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 /* Oscillation Prevention/Reduction: recommended by rfc3448bis, on by default */
53 static int do_osc_prev = true;
56 * Compute the initial sending rate X_init in the manner of RFC 3390:
58 * X_init = min(4 * MPS, max(2 * MPS, 4380 bytes)) / RTT
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 u32 mps = dccp_sk(sk)->dccps_mss_cache,
65 w_init = clamp(4380U, 2 * mps, 4 * mps);
67 return scaled_div(w_init << 6, ccid3_hc_tx_sk(sk)->rtt);
71 * ccid3_update_send_interval - Calculate new t_ipi = s / X
72 * This respects the granularity of X (64 * bytes/second) and enforces the
73 * scaled minimum of s * 64 / t_mbi = `s' bytes/second as per RFC 3448/4342.
75 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
77 if (unlikely(hctx->x <= hctx->s))
79 hctx->t_ipi = scaled_div32(((u64)hctx->s) << 6, hctx->x);
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);
121 } else if (ktime_us_delta(now, hctx->t_ld) - (s64)hctx->rtt >= 0) {
123 hctx->x = min(2 * hctx->x, min_rate);
124 hctx->x = max(hctx->x,
125 scaled_div(((u64)hctx->s) << 6, hctx->rtt));
129 if (hctx->x != old_x) {
130 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
131 "X_recv=%u\n", (unsigned)(old_x >> 6),
132 (unsigned)(hctx->x >> 6), hctx->x_calc,
133 (unsigned)(hctx->x_recv >> 6));
135 ccid3_update_send_interval(hctx);
140 * ccid3_hc_tx_measure_packet_size - Measuring the packet size `s' (sec 4.1)
141 * @new_len: DCCP payload size in bytes (not used by all methods)
143 static u32 ccid3_hc_tx_measure_packet_size(struct sock *sk, const u16 new_len)
145 #if defined(CONFIG_IP_DCCP_CCID3_MEASURE_S_AS_AVG)
146 return tfrc_ewma(ccid3_hc_tx_sk(sk)->s, new_len, 9);
147 #elif defined(CONFIG_IP_DCCP_CCID3_MEASURE_S_AS_MAX)
148 return max(ccid3_hc_tx_sk(sk)->s, new_len);
149 #else /* CONFIG_IP_DCCP_CCID3_MEASURE_S_AS_MPS */
150 return dccp_sk(sk)->dccps_mss_cache;
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 */
202 ccid3_update_send_interval(hctx);
206 * Modify the cached value of X_recv
208 * If (X_calc > 2 * X_recv)
209 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
211 * X_recv = X_calc / 4;
213 * Note that X_recv is scaled by 2^6 while X_calc is not
215 BUG_ON(hctx->p && !hctx->x_calc);
217 if (hctx->x_calc > (hctx->x_recv >> 5))
220 hctx->x_recv = hctx->x_calc;
223 ccid3_hc_tx_update_x(sk, NULL);
225 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
226 (unsigned long long)hctx->x);
229 * Set new timeout for the nofeedback timer.
230 * See comments in packet_recv() regarding the value of t_RTO.
232 if (unlikely(hctx->t_rto == 0)) /* no feedback received yet */
233 t_nfb = TFRC_INITIAL_TIMEOUT;
235 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
238 sk_reset_timer(sk, &hctx->no_feedback_timer,
239 jiffies + usecs_to_jiffies(t_nfb));
246 * ccid3_hc_tx_send_packet - Delay-based dequeueing of TX packets
247 * @skb: next packet candidate to send on @sk
248 * This function uses the convention of ccid_packet_dequeue_eval() and
249 * returns a millisecond-delay value between 0 and t_mbi = 64000 msec.
251 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
253 struct dccp_sock *dp = dccp_sk(sk);
254 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
255 ktime_t now = ktime_get_real();
259 * This function is called only for Data and DataAck packets. Sending
260 * zero-sized Data(Ack)s is theoretically possible, but for congestion
261 * control this case is pathological - ignore it.
263 if (unlikely(skb->len == 0))
267 sk_reset_timer(sk, &hctx->no_feedback_timer, (jiffies +
268 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
269 hctx->last_win_count = 0;
270 hctx->t_last_win_count = now;
272 /* Set t_0 for initial packet */
276 * Use initial RTT sample when available: recommended by erratum
277 * to RFC 4342. This implements the initialisation procedure of
278 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
280 if (dp->dccps_syn_rtt) {
281 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
282 hctx->rtt = dp->dccps_syn_rtt;
283 hctx->x = rfc3390_initial_rate(sk);
287 * Sender does not have RTT sample:
288 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
289 * is needed in several parts (e.g. window counter);
290 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
292 hctx->rtt = DCCP_FALLBACK_RTT;
293 hctx->x = dp->dccps_mss_cache;
297 /* Compute t_ipi = s / X */
298 hctx->s = ccid3_hc_tx_measure_packet_size(sk, skb->len);
299 ccid3_update_send_interval(hctx);
301 /* Seed value for Oscillation Prevention (sec. 4.5) */
302 hctx->r_sqmean = tfrc_scaled_sqrt(hctx->rtt);
305 delay = ktime_us_delta(hctx->t_nom, now);
306 ccid3_pr_debug("delay=%ld\n", (long)delay);
308 * Scheduling of packet transmissions [RFC 3448, 4.6]
310 * if (t_now > t_nom - delta)
311 * // send the packet now
313 * // send the packet in (t_nom - t_now) milliseconds.
315 if (delay >= TFRC_T_DELTA)
316 return (u32)delay / USEC_PER_MSEC;
318 ccid3_hc_tx_update_win_count(hctx, now);
321 /* prepare to send now (add options etc.) */
322 dp->dccps_hc_tx_insert_options = 1;
323 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->last_win_count;
325 /* set the nominal send time for the next following packet */
326 hctx->t_nom = ktime_add_us(hctx->t_nom, hctx->t_ipi);
327 return CCID_PACKET_SEND_AT_ONCE;
330 static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len)
332 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
334 /* Changes to s will become effective the next time X is computed */
335 hctx->s = ccid3_hc_tx_measure_packet_size(sk, len);
337 if (tfrc_tx_hist_add(&hctx->hist, dccp_sk(sk)->dccps_gss))
338 DCCP_CRIT("packet history - out of memory!");
341 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
343 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
344 struct tfrc_tx_hist_entry *acked;
349 /* we are only interested in ACKs */
350 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
351 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
354 * Locate the acknowledged packet in the TX history.
356 * Returning "entry not found" here can for instance happen when
357 * - the host has not sent out anything (e.g. a passive server),
358 * - the Ack is outdated (packet with higher Ack number was received),
359 * - it is a bogus Ack (for a packet not sent on this connection).
361 acked = tfrc_tx_hist_find_entry(hctx->hist, dccp_hdr_ack_seq(skb));
364 /* For the sake of RTT sampling, ignore/remove all older entries */
365 tfrc_tx_hist_purge(&acked->next);
367 /* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
368 now = ktime_get_real();
369 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
370 hctx->rtt = tfrc_ewma(hctx->rtt, r_sample, 9);
373 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
375 if (!hctx->feedback) {
376 hctx->feedback = true;
378 if (hctx->t_rto == 0) {
380 * Initial feedback packet: Larger Initial Windows (4.2)
382 hctx->x = rfc3390_initial_rate(sk);
385 ccid3_update_send_interval(hctx);
387 goto done_computing_x;
388 } else if (hctx->p == 0) {
390 * First feedback after nofeedback timer expiry (4.3)
392 goto done_computing_x;
396 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
398 hctx->x_calc = tfrc_calc_x(hctx->s, hctx->rtt, hctx->p);
399 ccid3_hc_tx_update_x(sk, &now);
402 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
403 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
404 dccp_role(sk), sk, hctx->rtt, r_sample,
405 hctx->s, hctx->p, hctx->x_calc,
406 (unsigned)(hctx->x_recv >> 6),
407 (unsigned)(hctx->x >> 6));
409 * Oscillation Reduction (RFC 3448, 4.5) - modifying t_ipi according to
410 * RTT changes, multiplying by X/X_inst = sqrt(R_sample)/R_sqmean. This
411 * can be useful if few connections share a link, avoiding that buffer
412 * fill levels (RTT) oscillate as a result of frequent adjustments to X.
413 * A useful presentation with background information is in
414 * Joerg Widmer, "Equation-Based Congestion Control",
415 * MSc Thesis, University of Mannheim, Germany, 2000
416 * (sec. 3.6.4), who calls this ISM ("Inter-packet Space Modulation").
419 r_sample = tfrc_scaled_sqrt(r_sample);
421 * The modulation can work in both ways: increase/decrease t_ipi
422 * according to long-term increases/decreases of the RTT. The
423 * former is a useful measure, since it works against queue
424 * build-up. The latter temporarily increases the sending rate,
425 * so that buffers fill up more quickly. This in turn causes
426 * the RTT to increase, so that either later reduction becomes
427 * necessary or the RTT stays at a very high level. Decreasing
428 * t_ipi is therefore not supported.
429 * Furthermore, during the initial slow-start phase the RTT
430 * naturally increases, where using the algorithm would cause
431 * delays. Hence it is disabled during the initial slow-start.
433 if (r_sample > hctx->r_sqmean && hctx->p > 0)
434 hctx->t_ipi = div_u64((u64)hctx->t_ipi * (u64)r_sample,
436 hctx->t_ipi = min_t(u32, hctx->t_ipi, TFRC_T_MBI);
437 /* update R_sqmean _after_ computing the modulation factor */
438 hctx->r_sqmean = tfrc_ewma(hctx->r_sqmean, r_sample, 9);
441 /* unschedule no feedback timer */
442 sk_stop_timer(sk, &hctx->no_feedback_timer);
445 * As we have calculated new ipi, delta, t_nom it is possible
446 * that we now can send a packet, so wake up dccp_wait_for_ccid
448 sk->sk_write_space(sk);
451 * Update timeout interval for the nofeedback timer.
452 * We use a configuration option to increase the lower bound.
453 * This can help avoid triggering the nofeedback timer too
454 * often ('spinning') on LANs with small RTTs.
456 hctx->t_rto = max_t(u32, 4 * hctx->rtt, (CONFIG_IP_DCCP_CCID3_RTO *
457 (USEC_PER_SEC / 1000)));
459 * Schedule no feedback timer to expire in
460 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
462 t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
464 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
465 "expire in %lu jiffies (%luus)\n",
466 dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
468 sk_reset_timer(sk, &hctx->no_feedback_timer,
469 jiffies + usecs_to_jiffies(t_nfb));
472 static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type,
473 u8 option, u8 *optval, u8 optlen)
475 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
479 case TFRC_OPT_RECEIVE_RATE:
480 case TFRC_OPT_LOSS_EVENT_RATE:
481 /* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
482 if (packet_type == DCCP_PKT_DATA)
484 if (unlikely(optlen != 4)) {
485 DCCP_WARN("%s(%p), invalid len %d for %u\n",
486 dccp_role(sk), sk, optlen, option);
489 opt_val = ntohl(get_unaligned((__be32 *)optval));
491 if (option == TFRC_OPT_RECEIVE_RATE) {
492 /* Receive Rate is kept in units of 64 bytes/second */
493 hctx->x_recv = opt_val;
496 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
497 dccp_role(sk), sk, opt_val);
499 /* Update the fixpoint Loss Event Rate fraction */
500 hctx->p = tfrc_invert_loss_event_rate(opt_val);
502 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
503 dccp_role(sk), sk, opt_val);
509 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
511 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
514 setup_timer(&hctx->no_feedback_timer,
515 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
519 static void ccid3_hc_tx_exit(struct sock *sk)
521 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
523 sk_stop_timer(sk, &hctx->no_feedback_timer);
524 tfrc_tx_hist_purge(&hctx->hist);
527 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
529 info->tcpi_rto = ccid3_hc_tx_sk(sk)->t_rto;
530 info->tcpi_rtt = ccid3_hc_tx_sk(sk)->rtt;
533 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
534 u32 __user *optval, int __user *optlen)
536 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
537 struct tfrc_tx_info tfrc;
541 case DCCP_SOCKOPT_CCID_TX_INFO:
542 if (len < sizeof(tfrc))
544 tfrc.tfrctx_x = hctx->x;
545 tfrc.tfrctx_x_recv = hctx->x_recv;
546 tfrc.tfrctx_x_calc = hctx->x_calc;
547 tfrc.tfrctx_rtt = hctx->rtt;
548 tfrc.tfrctx_p = hctx->p;
549 tfrc.tfrctx_rto = hctx->t_rto;
550 tfrc.tfrctx_ipi = hctx->t_ipi;
558 if (put_user(len, optlen) || copy_to_user(optval, val, len))
565 * Receiver Half-Connection Routines
567 static void ccid3_hc_rx_send_feedback(struct sock *sk,
568 const struct sk_buff *skb,
569 enum ccid3_fback_type fbtype)
571 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
574 case CCID3_FBACK_INITIAL:
576 hcrx->p_inverse = ~0U; /* see RFC 4342, 8.5 */
578 case CCID3_FBACK_PARAM_CHANGE:
579 if (unlikely(hcrx->feedback == CCID3_FBACK_NONE)) {
581 * rfc3448bis-06, 6.3.1: First packet(s) lost or marked
582 * FIXME: in rfc3448bis the receiver returns X_recv=0
583 * here as it normally would in the first feedback packet.
584 * However this is not possible yet, since the code still
585 * uses RFC 3448, i.e.
587 * Calculate X_calc using the TCP throughput equation.
588 * X = max(min(X_calc, 2*X_recv), s/t_mbi);
589 * would bring X down to s/t_mbi. That is why we return
590 * X_recv according to rfc3448bis-06 for the moment.
592 u32 s = tfrc_rx_hist_packet_size(&hcrx->hist),
593 rtt = tfrc_rx_hist_rtt(&hcrx->hist);
595 hcrx->x_recv = scaled_div32(s, 2 * rtt);
599 * When parameters change (new loss or p > p_prev), we do not
600 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
601 * always check whether at least RTT time units were covered.
603 hcrx->x_recv = tfrc_rx_hist_x_recv(&hcrx->hist, hcrx->x_recv);
605 case CCID3_FBACK_PERIODIC:
607 * Step (2) of rfc3448bis-06, 6.2:
608 * - if no data packets have been received, just restart timer
609 * - if data packets have been received, re-compute X_recv
611 if (hcrx->hist.bytes_recvd == 0)
612 goto prepare_for_next_time;
613 hcrx->x_recv = tfrc_rx_hist_x_recv(&hcrx->hist, hcrx->x_recv);
619 ccid3_pr_debug("X_recv=%u, 1/p=%u\n", hcrx->x_recv, hcrx->p_inverse);
621 dccp_sk(sk)->dccps_hc_rx_insert_options = 1;
624 prepare_for_next_time:
625 tfrc_rx_hist_restart_byte_counter(&hcrx->hist);
626 hcrx->last_counter = dccp_hdr(skb)->dccph_ccval;
627 hcrx->feedback = fbtype;
630 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
632 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
635 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
638 if (dccp_packet_without_ack(skb))
641 x_recv = htonl(hcrx->x_recv);
642 pinv = htonl(hcrx->p_inverse);
644 if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
645 &pinv, sizeof(pinv)) ||
646 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
647 &x_recv, sizeof(x_recv)))
653 /** ccid3_first_li - Implements [RFC 3448, 6.3.1]
655 * Determine the length of the first loss interval via inverse lookup.
656 * Assume that X_recv can be computed by the throughput equation
660 * Find some p such that f(p) = fval; return 1/p (scaled).
662 static u32 ccid3_first_li(struct sock *sk)
664 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
665 u32 s = tfrc_rx_hist_packet_size(&hcrx->hist),
666 rtt = tfrc_rx_hist_rtt(&hcrx->hist), x_recv, p;
670 * rfc3448bis-06, 6.3.1: First data packet(s) are marked or lost. Set p
671 * to give the equivalent of X_target = s/(2*R). Thus fval = 2 and so p
672 * is about 20.64%. This yields an interval length of 4.84 (rounded up).
674 if (unlikely(hcrx->feedback == CCID3_FBACK_NONE))
677 x_recv = tfrc_rx_hist_x_recv(&hcrx->hist, hcrx->x_recv);
681 fval = scaled_div32(scaled_div(s, rtt), x_recv);
682 p = tfrc_calc_x_reverse_lookup(fval);
684 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
685 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
688 return scaled_div(1, p);
693 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
695 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
696 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
697 const bool is_data_packet = dccp_data_packet(skb);
700 * Perform loss detection and handle pending losses
702 if (tfrc_rx_congestion_event(&hcrx->hist, &hcrx->li_hist,
703 skb, ndp, ccid3_first_li, sk))
704 ccid3_hc_rx_send_feedback(sk, skb, CCID3_FBACK_PARAM_CHANGE);
706 * Feedback for first non-empty data packet (RFC 3448, 6.3)
708 else if (unlikely(hcrx->feedback == CCID3_FBACK_NONE && is_data_packet))
709 ccid3_hc_rx_send_feedback(sk, skb, CCID3_FBACK_INITIAL);
711 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
713 else if (!tfrc_rx_hist_loss_pending(&hcrx->hist) && is_data_packet &&
714 SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->last_counter) > 3)
715 ccid3_hc_rx_send_feedback(sk, skb, CCID3_FBACK_PERIODIC);
718 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
720 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
722 tfrc_lh_init(&hcrx->li_hist);
723 return tfrc_rx_hist_init(&hcrx->hist, sk);
726 static void ccid3_hc_rx_exit(struct sock *sk)
728 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
730 tfrc_rx_hist_purge(&hcrx->hist);
731 tfrc_lh_cleanup(&hcrx->li_hist);
734 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
736 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
737 info->tcpi_rcv_rtt = tfrc_rx_hist_rtt(&ccid3_hc_rx_sk(sk)->hist);
740 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
741 u32 __user *optval, int __user *optlen)
743 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
744 struct tfrc_rx_info rx_info;
748 case DCCP_SOCKOPT_CCID_RX_INFO:
749 if (len < sizeof(rx_info))
751 rx_info.tfrcrx_x_recv = hcrx->x_recv;
752 rx_info.tfrcrx_rtt = tfrc_rx_hist_rtt(&hcrx->hist);
753 rx_info.tfrcrx_p = tfrc_invert_loss_event_rate(hcrx->p_inverse);
754 len = sizeof(rx_info);
761 if (put_user(len, optlen) || copy_to_user(optval, val, len))
767 static struct ccid_operations ccid3 = {
768 .ccid_id = DCCPC_CCID3,
769 .ccid_name = "TCP-Friendly Rate Control",
770 .ccid_owner = THIS_MODULE,
771 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
772 .ccid_hc_tx_init = ccid3_hc_tx_init,
773 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
774 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
775 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
776 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
777 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
778 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
779 .ccid_hc_rx_init = ccid3_hc_rx_init,
780 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
781 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
782 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
783 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
784 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
785 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
786 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
789 module_param(do_osc_prev, bool, 0644);
790 MODULE_PARM_DESC(do_osc_prev, "Use Oscillation Prevention (RFC 3448, 4.5)");
792 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
793 module_param(ccid3_debug, bool, 0644);
794 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
797 static __init int ccid3_module_init(void)
802 * Without a fine-grained clock resolution, RTTs/X_recv are not sampled
803 * correctly and feedback is sent either too early or too late.
805 hrtimer_get_res(CLOCK_MONOTONIC, &tp);
806 if (tp.tv_sec || tp.tv_nsec > DCCP_TIME_RESOLUTION * NSEC_PER_USEC) {
807 printk(KERN_ERR "%s: Timer too coarse (%ld usec), need %u-usec"
808 " resolution - check your clocksource.\n", __func__,
809 tp.tv_nsec/NSEC_PER_USEC, DCCP_TIME_RESOLUTION);
810 return -ESOCKTNOSUPPORT;
812 return ccid_register(&ccid3);
814 module_init(ccid3_module_init);
816 static __exit void ccid3_module_exit(void)
818 ccid_unregister(&ccid3);
820 module_exit(ccid3_module_exit);
822 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
823 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
824 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
825 MODULE_LICENSE("GPL");
826 MODULE_ALIAS("net-dccp-ccid-3");