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 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
44 static int ccid3_debug;
45 #define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
47 #define ccid3_pr_debug(format, a...)
50 static struct dccp_tx_hist *ccid3_tx_hist;
51 static struct dccp_rx_hist *ccid3_rx_hist;
54 * Transmitter Half-Connection Routines
56 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
57 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
59 static char *ccid3_state_names[] = {
60 [TFRC_SSTATE_NO_SENT] = "NO_SENT",
61 [TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
62 [TFRC_SSTATE_FBACK] = "FBACK",
63 [TFRC_SSTATE_TERM] = "TERM",
66 return ccid3_state_names[state];
70 static void ccid3_hc_tx_set_state(struct sock *sk,
71 enum ccid3_hc_tx_states state)
73 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
74 enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
76 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
77 dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
78 ccid3_tx_state_name(state));
79 WARN_ON(state == oldstate);
80 hctx->ccid3hctx_state = state;
84 * Compute the initial sending rate X_init according to RFC 3390:
85 * w_init = min(4 * MSS, max(2 * MSS, 4380 bytes))
86 * X_init = w_init / RTT
87 * For consistency with other parts of the code, X_init is scaled by 2^6.
89 static inline u64 rfc3390_initial_rate(struct sock *sk)
91 const struct dccp_sock *dp = dccp_sk(sk);
92 const __u32 w_init = min(4 * dp->dccps_mss_cache,
93 max(2 * dp->dccps_mss_cache, 4380U));
95 return scaled_div(w_init << 6, ccid3_hc_tx_sk(sk)->ccid3hctx_rtt);
99 * Recalculate t_ipi and delta (should be called whenever X changes)
101 static inline void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
103 /* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
104 hctx->ccid3hctx_t_ipi = scaled_div32(((u64)hctx->ccid3hctx_s) << 6,
107 /* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
108 hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
109 TFRC_OPSYS_HALF_TIME_GRAN);
111 ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n",
112 hctx->ccid3hctx_t_ipi, hctx->ccid3hctx_delta,
113 hctx->ccid3hctx_s, (unsigned)(hctx->ccid3hctx_x >> 6));
118 * ccid3_hc_tx_update_x - Update allowed sending rate X
119 * @stamp: most recent time if available - can be left NULL.
120 * This function tracks draft rfc3448bis, check there for latest details.
122 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
123 * fine-grained resolution of sending rates. This requires scaling by 2^6
124 * throughout the code. Only X_calc is unscaled (in bytes/second).
127 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
130 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
131 __u64 min_rate = 2 * hctx->ccid3hctx_x_recv;
132 const __u64 old_x = hctx->ccid3hctx_x;
133 ktime_t now = stamp? *stamp : ktime_get_real();
136 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
137 * when idling [RFC 4342, 5.1]. See also draft-ietf-dccp-rfc3448bis.
138 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
140 if (unlikely(hctx->ccid3hctx_idle)) {
141 min_rate = rfc3390_initial_rate(sk);
142 min_rate = max(min_rate, 2 * hctx->ccid3hctx_x_recv);
145 if (hctx->ccid3hctx_p > 0) {
147 hctx->ccid3hctx_x = min(((__u64)hctx->ccid3hctx_x_calc) << 6,
149 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
150 (((__u64)hctx->ccid3hctx_s) << 6) /
153 } else if (ktime_us_delta(now, hctx->ccid3hctx_t_ld)
154 - (s64)hctx->ccid3hctx_rtt >= 0) {
157 max(min(2 * hctx->ccid3hctx_x, min_rate),
158 scaled_div(((__u64)hctx->ccid3hctx_s) << 6,
159 hctx->ccid3hctx_rtt));
160 hctx->ccid3hctx_t_ld = now;
163 if (hctx->ccid3hctx_x != old_x) {
164 ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
165 "X_recv=%u\n", (unsigned)(old_x >> 6),
166 (unsigned)(hctx->ccid3hctx_x >> 6),
167 hctx->ccid3hctx_x_calc,
168 (unsigned)(hctx->ccid3hctx_x_recv >> 6));
170 ccid3_update_send_interval(hctx);
175 * Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
176 * @len: DCCP packet payload size in bytes
178 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
180 const u16 old_s = hctx->ccid3hctx_s;
182 hctx->ccid3hctx_s = old_s == 0 ? len : (9 * old_s + len) / 10;
184 if (hctx->ccid3hctx_s != old_s)
185 ccid3_update_send_interval(hctx);
189 * Update Window Counter using the algorithm from [RFC 4342, 8.1].
190 * The algorithm is not applicable if RTT < 4 microseconds.
192 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
197 if (unlikely(hctx->ccid3hctx_rtt < 4)) /* avoid divide-by-zero */
200 quarter_rtts = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count);
201 quarter_rtts /= hctx->ccid3hctx_rtt / 4;
203 if (quarter_rtts > 0) {
204 hctx->ccid3hctx_t_last_win_count = now;
205 hctx->ccid3hctx_last_win_count += min_t(u32, quarter_rtts, 5);
206 hctx->ccid3hctx_last_win_count &= 0xF; /* mod 16 */
210 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
212 struct sock *sk = (struct sock *)data;
213 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
214 unsigned long t_nfb = USEC_PER_SEC / 5;
217 if (sock_owned_by_user(sk)) {
218 /* Try again later. */
219 /* XXX: set some sensible MIB */
223 ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
224 ccid3_tx_state_name(hctx->ccid3hctx_state));
226 hctx->ccid3hctx_idle = 1;
228 switch (hctx->ccid3hctx_state) {
229 case TFRC_SSTATE_NO_FBACK:
230 /* RFC 3448, 4.4: Halve send rate directly */
231 hctx->ccid3hctx_x = max(hctx->ccid3hctx_x / 2,
232 (((__u64)hctx->ccid3hctx_s) << 6) /
235 ccid3_pr_debug("%s(%p, state=%s), updated tx rate to %u "
236 "bytes/s\n", dccp_role(sk), sk,
237 ccid3_tx_state_name(hctx->ccid3hctx_state),
238 (unsigned)(hctx->ccid3hctx_x >> 6));
239 /* The value of R is still undefined and so we can not recompute
240 * the timout value. Keep initial value as per [RFC 4342, 5]. */
241 t_nfb = TFRC_INITIAL_TIMEOUT;
242 ccid3_update_send_interval(hctx);
244 case TFRC_SSTATE_FBACK:
246 * Modify the cached value of X_recv [RFC 3448, 4.4]
248 * If (p == 0 || X_calc > 2 * X_recv)
249 * X_recv = max(X_recv / 2, s / (2 * t_mbi));
251 * X_recv = X_calc / 4;
253 * Note that X_recv is scaled by 2^6 while X_calc is not
255 BUG_ON(hctx->ccid3hctx_p && !hctx->ccid3hctx_x_calc);
257 if (hctx->ccid3hctx_p == 0 ||
258 (hctx->ccid3hctx_x_calc > (hctx->ccid3hctx_x_recv >> 5))) {
260 hctx->ccid3hctx_x_recv =
261 max(hctx->ccid3hctx_x_recv / 2,
262 (((__u64)hctx->ccid3hctx_s) << 6) /
265 hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc;
266 hctx->ccid3hctx_x_recv <<= 4;
268 /* Now recalculate X [RFC 3448, 4.3, step (4)] */
269 ccid3_hc_tx_update_x(sk, NULL);
271 * Schedule no feedback timer to expire in
272 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
273 * See comments in packet_recv() regarding the value of t_RTO.
275 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
277 case TFRC_SSTATE_NO_SENT:
278 DCCP_BUG("%s(%p) - Illegal state NO_SENT", dccp_role(sk), sk);
280 case TFRC_SSTATE_TERM:
285 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
286 jiffies + usecs_to_jiffies(t_nfb));
294 * > 0: delay (in msecs) that should pass before actually sending
295 * = 0: can send immediately
296 * < 0: error condition; do not send packet
298 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
300 struct dccp_sock *dp = dccp_sk(sk);
301 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
302 ktime_t now = ktime_get_real();
305 BUG_ON(hctx == NULL);
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 BUG_ON(hctx == NULL);
388 ccid3_hc_tx_update_s(hctx, len);
390 packet = dccp_tx_hist_entry_new(ccid3_tx_hist, GFP_ATOMIC);
391 if (unlikely(packet == NULL)) {
392 DCCP_CRIT("packet history - out of memory!");
395 dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, packet);
397 packet->dccphtx_tstamp = ktime_get_real();
398 packet->dccphtx_seqno = dccp_sk(sk)->dccps_gss;
399 packet->dccphtx_rtt = hctx->ccid3hctx_rtt;
400 packet->dccphtx_sent = 1;
403 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
405 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
406 struct ccid3_options_received *opt_recv;
407 struct dccp_tx_hist_entry *packet;
412 BUG_ON(hctx == NULL);
414 /* we are only interested in ACKs */
415 if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
416 DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
419 opt_recv = &hctx->ccid3hctx_options_received;
421 switch (hctx->ccid3hctx_state) {
422 case TFRC_SSTATE_NO_FBACK:
423 case TFRC_SSTATE_FBACK:
424 /* get packet from history to look up t_recvdata */
425 packet = dccp_tx_hist_find_entry(&hctx->ccid3hctx_hist,
426 DCCP_SKB_CB(skb)->dccpd_ack_seq);
427 if (unlikely(packet == NULL)) {
428 DCCP_WARN("%s(%p), seqno %llu(%s) doesn't exist "
429 "in history!\n", dccp_role(sk), sk,
430 (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq,
431 dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type));
435 /* Update receive rate in units of 64 * bytes/second */
436 hctx->ccid3hctx_x_recv = opt_recv->ccid3or_receive_rate;
437 hctx->ccid3hctx_x_recv <<= 6;
439 /* Update loss event rate */
440 pinv = opt_recv->ccid3or_loss_event_rate;
441 if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
442 hctx->ccid3hctx_p = 0;
443 else /* can not exceed 100% */
444 hctx->ccid3hctx_p = 1000000 / pinv;
446 now = ktime_get_real();
448 * Calculate new round trip sample as per [RFC 3448, 4.3] by
449 * R_sample = (now - t_recvdata) - t_elapsed
451 r_sample = dccp_sample_rtt(sk, now, &packet->dccphtx_tstamp);
454 * Update RTT estimate by
455 * If (No feedback recv)
458 * R = q * R + (1 - q) * R_sample;
460 * q is a constant, RFC 3448 recomments 0.9
462 if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
464 * Larger Initial Windows [RFC 4342, sec. 5]
466 hctx->ccid3hctx_rtt = r_sample;
467 hctx->ccid3hctx_x = rfc3390_initial_rate(sk);
468 hctx->ccid3hctx_t_ld = now;
470 ccid3_update_send_interval(hctx);
472 ccid3_pr_debug("%s(%p), s=%u, MSS=%u, "
473 "R_sample=%uus, X=%u\n", dccp_role(sk),
474 sk, hctx->ccid3hctx_s,
475 dccp_sk(sk)->dccps_mss_cache, r_sample,
476 (unsigned)(hctx->ccid3hctx_x >> 6));
478 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
480 hctx->ccid3hctx_rtt = (9 * hctx->ccid3hctx_rtt +
483 /* Update sending rate (step 4 of [RFC 3448, 4.3]) */
484 if (hctx->ccid3hctx_p > 0)
485 hctx->ccid3hctx_x_calc =
486 tfrc_calc_x(hctx->ccid3hctx_s,
489 ccid3_hc_tx_update_x(sk, &now);
491 ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
492 "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
494 sk, hctx->ccid3hctx_rtt, r_sample,
495 hctx->ccid3hctx_s, hctx->ccid3hctx_p,
496 hctx->ccid3hctx_x_calc,
497 (unsigned)(hctx->ccid3hctx_x_recv >> 6),
498 (unsigned)(hctx->ccid3hctx_x >> 6));
501 /* unschedule no feedback timer */
502 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
504 /* remove all packets older than the one acked from history */
505 dccp_tx_hist_purge_older(ccid3_tx_hist,
506 &hctx->ccid3hctx_hist, packet);
508 * As we have calculated new ipi, delta, t_nom it is possible
509 * that we now can send a packet, so wake up dccp_wait_for_ccid
511 sk->sk_write_space(sk);
514 * Update timeout interval for the nofeedback timer.
515 * We use a configuration option to increase the lower bound.
516 * This can help avoid triggering the nofeedback timer too
517 * often ('spinning') on LANs with small RTTs.
519 hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
520 CONFIG_IP_DCCP_CCID3_RTO *
521 (USEC_PER_SEC/1000));
523 * Schedule no feedback timer to expire in
524 * max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
526 t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
528 ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
529 "expire in %lu jiffies (%luus)\n",
531 sk, usecs_to_jiffies(t_nfb), t_nfb);
533 sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
534 jiffies + usecs_to_jiffies(t_nfb));
537 hctx->ccid3hctx_idle = 1;
539 case TFRC_SSTATE_NO_SENT: /* fall through */
540 case TFRC_SSTATE_TERM: /* ignore feedback when closing */
545 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
546 unsigned char len, u16 idx,
547 unsigned char *value)
550 const struct dccp_sock *dp = dccp_sk(sk);
551 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
552 struct ccid3_options_received *opt_recv;
554 BUG_ON(hctx == NULL);
556 opt_recv = &hctx->ccid3hctx_options_received;
558 if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
559 opt_recv->ccid3or_seqno = dp->dccps_gsr;
560 opt_recv->ccid3or_loss_event_rate = ~0;
561 opt_recv->ccid3or_loss_intervals_idx = 0;
562 opt_recv->ccid3or_loss_intervals_len = 0;
563 opt_recv->ccid3or_receive_rate = 0;
567 case TFRC_OPT_LOSS_EVENT_RATE:
568 if (unlikely(len != 4)) {
569 DCCP_WARN("%s(%p), invalid len %d "
570 "for TFRC_OPT_LOSS_EVENT_RATE\n",
571 dccp_role(sk), sk, len);
574 opt_recv->ccid3or_loss_event_rate =
575 ntohl(*(__be32 *)value);
576 ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
578 opt_recv->ccid3or_loss_event_rate);
581 case TFRC_OPT_LOSS_INTERVALS:
582 opt_recv->ccid3or_loss_intervals_idx = idx;
583 opt_recv->ccid3or_loss_intervals_len = len;
584 ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
586 opt_recv->ccid3or_loss_intervals_idx,
587 opt_recv->ccid3or_loss_intervals_len);
589 case TFRC_OPT_RECEIVE_RATE:
590 if (unlikely(len != 4)) {
591 DCCP_WARN("%s(%p), invalid len %d "
592 "for TFRC_OPT_RECEIVE_RATE\n",
593 dccp_role(sk), sk, len);
596 opt_recv->ccid3or_receive_rate =
597 ntohl(*(__be32 *)value);
598 ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
600 opt_recv->ccid3or_receive_rate);
608 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
610 struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
612 hctx->ccid3hctx_s = 0;
613 hctx->ccid3hctx_rtt = 0;
614 hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
615 INIT_LIST_HEAD(&hctx->ccid3hctx_hist);
617 hctx->ccid3hctx_no_feedback_timer.function =
618 ccid3_hc_tx_no_feedback_timer;
619 hctx->ccid3hctx_no_feedback_timer.data = (unsigned long)sk;
620 init_timer(&hctx->ccid3hctx_no_feedback_timer);
625 static void ccid3_hc_tx_exit(struct sock *sk)
627 struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
629 BUG_ON(hctx == NULL);
631 ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
632 sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
634 /* Empty packet history */
635 dccp_tx_hist_purge(ccid3_tx_hist, &hctx->ccid3hctx_hist);
638 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
640 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
642 /* Listen socks doesn't have a private CCID block */
643 if (sk->sk_state == DCCP_LISTEN)
646 BUG_ON(hctx == NULL);
648 info->tcpi_rto = hctx->ccid3hctx_t_rto;
649 info->tcpi_rtt = hctx->ccid3hctx_rtt;
652 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
653 u32 __user *optval, int __user *optlen)
655 const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
658 /* Listen socks doesn't have a private CCID block */
659 if (sk->sk_state == DCCP_LISTEN)
663 case DCCP_SOCKOPT_CCID_TX_INFO:
664 if (len < sizeof(hctx->ccid3hctx_tfrc))
666 len = sizeof(hctx->ccid3hctx_tfrc);
667 val = &hctx->ccid3hctx_tfrc;
673 if (put_user(len, optlen) || copy_to_user(optval, val, len))
680 * Receiver Half-Connection Routines
682 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
683 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
685 static char *ccid3_rx_state_names[] = {
686 [TFRC_RSTATE_NO_DATA] = "NO_DATA",
687 [TFRC_RSTATE_DATA] = "DATA",
688 [TFRC_RSTATE_TERM] = "TERM",
691 return ccid3_rx_state_names[state];
695 static void ccid3_hc_rx_set_state(struct sock *sk,
696 enum ccid3_hc_rx_states state)
698 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
699 enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
701 ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
702 dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
703 ccid3_rx_state_name(state));
704 WARN_ON(state == oldstate);
705 hcrx->ccid3hcrx_state = state;
708 static inline void ccid3_hc_rx_update_s(struct ccid3_hc_rx_sock *hcrx, int len)
710 if (unlikely(len == 0)) /* don't update on empty packets (e.g. ACKs) */
711 ccid3_pr_debug("Packet payload length is 0 - not updating\n");
713 hcrx->ccid3hcrx_s = hcrx->ccid3hcrx_s == 0 ? len :
714 (9 * hcrx->ccid3hcrx_s + len) / 10;
717 static void ccid3_hc_rx_send_feedback(struct sock *sk)
719 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
720 struct dccp_sock *dp = dccp_sk(sk);
721 struct dccp_rx_hist_entry *packet;
725 ccid3_pr_debug("%s(%p) - entry \n", dccp_role(sk), sk);
727 now = ktime_get_real();
729 switch (hcrx->ccid3hcrx_state) {
730 case TFRC_RSTATE_NO_DATA:
731 hcrx->ccid3hcrx_x_recv = 0;
733 case TFRC_RSTATE_DATA:
734 delta = ktime_us_delta(now,
735 hcrx->ccid3hcrx_tstamp_last_feedback);
736 DCCP_BUG_ON(delta < 0);
737 hcrx->ccid3hcrx_x_recv =
738 scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
740 case TFRC_RSTATE_TERM:
741 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
745 packet = dccp_rx_hist_find_data_packet(&hcrx->ccid3hcrx_hist);
746 if (unlikely(packet == NULL)) {
747 DCCP_WARN("%s(%p), no data packet in history!\n",
752 hcrx->ccid3hcrx_tstamp_last_feedback = now;
753 hcrx->ccid3hcrx_ccval_last_counter = packet->dccphrx_ccval;
754 hcrx->ccid3hcrx_bytes_recv = 0;
756 /* Elapsed time information [RFC 4340, 13.2] in units of 10 * usecs */
757 delta = ktime_us_delta(now, packet->dccphrx_tstamp);
758 DCCP_BUG_ON(delta < 0);
759 hcrx->ccid3hcrx_elapsed_time = delta / 10;
761 if (hcrx->ccid3hcrx_p == 0)
762 hcrx->ccid3hcrx_pinv = ~0U; /* see RFC 4342, 8.5 */
763 else if (hcrx->ccid3hcrx_p > 1000000) {
764 DCCP_WARN("p (%u) > 100%%\n", hcrx->ccid3hcrx_p);
765 hcrx->ccid3hcrx_pinv = 1; /* use 100% in this case */
767 hcrx->ccid3hcrx_pinv = 1000000 / hcrx->ccid3hcrx_p;
769 dp->dccps_hc_rx_insert_options = 1;
773 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
775 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
778 BUG_ON(hcrx == NULL);
780 if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
783 DCCP_SKB_CB(skb)->dccpd_ccval = hcrx->ccid3hcrx_ccval_last_counter;
785 if (dccp_packet_without_ack(skb))
788 x_recv = htonl(hcrx->ccid3hcrx_x_recv);
789 pinv = htonl(hcrx->ccid3hcrx_pinv);
791 if ((hcrx->ccid3hcrx_elapsed_time != 0 &&
792 dccp_insert_option_elapsed_time(sk, skb,
793 hcrx->ccid3hcrx_elapsed_time)) ||
794 dccp_insert_option_timestamp(sk, skb) ||
795 dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
796 &pinv, sizeof(pinv)) ||
797 dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
798 &x_recv, sizeof(x_recv)))
804 static int ccid3_hc_rx_detect_loss(struct sock *sk,
805 struct dccp_rx_hist_entry *packet)
807 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
808 struct dccp_rx_hist_entry *rx_hist =
809 dccp_rx_hist_head(&hcrx->ccid3hcrx_hist);
810 u64 seqno = packet->dccphrx_seqno;
816 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
819 follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
820 hcrx->ccid3hcrx_seqno_nonloss = seqno;
821 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
826 while (dccp_delta_seqno(hcrx->ccid3hcrx_seqno_nonloss, seqno)
827 > TFRC_RECV_NUM_LATE_LOSS) {
829 dccp_li_update_li(sk,
830 &hcrx->ccid3hcrx_li_hist,
831 &hcrx->ccid3hcrx_hist,
832 hcrx->ccid3hcrx_tstamp_last_feedback,
834 hcrx->ccid3hcrx_bytes_recv,
835 hcrx->ccid3hcrx_x_recv,
836 hcrx->ccid3hcrx_seqno_nonloss,
837 hcrx->ccid3hcrx_ccval_nonloss);
838 tmp_seqno = hcrx->ccid3hcrx_seqno_nonloss;
839 dccp_inc_seqno(&tmp_seqno);
840 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
841 dccp_inc_seqno(&tmp_seqno);
842 while (dccp_rx_hist_find_entry(&hcrx->ccid3hcrx_hist,
843 tmp_seqno, &ccval)) {
844 hcrx->ccid3hcrx_seqno_nonloss = tmp_seqno;
845 hcrx->ccid3hcrx_ccval_nonloss = ccval;
846 dccp_inc_seqno(&tmp_seqno);
850 /* FIXME - this code could be simplified with above while */
851 /* but works at moment */
852 if (follows48(packet->dccphrx_seqno, hcrx->ccid3hcrx_seqno_nonloss)) {
853 hcrx->ccid3hcrx_seqno_nonloss = seqno;
854 hcrx->ccid3hcrx_ccval_nonloss = packet->dccphrx_ccval;
858 dccp_rx_hist_add_packet(ccid3_rx_hist, &hcrx->ccid3hcrx_hist,
859 &hcrx->ccid3hcrx_li_hist, packet,
860 hcrx->ccid3hcrx_seqno_nonloss);
864 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
866 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
867 const struct dccp_options_received *opt_recv;
868 struct dccp_rx_hist_entry *packet;
869 u32 p_prev, r_sample, rtt_prev;
870 int loss, payload_size;
873 BUG_ON(hcrx == NULL);
875 opt_recv = &dccp_sk(sk)->dccps_options_received;
877 switch (DCCP_SKB_CB(skb)->dccpd_type) {
879 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
881 case DCCP_PKT_DATAACK:
882 if (opt_recv->dccpor_timestamp_echo == 0)
884 rtt_prev = hcrx->ccid3hcrx_rtt;
885 now = ktime_get_real();
886 r_sample = dccp_sample_rtt(sk, now, NULL);
888 if (hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)
889 hcrx->ccid3hcrx_rtt = r_sample;
891 hcrx->ccid3hcrx_rtt = (hcrx->ccid3hcrx_rtt * 9) / 10 +
894 if (rtt_prev != hcrx->ccid3hcrx_rtt)
895 ccid3_pr_debug("%s(%p), New RTT=%uus, elapsed time=%u\n",
896 dccp_role(sk), sk, hcrx->ccid3hcrx_rtt,
897 opt_recv->dccpor_elapsed_time);
901 default: /* We're not interested in other packet types, move along */
905 packet = dccp_rx_hist_entry_new(ccid3_rx_hist, opt_recv->dccpor_ndp,
907 if (unlikely(packet == NULL)) {
908 DCCP_WARN("%s(%p), Not enough mem to add rx packet "
909 "to history, consider it lost!\n", dccp_role(sk), sk);
913 loss = ccid3_hc_rx_detect_loss(sk, packet);
915 if (DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK)
918 payload_size = skb->len - dccp_hdr(skb)->dccph_doff * 4;
919 ccid3_hc_rx_update_s(hcrx, payload_size);
921 switch (hcrx->ccid3hcrx_state) {
922 case TFRC_RSTATE_NO_DATA:
923 ccid3_pr_debug("%s(%p, state=%s), skb=%p, sending initial "
924 "feedback\n", dccp_role(sk), sk,
925 dccp_state_name(sk->sk_state), skb);
926 ccid3_hc_rx_send_feedback(sk);
927 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
929 case TFRC_RSTATE_DATA:
930 hcrx->ccid3hcrx_bytes_recv += payload_size;
934 now = ktime_get_real();
935 if ((ktime_us_delta(now, hcrx->ccid3hcrx_tstamp_last_ack) -
936 (s64)hcrx->ccid3hcrx_rtt) >= 0) {
937 hcrx->ccid3hcrx_tstamp_last_ack = now;
938 ccid3_hc_rx_send_feedback(sk);
941 case TFRC_RSTATE_TERM:
942 DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
946 /* Dealing with packet loss */
947 ccid3_pr_debug("%s(%p, state=%s), data loss! Reacting...\n",
948 dccp_role(sk), sk, dccp_state_name(sk->sk_state));
950 p_prev = hcrx->ccid3hcrx_p;
952 /* Calculate loss event rate */
953 if (!list_empty(&hcrx->ccid3hcrx_li_hist)) {
954 u32 i_mean = dccp_li_hist_calc_i_mean(&hcrx->ccid3hcrx_li_hist);
956 /* Scaling up by 1000000 as fixed decimal */
958 hcrx->ccid3hcrx_p = 1000000 / i_mean;
960 DCCP_BUG("empty loss history");
962 if (hcrx->ccid3hcrx_p > p_prev) {
963 ccid3_hc_rx_send_feedback(sk);
968 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
970 struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
972 ccid3_pr_debug("entry\n");
974 hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
975 INIT_LIST_HEAD(&hcrx->ccid3hcrx_hist);
976 INIT_LIST_HEAD(&hcrx->ccid3hcrx_li_hist);
977 hcrx->ccid3hcrx_tstamp_last_feedback =
978 hcrx->ccid3hcrx_tstamp_last_ack = ktime_get_real();
979 hcrx->ccid3hcrx_s = 0;
980 hcrx->ccid3hcrx_rtt = 0;
984 static void ccid3_hc_rx_exit(struct sock *sk)
986 struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
988 BUG_ON(hcrx == NULL);
990 ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
992 /* Empty packet history */
993 dccp_rx_hist_purge(ccid3_rx_hist, &hcrx->ccid3hcrx_hist);
995 /* Empty loss interval history */
996 dccp_li_hist_purge(&hcrx->ccid3hcrx_li_hist);
999 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
1001 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1003 /* Listen socks doesn't have a private CCID block */
1004 if (sk->sk_state == DCCP_LISTEN)
1007 BUG_ON(hcrx == NULL);
1009 info->tcpi_ca_state = hcrx->ccid3hcrx_state;
1010 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1011 info->tcpi_rcv_rtt = hcrx->ccid3hcrx_rtt;
1014 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
1015 u32 __user *optval, int __user *optlen)
1017 const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
1020 /* Listen socks doesn't have a private CCID block */
1021 if (sk->sk_state == DCCP_LISTEN)
1025 case DCCP_SOCKOPT_CCID_RX_INFO:
1026 if (len < sizeof(hcrx->ccid3hcrx_tfrc))
1028 len = sizeof(hcrx->ccid3hcrx_tfrc);
1029 val = &hcrx->ccid3hcrx_tfrc;
1032 return -ENOPROTOOPT;
1035 if (put_user(len, optlen) || copy_to_user(optval, val, len))
1041 static struct ccid_operations ccid3 = {
1042 .ccid_id = DCCPC_CCID3,
1043 .ccid_name = "ccid3",
1044 .ccid_owner = THIS_MODULE,
1045 .ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
1046 .ccid_hc_tx_init = ccid3_hc_tx_init,
1047 .ccid_hc_tx_exit = ccid3_hc_tx_exit,
1048 .ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
1049 .ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
1050 .ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
1051 .ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
1052 .ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
1053 .ccid_hc_rx_init = ccid3_hc_rx_init,
1054 .ccid_hc_rx_exit = ccid3_hc_rx_exit,
1055 .ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
1056 .ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
1057 .ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
1058 .ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
1059 .ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
1060 .ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
1063 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
1064 module_param(ccid3_debug, int, 0444);
1065 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
1068 static __init int ccid3_module_init(void)
1072 ccid3_rx_hist = dccp_rx_hist_new("ccid3");
1073 if (ccid3_rx_hist == NULL)
1076 ccid3_tx_hist = dccp_tx_hist_new("ccid3");
1077 if (ccid3_tx_hist == NULL)
1080 rc = ccid_register(&ccid3);
1087 dccp_tx_hist_delete(ccid3_tx_hist);
1088 ccid3_tx_hist = NULL;
1090 dccp_rx_hist_delete(ccid3_rx_hist);
1091 ccid3_rx_hist = NULL;
1094 module_init(ccid3_module_init);
1096 static __exit void ccid3_module_exit(void)
1098 ccid_unregister(&ccid3);
1100 if (ccid3_tx_hist != NULL) {
1101 dccp_tx_hist_delete(ccid3_tx_hist);
1102 ccid3_tx_hist = NULL;
1104 if (ccid3_rx_hist != NULL) {
1105 dccp_rx_hist_delete(ccid3_rx_hist);
1106 ccid3_rx_hist = NULL;
1109 module_exit(ccid3_module_exit);
1111 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
1112 "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
1113 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
1114 MODULE_LICENSE("GPL");
1115 MODULE_ALIAS("net-dccp-ccid-3");