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
54 * Compute the initial sending rate X_init in the manner of RFC 3390:
56 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
58 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
59 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
60 * For consistency with other parts of the code, X_init is scaled by 2^6.
62 static inline u64 rfc3390_initial_rate(struct sock *sk)
64 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
65 const __u32 w_init = clamp_t(__u32, 4380U, 2 * hctx->s, 4 * hctx->s);
67 return scaled_div(w_init << 6, hctx->rtt);
71 * ccid3_update_send_interval - Calculate new t_ipi = s / X_inst
72 * This respects the granularity of X_inst (64 * bytes/second).
74 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
76 hctx->t_ipi = scaled_div32(((u64)hctx->s) << 6, hctx->x);
78 ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hctx->t_ipi,
79 hctx->s, (unsigned)(hctx->x >> 6));
82 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
84 u32 delta = ktime_us_delta(now, hctx->t_last_win_count);
86 return delta / hctx->rtt;
90 * ccid3_hc_tx_update_x - Update allowed sending rate X
91 * @stamp: most recent time if available - can be left NULL.
92 * This function tracks draft rfc3448bis, check there for latest details.
94 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
95 * fine-grained resolution of sending rates. This requires scaling by 2^6
96 * throughout the code. Only X_calc is unscaled (in bytes/second).
99 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
101 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
102 u64 min_rate = 2 * hctx->x_recv;
103 const u64 old_x = hctx->x;
104 ktime_t now = stamp ? *stamp : ktime_get_real();
107 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
108 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
109 * a sender is idle if it has not sent anything over a 2-RTT-period.
110 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
112 if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) {
113 min_rate = rfc3390_initial_rate(sk);
114 min_rate = max(min_rate, 2 * hctx->x_recv);
119 hctx->x = min(((u64)hctx->x_calc) << 6, min_rate);
120 hctx->x = max(hctx->x, (((u64)hctx->s) << 6) / TFRC_T_MBI);
122 } else if (ktime_us_delta(now, hctx->t_ld) - (s64)hctx->rtt >= 0) {
124 hctx->x = min(2 * hctx->x, min_rate);
125 hctx->x = max(hctx->x,
126 scaled_div(((u64)hctx->s) << 6, hctx->rtt));
130 if (hctx->x != old_x) {
131 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
132 "X_recv=%u\n", (unsigned)(old_x >> 6),
133 (unsigned)(hctx->x >> 6), hctx->x_calc,
134 (unsigned)(hctx->x_recv >> 6));
136 ccid3_update_send_interval(hctx);
141 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
142 * @len: DCCP packet payload size in bytes
144 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
146 const u16 old_s = hctx->s;
148 hctx->s = tfrc_ewma(hctx->s, len, 9);
150 if (hctx->s != old_s)
151 ccid3_update_send_interval(hctx);
155 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
156 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
158 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
161 u32 delta = ktime_us_delta(now, hctx->t_last_win_count),
162 quarter_rtts = (4 * delta) / hctx->rtt;
164 if (quarter_rtts > 0) {
165 hctx->t_last_win_count = now;
166 hctx->last_win_count += min(quarter_rtts, 5U);
167 hctx->last_win_count &= 0xF; /* mod 16 */
171 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
173 struct sock *sk = (struct sock *)data;
174 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
175 unsigned long t_nfb = USEC_PER_SEC / 5;
178 if (sock_owned_by_user(sk)) {
179 /* Try again later. */
180 /* XXX: set some sensible MIB */
184 ccid3_pr_debug("%s(%p) entry with%s feedback\n", dccp_role(sk), sk,
185 hctx->feedback ? "" : "out");
187 /* Ignore and do not restart after leaving the established state */
188 if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
191 /* Reset feedback state to "no feedback received" */
192 hctx->feedback = false;
195 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
196 * RTO is 0 if and only if no feedback has been received yet.
198 if (hctx->t_rto == 0 || hctx->p == 0) {
200 /* halve send rate directly */
201 hctx->x = max(hctx->x / 2, (((u64)hctx->s) << 6) / TFRC_T_MBI);
202 ccid3_update_send_interval(hctx);
205 * Modify the cached value of X_recv
207 * If (X_calc > 2 * X_recv)
208 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
210 * X_recv = X_calc / 4;
212 * Note that X_recv is scaled by 2^6 while X_calc is not
214 BUG_ON(hctx->p && !hctx->x_calc);
216 if (hctx->x_calc > (hctx->x_recv >> 5))
218 max(hctx->x_recv / 2,
219 (((__u64)hctx->s) << 6) / (2 * TFRC_T_MBI));
221 hctx->x_recv = hctx->x_calc;
224 ccid3_hc_tx_update_x(sk, NULL);
226 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
227 (unsigned long long)hctx->x);
230 * Set new timeout for the nofeedback timer.
231 * See comments in packet_recv() regarding the value of t_RTO.
233 if (unlikely(hctx->t_rto == 0)) /* no feedback received yet */
234 t_nfb = TFRC_INITIAL_TIMEOUT;
236 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
239 sk_reset_timer(sk, &hctx->no_feedback_timer,
240 jiffies + usecs_to_jiffies(t_nfb));
247 * ccid3_hc_tx_send_packet - Delay-based dequeueing of TX packets
248 * @skb: next packet candidate to send on @sk
249 * This function uses the convention of ccid_packet_dequeue_eval() and
250 * returns a millisecond-delay value between 0 and t_mbi = 64000 msec.
252 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
254 struct dccp_sock *dp = dccp_sk(sk);
255 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
256 ktime_t now = ktime_get_real();
260 * This function is called only for Data and DataAck packets. Sending
261 * zero-sized Data(Ack)s is theoretically possible, but for congestion
262 * control this case is pathological - ignore it.
264 if (unlikely(skb->len == 0))
268 sk_reset_timer(sk, &hctx->no_feedback_timer, (jiffies +
269 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
270 hctx->last_win_count = 0;
271 hctx->t_last_win_count = now;
273 /* Set t_0 for initial packet */
279 * Use initial RTT sample when available: recommended by erratum
280 * to RFC 4342. This implements the initialisation procedure of
281 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
283 if (dp->dccps_syn_rtt) {
284 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
285 hctx->rtt = dp->dccps_syn_rtt;
286 hctx->x = rfc3390_initial_rate(sk);
290 * Sender does not have RTT sample:
291 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
292 * is needed in several parts (e.g. window counter);
293 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
295 hctx->rtt = DCCP_FALLBACK_RTT;
299 ccid3_update_send_interval(hctx);
302 delay = ktime_us_delta(hctx->t_nom, now);
303 ccid3_pr_debug("delay=%ld\n", (long)delay);
305 * Scheduling of packet transmissions [RFC 3448, 4.6]
307 * if (t_now > t_nom - delta)
308 * // send the packet now
310 * // send the packet in (t_nom - t_now) milliseconds.
312 if (delay >= TFRC_T_DELTA)
313 return (u32)delay / USEC_PER_MSEC;
315 ccid3_hc_tx_update_win_count(hctx, now);
318 /* prepare to send now (add options etc.) */
319 dp->dccps_hc_tx_insert_options = 1;
320 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->last_win_count;
322 /* set the nominal send time for the next following packet */
323 hctx->t_nom = ktime_add_us(hctx->t_nom, hctx->t_ipi);
324 return CCID_PACKET_SEND_AT_ONCE;
327 static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len)
329 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
331 ccid3_hc_tx_update_s(hctx, len);
333 if (tfrc_tx_hist_add(&hctx->hist, dccp_sk(sk)->dccps_gss))
334 DCCP_CRIT("packet history - out of memory!");
337 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
339 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
340 struct tfrc_tx_hist_entry *acked;
345 /* we are only interested in ACKs */
346 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
347 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
350 * Locate the acknowledged packet in the TX history.
352 * Returning "entry not found" here can for instance happen when
353 * - the host has not sent out anything (e.g. a passive server),
354 * - the Ack is outdated (packet with higher Ack number was received),
355 * - it is a bogus Ack (for a packet not sent on this connection).
357 acked = tfrc_tx_hist_find_entry(hctx->hist, dccp_hdr_ack_seq(skb));
360 /* For the sake of RTT sampling, ignore/remove all older entries */
361 tfrc_tx_hist_purge(&acked->next);
363 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
364 now = ktime_get_real();
365 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
366 hctx->rtt = tfrc_ewma(hctx->rtt, r_sample, 9);
369 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
371 if (!hctx->feedback) {
372 hctx->feedback = true;
374 if (hctx->t_rto == 0) {
376 * Initial feedback packet: Larger Initial Windows (4.2)
378 hctx->x = rfc3390_initial_rate(sk);
381 ccid3_update_send_interval(hctx);
383 goto done_computing_x;
384 } else if (hctx->p == 0) {
386 * First feedback after nofeedback timer expiry (4.3)
388 goto done_computing_x;
392 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
394 hctx->x_calc = tfrc_calc_x(hctx->s, hctx->rtt, hctx->p);
395 ccid3_hc_tx_update_x(sk, &now);
398 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
399 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
400 dccp_role(sk), sk, hctx->rtt, r_sample,
401 hctx->s, hctx->p, hctx->x_calc,
402 (unsigned)(hctx->x_recv >> 6),
403 (unsigned)(hctx->x >> 6));
405 /* unschedule no feedback timer */
406 sk_stop_timer(sk, &hctx->no_feedback_timer);
409 * As we have calculated new ipi, delta, t_nom it is possible
410 * that we now can send a packet, so wake up dccp_wait_for_ccid
412 sk->sk_write_space(sk);
415 * Update timeout interval for the nofeedback timer.
416 * We use a configuration option to increase the lower bound.
417 * This can help avoid triggering the nofeedback timer too
418 * often ('spinning') on LANs with small RTTs.
420 hctx->t_rto = max_t(u32, 4 * hctx->rtt, (CONFIG_IP_DCCP_CCID3_RTO *
421 (USEC_PER_SEC / 1000)));
423 * Schedule no feedback timer to expire in
424 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
426 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
428 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
429 "expire in %lu jiffies (%luus)\n",
430 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
432 sk_reset_timer(sk, &hctx->no_feedback_timer,
433 jiffies + usecs_to_jiffies(t_nfb));
436 static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type,
437 u8 option, u8 *optval, u8 optlen)
439 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
443 case TFRC_OPT_RECEIVE_RATE:
444 case TFRC_OPT_LOSS_EVENT_RATE:
445 /* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
446 if (packet_type == DCCP_PKT_DATA)
448 if (unlikely(optlen != 4)) {
449 DCCP_WARN("%s(%p), invalid len %d for %u\n",
450 dccp_role(sk), sk, optlen, option);
453 opt_val = ntohl(get_unaligned((__be32 *)optval));
455 if (option == TFRC_OPT_RECEIVE_RATE) {
456 /* Receive Rate is kept in units of 64 bytes/second */
457 hctx->x_recv = opt_val;
460 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
461 dccp_role(sk), sk, opt_val);
463 /* Update the fixpoint Loss Event Rate fraction */
464 hctx->p = tfrc_invert_loss_event_rate(opt_val);
466 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
467 dccp_role(sk), sk, opt_val);
473 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
475 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
478 setup_timer(&hctx->no_feedback_timer,
479 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
483 static void ccid3_hc_tx_exit(struct sock *sk)
485 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
487 sk_stop_timer(sk, &hctx->no_feedback_timer);
488 tfrc_tx_hist_purge(&hctx->hist);
491 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
493 info->tcpi_rto = ccid3_hc_tx_sk(sk)->t_rto;
494 info->tcpi_rtt = ccid3_hc_tx_sk(sk)->rtt;
497 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
498 u32 __user *optval, int __user *optlen)
500 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
501 struct tfrc_tx_info tfrc;
505 case DCCP_SOCKOPT_CCID_TX_INFO:
506 if (len < sizeof(tfrc))
508 tfrc.tfrctx_x = hctx->x;
509 tfrc.tfrctx_x_recv = hctx->x_recv;
510 tfrc.tfrctx_x_calc = hctx->x_calc;
511 tfrc.tfrctx_rtt = hctx->rtt;
512 tfrc.tfrctx_p = hctx->p;
513 tfrc.tfrctx_rto = hctx->t_rto;
514 tfrc.tfrctx_ipi = hctx->t_ipi;
522 if (put_user(len, optlen) || copy_to_user(optval, val, len))
529 * Receiver Half-Connection Routines
532 /* CCID3 feedback types */
533 enum ccid3_fback_type {
534 CCID3_FBACK_NONE = 0,
536 CCID3_FBACK_PERIODIC,
537 CCID3_FBACK_PARAM_CHANGE
540 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
541 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
543 static char *ccid3_rx_state_names[] = {
544 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
545 [TFRC_RSTATE_DATA] = "DATA",
548 return ccid3_rx_state_names[state];
552 static void ccid3_hc_rx_set_state(struct sock *sk,
553 enum ccid3_hc_rx_states state)
555 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
556 enum ccid3_hc_rx_states oldstate = hcrx->state;
558 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
559 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
560 ccid3_rx_state_name(state));
561 WARN_ON(state == oldstate);
565 static void ccid3_hc_rx_send_feedback(struct sock *sk,
566 const struct sk_buff *skb,
567 enum ccid3_fback_type fbtype)
569 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
570 struct dccp_sock *dp = dccp_sk(sk);
571 ktime_t now = ktime_get_real();
575 case CCID3_FBACK_INITIAL:
577 hcrx->p_inverse = ~0U; /* see RFC 4342, 8.5 */
579 case CCID3_FBACK_PARAM_CHANGE:
581 * When parameters change (new loss or p > p_prev), we do not
582 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
583 * need to reuse the previous value of X_recv. However, when
584 * X_recv was 0 (due to early loss), this would kill X down to
585 * s/t_mbi (i.e. one packet in 64 seconds).
586 * To avoid such drastic reduction, we approximate X_recv as
587 * the number of bytes since last feedback.
588 * This is a safe fallback, since X is bounded above by X_calc.
590 if (hcrx->x_recv > 0)
593 case CCID3_FBACK_PERIODIC:
594 delta = ktime_us_delta(now, hcrx->tstamp_last_feedback);
596 DCCP_BUG("delta (%ld) <= 0", (long)delta);
598 hcrx->x_recv = scaled_div32(hcrx->bytes_recv, delta);
604 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n",
605 (long)delta, hcrx->x_recv, hcrx->p_inverse);
607 hcrx->tstamp_last_feedback = now;
608 hcrx->last_counter = dccp_hdr(skb)->dccph_ccval;
609 hcrx->bytes_recv = 0;
611 dp->dccps_hc_rx_insert_options = 1;
615 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
617 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
620 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
623 if (dccp_packet_without_ack(skb))
626 x_recv = htonl(hcrx->x_recv);
627 pinv = htonl(hcrx->p_inverse);
629 if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
630 &pinv, sizeof(pinv)) ||
631 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
632 &x_recv, sizeof(x_recv)))
638 /** ccid3_first_li - Implements [RFC 3448, 6.3.1]
640 * Determine the length of the first loss interval via inverse lookup.
641 * Assume that X_recv can be computed by the throughput equation
645 * Find some p such that f(p) = fval; return 1/p (scaled).
647 static u32 ccid3_first_li(struct sock *sk)
649 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
650 u32 x_recv, p, delta;
653 if (hcrx->rtt == 0) {
654 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
655 hcrx->rtt = DCCP_FALLBACK_RTT;
658 delta = ktime_to_us(net_timedelta(hcrx->tstamp_last_feedback));
659 x_recv = scaled_div32(hcrx->bytes_recv, delta);
660 if (x_recv == 0) { /* would also trigger divide-by-zero */
661 DCCP_WARN("X_recv==0\n");
662 if (hcrx->x_recv == 0) {
663 DCCP_BUG("stored value of X_recv is zero");
666 x_recv = hcrx->x_recv;
669 fval = scaled_div(hcrx->s, hcrx->rtt);
670 fval = scaled_div32(fval, x_recv);
671 p = tfrc_calc_x_reverse_lookup(fval);
673 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
674 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
676 return p == 0 ? ~0U : scaled_div(1, p);
679 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
681 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
682 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
683 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
684 const bool is_data_packet = dccp_data_packet(skb);
686 if (unlikely(hcrx->state == TFRC_RSTATE_NO_DATA)) {
687 if (is_data_packet) {
688 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
689 do_feedback = CCID3_FBACK_INITIAL;
690 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
693 * Not necessary to update bytes_recv here,
694 * since X_recv = 0 for the first feedback packet (cf.
695 * RFC 3448, 6.3) -- gerrit
701 if (tfrc_rx_hist_duplicate(&hcrx->hist, skb))
702 return; /* done receiving */
704 if (is_data_packet) {
705 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
707 * Update moving-average of s and the sum of received payload bytes
709 hcrx->s = tfrc_ewma(hcrx->s, payload, 9);
710 hcrx->bytes_recv += payload;
714 * Perform loss detection and handle pending losses
716 if (tfrc_rx_handle_loss(&hcrx->hist, &hcrx->li_hist,
717 skb, ndp, ccid3_first_li, sk)) {
718 do_feedback = CCID3_FBACK_PARAM_CHANGE;
722 if (tfrc_rx_hist_loss_pending(&hcrx->hist))
723 return; /* done receiving */
726 * Handle data packets: RTT sampling and monitoring p
728 if (unlikely(!is_data_packet))
731 if (!tfrc_lh_is_initialised(&hcrx->li_hist)) {
732 const u32 sample = tfrc_rx_hist_sample_rtt(&hcrx->hist, skb);
734 * Empty loss history: no loss so far, hence p stays 0.
735 * Sample RTT values, since an RTT estimate is required for the
736 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
739 hcrx->rtt = tfrc_ewma(hcrx->rtt, sample, 9);
741 } else if (tfrc_lh_update_i_mean(&hcrx->li_hist, skb)) {
743 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
744 * has decreased (resp. p has increased), send feedback now.
746 do_feedback = CCID3_FBACK_PARAM_CHANGE;
750 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
752 if (SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->last_counter) > 3)
753 do_feedback = CCID3_FBACK_PERIODIC;
756 tfrc_rx_hist_add_packet(&hcrx->hist, skb, ndp);
760 ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
763 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
765 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
767 hcrx->state = TFRC_RSTATE_NO_DATA;
768 tfrc_lh_init(&hcrx->li_hist);
769 return tfrc_rx_hist_init(&hcrx->hist, sk);
772 static void ccid3_hc_rx_exit(struct sock *sk)
774 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
776 tfrc_rx_hist_purge(&hcrx->hist);
777 tfrc_lh_cleanup(&hcrx->li_hist);
780 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
782 info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->state;
783 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
784 info->tcpi_rcv_rtt = ccid3_hc_rx_sk(sk)->rtt;
787 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
788 u32 __user *optval, int __user *optlen)
790 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
791 struct tfrc_rx_info rx_info;
795 case DCCP_SOCKOPT_CCID_RX_INFO:
796 if (len < sizeof(rx_info))
798 rx_info.tfrcrx_x_recv = hcrx->x_recv;
799 rx_info.tfrcrx_rtt = hcrx->rtt;
800 rx_info.tfrcrx_p = tfrc_invert_loss_event_rate(hcrx->p_inverse);
801 len = sizeof(rx_info);
808 if (put_user(len, optlen) || copy_to_user(optval, val, len))
814 static struct ccid_operations ccid3 = {
815 .ccid_id = DCCPC_CCID3,
816 .ccid_name = "TCP-Friendly Rate Control",
817 .ccid_owner = THIS_MODULE,
818 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
819 .ccid_hc_tx_init = ccid3_hc_tx_init,
820 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
821 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
822 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
823 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
824 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
825 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
826 .ccid_hc_rx_init = ccid3_hc_rx_init,
827 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
828 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
829 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
830 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
831 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
832 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
833 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
836 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
837 module_param(ccid3_debug, bool, 0644);
838 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
841 static __init int ccid3_module_init(void)
846 * Without a fine-grained clock resolution, RTTs/X_recv are not sampled
847 * correctly and feedback is sent either too early or too late.
849 hrtimer_get_res(CLOCK_MONOTONIC, &tp);
850 if (tp.tv_sec || tp.tv_nsec > DCCP_TIME_RESOLUTION * NSEC_PER_USEC) {
851 printk(KERN_ERR "%s: Timer too coarse (%ld usec), need %u-usec"
852 " resolution - check your clocksource.\n", __func__,
853 tp.tv_nsec/NSEC_PER_USEC, DCCP_TIME_RESOLUTION);
854 return -ESOCKTNOSUPPORT;
856 return ccid_register(&ccid3);
858 module_init(ccid3_module_init);
860 static __exit void ccid3_module_exit(void)
862 ccid_unregister(&ccid3);
864 module_exit(ccid3_module_exit);
866 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
867 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
868 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
869 MODULE_LICENSE("GPL");
870 MODULE_ALIAS("net-dccp-ccid-3");