2 * net/dccp/ccids/ccid3.c
4 * Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
5 * Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
7 * An implementation of the DCCP protocol
9 * This code has been developed by the University of Waikato WAND
10 * research group. For further information please see http://www.wand.net.nz/
12 * This code also uses code from Lulea University, rereleased as GPL by its
14 * Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
16 * Changes to meet Linux coding standards, to make it meet latest ccid3 draft
17 * and to make it work as a loadable module in the DCCP stack written by
18 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
20 * Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
22 * This program is free software; you can redistribute it and/or modify
23 * it under the terms of the GNU General Public License as published by
24 * the Free Software Foundation; either version 2 of the License, or
25 * (at your option) any later version.
27 * This program is distributed in the hope that it will be useful,
28 * but WITHOUT ANY WARRANTY; without even the implied warranty of
29 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
30 * GNU General Public License for more details.
32 * You should have received a copy of the GNU General Public License
33 * along with this program; if not, write to the Free Software
34 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
38 #include "lib/packet_history.h"
39 #include "lib/loss_interval.h"
43 #include <asm/unaligned.h>
45 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
46 static int ccid3_debug;
47 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
49 #define ccid3_pr_debug(format, a...)
52 static struct dccp_rx_hist *ccid3_rx_hist;
55 * Transmitter Half-Connection Routines
57 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
58 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
60 static char *ccid3_state_names[] = {
61 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
62 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
63 [TFRC_SSTATE_FBACK] = "FBACK",
64 [TFRC_SSTATE_TERM] = "TERM",
67 return ccid3_state_names[state];
71 static void ccid3_hc_tx_set_state(struct sock *sk,
72 enum ccid3_hc_tx_states state)
74 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
75 enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
77 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
78 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
79 ccid3_tx_state_name(state));
80 WARN_ON(state == oldstate);
81 hctx->ccid3hctx_state = state;
85 * Compute the initial sending rate X_init in the manner of RFC 3390:
87 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
89 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
90 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
91 * For consistency with other parts of the code, X_init is scaled by 2^6.
93 static inline u64 rfc3390_initial_rate(struct sock *sk)
95 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
96 const __u32 w_init = min_t(__u32, 4 * hctx->ccid3hctx_s,
97 max_t(__u32, 2 * hctx->ccid3hctx_s, 4380));
99 return scaled_div(w_init << 6, hctx->ccid3hctx_rtt);
103 * Recalculate t_ipi and delta (should be called whenever X changes)
105 static inline void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
107 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
108 hctx->ccid3hctx_t_ipi = scaled_div32(((u64)hctx->ccid3hctx_s) << 6,
111 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
112 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
113 TFRC_OPSYS_HALF_TIME_GRAN);
115 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n",
116 hctx->ccid3hctx_t_ipi, hctx->ccid3hctx_delta,
117 hctx->ccid3hctx_s, (unsigned)(hctx->ccid3hctx_x >> 6));
121 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
123 u32 delta = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count);
125 return delta / hctx->ccid3hctx_rtt;
129 * ccid3_hc_tx_update_x - Update allowed sending rate X
130 * @stamp: most recent time if available - can be left NULL.
131 * This function tracks draft rfc3448bis, check there for latest details.
133 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
134 * fine-grained resolution of sending rates. This requires scaling by 2^6
135 * throughout the code. Only X_calc is unscaled (in bytes/second).
138 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
141 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
142 __u64 min_rate = 2 * hctx->ccid3hctx_x_recv;
143 const __u64 old_x = hctx->ccid3hctx_x;
144 ktime_t now = stamp? *stamp : ktime_get_real();
147 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
148 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
149 * a sender is idle if it has not sent anything over a 2-RTT-period.
150 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
152 if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) {
153 min_rate = rfc3390_initial_rate(sk);
154 min_rate = max(min_rate, 2 * hctx->ccid3hctx_x_recv);
157 if (hctx->ccid3hctx_p > 0) {
159 hctx->ccid3hctx_x = min(((__u64)hctx->ccid3hctx_x_calc) << 6,
161 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
162 (((__u64)hctx->ccid3hctx_s) << 6) /
165 } else if (ktime_us_delta(now, hctx->ccid3hctx_t_ld)
166 - (s64)hctx->ccid3hctx_rtt >= 0) {
169 max(min(2 * hctx->ccid3hctx_x, min_rate),
170 scaled_div(((__u64)hctx->ccid3hctx_s) << 6,
171 hctx->ccid3hctx_rtt));
172 hctx->ccid3hctx_t_ld = now;
175 if (hctx->ccid3hctx_x != old_x) {
176 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
177 "X_recv=%u\n", (unsigned)(old_x >> 6),
178 (unsigned)(hctx->ccid3hctx_x >> 6),
179 hctx->ccid3hctx_x_calc,
180 (unsigned)(hctx->ccid3hctx_x_recv >> 6));
182 ccid3_update_send_interval(hctx);
187 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
188 * @len: DCCP packet payload size in bytes
190 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
192 const u16 old_s = hctx->ccid3hctx_s;
194 hctx->ccid3hctx_s = tfrc_ewma(hctx->ccid3hctx_s, len, 9);
196 if (hctx->ccid3hctx_s != old_s)
197 ccid3_update_send_interval(hctx);
201 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
202 * The algorithm is not applicable if RTT < 4 microseconds.
204 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
209 if (unlikely(hctx->ccid3hctx_rtt < 4)) /* avoid divide-by-zero */
212 quarter_rtts = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count);
213 quarter_rtts /= hctx->ccid3hctx_rtt / 4;
215 if (quarter_rtts > 0) {
216 hctx->ccid3hctx_t_last_win_count = now;
217 hctx->ccid3hctx_last_win_count += min_t(u32, quarter_rtts, 5);
218 hctx->ccid3hctx_last_win_count &= 0xF; /* mod 16 */
222 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
224 struct sock *sk = (struct sock *)data;
225 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
226 unsigned long t_nfb = USEC_PER_SEC / 5;
229 if (sock_owned_by_user(sk)) {
230 /* Try again later. */
231 /* XXX: set some sensible MIB */
235 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
236 ccid3_tx_state_name(hctx->ccid3hctx_state));
238 switch (hctx->ccid3hctx_state) {
239 case TFRC_SSTATE_NO_FBACK:
240 /* RFC 3448, 4.4: Halve send rate directly */
241 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x / 2,
242 (((__u64)hctx->ccid3hctx_s) << 6) /
245 ccid3_pr_debug("%s(%p, state=%s), updated tx rate to %u "
246 "bytes/s\n", dccp_role(sk), sk,
247 ccid3_tx_state_name(hctx->ccid3hctx_state),
248 (unsigned)(hctx->ccid3hctx_x >> 6));
249 /* The value of R is still undefined and so we can not recompute
250 * the timeout value. Keep initial value as per [RFC 4342, 5]. */
251 t_nfb = TFRC_INITIAL_TIMEOUT;
252 ccid3_update_send_interval(hctx);
254 case TFRC_SSTATE_FBACK:
256 * Modify the cached value of X_recv [RFC 3448, 4.4]
258 * If (p == 0 || X_calc > 2 * X_recv)
259 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
261 * X_recv = X_calc / 4;
263 * Note that X_recv is scaled by 2^6 while X_calc is not
265 BUG_ON(hctx->ccid3hctx_p && !hctx->ccid3hctx_x_calc);
267 if (hctx->ccid3hctx_p == 0 ||
268 (hctx->ccid3hctx_x_calc > (hctx->ccid3hctx_x_recv >> 5))) {
270 hctx->ccid3hctx_x_recv =
271 max(hctx->ccid3hctx_x_recv / 2,
272 (((__u64)hctx->ccid3hctx_s) << 6) /
275 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc;
276 hctx->ccid3hctx_x_recv <<= 4;
278 /* Now recalculate X [RFC 3448, 4.3, step (4)] */
279 ccid3_hc_tx_update_x(sk, NULL);
281 * Schedule no feedback timer to expire in
282 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
283 * See comments in packet_recv() regarding the value of t_RTO.
285 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
287 case TFRC_SSTATE_NO_SENT:
288 DCCP_BUG("%s(%p) - Illegal state NO_SENT", dccp_role(sk), sk);
290 case TFRC_SSTATE_TERM:
295 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
296 jiffies + usecs_to_jiffies(t_nfb));
304 * > 0: delay (in msecs) that should pass before actually sending
305 * = 0: can send immediately
306 * < 0: error condition; do not send packet
308 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
310 struct dccp_sock *dp = dccp_sk(sk);
311 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
312 ktime_t now = ktime_get_real();
316 * This function is called only for Data and DataAck packets. Sending
317 * zero-sized Data(Ack)s is theoretically possible, but for congestion
318 * control this case is pathological - ignore it.
320 if (unlikely(skb->len == 0))
323 switch (hctx->ccid3hctx_state) {
324 case TFRC_SSTATE_NO_SENT:
325 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
327 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
328 hctx->ccid3hctx_last_win_count = 0;
329 hctx->ccid3hctx_t_last_win_count = now;
331 /* Set t_0 for initial packet */
332 hctx->ccid3hctx_t_nom = now;
334 hctx->ccid3hctx_s = skb->len;
337 * Use initial RTT sample when available: recommended by erratum
338 * to RFC 4342. This implements the initialisation procedure of
339 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
341 if (dp->dccps_syn_rtt) {
342 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
343 hctx->ccid3hctx_rtt = dp->dccps_syn_rtt;
344 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
345 hctx->ccid3hctx_t_ld = now;
347 /* Sender does not have RTT sample: X_pps = 1 pkt/sec */
348 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
349 hctx->ccid3hctx_x <<= 6;
351 ccid3_update_send_interval(hctx);
353 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
355 case TFRC_SSTATE_NO_FBACK:
356 case TFRC_SSTATE_FBACK:
357 delay = ktime_us_delta(hctx->ccid3hctx_t_nom, now);
358 ccid3_pr_debug("delay=%ld\n", (long)delay);
360 * Scheduling of packet transmissions [RFC 3448, 4.6]
362 * if (t_now > t_nom - delta)
363 * // send the packet now
365 * // send the packet in (t_nom - t_now) milliseconds.
367 if (delay - (s64)hctx->ccid3hctx_delta >= 1000)
368 return (u32)delay / 1000L;
370 ccid3_hc_tx_update_win_count(hctx, now);
372 case TFRC_SSTATE_TERM:
373 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
377 /* prepare to send now (add options etc.) */
378 dp->dccps_hc_tx_insert_options = 1;
379 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
381 /* set the nominal send time for the next following packet */
382 hctx->ccid3hctx_t_nom = ktime_add_us(hctx->ccid3hctx_t_nom,
383 hctx->ccid3hctx_t_ipi);
387 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
390 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
392 ccid3_hc_tx_update_s(hctx, len);
394 if (tfrc_tx_hist_add(&hctx->ccid3hctx_hist, dccp_sk(sk)->dccps_gss))
395 DCCP_CRIT("packet history - out of memory!");
398 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
400 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
401 struct ccid3_options_received *opt_recv;
406 /* we are only interested in ACKs */
407 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
408 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
411 opt_recv = &hctx->ccid3hctx_options_received;
413 switch (hctx->ccid3hctx_state) {
414 case TFRC_SSTATE_NO_FBACK:
415 case TFRC_SSTATE_FBACK:
416 now = ktime_get_real();
418 /* estimate RTT from history if ACK number is valid */
419 r_sample = tfrc_tx_hist_rtt(hctx->ccid3hctx_hist,
420 DCCP_SKB_CB(skb)->dccpd_ack_seq, now);
422 DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk), sk,
423 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type),
424 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq);
428 /* Update receive rate in units of 64 * bytes/second */
429 hctx->ccid3hctx_x_recv = opt_recv->ccid3or_receive_rate;
430 hctx->ccid3hctx_x_recv <<= 6;
432 /* Update loss event rate */
433 pinv = opt_recv->ccid3or_loss_event_rate;
434 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
435 hctx->ccid3hctx_p = 0;
436 else /* can not exceed 100% */
437 hctx->ccid3hctx_p = 1000000 / pinv;
439 * Validate new RTT sample and update moving average
441 r_sample = dccp_sample_rtt(sk, r_sample);
442 hctx->ccid3hctx_rtt = tfrc_ewma(hctx->ccid3hctx_rtt, r_sample, 9);
444 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
446 * Larger Initial Windows [RFC 4342, sec. 5]
448 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
449 hctx->ccid3hctx_t_ld = now;
451 ccid3_update_send_interval(hctx);
453 ccid3_pr_debug("%s(%p), s=%u, MSS=%u, "
454 "R_sample=%uus, X=%u\n", dccp_role(sk),
455 sk, hctx->ccid3hctx_s,
456 dccp_sk(sk)->dccps_mss_cache, r_sample,
457 (unsigned)(hctx->ccid3hctx_x >> 6));
459 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
462 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
463 if (hctx->ccid3hctx_p > 0)
464 hctx->ccid3hctx_x_calc =
465 tfrc_calc_x(hctx->ccid3hctx_s,
468 ccid3_hc_tx_update_x(sk, &now);
470 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
471 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
473 sk, hctx->ccid3hctx_rtt, r_sample,
474 hctx->ccid3hctx_s, hctx->ccid3hctx_p,
475 hctx->ccid3hctx_x_calc,
476 (unsigned)(hctx->ccid3hctx_x_recv >> 6),
477 (unsigned)(hctx->ccid3hctx_x >> 6));
480 /* unschedule no feedback timer */
481 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
484 * As we have calculated new ipi, delta, t_nom it is possible
485 * that we now can send a packet, so wake up dccp_wait_for_ccid
487 sk->sk_write_space(sk);
490 * Update timeout interval for the nofeedback timer.
491 * We use a configuration option to increase the lower bound.
492 * This can help avoid triggering the nofeedback timer too
493 * often ('spinning') on LANs with small RTTs.
495 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
496 CONFIG_IP_DCCP_CCID3_RTO *
497 (USEC_PER_SEC/1000));
499 * Schedule no feedback timer to expire in
500 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
502 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
504 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
505 "expire in %lu jiffies (%luus)\n",
507 sk, usecs_to_jiffies(t_nfb), t_nfb);
509 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
510 jiffies + usecs_to_jiffies(t_nfb));
512 case TFRC_SSTATE_NO_SENT: /* fall through */
513 case TFRC_SSTATE_TERM: /* ignore feedback when closing */
518 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
519 unsigned char len, u16 idx,
520 unsigned char *value)
523 const struct dccp_sock *dp = dccp_sk(sk);
524 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
525 struct ccid3_options_received *opt_recv;
528 opt_recv = &hctx->ccid3hctx_options_received;
530 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
531 opt_recv->ccid3or_seqno = dp->dccps_gsr;
532 opt_recv->ccid3or_loss_event_rate = ~0;
533 opt_recv->ccid3or_loss_intervals_idx = 0;
534 opt_recv->ccid3or_loss_intervals_len = 0;
535 opt_recv->ccid3or_receive_rate = 0;
539 case TFRC_OPT_LOSS_EVENT_RATE:
540 if (unlikely(len != 4)) {
541 DCCP_WARN("%s(%p), invalid len %d "
542 "for TFRC_OPT_LOSS_EVENT_RATE\n",
543 dccp_role(sk), sk, len);
546 opt_val = get_unaligned((__be32 *)value);
547 opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
548 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
550 opt_recv->ccid3or_loss_event_rate);
553 case TFRC_OPT_LOSS_INTERVALS:
554 opt_recv->ccid3or_loss_intervals_idx = idx;
555 opt_recv->ccid3or_loss_intervals_len = len;
556 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
558 opt_recv->ccid3or_loss_intervals_idx,
559 opt_recv->ccid3or_loss_intervals_len);
561 case TFRC_OPT_RECEIVE_RATE:
562 if (unlikely(len != 4)) {
563 DCCP_WARN("%s(%p), invalid len %d "
564 "for TFRC_OPT_RECEIVE_RATE\n",
565 dccp_role(sk), sk, len);
568 opt_val = get_unaligned((__be32 *)value);
569 opt_recv->ccid3or_receive_rate = ntohl(opt_val);
570 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
572 opt_recv->ccid3or_receive_rate);
580 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
582 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
584 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
585 hctx->ccid3hctx_hist = NULL;
586 setup_timer(&hctx->ccid3hctx_no_feedback_timer,
587 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
592 static void ccid3_hc_tx_exit(struct sock *sk)
594 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
596 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
597 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
599 tfrc_tx_hist_purge(&hctx->ccid3hctx_hist);
602 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
604 struct ccid3_hc_tx_sock *hctx;
606 /* Listen socks doesn't have a private CCID block */
607 if (sk->sk_state == DCCP_LISTEN)
610 hctx = ccid3_hc_tx_sk(sk);
611 info->tcpi_rto = hctx->ccid3hctx_t_rto;
612 info->tcpi_rtt = hctx->ccid3hctx_rtt;
615 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
616 u32 __user *optval, int __user *optlen)
618 const struct ccid3_hc_tx_sock *hctx;
621 /* Listen socks doesn't have a private CCID block */
622 if (sk->sk_state == DCCP_LISTEN)
625 hctx = ccid3_hc_tx_sk(sk);
627 case DCCP_SOCKOPT_CCID_TX_INFO:
628 if (len < sizeof(hctx->ccid3hctx_tfrc))
630 len = sizeof(hctx->ccid3hctx_tfrc);
631 val = &hctx->ccid3hctx_tfrc;
637 if (put_user(len, optlen) || copy_to_user(optval, val, len))
644 * Receiver Half-Connection Routines
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 inline void ccid3_hc_rx_update_s(struct ccid3_hc_rx_sock *hcrx, int len)
674 if (likely(len > 0)) /* don't update on empty packets (e.g. ACKs) */
675 hcrx->ccid3hcrx_s = tfrc_ewma(hcrx->ccid3hcrx_s, len, 9);
678 static void ccid3_hc_rx_send_feedback(struct sock *sk)
680 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
681 struct dccp_sock *dp = dccp_sk(sk);
682 struct dccp_rx_hist_entry *packet;
686 ccid3_pr_debug("%s(%p) - entry \n", dccp_role(sk), sk);
688 now = ktime_get_real();
690 switch (hcrx->ccid3hcrx_state) {
691 case TFRC_RSTATE_NO_DATA:
692 hcrx->ccid3hcrx_x_recv = 0;
694 case TFRC_RSTATE_DATA:
695 delta = ktime_us_delta(now,
696 hcrx->ccid3hcrx_tstamp_last_feedback);
697 DCCP_BUG_ON(delta < 0);
698 hcrx->ccid3hcrx_x_recv =
699 scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
701 case TFRC_RSTATE_TERM:
702 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
706 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
707 if (unlikely(packet == NULL)) {
708 DCCP_WARN("%s(%p), no data packet in history!\n",
713 hcrx->ccid3hcrx_tstamp_last_feedback = now;
714 hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
715 hcrx->ccid3hcrx_bytes_recv = 0;
717 if (hcrx->ccid3hcrx_p == 0)
718 hcrx->ccid3hcrx_pinv = ~0U; /* see RFC 4342, 8.5 */
719 else if (hcrx->ccid3hcrx_p > 1000000) {
720 DCCP_WARN("p (%u) > 100%%\n", hcrx->ccid3hcrx_p);
721 hcrx->ccid3hcrx_pinv = 1; /* use 100% in this case */
723 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
725 dp->dccps_hc_rx_insert_options = 1;
729 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
731 const struct ccid3_hc_rx_sock *hcrx;
734 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
737 hcrx = ccid3_hc_rx_sk(sk);
738 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
740 if (dccp_packet_without_ack(skb))
743 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
744 pinv = htonl(hcrx->ccid3hcrx_pinv);
746 if (dccp_insert_option_timestamp(sk, skb) ||
747 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
748 &pinv, sizeof(pinv)) ||
749 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
750 &x_recv, sizeof(x_recv)))
756 static int ccid3_hc_rx_detect_loss(struct sock *sk,
757 struct dccp_rx_hist_entry *packet)
759 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
760 struct dccp_rx_hist_entry *rx_hist =
761 dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
762 u64 seqno = packet->dccphrx_seqno;
768 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
771 follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
772 hcrx->ccid3hcrx_seqno_nonloss = seqno;
773 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
778 while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
779 > TFRC_RECV_NUM_LATE_LOSS) {
781 dccp_li_update_li(sk,
782 &hcrx->ccid3hcrx_li_hist,
783 &hcrx->ccid3hcrx_hist,
784 hcrx->ccid3hcrx_tstamp_last_feedback,
786 hcrx->ccid3hcrx_bytes_recv,
787 hcrx->ccid3hcrx_x_recv,
788 hcrx->ccid3hcrx_seqno_nonloss,
789 hcrx->ccid3hcrx_ccval_nonloss);
790 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
791 dccp_inc_seqno(&tmp_seqno);
792 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
793 dccp_inc_seqno(&tmp_seqno);
794 while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
795 tmp_seqno, &ccval)) {
796 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
797 hcrx->ccid3hcrx_ccval_nonloss = ccval;
798 dccp_inc_seqno(&tmp_seqno);
802 /* FIXME - this code could be simplified with above while */
803 /* but works at moment */
804 if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
805 hcrx->ccid3hcrx_seqno_nonloss = seqno;
806 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
810 dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
811 &hcrx->ccid3hcrx_li_hist, packet,
812 hcrx->ccid3hcrx_seqno_nonloss);
816 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
818 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
819 const struct dccp_options_received *opt_recv;
820 struct dccp_rx_hist_entry *packet;
821 u32 p_prev, r_sample, rtt_prev;
822 int loss, payload_size;
825 opt_recv = &dccp_sk(sk)->dccps_options_received;
827 switch (DCCP_SKB_CB(skb)->dccpd_type) {
829 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
831 case DCCP_PKT_DATAACK:
832 if (opt_recv->dccpor_timestamp_echo == 0)
834 r_sample = dccp_timestamp() - opt_recv->dccpor_timestamp_echo;
835 rtt_prev = hcrx->ccid3hcrx_rtt;
836 r_sample = dccp_sample_rtt(sk, 10 * r_sample);
838 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
839 hcrx->ccid3hcrx_rtt = r_sample;
841 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
844 if (rtt_prev != hcrx->ccid3hcrx_rtt)
845 ccid3_pr_debug("%s(%p), New RTT=%uus, elapsed time=%u\n",
846 dccp_role(sk), sk, hcrx->ccid3hcrx_rtt,
847 opt_recv->dccpor_elapsed_time);
851 default: /* We're not interested in other packet types, move along */
855 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, opt_recv->dccpor_ndp,
857 if (unlikely(packet == NULL)) {
858 DCCP_WARN("%s(%p), Not enough mem to add rx packet "
859 "to history, consider it lost!\n", dccp_role(sk), sk);
863 loss = ccid3_hc_rx_detect_loss(sk, packet);
865 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
868 payload_size = skb->len - dccp_hdr(skb)->dccph_doff * 4;
869 ccid3_hc_rx_update_s(hcrx, payload_size);
871 switch (hcrx->ccid3hcrx_state) {
872 case TFRC_RSTATE_NO_DATA:
873 ccid3_pr_debug("%s(%p, state=%s), skb=%p, sending initial "
874 "feedback\n", dccp_role(sk), sk,
875 dccp_state_name(sk->sk_state), skb);
876 ccid3_hc_rx_send_feedback(sk);
877 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
879 case TFRC_RSTATE_DATA:
880 hcrx->ccid3hcrx_bytes_recv += payload_size;
884 now = ktime_get_real();
885 if ((ktime_us_delta(now, hcrx->ccid3hcrx_tstamp_last_ack) -
886 (s64)hcrx->ccid3hcrx_rtt) >= 0) {
887 hcrx->ccid3hcrx_tstamp_last_ack = now;
888 ccid3_hc_rx_send_feedback(sk);
891 case TFRC_RSTATE_TERM:
892 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
896 /* Dealing with packet loss */
897 ccid3_pr_debug("%s(%p, state=%s), data loss! Reacting...\n",
898 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
900 p_prev = hcrx->ccid3hcrx_p;
902 /* Calculate loss event rate */
903 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
904 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
906 /* Scaling up by 1000000 as fixed decimal */
908 hcrx->ccid3hcrx_p = 1000000 / i_mean;
910 DCCP_BUG("empty loss history");
912 if (hcrx->ccid3hcrx_p > p_prev) {
913 ccid3_hc_rx_send_feedback(sk);
918 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
920 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
922 ccid3_pr_debug("entry\n");
924 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
925 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
926 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
927 hcrx->ccid3hcrx_tstamp_last_feedback =
928 hcrx->ccid3hcrx_tstamp_last_ack = ktime_get_real();
932 static void ccid3_hc_rx_exit(struct sock *sk)
934 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
936 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
938 /* Empty packet history */
939 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
941 /* Empty loss interval history */
942 dccp_li_hist_purge(&hcrx->ccid3hcrx_li_hist);
945 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
947 const struct ccid3_hc_rx_sock *hcrx;
949 /* Listen socks doesn't have a private CCID block */
950 if (sk->sk_state == DCCP_LISTEN)
953 hcrx = ccid3_hc_rx_sk(sk);
954 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
955 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
956 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
959 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
960 u32 __user *optval, int __user *optlen)
962 const struct ccid3_hc_rx_sock *hcrx;
965 /* Listen socks doesn't have a private CCID block */
966 if (sk->sk_state == DCCP_LISTEN)
969 hcrx = ccid3_hc_rx_sk(sk);
971 case DCCP_SOCKOPT_CCID_RX_INFO:
972 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
974 len = sizeof(hcrx->ccid3hcrx_tfrc);
975 val = &hcrx->ccid3hcrx_tfrc;
981 if (put_user(len, optlen) || copy_to_user(optval, val, len))
987 static struct ccid_operations ccid3 = {
988 .ccid_id = DCCPC_CCID3,
989 .ccid_name = "ccid3",
990 .ccid_owner = THIS_MODULE,
991 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
992 .ccid_hc_tx_init = ccid3_hc_tx_init,
993 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
994 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
995 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
996 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
997 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
998 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
999 .ccid_hc_rx_init = ccid3_hc_rx_init,
1000 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1001 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1002 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1003 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1004 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1005 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1006 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1009 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
1010 module_param(ccid3_debug, bool, 0444);
1011 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1014 static __init int ccid3_module_init(void)
1018 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1019 if (ccid3_rx_hist == NULL)
1022 rc = ccid_register(&ccid3);
1029 dccp_rx_hist_delete(ccid3_rx_hist);
1030 ccid3_rx_hist = NULL;
1033 module_init(ccid3_module_init);
1035 static __exit void ccid3_module_exit(void)
1037 ccid_unregister(&ccid3);
1039 if (ccid3_rx_hist != NULL) {
1040 dccp_rx_hist_delete(ccid3_rx_hist);
1041 ccid3_rx_hist = NULL;
1044 module_exit(ccid3_module_exit);
1046 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
1047 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1048 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1049 MODULE_LICENSE("GPL");
1050 MODULE_ALIAS("net-dccp-ccid-3");