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_tx_hist *ccid3_tx_hist;
53 static struct dccp_rx_hist *ccid3_rx_hist;
56 * Transmitter Half-Connection Routines
58 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
59 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
61 static char *ccid3_state_names[] = {
62 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
63 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
64 [TFRC_SSTATE_FBACK] = "FBACK",
65 [TFRC_SSTATE_TERM] = "TERM",
68 return ccid3_state_names[state];
72 static void ccid3_hc_tx_set_state(struct sock *sk,
73 enum ccid3_hc_tx_states state)
75 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
76 enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
78 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
79 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
80 ccid3_tx_state_name(state));
81 WARN_ON(state == oldstate);
82 hctx->ccid3hctx_state = state;
86 * Compute the initial sending rate X_init in the manner of RFC 3390:
88 * X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
90 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
91 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
92 * For consistency with other parts of the code, X_init is scaled by 2^6.
94 static inline u64 rfc3390_initial_rate(struct sock *sk)
96 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
97 const __u32 w_init = min_t(__u32, 4 * hctx->ccid3hctx_s,
98 max_t(__u32, 2 * hctx->ccid3hctx_s, 4380));
100 return scaled_div(w_init << 6, hctx->ccid3hctx_rtt);
104 * Recalculate t_ipi and delta (should be called whenever X changes)
106 static inline void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
108 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
109 hctx->ccid3hctx_t_ipi = scaled_div32(((u64)hctx->ccid3hctx_s) << 6,
112 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
113 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
114 TFRC_OPSYS_HALF_TIME_GRAN);
116 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n",
117 hctx->ccid3hctx_t_ipi, hctx->ccid3hctx_delta,
118 hctx->ccid3hctx_s, (unsigned)(hctx->ccid3hctx_x >> 6));
122 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
124 u32 delta = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count);
126 return delta / hctx->ccid3hctx_rtt;
130 * ccid3_hc_tx_update_x - Update allowed sending rate X
131 * @stamp: most recent time if available - can be left NULL.
132 * This function tracks draft rfc3448bis, check there for latest details.
134 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
135 * fine-grained resolution of sending rates. This requires scaling by 2^6
136 * throughout the code. Only X_calc is unscaled (in bytes/second).
139 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
142 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
143 __u64 min_rate = 2 * hctx->ccid3hctx_x_recv;
144 const __u64 old_x = hctx->ccid3hctx_x;
145 ktime_t now = stamp? *stamp : ktime_get_real();
148 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
149 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
150 * a sender is idle if it has not sent anything over a 2-RTT-period.
151 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
153 if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) {
154 min_rate = rfc3390_initial_rate(sk);
155 min_rate = max(min_rate, 2 * hctx->ccid3hctx_x_recv);
158 if (hctx->ccid3hctx_p > 0) {
160 hctx->ccid3hctx_x = min(((__u64)hctx->ccid3hctx_x_calc) << 6,
162 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
163 (((__u64)hctx->ccid3hctx_s) << 6) /
166 } else if (ktime_us_delta(now, hctx->ccid3hctx_t_ld)
167 - (s64)hctx->ccid3hctx_rtt >= 0) {
170 max(min(2 * hctx->ccid3hctx_x, min_rate),
171 scaled_div(((__u64)hctx->ccid3hctx_s) << 6,
172 hctx->ccid3hctx_rtt));
173 hctx->ccid3hctx_t_ld = now;
176 if (hctx->ccid3hctx_x != old_x) {
177 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
178 "X_recv=%u\n", (unsigned)(old_x >> 6),
179 (unsigned)(hctx->ccid3hctx_x >> 6),
180 hctx->ccid3hctx_x_calc,
181 (unsigned)(hctx->ccid3hctx_x_recv >> 6));
183 ccid3_update_send_interval(hctx);
188 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
189 * @len: DCCP packet payload size in bytes
191 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
193 const u16 old_s = hctx->ccid3hctx_s;
195 hctx->ccid3hctx_s = tfrc_ewma(hctx->ccid3hctx_s, len, 9);
197 if (hctx->ccid3hctx_s != old_s)
198 ccid3_update_send_interval(hctx);
202 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
203 * The algorithm is not applicable if RTT < 4 microseconds.
205 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
210 if (unlikely(hctx->ccid3hctx_rtt < 4)) /* avoid divide-by-zero */
213 quarter_rtts = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count);
214 quarter_rtts /= hctx->ccid3hctx_rtt / 4;
216 if (quarter_rtts > 0) {
217 hctx->ccid3hctx_t_last_win_count = now;
218 hctx->ccid3hctx_last_win_count += min_t(u32, quarter_rtts, 5);
219 hctx->ccid3hctx_last_win_count &= 0xF; /* mod 16 */
223 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
225 struct sock *sk = (struct sock *)data;
226 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
227 unsigned long t_nfb = USEC_PER_SEC / 5;
230 if (sock_owned_by_user(sk)) {
231 /* Try again later. */
232 /* XXX: set some sensible MIB */
236 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
237 ccid3_tx_state_name(hctx->ccid3hctx_state));
239 switch (hctx->ccid3hctx_state) {
240 case TFRC_SSTATE_NO_FBACK:
241 /* RFC 3448, 4.4: Halve send rate directly */
242 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x / 2,
243 (((__u64)hctx->ccid3hctx_s) << 6) /
246 ccid3_pr_debug("%s(%p, state=%s), updated tx rate to %u "
247 "bytes/s\n", dccp_role(sk), sk,
248 ccid3_tx_state_name(hctx->ccid3hctx_state),
249 (unsigned)(hctx->ccid3hctx_x >> 6));
250 /* The value of R is still undefined and so we can not recompute
251 * the timeout value. Keep initial value as per [RFC 4342, 5]. */
252 t_nfb = TFRC_INITIAL_TIMEOUT;
253 ccid3_update_send_interval(hctx);
255 case TFRC_SSTATE_FBACK:
257 * Modify the cached value of X_recv [RFC 3448, 4.4]
259 * If (p == 0 || X_calc > 2 * X_recv)
260 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
262 * X_recv = X_calc / 4;
264 * Note that X_recv is scaled by 2^6 while X_calc is not
266 BUG_ON(hctx->ccid3hctx_p && !hctx->ccid3hctx_x_calc);
268 if (hctx->ccid3hctx_p == 0 ||
269 (hctx->ccid3hctx_x_calc > (hctx->ccid3hctx_x_recv >> 5))) {
271 hctx->ccid3hctx_x_recv =
272 max(hctx->ccid3hctx_x_recv / 2,
273 (((__u64)hctx->ccid3hctx_s) << 6) /
276 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc;
277 hctx->ccid3hctx_x_recv <<= 4;
279 /* Now recalculate X [RFC 3448, 4.3, step (4)] */
280 ccid3_hc_tx_update_x(sk, NULL);
282 * Schedule no feedback timer to expire in
283 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
284 * See comments in packet_recv() regarding the value of t_RTO.
286 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
288 case TFRC_SSTATE_NO_SENT:
289 DCCP_BUG("%s(%p) - Illegal state NO_SENT", dccp_role(sk), sk);
291 case TFRC_SSTATE_TERM:
296 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
297 jiffies + usecs_to_jiffies(t_nfb));
305 * > 0: delay (in msecs) that should pass before actually sending
306 * = 0: can send immediately
307 * < 0: error condition; do not send packet
309 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
311 struct dccp_sock *dp = dccp_sk(sk);
312 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
313 ktime_t now = ktime_get_real();
317 * This function is called only for Data and DataAck packets. Sending
318 * zero-sized Data(Ack)s is theoretically possible, but for congestion
319 * control this case is pathological - ignore it.
321 if (unlikely(skb->len == 0))
324 switch (hctx->ccid3hctx_state) {
325 case TFRC_SSTATE_NO_SENT:
326 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
328 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
329 hctx->ccid3hctx_last_win_count = 0;
330 hctx->ccid3hctx_t_last_win_count = now;
332 /* Set t_0 for initial packet */
333 hctx->ccid3hctx_t_nom = now;
335 hctx->ccid3hctx_s = skb->len;
338 * Use initial RTT sample when available: recommended by erratum
339 * to RFC 4342. This implements the initialisation procedure of
340 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
342 if (dp->dccps_syn_rtt) {
343 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
344 hctx->ccid3hctx_rtt = dp->dccps_syn_rtt;
345 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
346 hctx->ccid3hctx_t_ld = now;
348 /* Sender does not have RTT sample: X_pps = 1 pkt/sec */
349 hctx->ccid3hctx_x = hctx->ccid3hctx_s;
350 hctx->ccid3hctx_x <<= 6;
352 ccid3_update_send_interval(hctx);
354 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
356 case TFRC_SSTATE_NO_FBACK:
357 case TFRC_SSTATE_FBACK:
358 delay = ktime_us_delta(hctx->ccid3hctx_t_nom, now);
359 ccid3_pr_debug("delay=%ld\n", (long)delay);
361 * Scheduling of packet transmissions [RFC 3448, 4.6]
363 * if (t_now > t_nom - delta)
364 * // send the packet now
366 * // send the packet in (t_nom - t_now) milliseconds.
368 if (delay - (s64)hctx->ccid3hctx_delta >= 1000)
369 return (u32)delay / 1000L;
371 ccid3_hc_tx_update_win_count(hctx, now);
373 case TFRC_SSTATE_TERM:
374 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
378 /* prepare to send now (add options etc.) */
379 dp->dccps_hc_tx_insert_options = 1;
380 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
382 /* set the nominal send time for the next following packet */
383 hctx->ccid3hctx_t_nom = ktime_add_us(hctx->ccid3hctx_t_nom,
384 hctx->ccid3hctx_t_ipi);
388 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
391 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
392 struct dccp_tx_hist_entry *packet;
394 ccid3_hc_tx_update_s(hctx, len);
396 packet = dccp_tx_hist_entry_new(ccid3_tx_hist, GFP_ATOMIC);
397 if (unlikely(packet == NULL)) {
398 DCCP_CRIT("packet history - out of memory!");
401 dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, packet);
403 packet->dccphtx_tstamp = ktime_get_real();
404 packet->dccphtx_seqno = dccp_sk(sk)->dccps_gss;
405 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
406 packet->dccphtx_sent = 1;
409 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
411 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
412 struct ccid3_options_received *opt_recv;
413 struct dccp_tx_hist_entry *packet;
418 /* we are only interested in ACKs */
419 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
420 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
423 opt_recv = &hctx->ccid3hctx_options_received;
425 switch (hctx->ccid3hctx_state) {
426 case TFRC_SSTATE_NO_FBACK:
427 case TFRC_SSTATE_FBACK:
428 /* get packet from history to look up t_recvdata */
429 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
430 DCCP_SKB_CB(skb)->dccpd_ack_seq);
431 if (unlikely(packet == NULL)) {
432 DCCP_WARN("%s(%p), seqno %llu(%s) doesn't exist "
433 "in history!\n", dccp_role(sk), sk,
434 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
435 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
439 /* Update receive rate in units of 64 * bytes/second */
440 hctx->ccid3hctx_x_recv = opt_recv->ccid3or_receive_rate;
441 hctx->ccid3hctx_x_recv <<= 6;
443 /* Update loss event rate */
444 pinv = opt_recv->ccid3or_loss_event_rate;
445 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
446 hctx->ccid3hctx_p = 0;
447 else /* can not exceed 100% */
448 hctx->ccid3hctx_p = 1000000 / pinv;
450 now = ktime_get_real();
452 * Calculate new RTT sample and update moving average
454 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, packet->dccphtx_tstamp));
455 hctx->ccid3hctx_rtt = tfrc_ewma(hctx->ccid3hctx_rtt, r_sample, 9);
457 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
459 * Larger Initial Windows [RFC 4342, sec. 5]
461 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
462 hctx->ccid3hctx_t_ld = now;
464 ccid3_update_send_interval(hctx);
466 ccid3_pr_debug("%s(%p), s=%u, MSS=%u, "
467 "R_sample=%uus, X=%u\n", dccp_role(sk),
468 sk, hctx->ccid3hctx_s,
469 dccp_sk(sk)->dccps_mss_cache, r_sample,
470 (unsigned)(hctx->ccid3hctx_x >> 6));
472 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
475 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
476 if (hctx->ccid3hctx_p > 0)
477 hctx->ccid3hctx_x_calc =
478 tfrc_calc_x(hctx->ccid3hctx_s,
481 ccid3_hc_tx_update_x(sk, &now);
483 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
484 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
486 sk, hctx->ccid3hctx_rtt, r_sample,
487 hctx->ccid3hctx_s, hctx->ccid3hctx_p,
488 hctx->ccid3hctx_x_calc,
489 (unsigned)(hctx->ccid3hctx_x_recv >> 6),
490 (unsigned)(hctx->ccid3hctx_x >> 6));
493 /* unschedule no feedback timer */
494 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
496 /* remove all packets older than the one acked from history */
497 dccp_tx_hist_purge_older(ccid3_tx_hist,
498 &hctx->ccid3hctx_hist, packet);
500 * As we have calculated new ipi, delta, t_nom it is possible
501 * that we now can send a packet, so wake up dccp_wait_for_ccid
503 sk->sk_write_space(sk);
506 * Update timeout interval for the nofeedback timer.
507 * We use a configuration option to increase the lower bound.
508 * This can help avoid triggering the nofeedback timer too
509 * often ('spinning') on LANs with small RTTs.
511 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
512 CONFIG_IP_DCCP_CCID3_RTO *
513 (USEC_PER_SEC/1000));
515 * Schedule no feedback timer to expire in
516 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
518 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
520 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
521 "expire in %lu jiffies (%luus)\n",
523 sk, usecs_to_jiffies(t_nfb), t_nfb);
525 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
526 jiffies + usecs_to_jiffies(t_nfb));
528 case TFRC_SSTATE_NO_SENT: /* fall through */
529 case TFRC_SSTATE_TERM: /* ignore feedback when closing */
534 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
535 unsigned char len, u16 idx,
536 unsigned char *value)
539 const struct dccp_sock *dp = dccp_sk(sk);
540 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
541 struct ccid3_options_received *opt_recv;
544 opt_recv = &hctx->ccid3hctx_options_received;
546 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
547 opt_recv->ccid3or_seqno = dp->dccps_gsr;
548 opt_recv->ccid3or_loss_event_rate = ~0;
549 opt_recv->ccid3or_loss_intervals_idx = 0;
550 opt_recv->ccid3or_loss_intervals_len = 0;
551 opt_recv->ccid3or_receive_rate = 0;
555 case TFRC_OPT_LOSS_EVENT_RATE:
556 if (unlikely(len != 4)) {
557 DCCP_WARN("%s(%p), invalid len %d "
558 "for TFRC_OPT_LOSS_EVENT_RATE\n",
559 dccp_role(sk), sk, len);
562 opt_val = get_unaligned((__be32 *)value);
563 opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
564 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
566 opt_recv->ccid3or_loss_event_rate);
569 case TFRC_OPT_LOSS_INTERVALS:
570 opt_recv->ccid3or_loss_intervals_idx = idx;
571 opt_recv->ccid3or_loss_intervals_len = len;
572 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
574 opt_recv->ccid3or_loss_intervals_idx,
575 opt_recv->ccid3or_loss_intervals_len);
577 case TFRC_OPT_RECEIVE_RATE:
578 if (unlikely(len != 4)) {
579 DCCP_WARN("%s(%p), invalid len %d "
580 "for TFRC_OPT_RECEIVE_RATE\n",
581 dccp_role(sk), sk, len);
584 opt_val = get_unaligned((__be32 *)value);
585 opt_recv->ccid3or_receive_rate = ntohl(opt_val);
586 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
588 opt_recv->ccid3or_receive_rate);
596 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
598 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
600 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
601 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
602 setup_timer(&hctx->ccid3hctx_no_feedback_timer,
603 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
608 static void ccid3_hc_tx_exit(struct sock *sk)
610 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
612 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
613 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
615 /* Empty packet history */
616 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
619 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
621 struct ccid3_hc_tx_sock *hctx;
623 /* Listen socks doesn't have a private CCID block */
624 if (sk->sk_state == DCCP_LISTEN)
627 hctx = ccid3_hc_tx_sk(sk);
628 info->tcpi_rto = hctx->ccid3hctx_t_rto;
629 info->tcpi_rtt = hctx->ccid3hctx_rtt;
632 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
633 u32 __user *optval, int __user *optlen)
635 const struct ccid3_hc_tx_sock *hctx;
638 /* Listen socks doesn't have a private CCID block */
639 if (sk->sk_state == DCCP_LISTEN)
642 hctx = ccid3_hc_tx_sk(sk);
644 case DCCP_SOCKOPT_CCID_TX_INFO:
645 if (len < sizeof(hctx->ccid3hctx_tfrc))
647 len = sizeof(hctx->ccid3hctx_tfrc);
648 val = &hctx->ccid3hctx_tfrc;
654 if (put_user(len, optlen) || copy_to_user(optval, val, len))
661 * Receiver Half-Connection Routines
663 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
664 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
666 static char *ccid3_rx_state_names[] = {
667 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
668 [TFRC_RSTATE_DATA] = "DATA",
669 [TFRC_RSTATE_TERM] = "TERM",
672 return ccid3_rx_state_names[state];
676 static void ccid3_hc_rx_set_state(struct sock *sk,
677 enum ccid3_hc_rx_states state)
679 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
680 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
682 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
683 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
684 ccid3_rx_state_name(state));
685 WARN_ON(state == oldstate);
686 hcrx->ccid3hcrx_state = state;
689 static inline void ccid3_hc_rx_update_s(struct ccid3_hc_rx_sock *hcrx, int len)
691 if (likely(len > 0)) /* don't update on empty packets (e.g. ACKs) */
692 hcrx->ccid3hcrx_s = tfrc_ewma(hcrx->ccid3hcrx_s, len, 9);
695 static void ccid3_hc_rx_send_feedback(struct sock *sk)
697 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
698 struct dccp_sock *dp = dccp_sk(sk);
699 struct dccp_rx_hist_entry *packet;
703 ccid3_pr_debug("%s(%p) - entry \n", dccp_role(sk), sk);
705 now = ktime_get_real();
707 switch (hcrx->ccid3hcrx_state) {
708 case TFRC_RSTATE_NO_DATA:
709 hcrx->ccid3hcrx_x_recv = 0;
711 case TFRC_RSTATE_DATA:
712 delta = ktime_us_delta(now,
713 hcrx->ccid3hcrx_tstamp_last_feedback);
714 DCCP_BUG_ON(delta < 0);
715 hcrx->ccid3hcrx_x_recv =
716 scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
718 case TFRC_RSTATE_TERM:
719 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
723 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
724 if (unlikely(packet == NULL)) {
725 DCCP_WARN("%s(%p), no data packet in history!\n",
730 hcrx->ccid3hcrx_tstamp_last_feedback = now;
731 hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
732 hcrx->ccid3hcrx_bytes_recv = 0;
734 if (hcrx->ccid3hcrx_p == 0)
735 hcrx->ccid3hcrx_pinv = ~0U; /* see RFC 4342, 8.5 */
736 else if (hcrx->ccid3hcrx_p > 1000000) {
737 DCCP_WARN("p (%u) > 100%%\n", hcrx->ccid3hcrx_p);
738 hcrx->ccid3hcrx_pinv = 1; /* use 100% in this case */
740 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
742 dp->dccps_hc_rx_insert_options = 1;
746 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
748 const struct ccid3_hc_rx_sock *hcrx;
751 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
754 hcrx = ccid3_hc_rx_sk(sk);
755 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
757 if (dccp_packet_without_ack(skb))
760 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
761 pinv = htonl(hcrx->ccid3hcrx_pinv);
763 if (dccp_insert_option_timestamp(sk, skb) ||
764 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
765 &pinv, sizeof(pinv)) ||
766 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
767 &x_recv, sizeof(x_recv)))
773 static int ccid3_hc_rx_detect_loss(struct sock *sk,
774 struct dccp_rx_hist_entry *packet)
776 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
777 struct dccp_rx_hist_entry *rx_hist =
778 dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
779 u64 seqno = packet->dccphrx_seqno;
785 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
788 follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
789 hcrx->ccid3hcrx_seqno_nonloss = seqno;
790 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
795 while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
796 > TFRC_RECV_NUM_LATE_LOSS) {
798 dccp_li_update_li(sk,
799 &hcrx->ccid3hcrx_li_hist,
800 &hcrx->ccid3hcrx_hist,
801 hcrx->ccid3hcrx_tstamp_last_feedback,
803 hcrx->ccid3hcrx_bytes_recv,
804 hcrx->ccid3hcrx_x_recv,
805 hcrx->ccid3hcrx_seqno_nonloss,
806 hcrx->ccid3hcrx_ccval_nonloss);
807 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
808 dccp_inc_seqno(&tmp_seqno);
809 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
810 dccp_inc_seqno(&tmp_seqno);
811 while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
812 tmp_seqno, &ccval)) {
813 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
814 hcrx->ccid3hcrx_ccval_nonloss = ccval;
815 dccp_inc_seqno(&tmp_seqno);
819 /* FIXME - this code could be simplified with above while */
820 /* but works at moment */
821 if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
822 hcrx->ccid3hcrx_seqno_nonloss = seqno;
823 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
827 dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
828 &hcrx->ccid3hcrx_li_hist, packet,
829 hcrx->ccid3hcrx_seqno_nonloss);
833 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
835 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
836 const struct dccp_options_received *opt_recv;
837 struct dccp_rx_hist_entry *packet;
838 u32 p_prev, r_sample, rtt_prev;
839 int loss, payload_size;
842 opt_recv = &dccp_sk(sk)->dccps_options_received;
844 switch (DCCP_SKB_CB(skb)->dccpd_type) {
846 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
848 case DCCP_PKT_DATAACK:
849 if (opt_recv->dccpor_timestamp_echo == 0)
851 r_sample = dccp_timestamp() - opt_recv->dccpor_timestamp_echo;
852 rtt_prev = hcrx->ccid3hcrx_rtt;
853 r_sample = dccp_sample_rtt(sk, 10 * r_sample);
855 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
856 hcrx->ccid3hcrx_rtt = r_sample;
858 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
861 if (rtt_prev != hcrx->ccid3hcrx_rtt)
862 ccid3_pr_debug("%s(%p), New RTT=%uus, elapsed time=%u\n",
863 dccp_role(sk), sk, hcrx->ccid3hcrx_rtt,
864 opt_recv->dccpor_elapsed_time);
868 default: /* We're not interested in other packet types, move along */
872 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, opt_recv->dccpor_ndp,
874 if (unlikely(packet == NULL)) {
875 DCCP_WARN("%s(%p), Not enough mem to add rx packet "
876 "to history, consider it lost!\n", dccp_role(sk), sk);
880 loss = ccid3_hc_rx_detect_loss(sk, packet);
882 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
885 payload_size = skb->len - dccp_hdr(skb)->dccph_doff * 4;
886 ccid3_hc_rx_update_s(hcrx, payload_size);
888 switch (hcrx->ccid3hcrx_state) {
889 case TFRC_RSTATE_NO_DATA:
890 ccid3_pr_debug("%s(%p, state=%s), skb=%p, sending initial "
891 "feedback\n", dccp_role(sk), sk,
892 dccp_state_name(sk->sk_state), skb);
893 ccid3_hc_rx_send_feedback(sk);
894 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
896 case TFRC_RSTATE_DATA:
897 hcrx->ccid3hcrx_bytes_recv += payload_size;
901 now = ktime_get_real();
902 if ((ktime_us_delta(now, hcrx->ccid3hcrx_tstamp_last_ack) -
903 (s64)hcrx->ccid3hcrx_rtt) >= 0) {
904 hcrx->ccid3hcrx_tstamp_last_ack = now;
905 ccid3_hc_rx_send_feedback(sk);
908 case TFRC_RSTATE_TERM:
909 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
913 /* Dealing with packet loss */
914 ccid3_pr_debug("%s(%p, state=%s), data loss! Reacting...\n",
915 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
917 p_prev = hcrx->ccid3hcrx_p;
919 /* Calculate loss event rate */
920 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
921 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
923 /* Scaling up by 1000000 as fixed decimal */
925 hcrx->ccid3hcrx_p = 1000000 / i_mean;
927 DCCP_BUG("empty loss history");
929 if (hcrx->ccid3hcrx_p > p_prev) {
930 ccid3_hc_rx_send_feedback(sk);
935 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
937 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
939 ccid3_pr_debug("entry\n");
941 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
942 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
943 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
944 hcrx->ccid3hcrx_tstamp_last_feedback =
945 hcrx->ccid3hcrx_tstamp_last_ack = ktime_get_real();
949 static void ccid3_hc_rx_exit(struct sock *sk)
951 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
953 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
955 /* Empty packet history */
956 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
958 /* Empty loss interval history */
959 dccp_li_hist_purge(&hcrx->ccid3hcrx_li_hist);
962 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
964 const struct ccid3_hc_rx_sock *hcrx;
966 /* Listen socks doesn't have a private CCID block */
967 if (sk->sk_state == DCCP_LISTEN)
970 hcrx = ccid3_hc_rx_sk(sk);
971 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
972 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
973 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
976 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
977 u32 __user *optval, int __user *optlen)
979 const struct ccid3_hc_rx_sock *hcrx;
982 /* Listen socks doesn't have a private CCID block */
983 if (sk->sk_state == DCCP_LISTEN)
986 hcrx = ccid3_hc_rx_sk(sk);
988 case DCCP_SOCKOPT_CCID_RX_INFO:
989 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
991 len = sizeof(hcrx->ccid3hcrx_tfrc);
992 val = &hcrx->ccid3hcrx_tfrc;
998 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1004 static struct ccid_operations ccid3 = {
1005 .ccid_id = DCCPC_CCID3,
1006 .ccid_name = "ccid3",
1007 .ccid_owner = THIS_MODULE,
1008 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
1009 .ccid_hc_tx_init = ccid3_hc_tx_init,
1010 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1011 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1012 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1013 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1014 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1015 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
1016 .ccid_hc_rx_init = ccid3_hc_rx_init,
1017 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1018 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1019 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1020 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1021 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1022 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1023 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1026 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
1027 module_param(ccid3_debug, bool, 0444);
1028 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1031 static __init int ccid3_module_init(void)
1035 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1036 if (ccid3_rx_hist == NULL)
1039 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1040 if (ccid3_tx_hist == NULL)
1043 rc = ccid_register(&ccid3);
1050 dccp_tx_hist_delete(ccid3_tx_hist);
1051 ccid3_tx_hist = NULL;
1053 dccp_rx_hist_delete(ccid3_rx_hist);
1054 ccid3_rx_hist = NULL;
1057 module_init(ccid3_module_init);
1059 static __exit void ccid3_module_exit(void)
1061 ccid_unregister(&ccid3);
1063 if (ccid3_tx_hist != NULL) {
1064 dccp_tx_hist_delete(ccid3_tx_hist);
1065 ccid3_tx_hist = NULL;
1067 if (ccid3_rx_hist != NULL) {
1068 dccp_rx_hist_delete(ccid3_rx_hist);
1069 ccid3_rx_hist = NULL;
1072 module_exit(ccid3_module_exit);
1074 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
1075 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1076 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1077 MODULE_LICENSE("GPL");
1078 MODULE_ALIAS("net-dccp-ccid-3");