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 according to RFC 3390:
87 * w_init = min(4 * MSS, max(2 * MSS, 4380 bytes))
88 * X_init = w_init / RTT
89 * For consistency with other parts of the code, X_init is scaled by 2^6.
91 static inline u64 rfc3390_initial_rate(struct sock *sk)
93 const struct dccp_sock *dp = dccp_sk(sk);
94 const __u32 w_init = min(4 * dp->dccps_mss_cache,
95 max(2 * dp->dccps_mss_cache, 4380U));
97 return scaled_div(w_init << 6, ccid3_hc_tx_sk(sk)->ccid3hctx_rtt);
101 * Recalculate t_ipi and delta (should be called whenever X changes)
103 static inline void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
105 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
106 hctx->ccid3hctx_t_ipi = scaled_div32(((u64)hctx->ccid3hctx_s) << 6,
109 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
110 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
111 TFRC_OPSYS_HALF_TIME_GRAN);
113 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n",
114 hctx->ccid3hctx_t_ipi, hctx->ccid3hctx_delta,
115 hctx->ccid3hctx_s, (unsigned)(hctx->ccid3hctx_x >> 6));
120 * ccid3_hc_tx_update_x - Update allowed sending rate X
121 * @stamp: most recent time if available - can be left NULL.
122 * This function tracks draft rfc3448bis, check there for latest details.
124 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
125 * fine-grained resolution of sending rates. This requires scaling by 2^6
126 * throughout the code. Only X_calc is unscaled (in bytes/second).
129 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
132 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
133 __u64 min_rate = 2 * hctx->ccid3hctx_x_recv;
134 const __u64 old_x = hctx->ccid3hctx_x;
135 ktime_t now = stamp? *stamp : ktime_get_real();
138 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
139 * when idling [RFC 4342, 5.1]. See also draft-ietf-dccp-rfc3448bis.
140 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
142 if (unlikely(hctx->ccid3hctx_idle)) {
143 min_rate = rfc3390_initial_rate(sk);
144 min_rate = max(min_rate, 2 * hctx->ccid3hctx_x_recv);
147 if (hctx->ccid3hctx_p > 0) {
149 hctx->ccid3hctx_x = min(((__u64)hctx->ccid3hctx_x_calc) << 6,
151 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
152 (((__u64)hctx->ccid3hctx_s) << 6) /
155 } else if (ktime_us_delta(now, hctx->ccid3hctx_t_ld)
156 - (s64)hctx->ccid3hctx_rtt >= 0) {
159 max(min(2 * hctx->ccid3hctx_x, min_rate),
160 scaled_div(((__u64)hctx->ccid3hctx_s) << 6,
161 hctx->ccid3hctx_rtt));
162 hctx->ccid3hctx_t_ld = now;
165 if (hctx->ccid3hctx_x != old_x) {
166 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
167 "X_recv=%u\n", (unsigned)(old_x >> 6),
168 (unsigned)(hctx->ccid3hctx_x >> 6),
169 hctx->ccid3hctx_x_calc,
170 (unsigned)(hctx->ccid3hctx_x_recv >> 6));
172 ccid3_update_send_interval(hctx);
177 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
178 * @len: DCCP packet payload size in bytes
180 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
182 const u16 old_s = hctx->ccid3hctx_s;
184 hctx->ccid3hctx_s = old_s == 0 ? len : (9 * old_s + len) / 10;
186 if (hctx->ccid3hctx_s != old_s)
187 ccid3_update_send_interval(hctx);
191 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
192 * The algorithm is not applicable if RTT < 4 microseconds.
194 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
199 if (unlikely(hctx->ccid3hctx_rtt < 4)) /* avoid divide-by-zero */
202 quarter_rtts = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count);
203 quarter_rtts /= hctx->ccid3hctx_rtt / 4;
205 if (quarter_rtts > 0) {
206 hctx->ccid3hctx_t_last_win_count = now;
207 hctx->ccid3hctx_last_win_count += min_t(u32, quarter_rtts, 5);
208 hctx->ccid3hctx_last_win_count &= 0xF; /* mod 16 */
212 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
214 struct sock *sk = (struct sock *)data;
215 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
216 unsigned long t_nfb = USEC_PER_SEC / 5;
219 if (sock_owned_by_user(sk)) {
220 /* Try again later. */
221 /* XXX: set some sensible MIB */
225 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
226 ccid3_tx_state_name(hctx->ccid3hctx_state));
228 hctx->ccid3hctx_idle = 1;
230 switch (hctx->ccid3hctx_state) {
231 case TFRC_SSTATE_NO_FBACK:
232 /* RFC 3448, 4.4: Halve send rate directly */
233 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x / 2,
234 (((__u64)hctx->ccid3hctx_s) << 6) /
237 ccid3_pr_debug("%s(%p, state=%s), updated tx rate to %u "
238 "bytes/s\n", dccp_role(sk), sk,
239 ccid3_tx_state_name(hctx->ccid3hctx_state),
240 (unsigned)(hctx->ccid3hctx_x >> 6));
241 /* The value of R is still undefined and so we can not recompute
242 * the timeout value. Keep initial value as per [RFC 4342, 5]. */
243 t_nfb = TFRC_INITIAL_TIMEOUT;
244 ccid3_update_send_interval(hctx);
246 case TFRC_SSTATE_FBACK:
248 * Modify the cached value of X_recv [RFC 3448, 4.4]
250 * If (p == 0 || X_calc > 2 * X_recv)
251 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
253 * X_recv = X_calc / 4;
255 * Note that X_recv is scaled by 2^6 while X_calc is not
257 BUG_ON(hctx->ccid3hctx_p && !hctx->ccid3hctx_x_calc);
259 if (hctx->ccid3hctx_p == 0 ||
260 (hctx->ccid3hctx_x_calc > (hctx->ccid3hctx_x_recv >> 5))) {
262 hctx->ccid3hctx_x_recv =
263 max(hctx->ccid3hctx_x_recv / 2,
264 (((__u64)hctx->ccid3hctx_s) << 6) /
267 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc;
268 hctx->ccid3hctx_x_recv <<= 4;
270 /* Now recalculate X [RFC 3448, 4.3, step (4)] */
271 ccid3_hc_tx_update_x(sk, NULL);
273 * Schedule no feedback timer to expire in
274 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
275 * See comments in packet_recv() regarding the value of t_RTO.
277 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
279 case TFRC_SSTATE_NO_SENT:
280 DCCP_BUG("%s(%p) - Illegal state NO_SENT", dccp_role(sk), sk);
282 case TFRC_SSTATE_TERM:
287 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
288 jiffies + usecs_to_jiffies(t_nfb));
296 * > 0: delay (in msecs) that should pass before actually sending
297 * = 0: can send immediately
298 * < 0: error condition; do not send packet
300 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
302 struct dccp_sock *dp = dccp_sk(sk);
303 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
304 ktime_t now = ktime_get_real();
308 * This function is called only for Data and DataAck packets. Sending
309 * zero-sized Data(Ack)s is theoretically possible, but for congestion
310 * control this case is pathological - ignore it.
312 if (unlikely(skb->len == 0))
315 switch (hctx->ccid3hctx_state) {
316 case TFRC_SSTATE_NO_SENT:
317 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
319 usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
320 hctx->ccid3hctx_last_win_count = 0;
321 hctx->ccid3hctx_t_last_win_count = now;
323 /* Set t_0 for initial packet */
324 hctx->ccid3hctx_t_nom = now;
326 hctx->ccid3hctx_s = skb->len;
329 * Use initial RTT sample when available: recommended by erratum
330 * to RFC 4342. This implements the initialisation procedure of
331 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
333 if (dp->dccps_syn_rtt) {
334 ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
335 hctx->ccid3hctx_rtt = dp->dccps_syn_rtt;
336 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
337 hctx->ccid3hctx_t_ld = now;
339 /* Sender does not have RTT sample: X = MSS/second */
340 hctx->ccid3hctx_x = dp->dccps_mss_cache;
341 hctx->ccid3hctx_x <<= 6;
343 ccid3_update_send_interval(hctx);
345 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
347 case TFRC_SSTATE_NO_FBACK:
348 case TFRC_SSTATE_FBACK:
349 delay = ktime_us_delta(hctx->ccid3hctx_t_nom, now);
350 ccid3_pr_debug("delay=%ld\n", (long)delay);
352 * Scheduling of packet transmissions [RFC 3448, 4.6]
354 * if (t_now > t_nom - delta)
355 * // send the packet now
357 * // send the packet in (t_nom - t_now) milliseconds.
359 if (delay - (s64)hctx->ccid3hctx_delta >= 1000)
360 return (u32)delay / 1000L;
362 ccid3_hc_tx_update_win_count(hctx, now);
364 case TFRC_SSTATE_TERM:
365 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
369 /* prepare to send now (add options etc.) */
370 dp->dccps_hc_tx_insert_options = 1;
371 DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
372 hctx->ccid3hctx_idle = 0;
374 /* set the nominal send time for the next following packet */
375 hctx->ccid3hctx_t_nom = ktime_add_us(hctx->ccid3hctx_t_nom,
376 hctx->ccid3hctx_t_ipi);
380 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
383 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
384 struct dccp_tx_hist_entry *packet;
386 ccid3_hc_tx_update_s(hctx, len);
388 packet = dccp_tx_hist_entry_new(ccid3_tx_hist, GFP_ATOMIC);
389 if (unlikely(packet == NULL)) {
390 DCCP_CRIT("packet history - out of memory!");
393 dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, packet);
395 packet->dccphtx_tstamp = ktime_get_real();
396 packet->dccphtx_seqno = dccp_sk(sk)->dccps_gss;
397 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
398 packet->dccphtx_sent = 1;
401 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
403 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
404 struct ccid3_options_received *opt_recv;
405 struct dccp_tx_hist_entry *packet;
410 /* we are only interested in ACKs */
411 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
412 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
415 opt_recv = &hctx->ccid3hctx_options_received;
417 switch (hctx->ccid3hctx_state) {
418 case TFRC_SSTATE_NO_FBACK:
419 case TFRC_SSTATE_FBACK:
420 /* get packet from history to look up t_recvdata */
421 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
422 DCCP_SKB_CB(skb)->dccpd_ack_seq);
423 if (unlikely(packet == NULL)) {
424 DCCP_WARN("%s(%p), seqno %llu(%s) doesn't exist "
425 "in history!\n", dccp_role(sk), sk,
426 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
427 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
431 /* Update receive rate in units of 64 * bytes/second */
432 hctx->ccid3hctx_x_recv = opt_recv->ccid3or_receive_rate;
433 hctx->ccid3hctx_x_recv <<= 6;
435 /* Update loss event rate */
436 pinv = opt_recv->ccid3or_loss_event_rate;
437 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
438 hctx->ccid3hctx_p = 0;
439 else /* can not exceed 100% */
440 hctx->ccid3hctx_p = 1000000 / pinv;
442 now = ktime_get_real();
444 * Calculate new round trip sample as per [RFC 3448, 4.3] by
445 * R_sample = (now - t_recvdata) - t_elapsed
447 r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, packet->dccphtx_tstamp));
450 * Update RTT estimate by
451 * If (No feedback recv)
454 * R = q * R + (1 - q) * R_sample;
456 * q is a constant, RFC 3448 recomments 0.9
458 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
460 * Larger Initial Windows [RFC 4342, sec. 5]
462 hctx->ccid3hctx_rtt = r_sample;
463 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
464 hctx->ccid3hctx_t_ld = now;
466 ccid3_update_send_interval(hctx);
468 ccid3_pr_debug("%s(%p), s=%u, MSS=%u, "
469 "R_sample=%uus, X=%u\n", dccp_role(sk),
470 sk, hctx->ccid3hctx_s,
471 dccp_sk(sk)->dccps_mss_cache, r_sample,
472 (unsigned)(hctx->ccid3hctx_x >> 6));
474 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
476 hctx->ccid3hctx_rtt = (9 * hctx->ccid3hctx_rtt +
479 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
480 if (hctx->ccid3hctx_p > 0)
481 hctx->ccid3hctx_x_calc =
482 tfrc_calc_x(hctx->ccid3hctx_s,
485 ccid3_hc_tx_update_x(sk, &now);
487 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
488 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
490 sk, hctx->ccid3hctx_rtt, r_sample,
491 hctx->ccid3hctx_s, hctx->ccid3hctx_p,
492 hctx->ccid3hctx_x_calc,
493 (unsigned)(hctx->ccid3hctx_x_recv >> 6),
494 (unsigned)(hctx->ccid3hctx_x >> 6));
497 /* unschedule no feedback timer */
498 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
500 /* remove all packets older than the one acked from history */
501 dccp_tx_hist_purge_older(ccid3_tx_hist,
502 &hctx->ccid3hctx_hist, packet);
504 * As we have calculated new ipi, delta, t_nom it is possible
505 * that we now can send a packet, so wake up dccp_wait_for_ccid
507 sk->sk_write_space(sk);
510 * Update timeout interval for the nofeedback timer.
511 * We use a configuration option to increase the lower bound.
512 * This can help avoid triggering the nofeedback timer too
513 * often ('spinning') on LANs with small RTTs.
515 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
516 CONFIG_IP_DCCP_CCID3_RTO *
517 (USEC_PER_SEC/1000));
519 * Schedule no feedback timer to expire in
520 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
522 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
524 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
525 "expire in %lu jiffies (%luus)\n",
527 sk, usecs_to_jiffies(t_nfb), t_nfb);
529 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
530 jiffies + usecs_to_jiffies(t_nfb));
533 hctx->ccid3hctx_idle = 1;
535 case TFRC_SSTATE_NO_SENT: /* fall through */
536 case TFRC_SSTATE_TERM: /* ignore feedback when closing */
541 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
542 unsigned char len, u16 idx,
543 unsigned char *value)
546 const struct dccp_sock *dp = dccp_sk(sk);
547 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
548 struct ccid3_options_received *opt_recv;
551 opt_recv = &hctx->ccid3hctx_options_received;
553 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
554 opt_recv->ccid3or_seqno = dp->dccps_gsr;
555 opt_recv->ccid3or_loss_event_rate = ~0;
556 opt_recv->ccid3or_loss_intervals_idx = 0;
557 opt_recv->ccid3or_loss_intervals_len = 0;
558 opt_recv->ccid3or_receive_rate = 0;
562 case TFRC_OPT_LOSS_EVENT_RATE:
563 if (unlikely(len != 4)) {
564 DCCP_WARN("%s(%p), invalid len %d "
565 "for TFRC_OPT_LOSS_EVENT_RATE\n",
566 dccp_role(sk), sk, len);
569 opt_val = get_unaligned((__be32 *)value);
570 opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
571 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
573 opt_recv->ccid3or_loss_event_rate);
576 case TFRC_OPT_LOSS_INTERVALS:
577 opt_recv->ccid3or_loss_intervals_idx = idx;
578 opt_recv->ccid3or_loss_intervals_len = len;
579 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
581 opt_recv->ccid3or_loss_intervals_idx,
582 opt_recv->ccid3or_loss_intervals_len);
584 case TFRC_OPT_RECEIVE_RATE:
585 if (unlikely(len != 4)) {
586 DCCP_WARN("%s(%p), invalid len %d "
587 "for TFRC_OPT_RECEIVE_RATE\n",
588 dccp_role(sk), sk, len);
591 opt_val = get_unaligned((__be32 *)value);
592 opt_recv->ccid3or_receive_rate = ntohl(opt_val);
593 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
595 opt_recv->ccid3or_receive_rate);
603 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
605 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
607 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
608 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
609 setup_timer(&hctx->ccid3hctx_no_feedback_timer,
610 ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
615 static void ccid3_hc_tx_exit(struct sock *sk)
617 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
619 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
620 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
622 /* Empty packet history */
623 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
626 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
628 struct ccid3_hc_tx_sock *hctx;
630 /* Listen socks doesn't have a private CCID block */
631 if (sk->sk_state == DCCP_LISTEN)
634 hctx = ccid3_hc_tx_sk(sk);
635 info->tcpi_rto = hctx->ccid3hctx_t_rto;
636 info->tcpi_rtt = hctx->ccid3hctx_rtt;
639 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
640 u32 __user *optval, int __user *optlen)
642 const struct ccid3_hc_tx_sock *hctx;
645 /* Listen socks doesn't have a private CCID block */
646 if (sk->sk_state == DCCP_LISTEN)
649 hctx = ccid3_hc_tx_sk(sk);
651 case DCCP_SOCKOPT_CCID_TX_INFO:
652 if (len < sizeof(hctx->ccid3hctx_tfrc))
654 len = sizeof(hctx->ccid3hctx_tfrc);
655 val = &hctx->ccid3hctx_tfrc;
661 if (put_user(len, optlen) || copy_to_user(optval, val, len))
668 * Receiver Half-Connection Routines
670 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
671 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
673 static char *ccid3_rx_state_names[] = {
674 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
675 [TFRC_RSTATE_DATA] = "DATA",
676 [TFRC_RSTATE_TERM] = "TERM",
679 return ccid3_rx_state_names[state];
683 static void ccid3_hc_rx_set_state(struct sock *sk,
684 enum ccid3_hc_rx_states state)
686 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
687 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
689 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
690 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
691 ccid3_rx_state_name(state));
692 WARN_ON(state == oldstate);
693 hcrx->ccid3hcrx_state = state;
696 static inline void ccid3_hc_rx_update_s(struct ccid3_hc_rx_sock *hcrx, int len)
698 if (unlikely(len == 0)) /* don't update on empty packets (e.g. ACKs) */
699 ccid3_pr_debug("Packet payload length is 0 - not updating\n");
701 hcrx->ccid3hcrx_s = hcrx->ccid3hcrx_s == 0 ? len :
702 (9 * hcrx->ccid3hcrx_s + len) / 10;
705 static void ccid3_hc_rx_send_feedback(struct sock *sk)
707 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
708 struct dccp_sock *dp = dccp_sk(sk);
709 struct dccp_rx_hist_entry *packet;
713 ccid3_pr_debug("%s(%p) - entry \n", dccp_role(sk), sk);
715 now = ktime_get_real();
717 switch (hcrx->ccid3hcrx_state) {
718 case TFRC_RSTATE_NO_DATA:
719 hcrx->ccid3hcrx_x_recv = 0;
721 case TFRC_RSTATE_DATA:
722 delta = ktime_us_delta(now,
723 hcrx->ccid3hcrx_tstamp_last_feedback);
724 DCCP_BUG_ON(delta < 0);
725 hcrx->ccid3hcrx_x_recv =
726 scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
728 case TFRC_RSTATE_TERM:
729 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
733 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
734 if (unlikely(packet == NULL)) {
735 DCCP_WARN("%s(%p), no data packet in history!\n",
740 hcrx->ccid3hcrx_tstamp_last_feedback = now;
741 hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
742 hcrx->ccid3hcrx_bytes_recv = 0;
744 /* Elapsed time information [RFC 4340, 13.2] in units of 10 * usecs */
745 delta = ktime_us_delta(now, packet->dccphrx_tstamp);
746 DCCP_BUG_ON(delta < 0);
747 hcrx->ccid3hcrx_elapsed_time = delta / 10;
749 if (hcrx->ccid3hcrx_p == 0)
750 hcrx->ccid3hcrx_pinv = ~0U; /* see RFC 4342, 8.5 */
751 else if (hcrx->ccid3hcrx_p > 1000000) {
752 DCCP_WARN("p (%u) > 100%%\n", hcrx->ccid3hcrx_p);
753 hcrx->ccid3hcrx_pinv = 1; /* use 100% in this case */
755 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
757 dp->dccps_hc_rx_insert_options = 1;
761 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
763 const struct ccid3_hc_rx_sock *hcrx;
766 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
769 hcrx = ccid3_hc_rx_sk(sk);
770 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
772 if (dccp_packet_without_ack(skb))
775 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
776 pinv = htonl(hcrx->ccid3hcrx_pinv);
778 if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
779 dccp_insert_option_elapsed_time(sk, skb,
780 hcrx->ccid3hcrx_elapsed_time)) ||
781 dccp_insert_option_timestamp(sk, skb) ||
782 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
783 &pinv, sizeof(pinv)) ||
784 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
785 &x_recv, sizeof(x_recv)))
791 static int ccid3_hc_rx_detect_loss(struct sock *sk,
792 struct dccp_rx_hist_entry *packet)
794 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
795 struct dccp_rx_hist_entry *rx_hist =
796 dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
797 u64 seqno = packet->dccphrx_seqno;
803 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
806 follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
807 hcrx->ccid3hcrx_seqno_nonloss = seqno;
808 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
813 while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
814 > TFRC_RECV_NUM_LATE_LOSS) {
816 dccp_li_update_li(sk,
817 &hcrx->ccid3hcrx_li_hist,
818 &hcrx->ccid3hcrx_hist,
819 hcrx->ccid3hcrx_tstamp_last_feedback,
821 hcrx->ccid3hcrx_bytes_recv,
822 hcrx->ccid3hcrx_x_recv,
823 hcrx->ccid3hcrx_seqno_nonloss,
824 hcrx->ccid3hcrx_ccval_nonloss);
825 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
826 dccp_inc_seqno(&tmp_seqno);
827 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
828 dccp_inc_seqno(&tmp_seqno);
829 while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
830 tmp_seqno, &ccval)) {
831 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
832 hcrx->ccid3hcrx_ccval_nonloss = ccval;
833 dccp_inc_seqno(&tmp_seqno);
837 /* FIXME - this code could be simplified with above while */
838 /* but works at moment */
839 if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
840 hcrx->ccid3hcrx_seqno_nonloss = seqno;
841 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
845 dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
846 &hcrx->ccid3hcrx_li_hist, packet,
847 hcrx->ccid3hcrx_seqno_nonloss);
851 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
853 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
854 const struct dccp_options_received *opt_recv;
855 struct dccp_rx_hist_entry *packet;
856 u32 p_prev, r_sample, rtt_prev;
857 int loss, payload_size;
860 opt_recv = &dccp_sk(sk)->dccps_options_received;
862 switch (DCCP_SKB_CB(skb)->dccpd_type) {
864 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
866 case DCCP_PKT_DATAACK:
867 if (opt_recv->dccpor_timestamp_echo == 0)
869 r_sample = dccp_timestamp() - opt_recv->dccpor_timestamp_echo;
870 rtt_prev = hcrx->ccid3hcrx_rtt;
871 r_sample = dccp_sample_rtt(sk, 10 * r_sample);
873 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
874 hcrx->ccid3hcrx_rtt = r_sample;
876 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
879 if (rtt_prev != hcrx->ccid3hcrx_rtt)
880 ccid3_pr_debug("%s(%p), New RTT=%uus, elapsed time=%u\n",
881 dccp_role(sk), sk, hcrx->ccid3hcrx_rtt,
882 opt_recv->dccpor_elapsed_time);
886 default: /* We're not interested in other packet types, move along */
890 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, opt_recv->dccpor_ndp,
892 if (unlikely(packet == NULL)) {
893 DCCP_WARN("%s(%p), Not enough mem to add rx packet "
894 "to history, consider it lost!\n", dccp_role(sk), sk);
898 loss = ccid3_hc_rx_detect_loss(sk, packet);
900 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
903 payload_size = skb->len - dccp_hdr(skb)->dccph_doff * 4;
904 ccid3_hc_rx_update_s(hcrx, payload_size);
906 switch (hcrx->ccid3hcrx_state) {
907 case TFRC_RSTATE_NO_DATA:
908 ccid3_pr_debug("%s(%p, state=%s), skb=%p, sending initial "
909 "feedback\n", dccp_role(sk), sk,
910 dccp_state_name(sk->sk_state), skb);
911 ccid3_hc_rx_send_feedback(sk);
912 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
914 case TFRC_RSTATE_DATA:
915 hcrx->ccid3hcrx_bytes_recv += payload_size;
919 now = ktime_get_real();
920 if ((ktime_us_delta(now, hcrx->ccid3hcrx_tstamp_last_ack) -
921 (s64)hcrx->ccid3hcrx_rtt) >= 0) {
922 hcrx->ccid3hcrx_tstamp_last_ack = now;
923 ccid3_hc_rx_send_feedback(sk);
926 case TFRC_RSTATE_TERM:
927 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
931 /* Dealing with packet loss */
932 ccid3_pr_debug("%s(%p, state=%s), data loss! Reacting...\n",
933 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
935 p_prev = hcrx->ccid3hcrx_p;
937 /* Calculate loss event rate */
938 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
939 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
941 /* Scaling up by 1000000 as fixed decimal */
943 hcrx->ccid3hcrx_p = 1000000 / i_mean;
945 DCCP_BUG("empty loss history");
947 if (hcrx->ccid3hcrx_p > p_prev) {
948 ccid3_hc_rx_send_feedback(sk);
953 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
955 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
957 ccid3_pr_debug("entry\n");
959 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
960 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
961 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
962 hcrx->ccid3hcrx_tstamp_last_feedback =
963 hcrx->ccid3hcrx_tstamp_last_ack = ktime_get_real();
967 static void ccid3_hc_rx_exit(struct sock *sk)
969 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
971 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
973 /* Empty packet history */
974 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
976 /* Empty loss interval history */
977 dccp_li_hist_purge(&hcrx->ccid3hcrx_li_hist);
980 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
982 const struct ccid3_hc_rx_sock *hcrx;
984 /* Listen socks doesn't have a private CCID block */
985 if (sk->sk_state == DCCP_LISTEN)
988 hcrx = ccid3_hc_rx_sk(sk);
989 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
990 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
991 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
994 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
995 u32 __user *optval, int __user *optlen)
997 const struct ccid3_hc_rx_sock *hcrx;
1000 /* Listen socks doesn't have a private CCID block */
1001 if (sk->sk_state == DCCP_LISTEN)
1004 hcrx = ccid3_hc_rx_sk(sk);
1006 case DCCP_SOCKOPT_CCID_RX_INFO:
1007 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
1009 len = sizeof(hcrx->ccid3hcrx_tfrc);
1010 val = &hcrx->ccid3hcrx_tfrc;
1013 return -ENOPROTOOPT;
1016 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1022 static struct ccid_operations ccid3 = {
1023 .ccid_id = DCCPC_CCID3,
1024 .ccid_name = "ccid3",
1025 .ccid_owner = THIS_MODULE,
1026 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
1027 .ccid_hc_tx_init = ccid3_hc_tx_init,
1028 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1029 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1030 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1031 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1032 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1033 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
1034 .ccid_hc_rx_init = ccid3_hc_rx_init,
1035 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1036 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1037 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1038 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1039 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1040 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1041 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1044 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
1045 module_param(ccid3_debug, bool, 0444);
1046 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1049 static __init int ccid3_module_init(void)
1053 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1054 if (ccid3_rx_hist == NULL)
1057 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1058 if (ccid3_tx_hist == NULL)
1061 rc = ccid_register(&ccid3);
1068 dccp_tx_hist_delete(ccid3_tx_hist);
1069 ccid3_tx_hist = NULL;
1071 dccp_rx_hist_delete(ccid3_rx_hist);
1072 ccid3_rx_hist = NULL;
1075 module_init(ccid3_module_init);
1077 static __exit void ccid3_module_exit(void)
1079 ccid_unregister(&ccid3);
1081 if (ccid3_tx_hist != NULL) {
1082 dccp_tx_hist_delete(ccid3_tx_hist);
1083 ccid3_tx_hist = NULL;
1085 if (ccid3_rx_hist != NULL) {
1086 dccp_rx_hist_delete(ccid3_rx_hist);
1087 ccid3_rx_hist = NULL;
1090 module_exit(ccid3_module_exit);
1092 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
1093 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1094 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1095 MODULE_LICENSE("GPL");
1096 MODULE_ALIAS("net-dccp-ccid-3");