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",
59 return ccid3_state_names[state];
63 static void ccid3_hc_tx_set_state(struct sock *sk,
64 enum ccid3_hc_tx_states state)
66 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
67 enum ccid3_hc_tx_states oldstate = hc->tx_state;
69 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
70 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
71 ccid3_tx_state_name(state));
72 WARN_ON(state == oldstate);
77 * Compute the initial sending rate X_init in the manner of RFC 3390:
79 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
81 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
82 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
83 * For consistency with other parts of the code, X_init is scaled by 2^6.
85 static inline u64 rfc3390_initial_rate(struct sock *sk)
87 const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
88 const __u32 w_init = clamp_t(__u32, 4380U, 2 * hc->tx_s, 4 * hc->tx_s);
90 return scaled_div(w_init << 6, hc->tx_rtt);
94 * ccid3_update_send_interval - Calculate new t_ipi = s / X_inst
95 * This respects the granularity of X_inst (64 * bytes/second).
97 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hc)
99 hc->tx_t_ipi = scaled_div32(((u64)hc->tx_s) << 6, hc->tx_x);
101 ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hc->tx_t_ipi,
102 hc->tx_s, (unsigned)(hc->tx_x >> 6));
105 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hc, ktime_t now)
107 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count);
109 return delta / hc->tx_rtt;
113 * ccid3_hc_tx_update_x - Update allowed sending rate X
114 * @stamp: most recent time if available - can be left NULL.
115 * This function tracks draft rfc3448bis, check there for latest details.
117 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
118 * fine-grained resolution of sending rates. This requires scaling by 2^6
119 * throughout the code. Only X_calc is unscaled (in bytes/second).
122 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
124 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
125 __u64 min_rate = 2 * hc->tx_x_recv;
126 const __u64 old_x = hc->tx_x;
127 ktime_t now = stamp ? *stamp : ktime_get_real();
130 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
131 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
132 * a sender is idle if it has not sent anything over a 2-RTT-period.
133 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
135 if (ccid3_hc_tx_idle_rtt(hc, now) >= 2) {
136 min_rate = rfc3390_initial_rate(sk);
137 min_rate = max(min_rate, 2 * hc->tx_x_recv);
142 hc->tx_x = min(((__u64)hc->tx_x_calc) << 6, min_rate);
143 hc->tx_x = max(hc->tx_x, (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
145 } else if (ktime_us_delta(now, hc->tx_t_ld) - (s64)hc->tx_rtt >= 0) {
147 hc->tx_x = min(2 * hc->tx_x, min_rate);
148 hc->tx_x = max(hc->tx_x,
149 scaled_div(((__u64)hc->tx_s) << 6, hc->tx_rtt));
153 if (hc->tx_x != old_x) {
154 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
155 "X_recv=%u\n", (unsigned)(old_x >> 6),
156 (unsigned)(hc->tx_x >> 6), hc->tx_x_calc,
157 (unsigned)(hc->tx_x_recv >> 6));
159 ccid3_update_send_interval(hc);
164 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
165 * @len: DCCP packet payload size in bytes
167 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hc, int len)
169 const u16 old_s = hc->tx_s;
171 hc->tx_s = tfrc_ewma(hc->tx_s, len, 9);
173 if (hc->tx_s != old_s)
174 ccid3_update_send_interval(hc);
178 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
179 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
181 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hc,
184 u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count),
185 quarter_rtts = (4 * delta) / hc->tx_rtt;
187 if (quarter_rtts > 0) {
188 hc->tx_t_last_win_count = now;
189 hc->tx_last_win_count += min(quarter_rtts, 5U);
190 hc->tx_last_win_count &= 0xF; /* mod 16 */
194 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
196 struct sock *sk = (struct sock *)data;
197 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
198 unsigned long t_nfb = USEC_PER_SEC / 5;
201 if (sock_owned_by_user(sk)) {
202 /* Try again later. */
203 /* XXX: set some sensible MIB */
207 ccid3_pr_debug("%s(%p, state=%s) - entry\n", dccp_role(sk), sk,
208 ccid3_tx_state_name(hc->tx_state));
210 /* Ignore and do not restart after leaving the established state */
211 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
214 /* Reset feedback state to "no feedback received" */
215 if (hc->tx_state == TFRC_SSTATE_FBACK)
216 ccid3_hc_tx_set_state(sk, 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 (hc->tx_t_rto == 0 || hc->tx_p == 0) {
224 /* halve send rate directly */
225 hc->tx_x = max(hc->tx_x / 2,
226 (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
227 ccid3_update_send_interval(hc);
230 * Modify the cached value of X_recv
232 * If (X_calc > 2 * X_recv)
233 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
235 * X_recv = X_calc / 4;
237 * Note that X_recv is scaled by 2^6 while X_calc is not
239 BUG_ON(hc->tx_p && !hc->tx_x_calc);
241 if (hc->tx_x_calc > (hc->tx_x_recv >> 5))
243 max(hc->tx_x_recv / 2,
244 (((__u64)hc->tx_s) << 6) / (2*TFRC_T_MBI));
246 hc->tx_x_recv = hc->tx_x_calc;
249 ccid3_hc_tx_update_x(sk, NULL);
251 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
252 (unsigned long long)hc->tx_x);
255 * Set new timeout for the nofeedback timer.
256 * See comments in packet_recv() regarding the value of t_RTO.
258 if (unlikely(hc->tx_t_rto == 0)) /* no feedback received yet */
259 t_nfb = TFRC_INITIAL_TIMEOUT;
261 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
264 sk_reset_timer(sk, &hc->tx_no_feedback_timer,
265 jiffies + usecs_to_jiffies(t_nfb));
272 * ccid3_hc_tx_send_packet - Delay-based dequeueing of TX packets
273 * @skb: next packet candidate to send on @sk
274 * This function uses the convention of ccid_packet_dequeue_eval() and
275 * returns a millisecond-delay value between 0 and t_mbi = 64000 msec.
277 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
279 struct dccp_sock *dp = dccp_sk(sk);
280 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
281 ktime_t now = ktime_get_real();
285 * This function is called only for Data and DataAck packets. Sending
286 * zero-sized Data(Ack)s is theoretically possible, but for congestion
287 * control this case is pathological - ignore it.
289 if (unlikely(skb->len == 0))
292 if (hc->tx_state == TFRC_SSTATE_NO_SENT) {
293 sk_reset_timer(sk, &hc->tx_no_feedback_timer, (jiffies +
294 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
295 hc->tx_last_win_count = 0;
296 hc->tx_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 hc->tx_rtt = dp->dccps_syn_rtt;
311 hc->tx_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 hc->tx_rtt = DCCP_FALLBACK_RTT;
324 ccid3_update_send_interval(hc);
326 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_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);
345 /* prepare to send now (add options etc.) */
346 dp->dccps_hc_tx_insert_options = 1;
347 DCCP_SKB_CB(skb)->dccpd_ccval = hc->tx_last_win_count;
349 /* set the nominal send time for the next following packet */
350 hc->tx_t_nom = ktime_add_us(hc->tx_t_nom, hc->tx_t_ipi);
351 return CCID_PACKET_SEND_AT_ONCE;
354 static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len)
356 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
358 ccid3_hc_tx_update_s(hc, len);
360 if (tfrc_tx_hist_add(&hc->tx_hist, dccp_sk(sk)->dccps_gss))
361 DCCP_CRIT("packet history - out of memory!");
364 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
366 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
367 struct tfrc_tx_hist_entry *acked;
372 /* we are only interested in ACKs */
373 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
374 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
377 * Locate the acknowledged packet in the TX history.
379 * Returning "entry not found" here can for instance happen when
380 * - the host has not sent out anything (e.g. a passive server),
381 * - the Ack is outdated (packet with higher Ack number was received),
382 * - it is a bogus Ack (for a packet not sent on this connection).
384 acked = tfrc_tx_hist_find_entry(hc->tx_hist, dccp_hdr_ack_seq(skb));
387 /* For the sake of RTT sampling, ignore/remove all older entries */
388 tfrc_tx_hist_purge(&acked->next);
390 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
391 now = ktime_get_real();
392 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
393 hc->tx_rtt = tfrc_ewma(hc->tx_rtt, r_sample, 9);
396 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
398 if (hc->tx_state == TFRC_SSTATE_NO_FBACK) {
399 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
401 if (hc->tx_t_rto == 0) {
403 * Initial feedback packet: Larger Initial Windows (4.2)
405 hc->tx_x = rfc3390_initial_rate(sk);
408 ccid3_update_send_interval(hc);
410 goto done_computing_x;
411 } else if (hc->tx_p == 0) {
413 * First feedback after nofeedback timer expiry (4.3)
415 goto done_computing_x;
419 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
421 hc->tx_x_calc = tfrc_calc_x(hc->tx_s, hc->tx_rtt, hc->tx_p);
422 ccid3_hc_tx_update_x(sk, &now);
425 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
426 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
427 dccp_role(sk), sk, hc->tx_rtt, r_sample,
428 hc->tx_s, hc->tx_p, hc->tx_x_calc,
429 (unsigned)(hc->tx_x_recv >> 6),
430 (unsigned)(hc->tx_x >> 6));
432 /* unschedule no feedback timer */
433 sk_stop_timer(sk, &hc->tx_no_feedback_timer);
436 * As we have calculated new ipi, delta, t_nom it is possible
437 * that we now can send a packet, so wake up dccp_wait_for_ccid
439 sk->sk_write_space(sk);
442 * Update timeout interval for the nofeedback timer. In order to control
443 * rate halving on networks with very low RTTs (<= 1 ms), use per-route
444 * tunable RTAX_RTO_MIN value as the lower bound.
446 hc->tx_t_rto = max_t(u32, 4 * hc->tx_rtt,
447 USEC_PER_SEC/HZ * tcp_rto_min(sk));
449 * Schedule no feedback timer to expire in
450 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
452 t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
454 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
455 "expire in %lu jiffies (%luus)\n",
456 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
458 sk_reset_timer(sk, &hc->tx_no_feedback_timer,
459 jiffies + usecs_to_jiffies(t_nfb));
462 static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type,
463 u8 option, u8 *optval, u8 optlen)
465 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
469 case TFRC_OPT_RECEIVE_RATE:
470 case TFRC_OPT_LOSS_EVENT_RATE:
471 /* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
472 if (packet_type == DCCP_PKT_DATA)
474 if (unlikely(optlen != 4)) {
475 DCCP_WARN("%s(%p), invalid len %d for %u\n",
476 dccp_role(sk), sk, optlen, option);
479 opt_val = ntohl(get_unaligned((__be32 *)optval));
481 if (option == TFRC_OPT_RECEIVE_RATE) {
482 /* Receive Rate is kept in units of 64 bytes/second */
483 hc->tx_x_recv = opt_val;
486 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
487 dccp_role(sk), sk, opt_val);
489 /* Update the fixpoint Loss Event Rate fraction */
490 hc->tx_p = tfrc_invert_loss_event_rate(opt_val);
492 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
493 dccp_role(sk), sk, opt_val);
499 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
501 struct ccid3_hc_tx_sock *hc = ccid_priv(ccid);
503 hc->tx_state = TFRC_SSTATE_NO_SENT;
505 setup_timer(&hc->tx_no_feedback_timer,
506 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
510 static void ccid3_hc_tx_exit(struct sock *sk)
512 struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
514 sk_stop_timer(sk, &hc->tx_no_feedback_timer);
515 tfrc_tx_hist_purge(&hc->tx_hist);
518 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
520 info->tcpi_rto = ccid3_hc_tx_sk(sk)->tx_t_rto;
521 info->tcpi_rtt = ccid3_hc_tx_sk(sk)->tx_rtt;
524 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
525 u32 __user *optval, int __user *optlen)
527 const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
528 struct tfrc_tx_info tfrc;
532 case DCCP_SOCKOPT_CCID_TX_INFO:
533 if (len < sizeof(tfrc))
535 tfrc.tfrctx_x = hc->tx_x;
536 tfrc.tfrctx_x_recv = hc->tx_x_recv;
537 tfrc.tfrctx_x_calc = hc->tx_x_calc;
538 tfrc.tfrctx_rtt = hc->tx_rtt;
539 tfrc.tfrctx_p = hc->tx_p;
540 tfrc.tfrctx_rto = hc->tx_t_rto;
541 tfrc.tfrctx_ipi = hc->tx_t_ipi;
549 if (put_user(len, optlen) || copy_to_user(optval, val, len))
556 * Receiver Half-Connection Routines
559 /* CCID3 feedback types */
560 enum ccid3_fback_type {
561 CCID3_FBACK_NONE = 0,
563 CCID3_FBACK_PERIODIC,
564 CCID3_FBACK_PARAM_CHANGE
567 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
568 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
570 static const char *const ccid3_rx_state_names[] = {
571 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
572 [TFRC_RSTATE_DATA] = "DATA",
575 return ccid3_rx_state_names[state];
579 static void ccid3_hc_rx_set_state(struct sock *sk,
580 enum ccid3_hc_rx_states state)
582 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
583 enum ccid3_hc_rx_states oldstate = hc->rx_state;
585 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
586 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
587 ccid3_rx_state_name(state));
588 WARN_ON(state == oldstate);
589 hc->rx_state = state;
592 static void ccid3_hc_rx_send_feedback(struct sock *sk,
593 const struct sk_buff *skb,
594 enum ccid3_fback_type fbtype)
596 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
597 struct dccp_sock *dp = dccp_sk(sk);
598 ktime_t now = ktime_get_real();
602 case CCID3_FBACK_INITIAL:
604 hc->rx_pinv = ~0U; /* see RFC 4342, 8.5 */
606 case CCID3_FBACK_PARAM_CHANGE:
608 * When parameters change (new loss or p > p_prev), we do not
609 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
610 * need to reuse the previous value of X_recv. However, when
611 * X_recv was 0 (due to early loss), this would kill X down to
612 * s/t_mbi (i.e. one packet in 64 seconds).
613 * To avoid such drastic reduction, we approximate X_recv as
614 * the number of bytes since last feedback.
615 * This is a safe fallback, since X is bounded above by X_calc.
617 if (hc->rx_x_recv > 0)
620 case CCID3_FBACK_PERIODIC:
621 delta = ktime_us_delta(now, hc->rx_tstamp_last_feedback);
623 DCCP_BUG("delta (%ld) <= 0", (long)delta);
625 hc->rx_x_recv = scaled_div32(hc->rx_bytes_recv, delta);
631 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta,
632 hc->rx_x_recv, hc->rx_pinv);
634 hc->rx_tstamp_last_feedback = now;
635 hc->rx_last_counter = dccp_hdr(skb)->dccph_ccval;
636 hc->rx_bytes_recv = 0;
638 dp->dccps_hc_rx_insert_options = 1;
642 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
644 const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
647 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
650 if (dccp_packet_without_ack(skb))
653 x_recv = htonl(hc->rx_x_recv);
654 pinv = htonl(hc->rx_pinv);
656 if (dccp_insert_option(skb, TFRC_OPT_LOSS_EVENT_RATE,
657 &pinv, sizeof(pinv)) ||
658 dccp_insert_option(skb, TFRC_OPT_RECEIVE_RATE,
659 &x_recv, sizeof(x_recv)))
666 * ccid3_first_li - Implements [RFC 5348, 6.3.1]
668 * Determine the length of the first loss interval via inverse lookup.
669 * Assume that X_recv can be computed by the throughput equation
673 * Find some p such that f(p) = fval; return 1/p (scaled).
675 static u32 ccid3_first_li(struct sock *sk)
677 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
678 u32 x_recv, p, delta;
681 if (hc->rx_rtt == 0) {
682 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
683 hc->rx_rtt = DCCP_FALLBACK_RTT;
686 delta = ktime_to_us(net_timedelta(hc->rx_tstamp_last_feedback));
687 x_recv = scaled_div32(hc->rx_bytes_recv, delta);
688 if (x_recv == 0) { /* would also trigger divide-by-zero */
689 DCCP_WARN("X_recv==0\n");
690 if (hc->rx_x_recv == 0) {
691 DCCP_BUG("stored value of X_recv is zero");
694 x_recv = hc->rx_x_recv;
697 fval = scaled_div(hc->rx_s, hc->rx_rtt);
698 fval = scaled_div32(fval, x_recv);
699 p = tfrc_calc_x_reverse_lookup(fval);
701 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
702 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
704 return p == 0 ? ~0U : scaled_div(1, p);
707 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
709 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
710 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
711 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
712 const bool is_data_packet = dccp_data_packet(skb);
714 if (unlikely(hc->rx_state == TFRC_RSTATE_NO_DATA)) {
715 if (is_data_packet) {
716 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
717 do_feedback = CCID3_FBACK_INITIAL;
718 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
721 * Not necessary to update rx_bytes_recv here,
722 * since X_recv = 0 for the first feedback packet (cf.
723 * RFC 3448, 6.3) -- gerrit
729 if (tfrc_rx_hist_duplicate(&hc->rx_hist, skb))
730 return; /* done receiving */
732 if (is_data_packet) {
733 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
735 * Update moving-average of s and the sum of received payload bytes
737 hc->rx_s = tfrc_ewma(hc->rx_s, payload, 9);
738 hc->rx_bytes_recv += payload;
742 * Perform loss detection and handle pending losses
744 if (tfrc_rx_handle_loss(&hc->rx_hist, &hc->rx_li_hist,
745 skb, ndp, ccid3_first_li, sk)) {
746 do_feedback = CCID3_FBACK_PARAM_CHANGE;
750 if (tfrc_rx_hist_loss_pending(&hc->rx_hist))
751 return; /* done receiving */
754 * Handle data packets: RTT sampling and monitoring p
756 if (unlikely(!is_data_packet))
759 if (!tfrc_lh_is_initialised(&hc->rx_li_hist)) {
760 const u32 sample = tfrc_rx_hist_sample_rtt(&hc->rx_hist, skb);
762 * Empty loss history: no loss so far, hence p stays 0.
763 * Sample RTT values, since an RTT estimate is required for the
764 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
767 hc->rx_rtt = tfrc_ewma(hc->rx_rtt, sample, 9);
769 } else if (tfrc_lh_update_i_mean(&hc->rx_li_hist, skb)) {
771 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
772 * has decreased (resp. p has increased), send feedback now.
774 do_feedback = CCID3_FBACK_PARAM_CHANGE;
778 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
780 if (SUB16(dccp_hdr(skb)->dccph_ccval, hc->rx_last_counter) > 3)
781 do_feedback = CCID3_FBACK_PERIODIC;
784 tfrc_rx_hist_add_packet(&hc->rx_hist, skb, ndp);
788 ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
791 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
793 struct ccid3_hc_rx_sock *hc = ccid_priv(ccid);
795 hc->rx_state = TFRC_RSTATE_NO_DATA;
796 tfrc_lh_init(&hc->rx_li_hist);
797 return tfrc_rx_hist_alloc(&hc->rx_hist);
800 static void ccid3_hc_rx_exit(struct sock *sk)
802 struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
804 tfrc_rx_hist_purge(&hc->rx_hist);
805 tfrc_lh_cleanup(&hc->rx_li_hist);
808 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
810 info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->rx_state;
811 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
812 info->tcpi_rcv_rtt = ccid3_hc_rx_sk(sk)->rx_rtt;
815 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
816 u32 __user *optval, int __user *optlen)
818 const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
819 struct tfrc_rx_info rx_info;
823 case DCCP_SOCKOPT_CCID_RX_INFO:
824 if (len < sizeof(rx_info))
826 rx_info.tfrcrx_x_recv = hc->rx_x_recv;
827 rx_info.tfrcrx_rtt = hc->rx_rtt;
828 rx_info.tfrcrx_p = tfrc_invert_loss_event_rate(hc->rx_pinv);
829 len = sizeof(rx_info);
836 if (put_user(len, optlen) || copy_to_user(optval, val, len))
842 struct ccid_operations ccid3_ops = {
843 .ccid_id = DCCPC_CCID3,
844 .ccid_name = "TCP-Friendly Rate Control",
845 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
846 .ccid_hc_tx_init = ccid3_hc_tx_init,
847 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
848 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
849 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
850 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
851 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
852 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
853 .ccid_hc_rx_init = ccid3_hc_rx_init,
854 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
855 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
856 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
857 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
858 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
859 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
860 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
863 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
864 module_param(ccid3_debug, bool, 0644);
865 MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages");