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(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
249 * See comments in packet_recv() regarding the value of t_RTO.
251 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
253 case TFRC_SSTATE_NO_SENT:
254 DCCP_BUG("Illegal %s state NO_SENT, sk=%p", dccp_role(sk), sk);
256 case TFRC_SSTATE_TERM:
260 hctx->ccid3hctx_idle = 1;
263 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
264 jiffies + usecs_to_jiffies(t_nfb));
272 * > 0: delay (in msecs) that should pass before actually sending
273 * = 0: can send immediately
274 * < 0: error condition; do not send packet
276 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
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(skb->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, skb->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 - (long)hctx->ccid3hctx_delta >= 0)
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, unsigned int len)
361 const struct dccp_sock *dp = dccp_sk(sk);
362 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
364 unsigned long quarter_rtt;
365 struct dccp_tx_hist_entry *packet;
367 BUG_ON(hctx == NULL);
369 dccp_timestamp(sk, &now);
371 ccid3_hc_tx_update_s(hctx, len);
373 packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
374 if (unlikely(packet == NULL)) {
375 DCCP_WARN("packet doesn't exist in history!\n");
378 if (unlikely(packet->dccphtx_sent)) {
379 DCCP_WARN("no unsent packet in history!\n");
382 packet->dccphtx_tstamp = now;
383 packet->dccphtx_seqno = dp->dccps_gss;
385 * Check if win_count have changed
386 * Algorithm in "8.1. Window Counter Value" in RFC 4342.
388 quarter_rtt = timeval_delta(&now, &hctx->ccid3hctx_t_last_win_count);
389 if (likely(hctx->ccid3hctx_rtt > 8))
390 quarter_rtt /= hctx->ccid3hctx_rtt / 4;
392 if (quarter_rtt > 0) {
393 hctx->ccid3hctx_t_last_win_count = now;
394 hctx->ccid3hctx_last_win_count = (hctx->ccid3hctx_last_win_count +
395 min_t(unsigned long, quarter_rtt, 5)) % 16;
396 ccid3_pr_debug("%s, sk=%p, window changed from "
399 packet->dccphtx_ccval,
400 hctx->ccid3hctx_last_win_count);
403 hctx->ccid3hctx_idle = 0;
404 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
405 packet->dccphtx_sent = 1;
408 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
410 const struct dccp_sock *dp = dccp_sk(sk);
411 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
412 struct ccid3_options_received *opt_recv;
413 struct dccp_tx_hist_entry *packet;
421 BUG_ON(hctx == NULL);
423 /* we are only interested in ACKs */
424 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
425 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
428 opt_recv = &hctx->ccid3hctx_options_received;
430 t_elapsed = dp->dccps_options_received.dccpor_elapsed_time * 10;
431 x_recv = opt_recv->ccid3or_receive_rate;
432 pinv = opt_recv->ccid3or_loss_event_rate;
434 switch (hctx->ccid3hctx_state) {
435 case TFRC_SSTATE_NO_FBACK:
436 case TFRC_SSTATE_FBACK:
437 /* Calculate new round trip sample by
438 * R_sample = (now - t_recvdata) - t_delay */
439 /* get t_recvdata from history */
440 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
441 DCCP_SKB_CB(skb)->dccpd_ack_seq);
442 if (unlikely(packet == NULL)) {
443 DCCP_WARN("%s, sk=%p, seqno %llu(%s) does't exist "
444 "in history!\n", dccp_role(sk), sk,
445 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
446 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
451 dccp_timestamp(sk, &now);
452 r_sample = timeval_delta(&now, &packet->dccphtx_tstamp);
453 if (unlikely(r_sample <= t_elapsed))
454 DCCP_WARN("r_sample=%uus,t_elapsed=%uus\n",
455 r_sample, t_elapsed);
457 r_sample -= t_elapsed;
459 /* Update RTT estimate by
460 * If (No feedback recv)
463 * R = q * R + (1 - q) * R_sample;
465 * q is a constant, RFC 3448 recomments 0.9
467 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
468 /* Use Larger Initial Windows [RFC 4342, sec. 5]
469 * We deviate in that we use `s' instead of `MSS'. */
470 u16 w_init = max( 4 * hctx->ccid3hctx_s,
471 max(2 * hctx->ccid3hctx_s, 4380));
472 hctx->ccid3hctx_rtt = r_sample;
473 hctx->ccid3hctx_x = usecs_div(w_init, r_sample);
474 hctx->ccid3hctx_t_ld = now;
476 ccid3_update_send_time(hctx);
477 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
479 hctx->ccid3hctx_rtt = (hctx->ccid3hctx_rtt * 9) / 10 +
481 ccid3_hc_tx_update_x(sk, &now);
484 ccid3_pr_debug("%s, sk=%p, New RTT estimate=%uus, "
485 "r_sample=%us\n", dccp_role(sk), sk,
486 hctx->ccid3hctx_rtt, r_sample);
488 /* Update receive rate */
489 hctx->ccid3hctx_x_recv = x_recv;/* X_recv in bytes per sec */
491 /* Update loss event rate */
492 if (pinv == ~0 || pinv == 0)
493 hctx->ccid3hctx_p = 0;
495 hctx->ccid3hctx_p = 1000000 / pinv;
497 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P) {
498 hctx->ccid3hctx_p = TFRC_SMALLEST_P;
499 ccid3_pr_debug("%s, sk=%p, Smallest p used!\n",
504 /* unschedule no feedback timer */
505 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
507 /* remove all packets older than the one acked from history */
508 dccp_tx_hist_purge_older(ccid3_tx_hist,
509 &hctx->ccid3hctx_hist, packet);
511 * As we have calculated new ipi, delta, t_nom it is possible that
512 * we now can send a packet, so wake up dccp_wait_for_ccid
514 sk->sk_write_space(sk);
517 * Update timeout interval for the nofeedback timer.
518 * We use a configuration option to increase the lower bound.
519 * This can help avoid triggering the nofeedback timer too often
520 * ('spinning') on LANs with small RTTs.
522 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
523 CONFIG_IP_DCCP_CCID3_RTO *
524 (USEC_PER_SEC/1000) );
526 * Schedule no feedback timer to expire in
527 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
529 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
531 ccid3_pr_debug("%s, sk=%p, Scheduled no feedback timer to "
532 "expire in %lu jiffies (%luus)\n",
534 usecs_to_jiffies(t_nfb), t_nfb);
536 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
537 jiffies + usecs_to_jiffies(t_nfb));
540 hctx->ccid3hctx_idle = 1;
542 case TFRC_SSTATE_NO_SENT:
543 DCCP_WARN("Illegal ACK received - no packet has been sent\n");
545 case TFRC_SSTATE_TERM: /* ignore feedback when closing */
550 static int ccid3_hc_tx_insert_options(struct sock *sk, struct sk_buff *skb)
552 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
554 BUG_ON(hctx == NULL);
556 if (sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)
557 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
561 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
562 unsigned char len, u16 idx,
563 unsigned char *value)
566 const struct dccp_sock *dp = dccp_sk(sk);
567 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
568 struct ccid3_options_received *opt_recv;
570 BUG_ON(hctx == NULL);
572 opt_recv = &hctx->ccid3hctx_options_received;
574 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
575 opt_recv->ccid3or_seqno = dp->dccps_gsr;
576 opt_recv->ccid3or_loss_event_rate = ~0;
577 opt_recv->ccid3or_loss_intervals_idx = 0;
578 opt_recv->ccid3or_loss_intervals_len = 0;
579 opt_recv->ccid3or_receive_rate = 0;
583 case TFRC_OPT_LOSS_EVENT_RATE:
584 if (unlikely(len != 4)) {
585 DCCP_WARN("%s, sk=%p, invalid len %d "
586 "for TFRC_OPT_LOSS_EVENT_RATE\n",
587 dccp_role(sk), sk, len);
590 opt_recv->ccid3or_loss_event_rate = ntohl(*(__be32 *)value);
591 ccid3_pr_debug("%s, sk=%p, LOSS_EVENT_RATE=%u\n",
593 opt_recv->ccid3or_loss_event_rate);
596 case TFRC_OPT_LOSS_INTERVALS:
597 opt_recv->ccid3or_loss_intervals_idx = idx;
598 opt_recv->ccid3or_loss_intervals_len = len;
599 ccid3_pr_debug("%s, sk=%p, LOSS_INTERVALS=(%u, %u)\n",
601 opt_recv->ccid3or_loss_intervals_idx,
602 opt_recv->ccid3or_loss_intervals_len);
604 case TFRC_OPT_RECEIVE_RATE:
605 if (unlikely(len != 4)) {
606 DCCP_WARN("%s, sk=%p, invalid len %d "
607 "for TFRC_OPT_RECEIVE_RATE\n",
608 dccp_role(sk), sk, len);
611 opt_recv->ccid3or_receive_rate = ntohl(*(__be32 *)value);
612 ccid3_pr_debug("%s, sk=%p, RECEIVE_RATE=%u\n",
614 opt_recv->ccid3or_receive_rate);
622 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
624 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
626 hctx->ccid3hctx_s = 0;
627 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
628 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
630 hctx->ccid3hctx_no_feedback_timer.function = ccid3_hc_tx_no_feedback_timer;
631 hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
632 init_timer(&hctx->ccid3hctx_no_feedback_timer);
637 static void ccid3_hc_tx_exit(struct sock *sk)
639 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
641 BUG_ON(hctx == NULL);
643 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
644 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
646 /* Empty packet history */
647 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
651 * RX Half Connection methods
654 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
655 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
657 static char *ccid3_rx_state_names[] = {
658 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
659 [TFRC_RSTATE_DATA] = "DATA",
660 [TFRC_RSTATE_TERM] = "TERM",
663 return ccid3_rx_state_names[state];
667 static void ccid3_hc_rx_set_state(struct sock *sk,
668 enum ccid3_hc_rx_states state)
670 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
671 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
673 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
674 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
675 ccid3_rx_state_name(state));
676 WARN_ON(state == oldstate);
677 hcrx->ccid3hcrx_state = state;
680 static inline void ccid3_hc_rx_update_s(struct ccid3_hc_rx_sock *hcrx, int len)
682 if (unlikely(len == 0)) /* don't update on empty packets (e.g. ACKs) */
683 ccid3_pr_debug("Packet payload length is 0 - not updating\n");
685 hcrx->ccid3hcrx_s = hcrx->ccid3hcrx_s == 0 ? len :
686 (9 * hcrx->ccid3hcrx_s + len) / 10;
689 static void ccid3_hc_rx_send_feedback(struct sock *sk)
691 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
692 struct dccp_sock *dp = dccp_sk(sk);
693 struct dccp_rx_hist_entry *packet;
696 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
698 dccp_timestamp(sk, &now);
700 switch (hcrx->ccid3hcrx_state) {
701 case TFRC_RSTATE_NO_DATA:
702 hcrx->ccid3hcrx_x_recv = 0;
704 case TFRC_RSTATE_DATA: {
705 const u32 delta = timeval_delta(&now,
706 &hcrx->ccid3hcrx_tstamp_last_feedback);
707 hcrx->ccid3hcrx_x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv,
711 case TFRC_RSTATE_TERM:
712 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
716 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
717 if (unlikely(packet == NULL)) {
718 DCCP_WARN("%s, sk=%p, no data packet in history!\n",
723 hcrx->ccid3hcrx_tstamp_last_feedback = now;
724 hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
725 hcrx->ccid3hcrx_bytes_recv = 0;
727 /* Convert to multiples of 10us */
728 hcrx->ccid3hcrx_elapsed_time =
729 timeval_delta(&now, &packet->dccphrx_tstamp) / 10;
730 if (hcrx->ccid3hcrx_p == 0)
731 hcrx->ccid3hcrx_pinv = ~0;
733 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
734 dp->dccps_hc_rx_insert_options = 1;
738 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
740 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
743 BUG_ON(hcrx == NULL);
745 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
748 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
750 if (dccp_packet_without_ack(skb))
753 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
754 pinv = htonl(hcrx->ccid3hcrx_pinv);
756 if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
757 dccp_insert_option_elapsed_time(sk, skb,
758 hcrx->ccid3hcrx_elapsed_time)) ||
759 dccp_insert_option_timestamp(sk, skb) ||
760 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
761 &pinv, sizeof(pinv)) ||
762 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
763 &x_recv, sizeof(x_recv)))
769 /* calculate first loss interval
771 * returns estimated loss interval in usecs */
773 static u32 ccid3_hc_rx_calc_first_li(struct sock *sk)
775 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
776 struct dccp_rx_hist_entry *entry, *next, *tail = NULL;
777 u32 rtt, delta, x_recv, fval, p, tmp2;
778 struct timeval tstamp = { 0, };
784 list_for_each_entry_safe(entry, next, &hcrx->ccid3hcrx_hist,
786 if (dccp_rx_hist_entry_data_packet(entry)) {
791 tstamp = entry->dccphrx_tstamp;
792 win_count = entry->dccphrx_ccval;
796 interval = win_count - entry->dccphrx_ccval;
798 interval += TFRC_WIN_COUNT_LIMIT;
806 if (unlikely(step == 0)) {
807 DCCP_WARN("%s, sk=%p, packet history has no data packets!\n",
812 if (unlikely(interval == 0)) {
813 DCCP_WARN("%s, sk=%p, Could not find a win_count interval > 0."
814 "Defaulting to 1\n", dccp_role(sk), sk);
819 DCCP_CRIT("tail is null\n");
822 rtt = timeval_delta(&tstamp, &tail->dccphrx_tstamp) * 4 / interval;
823 ccid3_pr_debug("%s, sk=%p, approximated RTT to %uus\n",
824 dccp_role(sk), sk, rtt);
827 DCCP_WARN("RTT==0, setting to 1\n");
831 dccp_timestamp(sk, &tstamp);
832 delta = timeval_delta(&tstamp, &hcrx->ccid3hcrx_tstamp_last_feedback);
833 x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv, delta);
836 x_recv = hcrx->ccid3hcrx_x_recv;
838 tmp1 = (u64)x_recv * (u64)rtt;
839 do_div(tmp1,10000000);
843 DCCP_CRIT("tmp2 = 0, x_recv = %u, rtt =%u\n", x_recv, rtt);
847 fval = (hcrx->ccid3hcrx_s * 100000) / tmp2;
848 /* do not alter order above or you will get overflow on 32 bit */
849 p = tfrc_calc_x_reverse_lookup(fval);
850 ccid3_pr_debug("%s, sk=%p, receive rate=%u bytes/s, implied "
851 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
859 static void ccid3_hc_rx_update_li(struct sock *sk, u64 seq_loss, u8 win_loss)
861 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
862 struct dccp_li_hist_entry *head;
865 if (list_empty(&hcrx->ccid3hcrx_li_hist)) {
866 if (!dccp_li_hist_interval_new(ccid3_li_hist,
867 &hcrx->ccid3hcrx_li_hist, seq_loss, win_loss))
870 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
871 struct dccp_li_hist_entry, dccplih_node);
872 head->dccplih_interval = ccid3_hc_rx_calc_first_li(sk);
874 struct dccp_li_hist_entry *entry;
875 struct list_head *tail;
877 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
878 struct dccp_li_hist_entry, dccplih_node);
879 /* FIXME win count check removed as was wrong */
880 /* should make this check with receive history */
881 /* and compare there as per section 10.2 of RFC4342 */
883 /* new loss event detected */
884 /* calculate last interval length */
885 seq_temp = dccp_delta_seqno(head->dccplih_seqno, seq_loss);
886 entry = dccp_li_hist_entry_new(ccid3_li_hist, SLAB_ATOMIC);
889 DCCP_BUG("out of memory - can not allocate entry");
893 list_add(&entry->dccplih_node, &hcrx->ccid3hcrx_li_hist);
895 tail = hcrx->ccid3hcrx_li_hist.prev;
897 kmem_cache_free(ccid3_li_hist->dccplih_slab, tail);
899 /* Create the newest interval */
900 entry->dccplih_seqno = seq_loss;
901 entry->dccplih_interval = seq_temp;
902 entry->dccplih_win_count = win_loss;
906 static int ccid3_hc_rx_detect_loss(struct sock *sk,
907 struct dccp_rx_hist_entry *packet)
909 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
910 struct dccp_rx_hist_entry *rx_hist = dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
911 u64 seqno = packet->dccphrx_seqno;
917 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
920 follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
921 hcrx->ccid3hcrx_seqno_nonloss = seqno;
922 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
927 while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
928 > TFRC_RECV_NUM_LATE_LOSS) {
930 ccid3_hc_rx_update_li(sk, hcrx->ccid3hcrx_seqno_nonloss,
931 hcrx->ccid3hcrx_ccval_nonloss);
932 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
933 dccp_inc_seqno(&tmp_seqno);
934 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
935 dccp_inc_seqno(&tmp_seqno);
936 while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
937 tmp_seqno, &ccval)) {
938 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
939 hcrx->ccid3hcrx_ccval_nonloss = ccval;
940 dccp_inc_seqno(&tmp_seqno);
944 /* FIXME - this code could be simplified with above while */
945 /* but works at moment */
946 if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
947 hcrx->ccid3hcrx_seqno_nonloss = seqno;
948 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
952 dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
953 &hcrx->ccid3hcrx_li_hist, packet,
954 hcrx->ccid3hcrx_seqno_nonloss);
958 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
960 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
961 const struct dccp_options_received *opt_recv;
962 struct dccp_rx_hist_entry *packet;
964 u32 p_prev, rtt_prev, r_sample, t_elapsed;
965 int loss, payload_size;
967 BUG_ON(hcrx == NULL);
969 opt_recv = &dccp_sk(sk)->dccps_options_received;
971 switch (DCCP_SKB_CB(skb)->dccpd_type) {
973 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
975 case DCCP_PKT_DATAACK:
976 if (opt_recv->dccpor_timestamp_echo == 0)
978 rtt_prev = hcrx->ccid3hcrx_rtt;
979 dccp_timestamp(sk, &now);
980 timeval_sub_usecs(&now, opt_recv->dccpor_timestamp_echo * 10);
981 r_sample = timeval_usecs(&now);
982 t_elapsed = opt_recv->dccpor_elapsed_time * 10;
984 if (unlikely(r_sample <= t_elapsed))
985 DCCP_WARN("r_sample=%uus, t_elapsed=%uus\n",
986 r_sample, t_elapsed);
988 r_sample -= t_elapsed;
990 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
991 hcrx->ccid3hcrx_rtt = r_sample;
993 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
996 if (rtt_prev != hcrx->ccid3hcrx_rtt)
997 ccid3_pr_debug("%s, New RTT=%uus, elapsed time=%u\n",
998 dccp_role(sk), hcrx->ccid3hcrx_rtt,
999 opt_recv->dccpor_elapsed_time);
1003 default: /* We're not interested in other packet types, move along */
1007 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, sk, opt_recv->dccpor_ndp,
1009 if (unlikely(packet == NULL)) {
1010 DCCP_WARN("%s, sk=%p, Not enough mem to add rx packet "
1011 "to history, consider it lost!\n", dccp_role(sk), sk);
1015 loss = ccid3_hc_rx_detect_loss(sk, packet);
1017 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
1020 payload_size = skb->len - dccp_hdr(skb)->dccph_doff * 4;
1021 ccid3_hc_rx_update_s(hcrx, payload_size);
1023 switch (hcrx->ccid3hcrx_state) {
1024 case TFRC_RSTATE_NO_DATA:
1025 ccid3_pr_debug("%s, sk=%p(%s), skb=%p, sending initial "
1028 dccp_state_name(sk->sk_state), skb);
1029 ccid3_hc_rx_send_feedback(sk);
1030 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
1032 case TFRC_RSTATE_DATA:
1033 hcrx->ccid3hcrx_bytes_recv += payload_size;
1037 dccp_timestamp(sk, &now);
1038 if (timeval_delta(&now, &hcrx->ccid3hcrx_tstamp_last_ack) >=
1039 hcrx->ccid3hcrx_rtt) {
1040 hcrx->ccid3hcrx_tstamp_last_ack = now;
1041 ccid3_hc_rx_send_feedback(sk);
1044 case TFRC_RSTATE_TERM:
1045 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
1049 /* Dealing with packet loss */
1050 ccid3_pr_debug("%s, sk=%p(%s), data loss! Reacting...\n",
1051 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
1053 p_prev = hcrx->ccid3hcrx_p;
1055 /* Calculate loss event rate */
1056 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
1057 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
1059 /* Scaling up by 1000000 as fixed decimal */
1061 hcrx->ccid3hcrx_p = 1000000 / i_mean;
1063 DCCP_BUG("empty loss history");
1065 if (hcrx->ccid3hcrx_p > p_prev) {
1066 ccid3_hc_rx_send_feedback(sk);
1071 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
1073 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
1075 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
1077 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
1078 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
1079 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
1080 dccp_timestamp(sk, &hcrx->ccid3hcrx_tstamp_last_ack);
1081 hcrx->ccid3hcrx_tstamp_last_feedback = hcrx->ccid3hcrx_tstamp_last_ack;
1082 hcrx->ccid3hcrx_s = 0;
1083 hcrx->ccid3hcrx_rtt = 5000; /* XXX 5ms for now... */
1087 static void ccid3_hc_rx_exit(struct sock *sk)
1089 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1091 BUG_ON(hcrx == NULL);
1093 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
1095 /* Empty packet history */
1096 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
1098 /* Empty loss interval history */
1099 dccp_li_hist_purge(ccid3_li_hist, &hcrx->ccid3hcrx_li_hist);
1102 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
1104 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1106 /* Listen socks doesn't have a private CCID block */
1107 if (sk->sk_state == DCCP_LISTEN)
1110 BUG_ON(hcrx == NULL);
1112 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
1113 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1114 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
1117 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
1119 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1121 /* Listen socks doesn't have a private CCID block */
1122 if (sk->sk_state == DCCP_LISTEN)
1125 BUG_ON(hctx == NULL);
1127 info->tcpi_rto = hctx->ccid3hctx_t_rto;
1128 info->tcpi_rtt = hctx->ccid3hctx_rtt;
1131 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
1132 u32 __user *optval, int __user *optlen)
1134 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1137 /* Listen socks doesn't have a private CCID block */
1138 if (sk->sk_state == DCCP_LISTEN)
1142 case DCCP_SOCKOPT_CCID_RX_INFO:
1143 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
1145 len = sizeof(hcrx->ccid3hcrx_tfrc);
1146 val = &hcrx->ccid3hcrx_tfrc;
1149 return -ENOPROTOOPT;
1152 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1158 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
1159 u32 __user *optval, int __user *optlen)
1161 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1164 /* Listen socks doesn't have a private CCID block */
1165 if (sk->sk_state == DCCP_LISTEN)
1169 case DCCP_SOCKOPT_CCID_TX_INFO:
1170 if (len < sizeof(hctx->ccid3hctx_tfrc))
1172 len = sizeof(hctx->ccid3hctx_tfrc);
1173 val = &hctx->ccid3hctx_tfrc;
1176 return -ENOPROTOOPT;
1179 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1185 static struct ccid_operations ccid3 = {
1186 .ccid_id = DCCPC_CCID3,
1187 .ccid_name = "ccid3",
1188 .ccid_owner = THIS_MODULE,
1189 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
1190 .ccid_hc_tx_init = ccid3_hc_tx_init,
1191 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1192 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1193 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1194 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1195 .ccid_hc_tx_insert_options = ccid3_hc_tx_insert_options,
1196 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1197 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
1198 .ccid_hc_rx_init = ccid3_hc_rx_init,
1199 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1200 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1201 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1202 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1203 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1204 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1205 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1208 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
1209 module_param(ccid3_debug, int, 0444);
1210 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1213 static __init int ccid3_module_init(void)
1217 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1218 if (ccid3_rx_hist == NULL)
1221 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1222 if (ccid3_tx_hist == NULL)
1225 ccid3_li_hist = dccp_li_hist_new("ccid3");
1226 if (ccid3_li_hist == NULL)
1229 rc = ccid_register(&ccid3);
1231 goto out_free_loss_interval_history;
1235 out_free_loss_interval_history:
1236 dccp_li_hist_delete(ccid3_li_hist);
1237 ccid3_li_hist = NULL;
1239 dccp_tx_hist_delete(ccid3_tx_hist);
1240 ccid3_tx_hist = NULL;
1242 dccp_rx_hist_delete(ccid3_rx_hist);
1243 ccid3_rx_hist = NULL;
1246 module_init(ccid3_module_init);
1248 static __exit void ccid3_module_exit(void)
1250 ccid_unregister(&ccid3);
1252 if (ccid3_tx_hist != NULL) {
1253 dccp_tx_hist_delete(ccid3_tx_hist);
1254 ccid3_tx_hist = NULL;
1256 if (ccid3_rx_hist != NULL) {
1257 dccp_rx_hist_delete(ccid3_rx_hist);
1258 ccid3_rx_hist = NULL;
1260 if (ccid3_li_hist != NULL) {
1261 dccp_li_hist_delete(ccid3_li_hist);
1262 ccid3_li_hist = NULL;
1265 module_exit(ccid3_module_exit);
1267 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
1268 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1269 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1270 MODULE_LICENSE("GPL");
1271 MODULE_ALIAS("net-dccp-ccid-3");