2 * net/dccp/ccids/ccid3.c
4 * Copyright (c) 2005 The University of Waikato, Hamilton, New Zealand.
5 * Copyright (c) 2005 Ian McDonald <iam4@cs.waikato.ac.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.
37 #include <linux/config.h>
40 #include "lib/packet_history.h"
41 #include "lib/loss_interval.h"
46 * Reason for maths here is to avoid 32 bit overflow when a is big.
47 * With this we get close to the limit.
49 static inline u32 usecs_div(const u32 a, const u32 b)
51 const u32 div = a < (UINT_MAX / (USEC_PER_SEC / 10)) ? 10 :
52 a < (UINT_MAX / (USEC_PER_SEC / 50)) ? 50 :
53 a < (UINT_MAX / (USEC_PER_SEC / 100)) ? 100 :
54 a < (UINT_MAX / (USEC_PER_SEC / 500)) ? 500 :
55 a < (UINT_MAX / (USEC_PER_SEC / 1000)) ? 1000 :
56 a < (UINT_MAX / (USEC_PER_SEC / 5000)) ? 5000 :
57 a < (UINT_MAX / (USEC_PER_SEC / 10000)) ? 10000 :
58 a < (UINT_MAX / (USEC_PER_SEC / 50000)) ? 50000 :
60 const u32 tmp = a * (USEC_PER_SEC / div);
61 return (b >= 2 * div) ? tmp / (b / div) : tmp;
64 static int ccid3_debug;
67 #define ccid3_pr_debug(format, a...) \
68 do { if (ccid3_debug) \
69 printk(KERN_DEBUG "%s: " format, __FUNCTION__, ##a); \
72 #define ccid3_pr_debug(format, a...)
75 static struct dccp_tx_hist *ccid3_tx_hist;
76 static struct dccp_rx_hist *ccid3_rx_hist;
77 static struct dccp_li_hist *ccid3_li_hist;
79 static int ccid3_init(struct sock *sk)
81 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
85 static void ccid3_exit(struct sock *sk)
87 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
90 /* TFRC sender states */
91 enum ccid3_hc_tx_states {
92 TFRC_SSTATE_NO_SENT = 1,
99 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
101 static char *ccid3_state_names[] = {
102 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
103 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
104 [TFRC_SSTATE_FBACK] = "FBACK",
105 [TFRC_SSTATE_TERM] = "TERM",
108 return ccid3_state_names[state];
112 static inline void ccid3_hc_tx_set_state(struct sock *sk,
113 enum ccid3_hc_tx_states state)
115 struct dccp_sock *dp = dccp_sk(sk);
116 struct ccid3_hc_tx_sock *hctx = dp->dccps_hc_tx_ccid_private;
117 enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
119 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
120 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
121 ccid3_tx_state_name(state));
122 WARN_ON(state == oldstate);
123 hctx->ccid3hctx_state = state;
126 /* Calculate new t_ipi (inter packet interval) by t_ipi = s / X_inst */
127 static inline void ccid3_calc_new_t_ipi(struct ccid3_hc_tx_sock *hctx)
130 * If no feedback spec says t_ipi is 1 second (set elsewhere and then
131 * doubles after every no feedback timer (separate function)
133 if (hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
134 hctx->ccid3hctx_t_ipi = usecs_div(hctx->ccid3hctx_s,
138 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
139 static inline void ccid3_calc_new_delta(struct ccid3_hc_tx_sock *hctx)
141 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
142 TFRC_OPSYS_HALF_TIME_GRAN);
148 * x_calc = calcX(s, R, p);
149 * X = max(min(X_calc, 2 * X_recv), s / t_mbi);
151 * If (now - tld >= R)
152 * X = max(min(2 * X, 2 * X_recv), s / R);
155 static void ccid3_hc_tx_update_x(struct sock *sk)
157 struct dccp_sock *dp = dccp_sk(sk);
158 struct ccid3_hc_tx_sock *hctx = dp->dccps_hc_tx_ccid_private;
160 /* To avoid large error in calcX */
161 if (hctx->ccid3hctx_p >= TFRC_SMALLEST_P) {
162 hctx->ccid3hctx_x_calc = tfrc_calc_x(hctx->ccid3hctx_s,
165 hctx->ccid3hctx_x = max_t(u32, min_t(u32, hctx->ccid3hctx_x_calc,
166 2 * hctx->ccid3hctx_x_recv),
168 TFRC_MAX_BACK_OFF_TIME));
172 do_gettimeofday(&now);
173 if (timeval_delta(&now, &hctx->ccid3hctx_t_ld) >=
174 hctx->ccid3hctx_rtt) {
175 hctx->ccid3hctx_x = max_t(u32, min_t(u32, hctx->ccid3hctx_x_recv,
176 hctx->ccid3hctx_x) * 2,
177 usecs_div(hctx->ccid3hctx_s,
178 hctx->ccid3hctx_rtt));
179 hctx->ccid3hctx_t_ld = now;
184 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
186 struct sock *sk = (struct sock *)data;
187 struct dccp_sock *dp = dccp_sk(sk);
188 unsigned long next_tmout = 0;
189 struct ccid3_hc_tx_sock *hctx = dp->dccps_hc_tx_ccid_private;
192 if (sock_owned_by_user(sk)) {
193 /* Try again later. */
194 /* XXX: set some sensible MIB */
195 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
200 ccid3_pr_debug("%s, sk=%p, state=%s\n", dccp_role(sk), sk,
201 ccid3_tx_state_name(hctx->ccid3hctx_state));
203 switch (hctx->ccid3hctx_state) {
204 case TFRC_SSTATE_TERM:
206 case TFRC_SSTATE_NO_FBACK:
207 /* Halve send rate */
208 hctx->ccid3hctx_x /= 2;
209 if (hctx->ccid3hctx_x < (hctx->ccid3hctx_s /
210 TFRC_MAX_BACK_OFF_TIME))
211 hctx->ccid3hctx_x = (hctx->ccid3hctx_s /
212 TFRC_MAX_BACK_OFF_TIME);
214 ccid3_pr_debug("%s, sk=%p, state=%s, updated tx rate to %d "
217 ccid3_tx_state_name(hctx->ccid3hctx_state),
219 next_tmout = max_t(u32, 2 * usecs_div(hctx->ccid3hctx_s,
221 TFRC_INITIAL_TIMEOUT);
223 * FIXME - not sure above calculation is correct. See section
224 * 5 of CCID3 11 should adjust tx_t_ipi and double that to
228 case TFRC_SSTATE_FBACK:
230 * Check if IDLE since last timeout and recv rate is less than
233 if (!hctx->ccid3hctx_idle ||
234 (hctx->ccid3hctx_x_recv >=
235 4 * usecs_div(hctx->ccid3hctx_s, hctx->ccid3hctx_rtt))) {
236 ccid3_pr_debug("%s, sk=%p, state=%s, not idle\n",
238 ccid3_tx_state_name(hctx->ccid3hctx_state));
239 /* Halve sending rate */
241 /* If (X_calc > 2 * X_recv)
242 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
244 * X_recv = X_calc / 4;
246 BUG_ON(hctx->ccid3hctx_p >= TFRC_SMALLEST_P &&
247 hctx->ccid3hctx_x_calc == 0);
249 /* check also if p is zero -> x_calc is infinity? */
250 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P ||
251 hctx->ccid3hctx_x_calc > 2 * hctx->ccid3hctx_x_recv)
252 hctx->ccid3hctx_x_recv = max_t(u32, hctx->ccid3hctx_x_recv / 2,
253 hctx->ccid3hctx_s / (2 * TFRC_MAX_BACK_OFF_TIME));
255 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc / 4;
257 /* Update sending rate */
258 ccid3_hc_tx_update_x(sk);
261 * Schedule no feedback timer to expire in
262 * max(4 * R, 2 * s / X)
264 next_tmout = max_t(u32, hctx->ccid3hctx_t_rto,
265 2 * usecs_div(hctx->ccid3hctx_s,
269 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
270 __FUNCTION__, dccp_role(sk), sk, hctx->ccid3hctx_state);
275 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
276 jiffies + max_t(u32, 1, usecs_to_jiffies(next_tmout)));
277 hctx->ccid3hctx_idle = 1;
283 static int ccid3_hc_tx_send_packet(struct sock *sk,
284 struct sk_buff *skb, int len)
286 struct dccp_sock *dp = dccp_sk(sk);
287 struct ccid3_hc_tx_sock *hctx = dp->dccps_hc_tx_ccid_private;
288 struct dccp_tx_hist_entry *new_packet;
293 /* Check if pure ACK or Terminating*/
296 * XXX: We only call this function for DATA and DATAACK, on, these
297 * packets can have zero length, but why the comment about "pure ACK"?
299 if (hctx == NULL || len == 0 ||
300 hctx->ccid3hctx_state == TFRC_SSTATE_TERM)
303 /* See if last packet allocated was not sent */
304 new_packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
305 if (new_packet == NULL || new_packet->dccphtx_sent) {
306 new_packet = dccp_tx_hist_entry_new(ccid3_tx_hist,
310 if (new_packet == NULL) {
311 ccid3_pr_debug("%s, sk=%p, not enough mem to add "
312 "to history, send refused\n",
317 dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, new_packet);
320 do_gettimeofday(&now);
322 switch (hctx->ccid3hctx_state) {
323 case TFRC_SSTATE_NO_SENT:
324 ccid3_pr_debug("%s, sk=%p, first packet(%llu)\n",
325 dccp_role(sk), sk, dp->dccps_gss);
327 hctx->ccid3hctx_no_feedback_timer.function = ccid3_hc_tx_no_feedback_timer;
328 hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
329 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
330 jiffies + usecs_to_jiffies(TFRC_INITIAL_TIMEOUT));
331 hctx->ccid3hctx_last_win_count = 0;
332 hctx->ccid3hctx_t_last_win_count = now;
333 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
334 hctx->ccid3hctx_t_ipi = TFRC_INITIAL_TIMEOUT;
336 /* Set nominal send time for initial packet */
337 hctx->ccid3hctx_t_nom = now;
338 timeval_add_usecs(&hctx->ccid3hctx_t_nom,
339 hctx->ccid3hctx_t_ipi);
340 ccid3_calc_new_delta(hctx);
343 case TFRC_SSTATE_NO_FBACK:
344 case TFRC_SSTATE_FBACK:
345 delay = (timeval_delta(&now, &hctx->ccid3hctx_t_nom) -
346 hctx->ccid3hctx_delta);
347 ccid3_pr_debug("send_packet delay=%ld\n", delay);
349 /* divide by -1000 is to convert to ms and get sign right */
350 rc = delay > 0 ? delay : 0;
353 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
354 __FUNCTION__, dccp_role(sk), sk, hctx->ccid3hctx_state);
360 /* Can we send? if so add options and add to packet history */
362 dp->dccps_hc_tx_insert_options = 1;
363 new_packet->dccphtx_ccval =
364 DCCP_SKB_CB(skb)->dccpd_ccval =
365 hctx->ccid3hctx_last_win_count;
371 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more, int len)
373 struct dccp_sock *dp = dccp_sk(sk);
374 struct ccid3_hc_tx_sock *hctx = dp->dccps_hc_tx_ccid_private;
377 BUG_ON(hctx == NULL);
379 if (hctx->ccid3hctx_state == TFRC_SSTATE_TERM) {
380 ccid3_pr_debug("%s, sk=%p, while state is TFRC_SSTATE_TERM!\n",
385 do_gettimeofday(&now);
387 /* check if we have sent a data packet */
389 unsigned long quarter_rtt;
390 struct dccp_tx_hist_entry *packet;
392 packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
393 if (packet == NULL) {
394 printk(KERN_CRIT "%s: packet doesn't exists in "
395 "history!\n", __FUNCTION__);
398 if (packet->dccphtx_sent) {
399 printk(KERN_CRIT "%s: no unsent packet in history!\n",
403 packet->dccphtx_tstamp = now;
404 packet->dccphtx_seqno = dp->dccps_gss;
406 * Check if win_count have changed
407 * Algorithm in "8.1. Window Counter Valuer" in
408 * draft-ietf-dccp-ccid3-11.txt
410 quarter_rtt = timeval_delta(&now, &hctx->ccid3hctx_t_last_win_count);
411 if (likely(hctx->ccid3hctx_rtt > 8))
412 quarter_rtt /= hctx->ccid3hctx_rtt / 4;
414 if (quarter_rtt > 0) {
415 hctx->ccid3hctx_t_last_win_count = now;
416 hctx->ccid3hctx_last_win_count = (hctx->ccid3hctx_last_win_count +
417 min_t(unsigned long, quarter_rtt, 5)) % 16;
418 ccid3_pr_debug("%s, sk=%p, window changed from "
421 packet->dccphtx_ccval,
422 hctx->ccid3hctx_last_win_count);
425 hctx->ccid3hctx_idle = 0;
426 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
427 packet->dccphtx_sent = 1;
429 ccid3_pr_debug("%s, sk=%p, seqno=%llu NOT inserted!\n",
430 dccp_role(sk), sk, dp->dccps_gss);
432 switch (hctx->ccid3hctx_state) {
433 case TFRC_SSTATE_NO_SENT:
434 /* if first wasn't pure ack */
436 printk(KERN_CRIT "%s: %s, First packet sent is noted "
437 "as a data packet\n",
438 __FUNCTION__, dccp_role(sk));
440 case TFRC_SSTATE_NO_FBACK:
441 case TFRC_SSTATE_FBACK:
443 hctx->ccid3hctx_t_nom = now;
444 ccid3_calc_new_t_ipi(hctx);
445 ccid3_calc_new_delta(hctx);
446 timeval_add_usecs(&hctx->ccid3hctx_t_nom,
447 hctx->ccid3hctx_t_ipi);
451 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
452 __FUNCTION__, dccp_role(sk), sk, hctx->ccid3hctx_state);
458 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
460 struct dccp_sock *dp = dccp_sk(sk);
461 struct ccid3_hc_tx_sock *hctx = dp->dccps_hc_tx_ccid_private;
462 struct ccid3_options_received *opt_recv;
463 struct dccp_tx_hist_entry *packet;
464 unsigned long next_tmout;
473 if (hctx->ccid3hctx_state == TFRC_SSTATE_TERM) {
474 ccid3_pr_debug("%s, sk=%p, received a packet when "
475 "terminating!\n", dccp_role(sk), sk);
479 /* we are only interested in ACKs */
480 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
481 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
484 opt_recv = &hctx->ccid3hctx_options_received;
486 t_elapsed = dp->dccps_options_received.dccpor_elapsed_time * 10;
487 x_recv = opt_recv->ccid3or_receive_rate;
488 pinv = opt_recv->ccid3or_loss_event_rate;
490 switch (hctx->ccid3hctx_state) {
491 case TFRC_SSTATE_NO_SENT:
492 /* FIXME: what to do here? */
494 case TFRC_SSTATE_NO_FBACK:
495 case TFRC_SSTATE_FBACK:
496 /* Calculate new round trip sample by
497 * R_sample = (now - t_recvdata) - t_delay */
498 /* get t_recvdata from history */
499 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
500 DCCP_SKB_CB(skb)->dccpd_ack_seq);
501 if (packet == NULL) {
502 ccid3_pr_debug("%s, sk=%p, seqno %llu(%s) does't "
503 "exist in history!\n",
505 DCCP_SKB_CB(skb)->dccpd_ack_seq,
506 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
511 r_sample = timeval_now_delta(&packet->dccphtx_tstamp);
512 if (unlikely(r_sample <= t_elapsed))
513 LIMIT_NETDEBUG(KERN_WARNING
514 "%s: r_sample=%uus, t_elapsed=%uus\n",
515 __FUNCTION__, r_sample, t_elapsed);
517 r_sample -= t_elapsed;
519 /* Update RTT estimate by
520 * If (No feedback recv)
523 * R = q * R + (1 - q) * R_sample;
525 * q is a constant, RFC 3448 recomments 0.9
527 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
528 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
529 hctx->ccid3hctx_rtt = r_sample;
531 hctx->ccid3hctx_rtt = (hctx->ccid3hctx_rtt * 9) / 10 +
534 ccid3_pr_debug("%s, sk=%p, New RTT estimate=%uus, "
535 "r_sample=%us\n", dccp_role(sk), sk,
536 hctx->ccid3hctx_rtt, r_sample);
538 /* Update timeout interval */
539 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
542 /* Update receive rate */
543 hctx->ccid3hctx_x_recv = x_recv;/* X_recv in bytes per sec */
545 /* Update loss event rate */
546 if (pinv == ~0 || pinv == 0)
547 hctx->ccid3hctx_p = 0;
549 hctx->ccid3hctx_p = 1000000 / pinv;
551 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P) {
552 hctx->ccid3hctx_p = TFRC_SMALLEST_P;
553 ccid3_pr_debug("%s, sk=%p, Smallest p used!\n",
558 /* unschedule no feedback timer */
559 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
561 /* Update sending rate */
562 ccid3_hc_tx_update_x(sk);
564 /* Update next send time */
565 timeval_sub_usecs(&hctx->ccid3hctx_t_nom,
566 hctx->ccid3hctx_t_ipi);
567 ccid3_calc_new_t_ipi(hctx);
568 timeval_add_usecs(&hctx->ccid3hctx_t_nom,
569 hctx->ccid3hctx_t_ipi);
570 ccid3_calc_new_delta(hctx);
572 /* remove all packets older than the one acked from history */
573 dccp_tx_hist_purge_older(ccid3_tx_hist,
574 &hctx->ccid3hctx_hist, packet);
576 * As we have calculated new ipi, delta, t_nom it is possible that
577 * we now can send a packet, so wake up dccp_wait_for_ccids.
579 sk->sk_write_space(sk);
582 * Schedule no feedback timer to expire in
583 * max(4 * R, 2 * s / X)
585 next_tmout = max(hctx->ccid3hctx_t_rto,
586 2 * usecs_div(hctx->ccid3hctx_s,
589 ccid3_pr_debug("%s, sk=%p, Scheduled no feedback timer to "
590 "expire in %lu jiffies (%luus)\n",
592 usecs_to_jiffies(next_tmout), next_tmout);
594 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
595 jiffies + max_t(u32, 1, usecs_to_jiffies(next_tmout)));
598 hctx->ccid3hctx_idle = 1;
601 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
602 __FUNCTION__, dccp_role(sk), sk, hctx->ccid3hctx_state);
608 static void ccid3_hc_tx_insert_options(struct sock *sk, struct sk_buff *skb)
610 const struct dccp_sock *dp = dccp_sk(sk);
611 struct ccid3_hc_tx_sock *hctx = dp->dccps_hc_tx_ccid_private;
613 if (hctx == NULL || !(sk->sk_state == DCCP_OPEN ||
614 sk->sk_state == DCCP_PARTOPEN))
617 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
620 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
621 unsigned char len, u16 idx,
622 unsigned char *value)
625 struct dccp_sock *dp = dccp_sk(sk);
626 struct ccid3_hc_tx_sock *hctx = dp->dccps_hc_tx_ccid_private;
627 struct ccid3_options_received *opt_recv;
632 opt_recv = &hctx->ccid3hctx_options_received;
634 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
635 opt_recv->ccid3or_seqno = dp->dccps_gsr;
636 opt_recv->ccid3or_loss_event_rate = ~0;
637 opt_recv->ccid3or_loss_intervals_idx = 0;
638 opt_recv->ccid3or_loss_intervals_len = 0;
639 opt_recv->ccid3or_receive_rate = 0;
643 case TFRC_OPT_LOSS_EVENT_RATE:
645 ccid3_pr_debug("%s, sk=%p, invalid len for "
646 "TFRC_OPT_LOSS_EVENT_RATE\n",
650 opt_recv->ccid3or_loss_event_rate = ntohl(*(u32 *)value);
651 ccid3_pr_debug("%s, sk=%p, LOSS_EVENT_RATE=%u\n",
653 opt_recv->ccid3or_loss_event_rate);
656 case TFRC_OPT_LOSS_INTERVALS:
657 opt_recv->ccid3or_loss_intervals_idx = idx;
658 opt_recv->ccid3or_loss_intervals_len = len;
659 ccid3_pr_debug("%s, sk=%p, LOSS_INTERVALS=(%u, %u)\n",
661 opt_recv->ccid3or_loss_intervals_idx,
662 opt_recv->ccid3or_loss_intervals_len);
664 case TFRC_OPT_RECEIVE_RATE:
666 ccid3_pr_debug("%s, sk=%p, invalid len for "
667 "TFRC_OPT_RECEIVE_RATE\n",
671 opt_recv->ccid3or_receive_rate = ntohl(*(u32 *)value);
672 ccid3_pr_debug("%s, sk=%p, RECEIVE_RATE=%u\n",
674 opt_recv->ccid3or_receive_rate);
682 static int ccid3_hc_tx_init(struct sock *sk)
684 struct dccp_sock *dp = dccp_sk(sk);
685 struct ccid3_hc_tx_sock *hctx;
687 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
689 hctx = dp->dccps_hc_tx_ccid_private = kmalloc(sizeof(*hctx),
694 memset(hctx, 0, sizeof(*hctx));
696 if (dp->dccps_packet_size >= TFRC_MIN_PACKET_SIZE &&
697 dp->dccps_packet_size <= TFRC_MAX_PACKET_SIZE)
698 hctx->ccid3hctx_s = dp->dccps_packet_size;
700 hctx->ccid3hctx_s = TFRC_STD_PACKET_SIZE;
702 /* Set transmission rate to 1 packet per second */
703 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
704 hctx->ccid3hctx_t_rto = USEC_PER_SEC;
705 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
706 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
707 init_timer(&hctx->ccid3hctx_no_feedback_timer);
712 static void ccid3_hc_tx_exit(struct sock *sk)
714 struct dccp_sock *dp = dccp_sk(sk);
715 struct ccid3_hc_tx_sock *hctx = dp->dccps_hc_tx_ccid_private;
717 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
718 BUG_ON(hctx == NULL);
720 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
721 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
723 /* Empty packet history */
724 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
726 kfree(dp->dccps_hc_tx_ccid_private);
727 dp->dccps_hc_tx_ccid_private = NULL;
731 * RX Half Connection methods
734 /* TFRC receiver states */
735 enum ccid3_hc_rx_states {
736 TFRC_RSTATE_NO_DATA = 1,
738 TFRC_RSTATE_TERM = 127,
742 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
744 static char *ccid3_rx_state_names[] = {
745 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
746 [TFRC_RSTATE_DATA] = "DATA",
747 [TFRC_RSTATE_TERM] = "TERM",
750 return ccid3_rx_state_names[state];
754 static inline void ccid3_hc_rx_set_state(struct sock *sk,
755 enum ccid3_hc_rx_states state)
757 struct dccp_sock *dp = dccp_sk(sk);
758 struct ccid3_hc_rx_sock *hcrx = dp->dccps_hc_rx_ccid_private;
759 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
761 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
762 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
763 ccid3_rx_state_name(state));
764 WARN_ON(state == oldstate);
765 hcrx->ccid3hcrx_state = state;
768 static void ccid3_hc_rx_send_feedback(struct sock *sk)
770 struct dccp_sock *dp = dccp_sk(sk);
771 struct ccid3_hc_rx_sock *hcrx = dp->dccps_hc_rx_ccid_private;
772 struct dccp_rx_hist_entry *packet;
775 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
777 do_gettimeofday(&now);
779 switch (hcrx->ccid3hcrx_state) {
780 case TFRC_RSTATE_NO_DATA:
781 hcrx->ccid3hcrx_x_recv = 0;
783 case TFRC_RSTATE_DATA: {
784 const u32 delta = timeval_delta(&now,
785 &hcrx->ccid3hcrx_tstamp_last_feedback);
786 hcrx->ccid3hcrx_x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv,
791 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
792 __FUNCTION__, dccp_role(sk), sk, hcrx->ccid3hcrx_state);
797 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
798 if (packet == NULL) {
799 printk(KERN_CRIT "%s: %s, sk=%p, no data packet in history!\n",
800 __FUNCTION__, dccp_role(sk), sk);
805 hcrx->ccid3hcrx_tstamp_last_feedback = now;
806 hcrx->ccid3hcrx_last_counter = packet->dccphrx_ccval;
807 hcrx->ccid3hcrx_seqno_last_counter = packet->dccphrx_seqno;
808 hcrx->ccid3hcrx_bytes_recv = 0;
810 /* Convert to multiples of 10us */
811 hcrx->ccid3hcrx_elapsed_time =
812 timeval_delta(&now, &packet->dccphrx_tstamp) / 10;
813 if (hcrx->ccid3hcrx_p == 0)
814 hcrx->ccid3hcrx_pinv = ~0;
816 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
817 dp->dccps_hc_rx_insert_options = 1;
821 static void ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
823 const struct dccp_sock *dp = dccp_sk(sk);
825 struct ccid3_hc_rx_sock *hcrx = dp->dccps_hc_rx_ccid_private;
827 if (hcrx == NULL || !(sk->sk_state == DCCP_OPEN ||
828 sk->sk_state == DCCP_PARTOPEN))
831 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_last_counter;
833 if (dccp_packet_without_ack(skb))
836 if (hcrx->ccid3hcrx_elapsed_time != 0)
837 dccp_insert_option_elapsed_time(sk, skb,
838 hcrx->ccid3hcrx_elapsed_time);
839 dccp_insert_option_timestamp(sk, skb);
840 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
841 pinv = htonl(hcrx->ccid3hcrx_pinv);
842 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
843 &pinv, sizeof(pinv));
844 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
845 &x_recv, sizeof(x_recv));
848 /* calculate first loss interval
850 * returns estimated loss interval in usecs */
852 static u32 ccid3_hc_rx_calc_first_li(struct sock *sk)
854 struct dccp_sock *dp = dccp_sk(sk);
855 struct ccid3_hc_rx_sock *hcrx = dp->dccps_hc_rx_ccid_private;
856 struct dccp_rx_hist_entry *entry, *next, *tail = NULL;
857 u32 rtt, delta, x_recv, fval, p, tmp2;
858 struct timeval tstamp = { 0, };
864 list_for_each_entry_safe(entry, next, &hcrx->ccid3hcrx_hist,
866 if (dccp_rx_hist_entry_data_packet(entry)) {
871 tstamp = entry->dccphrx_tstamp;
872 win_count = entry->dccphrx_ccval;
876 interval = win_count - entry->dccphrx_ccval;
878 interval += TFRC_WIN_COUNT_LIMIT;
887 printk(KERN_CRIT "%s: %s, sk=%p, packet history contains no "
889 __FUNCTION__, dccp_role(sk), sk);
894 ccid3_pr_debug("%s, sk=%p, Could not find a win_count "
895 "interval > 0. Defaulting to 1\n",
900 rtt = timeval_delta(&tstamp, &tail->dccphrx_tstamp) * 4 / interval;
901 ccid3_pr_debug("%s, sk=%p, approximated RTT to %uus\n",
902 dccp_role(sk), sk, rtt);
906 delta = timeval_now_delta(&hcrx->ccid3hcrx_tstamp_last_feedback);
907 x_recv = hcrx->ccid3hcrx_bytes_recv * USEC_PER_SEC;
908 if (likely(delta > 1))
911 tmp1 = (u64)x_recv * (u64)rtt;
912 do_div(tmp1,10000000);
914 fval = (hcrx->ccid3hcrx_s * 100000) / tmp2;
915 /* do not alter order above or you will get overflow on 32 bit */
916 p = tfrc_calc_x_reverse_lookup(fval);
917 ccid3_pr_debug("%s, sk=%p, receive rate=%u bytes/s, implied "
918 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
926 static void ccid3_hc_rx_update_li(struct sock *sk, u64 seq_loss, u8 win_loss)
928 struct dccp_sock *dp = dccp_sk(sk);
929 struct ccid3_hc_rx_sock *hcrx = dp->dccps_hc_rx_ccid_private;
931 if (seq_loss != DCCP_MAX_SEQNO + 1 &&
932 list_empty(&hcrx->ccid3hcrx_li_hist)) {
933 struct dccp_li_hist_entry *li_tail;
935 li_tail = dccp_li_hist_interval_new(ccid3_li_hist,
936 &hcrx->ccid3hcrx_li_hist,
940 li_tail->dccplih_interval = ccid3_hc_rx_calc_first_li(sk);
942 /* FIXME: find end of interval */
945 static void ccid3_hc_rx_detect_loss(struct sock *sk)
947 struct dccp_sock *dp = dccp_sk(sk);
948 struct ccid3_hc_rx_sock *hcrx = dp->dccps_hc_rx_ccid_private;
950 const u64 seq_loss = dccp_rx_hist_detect_loss(&hcrx->ccid3hcrx_hist,
951 &hcrx->ccid3hcrx_li_hist,
954 ccid3_hc_rx_update_li(sk, seq_loss, win_loss);
957 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
959 struct dccp_sock *dp = dccp_sk(sk);
960 struct ccid3_hc_rx_sock *hcrx = dp->dccps_hc_rx_ccid_private;
961 const struct dccp_options_received *opt_recv;
962 struct dccp_rx_hist_entry *packet;
965 u32 p_prev, r_sample, t_elapsed;
971 BUG_ON(!(hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA ||
972 hcrx->ccid3hcrx_state == TFRC_RSTATE_DATA));
974 opt_recv = &dp->dccps_options_received;
976 switch (DCCP_SKB_CB(skb)->dccpd_type) {
978 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
980 case DCCP_PKT_DATAACK:
981 if (opt_recv->dccpor_timestamp_echo == 0)
983 p_prev = hcrx->ccid3hcrx_rtt;
984 do_gettimeofday(&now);
985 timeval_sub_usecs(&now, opt_recv->dccpor_timestamp_echo * 10);
986 r_sample = timeval_usecs(&now);
987 t_elapsed = opt_recv->dccpor_elapsed_time * 10;
989 if (unlikely(r_sample <= t_elapsed))
990 LIMIT_NETDEBUG(KERN_WARNING
991 "%s: r_sample=%uus, t_elapsed=%uus\n",
992 __FUNCTION__, r_sample, t_elapsed);
994 r_sample -= t_elapsed;
996 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
997 hcrx->ccid3hcrx_rtt = r_sample;
999 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
1002 if (p_prev != hcrx->ccid3hcrx_rtt)
1003 ccid3_pr_debug("%s, New RTT=%luus, elapsed time=%u\n",
1004 dccp_role(sk), hcrx->ccid3hcrx_rtt,
1005 opt_recv->dccpor_elapsed_time);
1010 ccid3_pr_debug("%s, sk=%p, not DATA/DATAACK/ACK packet(%s)\n",
1012 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
1016 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, opt_recv->dccpor_ndp,
1018 if (packet == NULL) {
1019 ccid3_pr_debug("%s, sk=%p, Not enough mem to add rx packet "
1020 "to history (consider it lost)!",
1025 win_count = packet->dccphrx_ccval;
1027 ins = dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
1028 &hcrx->ccid3hcrx_li_hist, packet);
1030 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
1033 switch (hcrx->ccid3hcrx_state) {
1034 case TFRC_RSTATE_NO_DATA:
1035 ccid3_pr_debug("%s, sk=%p(%s), skb=%p, sending initial "
1038 dccp_state_name(sk->sk_state), skb);
1039 ccid3_hc_rx_send_feedback(sk);
1040 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
1042 case TFRC_RSTATE_DATA:
1043 hcrx->ccid3hcrx_bytes_recv += skb->len -
1044 dccp_hdr(skb)->dccph_doff * 4;
1048 do_gettimeofday(&now);
1049 if (timeval_delta(&now, &hcrx->ccid3hcrx_tstamp_last_ack) >=
1050 hcrx->ccid3hcrx_rtt) {
1051 hcrx->ccid3hcrx_tstamp_last_ack = now;
1052 ccid3_hc_rx_send_feedback(sk);
1056 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
1057 __FUNCTION__, dccp_role(sk), sk, hcrx->ccid3hcrx_state);
1062 /* Dealing with packet loss */
1063 ccid3_pr_debug("%s, sk=%p(%s), data loss! Reacting...\n",
1064 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
1066 ccid3_hc_rx_detect_loss(sk);
1067 p_prev = hcrx->ccid3hcrx_p;
1069 /* Calculate loss event rate */
1070 if (!list_empty(&hcrx->ccid3hcrx_li_hist))
1071 /* Scaling up by 1000000 as fixed decimal */
1072 hcrx->ccid3hcrx_p = 1000000 / dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
1074 if (hcrx->ccid3hcrx_p > p_prev) {
1075 ccid3_hc_rx_send_feedback(sk);
1080 static int ccid3_hc_rx_init(struct sock *sk)
1082 struct dccp_sock *dp = dccp_sk(sk);
1083 struct ccid3_hc_rx_sock *hcrx;
1085 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
1087 hcrx = dp->dccps_hc_rx_ccid_private = kmalloc(sizeof(*hcrx),
1092 memset(hcrx, 0, sizeof(*hcrx));
1094 if (dp->dccps_packet_size >= TFRC_MIN_PACKET_SIZE &&
1095 dp->dccps_packet_size <= TFRC_MAX_PACKET_SIZE)
1096 hcrx->ccid3hcrx_s = dp->dccps_packet_size;
1098 hcrx->ccid3hcrx_s = TFRC_STD_PACKET_SIZE;
1100 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
1101 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
1102 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
1104 * XXX this seems to be paranoid, need to think more about this, for
1105 * now start with something different than zero. -acme
1107 hcrx->ccid3hcrx_rtt = USEC_PER_SEC / 5;
1111 static void ccid3_hc_rx_exit(struct sock *sk)
1113 struct dccp_sock *dp = dccp_sk(sk);
1114 struct ccid3_hc_rx_sock *hcrx = dp->dccps_hc_rx_ccid_private;
1116 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
1121 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
1123 /* Empty packet history */
1124 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
1126 /* Empty loss interval history */
1127 dccp_li_hist_purge(ccid3_li_hist, &hcrx->ccid3hcrx_li_hist);
1129 kfree(dp->dccps_hc_rx_ccid_private);
1130 dp->dccps_hc_rx_ccid_private = NULL;
1133 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
1135 const struct dccp_sock *dp = dccp_sk(sk);
1136 const struct ccid3_hc_rx_sock *hcrx = dp->dccps_hc_rx_ccid_private;
1141 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
1142 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1143 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
1146 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
1148 const struct dccp_sock *dp = dccp_sk(sk);
1149 const struct ccid3_hc_tx_sock *hctx = dp->dccps_hc_tx_ccid_private;
1154 info->tcpi_rto = hctx->ccid3hctx_t_rto;
1155 info->tcpi_rtt = hctx->ccid3hctx_rtt;
1158 static struct ccid ccid3 = {
1160 .ccid_name = "ccid3",
1161 .ccid_owner = THIS_MODULE,
1162 .ccid_init = ccid3_init,
1163 .ccid_exit = ccid3_exit,
1164 .ccid_hc_tx_init = ccid3_hc_tx_init,
1165 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1166 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1167 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1168 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1169 .ccid_hc_tx_insert_options = ccid3_hc_tx_insert_options,
1170 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1171 .ccid_hc_rx_init = ccid3_hc_rx_init,
1172 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1173 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1174 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1175 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1176 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1179 module_param(ccid3_debug, int, 0444);
1180 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1182 static __init int ccid3_module_init(void)
1186 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1187 if (ccid3_rx_hist == NULL)
1190 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1191 if (ccid3_tx_hist == NULL)
1194 ccid3_li_hist = dccp_li_hist_new("ccid3");
1195 if (ccid3_li_hist == NULL)
1198 rc = ccid_register(&ccid3);
1200 goto out_free_loss_interval_history;
1204 out_free_loss_interval_history:
1205 dccp_li_hist_delete(ccid3_li_hist);
1206 ccid3_li_hist = NULL;
1208 dccp_tx_hist_delete(ccid3_tx_hist);
1209 ccid3_tx_hist = NULL;
1211 dccp_rx_hist_delete(ccid3_rx_hist);
1212 ccid3_rx_hist = NULL;
1215 module_init(ccid3_module_init);
1217 static __exit void ccid3_module_exit(void)
1219 #ifdef CONFIG_IP_DCCP_UNLOAD_HACK
1221 * Hack to use while developing, so that we get rid of the control
1222 * sock, that is what keeps a refcount on dccp.ko -acme
1224 extern void dccp_ctl_sock_exit(void);
1226 dccp_ctl_sock_exit();
1228 ccid_unregister(&ccid3);
1230 if (ccid3_tx_hist != NULL) {
1231 dccp_tx_hist_delete(ccid3_tx_hist);
1232 ccid3_tx_hist = NULL;
1234 if (ccid3_rx_hist != NULL) {
1235 dccp_rx_hist_delete(ccid3_rx_hist);
1236 ccid3_rx_hist = NULL;
1238 if (ccid3_li_hist != NULL) {
1239 dccp_li_hist_delete(ccid3_li_hist);
1240 ccid3_li_hist = NULL;
1243 module_exit(ccid3_module_exit);
1245 MODULE_AUTHOR("Ian McDonald <iam4@cs.waikato.ac.nz>, "
1246 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1247 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1248 MODULE_LICENSE("GPL");
1249 MODULE_ALIAS("net-dccp-ccid-3");