xref: /openbmc/linux/net/dccp/ccids/ccid3.c (revision b627b4ed)
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 = clamp_t(__u32, 4380U,
92 			2 * hctx->ccid3hctx_s, 4 * hctx->ccid3hctx_s);
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 = min(2 * hctx->ccid3hctx_x, min_rate);
163 		hctx->ccid3hctx_x = max(hctx->ccid3hctx_x,
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  *	As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
197  */
198 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
199 						ktime_t now)
200 {
201 	u32 delta = ktime_us_delta(now, hctx->ccid3hctx_t_last_win_count),
202 	    quarter_rtts = (4 * delta) / hctx->ccid3hctx_rtt;
203 
204 	if (quarter_rtts > 0) {
205 		hctx->ccid3hctx_t_last_win_count = now;
206 		hctx->ccid3hctx_last_win_count  += min(quarter_rtts, 5U);
207 		hctx->ccid3hctx_last_win_count	&= 0xF;		/* mod 16 */
208 	}
209 }
210 
211 static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
212 {
213 	struct sock *sk = (struct sock *)data;
214 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
215 	unsigned long t_nfb = USEC_PER_SEC / 5;
216 
217 	bh_lock_sock(sk);
218 	if (sock_owned_by_user(sk)) {
219 		/* Try again later. */
220 		/* XXX: set some sensible MIB */
221 		goto restart_timer;
222 	}
223 
224 	ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
225 		       ccid3_tx_state_name(hctx->ccid3hctx_state));
226 
227 	if (hctx->ccid3hctx_state == TFRC_SSTATE_FBACK)
228 		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
229 	else if (hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
230 		goto out;
231 
232 	/*
233 	 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
234 	 */
235 	if (hctx->ccid3hctx_t_rto == 0 ||	/* no feedback received yet */
236 	    hctx->ccid3hctx_p == 0) {
237 
238 		/* halve send rate directly */
239 		hctx->ccid3hctx_x = max(hctx->ccid3hctx_x / 2,
240 					(((__u64)hctx->ccid3hctx_s) << 6) /
241 								    TFRC_T_MBI);
242 		ccid3_update_send_interval(hctx);
243 	} else {
244 		/*
245 		 *  Modify the cached value of X_recv
246 		 *
247 		 *  If (X_calc > 2 * X_recv)
248 		 *    X_recv = max(X_recv / 2, s / (2 * t_mbi));
249 		 *  Else
250 		 *    X_recv = X_calc / 4;
251 		 *
252 		 *  Note that X_recv is scaled by 2^6 while X_calc is not
253 		 */
254 		BUG_ON(hctx->ccid3hctx_p && !hctx->ccid3hctx_x_calc);
255 
256 		if (hctx->ccid3hctx_x_calc > (hctx->ccid3hctx_x_recv >> 5))
257 			hctx->ccid3hctx_x_recv =
258 				max(hctx->ccid3hctx_x_recv / 2,
259 				    (((__u64)hctx->ccid3hctx_s) << 6) /
260 							      (2 * TFRC_T_MBI));
261 		else {
262 			hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc;
263 			hctx->ccid3hctx_x_recv <<= 4;
264 		}
265 		ccid3_hc_tx_update_x(sk, NULL);
266 	}
267 	ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
268 			(unsigned long long)hctx->ccid3hctx_x);
269 
270 	/*
271 	 * Set new timeout for the nofeedback timer.
272 	 * See comments in packet_recv() regarding the value of t_RTO.
273 	 */
274 	if (unlikely(hctx->ccid3hctx_t_rto == 0))	/* no feedback yet */
275 		t_nfb = TFRC_INITIAL_TIMEOUT;
276 	else
277 		t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
278 
279 restart_timer:
280 	sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
281 			   jiffies + usecs_to_jiffies(t_nfb));
282 out:
283 	bh_unlock_sock(sk);
284 	sock_put(sk);
285 }
286 
287 /*
288  * returns
289  *   > 0: delay (in msecs) that should pass before actually sending
290  *   = 0: can send immediately
291  *   < 0: error condition; do not send packet
292  */
293 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
294 {
295 	struct dccp_sock *dp = dccp_sk(sk);
296 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
297 	ktime_t now = ktime_get_real();
298 	s64 delay;
299 
300 	/*
301 	 * This function is called only for Data and DataAck packets. Sending
302 	 * zero-sized Data(Ack)s is theoretically possible, but for congestion
303 	 * control this case is pathological - ignore it.
304 	 */
305 	if (unlikely(skb->len == 0))
306 		return -EBADMSG;
307 
308 	switch (hctx->ccid3hctx_state) {
309 	case TFRC_SSTATE_NO_SENT:
310 		sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
311 			       (jiffies +
312 				usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
313 		hctx->ccid3hctx_last_win_count	 = 0;
314 		hctx->ccid3hctx_t_last_win_count = now;
315 
316 		/* Set t_0 for initial packet */
317 		hctx->ccid3hctx_t_nom = now;
318 
319 		hctx->ccid3hctx_s = skb->len;
320 
321 		/*
322 		 * Use initial RTT sample when available: recommended by erratum
323 		 * to RFC 4342. This implements the initialisation procedure of
324 		 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
325 		 */
326 		if (dp->dccps_syn_rtt) {
327 			ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
328 			hctx->ccid3hctx_rtt  = dp->dccps_syn_rtt;
329 			hctx->ccid3hctx_x    = rfc3390_initial_rate(sk);
330 			hctx->ccid3hctx_t_ld = now;
331 		} else {
332 			/*
333 			 * Sender does not have RTT sample:
334 			 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
335 			 *   is needed in several parts (e.g.  window counter);
336 			 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
337 			 */
338 			hctx->ccid3hctx_rtt = DCCP_FALLBACK_RTT;
339 			hctx->ccid3hctx_x   = hctx->ccid3hctx_s;
340 			hctx->ccid3hctx_x <<= 6;
341 		}
342 		ccid3_update_send_interval(hctx);
343 
344 		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
345 		break;
346 	case TFRC_SSTATE_NO_FBACK:
347 	case TFRC_SSTATE_FBACK:
348 		delay = ktime_us_delta(hctx->ccid3hctx_t_nom, now);
349 		ccid3_pr_debug("delay=%ld\n", (long)delay);
350 		/*
351 		 *	Scheduling of packet transmissions [RFC 3448, 4.6]
352 		 *
353 		 * if (t_now > t_nom - delta)
354 		 *       // send the packet now
355 		 * else
356 		 *       // send the packet in (t_nom - t_now) milliseconds.
357 		 */
358 		if (delay - (s64)hctx->ccid3hctx_delta >= 1000)
359 			return (u32)delay / 1000L;
360 
361 		ccid3_hc_tx_update_win_count(hctx, now);
362 		break;
363 	case TFRC_SSTATE_TERM:
364 		DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
365 		return -EINVAL;
366 	}
367 
368 	/* prepare to send now (add options etc.) */
369 	dp->dccps_hc_tx_insert_options = 1;
370 	DCCP_SKB_CB(skb)->dccpd_ccval = hctx->ccid3hctx_last_win_count;
371 
372 	/* set the nominal send time for the next following packet */
373 	hctx->ccid3hctx_t_nom = ktime_add_us(hctx->ccid3hctx_t_nom,
374 					     hctx->ccid3hctx_t_ipi);
375 	return 0;
376 }
377 
378 static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
379 				    unsigned int len)
380 {
381 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
382 
383 	ccid3_hc_tx_update_s(hctx, len);
384 
385 	if (tfrc_tx_hist_add(&hctx->ccid3hctx_hist, dccp_sk(sk)->dccps_gss))
386 		DCCP_CRIT("packet history - out of memory!");
387 }
388 
389 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
390 {
391 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
392 	struct ccid3_options_received *opt_recv;
393 	ktime_t now;
394 	unsigned long t_nfb;
395 	u32 pinv, r_sample;
396 
397 	/* we are only interested in ACKs */
398 	if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
399 	      DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
400 		return;
401 	/* ... and only in the established state */
402 	if (hctx->ccid3hctx_state != TFRC_SSTATE_FBACK &&
403 	    hctx->ccid3hctx_state != TFRC_SSTATE_NO_FBACK)
404 		return;
405 
406 	opt_recv = &hctx->ccid3hctx_options_received;
407 	now = ktime_get_real();
408 
409 	/* Estimate RTT from history if ACK number is valid */
410 	r_sample = tfrc_tx_hist_rtt(hctx->ccid3hctx_hist,
411 				    DCCP_SKB_CB(skb)->dccpd_ack_seq, now);
412 	if (r_sample == 0) {
413 		DCCP_WARN("%s(%p): %s with bogus ACK-%llu\n", dccp_role(sk), sk,
414 			  dccp_packet_name(DCCP_SKB_CB(skb)->dccpd_type),
415 			  (unsigned long long)DCCP_SKB_CB(skb)->dccpd_ack_seq);
416 		return;
417 	}
418 
419 	/* Update receive rate in units of 64 * bytes/second */
420 	hctx->ccid3hctx_x_recv = opt_recv->ccid3or_receive_rate;
421 	hctx->ccid3hctx_x_recv <<= 6;
422 
423 	/* Update loss event rate (which is scaled by 1e6) */
424 	pinv = opt_recv->ccid3or_loss_event_rate;
425 	if (pinv == ~0U || pinv == 0)	       /* see RFC 4342, 8.5   */
426 		hctx->ccid3hctx_p = 0;
427 	else				       /* can not exceed 100% */
428 		hctx->ccid3hctx_p = scaled_div(1, pinv);
429 	/*
430 	 * Validate new RTT sample and update moving average
431 	 */
432 	r_sample = dccp_sample_rtt(sk, r_sample);
433 	hctx->ccid3hctx_rtt = tfrc_ewma(hctx->ccid3hctx_rtt, r_sample, 9);
434 	/*
435 	 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
436 	 */
437 	if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) {
438 		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
439 
440 		if (hctx->ccid3hctx_t_rto == 0) {
441 			/*
442 			 * Initial feedback packet: Larger Initial Windows (4.2)
443 			 */
444 			hctx->ccid3hctx_x    = rfc3390_initial_rate(sk);
445 			hctx->ccid3hctx_t_ld = now;
446 
447 			ccid3_update_send_interval(hctx);
448 
449 			goto done_computing_x;
450 		} else if (hctx->ccid3hctx_p == 0) {
451 			/*
452 			 * First feedback after nofeedback timer expiry (4.3)
453 			 */
454 			goto done_computing_x;
455 		}
456 	}
457 
458 	/* Update sending rate (step 4 of [RFC 3448, 4.3]) */
459 	if (hctx->ccid3hctx_p > 0)
460 		hctx->ccid3hctx_x_calc =
461 				tfrc_calc_x(hctx->ccid3hctx_s,
462 					    hctx->ccid3hctx_rtt,
463 					    hctx->ccid3hctx_p);
464 	ccid3_hc_tx_update_x(sk, &now);
465 
466 done_computing_x:
467 	ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
468 			       "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
469 			       dccp_role(sk),
470 			       sk, hctx->ccid3hctx_rtt, r_sample,
471 			       hctx->ccid3hctx_s, hctx->ccid3hctx_p,
472 			       hctx->ccid3hctx_x_calc,
473 			       (unsigned)(hctx->ccid3hctx_x_recv >> 6),
474 			       (unsigned)(hctx->ccid3hctx_x >> 6));
475 
476 	/* unschedule no feedback timer */
477 	sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
478 
479 	/*
480 	 * As we have calculated new ipi, delta, t_nom it is possible
481 	 * that we now can send a packet, so wake up dccp_wait_for_ccid
482 	 */
483 	sk->sk_write_space(sk);
484 
485 	/*
486 	 * Update timeout interval for the nofeedback timer.
487 	 * We use a configuration option to increase the lower bound.
488 	 * This can help avoid triggering the nofeedback timer too
489 	 * often ('spinning') on LANs with small RTTs.
490 	 */
491 	hctx->ccid3hctx_t_rto = max_t(u32, 4 * hctx->ccid3hctx_rtt,
492 					   (CONFIG_IP_DCCP_CCID3_RTO *
493 					    (USEC_PER_SEC / 1000)));
494 	/*
495 	 * Schedule no feedback timer to expire in
496 	 * max(t_RTO, 2 * s/X)  =  max(t_RTO, 2 * t_ipi)
497 	 */
498 	t_nfb = max(hctx->ccid3hctx_t_rto, 2 * hctx->ccid3hctx_t_ipi);
499 
500 	ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
501 		       "expire in %lu jiffies (%luus)\n",
502 		       dccp_role(sk),
503 		       sk, usecs_to_jiffies(t_nfb), t_nfb);
504 
505 	sk_reset_timer(sk, &hctx->ccid3hctx_no_feedback_timer,
506 			   jiffies + usecs_to_jiffies(t_nfb));
507 }
508 
509 static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
510 				     unsigned char len, u16 idx,
511 				     unsigned char *value)
512 {
513 	int rc = 0;
514 	const struct dccp_sock *dp = dccp_sk(sk);
515 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
516 	struct ccid3_options_received *opt_recv;
517 	__be32 opt_val;
518 
519 	opt_recv = &hctx->ccid3hctx_options_received;
520 
521 	if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
522 		opt_recv->ccid3or_seqno		     = dp->dccps_gsr;
523 		opt_recv->ccid3or_loss_event_rate    = ~0;
524 		opt_recv->ccid3or_loss_intervals_idx = 0;
525 		opt_recv->ccid3or_loss_intervals_len = 0;
526 		opt_recv->ccid3or_receive_rate	     = 0;
527 	}
528 
529 	switch (option) {
530 	case TFRC_OPT_LOSS_EVENT_RATE:
531 		if (unlikely(len != 4)) {
532 			DCCP_WARN("%s(%p), invalid len %d "
533 				  "for TFRC_OPT_LOSS_EVENT_RATE\n",
534 				  dccp_role(sk), sk, len);
535 			rc = -EINVAL;
536 		} else {
537 			opt_val = get_unaligned((__be32 *)value);
538 			opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
539 			ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
540 				       dccp_role(sk), sk,
541 				       opt_recv->ccid3or_loss_event_rate);
542 		}
543 		break;
544 	case TFRC_OPT_LOSS_INTERVALS:
545 		opt_recv->ccid3or_loss_intervals_idx = idx;
546 		opt_recv->ccid3or_loss_intervals_len = len;
547 		ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
548 			       dccp_role(sk), sk,
549 			       opt_recv->ccid3or_loss_intervals_idx,
550 			       opt_recv->ccid3or_loss_intervals_len);
551 		break;
552 	case TFRC_OPT_RECEIVE_RATE:
553 		if (unlikely(len != 4)) {
554 			DCCP_WARN("%s(%p), invalid len %d "
555 				  "for TFRC_OPT_RECEIVE_RATE\n",
556 				  dccp_role(sk), sk, len);
557 			rc = -EINVAL;
558 		} else {
559 			opt_val = get_unaligned((__be32 *)value);
560 			opt_recv->ccid3or_receive_rate = ntohl(opt_val);
561 			ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
562 				       dccp_role(sk), sk,
563 				       opt_recv->ccid3or_receive_rate);
564 		}
565 		break;
566 	}
567 
568 	return rc;
569 }
570 
571 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
572 {
573 	struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
574 
575 	hctx->ccid3hctx_state = TFRC_SSTATE_NO_SENT;
576 	hctx->ccid3hctx_hist = NULL;
577 	setup_timer(&hctx->ccid3hctx_no_feedback_timer,
578 			ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
579 
580 	return 0;
581 }
582 
583 static void ccid3_hc_tx_exit(struct sock *sk)
584 {
585 	struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
586 
587 	ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
588 	sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);
589 
590 	tfrc_tx_hist_purge(&hctx->ccid3hctx_hist);
591 }
592 
593 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
594 {
595 	struct ccid3_hc_tx_sock *hctx;
596 
597 	/* Listen socks doesn't have a private CCID block */
598 	if (sk->sk_state == DCCP_LISTEN)
599 		return;
600 
601 	hctx = ccid3_hc_tx_sk(sk);
602 	info->tcpi_rto = hctx->ccid3hctx_t_rto;
603 	info->tcpi_rtt = hctx->ccid3hctx_rtt;
604 }
605 
606 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
607 				  u32 __user *optval, int __user *optlen)
608 {
609 	const struct ccid3_hc_tx_sock *hctx;
610 	const void *val;
611 
612 	/* Listen socks doesn't have a private CCID block */
613 	if (sk->sk_state == DCCP_LISTEN)
614 		return -EINVAL;
615 
616 	hctx = ccid3_hc_tx_sk(sk);
617 	switch (optname) {
618 	case DCCP_SOCKOPT_CCID_TX_INFO:
619 		if (len < sizeof(hctx->ccid3hctx_tfrc))
620 			return -EINVAL;
621 		len = sizeof(hctx->ccid3hctx_tfrc);
622 		val = &hctx->ccid3hctx_tfrc;
623 		break;
624 	default:
625 		return -ENOPROTOOPT;
626 	}
627 
628 	if (put_user(len, optlen) || copy_to_user(optval, val, len))
629 		return -EFAULT;
630 
631 	return 0;
632 }
633 
634 /*
635  *	Receiver Half-Connection Routines
636  */
637 
638 /* CCID3 feedback types */
639 enum ccid3_fback_type {
640 	CCID3_FBACK_NONE = 0,
641 	CCID3_FBACK_INITIAL,
642 	CCID3_FBACK_PERIODIC,
643 	CCID3_FBACK_PARAM_CHANGE
644 };
645 
646 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
647 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
648 {
649 	static char *ccid3_rx_state_names[] = {
650 	[TFRC_RSTATE_NO_DATA] = "NO_DATA",
651 	[TFRC_RSTATE_DATA]    = "DATA",
652 	[TFRC_RSTATE_TERM]    = "TERM",
653 	};
654 
655 	return ccid3_rx_state_names[state];
656 }
657 #endif
658 
659 static void ccid3_hc_rx_set_state(struct sock *sk,
660 				  enum ccid3_hc_rx_states state)
661 {
662 	struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
663 	enum ccid3_hc_rx_states oldstate = hcrx->ccid3hcrx_state;
664 
665 	ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
666 		       dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
667 		       ccid3_rx_state_name(state));
668 	WARN_ON(state == oldstate);
669 	hcrx->ccid3hcrx_state = state;
670 }
671 
672 static void ccid3_hc_rx_send_feedback(struct sock *sk,
673 				      const struct sk_buff *skb,
674 				      enum ccid3_fback_type fbtype)
675 {
676 	struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
677 	struct dccp_sock *dp = dccp_sk(sk);
678 	ktime_t now;
679 	s64 delta = 0;
680 
681 	if (unlikely(hcrx->ccid3hcrx_state == TFRC_RSTATE_TERM))
682 		return;
683 
684 	now = ktime_get_real();
685 
686 	switch (fbtype) {
687 	case CCID3_FBACK_INITIAL:
688 		hcrx->ccid3hcrx_x_recv = 0;
689 		hcrx->ccid3hcrx_pinv   = ~0U;   /* see RFC 4342, 8.5 */
690 		break;
691 	case CCID3_FBACK_PARAM_CHANGE:
692 		/*
693 		 * When parameters change (new loss or p > p_prev), we do not
694 		 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
695 		 * need to  reuse the previous value of X_recv. However, when
696 		 * X_recv was 0 (due to early loss), this would kill X down to
697 		 * s/t_mbi (i.e. one packet in 64 seconds).
698 		 * To avoid such drastic reduction, we approximate X_recv as
699 		 * the number of bytes since last feedback.
700 		 * This is a safe fallback, since X is bounded above by X_calc.
701 		 */
702 		if (hcrx->ccid3hcrx_x_recv > 0)
703 			break;
704 		/* fall through */
705 	case CCID3_FBACK_PERIODIC:
706 		delta = ktime_us_delta(now, hcrx->ccid3hcrx_tstamp_last_feedback);
707 		if (delta <= 0)
708 			DCCP_BUG("delta (%ld) <= 0", (long)delta);
709 		else
710 			hcrx->ccid3hcrx_x_recv =
711 				scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
712 		break;
713 	default:
714 		return;
715 	}
716 
717 	ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n", (long)delta,
718 		       hcrx->ccid3hcrx_x_recv, hcrx->ccid3hcrx_pinv);
719 
720 	hcrx->ccid3hcrx_tstamp_last_feedback = now;
721 	hcrx->ccid3hcrx_last_counter	     = dccp_hdr(skb)->dccph_ccval;
722 	hcrx->ccid3hcrx_bytes_recv	     = 0;
723 
724 	dp->dccps_hc_rx_insert_options = 1;
725 	dccp_send_ack(sk);
726 }
727 
728 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
729 {
730 	const struct ccid3_hc_rx_sock *hcrx;
731 	__be32 x_recv, pinv;
732 
733 	if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
734 		return 0;
735 
736 	hcrx = ccid3_hc_rx_sk(sk);
737 
738 	if (dccp_packet_without_ack(skb))
739 		return 0;
740 
741 	x_recv = htonl(hcrx->ccid3hcrx_x_recv);
742 	pinv   = htonl(hcrx->ccid3hcrx_pinv);
743 
744 	if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
745 			       &pinv, sizeof(pinv)) ||
746 	    dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
747 			       &x_recv, sizeof(x_recv)))
748 		return -1;
749 
750 	return 0;
751 }
752 
753 /** ccid3_first_li  -  Implements [RFC 3448, 6.3.1]
754  *
755  * Determine the length of the first loss interval via inverse lookup.
756  * Assume that X_recv can be computed by the throughput equation
757  *		    s
758  *	X_recv = --------
759  *		 R * fval
760  * Find some p such that f(p) = fval; return 1/p (scaled).
761  */
762 static u32 ccid3_first_li(struct sock *sk)
763 {
764 	struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
765 	u32 x_recv, p, delta;
766 	u64 fval;
767 
768 	if (hcrx->ccid3hcrx_rtt == 0) {
769 		DCCP_WARN("No RTT estimate available, using fallback RTT\n");
770 		hcrx->ccid3hcrx_rtt = DCCP_FALLBACK_RTT;
771 	}
772 
773 	delta = ktime_to_us(net_timedelta(hcrx->ccid3hcrx_tstamp_last_feedback));
774 	x_recv = scaled_div32(hcrx->ccid3hcrx_bytes_recv, delta);
775 	if (x_recv == 0) {		/* would also trigger divide-by-zero */
776 		DCCP_WARN("X_recv==0\n");
777 		if ((x_recv = hcrx->ccid3hcrx_x_recv) == 0) {
778 			DCCP_BUG("stored value of X_recv is zero");
779 			return ~0U;
780 		}
781 	}
782 
783 	fval = scaled_div(hcrx->ccid3hcrx_s, hcrx->ccid3hcrx_rtt);
784 	fval = scaled_div32(fval, x_recv);
785 	p = tfrc_calc_x_reverse_lookup(fval);
786 
787 	ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
788 		       "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
789 
790 	return p == 0 ? ~0U : scaled_div(1, p);
791 }
792 
793 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
794 {
795 	struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
796 	enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
797 	const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
798 	const bool is_data_packet = dccp_data_packet(skb);
799 
800 	if (unlikely(hcrx->ccid3hcrx_state == TFRC_RSTATE_NO_DATA)) {
801 		if (is_data_packet) {
802 			const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
803 			do_feedback = CCID3_FBACK_INITIAL;
804 			ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
805 			hcrx->ccid3hcrx_s = payload;
806 			/*
807 			 * Not necessary to update ccid3hcrx_bytes_recv here,
808 			 * since X_recv = 0 for the first feedback packet (cf.
809 			 * RFC 3448, 6.3) -- gerrit
810 			 */
811 		}
812 		goto update_records;
813 	}
814 
815 	if (tfrc_rx_hist_duplicate(&hcrx->ccid3hcrx_hist, skb))
816 		return; /* done receiving */
817 
818 	if (is_data_packet) {
819 		const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
820 		/*
821 		 * Update moving-average of s and the sum of received payload bytes
822 		 */
823 		hcrx->ccid3hcrx_s = tfrc_ewma(hcrx->ccid3hcrx_s, payload, 9);
824 		hcrx->ccid3hcrx_bytes_recv += payload;
825 	}
826 
827 	/*
828 	 * Perform loss detection and handle pending losses
829 	 */
830 	if (tfrc_rx_handle_loss(&hcrx->ccid3hcrx_hist, &hcrx->ccid3hcrx_li_hist,
831 				skb, ndp, ccid3_first_li, sk)) {
832 		do_feedback = CCID3_FBACK_PARAM_CHANGE;
833 		goto done_receiving;
834 	}
835 
836 	if (tfrc_rx_hist_loss_pending(&hcrx->ccid3hcrx_hist))
837 		return; /* done receiving */
838 
839 	/*
840 	 * Handle data packets: RTT sampling and monitoring p
841 	 */
842 	if (unlikely(!is_data_packet))
843 		goto update_records;
844 
845 	if (!tfrc_lh_is_initialised(&hcrx->ccid3hcrx_li_hist)) {
846 		const u32 sample = tfrc_rx_hist_sample_rtt(&hcrx->ccid3hcrx_hist, skb);
847 		/*
848 		 * Empty loss history: no loss so far, hence p stays 0.
849 		 * Sample RTT values, since an RTT estimate is required for the
850 		 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
851 		 */
852 		if (sample != 0)
853 			hcrx->ccid3hcrx_rtt = tfrc_ewma(hcrx->ccid3hcrx_rtt, sample, 9);
854 
855 	} else if (tfrc_lh_update_i_mean(&hcrx->ccid3hcrx_li_hist, skb)) {
856 		/*
857 		 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
858 		 * has decreased (resp. p has increased), send feedback now.
859 		 */
860 		do_feedback = CCID3_FBACK_PARAM_CHANGE;
861 	}
862 
863 	/*
864 	 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
865 	 */
866 	if (SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->ccid3hcrx_last_counter) > 3)
867 		do_feedback = CCID3_FBACK_PERIODIC;
868 
869 update_records:
870 	tfrc_rx_hist_add_packet(&hcrx->ccid3hcrx_hist, skb, ndp);
871 
872 done_receiving:
873 	if (do_feedback)
874 		ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
875 }
876 
877 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
878 {
879 	struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
880 
881 	hcrx->ccid3hcrx_state = TFRC_RSTATE_NO_DATA;
882 	tfrc_lh_init(&hcrx->ccid3hcrx_li_hist);
883 	return tfrc_rx_hist_alloc(&hcrx->ccid3hcrx_hist);
884 }
885 
886 static void ccid3_hc_rx_exit(struct sock *sk)
887 {
888 	struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
889 
890 	ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
891 
892 	tfrc_rx_hist_purge(&hcrx->ccid3hcrx_hist);
893 	tfrc_lh_cleanup(&hcrx->ccid3hcrx_li_hist);
894 }
895 
896 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
897 {
898 	const struct ccid3_hc_rx_sock *hcrx;
899 
900 	/* Listen socks doesn't have a private CCID block */
901 	if (sk->sk_state == DCCP_LISTEN)
902 		return;
903 
904 	hcrx = ccid3_hc_rx_sk(sk);
905 	info->tcpi_ca_state = hcrx->ccid3hcrx_state;
906 	info->tcpi_options  |= TCPI_OPT_TIMESTAMPS;
907 	info->tcpi_rcv_rtt  = hcrx->ccid3hcrx_rtt;
908 }
909 
910 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
911 				  u32 __user *optval, int __user *optlen)
912 {
913 	const struct ccid3_hc_rx_sock *hcrx;
914 	struct tfrc_rx_info rx_info;
915 	const void *val;
916 
917 	/* Listen socks doesn't have a private CCID block */
918 	if (sk->sk_state == DCCP_LISTEN)
919 		return -EINVAL;
920 
921 	hcrx = ccid3_hc_rx_sk(sk);
922 	switch (optname) {
923 	case DCCP_SOCKOPT_CCID_RX_INFO:
924 		if (len < sizeof(rx_info))
925 			return -EINVAL;
926 		rx_info.tfrcrx_x_recv = hcrx->ccid3hcrx_x_recv;
927 		rx_info.tfrcrx_rtt    = hcrx->ccid3hcrx_rtt;
928 		rx_info.tfrcrx_p      = hcrx->ccid3hcrx_pinv == 0 ? ~0U :
929 					   scaled_div(1, hcrx->ccid3hcrx_pinv);
930 		len = sizeof(rx_info);
931 		val = &rx_info;
932 		break;
933 	default:
934 		return -ENOPROTOOPT;
935 	}
936 
937 	if (put_user(len, optlen) || copy_to_user(optval, val, len))
938 		return -EFAULT;
939 
940 	return 0;
941 }
942 
943 struct ccid_operations ccid3_ops = {
944 	.ccid_id		   = DCCPC_CCID3,
945 	.ccid_name		   = "TCP-Friendly Rate Control",
946 	.ccid_hc_tx_obj_size	   = sizeof(struct ccid3_hc_tx_sock),
947 	.ccid_hc_tx_init	   = ccid3_hc_tx_init,
948 	.ccid_hc_tx_exit	   = ccid3_hc_tx_exit,
949 	.ccid_hc_tx_send_packet	   = ccid3_hc_tx_send_packet,
950 	.ccid_hc_tx_packet_sent	   = ccid3_hc_tx_packet_sent,
951 	.ccid_hc_tx_packet_recv	   = ccid3_hc_tx_packet_recv,
952 	.ccid_hc_tx_parse_options  = ccid3_hc_tx_parse_options,
953 	.ccid_hc_rx_obj_size	   = sizeof(struct ccid3_hc_rx_sock),
954 	.ccid_hc_rx_init	   = ccid3_hc_rx_init,
955 	.ccid_hc_rx_exit	   = ccid3_hc_rx_exit,
956 	.ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
957 	.ccid_hc_rx_packet_recv	   = ccid3_hc_rx_packet_recv,
958 	.ccid_hc_rx_get_info	   = ccid3_hc_rx_get_info,
959 	.ccid_hc_tx_get_info	   = ccid3_hc_tx_get_info,
960 	.ccid_hc_rx_getsockopt	   = ccid3_hc_rx_getsockopt,
961 	.ccid_hc_tx_getsockopt	   = ccid3_hc_tx_getsockopt,
962 };
963 
964 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
965 module_param(ccid3_debug, bool, 0644);
966 MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages");
967 #endif
968