2 * net/dccp/ccids/ccid3.c
4 * Copyright (c) 2005 The University of Waikato, Hamilton, New Zealand.
5 * Copyright (c) 2005-6 Ian McDonald <imcdnzl@gmail.com>
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;
63 static int ccid3_debug;
66 #define ccid3_pr_debug(format, a...) \
67 do { if (ccid3_debug) \
68 printk(KERN_DEBUG "%s: " format, __FUNCTION__, ##a); \
71 #define ccid3_pr_debug(format, a...)
74 static struct dccp_tx_hist *ccid3_tx_hist;
75 static struct dccp_rx_hist *ccid3_rx_hist;
76 static struct dccp_li_hist *ccid3_li_hist;
78 /* TFRC sender states */
79 enum ccid3_hc_tx_states {
80 TFRC_SSTATE_NO_SENT = 1,
87 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
89 static char *ccid3_state_names[] = {
90 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
91 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
92 [TFRC_SSTATE_FBACK] = "FBACK",
93 [TFRC_SSTATE_TERM] = "TERM",
96 return ccid3_state_names[state];
100 static void ccid3_hc_tx_set_state(struct sock *sk,
101 enum ccid3_hc_tx_states state)
103 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
104 enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
106 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
107 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
108 ccid3_tx_state_name(state));
109 WARN_ON(state == oldstate);
110 hctx->ccid3hctx_state = state;
113 /* Calculate new t_ipi (inter packet interval) by t_ipi = s / X_inst */
114 static inline void ccid3_calc_new_t_ipi(struct ccid3_hc_tx_sock *hctx)
117 * If no feedback spec says t_ipi is 1 second (set elsewhere and then
118 * doubles after every no feedback timer (separate function)
120 if (hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
121 hctx->ccid3hctx_t_ipi = usecs_div(hctx->ccid3hctx_s,
125 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
126 static inline void ccid3_calc_new_delta(struct ccid3_hc_tx_sock *hctx)
128 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
129 TFRC_OPSYS_HALF_TIME_GRAN);
135 * x_calc = calcX(s, R, p);
136 * X = max(min(X_calc, 2 * X_recv), s / t_mbi);
138 * If (now - tld >= R)
139 * X = max(min(2 * X, 2 * X_recv), s / R);
142 static void ccid3_hc_tx_update_x(struct sock *sk)
144 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
146 /* To avoid large error in calcX */
147 if (hctx->ccid3hctx_p >= TFRC_SMALLEST_P) {
148 hctx->ccid3hctx_x_calc = tfrc_calc_x(hctx->ccid3hctx_s,
151 hctx->ccid3hctx_x = max_t(u32, min_t(u32, hctx->ccid3hctx_x_calc,
152 2 * hctx->ccid3hctx_x_recv),
154 TFRC_MAX_BACK_OFF_TIME));
158 dccp_timestamp(sk, &now);
159 if (timeval_delta(&now, &hctx->ccid3hctx_t_ld) >=
160 hctx->ccid3hctx_rtt) {
161 hctx->ccid3hctx_x = max_t(u32, min_t(u32, hctx->ccid3hctx_x_recv,
162 hctx->ccid3hctx_x) * 2,
163 usecs_div(hctx->ccid3hctx_s,
164 hctx->ccid3hctx_rtt));
165 hctx->ccid3hctx_t_ld = now;
170 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
172 struct sock *sk = (struct sock *)data;
173 unsigned long next_tmout = 0;
174 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
177 if (sock_owned_by_user(sk)) {
178 /* Try again later. */
179 /* XXX: set some sensible MIB */
180 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
185 ccid3_pr_debug("%s, sk=%p, state=%s\n", dccp_role(sk), sk,
186 ccid3_tx_state_name(hctx->ccid3hctx_state));
188 switch (hctx->ccid3hctx_state) {
189 case TFRC_SSTATE_TERM:
191 case TFRC_SSTATE_NO_FBACK:
192 /* Halve send rate */
193 hctx->ccid3hctx_x /= 2;
194 if (hctx->ccid3hctx_x < (hctx->ccid3hctx_s /
195 TFRC_MAX_BACK_OFF_TIME))
196 hctx->ccid3hctx_x = (hctx->ccid3hctx_s /
197 TFRC_MAX_BACK_OFF_TIME);
199 ccid3_pr_debug("%s, sk=%p, state=%s, updated tx rate to %d "
202 ccid3_tx_state_name(hctx->ccid3hctx_state),
204 next_tmout = max_t(u32, 2 * usecs_div(hctx->ccid3hctx_s,
206 TFRC_INITIAL_TIMEOUT);
208 * FIXME - not sure above calculation is correct. See section
209 * 5 of CCID3 11 should adjust tx_t_ipi and double that to
213 case TFRC_SSTATE_FBACK:
215 * Check if IDLE since last timeout and recv rate is less than
218 if (!hctx->ccid3hctx_idle ||
219 (hctx->ccid3hctx_x_recv >=
220 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_MAX_BACK_OFF_TIME));
240 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc / 4;
242 /* Update sending rate */
243 ccid3_hc_tx_update_x(sk);
246 * Schedule no feedback timer to expire in
247 * max(4 * R, 2 * s / X)
249 next_tmout = max_t(u32, hctx->ccid3hctx_t_rto,
250 2 * usecs_div(hctx->ccid3hctx_s,
254 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
255 __FUNCTION__, dccp_role(sk), sk, hctx->ccid3hctx_state);
260 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
261 jiffies + max_t(u32, 1, usecs_to_jiffies(next_tmout)));
262 hctx->ccid3hctx_idle = 1;
268 static int ccid3_hc_tx_send_packet(struct sock *sk,
269 struct sk_buff *skb, int len)
271 struct dccp_sock *dp = dccp_sk(sk);
272 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
273 struct dccp_tx_hist_entry *new_packet;
278 BUG_ON(hctx == NULL || hctx->ccid3hctx_state == TFRC_SSTATE_TERM);
280 /* Check if pure ACK or Terminating*/
282 * XXX: We only call this function for DATA and DATAACK, on, these
283 * packets can have zero length, but why the comment about "pure ACK"?
285 if (unlikely(len == 0))
288 /* See if last packet allocated was not sent */
289 new_packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
290 if (new_packet == NULL || new_packet->dccphtx_sent) {
291 new_packet = dccp_tx_hist_entry_new(ccid3_tx_hist,
295 if (unlikely(new_packet == NULL)) {
296 LIMIT_NETDEBUG(KERN_WARNING "%s: %s, sk=%p, not enough "
297 "mem to add to history, send refused\n",
298 __FUNCTION__, dccp_role(sk), sk);
302 dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, new_packet);
305 dccp_timestamp(sk, &now);
307 switch (hctx->ccid3hctx_state) {
308 case TFRC_SSTATE_NO_SENT:
309 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
310 jiffies + usecs_to_jiffies(TFRC_INITIAL_TIMEOUT));
311 hctx->ccid3hctx_last_win_count = 0;
312 hctx->ccid3hctx_t_last_win_count = now;
313 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
314 hctx->ccid3hctx_t_ipi = TFRC_INITIAL_IPI;
316 /* Set nominal send time for initial packet */
317 hctx->ccid3hctx_t_nom = now;
318 timeval_add_usecs(&hctx->ccid3hctx_t_nom,
319 hctx->ccid3hctx_t_ipi);
320 ccid3_calc_new_delta(hctx);
323 case TFRC_SSTATE_NO_FBACK:
324 case TFRC_SSTATE_FBACK:
325 delay = (timeval_delta(&now, &hctx->ccid3hctx_t_nom) -
326 hctx->ccid3hctx_delta);
328 /* divide by -1000 is to convert to ms and get sign right */
329 rc = delay > 0 ? delay : 0;
332 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
333 __FUNCTION__, dccp_role(sk), sk, hctx->ccid3hctx_state);
339 /* Can we send? if so add options and add to packet history */
341 dp->dccps_hc_tx_insert_options = 1;
342 new_packet->dccphtx_ccval =
343 DCCP_SKB_CB(skb)->dccpd_ccval =
344 hctx->ccid3hctx_last_win_count;
350 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more, int len)
352 const struct dccp_sock *dp = dccp_sk(sk);
353 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
356 BUG_ON(hctx == NULL || hctx->ccid3hctx_state == TFRC_SSTATE_TERM);
358 dccp_timestamp(sk, &now);
360 /* check if we have sent a data packet */
362 unsigned long quarter_rtt;
363 struct dccp_tx_hist_entry *packet;
365 packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
366 if (unlikely(packet == NULL)) {
367 LIMIT_NETDEBUG(KERN_WARNING "%s: packet doesn't "
368 "exists in history!\n", __FUNCTION__);
371 if (unlikely(packet->dccphtx_sent)) {
372 LIMIT_NETDEBUG(KERN_WARNING "%s: no unsent packet in "
373 "history!\n", __FUNCTION__);
376 packet->dccphtx_tstamp = now;
377 packet->dccphtx_seqno = dp->dccps_gss;
379 * Check if win_count have changed
380 * Algorithm in "8.1. Window Counter Valuer" in
381 * draft-ietf-dccp-ccid3-11.txt
383 quarter_rtt = timeval_delta(&now, &hctx->ccid3hctx_t_last_win_count);
384 if (likely(hctx->ccid3hctx_rtt > 8))
385 quarter_rtt /= hctx->ccid3hctx_rtt / 4;
387 if (quarter_rtt > 0) {
388 hctx->ccid3hctx_t_last_win_count = now;
389 hctx->ccid3hctx_last_win_count = (hctx->ccid3hctx_last_win_count +
390 min_t(unsigned long, quarter_rtt, 5)) % 16;
391 ccid3_pr_debug("%s, sk=%p, window changed from "
394 packet->dccphtx_ccval,
395 hctx->ccid3hctx_last_win_count);
398 hctx->ccid3hctx_idle = 0;
399 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
400 packet->dccphtx_sent = 1;
402 ccid3_pr_debug("%s, sk=%p, seqno=%llu NOT inserted!\n",
403 dccp_role(sk), sk, dp->dccps_gss);
405 switch (hctx->ccid3hctx_state) {
406 case TFRC_SSTATE_NO_SENT:
407 /* if first wasn't pure ack */
409 printk(KERN_CRIT "%s: %s, First packet sent is noted "
410 "as a data packet\n",
411 __FUNCTION__, dccp_role(sk));
413 case TFRC_SSTATE_NO_FBACK:
414 case TFRC_SSTATE_FBACK:
416 hctx->ccid3hctx_t_nom = now;
417 ccid3_calc_new_t_ipi(hctx);
418 ccid3_calc_new_delta(hctx);
419 timeval_add_usecs(&hctx->ccid3hctx_t_nom,
420 hctx->ccid3hctx_t_ipi);
424 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
425 __FUNCTION__, dccp_role(sk), sk, hctx->ccid3hctx_state);
431 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
433 const struct dccp_sock *dp = dccp_sk(sk);
434 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
435 struct ccid3_options_received *opt_recv;
436 struct dccp_tx_hist_entry *packet;
438 unsigned long next_tmout;
444 BUG_ON(hctx == NULL || hctx->ccid3hctx_state == TFRC_SSTATE_TERM);
446 /* we are only interested in ACKs */
447 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
448 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
451 opt_recv = &hctx->ccid3hctx_options_received;
453 t_elapsed = dp->dccps_options_received.dccpor_elapsed_time * 10;
454 x_recv = opt_recv->ccid3or_receive_rate;
455 pinv = opt_recv->ccid3or_loss_event_rate;
457 switch (hctx->ccid3hctx_state) {
458 case TFRC_SSTATE_NO_SENT:
459 /* FIXME: what to do here? */
461 case TFRC_SSTATE_NO_FBACK:
462 case TFRC_SSTATE_FBACK:
463 /* Calculate new round trip sample by
464 * R_sample = (now - t_recvdata) - t_delay */
465 /* get t_recvdata from history */
466 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
467 DCCP_SKB_CB(skb)->dccpd_ack_seq);
468 if (unlikely(packet == NULL)) {
469 LIMIT_NETDEBUG(KERN_WARNING "%s: %s, sk=%p, seqno "
470 "%llu(%s) does't exist in history!\n",
471 __FUNCTION__, dccp_role(sk), sk,
472 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
473 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
478 dccp_timestamp(sk, &now);
479 r_sample = timeval_delta(&now, &packet->dccphtx_tstamp);
480 if (unlikely(r_sample <= t_elapsed))
481 LIMIT_NETDEBUG(KERN_WARNING "%s: r_sample=%uus, "
483 __FUNCTION__, r_sample, t_elapsed);
485 r_sample -= t_elapsed;
487 /* Update RTT estimate by
488 * If (No feedback recv)
491 * R = q * R + (1 - q) * R_sample;
493 * q is a constant, RFC 3448 recomments 0.9
495 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
496 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
497 hctx->ccid3hctx_rtt = r_sample;
499 hctx->ccid3hctx_rtt = (hctx->ccid3hctx_rtt * 9) / 10 +
502 ccid3_pr_debug("%s, sk=%p, New RTT estimate=%uus, "
503 "r_sample=%us\n", dccp_role(sk), sk,
504 hctx->ccid3hctx_rtt, r_sample);
506 /* Update timeout interval */
507 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
510 /* Update receive rate */
511 hctx->ccid3hctx_x_recv = x_recv;/* X_recv in bytes per sec */
513 /* Update loss event rate */
514 if (pinv == ~0 || pinv == 0)
515 hctx->ccid3hctx_p = 0;
517 hctx->ccid3hctx_p = 1000000 / pinv;
519 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P) {
520 hctx->ccid3hctx_p = TFRC_SMALLEST_P;
521 ccid3_pr_debug("%s, sk=%p, Smallest p used!\n",
526 /* unschedule no feedback timer */
527 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
529 /* Update sending rate */
530 ccid3_hc_tx_update_x(sk);
532 /* Update next send time */
533 timeval_sub_usecs(&hctx->ccid3hctx_t_nom,
534 hctx->ccid3hctx_t_ipi);
535 ccid3_calc_new_t_ipi(hctx);
536 timeval_add_usecs(&hctx->ccid3hctx_t_nom,
537 hctx->ccid3hctx_t_ipi);
538 ccid3_calc_new_delta(hctx);
540 /* remove all packets older than the one acked from history */
541 dccp_tx_hist_purge_older(ccid3_tx_hist,
542 &hctx->ccid3hctx_hist, packet);
544 * As we have calculated new ipi, delta, t_nom it is possible that
545 * we now can send a packet, so wake up dccp_wait_for_ccids.
547 sk->sk_write_space(sk);
550 * Schedule no feedback timer to expire in
551 * max(4 * R, 2 * s / X)
553 next_tmout = max(hctx->ccid3hctx_t_rto,
554 2 * usecs_div(hctx->ccid3hctx_s,
557 ccid3_pr_debug("%s, sk=%p, Scheduled no feedback timer to "
558 "expire in %lu jiffies (%luus)\n",
560 usecs_to_jiffies(next_tmout), next_tmout);
562 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
563 jiffies + max_t(u32, 1, usecs_to_jiffies(next_tmout)));
566 hctx->ccid3hctx_idle = 1;
569 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
570 __FUNCTION__, dccp_role(sk), sk, hctx->ccid3hctx_state);
576 static int ccid3_hc_tx_insert_options(struct sock *sk, struct sk_buff *skb)
578 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
580 BUG_ON(hctx == NULL);
582 if (sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)
583 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
587 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
588 unsigned char len, u16 idx,
589 unsigned char *value)
592 const struct dccp_sock *dp = dccp_sk(sk);
593 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
594 struct ccid3_options_received *opt_recv;
596 BUG_ON(hctx == NULL);
598 opt_recv = &hctx->ccid3hctx_options_received;
600 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
601 opt_recv->ccid3or_seqno = dp->dccps_gsr;
602 opt_recv->ccid3or_loss_event_rate = ~0;
603 opt_recv->ccid3or_loss_intervals_idx = 0;
604 opt_recv->ccid3or_loss_intervals_len = 0;
605 opt_recv->ccid3or_receive_rate = 0;
609 case TFRC_OPT_LOSS_EVENT_RATE:
610 if (unlikely(len != 4)) {
611 LIMIT_NETDEBUG(KERN_WARNING "%s: %s, sk=%p, invalid "
612 "len for TFRC_OPT_LOSS_EVENT_RATE\n",
613 __FUNCTION__, dccp_role(sk), sk);
616 opt_recv->ccid3or_loss_event_rate = ntohl(*(__be32 *)value);
617 ccid3_pr_debug("%s, sk=%p, LOSS_EVENT_RATE=%u\n",
619 opt_recv->ccid3or_loss_event_rate);
622 case TFRC_OPT_LOSS_INTERVALS:
623 opt_recv->ccid3or_loss_intervals_idx = idx;
624 opt_recv->ccid3or_loss_intervals_len = len;
625 ccid3_pr_debug("%s, sk=%p, LOSS_INTERVALS=(%u, %u)\n",
627 opt_recv->ccid3or_loss_intervals_idx,
628 opt_recv->ccid3or_loss_intervals_len);
630 case TFRC_OPT_RECEIVE_RATE:
631 if (unlikely(len != 4)) {
632 LIMIT_NETDEBUG(KERN_WARNING "%s: %s, sk=%p, invalid "
633 "len for TFRC_OPT_RECEIVE_RATE\n",
634 __FUNCTION__, dccp_role(sk), sk);
637 opt_recv->ccid3or_receive_rate = ntohl(*(__be32 *)value);
638 ccid3_pr_debug("%s, sk=%p, RECEIVE_RATE=%u\n",
640 opt_recv->ccid3or_receive_rate);
648 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
650 struct dccp_sock *dp = dccp_sk(sk);
651 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
653 if (dp->dccps_packet_size >= TFRC_MIN_PACKET_SIZE &&
654 dp->dccps_packet_size <= TFRC_MAX_PACKET_SIZE)
655 hctx->ccid3hctx_s = dp->dccps_packet_size;
657 hctx->ccid3hctx_s = TFRC_STD_PACKET_SIZE;
659 /* Set transmission rate to 1 packet per second */
660 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
661 hctx->ccid3hctx_t_rto = USEC_PER_SEC;
662 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
663 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
665 hctx->ccid3hctx_no_feedback_timer.function = ccid3_hc_tx_no_feedback_timer;
666 hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
667 init_timer(&hctx->ccid3hctx_no_feedback_timer);
672 static void ccid3_hc_tx_exit(struct sock *sk)
674 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
676 BUG_ON(hctx == NULL);
678 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
679 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
681 /* Empty packet history */
682 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
686 * RX Half Connection methods
689 /* TFRC receiver states */
690 enum ccid3_hc_rx_states {
691 TFRC_RSTATE_NO_DATA = 1,
693 TFRC_RSTATE_TERM = 127,
697 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
699 static char *ccid3_rx_state_names[] = {
700 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
701 [TFRC_RSTATE_DATA] = "DATA",
702 [TFRC_RSTATE_TERM] = "TERM",
705 return ccid3_rx_state_names[state];
709 static void ccid3_hc_rx_set_state(struct sock *sk,
710 enum ccid3_hc_rx_states state)
712 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
713 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
715 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
716 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
717 ccid3_rx_state_name(state));
718 WARN_ON(state == oldstate);
719 hcrx->ccid3hcrx_state = state;
722 static void ccid3_hc_rx_send_feedback(struct sock *sk)
724 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
725 struct dccp_sock *dp = dccp_sk(sk);
726 struct dccp_rx_hist_entry *packet;
729 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
731 dccp_timestamp(sk, &now);
733 switch (hcrx->ccid3hcrx_state) {
734 case TFRC_RSTATE_NO_DATA:
735 hcrx->ccid3hcrx_x_recv = 0;
737 case TFRC_RSTATE_DATA: {
738 const u32 delta = timeval_delta(&now,
739 &hcrx->ccid3hcrx_tstamp_last_feedback);
740 hcrx->ccid3hcrx_x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv,
745 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
746 __FUNCTION__, dccp_role(sk), sk, hcrx->ccid3hcrx_state);
751 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
752 if (unlikely(packet == NULL)) {
753 LIMIT_NETDEBUG(KERN_WARNING "%s: %s, sk=%p, no data packet "
755 __FUNCTION__, dccp_role(sk), sk);
759 hcrx->ccid3hcrx_tstamp_last_feedback = now;
760 hcrx->ccid3hcrx_last_counter = packet->dccphrx_ccval;
761 hcrx->ccid3hcrx_seqno_last_counter = packet->dccphrx_seqno;
762 hcrx->ccid3hcrx_bytes_recv = 0;
764 /* Convert to multiples of 10us */
765 hcrx->ccid3hcrx_elapsed_time =
766 timeval_delta(&now, &packet->dccphrx_tstamp) / 10;
767 if (hcrx->ccid3hcrx_p == 0)
768 hcrx->ccid3hcrx_pinv = ~0;
770 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
771 dp->dccps_hc_rx_insert_options = 1;
775 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
777 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
780 BUG_ON(hcrx == NULL);
782 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
785 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_last_counter;
787 if (dccp_packet_without_ack(skb))
790 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
791 pinv = htonl(hcrx->ccid3hcrx_pinv);
793 if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
794 dccp_insert_option_elapsed_time(sk, skb,
795 hcrx->ccid3hcrx_elapsed_time)) ||
796 dccp_insert_option_timestamp(sk, skb) ||
797 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
798 &pinv, sizeof(pinv)) ||
799 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
800 &x_recv, sizeof(x_recv)))
806 /* calculate first loss interval
808 * returns estimated loss interval in usecs */
810 static u32 ccid3_hc_rx_calc_first_li(struct sock *sk)
812 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
813 struct dccp_rx_hist_entry *entry, *next, *tail = NULL;
814 u32 rtt, delta, x_recv, fval, p, tmp2;
815 struct timeval tstamp = { 0, };
821 list_for_each_entry_safe(entry, next, &hcrx->ccid3hcrx_hist,
823 if (dccp_rx_hist_entry_data_packet(entry)) {
828 tstamp = entry->dccphrx_tstamp;
829 win_count = entry->dccphrx_ccval;
833 interval = win_count - entry->dccphrx_ccval;
835 interval += TFRC_WIN_COUNT_LIMIT;
843 if (unlikely(step == 0)) {
844 LIMIT_NETDEBUG(KERN_WARNING "%s: %s, sk=%p, packet history "
845 "contains no data packets!\n",
846 __FUNCTION__, dccp_role(sk), sk);
850 if (unlikely(interval == 0)) {
851 LIMIT_NETDEBUG(KERN_WARNING "%s: %s, sk=%p, Could not find a "
852 "win_count interval > 0. Defaulting to 1\n",
853 __FUNCTION__, dccp_role(sk), sk);
857 rtt = timeval_delta(&tstamp, &tail->dccphrx_tstamp) * 4 / interval;
858 ccid3_pr_debug("%s, sk=%p, approximated RTT to %uus\n",
859 dccp_role(sk), sk, rtt);
863 dccp_timestamp(sk, &tstamp);
864 delta = timeval_delta(&tstamp, &hcrx->ccid3hcrx_tstamp_last_feedback);
865 x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv, delta);
867 tmp1 = (u64)x_recv * (u64)rtt;
868 do_div(tmp1,10000000);
870 fval = (hcrx->ccid3hcrx_s * 100000) / tmp2;
871 /* do not alter order above or you will get overflow on 32 bit */
872 p = tfrc_calc_x_reverse_lookup(fval);
873 ccid3_pr_debug("%s, sk=%p, receive rate=%u bytes/s, implied "
874 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
882 static void ccid3_hc_rx_update_li(struct sock *sk, u64 seq_loss, u8 win_loss)
884 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
886 if (seq_loss != DCCP_MAX_SEQNO + 1 &&
887 list_empty(&hcrx->ccid3hcrx_li_hist)) {
888 struct dccp_li_hist_entry *li_tail;
890 li_tail = dccp_li_hist_interval_new(ccid3_li_hist,
891 &hcrx->ccid3hcrx_li_hist,
895 li_tail->dccplih_interval = ccid3_hc_rx_calc_first_li(sk);
897 LIMIT_NETDEBUG(KERN_WARNING "%s: FIXME: find end of "
898 "interval\n", __FUNCTION__);
901 static void ccid3_hc_rx_detect_loss(struct sock *sk)
903 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
905 const u64 seq_loss = dccp_rx_hist_detect_loss(&hcrx->ccid3hcrx_hist,
906 &hcrx->ccid3hcrx_li_hist,
909 ccid3_hc_rx_update_li(sk, seq_loss, win_loss);
912 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
914 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
915 const struct dccp_options_received *opt_recv;
916 struct dccp_rx_hist_entry *packet;
919 u32 p_prev, r_sample, t_elapsed;
922 BUG_ON(hcrx == NULL ||
923 !(hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA ||
924 hcrx->ccid3hcrx_state == TFRC_RSTATE_DATA));
926 opt_recv = &dccp_sk(sk)->dccps_options_received;
928 switch (DCCP_SKB_CB(skb)->dccpd_type) {
930 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
932 case DCCP_PKT_DATAACK:
933 if (opt_recv->dccpor_timestamp_echo == 0)
935 p_prev = hcrx->ccid3hcrx_rtt;
936 dccp_timestamp(sk, &now);
937 timeval_sub_usecs(&now, opt_recv->dccpor_timestamp_echo * 10);
938 r_sample = timeval_usecs(&now);
939 t_elapsed = opt_recv->dccpor_elapsed_time * 10;
941 if (unlikely(r_sample <= t_elapsed))
942 LIMIT_NETDEBUG(KERN_WARNING "%s: r_sample=%uus, "
944 __FUNCTION__, r_sample, t_elapsed);
946 r_sample -= t_elapsed;
948 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
949 hcrx->ccid3hcrx_rtt = r_sample;
951 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
954 if (p_prev != hcrx->ccid3hcrx_rtt)
955 ccid3_pr_debug("%s, New RTT=%luus, elapsed time=%u\n",
956 dccp_role(sk), hcrx->ccid3hcrx_rtt,
957 opt_recv->dccpor_elapsed_time);
961 default: /* We're not interested in other packet types, move along */
965 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, sk, opt_recv->dccpor_ndp,
967 if (unlikely(packet == NULL)) {
968 LIMIT_NETDEBUG(KERN_WARNING "%s: %s, sk=%p, Not enough mem to "
969 "add rx packet to history, consider it lost!\n",
970 __FUNCTION__, dccp_role(sk), sk);
974 win_count = packet->dccphrx_ccval;
976 ins = dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
977 &hcrx->ccid3hcrx_li_hist, packet);
979 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
982 switch (hcrx->ccid3hcrx_state) {
983 case TFRC_RSTATE_NO_DATA:
984 ccid3_pr_debug("%s, sk=%p(%s), skb=%p, sending initial "
987 dccp_state_name(sk->sk_state), skb);
988 ccid3_hc_rx_send_feedback(sk);
989 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
991 case TFRC_RSTATE_DATA:
992 hcrx->ccid3hcrx_bytes_recv += skb->len -
993 dccp_hdr(skb)->dccph_doff * 4;
997 dccp_timestamp(sk, &now);
998 if (timeval_delta(&now, &hcrx->ccid3hcrx_tstamp_last_ack) >=
999 hcrx->ccid3hcrx_rtt) {
1000 hcrx->ccid3hcrx_tstamp_last_ack = now;
1001 ccid3_hc_rx_send_feedback(sk);
1005 printk(KERN_CRIT "%s: %s, sk=%p, Illegal state (%d)!\n",
1006 __FUNCTION__, dccp_role(sk), sk, hcrx->ccid3hcrx_state);
1011 /* Dealing with packet loss */
1012 ccid3_pr_debug("%s, sk=%p(%s), data loss! Reacting...\n",
1013 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
1015 ccid3_hc_rx_detect_loss(sk);
1016 p_prev = hcrx->ccid3hcrx_p;
1018 /* Calculate loss event rate */
1019 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
1020 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
1022 /* Scaling up by 1000000 as fixed decimal */
1024 hcrx->ccid3hcrx_p = 1000000 / i_mean;
1027 if (hcrx->ccid3hcrx_p > p_prev) {
1028 ccid3_hc_rx_send_feedback(sk);
1033 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
1035 struct dccp_sock *dp = dccp_sk(sk);
1036 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
1038 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
1040 if (dp->dccps_packet_size >= TFRC_MIN_PACKET_SIZE &&
1041 dp->dccps_packet_size <= TFRC_MAX_PACKET_SIZE)
1042 hcrx->ccid3hcrx_s = dp->dccps_packet_size;
1044 hcrx->ccid3hcrx_s = TFRC_STD_PACKET_SIZE;
1046 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
1047 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
1048 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
1049 dccp_timestamp(sk, &hcrx->ccid3hcrx_tstamp_last_ack);
1050 hcrx->ccid3hcrx_tstamp_last_feedback = hcrx->ccid3hcrx_tstamp_last_ack;
1051 hcrx->ccid3hcrx_rtt = 5000; /* XXX 5ms for now... */
1055 static void ccid3_hc_rx_exit(struct sock *sk)
1057 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1059 BUG_ON(hcrx == NULL);
1061 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
1063 /* Empty packet history */
1064 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
1066 /* Empty loss interval history */
1067 dccp_li_hist_purge(ccid3_li_hist, &hcrx->ccid3hcrx_li_hist);
1070 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
1072 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1074 /* Listen socks doesn't have a private CCID block */
1075 if (sk->sk_state == DCCP_LISTEN)
1078 BUG_ON(hcrx == NULL);
1080 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
1081 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1082 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
1085 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
1087 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1089 /* Listen socks doesn't have a private CCID block */
1090 if (sk->sk_state == DCCP_LISTEN)
1093 BUG_ON(hctx == NULL);
1095 info->tcpi_rto = hctx->ccid3hctx_t_rto;
1096 info->tcpi_rtt = hctx->ccid3hctx_rtt;
1099 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
1100 u32 __user *optval, int __user *optlen)
1102 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1105 /* Listen socks doesn't have a private CCID block */
1106 if (sk->sk_state == DCCP_LISTEN)
1110 case DCCP_SOCKOPT_CCID_RX_INFO:
1111 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
1113 len = sizeof(hcrx->ccid3hcrx_tfrc);
1114 val = &hcrx->ccid3hcrx_tfrc;
1117 return -ENOPROTOOPT;
1120 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1126 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
1127 u32 __user *optval, int __user *optlen)
1129 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1132 /* Listen socks doesn't have a private CCID block */
1133 if (sk->sk_state == DCCP_LISTEN)
1137 case DCCP_SOCKOPT_CCID_TX_INFO:
1138 if (len < sizeof(hctx->ccid3hctx_tfrc))
1140 len = sizeof(hctx->ccid3hctx_tfrc);
1141 val = &hctx->ccid3hctx_tfrc;
1144 return -ENOPROTOOPT;
1147 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1153 static struct ccid_operations ccid3 = {
1155 .ccid_name = "ccid3",
1156 .ccid_owner = THIS_MODULE,
1157 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
1158 .ccid_hc_tx_init = ccid3_hc_tx_init,
1159 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1160 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1161 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1162 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1163 .ccid_hc_tx_insert_options = ccid3_hc_tx_insert_options,
1164 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1165 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
1166 .ccid_hc_rx_init = ccid3_hc_rx_init,
1167 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1168 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1169 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1170 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1171 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1172 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1173 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1176 module_param(ccid3_debug, int, 0444);
1177 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1179 static __init int ccid3_module_init(void)
1183 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1184 if (ccid3_rx_hist == NULL)
1187 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1188 if (ccid3_tx_hist == NULL)
1191 ccid3_li_hist = dccp_li_hist_new("ccid3");
1192 if (ccid3_li_hist == NULL)
1195 rc = ccid_register(&ccid3);
1197 goto out_free_loss_interval_history;
1201 out_free_loss_interval_history:
1202 dccp_li_hist_delete(ccid3_li_hist);
1203 ccid3_li_hist = NULL;
1205 dccp_tx_hist_delete(ccid3_tx_hist);
1206 ccid3_tx_hist = NULL;
1208 dccp_rx_hist_delete(ccid3_rx_hist);
1209 ccid3_rx_hist = NULL;
1212 module_init(ccid3_module_init);
1214 static __exit void ccid3_module_exit(void)
1216 ccid_unregister(&ccid3);
1218 if (ccid3_tx_hist != NULL) {
1219 dccp_tx_hist_delete(ccid3_tx_hist);
1220 ccid3_tx_hist = NULL;
1222 if (ccid3_rx_hist != NULL) {
1223 dccp_rx_hist_delete(ccid3_rx_hist);
1224 ccid3_rx_hist = NULL;
1226 if (ccid3_li_hist != NULL) {
1227 dccp_li_hist_delete(ccid3_li_hist);
1228 ccid3_li_hist = NULL;
1231 module_exit(ccid3_module_exit);
1233 MODULE_AUTHOR("Ian McDonald <iam4@cs.waikato.ac.nz>, "
1234 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1235 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1236 MODULE_LICENSE("GPL");
1237 MODULE_ALIAS("net-dccp-ccid-3");