xref: /openbmc/linux/net/dccp/ccids/ccid3.c (revision 11788d9b)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  Copyright (c) 2007   The University of Aberdeen, Scotland, UK
4  *  Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
5  *  Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
6  *
7  *  An implementation of the DCCP protocol
8  *
9  *  This code has been developed by the University of Waikato WAND
10  *  research group. For further information please see https://www.wand.net.nz/
11  *
12  *  This code also uses code from Lulea University, rereleased as GPL by its
13  *  authors:
14  *  Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
15  *
16  *  Changes to meet Linux coding standards, to make it meet latest ccid3 draft
17  *  and to make it work as a loadable module in the DCCP stack written by
18  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
19  *
20  *  Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
21  */
22 #include "../dccp.h"
23 #include "ccid3.h"
24 
25 #include <asm/unaligned.h>
26 
27 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
28 static bool ccid3_debug;
29 #define ccid3_pr_debug(format, a...)	DCCP_PR_DEBUG(ccid3_debug, format, ##a)
30 #else
31 #define ccid3_pr_debug(format, a...)
32 #endif
33 
34 /*
35  *	Transmitter Half-Connection Routines
36  */
37 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
38 static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
39 {
40 	static const char *const ccid3_state_names[] = {
41 	[TFRC_SSTATE_NO_SENT]  = "NO_SENT",
42 	[TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
43 	[TFRC_SSTATE_FBACK]    = "FBACK",
44 	};
45 
46 	return ccid3_state_names[state];
47 }
48 #endif
49 
50 static void ccid3_hc_tx_set_state(struct sock *sk,
51 				  enum ccid3_hc_tx_states state)
52 {
53 	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
54 	enum ccid3_hc_tx_states oldstate = hc->tx_state;
55 
56 	ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
57 		       dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
58 		       ccid3_tx_state_name(state));
59 	WARN_ON(state == oldstate);
60 	hc->tx_state = state;
61 }
62 
63 /*
64  * Compute the initial sending rate X_init in the manner of RFC 3390:
65  *
66  *	X_init  =  min(4 * s, max(2 * s, 4380 bytes)) / RTT
67  *
68  * Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
69  * (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
70  * For consistency with other parts of the code, X_init is scaled by 2^6.
71  */
72 static inline u64 rfc3390_initial_rate(struct sock *sk)
73 {
74 	const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
75 	const __u32 w_init = clamp_t(__u32, 4380U, 2 * hc->tx_s, 4 * hc->tx_s);
76 
77 	return scaled_div(w_init << 6, hc->tx_rtt);
78 }
79 
80 /**
81  * ccid3_update_send_interval  -  Calculate new t_ipi = s / X_inst
82  * This respects the granularity of X_inst (64 * bytes/second).
83  */
84 static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hc)
85 {
86 	hc->tx_t_ipi = scaled_div32(((u64)hc->tx_s) << 6, hc->tx_x);
87 
88 	DCCP_BUG_ON(hc->tx_t_ipi == 0);
89 	ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hc->tx_t_ipi,
90 		       hc->tx_s, (unsigned int)(hc->tx_x >> 6));
91 }
92 
93 static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hc, ktime_t now)
94 {
95 	u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count);
96 
97 	return delta / hc->tx_rtt;
98 }
99 
100 /**
101  * ccid3_hc_tx_update_x  -  Update allowed sending rate X
102  * @stamp: most recent time if available - can be left NULL.
103  *
104  * This function tracks draft rfc3448bis, check there for latest details.
105  *
106  * Note: X and X_recv are both stored in units of 64 * bytes/second, to support
107  *       fine-grained resolution of sending rates. This requires scaling by 2^6
108  *       throughout the code. Only X_calc is unscaled (in bytes/second).
109  *
110  */
111 static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
112 {
113 	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
114 	__u64 min_rate = 2 * hc->tx_x_recv;
115 	const __u64 old_x = hc->tx_x;
116 	ktime_t now = stamp ? *stamp : ktime_get_real();
117 
118 	/*
119 	 * Handle IDLE periods: do not reduce below RFC3390 initial sending rate
120 	 * when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
121 	 * a sender is idle if it has not sent anything over a 2-RTT-period.
122 	 * For consistency with X and X_recv, min_rate is also scaled by 2^6.
123 	 */
124 	if (ccid3_hc_tx_idle_rtt(hc, now) >= 2) {
125 		min_rate = rfc3390_initial_rate(sk);
126 		min_rate = max(min_rate, 2 * hc->tx_x_recv);
127 	}
128 
129 	if (hc->tx_p > 0) {
130 
131 		hc->tx_x = min(((__u64)hc->tx_x_calc) << 6, min_rate);
132 		hc->tx_x = max(hc->tx_x, (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
133 
134 	} else if (ktime_us_delta(now, hc->tx_t_ld) - (s64)hc->tx_rtt >= 0) {
135 
136 		hc->tx_x = min(2 * hc->tx_x, min_rate);
137 		hc->tx_x = max(hc->tx_x,
138 			       scaled_div(((__u64)hc->tx_s) << 6, hc->tx_rtt));
139 		hc->tx_t_ld = now;
140 	}
141 
142 	if (hc->tx_x != old_x) {
143 		ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
144 			       "X_recv=%u\n", (unsigned int)(old_x >> 6),
145 			       (unsigned int)(hc->tx_x >> 6), hc->tx_x_calc,
146 			       (unsigned int)(hc->tx_x_recv >> 6));
147 
148 		ccid3_update_send_interval(hc);
149 	}
150 }
151 
152 /**
153  *	ccid3_hc_tx_update_s - Track the mean packet size `s'
154  *	@len: DCCP packet payload size in bytes
155  *
156  *	cf. RFC 4342, 5.3 and  RFC 3448, 4.1
157  */
158 static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hc, int len)
159 {
160 	const u16 old_s = hc->tx_s;
161 
162 	hc->tx_s = tfrc_ewma(hc->tx_s, len, 9);
163 
164 	if (hc->tx_s != old_s)
165 		ccid3_update_send_interval(hc);
166 }
167 
168 /*
169  *	Update Window Counter using the algorithm from [RFC 4342, 8.1].
170  *	As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
171  */
172 static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hc,
173 						ktime_t now)
174 {
175 	u32 delta = ktime_us_delta(now, hc->tx_t_last_win_count),
176 	    quarter_rtts = (4 * delta) / hc->tx_rtt;
177 
178 	if (quarter_rtts > 0) {
179 		hc->tx_t_last_win_count = now;
180 		hc->tx_last_win_count  += min(quarter_rtts, 5U);
181 		hc->tx_last_win_count  &= 0xF;		/* mod 16 */
182 	}
183 }
184 
185 static void ccid3_hc_tx_no_feedback_timer(struct timer_list *t)
186 {
187 	struct ccid3_hc_tx_sock *hc = from_timer(hc, t, tx_no_feedback_timer);
188 	struct sock *sk = hc->sk;
189 	unsigned long t_nfb = USEC_PER_SEC / 5;
190 
191 	bh_lock_sock(sk);
192 	if (sock_owned_by_user(sk)) {
193 		/* Try again later. */
194 		/* XXX: set some sensible MIB */
195 		goto restart_timer;
196 	}
197 
198 	ccid3_pr_debug("%s(%p, state=%s) - entry\n", dccp_role(sk), sk,
199 		       ccid3_tx_state_name(hc->tx_state));
200 
201 	/* Ignore and do not restart after leaving the established state */
202 	if ((1 << sk->sk_state) & ~(DCCPF_OPEN | DCCPF_PARTOPEN))
203 		goto out;
204 
205 	/* Reset feedback state to "no feedback received" */
206 	if (hc->tx_state == TFRC_SSTATE_FBACK)
207 		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
208 
209 	/*
210 	 * Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
211 	 * RTO is 0 if and only if no feedback has been received yet.
212 	 */
213 	if (hc->tx_t_rto == 0 || hc->tx_p == 0) {
214 
215 		/* halve send rate directly */
216 		hc->tx_x = max(hc->tx_x / 2,
217 			       (((__u64)hc->tx_s) << 6) / TFRC_T_MBI);
218 		ccid3_update_send_interval(hc);
219 	} else {
220 		/*
221 		 *  Modify the cached value of X_recv
222 		 *
223 		 *  If (X_calc > 2 * X_recv)
224 		 *    X_recv = max(X_recv / 2, s / (2 * t_mbi));
225 		 *  Else
226 		 *    X_recv = X_calc / 4;
227 		 *
228 		 *  Note that X_recv is scaled by 2^6 while X_calc is not
229 		 */
230 		if (hc->tx_x_calc > (hc->tx_x_recv >> 5))
231 			hc->tx_x_recv =
232 				max(hc->tx_x_recv / 2,
233 				    (((__u64)hc->tx_s) << 6) / (2*TFRC_T_MBI));
234 		else {
235 			hc->tx_x_recv = hc->tx_x_calc;
236 			hc->tx_x_recv <<= 4;
237 		}
238 		ccid3_hc_tx_update_x(sk, NULL);
239 	}
240 	ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
241 			(unsigned long long)hc->tx_x);
242 
243 	/*
244 	 * Set new timeout for the nofeedback timer.
245 	 * See comments in packet_recv() regarding the value of t_RTO.
246 	 */
247 	if (unlikely(hc->tx_t_rto == 0))	/* no feedback received yet */
248 		t_nfb = TFRC_INITIAL_TIMEOUT;
249 	else
250 		t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
251 
252 restart_timer:
253 	sk_reset_timer(sk, &hc->tx_no_feedback_timer,
254 			   jiffies + usecs_to_jiffies(t_nfb));
255 out:
256 	bh_unlock_sock(sk);
257 	sock_put(sk);
258 }
259 
260 /**
261  * ccid3_hc_tx_send_packet  -  Delay-based dequeueing of TX packets
262  * @skb: next packet candidate to send on @sk
263  *
264  * This function uses the convention of ccid_packet_dequeue_eval() and
265  * returns a millisecond-delay value between 0 and t_mbi = 64000 msec.
266  */
267 static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
268 {
269 	struct dccp_sock *dp = dccp_sk(sk);
270 	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
271 	ktime_t now = ktime_get_real();
272 	s64 delay;
273 
274 	/*
275 	 * This function is called only for Data and DataAck packets. Sending
276 	 * zero-sized Data(Ack)s is theoretically possible, but for congestion
277 	 * control this case is pathological - ignore it.
278 	 */
279 	if (unlikely(skb->len == 0))
280 		return -EBADMSG;
281 
282 	if (hc->tx_state == TFRC_SSTATE_NO_SENT) {
283 		sk_reset_timer(sk, &hc->tx_no_feedback_timer, (jiffies +
284 			       usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
285 		hc->tx_last_win_count	= 0;
286 		hc->tx_t_last_win_count = now;
287 
288 		/* Set t_0 for initial packet */
289 		hc->tx_t_nom = now;
290 
291 		hc->tx_s = skb->len;
292 
293 		/*
294 		 * Use initial RTT sample when available: recommended by erratum
295 		 * to RFC 4342. This implements the initialisation procedure of
296 		 * draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
297 		 */
298 		if (dp->dccps_syn_rtt) {
299 			ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
300 			hc->tx_rtt  = dp->dccps_syn_rtt;
301 			hc->tx_x    = rfc3390_initial_rate(sk);
302 			hc->tx_t_ld = now;
303 		} else {
304 			/*
305 			 * Sender does not have RTT sample:
306 			 * - set fallback RTT (RFC 4340, 3.4) since a RTT value
307 			 *   is needed in several parts (e.g.  window counter);
308 			 * - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
309 			 */
310 			hc->tx_rtt = DCCP_FALLBACK_RTT;
311 			hc->tx_x   = hc->tx_s;
312 			hc->tx_x <<= 6;
313 		}
314 		ccid3_update_send_interval(hc);
315 
316 		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
317 
318 	} else {
319 		delay = ktime_us_delta(hc->tx_t_nom, now);
320 		ccid3_pr_debug("delay=%ld\n", (long)delay);
321 		/*
322 		 *	Scheduling of packet transmissions (RFC 5348, 8.3)
323 		 *
324 		 * if (t_now > t_nom - delta)
325 		 *       // send the packet now
326 		 * else
327 		 *       // send the packet in (t_nom - t_now) milliseconds.
328 		 */
329 		if (delay >= TFRC_T_DELTA)
330 			return (u32)delay / USEC_PER_MSEC;
331 
332 		ccid3_hc_tx_update_win_count(hc, now);
333 	}
334 
335 	/* prepare to send now (add options etc.) */
336 	dp->dccps_hc_tx_insert_options = 1;
337 	DCCP_SKB_CB(skb)->dccpd_ccval  = hc->tx_last_win_count;
338 
339 	/* set the nominal send time for the next following packet */
340 	hc->tx_t_nom = ktime_add_us(hc->tx_t_nom, hc->tx_t_ipi);
341 	return CCID_PACKET_SEND_AT_ONCE;
342 }
343 
344 static void ccid3_hc_tx_packet_sent(struct sock *sk, unsigned int len)
345 {
346 	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
347 
348 	ccid3_hc_tx_update_s(hc, len);
349 
350 	if (tfrc_tx_hist_add(&hc->tx_hist, dccp_sk(sk)->dccps_gss))
351 		DCCP_CRIT("packet history - out of memory!");
352 }
353 
354 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
355 {
356 	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
357 	struct tfrc_tx_hist_entry *acked;
358 	ktime_t now;
359 	unsigned long t_nfb;
360 	u32 r_sample;
361 
362 	/* we are only interested in ACKs */
363 	if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
364 	      DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
365 		return;
366 	/*
367 	 * Locate the acknowledged packet in the TX history.
368 	 *
369 	 * Returning "entry not found" here can for instance happen when
370 	 *  - the host has not sent out anything (e.g. a passive server),
371 	 *  - the Ack is outdated (packet with higher Ack number was received),
372 	 *  - it is a bogus Ack (for a packet not sent on this connection).
373 	 */
374 	acked = tfrc_tx_hist_find_entry(hc->tx_hist, dccp_hdr_ack_seq(skb));
375 	if (acked == NULL)
376 		return;
377 	/* For the sake of RTT sampling, ignore/remove all older entries */
378 	tfrc_tx_hist_purge(&acked->next);
379 
380 	/* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
381 	now	  = ktime_get_real();
382 	r_sample  = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
383 	hc->tx_rtt = tfrc_ewma(hc->tx_rtt, r_sample, 9);
384 
385 	/*
386 	 * Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
387 	 */
388 	if (hc->tx_state == TFRC_SSTATE_NO_FBACK) {
389 		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
390 
391 		if (hc->tx_t_rto == 0) {
392 			/*
393 			 * Initial feedback packet: Larger Initial Windows (4.2)
394 			 */
395 			hc->tx_x    = rfc3390_initial_rate(sk);
396 			hc->tx_t_ld = now;
397 
398 			ccid3_update_send_interval(hc);
399 
400 			goto done_computing_x;
401 		} else if (hc->tx_p == 0) {
402 			/*
403 			 * First feedback after nofeedback timer expiry (4.3)
404 			 */
405 			goto done_computing_x;
406 		}
407 	}
408 
409 	/* Update sending rate (step 4 of [RFC 3448, 4.3]) */
410 	if (hc->tx_p > 0)
411 		hc->tx_x_calc = tfrc_calc_x(hc->tx_s, hc->tx_rtt, hc->tx_p);
412 	ccid3_hc_tx_update_x(sk, &now);
413 
414 done_computing_x:
415 	ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
416 			       "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
417 			       dccp_role(sk), sk, hc->tx_rtt, r_sample,
418 			       hc->tx_s, hc->tx_p, hc->tx_x_calc,
419 			       (unsigned int)(hc->tx_x_recv >> 6),
420 			       (unsigned int)(hc->tx_x >> 6));
421 
422 	/* unschedule no feedback timer */
423 	sk_stop_timer(sk, &hc->tx_no_feedback_timer);
424 
425 	/*
426 	 * As we have calculated new ipi, delta, t_nom it is possible
427 	 * that we now can send a packet, so wake up dccp_wait_for_ccid
428 	 */
429 	sk->sk_write_space(sk);
430 
431 	/*
432 	 * Update timeout interval for the nofeedback timer. In order to control
433 	 * rate halving on networks with very low RTTs (<= 1 ms), use per-route
434 	 * tunable RTAX_RTO_MIN value as the lower bound.
435 	 */
436 	hc->tx_t_rto = max_t(u32, 4 * hc->tx_rtt,
437 				  USEC_PER_SEC/HZ * tcp_rto_min(sk));
438 	/*
439 	 * Schedule no feedback timer to expire in
440 	 * max(t_RTO, 2 * s/X)  =  max(t_RTO, 2 * t_ipi)
441 	 */
442 	t_nfb = max(hc->tx_t_rto, 2 * hc->tx_t_ipi);
443 
444 	ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
445 		       "expire in %lu jiffies (%luus)\n",
446 		       dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
447 
448 	sk_reset_timer(sk, &hc->tx_no_feedback_timer,
449 			   jiffies + usecs_to_jiffies(t_nfb));
450 }
451 
452 static int ccid3_hc_tx_parse_options(struct sock *sk, u8 packet_type,
453 				     u8 option, u8 *optval, u8 optlen)
454 {
455 	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
456 	__be32 opt_val;
457 
458 	switch (option) {
459 	case TFRC_OPT_RECEIVE_RATE:
460 	case TFRC_OPT_LOSS_EVENT_RATE:
461 		/* Must be ignored on Data packets, cf. RFC 4342 8.3 and 8.5 */
462 		if (packet_type == DCCP_PKT_DATA)
463 			break;
464 		if (unlikely(optlen != 4)) {
465 			DCCP_WARN("%s(%p), invalid len %d for %u\n",
466 				  dccp_role(sk), sk, optlen, option);
467 			return -EINVAL;
468 		}
469 		opt_val = ntohl(get_unaligned((__be32 *)optval));
470 
471 		if (option == TFRC_OPT_RECEIVE_RATE) {
472 			/* Receive Rate is kept in units of 64 bytes/second */
473 			hc->tx_x_recv = opt_val;
474 			hc->tx_x_recv <<= 6;
475 
476 			ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
477 				       dccp_role(sk), sk, opt_val);
478 		} else {
479 			/* Update the fixpoint Loss Event Rate fraction */
480 			hc->tx_p = tfrc_invert_loss_event_rate(opt_val);
481 
482 			ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
483 				       dccp_role(sk), sk, opt_val);
484 		}
485 	}
486 	return 0;
487 }
488 
489 static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
490 {
491 	struct ccid3_hc_tx_sock *hc = ccid_priv(ccid);
492 
493 	hc->tx_state = TFRC_SSTATE_NO_SENT;
494 	hc->tx_hist  = NULL;
495 	hc->sk	     = sk;
496 	timer_setup(&hc->tx_no_feedback_timer,
497 		    ccid3_hc_tx_no_feedback_timer, 0);
498 	return 0;
499 }
500 
501 static void ccid3_hc_tx_exit(struct sock *sk)
502 {
503 	struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
504 
505 	sk_stop_timer(sk, &hc->tx_no_feedback_timer);
506 	tfrc_tx_hist_purge(&hc->tx_hist);
507 }
508 
509 static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
510 {
511 	info->tcpi_rto = ccid3_hc_tx_sk(sk)->tx_t_rto;
512 	info->tcpi_rtt = ccid3_hc_tx_sk(sk)->tx_rtt;
513 }
514 
515 static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
516 				  u32 __user *optval, int __user *optlen)
517 {
518 	const struct ccid3_hc_tx_sock *hc = ccid3_hc_tx_sk(sk);
519 	struct tfrc_tx_info tfrc;
520 	const void *val;
521 
522 	switch (optname) {
523 	case DCCP_SOCKOPT_CCID_TX_INFO:
524 		if (len < sizeof(tfrc))
525 			return -EINVAL;
526 		memset(&tfrc, 0, sizeof(tfrc));
527 		tfrc.tfrctx_x	   = hc->tx_x;
528 		tfrc.tfrctx_x_recv = hc->tx_x_recv;
529 		tfrc.tfrctx_x_calc = hc->tx_x_calc;
530 		tfrc.tfrctx_rtt	   = hc->tx_rtt;
531 		tfrc.tfrctx_p	   = hc->tx_p;
532 		tfrc.tfrctx_rto	   = hc->tx_t_rto;
533 		tfrc.tfrctx_ipi	   = hc->tx_t_ipi;
534 		len = sizeof(tfrc);
535 		val = &tfrc;
536 		break;
537 	default:
538 		return -ENOPROTOOPT;
539 	}
540 
541 	if (put_user(len, optlen) || copy_to_user(optval, val, len))
542 		return -EFAULT;
543 
544 	return 0;
545 }
546 
547 /*
548  *	Receiver Half-Connection Routines
549  */
550 
551 /* CCID3 feedback types */
552 enum ccid3_fback_type {
553 	CCID3_FBACK_NONE = 0,
554 	CCID3_FBACK_INITIAL,
555 	CCID3_FBACK_PERIODIC,
556 	CCID3_FBACK_PARAM_CHANGE
557 };
558 
559 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
560 static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
561 {
562 	static const char *const ccid3_rx_state_names[] = {
563 	[TFRC_RSTATE_NO_DATA] = "NO_DATA",
564 	[TFRC_RSTATE_DATA]    = "DATA",
565 	};
566 
567 	return ccid3_rx_state_names[state];
568 }
569 #endif
570 
571 static void ccid3_hc_rx_set_state(struct sock *sk,
572 				  enum ccid3_hc_rx_states state)
573 {
574 	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
575 	enum ccid3_hc_rx_states oldstate = hc->rx_state;
576 
577 	ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
578 		       dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
579 		       ccid3_rx_state_name(state));
580 	WARN_ON(state == oldstate);
581 	hc->rx_state = state;
582 }
583 
584 static void ccid3_hc_rx_send_feedback(struct sock *sk,
585 				      const struct sk_buff *skb,
586 				      enum ccid3_fback_type fbtype)
587 {
588 	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
589 	struct dccp_sock *dp = dccp_sk(sk);
590 	ktime_t now = ktime_get();
591 	s64 delta = 0;
592 
593 	switch (fbtype) {
594 	case CCID3_FBACK_INITIAL:
595 		hc->rx_x_recv = 0;
596 		hc->rx_pinv   = ~0U;   /* see RFC 4342, 8.5 */
597 		break;
598 	case CCID3_FBACK_PARAM_CHANGE:
599 		/*
600 		 * When parameters change (new loss or p > p_prev), we do not
601 		 * have a reliable estimate for R_m of [RFC 3448, 6.2] and so
602 		 * need to  reuse the previous value of X_recv. However, when
603 		 * X_recv was 0 (due to early loss), this would kill X down to
604 		 * s/t_mbi (i.e. one packet in 64 seconds).
605 		 * To avoid such drastic reduction, we approximate X_recv as
606 		 * the number of bytes since last feedback.
607 		 * This is a safe fallback, since X is bounded above by X_calc.
608 		 */
609 		if (hc->rx_x_recv > 0)
610 			break;
611 		/* fall through */
612 	case CCID3_FBACK_PERIODIC:
613 		delta = ktime_us_delta(now, hc->rx_tstamp_last_feedback);
614 		if (delta <= 0)
615 			delta = 1;
616 		hc->rx_x_recv = scaled_div32(hc->rx_bytes_recv, delta);
617 		break;
618 	default:
619 		return;
620 	}
621 
622 	ccid3_pr_debug("Interval %lldusec, X_recv=%u, 1/p=%u\n", delta,
623 		       hc->rx_x_recv, hc->rx_pinv);
624 
625 	hc->rx_tstamp_last_feedback = now;
626 	hc->rx_last_counter	    = dccp_hdr(skb)->dccph_ccval;
627 	hc->rx_bytes_recv	    = 0;
628 
629 	dp->dccps_hc_rx_insert_options = 1;
630 	dccp_send_ack(sk);
631 }
632 
633 static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
634 {
635 	const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
636 	__be32 x_recv, pinv;
637 
638 	if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
639 		return 0;
640 
641 	if (dccp_packet_without_ack(skb))
642 		return 0;
643 
644 	x_recv = htonl(hc->rx_x_recv);
645 	pinv   = htonl(hc->rx_pinv);
646 
647 	if (dccp_insert_option(skb, TFRC_OPT_LOSS_EVENT_RATE,
648 			       &pinv, sizeof(pinv)) ||
649 	    dccp_insert_option(skb, TFRC_OPT_RECEIVE_RATE,
650 			       &x_recv, sizeof(x_recv)))
651 		return -1;
652 
653 	return 0;
654 }
655 
656 /**
657  * ccid3_first_li  -  Implements [RFC 5348, 6.3.1]
658  *
659  * Determine the length of the first loss interval via inverse lookup.
660  * Assume that X_recv can be computed by the throughput equation
661  *		    s
662  *	X_recv = --------
663  *		 R * fval
664  * Find some p such that f(p) = fval; return 1/p (scaled).
665  */
666 static u32 ccid3_first_li(struct sock *sk)
667 {
668 	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
669 	u32 x_recv, p;
670 	s64 delta;
671 	u64 fval;
672 
673 	if (hc->rx_rtt == 0) {
674 		DCCP_WARN("No RTT estimate available, using fallback RTT\n");
675 		hc->rx_rtt = DCCP_FALLBACK_RTT;
676 	}
677 
678 	delta = ktime_us_delta(ktime_get(), hc->rx_tstamp_last_feedback);
679 	if (delta <= 0)
680 		delta = 1;
681 	x_recv = scaled_div32(hc->rx_bytes_recv, delta);
682 	if (x_recv == 0) {		/* would also trigger divide-by-zero */
683 		DCCP_WARN("X_recv==0\n");
684 		if (hc->rx_x_recv == 0) {
685 			DCCP_BUG("stored value of X_recv is zero");
686 			return ~0U;
687 		}
688 		x_recv = hc->rx_x_recv;
689 	}
690 
691 	fval = scaled_div(hc->rx_s, hc->rx_rtt);
692 	fval = scaled_div32(fval, x_recv);
693 	p = tfrc_calc_x_reverse_lookup(fval);
694 
695 	ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
696 		       "loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
697 
698 	return p == 0 ? ~0U : scaled_div(1, p);
699 }
700 
701 static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
702 {
703 	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
704 	enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
705 	const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
706 	const bool is_data_packet = dccp_data_packet(skb);
707 
708 	if (unlikely(hc->rx_state == TFRC_RSTATE_NO_DATA)) {
709 		if (is_data_packet) {
710 			const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
711 			do_feedback = CCID3_FBACK_INITIAL;
712 			ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
713 			hc->rx_s = payload;
714 			/*
715 			 * Not necessary to update rx_bytes_recv here,
716 			 * since X_recv = 0 for the first feedback packet (cf.
717 			 * RFC 3448, 6.3) -- gerrit
718 			 */
719 		}
720 		goto update_records;
721 	}
722 
723 	if (tfrc_rx_hist_duplicate(&hc->rx_hist, skb))
724 		return; /* done receiving */
725 
726 	if (is_data_packet) {
727 		const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
728 		/*
729 		 * Update moving-average of s and the sum of received payload bytes
730 		 */
731 		hc->rx_s = tfrc_ewma(hc->rx_s, payload, 9);
732 		hc->rx_bytes_recv += payload;
733 	}
734 
735 	/*
736 	 * Perform loss detection and handle pending losses
737 	 */
738 	if (tfrc_rx_handle_loss(&hc->rx_hist, &hc->rx_li_hist,
739 				skb, ndp, ccid3_first_li, sk)) {
740 		do_feedback = CCID3_FBACK_PARAM_CHANGE;
741 		goto done_receiving;
742 	}
743 
744 	if (tfrc_rx_hist_loss_pending(&hc->rx_hist))
745 		return; /* done receiving */
746 
747 	/*
748 	 * Handle data packets: RTT sampling and monitoring p
749 	 */
750 	if (unlikely(!is_data_packet))
751 		goto update_records;
752 
753 	if (!tfrc_lh_is_initialised(&hc->rx_li_hist)) {
754 		const u32 sample = tfrc_rx_hist_sample_rtt(&hc->rx_hist, skb);
755 		/*
756 		 * Empty loss history: no loss so far, hence p stays 0.
757 		 * Sample RTT values, since an RTT estimate is required for the
758 		 * computation of p when the first loss occurs; RFC 3448, 6.3.1.
759 		 */
760 		if (sample != 0)
761 			hc->rx_rtt = tfrc_ewma(hc->rx_rtt, sample, 9);
762 
763 	} else if (tfrc_lh_update_i_mean(&hc->rx_li_hist, skb)) {
764 		/*
765 		 * Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
766 		 * has decreased (resp. p has increased), send feedback now.
767 		 */
768 		do_feedback = CCID3_FBACK_PARAM_CHANGE;
769 	}
770 
771 	/*
772 	 * Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
773 	 */
774 	if (SUB16(dccp_hdr(skb)->dccph_ccval, hc->rx_last_counter) > 3)
775 		do_feedback = CCID3_FBACK_PERIODIC;
776 
777 update_records:
778 	tfrc_rx_hist_add_packet(&hc->rx_hist, skb, ndp);
779 
780 done_receiving:
781 	if (do_feedback)
782 		ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
783 }
784 
785 static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
786 {
787 	struct ccid3_hc_rx_sock *hc = ccid_priv(ccid);
788 
789 	hc->rx_state = TFRC_RSTATE_NO_DATA;
790 	tfrc_lh_init(&hc->rx_li_hist);
791 	return tfrc_rx_hist_alloc(&hc->rx_hist);
792 }
793 
794 static void ccid3_hc_rx_exit(struct sock *sk)
795 {
796 	struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
797 
798 	tfrc_rx_hist_purge(&hc->rx_hist);
799 	tfrc_lh_cleanup(&hc->rx_li_hist);
800 }
801 
802 static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
803 {
804 	info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->rx_state;
805 	info->tcpi_options  |= TCPI_OPT_TIMESTAMPS;
806 	info->tcpi_rcv_rtt  = ccid3_hc_rx_sk(sk)->rx_rtt;
807 }
808 
809 static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
810 				  u32 __user *optval, int __user *optlen)
811 {
812 	const struct ccid3_hc_rx_sock *hc = ccid3_hc_rx_sk(sk);
813 	struct tfrc_rx_info rx_info;
814 	const void *val;
815 
816 	switch (optname) {
817 	case DCCP_SOCKOPT_CCID_RX_INFO:
818 		if (len < sizeof(rx_info))
819 			return -EINVAL;
820 		rx_info.tfrcrx_x_recv = hc->rx_x_recv;
821 		rx_info.tfrcrx_rtt    = hc->rx_rtt;
822 		rx_info.tfrcrx_p      = tfrc_invert_loss_event_rate(hc->rx_pinv);
823 		len = sizeof(rx_info);
824 		val = &rx_info;
825 		break;
826 	default:
827 		return -ENOPROTOOPT;
828 	}
829 
830 	if (put_user(len, optlen) || copy_to_user(optval, val, len))
831 		return -EFAULT;
832 
833 	return 0;
834 }
835 
836 struct ccid_operations ccid3_ops = {
837 	.ccid_id		   = DCCPC_CCID3,
838 	.ccid_name		   = "TCP-Friendly Rate Control",
839 	.ccid_hc_tx_obj_size	   = sizeof(struct ccid3_hc_tx_sock),
840 	.ccid_hc_tx_init	   = ccid3_hc_tx_init,
841 	.ccid_hc_tx_exit	   = ccid3_hc_tx_exit,
842 	.ccid_hc_tx_send_packet	   = ccid3_hc_tx_send_packet,
843 	.ccid_hc_tx_packet_sent	   = ccid3_hc_tx_packet_sent,
844 	.ccid_hc_tx_packet_recv	   = ccid3_hc_tx_packet_recv,
845 	.ccid_hc_tx_parse_options  = ccid3_hc_tx_parse_options,
846 	.ccid_hc_rx_obj_size	   = sizeof(struct ccid3_hc_rx_sock),
847 	.ccid_hc_rx_init	   = ccid3_hc_rx_init,
848 	.ccid_hc_rx_exit	   = ccid3_hc_rx_exit,
849 	.ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
850 	.ccid_hc_rx_packet_recv	   = ccid3_hc_rx_packet_recv,
851 	.ccid_hc_rx_get_info	   = ccid3_hc_rx_get_info,
852 	.ccid_hc_tx_get_info	   = ccid3_hc_tx_get_info,
853 	.ccid_hc_rx_getsockopt	   = ccid3_hc_rx_getsockopt,
854 	.ccid_hc_tx_getsockopt	   = ccid3_hc_tx_getsockopt,
855 };
856 
857 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
858 module_param(ccid3_debug, bool, 0644);
859 MODULE_PARM_DESC(ccid3_debug, "Enable CCID-3 debug messages");
860 #endif
861