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",
61 return ccid3_state_names[state];
65 static void ccid3_hc_tx_set_state(struct sock *sk,
66 enum ccid3_hc_tx_states state)
68 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
69 enum ccid3_hc_tx_states oldstate = hctx->state;
71 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
72 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
73 ccid3_tx_state_name(state));
74 WARN_ON(state == oldstate);
79 * Compute the initial sending rate X_init in the manner of RFC 3390:
81 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
83 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
84 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
85 * For consistency with other parts of the code, X_init is scaled by 2^6.
87 static inline u64 rfc3390_initial_rate(struct sock *sk)
89 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
90 const __u32 w_init = clamp_t(__u32, 4380U, 2 * hctx->s, 4 * hctx->s);
92 return scaled_div(w_init << 6, hctx->rtt);
96 * ccid3_update_send_interval - Calculate new t_ipi = s / X_inst
97 * This respects the granularity of X_inst (64 * bytes/second).
99 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
101 hctx->t_ipi = scaled_div32(((u64)hctx->s) << 6, hctx->x);
103 ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hctx->t_ipi,
104 hctx->s, (unsigned)(hctx->x >> 6));
107 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
109 u32 delta = ktime_us_delta(now, hctx->t_last_win_count);
111 return delta / hctx->rtt;
115 * ccid3_hc_tx_update_x - Update allowed sending rate X
116 * @stamp: most recent time if available - can be left NULL.
117 * This function tracks draft rfc3448bis, check there for latest details.
119 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
120 * fine-grained resolution of sending rates. This requires scaling by 2^6
121 * throughout the code. Only X_calc is unscaled (in bytes/second).
124 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
126 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
127 u64 min_rate = 2 * hctx->x_recv;
128 const u64 old_x = hctx->x;
129 ktime_t now = stamp ? *stamp : ktime_get_real();
132 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
133 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
134 * a sender is idle if it has not sent anything over a 2-RTT-period.
135 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
137 if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) {
138 min_rate = rfc3390_initial_rate(sk);
139 min_rate = max(min_rate, 2 * hctx->x_recv);
144 hctx->x = min(((u64)hctx->x_calc) << 6, min_rate);
145 hctx->x = max(hctx->x, (((u64)hctx->s) << 6) / TFRC_T_MBI);
147 } else if (ktime_us_delta(now, hctx->t_ld) - (s64)hctx->rtt >= 0) {
149 hctx->x = min(2 * hctx->x, min_rate);
150 hctx->x = max(hctx->x,
151 scaled_div(((u64)hctx->s) << 6, hctx->rtt));
155 if (hctx->x != old_x) {
156 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
157 "X_recv=%u\n", (unsigned)(old_x >> 6),
158 (unsigned)(hctx->x >> 6), hctx->x_calc,
159 (unsigned)(hctx->x_recv >> 6));
161 ccid3_update_send_interval(hctx);
166 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
167 * @len: DCCP packet payload size in bytes
169 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
171 const u16 old_s = hctx->s;
173 hctx->s = tfrc_ewma(hctx->s, len, 9);
175 if (hctx->s != old_s)
176 ccid3_update_send_interval(hctx);
180 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
181 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
183 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
186 u32 delta = ktime_us_delta(now, hctx->t_last_win_count),
187 quarter_rtts = (4 * delta) / hctx->rtt;
189 if (quarter_rtts > 0) {
190 hctx->t_last_win_count = now;
191 hctx->last_win_count += min(quarter_rtts, 5U);
192 hctx->last_win_count &= 0xF; /* mod 16 */
196 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
198 struct sock *sk = (struct sock *)data;
199 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
200 unsigned long t_nfb = USEC_PER_SEC / 5;
203 if (sock_owned_by_user(sk)) {
204 /* Try again later. */
205 /* XXX: set some sensible MIB */
209 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
210 ccid3_tx_state_name(hctx->state));
212 /* Ignore and do not restart after leaving the established state */
213 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
216 /* Reset feedback state to "no feedback received" */
217 if (hctx->state == TFRC_SSTATE_FBACK)
218 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
221 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
222 * RTO is 0 if and only if no feedback has been received yet.
224 if (hctx->t_rto == 0 || hctx->p == 0) {
226 /* halve send rate directly */
227 hctx->x = max(hctx->x / 2, (((u64)hctx->s) << 6) / TFRC_T_MBI);
228 ccid3_update_send_interval(hctx);
231 * Modify the cached value of X_recv
233 * If (X_calc > 2 * X_recv)
234 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
236 * X_recv = X_calc / 4;
238 * Note that X_recv is scaled by 2^6 while X_calc is not
240 BUG_ON(hctx->p && !hctx->x_calc);
242 if (hctx->x_calc > (hctx->x_recv >> 5))
244 max(hctx->x_recv / 2,
245 (((__u64)hctx->s) << 6) / (2 * TFRC_T_MBI));
247 hctx->x_recv = hctx->x_calc;
250 ccid3_hc_tx_update_x(sk, NULL);
252 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
253 (unsigned long long)hctx->x);
256 * Set new timeout for the nofeedback timer.
257 * See comments in packet_recv() regarding the value of t_RTO.
259 if (unlikely(hctx->t_rto == 0)) /* no feedback received yet */
260 t_nfb = TFRC_INITIAL_TIMEOUT;
262 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
265 sk_reset_timer(sk, &hctx->no_feedback_timer,
266 jiffies + usecs_to_jiffies(t_nfb));
274 * > 0: delay (in msecs) that should pass before actually sending
275 * = 0: can send immediately
276 * < 0: error condition; do not send packet
278 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
280 struct dccp_sock *dp = dccp_sk(sk);
281 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
282 ktime_t now = ktime_get_real();
286 * This function is called only for Data and DataAck packets. Sending
287 * zero-sized Data(Ack)s is theoretically possible, but for congestion
288 * control this case is pathological - ignore it.
290 if (unlikely(skb->len == 0))
293 if (hctx->state == TFRC_SSTATE_NO_SENT) {
294 sk_reset_timer(sk, &hctx->no_feedback_timer, (jiffies +
295 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
296 hctx->last_win_count = 0;
297 hctx->t_last_win_count = now;
299 /* Set t_0 for initial packet */
305 * Use initial RTT sample when available: recommended by erratum
306 * to RFC 4342. This implements the initialisation procedure of
307 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
309 if (dp->dccps_syn_rtt) {
310 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
311 hctx->rtt = dp->dccps_syn_rtt;
312 hctx->x = rfc3390_initial_rate(sk);
316 * Sender does not have RTT sample:
317 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
318 * is needed in several parts (e.g. window counter);
319 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
321 hctx->rtt = DCCP_FALLBACK_RTT;
325 ccid3_update_send_interval(hctx);
327 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_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);
346 /* prepare to send now (add options etc.) */
347 dp->dccps_hc_tx_insert_options = 1;
348 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->last_win_count;
350 /* set the nominal send time for the next following packet */
351 hctx->t_nom = ktime_add_us(hctx->t_nom, hctx->t_ipi);
355 static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len)
357 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
359 ccid3_hc_tx_update_s(hctx, len);
361 if (tfrc_tx_hist_add(&hctx->hist, dccp_sk(sk)->dccps_gss))
362 DCCP_CRIT("packet history - out of memory!");
365 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
367 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
368 struct tfrc_tx_hist_entry *acked;
373 /* we are only interested in ACKs */
374 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
375 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
378 * Locate the acknowledged packet in the TX history.
380 * Returning "entry not found" here can for instance happen when
381 * - the host has not sent out anything (e.g. a passive server),
382 * - the Ack is outdated (packet with higher Ack number was received),
383 * - it is a bogus Ack (for a packet not sent on this connection).
385 acked = tfrc_tx_hist_find_entry(hctx->hist, dccp_hdr_ack_seq(skb));
388 /* For the sake of RTT sampling, ignore/remove all older entries */
389 tfrc_tx_hist_purge(&acked->next);
391 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
392 now = ktime_get_real();
393 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
394 hctx->rtt = tfrc_ewma(hctx->rtt, r_sample, 9);
397 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
399 if (hctx->state == TFRC_SSTATE_NO_FBACK) {
400 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
402 if (hctx->t_rto == 0) {
404 * Initial feedback packet: Larger Initial Windows (4.2)
406 hctx->x = rfc3390_initial_rate(sk);
409 ccid3_update_send_interval(hctx);
411 goto done_computing_x;
412 } else if (hctx->p == 0) {
414 * First feedback after nofeedback timer expiry (4.3)
416 goto done_computing_x;
420 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
422 hctx->x_calc = tfrc_calc_x(hctx->s, hctx->rtt, hctx->p);
423 ccid3_hc_tx_update_x(sk, &now);
426 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
427 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
428 dccp_role(sk), sk, hctx->rtt, r_sample,
429 hctx->s, hctx->p, hctx->x_calc,
430 (unsigned)(hctx->x_recv >> 6),
431 (unsigned)(hctx->x >> 6));
433 /* unschedule no feedback timer */
434 sk_stop_timer(sk, &hctx->no_feedback_timer);
437 * As we have calculated new ipi, delta, t_nom it is possible
438 * that we now can send a packet, so wake up dccp_wait_for_ccid
440 sk->sk_write_space(sk);
443 * Update timeout interval for the nofeedback timer.
444 * We use a configuration option to increase the lower bound.
445 * This can help avoid triggering the nofeedback timer too
446 * often ('spinning') on LANs with small RTTs.
448 hctx->t_rto = max_t(u32, 4 * hctx->rtt, (CONFIG_IP_DCCP_CCID3_RTO *
449 (USEC_PER_SEC / 1000)));
451 * Schedule no feedback timer to expire in
452 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
454 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
456 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
457 "expire in %lu jiffies (%luus)\n",
458 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
460 sk_reset_timer(sk, &hctx->no_feedback_timer,
461 jiffies + usecs_to_jiffies(t_nfb));
464 static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type,
465 u8 option, u8 *optval, u8 optlen)
467 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
471 case TFRC_OPT_RECEIVE_RATE:
472 case TFRC_OPT_LOSS_EVENT_RATE:
473 /* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
474 if (packet_type == DCCP_PKT_DATA)
476 if (unlikely(optlen != 4)) {
477 DCCP_WARN("%s(%p), invalid len %d for %u\n",
478 dccp_role(sk), sk, optlen, option);
481 opt_val = ntohl(get_unaligned((__be32 *)optval));
483 if (option == TFRC_OPT_RECEIVE_RATE) {
484 /* Receive Rate is kept in units of 64 bytes/second */
485 hctx->x_recv = opt_val;
488 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
489 dccp_role(sk), sk, opt_val);
491 /* Update the fixpoint Loss Event Rate fraction */
492 hctx->p = tfrc_invert_loss_event_rate(opt_val);
494 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
495 dccp_role(sk), sk, opt_val);
501 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
503 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
505 hctx->state = TFRC_SSTATE_NO_SENT;
507 setup_timer(&hctx->no_feedback_timer,
508 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
512 static void ccid3_hc_tx_exit(struct sock *sk)
514 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
516 sk_stop_timer(sk, &hctx->no_feedback_timer);
517 tfrc_tx_hist_purge(&hctx->hist);
520 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
522 info->tcpi_rto = ccid3_hc_tx_sk(sk)->t_rto;
523 info->tcpi_rtt = ccid3_hc_tx_sk(sk)->rtt;
526 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
527 u32 __user *optval, int __user *optlen)
529 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
530 struct tfrc_tx_info tfrc;
534 case DCCP_SOCKOPT_CCID_TX_INFO:
535 if (len < sizeof(tfrc))
537 tfrc.tfrctx_x = hctx->x;
538 tfrc.tfrctx_x_recv = hctx->x_recv;
539 tfrc.tfrctx_x_calc = hctx->x_calc;
540 tfrc.tfrctx_rtt = hctx->rtt;
541 tfrc.tfrctx_p = hctx->p;
542 tfrc.tfrctx_rto = hctx->t_rto;
543 tfrc.tfrctx_ipi = hctx->t_ipi;
551 if (put_user(len, optlen) || copy_to_user(optval, val, len))
558 * Receiver Half-Connection Routines
561 /* CCID3 feedback types */
562 enum ccid3_fback_type {
563 CCID3_FBACK_NONE = 0,
565 CCID3_FBACK_PERIODIC,
566 CCID3_FBACK_PARAM_CHANGE
569 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
570 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
572 static char *ccid3_rx_state_names[] = {
573 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
574 [TFRC_RSTATE_DATA] = "DATA",
577 return ccid3_rx_state_names[state];
581 static void ccid3_hc_rx_set_state(struct sock *sk,
582 enum ccid3_hc_rx_states state)
584 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
585 enum ccid3_hc_rx_states oldstate = hcrx->state;
587 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
588 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
589 ccid3_rx_state_name(state));
590 WARN_ON(state == oldstate);
594 static void ccid3_hc_rx_send_feedback(struct sock *sk,
595 const struct sk_buff *skb,
596 enum ccid3_fback_type fbtype)
598 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
599 struct dccp_sock *dp = dccp_sk(sk);
600 ktime_t now = ktime_get_real();
604 case CCID3_FBACK_INITIAL:
606 hcrx->p_inverse = ~0U; /* see RFC 4342, 8.5 */
608 case CCID3_FBACK_PARAM_CHANGE:
610 * When parameters change (new loss or p > p_prev), we do not
611 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
612 * need to reuse the previous value of X_recv. However, when
613 * X_recv was 0 (due to early loss), this would kill X down to
614 * s/t_mbi (i.e. one packet in 64 seconds).
615 * To avoid such drastic reduction, we approximate X_recv as
616 * the number of bytes since last feedback.
617 * This is a safe fallback, since X is bounded above by X_calc.
619 if (hcrx->x_recv > 0)
622 case CCID3_FBACK_PERIODIC:
623 delta = ktime_us_delta(now, hcrx->tstamp_last_feedback);
625 DCCP_BUG("delta (%ld) <= 0", (long)delta);
627 hcrx->x_recv = scaled_div32(hcrx->bytes_recv, delta);
633 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n",
634 (long)delta, hcrx->x_recv, hcrx->p_inverse);
636 hcrx->tstamp_last_feedback = now;
637 hcrx->last_counter = dccp_hdr(skb)->dccph_ccval;
638 hcrx->bytes_recv = 0;
640 dp->dccps_hc_rx_insert_options = 1;
644 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
646 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
649 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
652 if (dccp_packet_without_ack(skb))
655 x_recv = htonl(hcrx->x_recv);
656 pinv = htonl(hcrx->p_inverse);
658 if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
659 &pinv, sizeof(pinv)) ||
660 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
661 &x_recv, sizeof(x_recv)))
667 /** ccid3_first_li - Implements [RFC 3448, 6.3.1]
669 * Determine the length of the first loss interval via inverse lookup.
670 * Assume that X_recv can be computed by the throughput equation
674 * Find some p such that f(p) = fval; return 1/p (scaled).
676 static u32 ccid3_first_li(struct sock *sk)
678 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
679 u32 x_recv, p, delta;
682 if (hcrx->rtt == 0) {
683 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
684 hcrx->rtt = DCCP_FALLBACK_RTT;
687 delta = ktime_to_us(net_timedelta(hcrx->tstamp_last_feedback));
688 x_recv = scaled_div32(hcrx->bytes_recv, delta);
689 if (x_recv == 0) { /* would also trigger divide-by-zero */
690 DCCP_WARN("X_recv==0\n");
691 if (hcrx->x_recv == 0) {
692 DCCP_BUG("stored value of X_recv is zero");
695 x_recv = hcrx->x_recv;
698 fval = scaled_div(hcrx->s, hcrx->rtt);
699 fval = scaled_div32(fval, x_recv);
700 p = tfrc_calc_x_reverse_lookup(fval);
702 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
703 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
705 return p == 0 ? ~0U : scaled_div(1, p);
708 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
710 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
711 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
712 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
713 const bool is_data_packet = dccp_data_packet(skb);
715 if (unlikely(hcrx->state == TFRC_RSTATE_NO_DATA)) {
716 if (is_data_packet) {
717 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
718 do_feedback = CCID3_FBACK_INITIAL;
719 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
722 * Not necessary to update bytes_recv here,
723 * since X_recv = 0 for the first feedback packet (cf.
724 * RFC 3448, 6.3) -- gerrit
730 if (tfrc_rx_hist_duplicate(&hcrx->hist, skb))
731 return; /* done receiving */
733 if (is_data_packet) {
734 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
736 * Update moving-average of s and the sum of received payload bytes
738 hcrx->s = tfrc_ewma(hcrx->s, payload, 9);
739 hcrx->bytes_recv += payload;
743 * Perform loss detection and handle pending losses
745 if (tfrc_rx_handle_loss(&hcrx->hist, &hcrx->li_hist,
746 skb, ndp, ccid3_first_li, sk)) {
747 do_feedback = CCID3_FBACK_PARAM_CHANGE;
751 if (tfrc_rx_hist_loss_pending(&hcrx->hist))
752 return; /* done receiving */
755 * Handle data packets: RTT sampling and monitoring p
757 if (unlikely(!is_data_packet))
760 if (!tfrc_lh_is_initialised(&hcrx->li_hist)) {
761 const u32 sample = tfrc_rx_hist_sample_rtt(&hcrx->hist, skb);
763 * Empty loss history: no loss so far, hence p stays 0.
764 * Sample RTT values, since an RTT estimate is required for the
765 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
768 hcrx->rtt = tfrc_ewma(hcrx->rtt, sample, 9);
770 } else if (tfrc_lh_update_i_mean(&hcrx->li_hist, skb)) {
772 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
773 * has decreased (resp. p has increased), send feedback now.
775 do_feedback = CCID3_FBACK_PARAM_CHANGE;
779 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
781 if (SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->last_counter) > 3)
782 do_feedback = CCID3_FBACK_PERIODIC;
785 tfrc_rx_hist_add_packet(&hcrx->hist, skb, ndp);
789 ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
792 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
794 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
796 hcrx->state = TFRC_RSTATE_NO_DATA;
797 tfrc_lh_init(&hcrx->li_hist);
798 return tfrc_rx_hist_alloc(&hcrx->hist);
801 static void ccid3_hc_rx_exit(struct sock *sk)
803 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
805 tfrc_rx_hist_purge(&hcrx->hist);
806 tfrc_lh_cleanup(&hcrx->li_hist);
809 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
811 info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->state;
812 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
813 info->tcpi_rcv_rtt = ccid3_hc_rx_sk(sk)->rtt;
816 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
817 u32 __user *optval, int __user *optlen)
819 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
820 struct tfrc_rx_info rx_info;
824 case DCCP_SOCKOPT_CCID_RX_INFO:
825 if (len < sizeof(rx_info))
827 rx_info.tfrcrx_x_recv = hcrx->x_recv;
828 rx_info.tfrcrx_rtt = hcrx->rtt;
829 rx_info.tfrcrx_p = tfrc_invert_loss_event_rate(hcrx->p_inverse);
830 len = sizeof(rx_info);
837 if (put_user(len, optlen) || copy_to_user(optval, val, len))
843 static struct ccid_operations ccid3 = {
844 .ccid_id = DCCPC_CCID3,
845 .ccid_name = "TCP-Friendly Rate Control",
846 .ccid_owner = THIS_MODULE,
847 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
848 .ccid_hc_tx_init = ccid3_hc_tx_init,
849 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
850 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
851 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
852 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
853 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
854 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
855 .ccid_hc_rx_init = ccid3_hc_rx_init,
856 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
857 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
858 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
859 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
860 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
861 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
862 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
865 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
866 module_param(ccid3_debug, bool, 0644);
867 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
870 static __init int ccid3_module_init(void)
872 return ccid_register(&ccid3);
874 module_init(ccid3_module_init);
876 static __exit void ccid3_module_exit(void)
878 ccid_unregister(&ccid3);
880 module_exit(ccid3_module_exit);
882 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
883 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
884 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
885 MODULE_LICENSE("GPL");
886 MODULE_ALIAS("net-dccp-ccid-3");