xref: /openbmc/linux/net/dccp/ccids/ccid3.c (revision f42b3800)
1 /*
2  *  net/dccp/ccids/ccid3.c
3  *
4  *  Copyright (c) 2007   The University of Aberdeen, Scotland, UK
5  *  Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
6  *  Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
7  *
8  *  An implementation of the DCCP protocol
9  *
10  *  This code has been developed by the University of Waikato WAND
11  *  research group. For further information please see http://www.wand.net.nz/
12  *
13  *  This code also uses code from Lulea University, rereleased as GPL by its
14  *  authors:
15  *  Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
16  *
17  *  Changes to meet Linux coding standards, to make it meet latest ccid3 draft
18  *  and to make it work as a loadable module in the DCCP stack written by
19  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
20  *
21  *  Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
22  *
23  *  This program is free software; you can redistribute it and/or modify
24  *  it under the terms of the GNU General Public License as published by
25  *  the Free Software Foundation; either version 2 of the License, or
26  *  (at your option) any later version.
27  *
28  *  This program is distributed in the hope that it will be useful,
29  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
30  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31  *  GNU General Public License for more details.
32  *
33  *  You should have received a copy of the GNU General Public License
34  *  along with this program; if not, write to the Free Software
35  *  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
36  */
37 #include "../dccp.h"
38 #include "ccid3.h"
39 
40 #include <asm/unaligned.h>
41 
42 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
43 static int ccid3_debug;
44 #define ccid3_pr_debug(format, a...)	DCCP_PR_DEBUG(ccid3_debug, format, ##a)
45 #else
46 #define ccid3_pr_debug(format, a...)
47 #endif
48 
49 /*
50  *	Transmitter Half-Connection Routines
51  */
52 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
53 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
54 {
55 	static char *ccid3_state_names[] = {
56 	[TFRC_SSTATE_NO_SENT]  = "NO_SENT",
57 	[TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
58 	[TFRC_SSTATE_FBACK]    = "FBACK",
59 	[TFRC_SSTATE_TERM]     = "TERM",
60 	};
61 
62 	return ccid3_state_names[state];
63 }
64 #endif
65 
66 static void ccid3_hc_tx_set_state(struct sock *sk,
67 				  enum ccid3_hc_tx_states state)
68 {
69 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
70 	enum ccid3_hc_tx_states oldstate = hctx->ccid3hctx_state;
71 
72 	ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
73 		       dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
74 		       ccid3_tx_state_name(state));
75 	WARN_ON(state == oldstate);
76 	hctx->ccid3hctx_state = state;
77 }
78 
79 /*
80  * Compute the initial sending rate X_init in the manner of RFC 3390:
81  *
82  *	X_init  =  min(4 * s, max(2 * s, 4380 bytes)) / RTT
83  *
84  * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
85  * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
86  * For consistency with other parts of the code, X_init is scaled by 2^6.
87  */
88 static inline u64 rfc3390_initial_rate(struct sock *sk)
89 {
90 	const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
91 	const __u32 w_init = min_t(__u32, 4 * hctx->ccid3hctx_s,
92 				   max_t(__u32, 2 * hctx->ccid3hctx_s, 4380));
93 
94 	return scaled_div(w_init << 6, hctx->ccid3hctx_rtt);
95 }
96 
97 /*
98  * Recalculate t_ipi and delta (should be called whenever X changes)
99  */
100 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
101 {
102 	/* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
103 	hctx->ccid3hctx_t_ipi = scaled_div32(((u64)hctx->ccid3hctx_s) << 6,
104 					     hctx->ccid3hctx_x);
105 
106 	/* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
107 	hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
108 					   TFRC_OPSYS_HALF_TIME_GRAN);
109 
110 	ccid3_pr_debug("t_ipi=%u, delta=%u, s=%u, X=%u\n",
111 		       hctx->ccid3hctx_t_ipi, hctx->ccid3hctx_delta,
112 		       hctx->ccid3hctx_s, (unsigned)(hctx->ccid3hctx_x >> 6));
113 
114 }
115 
116 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
117 {
118 	u32 delta = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count);
119 
120 	return delta / hctx->ccid3hctx_rtt;
121 }
122 
123 /**
124  * ccid3_hc_tx_update_x  -  Update allowed sending rate X
125  * @stamp: most recent time if available - can be left NULL.
126  * This function tracks draft rfc3448bis, check there for latest details.
127  *
128  * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
129  *       fine-grained resolution of sending rates. This requires scaling by 2^6
130  *       throughout the code. Only X_calc is unscaled (in bytes/second).
131  *
132  */
133 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
134 {
135 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
136 	__u64 min_rate = 2 * hctx->ccid3hctx_x_recv;
137 	const  __u64 old_x = hctx->ccid3hctx_x;
138 	ktime_t now = stamp ? *stamp : ktime_get_real();
139 
140 	/*
141 	 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
142 	 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
143 	 * a sender is idle if it has not sent anything over a 2-RTT-period.
144 	 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
145 	 */
146 	if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) {
147 		min_rate = rfc3390_initial_rate(sk);
148 		min_rate = max(min_rate, 2 * hctx->ccid3hctx_x_recv);
149 	}
150 
151 	if (hctx->ccid3hctx_p > 0) {
152 
153 		hctx->ccid3hctx_x = min(((__u64)hctx->ccid3hctx_x_calc) << 6,
154 					min_rate);
155 		hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
156 					(((__u64)hctx->ccid3hctx_s) << 6) /
157 								TFRC_T_MBI);
158 
159 	} else if (ktime_us_delta(now, hctx->ccid3hctx_t_ld)
160 				- (s64)hctx->ccid3hctx_rtt >= 0) {
161 
162 		hctx->ccid3hctx_x =
163 			max(min(2 * hctx->ccid3hctx_x, min_rate),
164 			    scaled_div(((__u64)hctx->ccid3hctx_s) << 6,
165 				       hctx->ccid3hctx_rtt));
166 		hctx->ccid3hctx_t_ld = now;
167 	}
168 
169 	if (hctx->ccid3hctx_x != old_x) {
170 		ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
171 			       "X_recv=%u\n", (unsigned)(old_x >> 6),
172 			       (unsigned)(hctx->ccid3hctx_x >> 6),
173 			       hctx->ccid3hctx_x_calc,
174 			       (unsigned)(hctx->ccid3hctx_x_recv >> 6));
175 
176 		ccid3_update_send_interval(hctx);
177 	}
178 }
179 
180 /*
181  *	Track the mean packet size `s' (cf. RFC 4342, 5.3 and  RFC 3448, 4.1)
182  *	@len: DCCP packet payload size in bytes
183  */
184 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
185 {
186 	const u16 old_s = hctx->ccid3hctx_s;
187 
188 	hctx->ccid3hctx_s = tfrc_ewma(hctx->ccid3hctx_s, len, 9);
189 
190 	if (hctx->ccid3hctx_s != old_s)
191 		ccid3_update_send_interval(hctx);
192 }
193 
194 /*
195  *	Update Window Counter using the algorithm from [RFC 4342, 8.1].
196  *	The algorithm is not applicable if RTT < 4 microseconds.
197  */
198 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
199 						ktime_t now)
200 {
201 	u32 quarter_rtts;
202 
203 	if (unlikely(hctx->ccid3hctx_rtt < 4))	/* avoid divide-by-zero */
204 		return;
205 
206 	quarter_rtts = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count);
207 	quarter_rtts /= hctx->ccid3hctx_rtt / 4;
208 
209 	if (quarter_rtts > 0) {
210 		hctx->ccid3hctx_t_last_win_count = now;
211 		hctx->ccid3hctx_last_win_count	+= min_t(u32, quarter_rtts, 5);
212 		hctx->ccid3hctx_last_win_count	&= 0xF;		/* mod 16 */
213 	}
214 }
215 
216 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
217 {
218 	struct sock *sk = (struct sock *)data;
219 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
220 	unsigned long t_nfb = USEC_PER_SEC / 5;
221 
222 	bh_lock_sock(sk);
223 	if (sock_owned_by_user(sk)) {
224 		/* Try again later. */
225 		/* XXX: set some sensible MIB */
226 		goto restart_timer;
227 	}
228 
229 	ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
230 		       ccid3_tx_state_name(hctx->ccid3hctx_state));
231 
232 	if (hctx->ccid3hctx_state == TFRC_SSTATE_FBACK)
233 		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
234 	else if (hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
235 		goto out;
236 
237 	/*
238 	 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
239 	 */
240 	if (hctx->ccid3hctx_t_rto == 0 ||	/* no feedback received yet */
241 	    hctx->ccid3hctx_p == 0) {
242 
243 		/* halve send rate directly */
244 		hctx->ccid3hctx_x = max(hctx->ccid3hctx_x / 2,
245 					(((__u64)hctx->ccid3hctx_s) << 6) /
246 								    TFRC_T_MBI);
247 		ccid3_update_send_interval(hctx);
248 	} else {
249 		/*
250 		 *  Modify the cached value of X_recv
251 		 *
252 		 *  If (X_calc > 2 * X_recv)
253 		 *    X_recv = max(X_recv / 2, s / (2 * t_mbi));
254 		 *  Else
255 		 *    X_recv = X_calc / 4;
256 		 *
257 		 *  Note that X_recv is scaled by 2^6 while X_calc is not
258 		 */
259 		BUG_ON(hctx->ccid3hctx_p && !hctx->ccid3hctx_x_calc);
260 
261 		if (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) /
265 							      (2 * TFRC_T_MBI));
266 		else {
267 			hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc;
268 			hctx->ccid3hctx_x_recv <<= 4;
269 		}
270 		ccid3_hc_tx_update_x(sk, NULL);
271 	}
272 	ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
273 			(unsigned long long)hctx->ccid3hctx_x);
274 
275 	/*
276 	 * Set new timeout for the nofeedback timer.
277 	 * See comments in packet_recv() regarding the value of t_RTO.
278 	 */
279 	if (unlikely(hctx->ccid3hctx_t_rto == 0))	/* no feedback yet */
280 		t_nfb = TFRC_INITIAL_TIMEOUT;
281 	else
282 		t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
283 
284 restart_timer:
285 	sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
286 			   jiffies + usecs_to_jiffies(t_nfb));
287 out:
288 	bh_unlock_sock(sk);
289 	sock_put(sk);
290 }
291 
292 /*
293  * returns
294  *   > 0: delay (in msecs) that should pass before actually sending
295  *   = 0: can send immediately
296  *   < 0: error condition; do not send packet
297  */
298 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
299 {
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();
303 	s64 delay;
304 
305 	/*
306 	 * This function is called only for Data and DataAck packets. Sending
307 	 * zero-sized Data(Ack)s is theoretically possible, but for congestion
308 	 * control this case is pathological - ignore it.
309 	 */
310 	if (unlikely(skb->len == 0))
311 		return -EBADMSG;
312 
313 	switch (hctx->ccid3hctx_state) {
314 	case TFRC_SSTATE_NO_SENT:
315 		sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
316 			       (jiffies +
317 				usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
318 		hctx->ccid3hctx_last_win_count	 = 0;
319 		hctx->ccid3hctx_t_last_win_count = now;
320 
321 		/* Set t_0 for initial packet */
322 		hctx->ccid3hctx_t_nom = now;
323 
324 		hctx->ccid3hctx_s = skb->len;
325 
326 		/*
327 		 * Use initial RTT sample when available: recommended by erratum
328 		 * to RFC 4342. This implements the initialisation procedure of
329 		 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
330 		 */
331 		if (dp->dccps_syn_rtt) {
332 			ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
333 			hctx->ccid3hctx_rtt  = dp->dccps_syn_rtt;
334 			hctx->ccid3hctx_x    = rfc3390_initial_rate(sk);
335 			hctx->ccid3hctx_t_ld = now;
336 		} else {
337 			/* Sender does not have RTT sample: X_pps = 1 pkt/sec */
338 			hctx->ccid3hctx_x = hctx->ccid3hctx_s;
339 			hctx->ccid3hctx_x <<= 6;
340 		}
341 		ccid3_update_send_interval(hctx);
342 
343 		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
344 		break;
345 	case TFRC_SSTATE_NO_FBACK:
346 	case TFRC_SSTATE_FBACK:
347 		delay = ktime_us_delta(hctx->ccid3hctx_t_nom, now);
348 		ccid3_pr_debug("delay=%ld\n", (long)delay);
349 		/*
350 		 *	Scheduling of packet transmissions [RFC 3448, 4.6]
351 		 *
352 		 * if (t_now > t_nom - delta)
353 		 *       // send the packet now
354 		 * else
355 		 *       // send the packet in (t_nom - t_now) milliseconds.
356 		 */
357 		if (delay - (s64)hctx->ccid3hctx_delta >= 1000)
358 			return (u32)delay / 1000L;
359 
360 		ccid3_hc_tx_update_win_count(hctx, now);
361 		break;
362 	case TFRC_SSTATE_TERM:
363 		DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
364 		return -EINVAL;
365 	}
366 
367 	/* prepare to send now (add options etc.) */
368 	dp->dccps_hc_tx_insert_options = 1;
369 	DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
370 
371 	/* set the nominal send time for the next following packet */
372 	hctx->ccid3hctx_t_nom = ktime_add_us(hctx->ccid3hctx_t_nom,
373 					     hctx->ccid3hctx_t_ipi);
374 	return 0;
375 }
376 
377 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
378 				    unsigned int len)
379 {
380 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
381 
382 	ccid3_hc_tx_update_s(hctx, len);
383 
384 	if (tfrc_tx_hist_add(&hctx->ccid3hctx_hist, dccp_sk(sk)->dccps_gss))
385 		DCCP_CRIT("packet history - out of memory!");
386 }
387 
388 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
389 {
390 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
391 	struct ccid3_options_received *opt_recv;
392 	ktime_t now;
393 	unsigned long t_nfb;
394 	u32 pinv, r_sample;
395 
396 	/* we are only interested in ACKs */
397 	if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
398 	      DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
399 		return;
400 	/* ... and only in the established state */
401 	if (hctx->ccid3hctx_state != TFRC_SSTATE_FBACK &&
402 	    hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
403 		return;
404 
405 	opt_recv = &hctx->ccid3hctx_options_received;
406 	now = ktime_get_real();
407 
408 	/* Estimate RTT from history if ACK number is valid */
409 	r_sample = tfrc_tx_hist_rtt(hctx->ccid3hctx_hist,
410 				    DCCP_SKB_CB(skb)->dccpd_ack_seq, now);
411 	if (r_sample == 0) {
412 		DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk), sk,
413 			  dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type),
414 			  (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq);
415 		return;
416 	}
417 
418 	/* Update receive rate in units of 64 * bytes/second */
419 	hctx->ccid3hctx_x_recv = opt_recv->ccid3or_receive_rate;
420 	hctx->ccid3hctx_x_recv <<= 6;
421 
422 	/* Update loss event rate (which is scaled by 1e6) */
423 	pinv = opt_recv->ccid3or_loss_event_rate;
424 	if (pinv == ~0U || pinv == 0)	       /* see RFC 4342, 8.5   */
425 		hctx->ccid3hctx_p = 0;
426 	else				       /* can not exceed 100% */
427 		hctx->ccid3hctx_p = scaled_div(1, pinv);
428 	/*
429 	 * Validate new RTT sample and update moving average
430 	 */
431 	r_sample = dccp_sample_rtt(sk, r_sample);
432 	hctx->ccid3hctx_rtt = tfrc_ewma(hctx->ccid3hctx_rtt, r_sample, 9);
433 	/*
434 	 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
435 	 */
436 	if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
437 		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
438 
439 		if (hctx->ccid3hctx_t_rto == 0) {
440 			/*
441 			 * Initial feedback packet: Larger Initial Windows (4.2)
442 			 */
443 			hctx->ccid3hctx_x    = rfc3390_initial_rate(sk);
444 			hctx->ccid3hctx_t_ld = now;
445 
446 			ccid3_update_send_interval(hctx);
447 
448 			goto done_computing_x;
449 		} else if (hctx->ccid3hctx_p == 0) {
450 			/*
451 			 * First feedback after nofeedback timer expiry (4.3)
452 			 */
453 			goto done_computing_x;
454 		}
455 	}
456 
457 	/* Update sending rate (step 4 of [RFC 3448, 4.3]) */
458 	if (hctx->ccid3hctx_p > 0)
459 		hctx->ccid3hctx_x_calc =
460 				tfrc_calc_x(hctx->ccid3hctx_s,
461 					    hctx->ccid3hctx_rtt,
462 					    hctx->ccid3hctx_p);
463 	ccid3_hc_tx_update_x(sk, &now);
464 
465 done_computing_x:
466 	ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
467 			       "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
468 			       dccp_role(sk),
469 			       sk, hctx->ccid3hctx_rtt, r_sample,
470 			       hctx->ccid3hctx_s, hctx->ccid3hctx_p,
471 			       hctx->ccid3hctx_x_calc,
472 			       (unsigned)(hctx->ccid3hctx_x_recv >> 6),
473 			       (unsigned)(hctx->ccid3hctx_x >> 6));
474 
475 	/* unschedule no feedback timer */
476 	sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
477 
478 	/*
479 	 * As we have calculated new ipi, delta, t_nom it is possible
480 	 * that we now can send a packet, so wake up dccp_wait_for_ccid
481 	 */
482 	sk->sk_write_space(sk);
483 
484 	/*
485 	 * Update timeout interval for the nofeedback timer.
486 	 * We use a configuration option to increase the lower bound.
487 	 * This can help avoid triggering the nofeedback timer too
488 	 * often ('spinning') on LANs with small RTTs.
489 	 */
490 	hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
491 					   (CONFIG_IP_DCCP_CCID3_RTO *
492 					    (USEC_PER_SEC / 1000)));
493 	/*
494 	 * Schedule no feedback timer to expire in
495 	 * max(t_RTO, 2 * s/X)  =  max(t_RTO, 2 * t_ipi)
496 	 */
497 	t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
498 
499 	ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
500 		       "expire in %lu jiffies (%luus)\n",
501 		       dccp_role(sk),
502 		       sk, usecs_to_jiffies(t_nfb), t_nfb);
503 
504 	sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
505 			   jiffies + usecs_to_jiffies(t_nfb));
506 }
507 
508 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
509 				     unsigned char len, u16 idx,
510 				     unsigned char *value)
511 {
512 	int rc = 0;
513 	const struct dccp_sock *dp = dccp_sk(sk);
514 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
515 	struct ccid3_options_received *opt_recv;
516 	__be32 opt_val;
517 
518 	opt_recv = &hctx->ccid3hctx_options_received;
519 
520 	if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
521 		opt_recv->ccid3or_seqno		     = dp->dccps_gsr;
522 		opt_recv->ccid3or_loss_event_rate    = ~0;
523 		opt_recv->ccid3or_loss_intervals_idx = 0;
524 		opt_recv->ccid3or_loss_intervals_len = 0;
525 		opt_recv->ccid3or_receive_rate	     = 0;
526 	}
527 
528 	switch (option) {
529 	case TFRC_OPT_LOSS_EVENT_RATE:
530 		if (unlikely(len != 4)) {
531 			DCCP_WARN("%s(%p), invalid len %d "
532 				  "for TFRC_OPT_LOSS_EVENT_RATE\n",
533 				  dccp_role(sk), sk, len);
534 			rc = -EINVAL;
535 		} else {
536 			opt_val = get_unaligned((__be32 *)value);
537 			opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
538 			ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
539 				       dccp_role(sk), sk,
540 				       opt_recv->ccid3or_loss_event_rate);
541 		}
542 		break;
543 	case TFRC_OPT_LOSS_INTERVALS:
544 		opt_recv->ccid3or_loss_intervals_idx = idx;
545 		opt_recv->ccid3or_loss_intervals_len = len;
546 		ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
547 			       dccp_role(sk), sk,
548 			       opt_recv->ccid3or_loss_intervals_idx,
549 			       opt_recv->ccid3or_loss_intervals_len);
550 		break;
551 	case TFRC_OPT_RECEIVE_RATE:
552 		if (unlikely(len != 4)) {
553 			DCCP_WARN("%s(%p), invalid len %d "
554 				  "for TFRC_OPT_RECEIVE_RATE\n",
555 				  dccp_role(sk), sk, len);
556 			rc = -EINVAL;
557 		} else {
558 			opt_val = get_unaligned((__be32 *)value);
559 			opt_recv->ccid3or_receive_rate = ntohl(opt_val);
560 			ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
561 				       dccp_role(sk), sk,
562 				       opt_recv->ccid3or_receive_rate);
563 		}
564 		break;
565 	}
566 
567 	return rc;
568 }
569 
570 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
571 {
572 	struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
573 
574 	hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
575 	hctx->ccid3hctx_hist = NULL;
576 	setup_timer(&hctx->ccid3hctx_no_feedback_timer,
577 			ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
578 
579 	return 0;
580 }
581 
582 static void ccid3_hc_tx_exit(struct sock *sk)
583 {
584 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
585 
586 	ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
587 	sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
588 
589 	tfrc_tx_hist_purge(&hctx->ccid3hctx_hist);
590 }
591 
592 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
593 {
594 	struct ccid3_hc_tx_sock *hctx;
595 
596 	/* Listen socks doesn't have a private CCID block */
597 	if (sk->sk_state == DCCP_LISTEN)
598 		return;
599 
600 	hctx = ccid3_hc_tx_sk(sk);
601 	info->tcpi_rto = hctx->ccid3hctx_t_rto;
602 	info->tcpi_rtt = hctx->ccid3hctx_rtt;
603 }
604 
605 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
606 				  u32 __user *optval, int __user *optlen)
607 {
608 	const struct ccid3_hc_tx_sock *hctx;
609 	const void *val;
610 
611 	/* Listen socks doesn't have a private CCID block */
612 	if (sk->sk_state == DCCP_LISTEN)
613 		return -EINVAL;
614 
615 	hctx = ccid3_hc_tx_sk(sk);
616 	switch (optname) {
617 	case DCCP_SOCKOPT_CCID_TX_INFO:
618 		if (len < sizeof(hctx->ccid3hctx_tfrc))
619 			return -EINVAL;
620 		len = sizeof(hctx->ccid3hctx_tfrc);
621 		val = &hctx->ccid3hctx_tfrc;
622 		break;
623 	default:
624 		return -ENOPROTOOPT;
625 	}
626 
627 	if (put_user(len, optlen) || copy_to_user(optval, val, len))
628 		return -EFAULT;
629 
630 	return 0;
631 }
632 
633 /*
634  *	Receiver Half-Connection Routines
635  */
636 
637 /* CCID3 feedback types */
638 enum ccid3_fback_type {
639 	CCID3_FBACK_NONE = 0,
640 	CCID3_FBACK_INITIAL,
641 	CCID3_FBACK_PERIODIC,
642 	CCID3_FBACK_PARAM_CHANGE
643 };
644 
645 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
646 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
647 {
648 	static char *ccid3_rx_state_names[] = {
649 	[TFRC_RSTATE_NO_DATA] = "NO_DATA",
650 	[TFRC_RSTATE_DATA]    = "DATA",
651 	[TFRC_RSTATE_TERM]    = "TERM",
652 	};
653 
654 	return ccid3_rx_state_names[state];
655 }
656 #endif
657 
658 static void ccid3_hc_rx_set_state(struct sock *sk,
659 				  enum ccid3_hc_rx_states state)
660 {
661 	struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
662 	enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
663 
664 	ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
665 		       dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
666 		       ccid3_rx_state_name(state));
667 	WARN_ON(state == oldstate);
668 	hcrx->ccid3hcrx_state = state;
669 }
670 
671 static void ccid3_hc_rx_send_feedback(struct sock *sk,
672 				      const struct sk_buff *skb,
673 				      enum ccid3_fback_type fbtype)
674 {
675 	struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
676 	struct dccp_sock *dp = dccp_sk(sk);
677 	ktime_t now;
678 	s64 delta = 0;
679 
680 	if (unlikely(hcrx->ccid3hcrx_state == TFRC_RSTATE_TERM))
681 		return;
682 
683 	now = ktime_get_real();
684 
685 	switch (fbtype) {
686 	case CCID3_FBACK_INITIAL:
687 		hcrx->ccid3hcrx_x_recv = 0;
688 		hcrx->ccid3hcrx_pinv   = ~0U;   /* see RFC 4342, 8.5 */
689 		break;
690 	case CCID3_FBACK_PARAM_CHANGE:
691 		/*
692 		 * When parameters change (new loss or p > p_prev), we do not
693 		 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
694 		 * need to  reuse the previous value of X_recv. However, when
695 		 * X_recv was 0 (due to early loss), this would kill X down to
696 		 * s/t_mbi (i.e. one packet in 64 seconds).
697 		 * To avoid such drastic reduction, we approximate X_recv as
698 		 * the number of bytes since last feedback.
699 		 * This is a safe fallback, since X is bounded above by X_calc.
700 		 */
701 		if (hcrx->ccid3hcrx_x_recv > 0)
702 			break;
703 		/* fall through */
704 	case CCID3_FBACK_PERIODIC:
705 		delta = ktime_us_delta(now, hcrx->ccid3hcrx_tstamp_last_feedback);
706 		if (delta <= 0)
707 			DCCP_BUG("delta (%ld) <= 0", (long)delta);
708 		else
709 			hcrx->ccid3hcrx_x_recv =
710 				scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
711 		break;
712 	default:
713 		return;
714 	}
715 
716 	ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta,
717 		       hcrx->ccid3hcrx_x_recv, hcrx->ccid3hcrx_pinv);
718 
719 	hcrx->ccid3hcrx_tstamp_last_feedback = now;
720 	hcrx->ccid3hcrx_last_counter	     = dccp_hdr(skb)->dccph_ccval;
721 	hcrx->ccid3hcrx_bytes_recv	     = 0;
722 
723 	dp->dccps_hc_rx_insert_options = 1;
724 	dccp_send_ack(sk);
725 }
726 
727 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
728 {
729 	const struct ccid3_hc_rx_sock *hcrx;
730 	__be32 x_recv, pinv;
731 
732 	if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
733 		return 0;
734 
735 	hcrx = ccid3_hc_rx_sk(sk);
736 
737 	if (dccp_packet_without_ack(skb))
738 		return 0;
739 
740 	x_recv = htonl(hcrx->ccid3hcrx_x_recv);
741 	pinv   = htonl(hcrx->ccid3hcrx_pinv);
742 
743 	if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
744 			       &pinv, sizeof(pinv)) ||
745 	    dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
746 			       &x_recv, sizeof(x_recv)))
747 		return -1;
748 
749 	return 0;
750 }
751 
752 /** ccid3_first_li  -  Implements [RFC 3448, 6.3.1]
753  *
754  * Determine the length of the first loss interval via inverse lookup.
755  * Assume that X_recv can be computed by the throughput equation
756  *		    s
757  *	X_recv = --------
758  *		 R * fval
759  * Find some p such that f(p) = fval; return 1/p (scaled).
760  */
761 static u32 ccid3_first_li(struct sock *sk)
762 {
763 	struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
764 	u32 x_recv, p, delta;
765 	u64 fval;
766 
767 	if (hcrx->ccid3hcrx_rtt == 0) {
768 		DCCP_WARN("No RTT estimate available, using fallback RTT\n");
769 		hcrx->ccid3hcrx_rtt = DCCP_FALLBACK_RTT;
770 	}
771 
772 	delta = ktime_to_us(net_timedelta(hcrx->ccid3hcrx_tstamp_last_feedback));
773 	x_recv = scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
774 	if (x_recv == 0) {		/* would also trigger divide-by-zero */
775 		DCCP_WARN("X_recv==0\n");
776 		if ((x_recv = hcrx->ccid3hcrx_x_recv) == 0) {
777 			DCCP_BUG("stored value of X_recv is zero");
778 			return ~0U;
779 		}
780 	}
781 
782 	fval = scaled_div(hcrx->ccid3hcrx_s, hcrx->ccid3hcrx_rtt);
783 	fval = scaled_div32(fval, x_recv);
784 	p = tfrc_calc_x_reverse_lookup(fval);
785 
786 	ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
787 		       "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
788 
789 	return p == 0 ? ~0U : scaled_div(1, p);
790 }
791 
792 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
793 {
794 	struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
795 	enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
796 	const u32 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
797 	const bool is_data_packet = dccp_data_packet(skb);
798 
799 	if (unlikely(hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)) {
800 		if (is_data_packet) {
801 			const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
802 			do_feedback = CCID3_FBACK_INITIAL;
803 			ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
804 			hcrx->ccid3hcrx_s = payload;
805 			/*
806 			 * Not necessary to update ccid3hcrx_bytes_recv here,
807 			 * since X_recv = 0 for the first feedback packet (cf.
808 			 * RFC 3448, 6.3) -- gerrit
809 			 */
810 		}
811 		goto update_records;
812 	}
813 
814 	if (tfrc_rx_hist_duplicate(&hcrx->ccid3hcrx_hist, skb))
815 		return; /* done receiving */
816 
817 	if (is_data_packet) {
818 		const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
819 		/*
820 		 * Update moving-average of s and the sum of received payload bytes
821 		 */
822 		hcrx->ccid3hcrx_s = tfrc_ewma(hcrx->ccid3hcrx_s, payload, 9);
823 		hcrx->ccid3hcrx_bytes_recv += payload;
824 	}
825 
826 	/*
827 	 * Handle pending losses and otherwise check for new loss
828 	 */
829 	if (tfrc_rx_hist_loss_pending(&hcrx->ccid3hcrx_hist) &&
830 	    tfrc_rx_handle_loss(&hcrx->ccid3hcrx_hist,
831 				&hcrx->ccid3hcrx_li_hist,
832 				skb, ndp, ccid3_first_li, sk) ) {
833 		do_feedback = CCID3_FBACK_PARAM_CHANGE;
834 		goto done_receiving;
835 	}
836 
837 	if (tfrc_rx_hist_new_loss_indicated(&hcrx->ccid3hcrx_hist, skb, ndp))
838 		goto update_records;
839 
840 	/*
841 	 * Handle data packets: RTT sampling and monitoring p
842 	 */
843 	if (unlikely(!is_data_packet))
844 		goto update_records;
845 
846 	if (!tfrc_lh_is_initialised(&hcrx->ccid3hcrx_li_hist)) {
847 		const u32 sample = tfrc_rx_hist_sample_rtt(&hcrx->ccid3hcrx_hist, skb);
848 		/*
849 		 * Empty loss history: no loss so far, hence p stays 0.
850 		 * Sample RTT values, since an RTT estimate is required for the
851 		 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
852 		 */
853 		if (sample != 0)
854 			hcrx->ccid3hcrx_rtt = tfrc_ewma(hcrx->ccid3hcrx_rtt, sample, 9);
855 
856 	} else if (tfrc_lh_update_i_mean(&hcrx->ccid3hcrx_li_hist, skb)) {
857 		/*
858 		 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
859 		 * has decreased (resp. p has increased), send feedback now.
860 		 */
861 		do_feedback = CCID3_FBACK_PARAM_CHANGE;
862 	}
863 
864 	/*
865 	 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
866 	 */
867 	if (SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->ccid3hcrx_last_counter) > 3)
868 		do_feedback = CCID3_FBACK_PERIODIC;
869 
870 update_records:
871 	tfrc_rx_hist_add_packet(&hcrx->ccid3hcrx_hist, skb, ndp);
872 
873 done_receiving:
874 	if (do_feedback)
875 		ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
876 }
877 
878 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
879 {
880 	struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
881 
882 	hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
883 	tfrc_lh_init(&hcrx->ccid3hcrx_li_hist);
884 	return tfrc_rx_hist_alloc(&hcrx->ccid3hcrx_hist);
885 }
886 
887 static void ccid3_hc_rx_exit(struct sock *sk)
888 {
889 	struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
890 
891 	ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
892 
893 	tfrc_rx_hist_purge(&hcrx->ccid3hcrx_hist);
894 	tfrc_lh_cleanup(&hcrx->ccid3hcrx_li_hist);
895 }
896 
897 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
898 {
899 	const struct ccid3_hc_rx_sock *hcrx;
900 
901 	/* Listen socks doesn't have a private CCID block */
902 	if (sk->sk_state == DCCP_LISTEN)
903 		return;
904 
905 	hcrx = ccid3_hc_rx_sk(sk);
906 	info->tcpi_ca_state = hcrx->ccid3hcrx_state;
907 	info->tcpi_options  |= TCPI_OPT_TIMESTAMPS;
908 	info->tcpi_rcv_rtt  = hcrx->ccid3hcrx_rtt;
909 }
910 
911 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
912 				  u32 __user *optval, int __user *optlen)
913 {
914 	const struct ccid3_hc_rx_sock *hcrx;
915 	struct tfrc_rx_info rx_info;
916 	const void *val;
917 
918 	/* Listen socks doesn't have a private CCID block */
919 	if (sk->sk_state == DCCP_LISTEN)
920 		return -EINVAL;
921 
922 	hcrx = ccid3_hc_rx_sk(sk);
923 	switch (optname) {
924 	case DCCP_SOCKOPT_CCID_RX_INFO:
925 		if (len < sizeof(rx_info))
926 			return -EINVAL;
927 		rx_info.tfrcrx_x_recv = hcrx->ccid3hcrx_x_recv;
928 		rx_info.tfrcrx_rtt    = hcrx->ccid3hcrx_rtt;
929 		rx_info.tfrcrx_p      = hcrx->ccid3hcrx_pinv == 0 ? ~0U :
930 					   scaled_div(1, hcrx->ccid3hcrx_pinv);
931 		len = sizeof(rx_info);
932 		val = &rx_info;
933 		break;
934 	default:
935 		return -ENOPROTOOPT;
936 	}
937 
938 	if (put_user(len, optlen) || copy_to_user(optval, val, len))
939 		return -EFAULT;
940 
941 	return 0;
942 }
943 
944 static struct ccid_operations ccid3 = {
945 	.ccid_id		   = DCCPC_CCID3,
946 	.ccid_name		   = "TCP-Friendly Rate Control",
947 	.ccid_owner		   = THIS_MODULE,
948 	.ccid_hc_tx_obj_size	   = sizeof(struct ccid3_hc_tx_sock),
949 	.ccid_hc_tx_init	   = ccid3_hc_tx_init,
950 	.ccid_hc_tx_exit	   = ccid3_hc_tx_exit,
951 	.ccid_hc_tx_send_packet	   = ccid3_hc_tx_send_packet,
952 	.ccid_hc_tx_packet_sent	   = ccid3_hc_tx_packet_sent,
953 	.ccid_hc_tx_packet_recv	   = ccid3_hc_tx_packet_recv,
954 	.ccid_hc_tx_parse_options  = ccid3_hc_tx_parse_options,
955 	.ccid_hc_rx_obj_size	   = sizeof(struct ccid3_hc_rx_sock),
956 	.ccid_hc_rx_init	   = ccid3_hc_rx_init,
957 	.ccid_hc_rx_exit	   = ccid3_hc_rx_exit,
958 	.ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
959 	.ccid_hc_rx_packet_recv	   = ccid3_hc_rx_packet_recv,
960 	.ccid_hc_rx_get_info	   = ccid3_hc_rx_get_info,
961 	.ccid_hc_tx_get_info	   = ccid3_hc_tx_get_info,
962 	.ccid_hc_rx_getsockopt	   = ccid3_hc_rx_getsockopt,
963 	.ccid_hc_tx_getsockopt	   = ccid3_hc_tx_getsockopt,
964 };
965 
966 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
967 module_param(ccid3_debug, bool, 0444);
968 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
969 #endif
970 
971 static __init int ccid3_module_init(void)
972 {
973 	return ccid_register(&ccid3);
974 }
975 module_init(ccid3_module_init);
976 
977 static __exit void ccid3_module_exit(void)
978 {
979 	ccid_unregister(&ccid3);
980 }
981 module_exit(ccid3_module_exit);
982 
983 MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
984 	      "Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
985 MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
986 MODULE_LICENSE("GPL");
987 MODULE_ALIAS("net-dccp-ccid-3");
988