Linux Audio

Check our new training course

Loading...
v3.1
  1/*
  2 *  Copyright (c) 2007   The University of Aberdeen, Scotland, UK
  3 *  Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
  4 *  Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
  5 *
  6 *  An implementation of the DCCP protocol
  7 *
  8 *  This code has been developed by the University of Waikato WAND
  9 *  research group. For further information please see http://www.wand.net.nz/
 10 *
 11 *  This code also uses code from Lulea University, rereleased as GPL by its
 12 *  authors:
 13 *  Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
 14 *
 15 *  Changes to meet Linux coding standards, to make it meet latest ccid3 draft
 16 *  and to make it work as a loadable module in the DCCP stack written by
 17 *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
 18 *
 19 *  Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
 20 *
 21 *  This program is free software; you can redistribute it and/or modify
 22 *  it under the terms of the GNU General Public License as published by
 23 *  the Free Software Foundation; either version 2 of the License, or
 24 *  (at your option) any later version.
 25 *
 26 *  This program is distributed in the hope that it will be useful,
 27 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 28 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 29 *  GNU General Public License for more details.
 30 *
 31 *  You should have received a copy of the GNU General Public License
 32 *  along with this program; if not, write to the Free Software
 33 *  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 34 */
 35#include "../dccp.h"
 36#include "ccid3.h"
 37
 38#include <asm/unaligned.h>
 39
 40#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
 41static int ccid3_debug;
 42#define ccid3_pr_debug(format, a...)	DCCP_PR_DEBUG(ccid3_debug, format, ##a)
 43#else
 44#define ccid3_pr_debug(format, a...)
 45#endif
 46
 47/*
 48 *	Transmitter Half-Connection Routines
 49 */
 50#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
 51static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
 52{
 53	static const char *const ccid3_state_names[] = {
 54	[TFRC_SSTATE_NO_SENT]  = "NO_SENT",
 55	[TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
 56	[TFRC_SSTATE_FBACK]    = "FBACK",
 57	};
 58
 59	return ccid3_state_names[state];
 60}
 61#endif
 62
 63static void ccid3_hc_tx_set_state(struct sock *sk,
 64				  enum ccid3_hc_tx_states state)
 65{
 66	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
 67	enum ccid3_hc_tx_states oldstate = hc->tx_state;
 68
 69	ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
 70		       dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
 71		       ccid3_tx_state_name(state));
 72	WARN_ON(state == oldstate);
 73	hc->tx_state = state;
 74}
 75
 76/*
 77 * Compute the initial sending rate X_init in the manner of RFC 3390:
 78 *
 79 *	X_init  =  min(4 * s, max(2 * s, 4380 bytes)) / RTT
 80 *
 81 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
 82 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
 83 * For consistency with other parts of the code, X_init is scaled by 2^6.
 84 */
 85static inline u64 rfc3390_initial_rate(struct sock *sk)
 86{
 87	const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
 88	const __u32 w_init = clamp_t(__u32, 4380U, 2 * hc->tx_s, 4 * hc->tx_s);
 89
 90	return scaled_div(w_init << 6, hc->tx_rtt);
 91}
 92
 93/**
 94 * ccid3_update_send_interval  -  Calculate new t_ipi = s / X_inst
 95 * This respects the granularity of X_inst (64 * bytes/second).
 96 */
 97static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hc)
 98{
 99	hc->tx_t_ipi = scaled_div32(((u64)hc->tx_s) << 6, hc->tx_x);
100
 
101	ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hc->tx_t_ipi,
102		       hc->tx_s, (unsigned)(hc->tx_x >> 6));
103}
104
105static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hc, ktime_t now)
106{
107	u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count);
108
109	return delta / hc->tx_rtt;
110}
111
112/**
113 * ccid3_hc_tx_update_x  -  Update allowed sending rate X
114 * @stamp: most recent time if available - can be left NULL.
 
115 * This function tracks draft rfc3448bis, check there for latest details.
116 *
117 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
118 *       fine-grained resolution of sending rates. This requires scaling by 2^6
119 *       throughout the code. Only X_calc is unscaled (in bytes/second).
120 *
121 */
122static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
123{
124	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
125	__u64 min_rate = 2 * hc->tx_x_recv;
126	const __u64 old_x = hc->tx_x;
127	ktime_t now = stamp ? *stamp : ktime_get_real();
128
129	/*
130	 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
131	 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
132	 * a sender is idle if it has not sent anything over a 2-RTT-period.
133	 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
134	 */
135	if (ccid3_hc_tx_idle_rtt(hc, now) >= 2) {
136		min_rate = rfc3390_initial_rate(sk);
137		min_rate = max(min_rate, 2 * hc->tx_x_recv);
138	}
139
140	if (hc->tx_p > 0) {
141
142		hc->tx_x = min(((__u64)hc->tx_x_calc) << 6, min_rate);
143		hc->tx_x = max(hc->tx_x, (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
144
145	} else if (ktime_us_delta(now, hc->tx_t_ld) - (s64)hc->tx_rtt >= 0) {
146
147		hc->tx_x = min(2 * hc->tx_x, min_rate);
148		hc->tx_x = max(hc->tx_x,
149			       scaled_div(((__u64)hc->tx_s) << 6, hc->tx_rtt));
150		hc->tx_t_ld = now;
151	}
152
153	if (hc->tx_x != old_x) {
154		ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
155			       "X_recv=%u\n", (unsigned)(old_x >> 6),
156			       (unsigned)(hc->tx_x >> 6), hc->tx_x_calc,
157			       (unsigned)(hc->tx_x_recv >> 6));
158
159		ccid3_update_send_interval(hc);
160	}
161}
162
163/*
164 *	Track the mean packet size `s' (cf. RFC 4342, 5.3 and  RFC 3448, 4.1)
165 *	@len: DCCP packet payload size in bytes
 
 
166 */
167static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hc, int len)
168{
169	const u16 old_s = hc->tx_s;
170
171	hc->tx_s = tfrc_ewma(hc->tx_s, len, 9);
172
173	if (hc->tx_s != old_s)
174		ccid3_update_send_interval(hc);
175}
176
177/*
178 *	Update Window Counter using the algorithm from [RFC 4342, 8.1].
179 *	As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
180 */
181static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hc,
182						ktime_t now)
183{
184	u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count),
185	    quarter_rtts = (4 * delta) / hc->tx_rtt;
186
187	if (quarter_rtts > 0) {
188		hc->tx_t_last_win_count = now;
189		hc->tx_last_win_count  += min(quarter_rtts, 5U);
190		hc->tx_last_win_count  &= 0xF;		/* mod 16 */
191	}
192}
193
194static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
195{
196	struct sock *sk = (struct sock *)data;
197	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
198	unsigned long t_nfb = USEC_PER_SEC / 5;
199
200	bh_lock_sock(sk);
201	if (sock_owned_by_user(sk)) {
202		/* Try again later. */
203		/* XXX: set some sensible MIB */
204		goto restart_timer;
205	}
206
207	ccid3_pr_debug("%s(%p, state=%s) - entry\n", dccp_role(sk), sk,
208		       ccid3_tx_state_name(hc->tx_state));
209
210	/* Ignore and do not restart after leaving the established state */
211	if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
212		goto out;
213
214	/* Reset feedback state to "no feedback received" */
215	if (hc->tx_state == TFRC_SSTATE_FBACK)
216		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
217
218	/*
219	 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
220	 * RTO is 0 if and only if no feedback has been received yet.
221	 */
222	if (hc->tx_t_rto == 0 || hc->tx_p == 0) {
223
224		/* halve send rate directly */
225		hc->tx_x = max(hc->tx_x / 2,
226			       (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
227		ccid3_update_send_interval(hc);
228	} else {
229		/*
230		 *  Modify the cached value of X_recv
231		 *
232		 *  If (X_calc > 2 * X_recv)
233		 *    X_recv = max(X_recv / 2, s / (2 * t_mbi));
234		 *  Else
235		 *    X_recv = X_calc / 4;
236		 *
237		 *  Note that X_recv is scaled by 2^6 while X_calc is not
238		 */
239		BUG_ON(hc->tx_p && !hc->tx_x_calc);
240
241		if (hc->tx_x_calc > (hc->tx_x_recv >> 5))
242			hc->tx_x_recv =
243				max(hc->tx_x_recv / 2,
244				    (((__u64)hc->tx_s) << 6) / (2*TFRC_T_MBI));
245		else {
246			hc->tx_x_recv = hc->tx_x_calc;
247			hc->tx_x_recv <<= 4;
248		}
249		ccid3_hc_tx_update_x(sk, NULL);
250	}
251	ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
252			(unsigned long long)hc->tx_x);
253
254	/*
255	 * Set new timeout for the nofeedback timer.
256	 * See comments in packet_recv() regarding the value of t_RTO.
257	 */
258	if (unlikely(hc->tx_t_rto == 0))	/* no feedback received yet */
259		t_nfb = TFRC_INITIAL_TIMEOUT;
260	else
261		t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
262
263restart_timer:
264	sk_reset_timer(sk, &hc->tx_no_feedback_timer,
265			   jiffies + usecs_to_jiffies(t_nfb));
266out:
267	bh_unlock_sock(sk);
268	sock_put(sk);
269}
270
271/**
272 * ccid3_hc_tx_send_packet  -  Delay-based dequeueing of TX packets
273 * @skb: next packet candidate to send on @sk
 
274 * This function uses the convention of ccid_packet_dequeue_eval() and
275 * returns a millisecond-delay value between 0 and t_mbi = 64000 msec.
276 */
277static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
278{
279	struct dccp_sock *dp = dccp_sk(sk);
280	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
281	ktime_t now = ktime_get_real();
282	s64 delay;
283
284	/*
285	 * This function is called only for Data and DataAck packets. Sending
286	 * zero-sized Data(Ack)s is theoretically possible, but for congestion
287	 * control this case is pathological - ignore it.
288	 */
289	if (unlikely(skb->len == 0))
290		return -EBADMSG;
291
292	if (hc->tx_state == TFRC_SSTATE_NO_SENT) {
293		sk_reset_timer(sk, &hc->tx_no_feedback_timer, (jiffies +
294			       usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
295		hc->tx_last_win_count	= 0;
296		hc->tx_t_last_win_count = now;
297
298		/* Set t_0 for initial packet */
299		hc->tx_t_nom = now;
300
301		hc->tx_s = skb->len;
302
303		/*
304		 * Use initial RTT sample when available: recommended by erratum
305		 * to RFC 4342. This implements the initialisation procedure of
306		 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
307		 */
308		if (dp->dccps_syn_rtt) {
309			ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
310			hc->tx_rtt  = dp->dccps_syn_rtt;
311			hc->tx_x    = rfc3390_initial_rate(sk);
312			hc->tx_t_ld = now;
313		} else {
314			/*
315			 * Sender does not have RTT sample:
316			 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
317			 *   is needed in several parts (e.g.  window counter);
318			 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
319			 */
320			hc->tx_rtt = DCCP_FALLBACK_RTT;
321			hc->tx_x   = hc->tx_s;
322			hc->tx_x <<= 6;
323		}
324		ccid3_update_send_interval(hc);
325
326		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
327
328	} else {
329		delay = ktime_us_delta(hc->tx_t_nom, now);
330		ccid3_pr_debug("delay=%ld\n", (long)delay);
331		/*
332		 *	Scheduling of packet transmissions (RFC 5348, 8.3)
333		 *
334		 * if (t_now > t_nom - delta)
335		 *       // send the packet now
336		 * else
337		 *       // send the packet in (t_nom - t_now) milliseconds.
338		 */
339		if (delay >= TFRC_T_DELTA)
340			return (u32)delay / USEC_PER_MSEC;
341
342		ccid3_hc_tx_update_win_count(hc, now);
343	}
344
345	/* prepare to send now (add options etc.) */
346	dp->dccps_hc_tx_insert_options = 1;
347	DCCP_SKB_CB(skb)->dccpd_ccval  = hc->tx_last_win_count;
348
349	/* set the nominal send time for the next following packet */
350	hc->tx_t_nom = ktime_add_us(hc->tx_t_nom, hc->tx_t_ipi);
351	return CCID_PACKET_SEND_AT_ONCE;
352}
353
354static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len)
355{
356	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
357
358	ccid3_hc_tx_update_s(hc, len);
359
360	if (tfrc_tx_hist_add(&hc->tx_hist, dccp_sk(sk)->dccps_gss))
361		DCCP_CRIT("packet history - out of memory!");
362}
363
364static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
365{
366	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
367	struct tfrc_tx_hist_entry *acked;
368	ktime_t now;
369	unsigned long t_nfb;
370	u32 r_sample;
371
372	/* we are only interested in ACKs */
373	if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
374	      DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
375		return;
376	/*
377	 * Locate the acknowledged packet in the TX history.
378	 *
379	 * Returning "entry not found" here can for instance happen when
380	 *  - the host has not sent out anything (e.g. a passive server),
381	 *  - the Ack is outdated (packet with higher Ack number was received),
382	 *  - it is a bogus Ack (for a packet not sent on this connection).
383	 */
384	acked = tfrc_tx_hist_find_entry(hc->tx_hist, dccp_hdr_ack_seq(skb));
385	if (acked == NULL)
386		return;
387	/* For the sake of RTT sampling, ignore/remove all older entries */
388	tfrc_tx_hist_purge(&acked->next);
389
390	/* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
391	now	  = ktime_get_real();
392	r_sample  = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
393	hc->tx_rtt = tfrc_ewma(hc->tx_rtt, r_sample, 9);
394
395	/*
396	 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
397	 */
398	if (hc->tx_state == TFRC_SSTATE_NO_FBACK) {
399		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
400
401		if (hc->tx_t_rto == 0) {
402			/*
403			 * Initial feedback packet: Larger Initial Windows (4.2)
404			 */
405			hc->tx_x    = rfc3390_initial_rate(sk);
406			hc->tx_t_ld = now;
407
408			ccid3_update_send_interval(hc);
409
410			goto done_computing_x;
411		} else if (hc->tx_p == 0) {
412			/*
413			 * First feedback after nofeedback timer expiry (4.3)
414			 */
415			goto done_computing_x;
416		}
417	}
418
419	/* Update sending rate (step 4 of [RFC 3448, 4.3]) */
420	if (hc->tx_p > 0)
421		hc->tx_x_calc = tfrc_calc_x(hc->tx_s, hc->tx_rtt, hc->tx_p);
422	ccid3_hc_tx_update_x(sk, &now);
423
424done_computing_x:
425	ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
426			       "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
427			       dccp_role(sk), sk, hc->tx_rtt, r_sample,
428			       hc->tx_s, hc->tx_p, hc->tx_x_calc,
429			       (unsigned)(hc->tx_x_recv >> 6),
430			       (unsigned)(hc->tx_x >> 6));
431
432	/* unschedule no feedback timer */
433	sk_stop_timer(sk, &hc->tx_no_feedback_timer);
434
435	/*
436	 * As we have calculated new ipi, delta, t_nom it is possible
437	 * that we now can send a packet, so wake up dccp_wait_for_ccid
438	 */
439	sk->sk_write_space(sk);
440
441	/*
442	 * Update timeout interval for the nofeedback timer. In order to control
443	 * rate halving on networks with very low RTTs (<= 1 ms), use per-route
444	 * tunable RTAX_RTO_MIN value as the lower bound.
445	 */
446	hc->tx_t_rto = max_t(u32, 4 * hc->tx_rtt,
447				  USEC_PER_SEC/HZ * tcp_rto_min(sk));
448	/*
449	 * Schedule no feedback timer to expire in
450	 * max(t_RTO, 2 * s/X)  =  max(t_RTO, 2 * t_ipi)
451	 */
452	t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
453
454	ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
455		       "expire in %lu jiffies (%luus)\n",
456		       dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
457
458	sk_reset_timer(sk, &hc->tx_no_feedback_timer,
459			   jiffies + usecs_to_jiffies(t_nfb));
460}
461
462static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type,
463				     u8 option, u8 *optval, u8 optlen)
464{
465	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
466	__be32 opt_val;
467
468	switch (option) {
469	case TFRC_OPT_RECEIVE_RATE:
470	case TFRC_OPT_LOSS_EVENT_RATE:
471		/* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
472		if (packet_type == DCCP_PKT_DATA)
473			break;
474		if (unlikely(optlen != 4)) {
475			DCCP_WARN("%s(%p), invalid len %d for %u\n",
476				  dccp_role(sk), sk, optlen, option);
477			return -EINVAL;
478		}
479		opt_val = ntohl(get_unaligned((__be32 *)optval));
480
481		if (option == TFRC_OPT_RECEIVE_RATE) {
482			/* Receive Rate is kept in units of 64 bytes/second */
483			hc->tx_x_recv = opt_val;
484			hc->tx_x_recv <<= 6;
485
486			ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
487				       dccp_role(sk), sk, opt_val);
488		} else {
489			/* Update the fixpoint Loss Event Rate fraction */
490			hc->tx_p = tfrc_invert_loss_event_rate(opt_val);
491
492			ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
493				       dccp_role(sk), sk, opt_val);
494		}
495	}
496	return 0;
497}
498
499static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
500{
501	struct ccid3_hc_tx_sock *hc = ccid_priv(ccid);
502
503	hc->tx_state = TFRC_SSTATE_NO_SENT;
504	hc->tx_hist  = NULL;
505	setup_timer(&hc->tx_no_feedback_timer,
506			ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
 
507	return 0;
508}
509
510static void ccid3_hc_tx_exit(struct sock *sk)
511{
512	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
513
514	sk_stop_timer(sk, &hc->tx_no_feedback_timer);
515	tfrc_tx_hist_purge(&hc->tx_hist);
516}
517
518static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
519{
520	info->tcpi_rto = ccid3_hc_tx_sk(sk)->tx_t_rto;
521	info->tcpi_rtt = ccid3_hc_tx_sk(sk)->tx_rtt;
522}
523
524static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
525				  u32 __user *optval, int __user *optlen)
526{
527	const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
528	struct tfrc_tx_info tfrc;
529	const void *val;
530
531	switch (optname) {
532	case DCCP_SOCKOPT_CCID_TX_INFO:
533		if (len < sizeof(tfrc))
534			return -EINVAL;
 
535		tfrc.tfrctx_x	   = hc->tx_x;
536		tfrc.tfrctx_x_recv = hc->tx_x_recv;
537		tfrc.tfrctx_x_calc = hc->tx_x_calc;
538		tfrc.tfrctx_rtt	   = hc->tx_rtt;
539		tfrc.tfrctx_p	   = hc->tx_p;
540		tfrc.tfrctx_rto	   = hc->tx_t_rto;
541		tfrc.tfrctx_ipi	   = hc->tx_t_ipi;
542		len = sizeof(tfrc);
543		val = &tfrc;
544		break;
545	default:
546		return -ENOPROTOOPT;
547	}
548
549	if (put_user(len, optlen) || copy_to_user(optval, val, len))
550		return -EFAULT;
551
552	return 0;
553}
554
555/*
556 *	Receiver Half-Connection Routines
557 */
558
559/* CCID3 feedback types */
560enum ccid3_fback_type {
561	CCID3_FBACK_NONE = 0,
562	CCID3_FBACK_INITIAL,
563	CCID3_FBACK_PERIODIC,
564	CCID3_FBACK_PARAM_CHANGE
565};
566
567#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
568static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
569{
570	static const char *const ccid3_rx_state_names[] = {
571	[TFRC_RSTATE_NO_DATA] = "NO_DATA",
572	[TFRC_RSTATE_DATA]    = "DATA",
573	};
574
575	return ccid3_rx_state_names[state];
576}
577#endif
578
579static void ccid3_hc_rx_set_state(struct sock *sk,
580				  enum ccid3_hc_rx_states state)
581{
582	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
583	enum ccid3_hc_rx_states oldstate = hc->rx_state;
584
585	ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
586		       dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
587		       ccid3_rx_state_name(state));
588	WARN_ON(state == oldstate);
589	hc->rx_state = state;
590}
591
592static void ccid3_hc_rx_send_feedback(struct sock *sk,
593				      const struct sk_buff *skb,
594				      enum ccid3_fback_type fbtype)
595{
596	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
597	struct dccp_sock *dp = dccp_sk(sk);
598	ktime_t now = ktime_get_real();
599	s64 delta = 0;
600
601	switch (fbtype) {
602	case CCID3_FBACK_INITIAL:
603		hc->rx_x_recv = 0;
604		hc->rx_pinv   = ~0U;   /* see RFC 4342, 8.5 */
605		break;
606	case CCID3_FBACK_PARAM_CHANGE:
607		/*
608		 * When parameters change (new loss or p > p_prev), we do not
609		 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
610		 * need to  reuse the previous value of X_recv. However, when
611		 * X_recv was 0 (due to early loss), this would kill X down to
612		 * s/t_mbi (i.e. one packet in 64 seconds).
613		 * To avoid such drastic reduction, we approximate X_recv as
614		 * the number of bytes since last feedback.
615		 * This is a safe fallback, since X is bounded above by X_calc.
616		 */
617		if (hc->rx_x_recv > 0)
618			break;
619		/* fall through */
620	case CCID3_FBACK_PERIODIC:
621		delta = ktime_us_delta(now, hc->rx_tstamp_last_feedback);
622		if (delta <= 0)
623			DCCP_BUG("delta (%ld) <= 0", (long)delta);
624		else
625			hc->rx_x_recv = scaled_div32(hc->rx_bytes_recv, delta);
626		break;
627	default:
628		return;
629	}
630
631	ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta,
632		       hc->rx_x_recv, hc->rx_pinv);
633
634	hc->rx_tstamp_last_feedback = now;
635	hc->rx_last_counter	    = dccp_hdr(skb)->dccph_ccval;
636	hc->rx_bytes_recv	    = 0;
637
638	dp->dccps_hc_rx_insert_options = 1;
639	dccp_send_ack(sk);
640}
641
642static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
643{
644	const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
645	__be32 x_recv, pinv;
646
647	if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
648		return 0;
649
650	if (dccp_packet_without_ack(skb))
651		return 0;
652
653	x_recv = htonl(hc->rx_x_recv);
654	pinv   = htonl(hc->rx_pinv);
655
656	if (dccp_insert_option(skb, TFRC_OPT_LOSS_EVENT_RATE,
657			       &pinv, sizeof(pinv)) ||
658	    dccp_insert_option(skb, TFRC_OPT_RECEIVE_RATE,
659			       &x_recv, sizeof(x_recv)))
660		return -1;
661
662	return 0;
663}
664
665/**
666 * ccid3_first_li  -  Implements [RFC 5348, 6.3.1]
667 *
668 * Determine the length of the first loss interval via inverse lookup.
669 * Assume that X_recv can be computed by the throughput equation
670 *		    s
671 *	X_recv = --------
672 *		 R * fval
673 * Find some p such that f(p) = fval; return 1/p (scaled).
674 */
675static u32 ccid3_first_li(struct sock *sk)
676{
677	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
678	u32 x_recv, p, delta;
679	u64 fval;
680
681	if (hc->rx_rtt == 0) {
682		DCCP_WARN("No RTT estimate available, using fallback RTT\n");
683		hc->rx_rtt = DCCP_FALLBACK_RTT;
684	}
685
686	delta  = ktime_to_us(net_timedelta(hc->rx_tstamp_last_feedback));
687	x_recv = scaled_div32(hc->rx_bytes_recv, delta);
688	if (x_recv == 0) {		/* would also trigger divide-by-zero */
689		DCCP_WARN("X_recv==0\n");
690		if (hc->rx_x_recv == 0) {
691			DCCP_BUG("stored value of X_recv is zero");
692			return ~0U;
693		}
694		x_recv = hc->rx_x_recv;
695	}
696
697	fval = scaled_div(hc->rx_s, hc->rx_rtt);
698	fval = scaled_div32(fval, x_recv);
699	p = tfrc_calc_x_reverse_lookup(fval);
700
701	ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
702		       "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
703
704	return p == 0 ? ~0U : scaled_div(1, p);
705}
706
707static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
708{
709	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
710	enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
711	const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
712	const bool is_data_packet = dccp_data_packet(skb);
713
714	if (unlikely(hc->rx_state == TFRC_RSTATE_NO_DATA)) {
715		if (is_data_packet) {
716			const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
717			do_feedback = CCID3_FBACK_INITIAL;
718			ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
719			hc->rx_s = payload;
720			/*
721			 * Not necessary to update rx_bytes_recv here,
722			 * since X_recv = 0 for the first feedback packet (cf.
723			 * RFC 3448, 6.3) -- gerrit
724			 */
725		}
726		goto update_records;
727	}
728
729	if (tfrc_rx_hist_duplicate(&hc->rx_hist, skb))
730		return; /* done receiving */
731
732	if (is_data_packet) {
733		const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
734		/*
735		 * Update moving-average of s and the sum of received payload bytes
736		 */
737		hc->rx_s = tfrc_ewma(hc->rx_s, payload, 9);
738		hc->rx_bytes_recv += payload;
739	}
740
741	/*
742	 * Perform loss detection and handle pending losses
743	 */
744	if (tfrc_rx_handle_loss(&hc->rx_hist, &hc->rx_li_hist,
745				skb, ndp, ccid3_first_li, sk)) {
746		do_feedback = CCID3_FBACK_PARAM_CHANGE;
747		goto done_receiving;
748	}
749
750	if (tfrc_rx_hist_loss_pending(&hc->rx_hist))
751		return; /* done receiving */
752
753	/*
754	 * Handle data packets: RTT sampling and monitoring p
755	 */
756	if (unlikely(!is_data_packet))
757		goto update_records;
758
759	if (!tfrc_lh_is_initialised(&hc->rx_li_hist)) {
760		const u32 sample = tfrc_rx_hist_sample_rtt(&hc->rx_hist, skb);
761		/*
762		 * Empty loss history: no loss so far, hence p stays 0.
763		 * Sample RTT values, since an RTT estimate is required for the
764		 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
765		 */
766		if (sample != 0)
767			hc->rx_rtt = tfrc_ewma(hc->rx_rtt, sample, 9);
768
769	} else if (tfrc_lh_update_i_mean(&hc->rx_li_hist, skb)) {
770		/*
771		 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
772		 * has decreased (resp. p has increased), send feedback now.
773		 */
774		do_feedback = CCID3_FBACK_PARAM_CHANGE;
775	}
776
777	/*
778	 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
779	 */
780	if (SUB16(dccp_hdr(skb)->dccph_ccval, hc->rx_last_counter) > 3)
781		do_feedback = CCID3_FBACK_PERIODIC;
782
783update_records:
784	tfrc_rx_hist_add_packet(&hc->rx_hist, skb, ndp);
785
786done_receiving:
787	if (do_feedback)
788		ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
789}
790
791static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
792{
793	struct ccid3_hc_rx_sock *hc = ccid_priv(ccid);
794
795	hc->rx_state = TFRC_RSTATE_NO_DATA;
796	tfrc_lh_init(&hc->rx_li_hist);
797	return tfrc_rx_hist_alloc(&hc->rx_hist);
798}
799
800static void ccid3_hc_rx_exit(struct sock *sk)
801{
802	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
803
804	tfrc_rx_hist_purge(&hc->rx_hist);
805	tfrc_lh_cleanup(&hc->rx_li_hist);
806}
807
808static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
809{
810	info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->rx_state;
811	info->tcpi_options  |= TCPI_OPT_TIMESTAMPS;
812	info->tcpi_rcv_rtt  = ccid3_hc_rx_sk(sk)->rx_rtt;
813}
814
815static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
816				  u32 __user *optval, int __user *optlen)
817{
818	const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
819	struct tfrc_rx_info rx_info;
820	const void *val;
821
822	switch (optname) {
823	case DCCP_SOCKOPT_CCID_RX_INFO:
824		if (len < sizeof(rx_info))
825			return -EINVAL;
826		rx_info.tfrcrx_x_recv = hc->rx_x_recv;
827		rx_info.tfrcrx_rtt    = hc->rx_rtt;
828		rx_info.tfrcrx_p      = tfrc_invert_loss_event_rate(hc->rx_pinv);
829		len = sizeof(rx_info);
830		val = &rx_info;
831		break;
832	default:
833		return -ENOPROTOOPT;
834	}
835
836	if (put_user(len, optlen) || copy_to_user(optval, val, len))
837		return -EFAULT;
838
839	return 0;
840}
841
842struct ccid_operations ccid3_ops = {
843	.ccid_id		   = DCCPC_CCID3,
844	.ccid_name		   = "TCP-Friendly Rate Control",
845	.ccid_hc_tx_obj_size	   = sizeof(struct ccid3_hc_tx_sock),
846	.ccid_hc_tx_init	   = ccid3_hc_tx_init,
847	.ccid_hc_tx_exit	   = ccid3_hc_tx_exit,
848	.ccid_hc_tx_send_packet	   = ccid3_hc_tx_send_packet,
849	.ccid_hc_tx_packet_sent	   = ccid3_hc_tx_packet_sent,
850	.ccid_hc_tx_packet_recv	   = ccid3_hc_tx_packet_recv,
851	.ccid_hc_tx_parse_options  = ccid3_hc_tx_parse_options,
852	.ccid_hc_rx_obj_size	   = sizeof(struct ccid3_hc_rx_sock),
853	.ccid_hc_rx_init	   = ccid3_hc_rx_init,
854	.ccid_hc_rx_exit	   = ccid3_hc_rx_exit,
855	.ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
856	.ccid_hc_rx_packet_recv	   = ccid3_hc_rx_packet_recv,
857	.ccid_hc_rx_get_info	   = ccid3_hc_rx_get_info,
858	.ccid_hc_tx_get_info	   = ccid3_hc_tx_get_info,
859	.ccid_hc_rx_getsockopt	   = ccid3_hc_rx_getsockopt,
860	.ccid_hc_tx_getsockopt	   = ccid3_hc_tx_getsockopt,
861};
862
863#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
864module_param(ccid3_debug, bool, 0644);
865MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages");
866#endif
v4.17
  1/*
  2 *  Copyright (c) 2007   The University of Aberdeen, Scotland, UK
  3 *  Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
  4 *  Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
  5 *
  6 *  An implementation of the DCCP protocol
  7 *
  8 *  This code has been developed by the University of Waikato WAND
  9 *  research group. For further information please see http://www.wand.net.nz/
 10 *
 11 *  This code also uses code from Lulea University, rereleased as GPL by its
 12 *  authors:
 13 *  Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
 14 *
 15 *  Changes to meet Linux coding standards, to make it meet latest ccid3 draft
 16 *  and to make it work as a loadable module in the DCCP stack written by
 17 *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
 18 *
 19 *  Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
 20 *
 21 *  This program is free software; you can redistribute it and/or modify
 22 *  it under the terms of the GNU General Public License as published by
 23 *  the Free Software Foundation; either version 2 of the License, or
 24 *  (at your option) any later version.
 25 *
 26 *  This program is distributed in the hope that it will be useful,
 27 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 28 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 29 *  GNU General Public License for more details.
 30 *
 31 *  You should have received a copy of the GNU General Public License
 32 *  along with this program; if not, write to the Free Software
 33 *  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 34 */
 35#include "../dccp.h"
 36#include "ccid3.h"
 37
 38#include <asm/unaligned.h>
 39
 40#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
 41static bool ccid3_debug;
 42#define ccid3_pr_debug(format, a...)	DCCP_PR_DEBUG(ccid3_debug, format, ##a)
 43#else
 44#define ccid3_pr_debug(format, a...)
 45#endif
 46
 47/*
 48 *	Transmitter Half-Connection Routines
 49 */
 50#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
 51static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
 52{
 53	static const char *const ccid3_state_names[] = {
 54	[TFRC_SSTATE_NO_SENT]  = "NO_SENT",
 55	[TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
 56	[TFRC_SSTATE_FBACK]    = "FBACK",
 57	};
 58
 59	return ccid3_state_names[state];
 60}
 61#endif
 62
 63static void ccid3_hc_tx_set_state(struct sock *sk,
 64				  enum ccid3_hc_tx_states state)
 65{
 66	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
 67	enum ccid3_hc_tx_states oldstate = hc->tx_state;
 68
 69	ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
 70		       dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
 71		       ccid3_tx_state_name(state));
 72	WARN_ON(state == oldstate);
 73	hc->tx_state = state;
 74}
 75
 76/*
 77 * Compute the initial sending rate X_init in the manner of RFC 3390:
 78 *
 79 *	X_init  =  min(4 * s, max(2 * s, 4380 bytes)) / RTT
 80 *
 81 * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
 82 * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
 83 * For consistency with other parts of the code, X_init is scaled by 2^6.
 84 */
 85static inline u64 rfc3390_initial_rate(struct sock *sk)
 86{
 87	const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
 88	const __u32 w_init = clamp_t(__u32, 4380U, 2 * hc->tx_s, 4 * hc->tx_s);
 89
 90	return scaled_div(w_init << 6, hc->tx_rtt);
 91}
 92
 93/**
 94 * ccid3_update_send_interval  -  Calculate new t_ipi = s / X_inst
 95 * This respects the granularity of X_inst (64 * bytes/second).
 96 */
 97static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hc)
 98{
 99	hc->tx_t_ipi = scaled_div32(((u64)hc->tx_s) << 6, hc->tx_x);
100
101	DCCP_BUG_ON(hc->tx_t_ipi == 0);
102	ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hc->tx_t_ipi,
103		       hc->tx_s, (unsigned int)(hc->tx_x >> 6));
104}
105
106static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hc, ktime_t now)
107{
108	u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count);
109
110	return delta / hc->tx_rtt;
111}
112
113/**
114 * ccid3_hc_tx_update_x  -  Update allowed sending rate X
115 * @stamp: most recent time if available - can be left NULL.
116 *
117 * This function tracks draft rfc3448bis, check there for latest details.
118 *
119 * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
120 *       fine-grained resolution of sending rates. This requires scaling by 2^6
121 *       throughout the code. Only X_calc is unscaled (in bytes/second).
122 *
123 */
124static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
125{
126	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
127	__u64 min_rate = 2 * hc->tx_x_recv;
128	const __u64 old_x = hc->tx_x;
129	ktime_t now = stamp ? *stamp : ktime_get_real();
130
131	/*
132	 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
133	 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
134	 * a sender is idle if it has not sent anything over a 2-RTT-period.
135	 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
136	 */
137	if (ccid3_hc_tx_idle_rtt(hc, now) >= 2) {
138		min_rate = rfc3390_initial_rate(sk);
139		min_rate = max(min_rate, 2 * hc->tx_x_recv);
140	}
141
142	if (hc->tx_p > 0) {
143
144		hc->tx_x = min(((__u64)hc->tx_x_calc) << 6, min_rate);
145		hc->tx_x = max(hc->tx_x, (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
146
147	} else if (ktime_us_delta(now, hc->tx_t_ld) - (s64)hc->tx_rtt >= 0) {
148
149		hc->tx_x = min(2 * hc->tx_x, min_rate);
150		hc->tx_x = max(hc->tx_x,
151			       scaled_div(((__u64)hc->tx_s) << 6, hc->tx_rtt));
152		hc->tx_t_ld = now;
153	}
154
155	if (hc->tx_x != old_x) {
156		ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
157			       "X_recv=%u\n", (unsigned int)(old_x >> 6),
158			       (unsigned int)(hc->tx_x >> 6), hc->tx_x_calc,
159			       (unsigned int)(hc->tx_x_recv >> 6));
160
161		ccid3_update_send_interval(hc);
162	}
163}
164
165/**
166 *	ccid3_hc_tx_update_s - Track the mean packet size `s'
167 *	@len: DCCP packet payload size in bytes
168 *
169 *	cf. RFC 4342, 5.3 and  RFC 3448, 4.1
170 */
171static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hc, int len)
172{
173	const u16 old_s = hc->tx_s;
174
175	hc->tx_s = tfrc_ewma(hc->tx_s, len, 9);
176
177	if (hc->tx_s != old_s)
178		ccid3_update_send_interval(hc);
179}
180
181/*
182 *	Update Window Counter using the algorithm from [RFC 4342, 8.1].
183 *	As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
184 */
185static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hc,
186						ktime_t now)
187{
188	u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count),
189	    quarter_rtts = (4 * delta) / hc->tx_rtt;
190
191	if (quarter_rtts > 0) {
192		hc->tx_t_last_win_count = now;
193		hc->tx_last_win_count  += min(quarter_rtts, 5U);
194		hc->tx_last_win_count  &= 0xF;		/* mod 16 */
195	}
196}
197
198static void ccid3_hc_tx_no_feedback_timer(struct timer_list *t)
199{
200	struct ccid3_hc_tx_sock *hc = from_timer(hc, t, tx_no_feedback_timer);
201	struct sock *sk = hc->sk;
202	unsigned long t_nfb = USEC_PER_SEC / 5;
203
204	bh_lock_sock(sk);
205	if (sock_owned_by_user(sk)) {
206		/* Try again later. */
207		/* XXX: set some sensible MIB */
208		goto restart_timer;
209	}
210
211	ccid3_pr_debug("%s(%p, state=%s) - entry\n", dccp_role(sk), sk,
212		       ccid3_tx_state_name(hc->tx_state));
213
214	/* Ignore and do not restart after leaving the established state */
215	if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
216		goto out;
217
218	/* Reset feedback state to "no feedback received" */
219	if (hc->tx_state == TFRC_SSTATE_FBACK)
220		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
221
222	/*
223	 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
224	 * RTO is 0 if and only if no feedback has been received yet.
225	 */
226	if (hc->tx_t_rto == 0 || hc->tx_p == 0) {
227
228		/* halve send rate directly */
229		hc->tx_x = max(hc->tx_x / 2,
230			       (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
231		ccid3_update_send_interval(hc);
232	} else {
233		/*
234		 *  Modify the cached value of X_recv
235		 *
236		 *  If (X_calc > 2 * X_recv)
237		 *    X_recv = max(X_recv / 2, s / (2 * t_mbi));
238		 *  Else
239		 *    X_recv = X_calc / 4;
240		 *
241		 *  Note that X_recv is scaled by 2^6 while X_calc is not
242		 */
 
 
243		if (hc->tx_x_calc > (hc->tx_x_recv >> 5))
244			hc->tx_x_recv =
245				max(hc->tx_x_recv / 2,
246				    (((__u64)hc->tx_s) << 6) / (2*TFRC_T_MBI));
247		else {
248			hc->tx_x_recv = hc->tx_x_calc;
249			hc->tx_x_recv <<= 4;
250		}
251		ccid3_hc_tx_update_x(sk, NULL);
252	}
253	ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
254			(unsigned long long)hc->tx_x);
255
256	/*
257	 * Set new timeout for the nofeedback timer.
258	 * See comments in packet_recv() regarding the value of t_RTO.
259	 */
260	if (unlikely(hc->tx_t_rto == 0))	/* no feedback received yet */
261		t_nfb = TFRC_INITIAL_TIMEOUT;
262	else
263		t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
264
265restart_timer:
266	sk_reset_timer(sk, &hc->tx_no_feedback_timer,
267			   jiffies + usecs_to_jiffies(t_nfb));
268out:
269	bh_unlock_sock(sk);
270	sock_put(sk);
271}
272
273/**
274 * ccid3_hc_tx_send_packet  -  Delay-based dequeueing of TX packets
275 * @skb: next packet candidate to send on @sk
276 *
277 * This function uses the convention of ccid_packet_dequeue_eval() and
278 * returns a millisecond-delay value between 0 and t_mbi = 64000 msec.
279 */
280static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
281{
282	struct dccp_sock *dp = dccp_sk(sk);
283	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
284	ktime_t now = ktime_get_real();
285	s64 delay;
286
287	/*
288	 * This function is called only for Data and DataAck packets. Sending
289	 * zero-sized Data(Ack)s is theoretically possible, but for congestion
290	 * control this case is pathological - ignore it.
291	 */
292	if (unlikely(skb->len == 0))
293		return -EBADMSG;
294
295	if (hc->tx_state == TFRC_SSTATE_NO_SENT) {
296		sk_reset_timer(sk, &hc->tx_no_feedback_timer, (jiffies +
297			       usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
298		hc->tx_last_win_count	= 0;
299		hc->tx_t_last_win_count = now;
300
301		/* Set t_0 for initial packet */
302		hc->tx_t_nom = now;
303
304		hc->tx_s = skb->len;
305
306		/*
307		 * Use initial RTT sample when available: recommended by erratum
308		 * to RFC 4342. This implements the initialisation procedure of
309		 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
310		 */
311		if (dp->dccps_syn_rtt) {
312			ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
313			hc->tx_rtt  = dp->dccps_syn_rtt;
314			hc->tx_x    = rfc3390_initial_rate(sk);
315			hc->tx_t_ld = now;
316		} else {
317			/*
318			 * Sender does not have RTT sample:
319			 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
320			 *   is needed in several parts (e.g.  window counter);
321			 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
322			 */
323			hc->tx_rtt = DCCP_FALLBACK_RTT;
324			hc->tx_x   = hc->tx_s;
325			hc->tx_x <<= 6;
326		}
327		ccid3_update_send_interval(hc);
328
329		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
330
331	} else {
332		delay = ktime_us_delta(hc->tx_t_nom, now);
333		ccid3_pr_debug("delay=%ld\n", (long)delay);
334		/*
335		 *	Scheduling of packet transmissions (RFC 5348, 8.3)
336		 *
337		 * if (t_now > t_nom - delta)
338		 *       // send the packet now
339		 * else
340		 *       // send the packet in (t_nom - t_now) milliseconds.
341		 */
342		if (delay >= TFRC_T_DELTA)
343			return (u32)delay / USEC_PER_MSEC;
344
345		ccid3_hc_tx_update_win_count(hc, now);
346	}
347
348	/* prepare to send now (add options etc.) */
349	dp->dccps_hc_tx_insert_options = 1;
350	DCCP_SKB_CB(skb)->dccpd_ccval  = hc->tx_last_win_count;
351
352	/* set the nominal send time for the next following packet */
353	hc->tx_t_nom = ktime_add_us(hc->tx_t_nom, hc->tx_t_ipi);
354	return CCID_PACKET_SEND_AT_ONCE;
355}
356
357static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len)
358{
359	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
360
361	ccid3_hc_tx_update_s(hc, len);
362
363	if (tfrc_tx_hist_add(&hc->tx_hist, dccp_sk(sk)->dccps_gss))
364		DCCP_CRIT("packet history - out of memory!");
365}
366
367static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
368{
369	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
370	struct tfrc_tx_hist_entry *acked;
371	ktime_t now;
372	unsigned long t_nfb;
373	u32 r_sample;
374
375	/* we are only interested in ACKs */
376	if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
377	      DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
378		return;
379	/*
380	 * Locate the acknowledged packet in the TX history.
381	 *
382	 * Returning "entry not found" here can for instance happen when
383	 *  - the host has not sent out anything (e.g. a passive server),
384	 *  - the Ack is outdated (packet with higher Ack number was received),
385	 *  - it is a bogus Ack (for a packet not sent on this connection).
386	 */
387	acked = tfrc_tx_hist_find_entry(hc->tx_hist, dccp_hdr_ack_seq(skb));
388	if (acked == NULL)
389		return;
390	/* For the sake of RTT sampling, ignore/remove all older entries */
391	tfrc_tx_hist_purge(&acked->next);
392
393	/* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
394	now	  = ktime_get_real();
395	r_sample  = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
396	hc->tx_rtt = tfrc_ewma(hc->tx_rtt, r_sample, 9);
397
398	/*
399	 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
400	 */
401	if (hc->tx_state == TFRC_SSTATE_NO_FBACK) {
402		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
403
404		if (hc->tx_t_rto == 0) {
405			/*
406			 * Initial feedback packet: Larger Initial Windows (4.2)
407			 */
408			hc->tx_x    = rfc3390_initial_rate(sk);
409			hc->tx_t_ld = now;
410
411			ccid3_update_send_interval(hc);
412
413			goto done_computing_x;
414		} else if (hc->tx_p == 0) {
415			/*
416			 * First feedback after nofeedback timer expiry (4.3)
417			 */
418			goto done_computing_x;
419		}
420	}
421
422	/* Update sending rate (step 4 of [RFC 3448, 4.3]) */
423	if (hc->tx_p > 0)
424		hc->tx_x_calc = tfrc_calc_x(hc->tx_s, hc->tx_rtt, hc->tx_p);
425	ccid3_hc_tx_update_x(sk, &now);
426
427done_computing_x:
428	ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
429			       "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
430			       dccp_role(sk), sk, hc->tx_rtt, r_sample,
431			       hc->tx_s, hc->tx_p, hc->tx_x_calc,
432			       (unsigned int)(hc->tx_x_recv >> 6),
433			       (unsigned int)(hc->tx_x >> 6));
434
435	/* unschedule no feedback timer */
436	sk_stop_timer(sk, &hc->tx_no_feedback_timer);
437
438	/*
439	 * As we have calculated new ipi, delta, t_nom it is possible
440	 * that we now can send a packet, so wake up dccp_wait_for_ccid
441	 */
442	sk->sk_write_space(sk);
443
444	/*
445	 * Update timeout interval for the nofeedback timer. In order to control
446	 * rate halving on networks with very low RTTs (<= 1 ms), use per-route
447	 * tunable RTAX_RTO_MIN value as the lower bound.
448	 */
449	hc->tx_t_rto = max_t(u32, 4 * hc->tx_rtt,
450				  USEC_PER_SEC/HZ * tcp_rto_min(sk));
451	/*
452	 * Schedule no feedback timer to expire in
453	 * max(t_RTO, 2 * s/X)  =  max(t_RTO, 2 * t_ipi)
454	 */
455	t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
456
457	ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
458		       "expire in %lu jiffies (%luus)\n",
459		       dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
460
461	sk_reset_timer(sk, &hc->tx_no_feedback_timer,
462			   jiffies + usecs_to_jiffies(t_nfb));
463}
464
465static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type,
466				     u8 option, u8 *optval, u8 optlen)
467{
468	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
469	__be32 opt_val;
470
471	switch (option) {
472	case TFRC_OPT_RECEIVE_RATE:
473	case TFRC_OPT_LOSS_EVENT_RATE:
474		/* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
475		if (packet_type == DCCP_PKT_DATA)
476			break;
477		if (unlikely(optlen != 4)) {
478			DCCP_WARN("%s(%p), invalid len %d for %u\n",
479				  dccp_role(sk), sk, optlen, option);
480			return -EINVAL;
481		}
482		opt_val = ntohl(get_unaligned((__be32 *)optval));
483
484		if (option == TFRC_OPT_RECEIVE_RATE) {
485			/* Receive Rate is kept in units of 64 bytes/second */
486			hc->tx_x_recv = opt_val;
487			hc->tx_x_recv <<= 6;
488
489			ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
490				       dccp_role(sk), sk, opt_val);
491		} else {
492			/* Update the fixpoint Loss Event Rate fraction */
493			hc->tx_p = tfrc_invert_loss_event_rate(opt_val);
494
495			ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
496				       dccp_role(sk), sk, opt_val);
497		}
498	}
499	return 0;
500}
501
502static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
503{
504	struct ccid3_hc_tx_sock *hc = ccid_priv(ccid);
505
506	hc->tx_state = TFRC_SSTATE_NO_SENT;
507	hc->tx_hist  = NULL;
508	hc->sk	     = sk;
509	timer_setup(&hc->tx_no_feedback_timer,
510		    ccid3_hc_tx_no_feedback_timer, 0);
511	return 0;
512}
513
514static void ccid3_hc_tx_exit(struct sock *sk)
515{
516	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
517
518	sk_stop_timer(sk, &hc->tx_no_feedback_timer);
519	tfrc_tx_hist_purge(&hc->tx_hist);
520}
521
522static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
523{
524	info->tcpi_rto = ccid3_hc_tx_sk(sk)->tx_t_rto;
525	info->tcpi_rtt = ccid3_hc_tx_sk(sk)->tx_rtt;
526}
527
528static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
529				  u32 __user *optval, int __user *optlen)
530{
531	const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
532	struct tfrc_tx_info tfrc;
533	const void *val;
534
535	switch (optname) {
536	case DCCP_SOCKOPT_CCID_TX_INFO:
537		if (len < sizeof(tfrc))
538			return -EINVAL;
539		memset(&tfrc, 0, sizeof(tfrc));
540		tfrc.tfrctx_x	   = hc->tx_x;
541		tfrc.tfrctx_x_recv = hc->tx_x_recv;
542		tfrc.tfrctx_x_calc = hc->tx_x_calc;
543		tfrc.tfrctx_rtt	   = hc->tx_rtt;
544		tfrc.tfrctx_p	   = hc->tx_p;
545		tfrc.tfrctx_rto	   = hc->tx_t_rto;
546		tfrc.tfrctx_ipi	   = hc->tx_t_ipi;
547		len = sizeof(tfrc);
548		val = &tfrc;
549		break;
550	default:
551		return -ENOPROTOOPT;
552	}
553
554	if (put_user(len, optlen) || copy_to_user(optval, val, len))
555		return -EFAULT;
556
557	return 0;
558}
559
560/*
561 *	Receiver Half-Connection Routines
562 */
563
564/* CCID3 feedback types */
565enum ccid3_fback_type {
566	CCID3_FBACK_NONE = 0,
567	CCID3_FBACK_INITIAL,
568	CCID3_FBACK_PERIODIC,
569	CCID3_FBACK_PARAM_CHANGE
570};
571
572#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
573static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
574{
575	static const char *const ccid3_rx_state_names[] = {
576	[TFRC_RSTATE_NO_DATA] = "NO_DATA",
577	[TFRC_RSTATE_DATA]    = "DATA",
578	};
579
580	return ccid3_rx_state_names[state];
581}
582#endif
583
584static void ccid3_hc_rx_set_state(struct sock *sk,
585				  enum ccid3_hc_rx_states state)
586{
587	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
588	enum ccid3_hc_rx_states oldstate = hc->rx_state;
589
590	ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
591		       dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
592		       ccid3_rx_state_name(state));
593	WARN_ON(state == oldstate);
594	hc->rx_state = state;
595}
596
597static void ccid3_hc_rx_send_feedback(struct sock *sk,
598				      const struct sk_buff *skb,
599				      enum ccid3_fback_type fbtype)
600{
601	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
602	struct dccp_sock *dp = dccp_sk(sk);
603	ktime_t now = ktime_get_real();
604	s64 delta = 0;
605
606	switch (fbtype) {
607	case CCID3_FBACK_INITIAL:
608		hc->rx_x_recv = 0;
609		hc->rx_pinv   = ~0U;   /* see RFC 4342, 8.5 */
610		break;
611	case CCID3_FBACK_PARAM_CHANGE:
612		/*
613		 * When parameters change (new loss or p > p_prev), we do not
614		 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
615		 * need to  reuse the previous value of X_recv. However, when
616		 * X_recv was 0 (due to early loss), this would kill X down to
617		 * s/t_mbi (i.e. one packet in 64 seconds).
618		 * To avoid such drastic reduction, we approximate X_recv as
619		 * the number of bytes since last feedback.
620		 * This is a safe fallback, since X is bounded above by X_calc.
621		 */
622		if (hc->rx_x_recv > 0)
623			break;
624		/* fall through */
625	case CCID3_FBACK_PERIODIC:
626		delta = ktime_us_delta(now, hc->rx_tstamp_last_feedback);
627		if (delta <= 0)
628			DCCP_BUG("delta (%ld) <= 0", (long)delta);
629		else
630			hc->rx_x_recv = scaled_div32(hc->rx_bytes_recv, delta);
631		break;
632	default:
633		return;
634	}
635
636	ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta,
637		       hc->rx_x_recv, hc->rx_pinv);
638
639	hc->rx_tstamp_last_feedback = now;
640	hc->rx_last_counter	    = dccp_hdr(skb)->dccph_ccval;
641	hc->rx_bytes_recv	    = 0;
642
643	dp->dccps_hc_rx_insert_options = 1;
644	dccp_send_ack(sk);
645}
646
647static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
648{
649	const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
650	__be32 x_recv, pinv;
651
652	if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
653		return 0;
654
655	if (dccp_packet_without_ack(skb))
656		return 0;
657
658	x_recv = htonl(hc->rx_x_recv);
659	pinv   = htonl(hc->rx_pinv);
660
661	if (dccp_insert_option(skb, TFRC_OPT_LOSS_EVENT_RATE,
662			       &pinv, sizeof(pinv)) ||
663	    dccp_insert_option(skb, TFRC_OPT_RECEIVE_RATE,
664			       &x_recv, sizeof(x_recv)))
665		return -1;
666
667	return 0;
668}
669
670/**
671 * ccid3_first_li  -  Implements [RFC 5348, 6.3.1]
672 *
673 * Determine the length of the first loss interval via inverse lookup.
674 * Assume that X_recv can be computed by the throughput equation
675 *		    s
676 *	X_recv = --------
677 *		 R * fval
678 * Find some p such that f(p) = fval; return 1/p (scaled).
679 */
680static u32 ccid3_first_li(struct sock *sk)
681{
682	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
683	u32 x_recv, p, delta;
684	u64 fval;
685
686	if (hc->rx_rtt == 0) {
687		DCCP_WARN("No RTT estimate available, using fallback RTT\n");
688		hc->rx_rtt = DCCP_FALLBACK_RTT;
689	}
690
691	delta  = ktime_to_us(net_timedelta(hc->rx_tstamp_last_feedback));
692	x_recv = scaled_div32(hc->rx_bytes_recv, delta);
693	if (x_recv == 0) {		/* would also trigger divide-by-zero */
694		DCCP_WARN("X_recv==0\n");
695		if (hc->rx_x_recv == 0) {
696			DCCP_BUG("stored value of X_recv is zero");
697			return ~0U;
698		}
699		x_recv = hc->rx_x_recv;
700	}
701
702	fval = scaled_div(hc->rx_s, hc->rx_rtt);
703	fval = scaled_div32(fval, x_recv);
704	p = tfrc_calc_x_reverse_lookup(fval);
705
706	ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
707		       "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
708
709	return p == 0 ? ~0U : scaled_div(1, p);
710}
711
712static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
713{
714	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
715	enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
716	const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
717	const bool is_data_packet = dccp_data_packet(skb);
718
719	if (unlikely(hc->rx_state == TFRC_RSTATE_NO_DATA)) {
720		if (is_data_packet) {
721			const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
722			do_feedback = CCID3_FBACK_INITIAL;
723			ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
724			hc->rx_s = payload;
725			/*
726			 * Not necessary to update rx_bytes_recv here,
727			 * since X_recv = 0 for the first feedback packet (cf.
728			 * RFC 3448, 6.3) -- gerrit
729			 */
730		}
731		goto update_records;
732	}
733
734	if (tfrc_rx_hist_duplicate(&hc->rx_hist, skb))
735		return; /* done receiving */
736
737	if (is_data_packet) {
738		const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
739		/*
740		 * Update moving-average of s and the sum of received payload bytes
741		 */
742		hc->rx_s = tfrc_ewma(hc->rx_s, payload, 9);
743		hc->rx_bytes_recv += payload;
744	}
745
746	/*
747	 * Perform loss detection and handle pending losses
748	 */
749	if (tfrc_rx_handle_loss(&hc->rx_hist, &hc->rx_li_hist,
750				skb, ndp, ccid3_first_li, sk)) {
751		do_feedback = CCID3_FBACK_PARAM_CHANGE;
752		goto done_receiving;
753	}
754
755	if (tfrc_rx_hist_loss_pending(&hc->rx_hist))
756		return; /* done receiving */
757
758	/*
759	 * Handle data packets: RTT sampling and monitoring p
760	 */
761	if (unlikely(!is_data_packet))
762		goto update_records;
763
764	if (!tfrc_lh_is_initialised(&hc->rx_li_hist)) {
765		const u32 sample = tfrc_rx_hist_sample_rtt(&hc->rx_hist, skb);
766		/*
767		 * Empty loss history: no loss so far, hence p stays 0.
768		 * Sample RTT values, since an RTT estimate is required for the
769		 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
770		 */
771		if (sample != 0)
772			hc->rx_rtt = tfrc_ewma(hc->rx_rtt, sample, 9);
773
774	} else if (tfrc_lh_update_i_mean(&hc->rx_li_hist, skb)) {
775		/*
776		 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
777		 * has decreased (resp. p has increased), send feedback now.
778		 */
779		do_feedback = CCID3_FBACK_PARAM_CHANGE;
780	}
781
782	/*
783	 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
784	 */
785	if (SUB16(dccp_hdr(skb)->dccph_ccval, hc->rx_last_counter) > 3)
786		do_feedback = CCID3_FBACK_PERIODIC;
787
788update_records:
789	tfrc_rx_hist_add_packet(&hc->rx_hist, skb, ndp);
790
791done_receiving:
792	if (do_feedback)
793		ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
794}
795
796static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
797{
798	struct ccid3_hc_rx_sock *hc = ccid_priv(ccid);
799
800	hc->rx_state = TFRC_RSTATE_NO_DATA;
801	tfrc_lh_init(&hc->rx_li_hist);
802	return tfrc_rx_hist_alloc(&hc->rx_hist);
803}
804
805static void ccid3_hc_rx_exit(struct sock *sk)
806{
807	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
808
809	tfrc_rx_hist_purge(&hc->rx_hist);
810	tfrc_lh_cleanup(&hc->rx_li_hist);
811}
812
813static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
814{
815	info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->rx_state;
816	info->tcpi_options  |= TCPI_OPT_TIMESTAMPS;
817	info->tcpi_rcv_rtt  = ccid3_hc_rx_sk(sk)->rx_rtt;
818}
819
820static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
821				  u32 __user *optval, int __user *optlen)
822{
823	const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
824	struct tfrc_rx_info rx_info;
825	const void *val;
826
827	switch (optname) {
828	case DCCP_SOCKOPT_CCID_RX_INFO:
829		if (len < sizeof(rx_info))
830			return -EINVAL;
831		rx_info.tfrcrx_x_recv = hc->rx_x_recv;
832		rx_info.tfrcrx_rtt    = hc->rx_rtt;
833		rx_info.tfrcrx_p      = tfrc_invert_loss_event_rate(hc->rx_pinv);
834		len = sizeof(rx_info);
835		val = &rx_info;
836		break;
837	default:
838		return -ENOPROTOOPT;
839	}
840
841	if (put_user(len, optlen) || copy_to_user(optval, val, len))
842		return -EFAULT;
843
844	return 0;
845}
846
847struct ccid_operations ccid3_ops = {
848	.ccid_id		   = DCCPC_CCID3,
849	.ccid_name		   = "TCP-Friendly Rate Control",
850	.ccid_hc_tx_obj_size	   = sizeof(struct ccid3_hc_tx_sock),
851	.ccid_hc_tx_init	   = ccid3_hc_tx_init,
852	.ccid_hc_tx_exit	   = ccid3_hc_tx_exit,
853	.ccid_hc_tx_send_packet	   = ccid3_hc_tx_send_packet,
854	.ccid_hc_tx_packet_sent	   = ccid3_hc_tx_packet_sent,
855	.ccid_hc_tx_packet_recv	   = ccid3_hc_tx_packet_recv,
856	.ccid_hc_tx_parse_options  = ccid3_hc_tx_parse_options,
857	.ccid_hc_rx_obj_size	   = sizeof(struct ccid3_hc_rx_sock),
858	.ccid_hc_rx_init	   = ccid3_hc_rx_init,
859	.ccid_hc_rx_exit	   = ccid3_hc_rx_exit,
860	.ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
861	.ccid_hc_rx_packet_recv	   = ccid3_hc_rx_packet_recv,
862	.ccid_hc_rx_get_info	   = ccid3_hc_rx_get_info,
863	.ccid_hc_tx_get_info	   = ccid3_hc_tx_get_info,
864	.ccid_hc_rx_getsockopt	   = ccid3_hc_rx_getsockopt,
865	.ccid_hc_tx_getsockopt	   = ccid3_hc_tx_getsockopt,
866};
867
868#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
869module_param(ccid3_debug, bool, 0644);
870MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages");
871#endif