2 * net/dccp/ccids/ccid3.c
4 * Copyright (c) 2007 The University of Aberdeen, Scotland, UK
5 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
6 * Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
8 * An implementation of the DCCP protocol
10 * This code has been developed by the University of Waikato WAND
11 * research group. For further information please see http://www.wand.net.nz/
13 * This code also uses code from Lulea University, rereleased as GPL by its
15 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
17 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
18 * and to make it work as a loadable module in the DCCP stack written by
19 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
21 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
23 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
33 * You should have received a copy of the GNU General Public License
34 * along with this program; if not, write to the Free Software
35 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
40 #include <asm/unaligned.h>
42 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
43 static int ccid3_debug;
44 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
46 #define ccid3_pr_debug(format, a...)
50 * Transmitter Half-Connection Routines
52 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
53 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
55 static char *ccid3_state_names[] = {
56 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
57 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
58 [TFRC_SSTATE_FBACK] = "FBACK",
59 [TFRC_SSTATE_TERM] = "TERM",
62 return ccid3_state_names[state];
66 static void ccid3_hc_tx_set_state(struct sock *sk,
67 enum ccid3_hc_tx_states state)
69 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
70 enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
72 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
73 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
74 ccid3_tx_state_name(state));
75 WARN_ON(state == oldstate);
76 hctx->ccid3hctx_state = state;
80 * Compute the initial sending rate X_init in the manner of RFC 3390:
82 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
84 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
85 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
86 * For consistency with other parts of the code, X_init is scaled by 2^6.
88 static inline u64 rfc3390_initial_rate(struct sock *sk)
90 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
91 const __u32 w_init = clamp_t(__u32, 4380U,
92 2 * hctx->ccid3hctx_s, 4 * hctx->ccid3hctx_s);
94 return scaled_div(w_init << 6, hctx->ccid3hctx_rtt);
98 * Recalculate t_ipi and delta (should be called whenever X changes)
100 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
102 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
103 hctx->ccid3hctx_t_ipi = scaled_div32(((u64)hctx->ccid3hctx_s) << 6,
106 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
107 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
108 TFRC_OPSYS_HALF_TIME_GRAN);
110 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n",
111 hctx->ccid3hctx_t_ipi, hctx->ccid3hctx_delta,
112 hctx->ccid3hctx_s, (unsigned)(hctx->ccid3hctx_x >> 6));
116 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
118 u32 delta = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count);
120 return delta / hctx->ccid3hctx_rtt;
124 * ccid3_hc_tx_update_x - Update allowed sending rate X
125 * @stamp: most recent time if available - can be left NULL.
126 * This function tracks draft rfc3448bis, check there for latest details.
128 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
129 * fine-grained resolution of sending rates. This requires scaling by 2^6
130 * throughout the code. Only X_calc is unscaled (in bytes/second).
133 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
135 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
136 __u64 min_rate = 2 * hctx->ccid3hctx_x_recv;
137 const __u64 old_x = hctx->ccid3hctx_x;
138 ktime_t now = stamp ? *stamp : ktime_get_real();
141 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
142 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
143 * a sender is idle if it has not sent anything over a 2-RTT-period.
144 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
146 if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) {
147 min_rate = rfc3390_initial_rate(sk);
148 min_rate = max(min_rate, 2 * hctx->ccid3hctx_x_recv);
151 if (hctx->ccid3hctx_p > 0) {
153 hctx->ccid3hctx_x = min(((__u64)hctx->ccid3hctx_x_calc) << 6,
155 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
156 (((__u64)hctx->ccid3hctx_s) << 6) /
159 } else if (ktime_us_delta(now, hctx->ccid3hctx_t_ld)
160 - (s64)hctx->ccid3hctx_rtt >= 0) {
162 hctx->ccid3hctx_x = min(2 * hctx->ccid3hctx_x, min_rate);
163 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
164 scaled_div(((__u64)hctx->ccid3hctx_s) << 6,
165 hctx->ccid3hctx_rtt));
166 hctx->ccid3hctx_t_ld = now;
169 if (hctx->ccid3hctx_x != old_x) {
170 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
171 "X_recv=%u\n", (unsigned)(old_x >> 6),
172 (unsigned)(hctx->ccid3hctx_x >> 6),
173 hctx->ccid3hctx_x_calc,
174 (unsigned)(hctx->ccid3hctx_x_recv >> 6));
176 ccid3_update_send_interval(hctx);
181 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
182 * @len: DCCP packet payload size in bytes
184 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
186 const u16 old_s = hctx->ccid3hctx_s;
188 hctx->ccid3hctx_s = tfrc_ewma(hctx->ccid3hctx_s, len, 9);
190 if (hctx->ccid3hctx_s != old_s)
191 ccid3_update_send_interval(hctx);
195 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
196 * As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
198 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
201 u32 delta = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count),
202 quarter_rtts = (4 * delta) / hctx->ccid3hctx_rtt;
204 if (quarter_rtts > 0) {
205 hctx->ccid3hctx_t_last_win_count = now;
206 hctx->ccid3hctx_last_win_count += min(quarter_rtts, 5U);
207 hctx->ccid3hctx_last_win_count &= 0xF; /* mod 16 */
211 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
213 struct sock *sk = (struct sock *)data;
214 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
215 unsigned long t_nfb = USEC_PER_SEC / 5;
218 if (sock_owned_by_user(sk)) {
219 /* Try again later. */
220 /* XXX: set some sensible MIB */
224 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
225 ccid3_tx_state_name(hctx->ccid3hctx_state));
227 if (hctx->ccid3hctx_state == TFRC_SSTATE_FBACK)
228 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
229 else if (hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
233 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
235 if (hctx->ccid3hctx_t_rto == 0 || /* no feedback received yet */
236 hctx->ccid3hctx_p == 0) {
238 /* halve send rate directly */
239 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x / 2,
240 (((__u64)hctx->ccid3hctx_s) << 6) /
242 ccid3_update_send_interval(hctx);
245 * Modify the cached value of X_recv
247 * If (X_calc > 2 * X_recv)
248 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
250 * X_recv = X_calc / 4;
252 * Note that X_recv is scaled by 2^6 while X_calc is not
254 BUG_ON(hctx->ccid3hctx_p && !hctx->ccid3hctx_x_calc);
256 if (hctx->ccid3hctx_x_calc > (hctx->ccid3hctx_x_recv >> 5))
257 hctx->ccid3hctx_x_recv =
258 max(hctx->ccid3hctx_x_recv / 2,
259 (((__u64)hctx->ccid3hctx_s) << 6) /
262 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc;
263 hctx->ccid3hctx_x_recv <<= 4;
265 ccid3_hc_tx_update_x(sk, NULL);
267 ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
268 (unsigned long long)hctx->ccid3hctx_x);
271 * Set new timeout for the nofeedback timer.
272 * See comments in packet_recv() regarding the value of t_RTO.
274 if (unlikely(hctx->ccid3hctx_t_rto == 0)) /* no feedback yet */
275 t_nfb = TFRC_INITIAL_TIMEOUT;
277 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
280 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
281 jiffies + usecs_to_jiffies(t_nfb));
289 * > 0: delay (in msecs) that should pass before actually sending
290 * = 0: can send immediately
291 * < 0: error condition; do not send packet
293 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
295 struct dccp_sock *dp = dccp_sk(sk);
296 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
297 ktime_t now = ktime_get_real();
301 * This function is called only for Data and DataAck packets. Sending
302 * zero-sized Data(Ack)s is theoretically possible, but for congestion
303 * control this case is pathological - ignore it.
305 if (unlikely(skb->len == 0))
308 switch (hctx->ccid3hctx_state) {
309 case TFRC_SSTATE_NO_SENT:
310 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
312 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
313 hctx->ccid3hctx_last_win_count = 0;
314 hctx->ccid3hctx_t_last_win_count = now;
316 /* Set t_0 for initial packet */
317 hctx->ccid3hctx_t_nom = now;
319 hctx->ccid3hctx_s = skb->len;
322 * Use initial RTT sample when available: recommended by erratum
323 * to RFC 4342. This implements the initialisation procedure of
324 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
326 if (dp->dccps_syn_rtt) {
327 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
328 hctx->ccid3hctx_rtt = dp->dccps_syn_rtt;
329 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
330 hctx->ccid3hctx_t_ld = now;
333 * Sender does not have RTT sample:
334 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
335 * is needed in several parts (e.g. window counter);
336 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
338 hctx->ccid3hctx_rtt = DCCP_FALLBACK_RTT;
339 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
340 hctx->ccid3hctx_x <<= 6;
342 ccid3_update_send_interval(hctx);
344 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
346 case TFRC_SSTATE_NO_FBACK:
347 case TFRC_SSTATE_FBACK:
348 delay = ktime_us_delta(hctx->ccid3hctx_t_nom, now);
349 ccid3_pr_debug("delay=%ld\n", (long)delay);
351 * Scheduling of packet transmissions [RFC 3448, 4.6]
353 * if (t_now > t_nom - delta)
354 * // send the packet now
356 * // send the packet in (t_nom - t_now) milliseconds.
358 if (delay - (s64)hctx->ccid3hctx_delta >= 1000)
359 return (u32)delay / 1000L;
361 ccid3_hc_tx_update_win_count(hctx, now);
363 case TFRC_SSTATE_TERM:
364 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
368 /* prepare to send now (add options etc.) */
369 dp->dccps_hc_tx_insert_options = 1;
370 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
372 /* set the nominal send time for the next following packet */
373 hctx->ccid3hctx_t_nom = ktime_add_us(hctx->ccid3hctx_t_nom,
374 hctx->ccid3hctx_t_ipi);
378 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
381 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
383 ccid3_hc_tx_update_s(hctx, len);
385 if (tfrc_tx_hist_add(&hctx->ccid3hctx_hist, dccp_sk(sk)->dccps_gss))
386 DCCP_CRIT("packet history - out of memory!");
389 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
391 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
392 struct ccid3_options_received *opt_recv;
397 /* we are only interested in ACKs */
398 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
399 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
401 /* ... and only in the established state */
402 if (hctx->ccid3hctx_state != TFRC_SSTATE_FBACK &&
403 hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
406 opt_recv = &hctx->ccid3hctx_options_received;
407 now = ktime_get_real();
409 /* Estimate RTT from history if ACK number is valid */
410 r_sample = tfrc_tx_hist_rtt(hctx->ccid3hctx_hist,
411 DCCP_SKB_CB(skb)->dccpd_ack_seq, now);
413 DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk), sk,
414 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type),
415 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq);
419 /* Update receive rate in units of 64 * bytes/second */
420 hctx->ccid3hctx_x_recv = opt_recv->ccid3or_receive_rate;
421 hctx->ccid3hctx_x_recv <<= 6;
423 /* Update loss event rate (which is scaled by 1e6) */
424 pinv = opt_recv->ccid3or_loss_event_rate;
425 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
426 hctx->ccid3hctx_p = 0;
427 else /* can not exceed 100% */
428 hctx->ccid3hctx_p = scaled_div(1, pinv);
430 * Validate new RTT sample and update moving average
432 r_sample = dccp_sample_rtt(sk, r_sample);
433 hctx->ccid3hctx_rtt = tfrc_ewma(hctx->ccid3hctx_rtt, r_sample, 9);
435 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
437 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
438 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
440 if (hctx->ccid3hctx_t_rto == 0) {
442 * Initial feedback packet: Larger Initial Windows (4.2)
444 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
445 hctx->ccid3hctx_t_ld = now;
447 ccid3_update_send_interval(hctx);
449 goto done_computing_x;
450 } else if (hctx->ccid3hctx_p == 0) {
452 * First feedback after nofeedback timer expiry (4.3)
454 goto done_computing_x;
458 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
459 if (hctx->ccid3hctx_p > 0)
460 hctx->ccid3hctx_x_calc =
461 tfrc_calc_x(hctx->ccid3hctx_s,
464 ccid3_hc_tx_update_x(sk, &now);
467 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
468 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
470 sk, hctx->ccid3hctx_rtt, r_sample,
471 hctx->ccid3hctx_s, hctx->ccid3hctx_p,
472 hctx->ccid3hctx_x_calc,
473 (unsigned)(hctx->ccid3hctx_x_recv >> 6),
474 (unsigned)(hctx->ccid3hctx_x >> 6));
476 /* unschedule no feedback timer */
477 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
480 * As we have calculated new ipi, delta, t_nom it is possible
481 * that we now can send a packet, so wake up dccp_wait_for_ccid
483 sk->sk_write_space(sk);
486 * Update timeout interval for the nofeedback timer.
487 * We use a configuration option to increase the lower bound.
488 * This can help avoid triggering the nofeedback timer too
489 * often ('spinning') on LANs with small RTTs.
491 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
492 (CONFIG_IP_DCCP_CCID3_RTO *
493 (USEC_PER_SEC / 1000)));
495 * Schedule no feedback timer to expire in
496 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
498 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
500 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
501 "expire in %lu jiffies (%luus)\n",
503 sk, usecs_to_jiffies(t_nfb), t_nfb);
505 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
506 jiffies + usecs_to_jiffies(t_nfb));
509 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
510 unsigned char len, u16 idx,
511 unsigned char *value)
514 const struct dccp_sock *dp = dccp_sk(sk);
515 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
516 struct ccid3_options_received *opt_recv;
519 opt_recv = &hctx->ccid3hctx_options_received;
521 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
522 opt_recv->ccid3or_seqno = dp->dccps_gsr;
523 opt_recv->ccid3or_loss_event_rate = ~0;
524 opt_recv->ccid3or_loss_intervals_idx = 0;
525 opt_recv->ccid3or_loss_intervals_len = 0;
526 opt_recv->ccid3or_receive_rate = 0;
530 case TFRC_OPT_LOSS_EVENT_RATE:
531 if (unlikely(len != 4)) {
532 DCCP_WARN("%s(%p), invalid len %d "
533 "for TFRC_OPT_LOSS_EVENT_RATE\n",
534 dccp_role(sk), sk, len);
537 opt_val = get_unaligned((__be32 *)value);
538 opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
539 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
541 opt_recv->ccid3or_loss_event_rate);
544 case TFRC_OPT_LOSS_INTERVALS:
545 opt_recv->ccid3or_loss_intervals_idx = idx;
546 opt_recv->ccid3or_loss_intervals_len = len;
547 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
549 opt_recv->ccid3or_loss_intervals_idx,
550 opt_recv->ccid3or_loss_intervals_len);
552 case TFRC_OPT_RECEIVE_RATE:
553 if (unlikely(len != 4)) {
554 DCCP_WARN("%s(%p), invalid len %d "
555 "for TFRC_OPT_RECEIVE_RATE\n",
556 dccp_role(sk), sk, len);
559 opt_val = get_unaligned((__be32 *)value);
560 opt_recv->ccid3or_receive_rate = ntohl(opt_val);
561 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
563 opt_recv->ccid3or_receive_rate);
571 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
573 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
575 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
576 hctx->ccid3hctx_hist = NULL;
577 setup_timer(&hctx->ccid3hctx_no_feedback_timer,
578 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
583 static void ccid3_hc_tx_exit(struct sock *sk)
585 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
587 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
588 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
590 tfrc_tx_hist_purge(&hctx->ccid3hctx_hist);
593 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
595 struct ccid3_hc_tx_sock *hctx;
597 /* Listen socks doesn't have a private CCID block */
598 if (sk->sk_state == DCCP_LISTEN)
601 hctx = ccid3_hc_tx_sk(sk);
602 info->tcpi_rto = hctx->ccid3hctx_t_rto;
603 info->tcpi_rtt = hctx->ccid3hctx_rtt;
606 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
607 u32 __user *optval, int __user *optlen)
609 const struct ccid3_hc_tx_sock *hctx;
612 /* Listen socks doesn't have a private CCID block */
613 if (sk->sk_state == DCCP_LISTEN)
616 hctx = ccid3_hc_tx_sk(sk);
618 case DCCP_SOCKOPT_CCID_TX_INFO:
619 if (len < sizeof(hctx->ccid3hctx_tfrc))
621 len = sizeof(hctx->ccid3hctx_tfrc);
622 val = &hctx->ccid3hctx_tfrc;
628 if (put_user(len, optlen) || copy_to_user(optval, val, len))
635 * Receiver Half-Connection Routines
638 /* CCID3 feedback types */
639 enum ccid3_fback_type {
640 CCID3_FBACK_NONE = 0,
642 CCID3_FBACK_PERIODIC,
643 CCID3_FBACK_PARAM_CHANGE
646 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
647 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
649 static char *ccid3_rx_state_names[] = {
650 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
651 [TFRC_RSTATE_DATA] = "DATA",
652 [TFRC_RSTATE_TERM] = "TERM",
655 return ccid3_rx_state_names[state];
659 static void ccid3_hc_rx_set_state(struct sock *sk,
660 enum ccid3_hc_rx_states state)
662 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
663 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
665 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
666 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
667 ccid3_rx_state_name(state));
668 WARN_ON(state == oldstate);
669 hcrx->ccid3hcrx_state = state;
672 static void ccid3_hc_rx_send_feedback(struct sock *sk,
673 const struct sk_buff *skb,
674 enum ccid3_fback_type fbtype)
676 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
677 struct dccp_sock *dp = dccp_sk(sk);
681 if (unlikely(hcrx->ccid3hcrx_state == TFRC_RSTATE_TERM))
684 now = ktime_get_real();
687 case CCID3_FBACK_INITIAL:
688 hcrx->ccid3hcrx_x_recv = 0;
689 hcrx->ccid3hcrx_pinv = ~0U; /* see RFC 4342, 8.5 */
691 case CCID3_FBACK_PARAM_CHANGE:
693 * When parameters change (new loss or p > p_prev), we do not
694 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
695 * need to reuse the previous value of X_recv. However, when
696 * X_recv was 0 (due to early loss), this would kill X down to
697 * s/t_mbi (i.e. one packet in 64 seconds).
698 * To avoid such drastic reduction, we approximate X_recv as
699 * the number of bytes since last feedback.
700 * This is a safe fallback, since X is bounded above by X_calc.
702 if (hcrx->ccid3hcrx_x_recv > 0)
705 case CCID3_FBACK_PERIODIC:
706 delta = ktime_us_delta(now, hcrx->ccid3hcrx_tstamp_last_feedback);
708 DCCP_BUG("delta (%ld) <= 0", (long)delta);
710 hcrx->ccid3hcrx_x_recv =
711 scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
717 ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta,
718 hcrx->ccid3hcrx_x_recv, hcrx->ccid3hcrx_pinv);
720 hcrx->ccid3hcrx_tstamp_last_feedback = now;
721 hcrx->ccid3hcrx_last_counter = dccp_hdr(skb)->dccph_ccval;
722 hcrx->ccid3hcrx_bytes_recv = 0;
724 dp->dccps_hc_rx_insert_options = 1;
728 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
730 const struct ccid3_hc_rx_sock *hcrx;
733 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
736 hcrx = ccid3_hc_rx_sk(sk);
738 if (dccp_packet_without_ack(skb))
741 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
742 pinv = htonl(hcrx->ccid3hcrx_pinv);
744 if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
745 &pinv, sizeof(pinv)) ||
746 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
747 &x_recv, sizeof(x_recv)))
753 /** ccid3_first_li - Implements [RFC 3448, 6.3.1]
755 * Determine the length of the first loss interval via inverse lookup.
756 * Assume that X_recv can be computed by the throughput equation
760 * Find some p such that f(p) = fval; return 1/p (scaled).
762 static u32 ccid3_first_li(struct sock *sk)
764 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
765 u32 x_recv, p, delta;
768 if (hcrx->ccid3hcrx_rtt == 0) {
769 DCCP_WARN("No RTT estimate available, using fallback RTT\n");
770 hcrx->ccid3hcrx_rtt = DCCP_FALLBACK_RTT;
773 delta = ktime_to_us(net_timedelta(hcrx->ccid3hcrx_tstamp_last_feedback));
774 x_recv = scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
775 if (x_recv == 0) { /* would also trigger divide-by-zero */
776 DCCP_WARN("X_recv==0\n");
777 if ((x_recv = hcrx->ccid3hcrx_x_recv) == 0) {
778 DCCP_BUG("stored value of X_recv is zero");
783 fval = scaled_div(hcrx->ccid3hcrx_s, hcrx->ccid3hcrx_rtt);
784 fval = scaled_div32(fval, x_recv);
785 p = tfrc_calc_x_reverse_lookup(fval);
787 ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
788 "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
790 return p == 0 ? ~0U : scaled_div(1, p);
793 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
795 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
796 enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
797 const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
798 const bool is_data_packet = dccp_data_packet(skb);
800 if (unlikely(hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)) {
801 if (is_data_packet) {
802 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
803 do_feedback = CCID3_FBACK_INITIAL;
804 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
805 hcrx->ccid3hcrx_s = payload;
807 * Not necessary to update ccid3hcrx_bytes_recv here,
808 * since X_recv = 0 for the first feedback packet (cf.
809 * RFC 3448, 6.3) -- gerrit
815 if (tfrc_rx_hist_duplicate(&hcrx->ccid3hcrx_hist, skb))
816 return; /* done receiving */
818 if (is_data_packet) {
819 const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
821 * Update moving-average of s and the sum of received payload bytes
823 hcrx->ccid3hcrx_s = tfrc_ewma(hcrx->ccid3hcrx_s, payload, 9);
824 hcrx->ccid3hcrx_bytes_recv += payload;
828 * Perform loss detection and handle pending losses
830 if (tfrc_rx_handle_loss(&hcrx->ccid3hcrx_hist, &hcrx->ccid3hcrx_li_hist,
831 skb, ndp, ccid3_first_li, sk)) {
832 do_feedback = CCID3_FBACK_PARAM_CHANGE;
836 if (tfrc_rx_hist_loss_pending(&hcrx->ccid3hcrx_hist))
837 return; /* done receiving */
840 * Handle data packets: RTT sampling and monitoring p
842 if (unlikely(!is_data_packet))
845 if (!tfrc_lh_is_initialised(&hcrx->ccid3hcrx_li_hist)) {
846 const u32 sample = tfrc_rx_hist_sample_rtt(&hcrx->ccid3hcrx_hist, skb);
848 * Empty loss history: no loss so far, hence p stays 0.
849 * Sample RTT values, since an RTT estimate is required for the
850 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
853 hcrx->ccid3hcrx_rtt = tfrc_ewma(hcrx->ccid3hcrx_rtt, sample, 9);
855 } else if (tfrc_lh_update_i_mean(&hcrx->ccid3hcrx_li_hist, skb)) {
857 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
858 * has decreased (resp. p has increased), send feedback now.
860 do_feedback = CCID3_FBACK_PARAM_CHANGE;
864 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
866 if (SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->ccid3hcrx_last_counter) > 3)
867 do_feedback = CCID3_FBACK_PERIODIC;
870 tfrc_rx_hist_add_packet(&hcrx->ccid3hcrx_hist, skb, ndp);
874 ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
877 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
879 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
881 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
882 tfrc_lh_init(&hcrx->ccid3hcrx_li_hist);
883 return tfrc_rx_hist_alloc(&hcrx->ccid3hcrx_hist);
886 static void ccid3_hc_rx_exit(struct sock *sk)
888 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
890 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
892 tfrc_rx_hist_purge(&hcrx->ccid3hcrx_hist);
893 tfrc_lh_cleanup(&hcrx->ccid3hcrx_li_hist);
896 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
898 const struct ccid3_hc_rx_sock *hcrx;
900 /* Listen socks doesn't have a private CCID block */
901 if (sk->sk_state == DCCP_LISTEN)
904 hcrx = ccid3_hc_rx_sk(sk);
905 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
906 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
907 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
910 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
911 u32 __user *optval, int __user *optlen)
913 const struct ccid3_hc_rx_sock *hcrx;
914 struct tfrc_rx_info rx_info;
917 /* Listen socks doesn't have a private CCID block */
918 if (sk->sk_state == DCCP_LISTEN)
921 hcrx = ccid3_hc_rx_sk(sk);
923 case DCCP_SOCKOPT_CCID_RX_INFO:
924 if (len < sizeof(rx_info))
926 rx_info.tfrcrx_x_recv = hcrx->ccid3hcrx_x_recv;
927 rx_info.tfrcrx_rtt = hcrx->ccid3hcrx_rtt;
928 rx_info.tfrcrx_p = hcrx->ccid3hcrx_pinv == 0 ? ~0U :
929 scaled_div(1, hcrx->ccid3hcrx_pinv);
930 len = sizeof(rx_info);
937 if (put_user(len, optlen) || copy_to_user(optval, val, len))
943 static struct ccid_operations ccid3 = {
944 .ccid_id = DCCPC_CCID3,
945 .ccid_name = "TCP-Friendly Rate Control",
946 .ccid_owner = THIS_MODULE,
947 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
948 .ccid_hc_tx_init = ccid3_hc_tx_init,
949 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
950 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
951 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
952 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
953 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
954 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
955 .ccid_hc_rx_init = ccid3_hc_rx_init,
956 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
957 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
958 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
959 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
960 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
961 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
962 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
965 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
966 module_param(ccid3_debug, bool, 0644);
967 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
970 static __init int ccid3_module_init(void)
972 return ccid_register(&ccid3);
974 module_init(ccid3_module_init);
976 static __exit void ccid3_module_exit(void)
978 ccid_unregister(&ccid3);
980 module_exit(ccid3_module_exit);
982 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
983 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
984 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
985 MODULE_LICENSE("GPL");
986 MODULE_ALIAS("net-dccp-ccid-3");