_tfrc;
1151 return -ENOPROTOOPT;
1154 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1160 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
1161 u32 __user *optval, int __user *optlen)
1163 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1166 /* Listen socks doesn't have a private CCID block */
1167 if (sk->sk_state == DCCP_LISTEN)
1171 case DCCP_SOCKOPT_CCID_TX_INFO:
1172 if (len < sizeof(hctx->ccid3hctx_tfrc))
1174 len = sizeof(hctx->ccid3hctx_tfrc);
1175 val = &hctx->ccid3hctx_tfrc;
1178 return -ENOPROTOOPT;
1181 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1187 static struct ccid_operations ccid3 = {
1188 .ccid_id = DCCPC_CCID3,
1189 .ccid_name = "ccid3",
1190 .ccid_owner = THIS_MODULE,
1191 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
1192 .ccid_hc_tx_init = ccid3_hc_tx_init,
1193 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1194 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1195 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1196 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1197 .ccid_hc_tx_insert_options = ccid3_hc_tx_insert_options,
1198 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1199 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
1200 .ccid_hc_rx_init = ccid3_hc_rx_init,
1201 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1202 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1203 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1204 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1205 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1206 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1207 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1210 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
1211 module_param(ccid3_debug, int, 0444);
1212 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1215 static __init int ccid3_module_init(void)
1219 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1220 if (ccid3_rx_hist == NULL)
1223 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1224 if (ccid3_tx_hist == NULL)
1227 ccid3_li_hist = dccp_li_hist_new("ccid3");
1228 if (ccid3_li_hist == NULL)
1231 rc = ccid_register(&ccid3);
1233 goto out_free_loss_interval_history;
1237 out_free_loss_interval_history:
1238 dccp_li_hist_delete(ccid3_li_hist);
1239 ccid3_li_hist = NULL;
1241 dccp_tx_hist_delete(ccid3_tx_hist);
1242 ccid3_tx_hist = NULL;
1244 dccp_rx_hist_delete(ccid3_rx_hist);
1245 ccid3_rx_hist = NULL;
1248 module_init(ccid3_module_init);
1250 static __exit void ccid3_module_exit(void)
1252 ccid_unregister(&ccid3);
1254 if (ccid3_tx_hist != NULL) {
1255 dccp_tx_hist_delete(ccid3_tx_hist);
1256 ccid3_tx_hist = NULL;
1258 if (ccid3_rx_hist != NULL) {
1259 dccp_rx_hist_delete(ccid3_rx_hist);
1260 ccid3_rx_hist = NULL;
1262 if (ccid3_li_hist != NULL) {
1263 dccp_li_hist_delete(ccid3_li_hist);
1264 ccid3_li_hist = NULL;
1267 module_exit(ccid3_module_exit);
1269 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
1270 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1271 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1272 MODULE_LICENSE("GPL");
1273 MODULE_ALIAS("net-dccp-ccid-3");