2 * net/dccp/ccids/ccid3.c
4 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
5 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
6 * Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
8 * An implementation of the DCCP protocol
10 * This code has been developed by the University of Waikato WAND
11 * research group. For further information please see http://www.wand.net.nz/
13 * This code also uses code from Lulea University, rereleased as GPL by its
15 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
17 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
18 * and to make it work as a loadable module in the DCCP stack written by
19 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
21 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
23 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
33 * You should have received a copy of the GNU General Public License
34 * along with this program; if not, write to the Free Software
35 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
40 #include <asm/unaligned.h>
42 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
43 static int ccid3_debug;
44 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
46 #define ccid3_pr_debug(format, a...)
50 * Transmitter Half-Connection Routines
52 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
53 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
55 static char *ccid3_state_names[] = {
56 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
57 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
58 [TFRC_SSTATE_FBACK] = "FBACK",
59 [TFRC_SSTATE_TERM] = "TERM",
62 return ccid3_state_names[state];
66 static void ccid3_hc_tx_set_state(struct sock *sk,
67 enum ccid3_hc_tx_states state)
69 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
70 enum ccid3_hc_tx_states oldstate = hctx->state;
72 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
73 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
74 ccid3_tx_state_name(state));
75 WARN_ON(state == oldstate);
80 * Compute the initial sending rate X_init in the manner of RFC 3390:
82 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
84 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
85 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
86 * For consistency with other parts of the code, X_init is scaled by 2^6.
88 static inline u64 rfc3390_initial_rate(struct sock *sk)
90 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
91 const __u32 w_init = clamp_t(__u32, 4380U, 2 * hctx->s, 4 * hctx->s);
93 return scaled_div(w_init << 6, hctx->rtt);
97 * ccid3_update_send_interval - Calculate new t_ipi = s / X_inst
98 * This respects the granularity of X_inst (64 * bytes/second).
100 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
102 hctx->t_ipi = scaled_div32(((u64)hctx->s) << 6, hctx->x);
104 ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hctx->t_ipi,
105 hctx->s, (unsigned)(hctx->x >> 6));
108 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
110 u32 delta = ktime_us_delta(now, hctx->t_last_win_count);
112 return delta / hctx->rtt;
116 * ccid3_hc_tx_update_x - Update allowed sending rate X
117 * @stamp: most recent time if available - can be left NULL.
118 * This function tracks draft rfc3448bis, check there for latest details.
120 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
121 * fine-grained resolution of sending rates. This requires scaling by 2^6
122 * throughout the code. Only X_calc is unscaled (in bytes/second).
125 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
127 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
128 u64 min_rate = 2 * hctx->x_recv;
129 const u64 old_x = hctx->x;
130 ktime_t now = stamp ? *stamp : ktime_get_real();
133 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
134 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
135 * a sender is idle if it has not sent anything over a 2-RTT-period.
136 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
138 if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) {
139 min_rate = rfc3390_initial_rate(sk);
140 min_rate = max(min_rate, 2 * hctx->x_recv);
145 hctx->x = min(((u64)hctx->x_calc) << 6, min_rate);
146 hctx->x = max(hctx->x, (((u64)hctx->s) << 6) / TFRC_T_MBI);
148 } else if (ktime_us_delta(now, hctx->t_ld) - (s64)hctx->rtt >= 0) {
150 hctx->x = min(2 * hctx->x, min_rate);
151 hctx->x = max(hctx->x,
152 scaled_div(((u64)hctx->s) << 6, hctx->rtt));
156 if (hctx->x != old_x) {
157 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
158 "X_recv=%u\n", (unsigned)(old_x >> 6),
159 (unsigned)(hctx->x >> 6), hctx->x_calc,
160 (unsigned)(hctx->x_recv >> 6));
162 ccid3_update_send_interval(hctx);
167 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
168 * @len: DCCP packet payload size in bytes
170 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
172 const u16 old_s = hctx->s;
174 hctx->s = tfrc_ewma(hctx->s, len, 9);
176 if (hctx->s != old_s)
177 ccid3_update_send_interval(hctx);
181 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
182 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
184 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
187 u32 delta = ktime_us_delta(now, hctx->t_last_win_count),
188 quarter_rtts = (4 * delta) / hctx->rtt;
190 if (quarter_rtts > 0) {
191 hctx->t_last_win_count = now;
192 hctx->last_win_count += min(quarter_rtts, 5U);
193 hctx->last_win_count &= 0xF; /* mod 16 */
197 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
199 struct sock *sk = (struct sock *)data;
200 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
201 unsigned long t_nfb = USEC_PER_SEC / 5;
204 if (sock_owned_by_user(sk)) {
205 /* Try again later. */
206 /* XXX: set some sensible MIB */
210 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
211 ccid3_tx_state_name(hctx->state));
213 if (hctx->state == TFRC_SSTATE_FBACK)
214 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
215 else if (hctx->state != TFRC_SSTATE_NO_FBACK)
219 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
220 * RTO is 0 if and only if no feedback has been received yet.
222 if (hctx->t_rto == 0 || hctx->p == 0) {
224 /* halve send rate directly */
225 hctx->x = max(hctx->x / 2, (((u64)hctx->s) << 6) / TFRC_T_MBI);
226 ccid3_update_send_interval(hctx);
229 * Modify the cached value of X_recv
231 * If (X_calc > 2 * X_recv)
232 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
234 * X_recv = X_calc / 4;
236 * Note that X_recv is scaled by 2^6 while X_calc is not
238 BUG_ON(hctx->p && !hctx->x_calc);
240 if (hctx->x_calc > (hctx->x_recv >> 5))
242 max(hctx->x_recv / 2,
243 (((__u64)hctx->s) << 6) / (2 * TFRC_T_MBI));
245 hctx->x_recv = hctx->x_calc;
248 ccid3_hc_tx_update_x(sk, NULL);
250 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
251 (unsigned long long)hctx->x);
254 * Set new timeout for the nofeedback timer.
255 * See comments in packet_recv() regarding the value of t_RTO.
257 if (unlikely(hctx->t_rto == 0)) /* no feedback received yet */
258 t_nfb = TFRC_INITIAL_TIMEOUT;
260 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
263 sk_reset_timer(sk, &hctx->no_feedback_timer,
264 jiffies + usecs_to_jiffies(t_nfb));
272 * > 0: delay (in msecs) that should pass before actually sending
273 * = 0: can send immediately
274 * < 0: error condition; do not send packet
276 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
278 struct dccp_sock *dp = dccp_sk(sk);
279 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
280 ktime_t now = ktime_get_real();
284 * This function is called only for Data and DataAck packets. Sending
285 * zero-sized Data(Ack)s is theoretically possible, but for congestion
286 * control this case is pathological - ignore it.
288 if (unlikely(skb->len == 0))
291 switch (hctx->state) {
292 case TFRC_SSTATE_NO_SENT:
293 sk_reset_timer(sk, &hctx->no_feedback_timer, (jiffies +
294 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
295 hctx->last_win_count = 0;
296 hctx->t_last_win_count = now;
298 /* Set t_0 for initial packet */
304 * Use initial RTT sample when available: recommended by erratum
305 * to RFC 4342. This implements the initialisation procedure of
306 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
308 if (dp->dccps_syn_rtt) {
309 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
310 hctx->rtt = dp->dccps_syn_rtt;
311 hctx->x = rfc3390_initial_rate(sk);
315 * Sender does not have RTT sample:
316 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
317 * is needed in several parts (e.g. window counter);
318 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
320 hctx->rtt = DCCP_FALLBACK_RTT;
324 ccid3_update_send_interval(hctx);
326 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
328 case TFRC_SSTATE_NO_FBACK:
329 case TFRC_SSTATE_FBACK:
330 delay = ktime_us_delta(hctx->t_nom, now);
331 ccid3_pr_debug("delay=%ld\n", (long)delay);
333 * Scheduling of packet transmissions [RFC 3448, 4.6]
335 * if (t_now > t_nom - delta)
336 * // send the packet now
338 * // send the packet in (t_nom - t_now) milliseconds.
340 if (delay >= TFRC_T_DELTA)
341 return (u32)delay / USEC_PER_MSEC;
343 ccid3_hc_tx_update_win_count(hctx, now);
345 case TFRC_SSTATE_TERM:
346 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
350 /* prepare to send now (add options etc.) */
351 dp->dccps_hc_tx_insert_options = 1;
352 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->last_win_count;
354 /* set the nominal send time for the next following packet */
355 hctx->t_nom = ktime_add_us(hctx->t_nom, hctx->t_ipi);
359 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
362 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
364 ccid3_hc_tx_update_s(hctx, len);
366 if (tfrc_tx_hist_add(&hctx->hist, dccp_sk(sk)->dccps_gss))
367 DCCP_CRIT("packet history - out of memory!");
370 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
372 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
373 struct ccid3_options_received *opt_recv = &hctx->options_received;
374 struct tfrc_tx_hist_entry *acked;
379 /* we are only interested in ACKs */
380 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
381 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
383 /* ... and only in the established state */
384 if (hctx->state != TFRC_SSTATE_FBACK &&
385 hctx->state != TFRC_SSTATE_NO_FBACK)
389 * Locate the acknowledged packet in the TX history.
391 * Returning "entry not found" here can for instance happen when
392 * - the host has not sent out anything (e.g. a passive server),
393 * - the Ack is outdated (packet with higher Ack number was received),
394 * - it is a bogus Ack (for a packet not sent on this connection).
396 acked = tfrc_tx_hist_find_entry(hctx->hist, dccp_hdr_ack_seq(skb));
399 /* For the sake of RTT sampling, ignore/remove all older entries */
400 tfrc_tx_hist_purge(&acked->next);
402 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
403 now = ktime_get_real();
404 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
405 hctx->rtt = tfrc_ewma(hctx->rtt, r_sample, 9);
407 /* Update receive rate in units of 64 * bytes/second */
408 hctx->x_recv = opt_recv->ccid3or_receive_rate;
411 /* Update loss event rate (which is scaled by 1e6) */
412 pinv = opt_recv->ccid3or_loss_event_rate;
413 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
415 else /* can not exceed 100% */
416 hctx->p = scaled_div(1, pinv);
419 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
421 if (hctx->state == TFRC_SSTATE_NO_FBACK) {
422 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
424 if (hctx->t_rto == 0) {
426 * Initial feedback packet: Larger Initial Windows (4.2)
428 hctx->x = rfc3390_initial_rate(sk);
431 ccid3_update_send_interval(hctx);
433 goto done_computing_x;
434 } else if (hctx->p == 0) {
436 * First feedback after nofeedback timer expiry (4.3)
438 goto done_computing_x;
442 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
444 hctx->x_calc = tfrc_calc_x(hctx->s, hctx->rtt, hctx->p);
445 ccid3_hc_tx_update_x(sk, &now);
448 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
449 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
450 dccp_role(sk), sk, hctx->rtt, r_sample,
451 hctx->s, hctx->p, hctx->x_calc,
452 (unsigned)(hctx->x_recv >> 6),
453 (unsigned)(hctx->x >> 6));
455 /* unschedule no feedback timer */
456 sk_stop_timer(sk, &hctx->no_feedback_timer);
459 * As we have calculated new ipi, delta, t_nom it is possible
460 * that we now can send a packet, so wake up dccp_wait_for_ccid
462 sk->sk_write_space(sk);
465 * Update timeout interval for the nofeedback timer.
466 * We use a configuration option to increase the lower bound.
467 * This can help avoid triggering the nofeedback timer too
468 * often ('spinning') on LANs with small RTTs.
470 hctx->t_rto = max_t(u32, 4 * hctx->rtt, (CONFIG_IP_DCCP_CCID3_RTO *
471 (USEC_PER_SEC / 1000)));
473 * Schedule no feedback timer to expire in
474 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
476 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
478 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
479 "expire in %lu jiffies (%luus)\n",
480 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
482 sk_reset_timer(sk, &hctx->no_feedback_timer,
483 jiffies + usecs_to_jiffies(t_nfb));
486 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
487 unsigned char len, u16 idx,
488 unsigned char *value)
491 const struct dccp_sock *dp = dccp_sk(sk);
492 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
493 struct ccid3_options_received *opt_recv = &hctx->options_received;
496 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
497 opt_recv->ccid3or_seqno = dp->dccps_gsr;
498 opt_recv->ccid3or_loss_event_rate = ~0;
499 opt_recv->ccid3or_loss_intervals_idx = 0;
500 opt_recv->ccid3or_loss_intervals_len = 0;
501 opt_recv->ccid3or_receive_rate = 0;
505 case TFRC_OPT_LOSS_EVENT_RATE:
506 if (unlikely(len != 4)) {
507 DCCP_WARN("%s(%p), invalid len %d "
508 "for TFRC_OPT_LOSS_EVENT_RATE\n",
509 dccp_role(sk), sk, len);
512 opt_val = get_unaligned((__be32 *)value);
513 opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
514 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
516 opt_recv->ccid3or_loss_event_rate);
519 case TFRC_OPT_LOSS_INTERVALS:
520 opt_recv->ccid3or_loss_intervals_idx = idx;
521 opt_recv->ccid3or_loss_intervals_len = len;
522 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
524 opt_recv->ccid3or_loss_intervals_idx,
525 opt_recv->ccid3or_loss_intervals_len);
527 case TFRC_OPT_RECEIVE_RATE:
528 if (unlikely(len != 4)) {
529 DCCP_WARN("%s(%p), invalid len %d "
530 "for TFRC_OPT_RECEIVE_RATE\n",
531 dccp_role(sk), sk, len);
534 opt_val = get_unaligned((__be32 *)value);
535 opt_recv->ccid3or_receive_rate = ntohl(opt_val);
536 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
538 opt_recv->ccid3or_receive_rate);
546 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
548 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
550 hctx->state = TFRC_SSTATE_NO_SENT;
552 setup_timer(&hctx->no_feedback_timer,
553 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
557 static void ccid3_hc_tx_exit(struct sock *sk)
559 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
561 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
562 sk_stop_timer(sk, &hctx->no_feedback_timer);
564 tfrc_tx_hist_purge(&hctx->hist);
567 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
569 info->tcpi_rto = ccid3_hc_tx_sk(sk)->t_rto;
570 info->tcpi_rtt = ccid3_hc_tx_sk(sk)->rtt;
573 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
574 u32 __user *optval, int __user *optlen)
576 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
577 struct tfrc_tx_info tfrc;
581 case DCCP_SOCKOPT_CCID_TX_INFO:
582 if (len < sizeof(tfrc))
584 tfrc.tfrctx_x = hctx->x;
585 tfrc.tfrctx_x_recv = hctx->x_recv;
586 tfrc.tfrctx_x_calc = hctx->x_calc;
587 tfrc.tfrctx_rtt = hctx->rtt;
588 tfrc.tfrctx_p = hctx->p;
589 tfrc.tfrctx_rto = hctx->t_rto;
590 tfrc.tfrctx_ipi = hctx->t_ipi;
598 if (put_user(len, optlen) || copy_to_user(optval, val, len))
605 * Receiver Half-Connection Routines
608 /* CCID3 feedback types */
609 enum ccid3_fback_type {
610 CCID3_FBACK_NONE = 0,
612 CCID3_FBACK_PERIODIC,
613 CCID3_FBACK_PARAM_CHANGE
616 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
617 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
619 static char *ccid3_rx_state_names[] = {
620 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
621 [TFRC_RSTATE_DATA] = "DATA",
622 [TFRC_RSTATE_TERM] = "TERM",
625 return ccid3_rx_state_names[state];
629 static void ccid3_hc_rx_set_state(struct sock *sk,
630 enum ccid3_hc_rx_states state)
632 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
633 enum ccid3_hc_rx_states oldstate = hcrx->state;
635 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
636 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
637 ccid3_rx_state_name(state));
638 WARN_ON(state == oldstate);
642 static void ccid3_hc_rx_send_feedback(struct sock *sk,
643 const struct sk_buff *skb,
644 enum ccid3_fback_type fbtype)
646 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
647 struct dccp_sock *dp = dccp_sk(sk);
651 if (unlikely(hcrx->state == TFRC_RSTATE_TERM))
654 now = ktime_get_real();
657 case CCID3_FBACK_INITIAL:
659 hcrx->p_inverse = ~0U; /* see RFC 4342, 8.5 */
661 case CCID3_FBACK_PARAM_CHANGE:
663 * When parameters change (new loss or p > p_prev), we do not
664 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
665 * need to reuse the previous value of X_recv. However, when
666 * X_recv was 0 (due to early loss), this would kill X down to
667 * s/t_mbi (i.e. one packet in 64 seconds).
668 * To avoid such drastic reduction, we approximate X_recv as
669 * the number of bytes since last feedback.
670 * This is a safe fallback, since X is bounded above by X_calc.
672 if (hcrx->x_recv > 0)
675 case CCID3_FBACK_PERIODIC:
676 delta = ktime_us_delta(now, hcrx->tstamp_last_feedback);
678 DCCP_BUG("delta (%ld) <= 0", (long)delta);
680 hcrx->x_recv = scaled_div32(hcrx->bytes_recv, delta);
686 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n",
687 (long)delta, hcrx->x_recv, hcrx->p_inverse);
689 hcrx->tstamp_last_feedback = now;
690 hcrx->last_counter = dccp_hdr(skb)->dccph_ccval;
691 hcrx->bytes_recv = 0;
693 dp->dccps_hc_rx_insert_options = 1;
697 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
699 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
702 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
705 if (dccp_packet_without_ack(skb))
708 x_recv = htonl(hcrx->x_recv);
709 pinv = htonl(hcrx->p_inverse);
711 if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
712 &pinv, sizeof(pinv)) ||
713 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
714 &x_recv, sizeof(x_recv)))
720 /** ccid3_first_li - Implements [RFC 3448, 6.3.1]
722 * Determine the length of the first loss interval via inverse lookup.
723 * Assume that X_recv can be computed by the throughput equation
727 * Find some p such that f(p) = fval; return 1/p (scaled).
729 static u32 ccid3_first_li(struct sock *sk)
731 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
732 u32 x_recv, p, delta;
735 if (hcrx->rtt == 0) {
736 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
737 hcrx->rtt = DCCP_FALLBACK_RTT;
740 delta = ktime_to_us(net_timedelta(hcrx->tstamp_last_feedback));
741 x_recv = scaled_div32(hcrx->bytes_recv, delta);
742 if (x_recv == 0) { /* would also trigger divide-by-zero */
743 DCCP_WARN("X_recv==0\n");
744 if (hcrx->x_recv == 0) {
745 DCCP_BUG("stored value of X_recv is zero");
748 x_recv = hcrx->x_recv;
751 fval = scaled_div(hcrx->s, hcrx->rtt);
752 fval = scaled_div32(fval, x_recv);
753 p = tfrc_calc_x_reverse_lookup(fval);
755 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
756 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
758 return p == 0 ? ~0U : scaled_div(1, p);
761 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
763 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
764 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
765 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
766 const bool is_data_packet = dccp_data_packet(skb);
768 if (unlikely(hcrx->state == TFRC_RSTATE_NO_DATA)) {
769 if (is_data_packet) {
770 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
771 do_feedback = CCID3_FBACK_INITIAL;
772 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
775 * Not necessary to update bytes_recv here,
776 * since X_recv = 0 for the first feedback packet (cf.
777 * RFC 3448, 6.3) -- gerrit
783 if (tfrc_rx_hist_duplicate(&hcrx->hist, skb))
784 return; /* done receiving */
786 if (is_data_packet) {
787 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
789 * Update moving-average of s and the sum of received payload bytes
791 hcrx->s = tfrc_ewma(hcrx->s, payload, 9);
792 hcrx->bytes_recv += payload;
796 * Perform loss detection and handle pending losses
798 if (tfrc_rx_handle_loss(&hcrx->hist, &hcrx->li_hist,
799 skb, ndp, ccid3_first_li, sk)) {
800 do_feedback = CCID3_FBACK_PARAM_CHANGE;
804 if (tfrc_rx_hist_loss_pending(&hcrx->hist))
805 return; /* done receiving */
808 * Handle data packets: RTT sampling and monitoring p
810 if (unlikely(!is_data_packet))
813 if (!tfrc_lh_is_initialised(&hcrx->li_hist)) {
814 const u32 sample = tfrc_rx_hist_sample_rtt(&hcrx->hist, skb);
816 * Empty loss history: no loss so far, hence p stays 0.
817 * Sample RTT values, since an RTT estimate is required for the
818 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
821 hcrx->rtt = tfrc_ewma(hcrx->rtt, sample, 9);
823 } else if (tfrc_lh_update_i_mean(&hcrx->li_hist, skb)) {
825 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
826 * has decreased (resp. p has increased), send feedback now.
828 do_feedback = CCID3_FBACK_PARAM_CHANGE;
832 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
834 if (SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->last_counter) > 3)
835 do_feedback = CCID3_FBACK_PERIODIC;
838 tfrc_rx_hist_add_packet(&hcrx->hist, skb, ndp);
842 ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
845 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
847 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
849 hcrx->state = TFRC_RSTATE_NO_DATA;
850 tfrc_lh_init(&hcrx->li_hist);
851 return tfrc_rx_hist_alloc(&hcrx->hist);
854 static void ccid3_hc_rx_exit(struct sock *sk)
856 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
858 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
860 tfrc_rx_hist_purge(&hcrx->hist);
861 tfrc_lh_cleanup(&hcrx->li_hist);
864 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
866 info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->state;
867 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
868 info->tcpi_rcv_rtt = ccid3_hc_rx_sk(sk)->rtt;
871 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
872 u32 __user *optval, int __user *optlen)
874 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
875 struct tfrc_rx_info rx_info;
879 case DCCP_SOCKOPT_CCID_RX_INFO:
880 if (len < sizeof(rx_info))
882 rx_info.tfrcrx_x_recv = hcrx->x_recv;
883 rx_info.tfrcrx_rtt = hcrx->rtt;
884 rx_info.tfrcrx_p = hcrx->p_inverse == 0 ? ~0U :
885 scaled_div(1, hcrx->p_inverse);
886 len = sizeof(rx_info);
893 if (put_user(len, optlen) || copy_to_user(optval, val, len))
899 static struct ccid_operations ccid3 = {
900 .ccid_id = DCCPC_CCID3,
901 .ccid_name = "TCP-Friendly Rate Control",
902 .ccid_owner = THIS_MODULE,
903 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
904 .ccid_hc_tx_init = ccid3_hc_tx_init,
905 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
906 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
907 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
908 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
909 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
910 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
911 .ccid_hc_rx_init = ccid3_hc_rx_init,
912 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
913 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
914 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
915 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
916 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
917 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
918 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
921 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
922 module_param(ccid3_debug, bool, 0644);
923 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
926 static __init int ccid3_module_init(void)
928 return ccid_register(&ccid3);
930 module_init(ccid3_module_init);
932 static __exit void ccid3_module_exit(void)
934 ccid_unregister(&ccid3);
936 module_exit(ccid3_module_exit);
938 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
939 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
940 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
941 MODULE_LICENSE("GPL");
942 MODULE_ALIAS("net-dccp-ccid-3");