2 * net/dccp/ccids/ccid3.c
4 * Copyright (c) 2005 The University of Waikato, Hamilton, New Zealand.
5 * Copyright (c) 2005-6 Ian McDonald <ian.mcdonald@jandi.co.nz>
7 * An implementation of the DCCP protocol
9 * This code has been developed by the University of Waikato WAND
10 * research group. For further information please see http://www.wand.net.nz/
12 * This code also uses code from Lulea University, rereleased as GPL by its
14 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
16 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
17 * and to make it work as a loadable module in the DCCP stack written by
18 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
20 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
22 * This program is free software; you can redistribute it and/or modify
23 * it under the terms of the GNU General Public License as published by
24 * the Free Software Foundation; either version 2 of the License, or
25 * (at your option) any later version.
27 * This program is distributed in the hope that it will be useful,
28 * but WITHOUT ANY WARRANTY; without even the implied warranty of
29 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
30 * GNU General Public License for more details.
32 * You should have received a copy of the GNU General Public License
33 * along with this program; if not, write to the Free Software
34 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
39 #include "lib/packet_history.h"
40 #include "lib/loss_interval.h"
45 * Reason for maths here is to avoid 32 bit overflow when a is big.
46 * With this we get close to the limit.
48 static u32 usecs_div(const u32 a, const u32 b)
50 const u32 div = a < (UINT_MAX / (USEC_PER_SEC / 10)) ? 10 :
51 a < (UINT_MAX / (USEC_PER_SEC / 50)) ? 50 :
52 a < (UINT_MAX / (USEC_PER_SEC / 100)) ? 100 :
53 a < (UINT_MAX / (USEC_PER_SEC / 500)) ? 500 :
54 a < (UINT_MAX / (USEC_PER_SEC / 1000)) ? 1000 :
55 a < (UINT_MAX / (USEC_PER_SEC / 5000)) ? 5000 :
56 a < (UINT_MAX / (USEC_PER_SEC / 10000)) ? 10000 :
57 a < (UINT_MAX / (USEC_PER_SEC / 50000)) ? 50000 :
59 const u32 tmp = a * (USEC_PER_SEC / div);
60 return (b >= 2 * div) ? tmp / (b / div) : tmp;
65 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
66 static int ccid3_debug;
67 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
69 #define ccid3_pr_debug(format, a...)
72 static struct dccp_tx_hist *ccid3_tx_hist;
73 static struct dccp_rx_hist *ccid3_rx_hist;
74 static struct dccp_li_hist *ccid3_li_hist;
76 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
77 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
79 static char *ccid3_state_names[] = {
80 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
81 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
82 [TFRC_SSTATE_FBACK] = "FBACK",
83 [TFRC_SSTATE_TERM] = "TERM",
86 return ccid3_state_names[state];
90 static void ccid3_hc_tx_set_state(struct sock *sk,
91 enum ccid3_hc_tx_states state)
93 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
94 enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
96 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
97 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
98 ccid3_tx_state_name(state));
99 WARN_ON(state == oldstate);
100 hctx->ccid3hctx_state = state;
104 * Recalculate scheduled nominal send time t_nom, inter-packet interval
105 * t_ipi, and delta value. Should be called after each change to X.
107 static inline void ccid3_update_send_time(struct ccid3_hc_tx_sock *hctx)
109 timeval_sub_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
111 /* Calculate new t_ipi (inter packet interval) by t_ipi = s / X_inst */
112 hctx->ccid3hctx_t_ipi = usecs_div(hctx->ccid3hctx_s, hctx->ccid3hctx_x);
114 /* Update nominal send time with regard to the new t_ipi */
115 timeval_add_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
117 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
118 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
119 TFRC_OPSYS_HALF_TIME_GRAN);
124 * x_calc = calcX(s, R, p);
125 * X = max(min(X_calc, 2 * X_recv), s / t_mbi);
127 * If (now - tld >= R)
128 * X = max(min(2 * X, 2 * X_recv), s / R);
131 static void ccid3_hc_tx_update_x(struct sock *sk)
133 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
134 const __u32 old_x = hctx->ccid3hctx_x;
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;
158 if (hctx->ccid3hctx_x != old_x)
159 ccid3_update_send_time(hctx);
162 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
164 struct sock *sk = (struct sock *)data;
165 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
166 unsigned long t_nfb = USEC_PER_SEC / 5;
169 if (sock_owned_by_user(sk)) {
170 /* Try again later. */
171 /* XXX: set some sensible MIB */
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 /* The value of R is still undefined and so we can not recompute
193 * the timout value. Keep initial value as per [RFC 4342, 5]. */
194 t_nfb = TFRC_INITIAL_TIMEOUT;
195 ccid3_update_send_time(hctx);
197 case TFRC_SSTATE_FBACK:
199 * Check if IDLE since last timeout and recv rate is less than
202 if (!hctx->ccid3hctx_idle ||
203 (hctx->ccid3hctx_x_recv >=
204 4 * usecs_div(hctx->ccid3hctx_s, hctx->ccid3hctx_rtt))) {
205 ccid3_pr_debug("%s, sk=%p, state=%s, not idle\n",
207 ccid3_tx_state_name(hctx->ccid3hctx_state));
208 /* Halve sending rate */
210 /* If (X_calc > 2 * X_recv)
211 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
213 * X_recv = X_calc / 4;
215 BUG_ON(hctx->ccid3hctx_p >= TFRC_SMALLEST_P &&
216 hctx->ccid3hctx_x_calc == 0);
218 /* check also if p is zero -> x_calc is infinity? */
219 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P ||
220 hctx->ccid3hctx_x_calc > 2 * hctx->ccid3hctx_x_recv)
221 hctx->ccid3hctx_x_recv = max_t(u32, hctx->ccid3hctx_x_recv / 2,
222 hctx->ccid3hctx_s / (2 * TFRC_MAX_BACK_OFF_TIME));
224 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc / 4;
226 /* Update sending rate */
227 ccid3_hc_tx_update_x(sk);
230 * Schedule no feedback timer to expire in
231 * max(4 * R, 2 * s/X) = max(4 * R, 2 * t_ipi)
233 t_nfb = max(4 * hctx->ccid3hctx_rtt, 2 * hctx->ccid3hctx_t_ipi);
235 case TFRC_SSTATE_NO_SENT:
236 DCCP_BUG("Illegal %s state NO_SENT, sk=%p", dccp_role(sk), sk);
238 case TFRC_SSTATE_TERM:
242 hctx->ccid3hctx_idle = 1;
245 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
246 jiffies + usecs_to_jiffies(t_nfb));
254 * > 0: delay (in msecs) that should pass before actually sending
255 * = 0: can send immediately
256 * < 0: error condition; do not send packet
258 static int ccid3_hc_tx_send_packet(struct sock *sk,
259 struct sk_buff *skb, int len)
261 struct dccp_sock *dp = dccp_sk(sk);
262 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
263 struct dccp_tx_hist_entry *new_packet;
267 BUG_ON(hctx == NULL);
270 * This function is called only for Data and DataAck packets. Sending
271 * zero-sized Data(Ack)s is theoretically possible, but for congestion
272 * control this case is pathological - ignore it.
274 if (unlikely(len == 0))
277 /* See if last packet allocated was not sent */
278 new_packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
279 if (new_packet == NULL || new_packet->dccphtx_sent) {
280 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);
302 /* First timeout, according to [RFC 3448, 4.2], is 1 second */
303 hctx->ccid3hctx_t_ipi = USEC_PER_SEC;
304 /* Initial delta: minimum of 0.5 sec and t_gran/2 */
305 hctx->ccid3hctx_delta = TFRC_OPSYS_HALF_TIME_GRAN;
307 /* Set t_0 for initial packet */
308 hctx->ccid3hctx_t_nom = now;
310 case TFRC_SSTATE_NO_FBACK:
311 case TFRC_SSTATE_FBACK:
312 delay = timeval_delta(&hctx->ccid3hctx_t_nom, &now);
314 * Scheduling of packet transmissions [RFC 3448, 4.6]
316 * if (t_now > t_nom - delta)
317 * // send the packet now
319 * // send the packet in (t_nom - t_now) milliseconds.
321 if (delay >= hctx->ccid3hctx_delta)
322 return delay / 1000L;
324 case TFRC_SSTATE_TERM:
325 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
329 /* prepare to send now (add options etc.) */
330 dp->dccps_hc_tx_insert_options = 1;
331 new_packet->dccphtx_ccval = DCCP_SKB_CB(skb)->dccpd_ccval =
332 hctx->ccid3hctx_last_win_count;
333 timeval_add_usecs(&hctx->ccid3hctx_t_nom, hctx->ccid3hctx_t_ipi);
338 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more, int len)
340 const struct dccp_sock *dp = dccp_sk(sk);
341 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
344 BUG_ON(hctx == NULL);
346 dccp_timestamp(sk, &now);
348 /* check if we have sent a data packet */
350 unsigned long quarter_rtt;
351 struct dccp_tx_hist_entry *packet;
353 packet = dccp_tx_hist_head(&hctx->ccid3hctx_hist);
354 if (unlikely(packet == NULL)) {
355 DCCP_WARN("packet doesn't exist in history!\n");
358 if (unlikely(packet->dccphtx_sent)) {
359 DCCP_WARN("no unsent packet in history!\n");
362 packet->dccphtx_tstamp = now;
363 packet->dccphtx_seqno = dp->dccps_gss;
365 * Check if win_count have changed
366 * Algorithm in "8.1. Window Counter Value" in RFC 4342.
368 quarter_rtt = timeval_delta(&now, &hctx->ccid3hctx_t_last_win_count);
369 if (likely(hctx->ccid3hctx_rtt > 8))
370 quarter_rtt /= hctx->ccid3hctx_rtt / 4;
372 if (quarter_rtt > 0) {
373 hctx->ccid3hctx_t_last_win_count = now;
374 hctx->ccid3hctx_last_win_count = (hctx->ccid3hctx_last_win_count +
375 min_t(unsigned long, quarter_rtt, 5)) % 16;
376 ccid3_pr_debug("%s, sk=%p, window changed from "
379 packet->dccphtx_ccval,
380 hctx->ccid3hctx_last_win_count);
383 hctx->ccid3hctx_idle = 0;
384 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
385 packet->dccphtx_sent = 1;
387 ccid3_pr_debug("%s, sk=%p, seqno=%llu NOT inserted!\n",
388 dccp_role(sk), sk, dp->dccps_gss);
391 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
393 const struct dccp_sock *dp = dccp_sk(sk);
394 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
395 struct ccid3_options_received *opt_recv;
396 struct dccp_tx_hist_entry *packet;
404 BUG_ON(hctx == NULL);
406 /* we are only interested in ACKs */
407 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
408 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
411 opt_recv = &hctx->ccid3hctx_options_received;
413 t_elapsed = dp->dccps_options_received.dccpor_elapsed_time * 10;
414 x_recv = opt_recv->ccid3or_receive_rate;
415 pinv = opt_recv->ccid3or_loss_event_rate;
417 switch (hctx->ccid3hctx_state) {
418 case TFRC_SSTATE_NO_FBACK:
419 case TFRC_SSTATE_FBACK:
420 /* Calculate new round trip sample by
421 * R_sample = (now - t_recvdata) - t_delay */
422 /* get t_recvdata from history */
423 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
424 DCCP_SKB_CB(skb)->dccpd_ack_seq);
425 if (unlikely(packet == NULL)) {
426 DCCP_WARN("%s, sk=%p, seqno %llu(%s) does't exist "
427 "in history!\n", dccp_role(sk), sk,
428 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
429 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
434 dccp_timestamp(sk, &now);
435 r_sample = timeval_delta(&now, &packet->dccphtx_tstamp);
436 if (unlikely(r_sample <= t_elapsed))
437 DCCP_WARN("r_sample=%uus,t_elapsed=%uus\n",
438 r_sample, t_elapsed);
440 r_sample -= t_elapsed;
442 /* Update RTT estimate by
443 * If (No feedback recv)
446 * R = q * R + (1 - q) * R_sample;
448 * q is a constant, RFC 3448 recomments 0.9
450 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
451 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
452 hctx->ccid3hctx_rtt = r_sample;
454 hctx->ccid3hctx_rtt = (hctx->ccid3hctx_rtt * 9) / 10 +
457 ccid3_pr_debug("%s, sk=%p, New RTT estimate=%uus, "
458 "r_sample=%us\n", dccp_role(sk), sk,
459 hctx->ccid3hctx_rtt, r_sample);
461 /* Update receive rate */
462 hctx->ccid3hctx_x_recv = x_recv;/* X_recv in bytes per sec */
464 /* Update loss event rate */
465 if (pinv == ~0 || pinv == 0)
466 hctx->ccid3hctx_p = 0;
468 hctx->ccid3hctx_p = 1000000 / pinv;
470 if (hctx->ccid3hctx_p < TFRC_SMALLEST_P) {
471 hctx->ccid3hctx_p = TFRC_SMALLEST_P;
472 ccid3_pr_debug("%s, sk=%p, Smallest p used!\n",
477 /* unschedule no feedback timer */
478 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
480 /* Update sending rate (and likely t_ipi, t_nom, and delta) */
481 ccid3_hc_tx_update_x(sk);
483 /* remove all packets older than the one acked from history */
484 dccp_tx_hist_purge_older(ccid3_tx_hist,
485 &hctx->ccid3hctx_hist, packet);
487 * As we have calculated new ipi, delta, t_nom it is possible that
488 * we now can send a packet, so wake up dccp_wait_for_ccid
490 sk->sk_write_space(sk);
493 /* Update timeout interval. We use the alternative variant of
494 * [RFC 3448, 3.1] which sets the upper bound of t_rto to one
495 * second, as it is suggested for TCP (see RFC 2988, 2.4). */
496 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
499 * Schedule no feedback timer to expire in
500 * max(4 * R, 2 * s/X) = max(4 * R, 2 * t_ipi)
502 t_nfb = max(4 * hctx->ccid3hctx_rtt, 2 * hctx->ccid3hctx_t_ipi);
504 ccid3_pr_debug("%s, sk=%p, Scheduled no feedback timer to "
505 "expire in %lu jiffies (%luus)\n",
507 usecs_to_jiffies(t_nfb), t_nfb);
509 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
510 jiffies + usecs_to_jiffies(t_nfb));
513 hctx->ccid3hctx_idle = 1;
515 case TFRC_SSTATE_NO_SENT:
516 DCCP_WARN("Illegal ACK received - no packet has been sent\n");
518 case TFRC_SSTATE_TERM: /* ignore feedback when closing */
523 static int ccid3_hc_tx_insert_options(struct sock *sk, struct sk_buff *skb)
525 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
527 BUG_ON(hctx == NULL);
529 if (sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN)
530 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
534 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
535 unsigned char len, u16 idx,
536 unsigned char *value)
539 const struct dccp_sock *dp = dccp_sk(sk);
540 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
541 struct ccid3_options_received *opt_recv;
543 BUG_ON(hctx == NULL);
545 opt_recv = &hctx->ccid3hctx_options_received;
547 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
548 opt_recv->ccid3or_seqno = dp->dccps_gsr;
549 opt_recv->ccid3or_loss_event_rate = ~0;
550 opt_recv->ccid3or_loss_intervals_idx = 0;
551 opt_recv->ccid3or_loss_intervals_len = 0;
552 opt_recv->ccid3or_receive_rate = 0;
556 case TFRC_OPT_LOSS_EVENT_RATE:
557 if (unlikely(len != 4)) {
558 DCCP_WARN("%s, sk=%p, invalid len %d "
559 "for TFRC_OPT_LOSS_EVENT_RATE\n",
560 dccp_role(sk), sk, len);
563 opt_recv->ccid3or_loss_event_rate = ntohl(*(__be32 *)value);
564 ccid3_pr_debug("%s, sk=%p, LOSS_EVENT_RATE=%u\n",
566 opt_recv->ccid3or_loss_event_rate);
569 case TFRC_OPT_LOSS_INTERVALS:
570 opt_recv->ccid3or_loss_intervals_idx = idx;
571 opt_recv->ccid3or_loss_intervals_len = len;
572 ccid3_pr_debug("%s, sk=%p, LOSS_INTERVALS=(%u, %u)\n",
574 opt_recv->ccid3or_loss_intervals_idx,
575 opt_recv->ccid3or_loss_intervals_len);
577 case TFRC_OPT_RECEIVE_RATE:
578 if (unlikely(len != 4)) {
579 DCCP_WARN("%s, sk=%p, invalid len %d "
580 "for TFRC_OPT_RECEIVE_RATE\n",
581 dccp_role(sk), sk, len);
584 opt_recv->ccid3or_receive_rate = ntohl(*(__be32 *)value);
585 ccid3_pr_debug("%s, sk=%p, RECEIVE_RATE=%u\n",
587 opt_recv->ccid3or_receive_rate);
595 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
597 struct dccp_sock *dp = dccp_sk(sk);
598 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
600 if (dp->dccps_packet_size >= TFRC_MIN_PACKET_SIZE &&
601 dp->dccps_packet_size <= TFRC_MAX_PACKET_SIZE)
602 hctx->ccid3hctx_s = dp->dccps_packet_size;
604 hctx->ccid3hctx_s = TFRC_STD_PACKET_SIZE;
606 /* Set transmission rate to 1 packet per second */
607 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
608 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
609 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
611 hctx->ccid3hctx_no_feedback_timer.function = ccid3_hc_tx_no_feedback_timer;
612 hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
613 init_timer(&hctx->ccid3hctx_no_feedback_timer);
618 static void ccid3_hc_tx_exit(struct sock *sk)
620 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
622 BUG_ON(hctx == NULL);
624 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
625 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
627 /* Empty packet history */
628 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
632 * RX Half Connection methods
635 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
636 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
638 static char *ccid3_rx_state_names[] = {
639 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
640 [TFRC_RSTATE_DATA] = "DATA",
641 [TFRC_RSTATE_TERM] = "TERM",
644 return ccid3_rx_state_names[state];
648 static void ccid3_hc_rx_set_state(struct sock *sk,
649 enum ccid3_hc_rx_states state)
651 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
652 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
654 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
655 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
656 ccid3_rx_state_name(state));
657 WARN_ON(state == oldstate);
658 hcrx->ccid3hcrx_state = state;
661 static void ccid3_hc_rx_send_feedback(struct sock *sk)
663 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
664 struct dccp_sock *dp = dccp_sk(sk);
665 struct dccp_rx_hist_entry *packet;
668 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
670 dccp_timestamp(sk, &now);
672 switch (hcrx->ccid3hcrx_state) {
673 case TFRC_RSTATE_NO_DATA:
674 hcrx->ccid3hcrx_x_recv = 0;
676 case TFRC_RSTATE_DATA: {
677 const u32 delta = timeval_delta(&now,
678 &hcrx->ccid3hcrx_tstamp_last_feedback);
679 hcrx->ccid3hcrx_x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv,
683 case TFRC_RSTATE_TERM:
684 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
688 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
689 if (unlikely(packet == NULL)) {
690 DCCP_WARN("%s, sk=%p, no data packet in history!\n",
695 hcrx->ccid3hcrx_tstamp_last_feedback = now;
696 hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
697 hcrx->ccid3hcrx_bytes_recv = 0;
699 /* Convert to multiples of 10us */
700 hcrx->ccid3hcrx_elapsed_time =
701 timeval_delta(&now, &packet->dccphrx_tstamp) / 10;
702 if (hcrx->ccid3hcrx_p == 0)
703 hcrx->ccid3hcrx_pinv = ~0;
705 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
706 dp->dccps_hc_rx_insert_options = 1;
710 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
712 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
715 BUG_ON(hcrx == NULL);
717 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
720 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
722 if (dccp_packet_without_ack(skb))
725 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
726 pinv = htonl(hcrx->ccid3hcrx_pinv);
728 if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
729 dccp_insert_option_elapsed_time(sk, skb,
730 hcrx->ccid3hcrx_elapsed_time)) ||
731 dccp_insert_option_timestamp(sk, skb) ||
732 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
733 &pinv, sizeof(pinv)) ||
734 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
735 &x_recv, sizeof(x_recv)))
741 /* calculate first loss interval
743 * returns estimated loss interval in usecs */
745 static u32 ccid3_hc_rx_calc_first_li(struct sock *sk)
747 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
748 struct dccp_rx_hist_entry *entry, *next, *tail = NULL;
749 u32 rtt, delta, x_recv, fval, p, tmp2;
750 struct timeval tstamp = { 0, };
756 list_for_each_entry_safe(entry, next, &hcrx->ccid3hcrx_hist,
758 if (dccp_rx_hist_entry_data_packet(entry)) {
763 tstamp = entry->dccphrx_tstamp;
764 win_count = entry->dccphrx_ccval;
768 interval = win_count - entry->dccphrx_ccval;
770 interval += TFRC_WIN_COUNT_LIMIT;
778 if (unlikely(step == 0)) {
779 DCCP_WARN("%s, sk=%p, packet history has no data packets!\n",
784 if (unlikely(interval == 0)) {
785 DCCP_WARN("%s, sk=%p, Could not find a win_count interval > 0."
786 "Defaulting to 1\n", dccp_role(sk), sk);
791 DCCP_CRIT("tail is null\n");
794 rtt = timeval_delta(&tstamp, &tail->dccphrx_tstamp) * 4 / interval;
795 ccid3_pr_debug("%s, sk=%p, approximated RTT to %uus\n",
796 dccp_role(sk), sk, rtt);
799 DCCP_WARN("RTT==0, setting to 1\n");
803 dccp_timestamp(sk, &tstamp);
804 delta = timeval_delta(&tstamp, &hcrx->ccid3hcrx_tstamp_last_feedback);
805 x_recv = usecs_div(hcrx->ccid3hcrx_bytes_recv, delta);
808 x_recv = hcrx->ccid3hcrx_x_recv;
810 tmp1 = (u64)x_recv * (u64)rtt;
811 do_div(tmp1,10000000);
815 DCCP_CRIT("tmp2 = 0, x_recv = %u, rtt =%u\n", x_recv, rtt);
819 fval = (hcrx->ccid3hcrx_s * 100000) / tmp2;
820 /* do not alter order above or you will get overflow on 32 bit */
821 p = tfrc_calc_x_reverse_lookup(fval);
822 ccid3_pr_debug("%s, sk=%p, receive rate=%u bytes/s, implied "
823 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
831 static void ccid3_hc_rx_update_li(struct sock *sk, u64 seq_loss, u8 win_loss)
833 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
834 struct dccp_li_hist_entry *head;
837 if (list_empty(&hcrx->ccid3hcrx_li_hist)) {
838 if (!dccp_li_hist_interval_new(ccid3_li_hist,
839 &hcrx->ccid3hcrx_li_hist, seq_loss, win_loss))
842 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
843 struct dccp_li_hist_entry, dccplih_node);
844 head->dccplih_interval = ccid3_hc_rx_calc_first_li(sk);
846 struct dccp_li_hist_entry *entry;
847 struct list_head *tail;
849 head = list_entry(hcrx->ccid3hcrx_li_hist.next,
850 struct dccp_li_hist_entry, dccplih_node);
851 /* FIXME win count check removed as was wrong */
852 /* should make this check with receive history */
853 /* and compare there as per section 10.2 of RFC4342 */
855 /* new loss event detected */
856 /* calculate last interval length */
857 seq_temp = dccp_delta_seqno(head->dccplih_seqno, seq_loss);
858 entry = dccp_li_hist_entry_new(ccid3_li_hist, SLAB_ATOMIC);
861 DCCP_BUG("out of memory - can not allocate entry");
865 list_add(&entry->dccplih_node, &hcrx->ccid3hcrx_li_hist);
867 tail = hcrx->ccid3hcrx_li_hist.prev;
869 kmem_cache_free(ccid3_li_hist->dccplih_slab, tail);
871 /* Create the newest interval */
872 entry->dccplih_seqno = seq_loss;
873 entry->dccplih_interval = seq_temp;
874 entry->dccplih_win_count = win_loss;
878 static int ccid3_hc_rx_detect_loss(struct sock *sk,
879 struct dccp_rx_hist_entry *packet)
881 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
882 struct dccp_rx_hist_entry *rx_hist = dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
883 u64 seqno = packet->dccphrx_seqno;
889 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
892 follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
893 hcrx->ccid3hcrx_seqno_nonloss = seqno;
894 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
899 while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
900 > TFRC_RECV_NUM_LATE_LOSS) {
902 ccid3_hc_rx_update_li(sk, hcrx->ccid3hcrx_seqno_nonloss,
903 hcrx->ccid3hcrx_ccval_nonloss);
904 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
905 dccp_inc_seqno(&tmp_seqno);
906 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
907 dccp_inc_seqno(&tmp_seqno);
908 while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
909 tmp_seqno, &ccval)) {
910 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
911 hcrx->ccid3hcrx_ccval_nonloss = ccval;
912 dccp_inc_seqno(&tmp_seqno);
916 /* FIXME - this code could be simplified with above while */
917 /* but works at moment */
918 if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
919 hcrx->ccid3hcrx_seqno_nonloss = seqno;
920 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
924 dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
925 &hcrx->ccid3hcrx_li_hist, packet,
926 hcrx->ccid3hcrx_seqno_nonloss);
930 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
932 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
933 const struct dccp_options_received *opt_recv;
934 struct dccp_rx_hist_entry *packet;
936 u32 p_prev, rtt_prev, r_sample, t_elapsed;
939 BUG_ON(hcrx == NULL);
941 opt_recv = &dccp_sk(sk)->dccps_options_received;
943 switch (DCCP_SKB_CB(skb)->dccpd_type) {
945 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
947 case DCCP_PKT_DATAACK:
948 if (opt_recv->dccpor_timestamp_echo == 0)
950 rtt_prev = hcrx->ccid3hcrx_rtt;
951 dccp_timestamp(sk, &now);
952 timeval_sub_usecs(&now, opt_recv->dccpor_timestamp_echo * 10);
953 r_sample = timeval_usecs(&now);
954 t_elapsed = opt_recv->dccpor_elapsed_time * 10;
956 if (unlikely(r_sample <= t_elapsed))
957 DCCP_WARN("r_sample=%uus, t_elapsed=%uus\n",
958 r_sample, t_elapsed);
960 r_sample -= t_elapsed;
962 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
963 hcrx->ccid3hcrx_rtt = r_sample;
965 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
968 if (rtt_prev != hcrx->ccid3hcrx_rtt)
969 ccid3_pr_debug("%s, New RTT=%uus, elapsed time=%u\n",
970 dccp_role(sk), hcrx->ccid3hcrx_rtt,
971 opt_recv->dccpor_elapsed_time);
975 default: /* We're not interested in other packet types, move along */
979 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, sk, opt_recv->dccpor_ndp,
981 if (unlikely(packet == NULL)) {
982 DCCP_WARN("%s, sk=%p, Not enough mem to add rx packet "
983 "to history, consider it lost!\n", dccp_role(sk), sk);
987 loss = ccid3_hc_rx_detect_loss(sk, packet);
989 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
992 switch (hcrx->ccid3hcrx_state) {
993 case TFRC_RSTATE_NO_DATA:
994 ccid3_pr_debug("%s, sk=%p(%s), skb=%p, sending initial "
997 dccp_state_name(sk->sk_state), skb);
998 ccid3_hc_rx_send_feedback(sk);
999 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
1001 case TFRC_RSTATE_DATA:
1002 hcrx->ccid3hcrx_bytes_recv += skb->len -
1003 dccp_hdr(skb)->dccph_doff * 4;
1007 dccp_timestamp(sk, &now);
1008 if (timeval_delta(&now, &hcrx->ccid3hcrx_tstamp_last_ack) >=
1009 hcrx->ccid3hcrx_rtt) {
1010 hcrx->ccid3hcrx_tstamp_last_ack = now;
1011 ccid3_hc_rx_send_feedback(sk);
1014 case TFRC_RSTATE_TERM:
1015 DCCP_BUG("Illegal %s state TERM, sk=%p", dccp_role(sk), sk);
1019 /* Dealing with packet loss */
1020 ccid3_pr_debug("%s, sk=%p(%s), data loss! Reacting...\n",
1021 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
1023 p_prev = hcrx->ccid3hcrx_p;
1025 /* Calculate loss event rate */
1026 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
1027 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
1029 /* Scaling up by 1000000 as fixed decimal */
1031 hcrx->ccid3hcrx_p = 1000000 / i_mean;
1033 DCCP_BUG("empty loss history");
1035 if (hcrx->ccid3hcrx_p > p_prev) {
1036 ccid3_hc_rx_send_feedback(sk);
1041 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
1043 struct dccp_sock *dp = dccp_sk(sk);
1044 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
1046 ccid3_pr_debug("%s, sk=%p\n", dccp_role(sk), sk);
1048 if (dp->dccps_packet_size >= TFRC_MIN_PACKET_SIZE &&
1049 dp->dccps_packet_size <= TFRC_MAX_PACKET_SIZE)
1050 hcrx->ccid3hcrx_s = dp->dccps_packet_size;
1052 hcrx->ccid3hcrx_s = TFRC_STD_PACKET_SIZE;
1054 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
1055 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
1056 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
1057 dccp_timestamp(sk, &hcrx->ccid3hcrx_tstamp_last_ack);
1058 hcrx->ccid3hcrx_tstamp_last_feedback = hcrx->ccid3hcrx_tstamp_last_ack;
1059 hcrx->ccid3hcrx_rtt = 5000; /* XXX 5ms for now... */
1063 static void ccid3_hc_rx_exit(struct sock *sk)
1065 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1067 BUG_ON(hcrx == NULL);
1069 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
1071 /* Empty packet history */
1072 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
1074 /* Empty loss interval history */
1075 dccp_li_hist_purge(ccid3_li_hist, &hcrx->ccid3hcrx_li_hist);
1078 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
1080 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1082 /* Listen socks doesn't have a private CCID block */
1083 if (sk->sk_state == DCCP_LISTEN)
1086 BUG_ON(hcrx == NULL);
1088 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
1089 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1090 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
1093 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
1095 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1097 /* Listen socks doesn't have a private CCID block */
1098 if (sk->sk_state == DCCP_LISTEN)
1101 BUG_ON(hctx == NULL);
1103 info->tcpi_rto = hctx->ccid3hctx_t_rto;
1104 info->tcpi_rtt = hctx->ccid3hctx_rtt;
1107 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
1108 u32 __user *optval, int __user *optlen)
1110 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1113 /* Listen socks doesn't have a private CCID block */
1114 if (sk->sk_state == DCCP_LISTEN)
1118 case DCCP_SOCKOPT_CCID_RX_INFO:
1119 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
1121 len = sizeof(hcrx->ccid3hcrx_tfrc);
1122 val = &hcrx->ccid3hcrx_tfrc;
1125 return -ENOPROTOOPT;
1128 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1134 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
1135 u32 __user *optval, int __user *optlen)
1137 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
1140 /* Listen socks doesn't have a private CCID block */
1141 if (sk->sk_state == DCCP_LISTEN)
1145 case DCCP_SOCKOPT_CCID_TX_INFO:
1146 if (len < sizeof(hctx->ccid3hctx_tfrc))
1148 len = sizeof(hctx->ccid3hctx_tfrc);
1149 val = &hctx->ccid3hctx_tfrc;
1152 return -ENOPROTOOPT;
1155 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1161 static struct ccid_operations ccid3 = {
1162 .ccid_id = DCCPC_CCID3,
1163 .ccid_name = "ccid3",
1164 .ccid_owner = THIS_MODULE,
1165 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
1166 .ccid_hc_tx_init = ccid3_hc_tx_init,
1167 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1168 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1169 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1170 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1171 .ccid_hc_tx_insert_options = ccid3_hc_tx_insert_options,
1172 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1173 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
1174 .ccid_hc_rx_init = ccid3_hc_rx_init,
1175 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1176 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1177 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1178 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1179 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1180 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1181 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1184 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
1185 module_param(ccid3_debug, int, 0444);
1186 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1189 static __init int ccid3_module_init(void)
1193 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1194 if (ccid3_rx_hist == NULL)
1197 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1198 if (ccid3_tx_hist == NULL)
1201 ccid3_li_hist = dccp_li_hist_new("ccid3");
1202 if (ccid3_li_hist == NULL)
1205 rc = ccid_register(&ccid3);
1207 goto out_free_loss_interval_history;
1211 out_free_loss_interval_history:
1212 dccp_li_hist_delete(ccid3_li_hist);
1213 ccid3_li_hist = NULL;
1215 dccp_tx_hist_delete(ccid3_tx_hist);
1216 ccid3_tx_hist = NULL;
1218 dccp_rx_hist_delete(ccid3_rx_hist);
1219 ccid3_rx_hist = NULL;
1222 module_init(ccid3_module_init);
1224 static __exit void ccid3_module_exit(void)
1226 ccid_unregister(&ccid3);
1228 if (ccid3_tx_hist != NULL) {
1229 dccp_tx_hist_delete(ccid3_tx_hist);
1230 ccid3_tx_hist = NULL;
1232 if (ccid3_rx_hist != NULL) {
1233 dccp_rx_hist_delete(ccid3_rx_hist);
1234 ccid3_rx_hist = NULL;
1236 if (ccid3_li_hist != NULL) {
1237 dccp_li_hist_delete(ccid3_li_hist);
1238 ccid3_li_hist = NULL;
1241 module_exit(ccid3_module_exit);
1243 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
1244 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1245 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1246 MODULE_LICENSE("GPL");
1247 MODULE_ALIAS("net-dccp-ccid-3");