2 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
3 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
4 * Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
6 * An implementation of the DCCP protocol
8 * This code has been developed by the University of Waikato WAND
9 * research group. For further information please see http://www.wand.net.nz/
11 * This code also uses code from Lulea University, rereleased as GPL by its
13 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
15 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
16 * and to make it work as a loadable module in the DCCP stack written by
17 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
19 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
21 * This program is free software; you can redistribute it and/or modify
22 * it under the terms of the GNU General Public License as published by
23 * the Free Software Foundation; either version 2 of the License, or
24 * (at your option) any later version.
26 * This program is distributed in the hope that it will be useful,
27 * but WITHOUT ANY WARRANTY; without even the implied warranty of
28 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
29 * GNU General Public License for more details.
31 * You should have received a copy of the GNU General Public License
32 * along with this program; if not, write to the Free Software
33 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
38 #include <asm/unaligned.h>
40 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
41 static int ccid3_debug;
42 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
44 #define ccid3_pr_debug(format, a...)
48 * Transmitter Half-Connection Routines
50 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
51 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
53 static const char *const ccid3_state_names[] = {
54 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
55 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
56 [TFRC_SSTATE_FBACK] = "FBACK",
57 [TFRC_SSTATE_TERM] = "TERM",
60 return ccid3_state_names[state];
64 static void ccid3_hc_tx_set_state(struct sock *sk,
65 enum ccid3_hc_tx_states state)
67 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
68 enum ccid3_hc_tx_states oldstate = hc->tx_state;
70 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
71 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
72 ccid3_tx_state_name(state));
73 WARN_ON(state == oldstate);
78 * Compute the initial sending rate X_init in the manner of RFC 3390:
80 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
82 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
83 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
84 * For consistency with other parts of the code, X_init is scaled by 2^6.
86 static inline u64 rfc3390_initial_rate(struct sock *sk)
88 const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
89 const __u32 w_init = clamp_t(__u32, 4380U, 2 * hc->tx_s, 4 * hc->tx_s);
91 return scaled_div(w_init << 6, hc->tx_rtt);
95 * ccid3_update_send_interval - Calculate new t_ipi = s / X_inst
96 * This respects the granularity of X_inst (64 * bytes/second).
98 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hc)
100 hc->tx_t_ipi = scaled_div32(((u64)hc->tx_s) << 6, hc->tx_x);
102 ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hc->tx_t_ipi,
103 hc->tx_s, (unsigned)(hc->tx_x >> 6));
106 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hc, ktime_t now)
108 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count);
110 return delta / hc->tx_rtt;
114 * ccid3_hc_tx_update_x - Update allowed sending rate X
115 * @stamp: most recent time if available - can be left NULL.
116 * This function tracks draft rfc3448bis, check there for latest details.
118 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
119 * fine-grained resolution of sending rates. This requires scaling by 2^6
120 * throughout the code. Only X_calc is unscaled (in bytes/second).
123 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
125 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
126 __u64 min_rate = 2 * hc->tx_x_recv;
127 const __u64 old_x = hc->tx_x;
128 ktime_t now = stamp ? *stamp : ktime_get_real();
131 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
132 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
133 * a sender is idle if it has not sent anything over a 2-RTT-period.
134 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
136 if (ccid3_hc_tx_idle_rtt(hc, now) >= 2) {
137 min_rate = rfc3390_initial_rate(sk);
138 min_rate = max(min_rate, 2 * hc->tx_x_recv);
143 hc->tx_x = min(((__u64)hc->tx_x_calc) << 6, min_rate);
144 hc->tx_x = max(hc->tx_x, (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
146 } else if (ktime_us_delta(now, hc->tx_t_ld) - (s64)hc->tx_rtt >= 0) {
148 hc->tx_x = min(2 * hc->tx_x, min_rate);
149 hc->tx_x = max(hc->tx_x,
150 scaled_div(((__u64)hc->tx_s) << 6, hc->tx_rtt));
154 if (hc->tx_x != old_x) {
155 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
156 "X_recv=%u\n", (unsigned)(old_x >> 6),
157 (unsigned)(hc->tx_x >> 6), hc->tx_x_calc,
158 (unsigned)(hc->tx_x_recv >> 6));
160 ccid3_update_send_interval(hc);
165 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
166 * @len: DCCP packet payload size in bytes
168 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hc, int len)
170 const u16 old_s = hc->tx_s;
172 hc->tx_s = tfrc_ewma(hc->tx_s, len, 9);
174 if (hc->tx_s != old_s)
175 ccid3_update_send_interval(hc);
179 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
180 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
182 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hc,
185 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count),
186 quarter_rtts = (4 * delta) / hc->tx_rtt;
188 if (quarter_rtts > 0) {
189 hc->tx_t_last_win_count = now;
190 hc->tx_last_win_count += min(quarter_rtts, 5U);
191 hc->tx_last_win_count &= 0xF; /* mod 16 */
195 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
197 struct sock *sk = (struct sock *)data;
198 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
199 unsigned long t_nfb = USEC_PER_SEC / 5;
202 if (sock_owned_by_user(sk)) {
203 /* Try again later. */
204 /* XXX: set some sensible MIB */
208 ccid3_pr_debug("%s(%p, state=%s) - entry\n", dccp_role(sk), sk,
209 ccid3_tx_state_name(hc->tx_state));
211 if (hc->tx_state == TFRC_SSTATE_FBACK)
212 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
213 else if (hc->tx_state != TFRC_SSTATE_NO_FBACK)
217 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
218 * RTO is 0 if and only if no feedback has been received yet.
220 if (hc->tx_t_rto == 0 || hc->tx_p == 0) {
222 /* halve send rate directly */
223 hc->tx_x = max(hc->tx_x / 2,
224 (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
225 ccid3_update_send_interval(hc);
228 * Modify the cached value of X_recv
230 * If (X_calc > 2 * X_recv)
231 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
233 * X_recv = X_calc / 4;
235 * Note that X_recv is scaled by 2^6 while X_calc is not
237 BUG_ON(hc->tx_p && !hc->tx_x_calc);
239 if (hc->tx_x_calc > (hc->tx_x_recv >> 5))
241 max(hc->tx_x_recv / 2,
242 (((__u64)hc->tx_s) << 6) / (2*TFRC_T_MBI));
244 hc->tx_x_recv = hc->tx_x_calc;
247 ccid3_hc_tx_update_x(sk, NULL);
249 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
250 (unsigned long long)hc->tx_x);
253 * Set new timeout for the nofeedback timer.
254 * See comments in packet_recv() regarding the value of t_RTO.
256 if (unlikely(hc->tx_t_rto == 0)) /* no feedback received yet */
257 t_nfb = TFRC_INITIAL_TIMEOUT;
259 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
262 sk_reset_timer(sk, &hc->tx_no_feedback_timer,
263 jiffies + usecs_to_jiffies(t_nfb));
271 * > 0: delay (in msecs) that should pass before actually sending
272 * = 0: can send immediately
273 * < 0: error condition; do not send packet
275 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
277 struct dccp_sock *dp = dccp_sk(sk);
278 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
279 ktime_t now = ktime_get_real();
283 * This function is called only for Data and DataAck packets. Sending
284 * zero-sized Data(Ack)s is theoretically possible, but for congestion
285 * control this case is pathological - ignore it.
287 if (unlikely(skb->len == 0))
290 switch (hc->tx_state) {
291 case TFRC_SSTATE_NO_SENT:
292 sk_reset_timer(sk, &hc->tx_no_feedback_timer, (jiffies +
293 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
294 hc->tx_last_win_count = 0;
295 hc->tx_t_last_win_count = now;
297 /* Set t_0 for initial packet */
303 * Use initial RTT sample when available: recommended by erratum
304 * to RFC 4342. This implements the initialisation procedure of
305 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
307 if (dp->dccps_syn_rtt) {
308 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
309 hc->tx_rtt = dp->dccps_syn_rtt;
310 hc->tx_x = rfc3390_initial_rate(sk);
314 * Sender does not have RTT sample:
315 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
316 * is needed in several parts (e.g. window counter);
317 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
319 hc->tx_rtt = DCCP_FALLBACK_RTT;
323 ccid3_update_send_interval(hc);
325 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
327 case TFRC_SSTATE_NO_FBACK:
328 case TFRC_SSTATE_FBACK:
329 delay = ktime_us_delta(hc->tx_t_nom, now);
330 ccid3_pr_debug("delay=%ld\n", (long)delay);
332 * Scheduling of packet transmissions (RFC 5348, 8.3)
334 * if (t_now > t_nom - delta)
335 * // send the packet now
337 * // send the packet in (t_nom - t_now) milliseconds.
339 if (delay >= TFRC_T_DELTA)
340 return (u32)delay / USEC_PER_MSEC;
342 ccid3_hc_tx_update_win_count(hc, now);
344 case TFRC_SSTATE_TERM:
345 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
349 /* prepare to send now (add options etc.) */
350 dp->dccps_hc_tx_insert_options = 1;
351 DCCP_SKB_CB(skb)->dccpd_ccval = hc->tx_last_win_count;
353 /* set the nominal send time for the next following packet */
354 hc->tx_t_nom = ktime_add_us(hc->tx_t_nom, hc->tx_t_ipi);
358 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
361 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
363 ccid3_hc_tx_update_s(hc, len);
365 if (tfrc_tx_hist_add(&hc->tx_hist, dccp_sk(sk)->dccps_gss))
366 DCCP_CRIT("packet history - out of memory!");
369 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
371 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
372 struct ccid3_options_received *opt_recv = &hc->tx_options_received;
373 struct tfrc_tx_hist_entry *acked;
378 /* we are only interested in ACKs */
379 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
380 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
382 /* ... and only in the established state */
383 if (hc->tx_state != TFRC_SSTATE_FBACK &&
384 hc->tx_state != TFRC_SSTATE_NO_FBACK)
388 * Locate the acknowledged packet in the TX history.
390 * Returning "entry not found" here can for instance happen when
391 * - the host has not sent out anything (e.g. a passive server),
392 * - the Ack is outdated (packet with higher Ack number was received),
393 * - it is a bogus Ack (for a packet not sent on this connection).
395 acked = tfrc_tx_hist_find_entry(hc->tx_hist, dccp_hdr_ack_seq(skb));
398 /* For the sake of RTT sampling, ignore/remove all older entries */
399 tfrc_tx_hist_purge(&acked->next);
401 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
402 now = ktime_get_real();
403 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
404 hc->tx_rtt = tfrc_ewma(hc->tx_rtt, r_sample, 9);
406 /* Update receive rate in units of 64 * bytes/second */
407 hc->tx_x_recv = opt_recv->ccid3or_receive_rate;
410 /* Update loss event rate (which is scaled by 1e6) */
411 pinv = opt_recv->ccid3or_loss_event_rate;
412 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
414 else /* can not exceed 100% */
415 hc->tx_p = scaled_div(1, pinv);
418 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
420 if (hc->tx_state == TFRC_SSTATE_NO_FBACK) {
421 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
423 if (hc->tx_t_rto == 0) {
425 * Initial feedback packet: Larger Initial Windows (4.2)
427 hc->tx_x = rfc3390_initial_rate(sk);
430 ccid3_update_send_interval(hc);
432 goto done_computing_x;
433 } else if (hc->tx_p == 0) {
435 * First feedback after nofeedback timer expiry (4.3)
437 goto done_computing_x;
441 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
443 hc->tx_x_calc = tfrc_calc_x(hc->tx_s, hc->tx_rtt, hc->tx_p);
444 ccid3_hc_tx_update_x(sk, &now);
447 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
448 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
449 dccp_role(sk), sk, hc->tx_rtt, r_sample,
450 hc->tx_s, hc->tx_p, hc->tx_x_calc,
451 (unsigned)(hc->tx_x_recv >> 6),
452 (unsigned)(hc->tx_x >> 6));
454 /* unschedule no feedback timer */
455 sk_stop_timer(sk, &hc->tx_no_feedback_timer);
458 * As we have calculated new ipi, delta, t_nom it is possible
459 * that we now can send a packet, so wake up dccp_wait_for_ccid
461 sk->sk_write_space(sk);
464 * Update timeout interval for the nofeedback timer. In order to control
465 * rate halving on networks with very low RTTs (<= 1 ms), use per-route
466 * tunable RTAX_RTO_MIN value as the lower bound.
468 hc->tx_t_rto = max_t(u32, 4 * hc->tx_rtt,
469 USEC_PER_SEC/HZ * tcp_rto_min(sk));
471 * Schedule no feedback timer to expire in
472 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
474 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
476 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
477 "expire in %lu jiffies (%luus)\n",
478 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
480 sk_reset_timer(sk, &hc->tx_no_feedback_timer,
481 jiffies + usecs_to_jiffies(t_nfb));
484 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
485 unsigned char len, u16 idx,
486 unsigned char *value)
488 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
489 struct ccid3_options_received *opt_recv = &hc->tx_options_received;
493 case TFRC_OPT_RECEIVE_RATE:
494 case TFRC_OPT_LOSS_EVENT_RATE:
495 if (unlikely(len != 4)) {
496 DCCP_WARN("%s(%p), invalid len %d for %u\n",
497 dccp_role(sk), sk, len, option);
500 opt_val = ntohl(get_unaligned((__be32 *)value));
502 if (option == TFRC_OPT_RECEIVE_RATE) {
503 opt_recv->ccid3or_receive_rate = opt_val;
504 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
505 dccp_role(sk), sk, opt_val);
507 opt_recv->ccid3or_loss_event_rate = opt_val;
508 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
509 dccp_role(sk), sk, opt_val);
515 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
517 struct ccid3_hc_tx_sock *hc = ccid_priv(ccid);
519 hc->tx_state = TFRC_SSTATE_NO_SENT;
521 setup_timer(&hc->tx_no_feedback_timer,
522 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
526 static void ccid3_hc_tx_exit(struct sock *sk)
528 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
530 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
531 sk_stop_timer(sk, &hc->tx_no_feedback_timer);
533 tfrc_tx_hist_purge(&hc->tx_hist);
536 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
538 info->tcpi_rto = ccid3_hc_tx_sk(sk)->tx_t_rto;
539 info->tcpi_rtt = ccid3_hc_tx_sk(sk)->tx_rtt;
542 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
543 u32 __user *optval, int __user *optlen)
545 const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
546 struct tfrc_tx_info tfrc;
550 case DCCP_SOCKOPT_CCID_TX_INFO:
551 if (len < sizeof(tfrc))
553 tfrc.tfrctx_x = hc->tx_x;
554 tfrc.tfrctx_x_recv = hc->tx_x_recv;
555 tfrc.tfrctx_x_calc = hc->tx_x_calc;
556 tfrc.tfrctx_rtt = hc->tx_rtt;
557 tfrc.tfrctx_p = hc->tx_p;
558 tfrc.tfrctx_rto = hc->tx_t_rto;
559 tfrc.tfrctx_ipi = hc->tx_t_ipi;
567 if (put_user(len, optlen) || copy_to_user(optval, val, len))
574 * Receiver Half-Connection Routines
577 /* CCID3 feedback types */
578 enum ccid3_fback_type {
579 CCID3_FBACK_NONE = 0,
581 CCID3_FBACK_PERIODIC,
582 CCID3_FBACK_PARAM_CHANGE
585 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
586 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
588 static const char *const ccid3_rx_state_names[] = {
589 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
590 [TFRC_RSTATE_DATA] = "DATA",
591 [TFRC_RSTATE_TERM] = "TERM",
594 return ccid3_rx_state_names[state];
598 static void ccid3_hc_rx_set_state(struct sock *sk,
599 enum ccid3_hc_rx_states state)
601 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
602 enum ccid3_hc_rx_states oldstate = hc->rx_state;
604 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
605 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
606 ccid3_rx_state_name(state));
607 WARN_ON(state == oldstate);
608 hc->rx_state = state;
611 static void ccid3_hc_rx_send_feedback(struct sock *sk,
612 const struct sk_buff *skb,
613 enum ccid3_fback_type fbtype)
615 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
616 struct dccp_sock *dp = dccp_sk(sk);
620 if (unlikely(hc->rx_state == TFRC_RSTATE_TERM))
623 now = ktime_get_real();
626 case CCID3_FBACK_INITIAL:
628 hc->rx_pinv = ~0U; /* see RFC 4342, 8.5 */
630 case CCID3_FBACK_PARAM_CHANGE:
632 * When parameters change (new loss or p > p_prev), we do not
633 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
634 * need to reuse the previous value of X_recv. However, when
635 * X_recv was 0 (due to early loss), this would kill X down to
636 * s/t_mbi (i.e. one packet in 64 seconds).
637 * To avoid such drastic reduction, we approximate X_recv as
638 * the number of bytes since last feedback.
639 * This is a safe fallback, since X is bounded above by X_calc.
641 if (hc->rx_x_recv > 0)
644 case CCID3_FBACK_PERIODIC:
645 delta = ktime_us_delta(now, hc->rx_tstamp_last_feedback);
647 DCCP_BUG("delta (%ld) <= 0", (long)delta);
649 hc->rx_x_recv = scaled_div32(hc->rx_bytes_recv, delta);
655 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta,
656 hc->rx_x_recv, hc->rx_pinv);
658 hc->rx_tstamp_last_feedback = now;
659 hc->rx_last_counter = dccp_hdr(skb)->dccph_ccval;
660 hc->rx_bytes_recv = 0;
662 dp->dccps_hc_rx_insert_options = 1;
666 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
668 const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
671 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
674 if (dccp_packet_without_ack(skb))
677 x_recv = htonl(hc->rx_x_recv);
678 pinv = htonl(hc->rx_pinv);
680 if (dccp_insert_option(skb, TFRC_OPT_LOSS_EVENT_RATE,
681 &pinv, sizeof(pinv)) ||
682 dccp_insert_option(skb, TFRC_OPT_RECEIVE_RATE,
683 &x_recv, sizeof(x_recv)))
690 * ccid3_first_li - Implements [RFC 5348, 6.3.1]
692 * Determine the length of the first loss interval via inverse lookup.
693 * Assume that X_recv can be computed by the throughput equation
697 * Find some p such that f(p) = fval; return 1/p (scaled).
699 static u32 ccid3_first_li(struct sock *sk)
701 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
702 u32 x_recv, p, delta;
705 if (hc->rx_rtt == 0) {
706 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
707 hc->rx_rtt = DCCP_FALLBACK_RTT;
710 delta = ktime_to_us(net_timedelta(hc->rx_tstamp_last_feedback));
711 x_recv = scaled_div32(hc->rx_bytes_recv, delta);
712 if (x_recv == 0) { /* would also trigger divide-by-zero */
713 DCCP_WARN("X_recv==0\n");
714 if (hc->rx_x_recv == 0) {
715 DCCP_BUG("stored value of X_recv is zero");
718 x_recv = hc->rx_x_recv;
721 fval = scaled_div(hc->rx_s, hc->rx_rtt);
722 fval = scaled_div32(fval, x_recv);
723 p = tfrc_calc_x_reverse_lookup(fval);
725 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
726 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
728 return p == 0 ? ~0U : scaled_div(1, p);
731 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
733 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
734 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
735 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
736 const bool is_data_packet = dccp_data_packet(skb);
738 if (unlikely(hc->rx_state == TFRC_RSTATE_NO_DATA)) {
739 if (is_data_packet) {
740 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
741 do_feedback = CCID3_FBACK_INITIAL;
742 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
745 * Not necessary to update rx_bytes_recv here,
746 * since X_recv = 0 for the first feedback packet (cf.
747 * RFC 3448, 6.3) -- gerrit
753 if (tfrc_rx_hist_duplicate(&hc->rx_hist, skb))
754 return; /* done receiving */
756 if (is_data_packet) {
757 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
759 * Update moving-average of s and the sum of received payload bytes
761 hc->rx_s = tfrc_ewma(hc->rx_s, payload, 9);
762 hc->rx_bytes_recv += payload;
766 * Perform loss detection and handle pending losses
768 if (tfrc_rx_handle_loss(&hc->rx_hist, &hc->rx_li_hist,
769 skb, ndp, ccid3_first_li, sk)) {
770 do_feedback = CCID3_FBACK_PARAM_CHANGE;
774 if (tfrc_rx_hist_loss_pending(&hc->rx_hist))
775 return; /* done receiving */
778 * Handle data packets: RTT sampling and monitoring p
780 if (unlikely(!is_data_packet))
783 if (!tfrc_lh_is_initialised(&hc->rx_li_hist)) {
784 const u32 sample = tfrc_rx_hist_sample_rtt(&hc->rx_hist, skb);
786 * Empty loss history: no loss so far, hence p stays 0.
787 * Sample RTT values, since an RTT estimate is required for the
788 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
791 hc->rx_rtt = tfrc_ewma(hc->rx_rtt, sample, 9);
793 } else if (tfrc_lh_update_i_mean(&hc->rx_li_hist, skb)) {
795 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
796 * has decreased (resp. p has increased), send feedback now.
798 do_feedback = CCID3_FBACK_PARAM_CHANGE;
802 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
804 if (SUB16(dccp_hdr(skb)->dccph_ccval, hc->rx_last_counter) > 3)
805 do_feedback = CCID3_FBACK_PERIODIC;
808 tfrc_rx_hist_add_packet(&hc->rx_hist, skb, ndp);
812 ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
815 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
817 struct ccid3_hc_rx_sock *hc = ccid_priv(ccid);
819 hc->rx_state = TFRC_RSTATE_NO_DATA;
820 tfrc_lh_init(&hc->rx_li_hist);
821 return tfrc_rx_hist_alloc(&hc->rx_hist);
824 static void ccid3_hc_rx_exit(struct sock *sk)
826 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
828 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
830 tfrc_rx_hist_purge(&hc->rx_hist);
831 tfrc_lh_cleanup(&hc->rx_li_hist);
834 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
836 info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->rx_state;
837 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
838 info->tcpi_rcv_rtt = ccid3_hc_rx_sk(sk)->rx_rtt;
841 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
842 u32 __user *optval, int __user *optlen)
844 const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
845 struct tfrc_rx_info rx_info;
849 case DCCP_SOCKOPT_CCID_RX_INFO:
850 if (len < sizeof(rx_info))
852 rx_info.tfrcrx_x_recv = hc->rx_x_recv;
853 rx_info.tfrcrx_rtt = hc->rx_rtt;
854 rx_info.tfrcrx_p = hc->rx_pinv == 0 ? ~0U :
855 scaled_div(1, hc->rx_pinv);
856 len = sizeof(rx_info);
863 if (put_user(len, optlen) || copy_to_user(optval, val, len))
869 struct ccid_operations ccid3_ops = {
870 .ccid_id = DCCPC_CCID3,
871 .ccid_name = "TCP-Friendly Rate Control",
872 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
873 .ccid_hc_tx_init = ccid3_hc_tx_init,
874 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
875 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
876 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
877 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
878 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
879 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
880 .ccid_hc_rx_init = ccid3_hc_rx_init,
881 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
882 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
883 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
884 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
885 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
886 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
887 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
890 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
891 module_param(ccid3_debug, bool, 0644);
892 MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages");