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;
103 /* Calculate new t_ipi (inter packet interval) by t_ipi = s / X_inst */
104 static inline void ccid3_calc_new_t_ipi(struct ccid3_hc_tx_sock *hctx)
107 * If no feedback spec says t_ipi is 1 second (set elsewhere and then
108 * doubles after every no feedback timer (separate function)
110 if (hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
111 hctx->ccid3hctx_t_ipi = usecs_div(hctx->ccid3hctx_s,
115 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
116 static inline void ccid3_calc_new_delta(struct ccid3_hc_tx_sock *hctx)
118 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
119 TFRC_OPSYS_HALF_TIME_GRAN);
125 * x_calc = calcX(s, R, p);
126 * X = max(min(X_calc, 2 * X_recv), s / t_mbi);
128 * If (now - tld >= R)
129 * X = max(min(2 * X, 2 * X_recv), s / R);
132 static void ccid3_hc_tx_update_x(struct sock *sk)
134 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
136 /* To avoid large error in calcX */
137 if (hctx->ccid3hctx_p >= TFRC_SMALLEST_P) {
138 hctx->ccid3hctx_x_calc = tfrc_calc_x(hctx->ccid3hctx_s,
141 hctx->ccid3hctx_x = max_t(u32, min_t(u32, hctx->ccid3hctx_x_calc,
142 2 * hctx->ccid3hctx_x_recv),
144 TFRC_MAX_BACK_OFF_TIME));
148 dccp_timestamp(sk, &now);
149 if (timeval_delta(&now, &hctx->ccid3hctx_t_ld) >=
150 hctx->ccid3hctx_rtt) {
151 hctx->ccid3hctx_x = max_t(u32, min_t(u32, hctx->ccid3hctx_x_recv,
152 hctx->ccid3hctx_x) * 2,
153 usecs_div(hctx->ccid3hctx_s,
154 hctx->ccid3hctx_rtt));
155 hctx->ccid3hctx_t_ld = now;
160 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
162 struct sock *sk = (struct sock *)data;
163 unsigned long next_tmout = 0;
164 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
167 if (sock_owned_by_user(sk)) {
168 /* Try again later. */
169 /* XXX: set some sensible MIB */
170 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
175 ccid3_pr_debug("%s, sk=%p, state=%s\n", dccp_role(sk), sk,
176 ccid3_tx_state_name(hctx->ccid3hctx_state));
178 switch (hctx->ccid3hctx_state) {
179 case TFRC_SSTATE_NO_FBACK:
180 /* Halve send rate */
181 hctx->ccid3hctx_x /= 2;
182 if (hctx->ccid3hctx_x < (hctx->ccid3hctx_s /
183 TFRC_MAX_BACK_OFF_TIME))
184 hctx->ccid3hctx_x = (hctx->ccid3hctx_s /
185 TFRC_MAX_BACK_OFF_TIME);
187 ccid3_pr_debug("%s, sk=%p, state=%s, updated tx rate to %d "
190 ccid3_tx_state_name(hctx->ccid3hctx_state),
192 next_tmout = max_t(u32, 2 * usecs_div(hctx->ccid3hctx_s,
194 TFRC_INITIAL_TIMEOUT);
196 * FIXME - not sure above calculation is correct. See section
197 * 5 of CCID3 11 should adjust tx_t_ipi and double that to
201 case TFRC_SSTATE_FBACK:
203 * Check if IDLE since last timeout and recv rate is less than
206 if (!hctx->ccid3hctx_idle ||
207 (hctx->ccid3hctx_x_recv >=
208 4 * usecs_div(hctx->ccid3hctx_s, hctx->ccid3hctx_rtt))) {
209 ccid3_pr_debug("%s, sk=%p, state=%s, not idle\n",
211 ccid3_tx_state_name(hctx->ccid3hctx_state));
212 /* Halve sending rate */
214 /* If (X_calc > 2 * X_recv)
215 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
217 * X_recv = X_calc / 4;
219 BUG_ON(hctx->ccid3hctx_p >= TFRC_SMALLEST_P &&
220 hctx->ccid3hctx_x_calc == 0);
222 /* check also if p is zero -> x_calc is infinity? */
223 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P ||
224 hctx->ccid3hctx_x_calc > 2 * hctx->ccid3hctx_x_recv)
225 hctx->ccid3hctx_x_recv = max_t(u32, hctx->ccid3hctx_x_recv / 2,
226 hctx->ccid3hctx_s / (2 * TFRC_MAX_BACK_OFF_TIME));
228 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc / 4;
230 /* Update sending rate */
231 ccid3_hc_tx_update_x(sk);
234 * Schedule no feedback timer to expire in
235 * max(4 * R, 2 * s / X)
237 next_tmout = max_t(u32, hctx->ccid3hctx_t_rto,
238 2 * usecs_div(hctx->ccid3hctx_s,
241 case TFRC_SSTATE_NO_SENT:
242 DCCP_BUG("Illegal %s state NO_SENT, sk=%p", dccp_role(sk), sk);
244 case TFRC_SSTATE_TERM:
248 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
249 jiffies + max_t(u32, 1, usecs_to_jiffies(next_tmout)));
250 hctx->ccid3hctx_idle = 1;
256 static int ccid3_hc_tx_send_packet(struct sock *sk,
257 struct sk_buff *skb, int len)
259 struct dccp_sock *dp = dccp_sk(sk);
260 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
261 struct dccp_tx_hist_entry *new_packet;
266 BUG_ON(hctx == NULL);
268 /* Check if pure ACK or Terminating*/
270 * XXX: We only call this function for DATA and DATAACK, on, these
271 * packets can have zero length, but why the comment about "pure ACK"?
273 if (unlikely(len == 0))
276 /* See if last packet allocated was not sent */
277 new_packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
278 if (new_packet == NULL || new_packet->dccphtx_sent) {
279 new_packet = dccp_tx_hist_entry_new(ccid3_tx_hist,
283 if (unlikely(new_packet == NULL)) {
284 DCCP_WARN("%s, sk=%p, not enough mem to add to history,"
285 "send refused\n", dccp_role(sk), sk);
289 dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, new_packet);
292 dccp_timestamp(sk, &now);
294 switch (hctx->ccid3hctx_state) {
295 case TFRC_SSTATE_NO_SENT:
296 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
297 jiffies + usecs_to_jiffies(TFRC_INITIAL_TIMEOUT));
298 hctx->ccid3hctx_last_win_count = 0;
299 hctx->ccid3hctx_t_last_win_count = now;
300 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
301 hctx->ccid3hctx_t_ipi = TFRC_INITIAL_IPI;
303 /* Set nominal send time for initial packet */
304 hctx->ccid3hctx_t_nom = now;
305 timeval_add_usecs(&hctx->ccid3hctx_t_nom,
306 hctx->ccid3hctx_t_ipi);
307 ccid3_calc_new_delta(hctx);
310 case TFRC_SSTATE_NO_FBACK:
311 case TFRC_SSTATE_FBACK:
312 delay = (timeval_delta(&now, &hctx->ccid3hctx_t_nom) -
313 hctx->ccid3hctx_delta);
315 /* divide by -1000 is to convert to ms and get sign right */
316 rc = delay > 0 ? delay : 0;
318 case TFRC_SSTATE_TERM:
319 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
324 /* Can we send? if so add options and add to packet history */
326 dp->dccps_hc_tx_insert_options = 1;
327 new_packet->dccphtx_ccval =
328 DCCP_SKB_CB(skb)->dccpd_ccval =
329 hctx->ccid3hctx_last_win_count;
330 timeval_add_usecs(&hctx->ccid3hctx_t_nom,
331 hctx->ccid3hctx_t_ipi);
337 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more, int len)
339 const struct dccp_sock *dp = dccp_sk(sk);
340 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
343 BUG_ON(hctx == NULL);
345 dccp_timestamp(sk, &now);
347 /* check if we have sent a data packet */
349 unsigned long quarter_rtt;
350 struct dccp_tx_hist_entry *packet;
352 packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
353 if (unlikely(packet == NULL)) {
354 DCCP_WARN("packet doesn't exist in history!\n");
357 if (unlikely(packet->dccphtx_sent)) {
358 DCCP_WARN("no unsent packet in history!\n");
361 packet->dccphtx_tstamp = now;
362 packet->dccphtx_seqno = dp->dccps_gss;
364 * Check if win_count have changed
365 * Algorithm in "8.1. Window Counter Value" in RFC 4342.
367 quarter_rtt = timeval_delta(&now, &hctx->ccid3hctx_t_last_win_count);
368 if (likely(hctx->ccid3hctx_rtt > 8))
369 quarter_rtt /= hctx->ccid3hctx_rtt / 4;
371 if (quarter_rtt > 0) {
372 hctx->ccid3hctx_t_last_win_count = now;
373 hctx->ccid3hctx_last_win_count = (hctx->ccid3hctx_last_win_count +
374 min_t(unsigned long, quarter_rtt, 5)) % 16;
375 ccid3_pr_debug("%s, sk=%p, window changed from "
378 packet->dccphtx_ccval,
379 hctx->ccid3hctx_last_win_count);
382 hctx->ccid3hctx_idle = 0;
383 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
384 packet->dccphtx_sent = 1;
386 ccid3_pr_debug("%s, sk=%p, seqno=%llu NOT inserted!\n",
387 dccp_role(sk), sk, dp->dccps_gss);
389 switch (hctx->ccid3hctx_state) {
390 case TFRC_SSTATE_NO_SENT:
391 /* if first wasn't pure ack */
393 DCCP_CRIT("%s, First packet sent is noted "
394 "as a data packet", dccp_role(sk));
396 case TFRC_SSTATE_NO_FBACK:
397 case TFRC_SSTATE_FBACK:
399 timeval_sub_usecs(&hctx->ccid3hctx_t_nom,
400 hctx->ccid3hctx_t_ipi);
401 ccid3_calc_new_t_ipi(hctx);
402 ccid3_calc_new_delta(hctx);
403 timeval_add_usecs(&hctx->ccid3hctx_t_nom,
404 hctx->ccid3hctx_t_ipi);
407 case TFRC_SSTATE_TERM:
408 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
413 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
415 const struct dccp_sock *dp = dccp_sk(sk);
416 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
417 struct ccid3_options_received *opt_recv;
418 struct dccp_tx_hist_entry *packet;
420 unsigned long next_tmout;
426 BUG_ON(hctx == NULL);
428 /* we are only interested in ACKs */
429 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
430 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
433 opt_recv = &hctx->ccid3hctx_options_received;
435 t_elapsed = dp->dccps_options_received.dccpor_elapsed_time * 10;
436 x_recv = opt_recv->ccid3or_receive_rate;
437 pinv = opt_recv->ccid3or_loss_event_rate;
439 switch (hctx->ccid3hctx_state) {
440 case TFRC_SSTATE_NO_SENT:
441 /* FIXME: what to do here? */
443 case TFRC_SSTATE_NO_FBACK:
444 case TFRC_SSTATE_FBACK:
445 /* Calculate new round trip sample by
446 * R_sample = (now - t_recvdata) - t_delay */
447 /* get t_recvdata from history */
448 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
449 DCCP_SKB_CB(skb)->dccpd_ack_seq);
450 if (unlikely(packet == NULL)) {
451 DCCP_WARN("%s, sk=%p, seqno %llu(%s) does't exist "
452 "in history!\n", dccp_role(sk), sk,
453 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
454 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
459 dccp_timestamp(sk, &now);
460 r_sample = timeval_delta(&now, &packet->dccphtx_tstamp);
461 if (unlikely(r_sample <= t_elapsed))
462 DCCP_WARN("r_sample=%uus,t_elapsed=%uus\n",
463 r_sample, t_elapsed);
465 r_sample -= t_elapsed;
467 /* Update RTT estimate by
468 * If (No feedback recv)
471 * R = q * R + (1 - q) * R_sample;
473 * q is a constant, RFC 3448 recomments 0.9
475 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
476 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
477 hctx->ccid3hctx_rtt = r_sample;
479 hctx->ccid3hctx_rtt = (hctx->ccid3hctx_rtt * 9) / 10 +
482 ccid3_pr_debug("%s, sk=%p, New RTT estimate=%uus, "
483 "r_sample=%us\n", dccp_role(sk), sk,
484 hctx->ccid3hctx_rtt, r_sample);
486 /* Update timeout interval */
487 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
490 /* Update receive rate */
491 hctx->ccid3hctx_x_recv = x_recv;/* X_recv in bytes per sec */
493 /* Update loss event rate */
494 if (pinv == ~0 || pinv == 0)
495 hctx->ccid3hctx_p = 0;
497 hctx->ccid3hctx_p = 1000000 / pinv;
499 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P) {
500 hctx->ccid3hctx_p = TFRC_SMALLEST_P;
501 ccid3_pr_debug("%s, sk=%p, Smallest p used!\n",
506 /* unschedule no feedback timer */
507 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
509 /* Update sending rate */
510 ccid3_hc_tx_update_x(sk);
512 /* Update next send time */
513 timeval_sub_usecs(&hctx->ccid3hctx_t_nom,
514 hctx->ccid3hctx_t_ipi);
515 ccid3_calc_new_t_ipi(hctx);
516 timeval_add_usecs(&hctx->ccid3hctx_t_nom,
517 hctx->ccid3hctx_t_ipi);
518 ccid3_calc_new_delta(hctx);
520 /* remove all packets older than the one acked from history */
521 dccp_tx_hist_purge_older(ccid3_tx_hist,
522 &hctx->ccid3hctx_hist, packet);
524 * As we have calculated new ipi, delta, t_nom it is possible that
525 * we now can send a packet, so wake up dccp_wait_for_ccids.
527 sk->sk_write_space(sk);
530 * Schedule no feedback timer to expire in
531 * max(4 * R, 2 * s / X)
533 next_tmout = max(hctx->ccid3hctx_t_rto,
534 2 * usecs_div(hctx->ccid3hctx_s,
537 ccid3_pr_debug("%s, sk=%p, Scheduled no feedback timer to "
538 "expire in %lu jiffies (%luus)\n",
540 usecs_to_jiffies(next_tmout), next_tmout);
542 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
543 jiffies + max_t(u32, 1, usecs_to_jiffies(next_tmout)));
546 hctx->ccid3hctx_idle = 1;
548 case TFRC_SSTATE_TERM:
549 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
554 static int ccid3_hc_tx_insert_options(struct sock *sk, struct sk_buff *skb)
556 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
558 BUG_ON(hctx == NULL);
560 if (sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)
561 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
565 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
566 unsigned char len, u16 idx,
567 unsigned char *value)
570 const struct dccp_sock *dp = dccp_sk(sk);
571 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
572 struct ccid3_options_received *opt_recv;
574 BUG_ON(hctx == NULL);
576 opt_recv = &hctx->ccid3hctx_options_received;
578 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
579 opt_recv->ccid3or_seqno = dp->dccps_gsr;
580 opt_recv->ccid3or_loss_event_rate = ~0;
581 opt_recv->ccid3or_loss_intervals_idx = 0;
582 opt_recv->ccid3or_loss_intervals_len = 0;
583 opt_recv->ccid3or_receive_rate = 0;
587 case TFRC_OPT_LOSS_EVENT_RATE:
588 if (unlikely(len != 4)) {
589 DCCP_WARN("%s, sk=%p, invalid len %d "
590 "for TFRC_OPT_LOSS_EVENT_RATE\n",
591 dccp_role(sk), sk, len);
594 opt_recv->ccid3or_loss_event_rate = ntohl(*(__be32 *)value);
595 ccid3_pr_debug("%s, sk=%p, LOSS_EVENT_RATE=%u\n",
597 opt_recv->ccid3or_loss_event_rate);
600 case TFRC_OPT_LOSS_INTERVALS:
601 opt_recv->ccid3or_loss_intervals_idx = idx;
602 opt_recv->ccid3or_loss_intervals_len = len;
603 ccid3_pr_debug("%s, sk=%p, LOSS_INTERVALS=(%u, %u)\n",
605 opt_recv->ccid3or_loss_intervals_idx,
606 opt_recv->ccid3or_loss_intervals_len);
608 case TFRC_OPT_RECEIVE_RATE:
609 if (unlikely(len != 4)) {
610 DCCP_WARN("%s, sk=%p, invalid len %d "
611 "for TFRC_OPT_RECEIVE_RATE\n",
612 dccp_role(sk), sk, len);
615 opt_recv->ccid3or_receive_rate = ntohl(*(__be32 *)value);
616 ccid3_pr_debug("%s, sk=%p, RECEIVE_RATE=%u\n",
618 opt_recv->ccid3or_receive_rate);
626 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
628 struct dccp_sock *dp = dccp_sk(sk);
629 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
631 if (dp->dccps_packet_size >= TFRC_MIN_PACKET_SIZE &&
632 dp->dccps_packet_size <= TFRC_MAX_PACKET_SIZE)
633 hctx->ccid3hctx_s = dp->dccps_packet_size;
635 hctx->ccid3hctx_s = TFRC_STD_PACKET_SIZE;
637 /* Set transmission rate to 1 packet per second */
638 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
639 hctx->ccid3hctx_t_rto = USEC_PER_SEC;
640 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
641 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
643 hctx->ccid3hctx_no_feedback_timer.function = ccid3_hc_tx_no_feedback_timer;
644 hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
645 init_timer(&hctx->ccid3hctx_no_feedback_timer);
650 static void ccid3_hc_tx_exit(struct sock *sk)
652 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
654 BUG_ON(hctx == NULL);
656 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
657 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
659 /* Empty packet history */
660 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
664 * RX Half Connection methods
667 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
668 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
670 static char *ccid3_rx_state_names[] = {
671 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
672 [TFRC_RSTATE_DATA] = "DATA",
673 [TFRC_RSTATE_TERM] = "TERM",
676 return ccid3_rx_state_names[state];
680 static void ccid3_hc_rx_set_state(struct sock *sk,
681 enum ccid3_hc_rx_states state)
683 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
684 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
686 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
687 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
688 ccid3_rx_state_name(state));
689 WARN_ON(state == oldstate);
690 hcrx->ccid3hcrx_state = state;
693 static void ccid3_hc_rx_send_feedback(struct sock *sk)
695 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
696 struct dccp_sock *dp = dccp_sk(sk);
697 struct dccp_rx_hist_entry *packet;
700 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
702 dccp_timestamp(sk, &now);
704 switch (hcrx->ccid3hcrx_state) {
705 case TFRC_RSTATE_NO_DATA:
706 hcrx->ccid3hcrx_x_recv = 0;
708 case TFRC_RSTATE_DATA: {
709 const u32 delta = timeval_delta(&now,
710 &hcrx->ccid3hcrx_tstamp_last_feedback);
711 hcrx->ccid3hcrx_x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv,
715 case TFRC_RSTATE_TERM:
716 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
720 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
721 if (unlikely(packet == NULL)) {
722 DCCP_WARN("%s, sk=%p, no data packet in history!\n",
727 hcrx->ccid3hcrx_tstamp_last_feedback = now;
728 hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
729 hcrx->ccid3hcrx_bytes_recv = 0;
731 /* Convert to multiples of 10us */
732 hcrx->ccid3hcrx_elapsed_time =
733 timeval_delta(&now, &packet->dccphrx_tstamp) / 10;
734 if (hcrx->ccid3hcrx_p == 0)
735 hcrx->ccid3hcrx_pinv = ~0;
737 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
738 dp->dccps_hc_rx_insert_options = 1;
742 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
744 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
747 BUG_ON(hcrx == NULL);
749 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
752 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
754 if (dccp_packet_without_ack(skb))
757 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
758 pinv = htonl(hcrx->ccid3hcrx_pinv);
760 if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
761 dccp_insert_option_elapsed_time(sk, skb,
762 hcrx->ccid3hcrx_elapsed_time)) ||
763 dccp_insert_option_timestamp(sk, skb) ||
764 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
765 &pinv, sizeof(pinv)) ||
766 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
767 &x_recv, sizeof(x_recv)))
773 /* calculate first loss interval
775 * returns estimated loss interval in usecs */
777 static u32 ccid3_hc_rx_calc_first_li(struct sock *sk)
779 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
780 struct dccp_rx_hist_entry *entry, *next, *tail = NULL;
781 u32 rtt, delta, x_recv, fval, p, tmp2;
782 struct timeval tstamp = { 0, };
788 list_for_each_entry_safe(entry, next, &hcrx->ccid3hcrx_hist,
790 if (dccp_rx_hist_entry_data_packet(entry)) {
795 tstamp = entry->dccphrx_tstamp;
796 win_count = entry->dccphrx_ccval;
800 interval = win_count - entry->dccphrx_ccval;
802 interval += TFRC_WIN_COUNT_LIMIT;
810 if (unlikely(step == 0)) {
811 DCCP_WARN("%s, sk=%p, packet history has no data packets!\n",
816 if (unlikely(interval == 0)) {
817 DCCP_WARN("%s, sk=%p, Could not find a win_count interval > 0."
818 "Defaulting to 1\n", dccp_role(sk), sk);
823 DCCP_CRIT("tail is null\n");
826 rtt = timeval_delta(&tstamp, &tail->dccphrx_tstamp) * 4 / interval;
827 ccid3_pr_debug("%s, sk=%p, approximated RTT to %uus\n",
828 dccp_role(sk), sk, rtt);
831 DCCP_WARN("RTT==0, setting to 1\n");
835 dccp_timestamp(sk, &tstamp);
836 delta = timeval_delta(&tstamp, &hcrx->ccid3hcrx_tstamp_last_feedback);
837 x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv, delta);
840 x_recv = hcrx->ccid3hcrx_x_recv;
842 tmp1 = (u64)x_recv * (u64)rtt;
843 do_div(tmp1,10000000);
847 DCCP_CRIT("tmp2 = 0, x_recv = %u, rtt =%u\n", x_recv, rtt);
851 fval = (hcrx->ccid3hcrx_s * 100000) / tmp2;
852 /* do not alter order above or you will get overflow on 32 bit */
853 p = tfrc_calc_x_reverse_lookup(fval);
854 ccid3_pr_debug("%s, sk=%p, receive rate=%u bytes/s, implied "
855 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
863 static void ccid3_hc_rx_update_li(struct sock *sk, u64 seq_loss, u8 win_loss)
865 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
866 struct dccp_li_hist_entry *head;
869 if (list_empty(&hcrx->ccid3hcrx_li_hist)) {
870 if (!dccp_li_hist_interval_new(ccid3_li_hist,
871 &hcrx->ccid3hcrx_li_hist, seq_loss, win_loss))
874 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
875 struct dccp_li_hist_entry, dccplih_node);
876 head->dccplih_interval = ccid3_hc_rx_calc_first_li(sk);
878 struct dccp_li_hist_entry *entry;
879 struct list_head *tail;
881 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
882 struct dccp_li_hist_entry, dccplih_node);
883 /* FIXME win count check removed as was wrong */
884 /* should make this check with receive history */
885 /* and compare there as per section 10.2 of RFC4342 */
887 /* new loss event detected */
888 /* calculate last interval length */
889 seq_temp = dccp_delta_seqno(head->dccplih_seqno, seq_loss);
890 entry = dccp_li_hist_entry_new(ccid3_li_hist, SLAB_ATOMIC);
893 DCCP_BUG("out of memory - can not allocate entry");
897 list_add(&entry->dccplih_node, &hcrx->ccid3hcrx_li_hist);
899 tail = hcrx->ccid3hcrx_li_hist.prev;
901 kmem_cache_free(ccid3_li_hist->dccplih_slab, tail);
903 /* Create the newest interval */
904 entry->dccplih_seqno = seq_loss;
905 entry->dccplih_interval = seq_temp;
906 entry->dccplih_win_count = win_loss;
910 static int ccid3_hc_rx_detect_loss(struct sock *sk,
911 struct dccp_rx_hist_entry *packet)
913 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
914 struct dccp_rx_hist_entry *rx_hist = dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
915 u64 seqno = packet->dccphrx_seqno;
921 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
924 follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
925 hcrx->ccid3hcrx_seqno_nonloss = seqno;
926 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
931 while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
932 > TFRC_RECV_NUM_LATE_LOSS) {
934 ccid3_hc_rx_update_li(sk, hcrx->ccid3hcrx_seqno_nonloss,
935 hcrx->ccid3hcrx_ccval_nonloss);
936 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
937 dccp_inc_seqno(&tmp_seqno);
938 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
939 dccp_inc_seqno(&tmp_seqno);
940 while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
941 tmp_seqno, &ccval)) {
942 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
943 hcrx->ccid3hcrx_ccval_nonloss = ccval;
944 dccp_inc_seqno(&tmp_seqno);
948 /* FIXME - this code could be simplified with above while */
949 /* but works at moment */
950 if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
951 hcrx->ccid3hcrx_seqno_nonloss = seqno;
952 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
956 dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
957 &hcrx->ccid3hcrx_li_hist, packet,
958 hcrx->ccid3hcrx_seqno_nonloss);
962 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
964 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
965 const struct dccp_options_received *opt_recv;
966 struct dccp_rx_hist_entry *packet;
969 u32 p_prev, rtt_prev, r_sample, t_elapsed;
972 BUG_ON(hcrx == NULL);
974 opt_recv = &dccp_sk(sk)->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 rtt_prev = hcrx->ccid3hcrx_rtt;
984 dccp_timestamp(sk, &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 DCCP_WARN("r_sample=%uus, t_elapsed=%uus\n",
991 r_sample, t_elapsed);
993 r_sample -= t_elapsed;
995 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
996 hcrx->ccid3hcrx_rtt = r_sample;
998 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
1001 if (rtt_prev != hcrx->ccid3hcrx_rtt)
1002 ccid3_pr_debug("%s, New RTT=%uus, elapsed time=%u\n",
1003 dccp_role(sk), hcrx->ccid3hcrx_rtt,
1004 opt_recv->dccpor_elapsed_time);
1008 default: /* We're not interested in other packet types, move along */
1012 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, sk, opt_recv->dccpor_ndp,
1014 if (unlikely(packet == NULL)) {
1015 DCCP_WARN("%s, sk=%p, Not enough mem to add rx packet "
1016 "to history, consider it lost!\n", dccp_role(sk), sk);
1020 win_count = packet->dccphrx_ccval;
1022 loss = ccid3_hc_rx_detect_loss(sk, packet);
1024 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
1027 switch (hcrx->ccid3hcrx_state) {
1028 case TFRC_RSTATE_NO_DATA:
1029 ccid3_pr_debug("%s, sk=%p(%s), skb=%p, sending initial "
1032 dccp_state_name(sk->sk_state), skb);
1033 ccid3_hc_rx_send_feedback(sk);
1034 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
1036 case TFRC_RSTATE_DATA:
1037 hcrx->ccid3hcrx_bytes_recv += skb->len -
1038 dccp_hdr(skb)->dccph_doff * 4;
1042 dccp_timestamp(sk, &now);
1043 if (timeval_delta(&now, &hcrx->ccid3hcrx_tstamp_last_ack) >=
1044 hcrx->ccid3hcrx_rtt) {
1045 hcrx->ccid3hcrx_tstamp_last_ack = now;
1046 ccid3_hc_rx_send_feedback(sk);
1049 case TFRC_RSTATE_TERM:
1050 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
1054 /* Dealing with packet loss */
1055 ccid3_pr_debug("%s, sk=%p(%s), data loss! Reacting...\n",
1056 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
1058 p_prev = hcrx->ccid3hcrx_p;
1060 /* Calculate loss event rate */
1061 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
1062 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
1064 /* Scaling up by 1000000 as fixed decimal */
1066 hcrx->ccid3hcrx_p = 1000000 / i_mean;
1068 DCCP_BUG("empty loss history");
1070 if (hcrx->ccid3hcrx_p > p_prev) {
1071 ccid3_hc_rx_send_feedback(sk);
1076 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
1078 struct dccp_sock *dp = dccp_sk(sk);
1079 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
1081 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
1083 if (dp->dccps_packet_size >= TFRC_MIN_PACKET_SIZE &&
1084 dp->dccps_packet_size <= TFRC_MAX_PACKET_SIZE)
1085 hcrx->ccid3hcrx_s = dp->dccps_packet_size;
1087 hcrx->ccid3hcrx_s = TFRC_STD_PACKET_SIZE;
1089 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
1090 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
1091 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
1092 dccp_timestamp(sk, &hcrx->ccid3hcrx_tstamp_last_ack);
1093 hcrx->ccid3hcrx_tstamp_last_feedback = hcrx->ccid3hcrx_tstamp_last_ack;
1094 hcrx->ccid3hcrx_rtt = 5000; /* XXX 5ms for now... */
1098 static void ccid3_hc_rx_exit(struct sock *sk)
1100 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1102 BUG_ON(hcrx == NULL);
1104 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
1106 /* Empty packet history */
1107 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
1109 /* Empty loss interval history */
1110 dccp_li_hist_purge(ccid3_li_hist, &hcrx->ccid3hcrx_li_hist);
1113 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
1115 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1117 /* Listen socks doesn't have a private CCID block */
1118 if (sk->sk_state == DCCP_LISTEN)
1121 BUG_ON(hcrx == NULL);
1123 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
1124 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1125 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
1128 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
1130 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)
1136 BUG_ON(hctx == NULL);
1138 info->tcpi_rto = hctx->ccid3hctx_t_rto;
1139 info->tcpi_rtt = hctx->ccid3hctx_rtt;
1142 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
1143 u32 __user *optval, int __user *optlen)
1145 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1148 /* Listen socks doesn't have a private CCID block */
1149 if (sk->sk_state == DCCP_LISTEN)
1153 case DCCP_SOCKOPT_CCID_RX_INFO:
1154 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
1156 len = sizeof(hcrx->ccid3hcrx_tfrc);
1157 val = &hcrx->ccid3hcrx_tfrc;
1160 return -ENOPROTOOPT;
1163 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1169 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
1170 u32 __user *optval, int __user *optlen)
1172 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1175 /* Listen socks doesn't have a private CCID block */
1176 if (sk->sk_state == DCCP_LISTEN)
1180 case DCCP_SOCKOPT_CCID_TX_INFO:
1181 if (len < sizeof(hctx->ccid3hctx_tfrc))
1183 len = sizeof(hctx->ccid3hctx_tfrc);
1184 val = &hctx->ccid3hctx_tfrc;
1187 return -ENOPROTOOPT;
1190 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1196 static struct ccid_operations ccid3 = {
1197 .ccid_id = DCCPC_CCID3,
1198 .ccid_name = "ccid3",
1199 .ccid_owner = THIS_MODULE,
1200 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
1201 .ccid_hc_tx_init = ccid3_hc_tx_init,
1202 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1203 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1204 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1205 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1206 .ccid_hc_tx_insert_options = ccid3_hc_tx_insert_options,
1207 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1208 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
1209 .ccid_hc_rx_init = ccid3_hc_rx_init,
1210 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1211 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1212 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1213 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1214 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1215 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1216 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1219 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
1220 module_param(ccid3_debug, int, 0444);
1221 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1224 static __init int ccid3_module_init(void)
1228 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1229 if (ccid3_rx_hist == NULL)
1232 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1233 if (ccid3_tx_hist == NULL)
1236 ccid3_li_hist = dccp_li_hist_new("ccid3");
1237 if (ccid3_li_hist == NULL)
1240 rc = ccid_register(&ccid3);
1242 goto out_free_loss_interval_history;
1246 out_free_loss_interval_history:
1247 dccp_li_hist_delete(ccid3_li_hist);
1248 ccid3_li_hist = NULL;
1250 dccp_tx_hist_delete(ccid3_tx_hist);
1251 ccid3_tx_hist = NULL;
1253 dccp_rx_hist_delete(ccid3_rx_hist);
1254 ccid3_rx_hist = NULL;
1257 module_init(ccid3_module_init);
1259 static __exit void ccid3_module_exit(void)
1261 ccid_unregister(&ccid3);
1263 if (ccid3_tx_hist != NULL) {
1264 dccp_tx_hist_delete(ccid3_tx_hist);
1265 ccid3_tx_hist = NULL;
1267 if (ccid3_rx_hist != NULL) {
1268 dccp_rx_hist_delete(ccid3_rx_hist);
1269 ccid3_rx_hist = NULL;
1271 if (ccid3_li_hist != NULL) {
1272 dccp_li_hist_delete(ccid3_li_hist);
1273 ccid3_li_hist = NULL;
1276 module_exit(ccid3_module_exit);
1278 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
1279 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1280 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1281 MODULE_LICENSE("GPL");
1282 MODULE_ALIAS("net-dccp-ccid-3");