xref: /openbmc/linux/net/dccp/ccids/ccid2.c (revision e8e0929d)
1 /*
2  *  Copyright (c) 2005, 2006 Andrea Bittau <a.bittau@cs.ucl.ac.uk>
3  *
4  *  Changes to meet Linux coding standards, and DCCP infrastructure fixes.
5  *
6  *  Copyright (c) 2006 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7  *
8  *  This program is free software; you can redistribute it and/or modify
9  *  it under the terms of the GNU General Public License as published by
10  *  the Free Software Foundation; either version 2 of the License, or
11  *  (at your option) any later version.
12  *
13  *  This program is distributed in the hope that it will be useful,
14  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
15  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  *  GNU General Public License for more details.
17  *
18  *  You should have received a copy of the GNU General Public License
19  *  along with this program; if not, write to the Free Software
20  *  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21  */
22 
23 /*
24  * This implementation should follow RFC 4341
25  */
26 #include "../feat.h"
27 #include "../ccid.h"
28 #include "../dccp.h"
29 #include "ccid2.h"
30 
31 
32 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
33 static int ccid2_debug;
34 #define ccid2_pr_debug(format, a...)	DCCP_PR_DEBUG(ccid2_debug, format, ##a)
35 
36 static void ccid2_hc_tx_check_sanity(const struct ccid2_hc_tx_sock *hctx)
37 {
38 	int len = 0;
39 	int pipe = 0;
40 	struct ccid2_seq *seqp = hctx->ccid2hctx_seqh;
41 
42 	/* there is data in the chain */
43 	if (seqp != hctx->ccid2hctx_seqt) {
44 		seqp = seqp->ccid2s_prev;
45 		len++;
46 		if (!seqp->ccid2s_acked)
47 			pipe++;
48 
49 		while (seqp != hctx->ccid2hctx_seqt) {
50 			struct ccid2_seq *prev = seqp->ccid2s_prev;
51 
52 			len++;
53 			if (!prev->ccid2s_acked)
54 				pipe++;
55 
56 			/* packets are sent sequentially */
57 			BUG_ON(dccp_delta_seqno(seqp->ccid2s_seq,
58 						prev->ccid2s_seq ) >= 0);
59 			BUG_ON(time_before(seqp->ccid2s_sent,
60 					   prev->ccid2s_sent));
61 
62 			seqp = prev;
63 		}
64 	}
65 
66 	BUG_ON(pipe != hctx->ccid2hctx_pipe);
67 	ccid2_pr_debug("len of chain=%d\n", len);
68 
69 	do {
70 		seqp = seqp->ccid2s_prev;
71 		len++;
72 	} while (seqp != hctx->ccid2hctx_seqh);
73 
74 	ccid2_pr_debug("total len=%d\n", len);
75 	BUG_ON(len != hctx->ccid2hctx_seqbufc * CCID2_SEQBUF_LEN);
76 }
77 #else
78 #define ccid2_pr_debug(format, a...)
79 #define ccid2_hc_tx_check_sanity(hctx)
80 #endif
81 
82 static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hctx)
83 {
84 	struct ccid2_seq *seqp;
85 	int i;
86 
87 	/* check if we have space to preserve the pointer to the buffer */
88 	if (hctx->ccid2hctx_seqbufc >= (sizeof(hctx->ccid2hctx_seqbuf) /
89 					sizeof(struct ccid2_seq*)))
90 		return -ENOMEM;
91 
92 	/* allocate buffer and initialize linked list */
93 	seqp = kmalloc(CCID2_SEQBUF_LEN * sizeof(struct ccid2_seq), gfp_any());
94 	if (seqp == NULL)
95 		return -ENOMEM;
96 
97 	for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) {
98 		seqp[i].ccid2s_next = &seqp[i + 1];
99 		seqp[i + 1].ccid2s_prev = &seqp[i];
100 	}
101 	seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp;
102 	seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
103 
104 	/* This is the first allocation.  Initiate the head and tail.  */
105 	if (hctx->ccid2hctx_seqbufc == 0)
106 		hctx->ccid2hctx_seqh = hctx->ccid2hctx_seqt = seqp;
107 	else {
108 		/* link the existing list with the one we just created */
109 		hctx->ccid2hctx_seqh->ccid2s_next = seqp;
110 		seqp->ccid2s_prev = hctx->ccid2hctx_seqh;
111 
112 		hctx->ccid2hctx_seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
113 		seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hctx->ccid2hctx_seqt;
114 	}
115 
116 	/* store the original pointer to the buffer so we can free it */
117 	hctx->ccid2hctx_seqbuf[hctx->ccid2hctx_seqbufc] = seqp;
118 	hctx->ccid2hctx_seqbufc++;
119 
120 	return 0;
121 }
122 
123 static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
124 {
125 	struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
126 
127 	if (hctx->ccid2hctx_pipe < hctx->ccid2hctx_cwnd)
128 		return 0;
129 
130 	return 1; /* XXX CCID should dequeue when ready instead of polling */
131 }
132 
133 static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
134 {
135 	struct dccp_sock *dp = dccp_sk(sk);
136 	u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->ccid2hctx_cwnd, 2);
137 
138 	/*
139 	 * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from
140 	 * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always
141 	 * acceptable since this causes starvation/deadlock whenever cwnd < 2.
142 	 * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled).
143 	 */
144 	if (val == 0 || val > max_ratio) {
145 		DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio);
146 		val = max_ratio;
147 	}
148 	if (val > DCCPF_ACK_RATIO_MAX)
149 		val = DCCPF_ACK_RATIO_MAX;
150 
151 	if (val == dp->dccps_l_ack_ratio)
152 		return;
153 
154 	ccid2_pr_debug("changing local ack ratio to %u\n", val);
155 	dp->dccps_l_ack_ratio = val;
156 }
157 
158 static void ccid2_change_srtt(struct ccid2_hc_tx_sock *hctx, long val)
159 {
160 	ccid2_pr_debug("change SRTT to %ld\n", val);
161 	hctx->ccid2hctx_srtt = val;
162 }
163 
164 static void ccid2_start_rto_timer(struct sock *sk);
165 
166 static void ccid2_hc_tx_rto_expire(unsigned long data)
167 {
168 	struct sock *sk = (struct sock *)data;
169 	struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
170 	long s;
171 
172 	bh_lock_sock(sk);
173 	if (sock_owned_by_user(sk)) {
174 		sk_reset_timer(sk, &hctx->ccid2hctx_rtotimer,
175 			       jiffies + HZ / 5);
176 		goto out;
177 	}
178 
179 	ccid2_pr_debug("RTO_EXPIRE\n");
180 
181 	ccid2_hc_tx_check_sanity(hctx);
182 
183 	/* back-off timer */
184 	hctx->ccid2hctx_rto <<= 1;
185 
186 	s = hctx->ccid2hctx_rto / HZ;
187 	if (s > 60)
188 		hctx->ccid2hctx_rto = 60 * HZ;
189 
190 	ccid2_start_rto_timer(sk);
191 
192 	/* adjust pipe, cwnd etc */
193 	hctx->ccid2hctx_ssthresh = hctx->ccid2hctx_cwnd / 2;
194 	if (hctx->ccid2hctx_ssthresh < 2)
195 		hctx->ccid2hctx_ssthresh = 2;
196 	hctx->ccid2hctx_cwnd	 = 1;
197 	hctx->ccid2hctx_pipe	 = 0;
198 
199 	/* clear state about stuff we sent */
200 	hctx->ccid2hctx_seqt = hctx->ccid2hctx_seqh;
201 	hctx->ccid2hctx_packets_acked = 0;
202 
203 	/* clear ack ratio state. */
204 	hctx->ccid2hctx_rpseq	 = 0;
205 	hctx->ccid2hctx_rpdupack = -1;
206 	ccid2_change_l_ack_ratio(sk, 1);
207 	ccid2_hc_tx_check_sanity(hctx);
208 out:
209 	bh_unlock_sock(sk);
210 	sock_put(sk);
211 }
212 
213 static void ccid2_start_rto_timer(struct sock *sk)
214 {
215 	struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
216 
217 	ccid2_pr_debug("setting RTO timeout=%ld\n", hctx->ccid2hctx_rto);
218 
219 	BUG_ON(timer_pending(&hctx->ccid2hctx_rtotimer));
220 	sk_reset_timer(sk, &hctx->ccid2hctx_rtotimer,
221 		       jiffies + hctx->ccid2hctx_rto);
222 }
223 
224 static void ccid2_hc_tx_packet_sent(struct sock *sk, int more, unsigned int len)
225 {
226 	struct dccp_sock *dp = dccp_sk(sk);
227 	struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
228 	struct ccid2_seq *next;
229 
230 	hctx->ccid2hctx_pipe++;
231 
232 	hctx->ccid2hctx_seqh->ccid2s_seq   = dp->dccps_gss;
233 	hctx->ccid2hctx_seqh->ccid2s_acked = 0;
234 	hctx->ccid2hctx_seqh->ccid2s_sent  = jiffies;
235 
236 	next = hctx->ccid2hctx_seqh->ccid2s_next;
237 	/* check if we need to alloc more space */
238 	if (next == hctx->ccid2hctx_seqt) {
239 		if (ccid2_hc_tx_alloc_seq(hctx)) {
240 			DCCP_CRIT("packet history - out of memory!");
241 			/* FIXME: find a more graceful way to bail out */
242 			return;
243 		}
244 		next = hctx->ccid2hctx_seqh->ccid2s_next;
245 		BUG_ON(next == hctx->ccid2hctx_seqt);
246 	}
247 	hctx->ccid2hctx_seqh = next;
248 
249 	ccid2_pr_debug("cwnd=%d pipe=%d\n", hctx->ccid2hctx_cwnd,
250 		       hctx->ccid2hctx_pipe);
251 
252 	/*
253 	 * FIXME: The code below is broken and the variables have been removed
254 	 * from the socket struct. The `ackloss' variable was always set to 0,
255 	 * and with arsent there are several problems:
256 	 *  (i) it doesn't just count the number of Acks, but all sent packets;
257 	 *  (ii) it is expressed in # of packets, not # of windows, so the
258 	 *  comparison below uses the wrong formula: Appendix A of RFC 4341
259 	 *  comes up with the number K = cwnd / (R^2 - R) of consecutive windows
260 	 *  of data with no lost or marked Ack packets. If arsent were the # of
261 	 *  consecutive Acks received without loss, then Ack Ratio needs to be
262 	 *  decreased by 1 when
263 	 *	      arsent >=  K * cwnd / R  =  cwnd^2 / (R^3 - R^2)
264 	 *  where cwnd / R is the number of Acks received per window of data
265 	 *  (cf. RFC 4341, App. A). The problems are that
266 	 *  - arsent counts other packets as well;
267 	 *  - the comparison uses a formula different from RFC 4341;
268 	 *  - computing a cubic/quadratic equation each time is too complicated.
269 	 *  Hence a different algorithm is needed.
270 	 */
271 #if 0
272 	/* Ack Ratio.  Need to maintain a concept of how many windows we sent */
273 	hctx->ccid2hctx_arsent++;
274 	/* We had an ack loss in this window... */
275 	if (hctx->ccid2hctx_ackloss) {
276 		if (hctx->ccid2hctx_arsent >= hctx->ccid2hctx_cwnd) {
277 			hctx->ccid2hctx_arsent	= 0;
278 			hctx->ccid2hctx_ackloss	= 0;
279 		}
280 	} else {
281 		/* No acks lost up to now... */
282 		/* decrease ack ratio if enough packets were sent */
283 		if (dp->dccps_l_ack_ratio > 1) {
284 			/* XXX don't calculate denominator each time */
285 			int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
286 				    dp->dccps_l_ack_ratio;
287 
288 			denom = hctx->ccid2hctx_cwnd * hctx->ccid2hctx_cwnd / denom;
289 
290 			if (hctx->ccid2hctx_arsent >= denom) {
291 				ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
292 				hctx->ccid2hctx_arsent = 0;
293 			}
294 		} else {
295 			/* we can't increase ack ratio further [1] */
296 			hctx->ccid2hctx_arsent = 0; /* or maybe set it to cwnd*/
297 		}
298 	}
299 #endif
300 
301 	/* setup RTO timer */
302 	if (!timer_pending(&hctx->ccid2hctx_rtotimer))
303 		ccid2_start_rto_timer(sk);
304 
305 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
306 	do {
307 		struct ccid2_seq *seqp = hctx->ccid2hctx_seqt;
308 
309 		while (seqp != hctx->ccid2hctx_seqh) {
310 			ccid2_pr_debug("out seq=%llu acked=%d time=%lu\n",
311 				       (unsigned long long)seqp->ccid2s_seq,
312 				       seqp->ccid2s_acked, seqp->ccid2s_sent);
313 			seqp = seqp->ccid2s_next;
314 		}
315 	} while (0);
316 	ccid2_pr_debug("=========\n");
317 	ccid2_hc_tx_check_sanity(hctx);
318 #endif
319 }
320 
321 /* XXX Lame code duplication!
322  * returns -1 if none was found.
323  * else returns the next offset to use in the function call.
324  */
325 static int ccid2_ackvector(struct sock *sk, struct sk_buff *skb, int offset,
326 			   unsigned char **vec, unsigned char *veclen)
327 {
328 	const struct dccp_hdr *dh = dccp_hdr(skb);
329 	unsigned char *options = (unsigned char *)dh + dccp_hdr_len(skb);
330 	unsigned char *opt_ptr;
331 	const unsigned char *opt_end = (unsigned char *)dh +
332 					(dh->dccph_doff * 4);
333 	unsigned char opt, len;
334 	unsigned char *value;
335 
336 	BUG_ON(offset < 0);
337 	options += offset;
338 	opt_ptr = options;
339 	if (opt_ptr >= opt_end)
340 		return -1;
341 
342 	while (opt_ptr != opt_end) {
343 		opt   = *opt_ptr++;
344 		len   = 0;
345 		value = NULL;
346 
347 		/* Check if this isn't a single byte option */
348 		if (opt > DCCPO_MAX_RESERVED) {
349 			if (opt_ptr == opt_end)
350 				goto out_invalid_option;
351 
352 			len = *opt_ptr++;
353 			if (len < 3)
354 				goto out_invalid_option;
355 			/*
356 			 * Remove the type and len fields, leaving
357 			 * just the value size
358 			 */
359 			len     -= 2;
360 			value   = opt_ptr;
361 			opt_ptr += len;
362 
363 			if (opt_ptr > opt_end)
364 				goto out_invalid_option;
365 		}
366 
367 		switch (opt) {
368 		case DCCPO_ACK_VECTOR_0:
369 		case DCCPO_ACK_VECTOR_1:
370 			*vec	= value;
371 			*veclen = len;
372 			return offset + (opt_ptr - options);
373 		}
374 	}
375 
376 	return -1;
377 
378 out_invalid_option:
379 	DCCP_BUG("Invalid option - this should not happen (previous parsing)!");
380 	return -1;
381 }
382 
383 static void ccid2_hc_tx_kill_rto_timer(struct sock *sk)
384 {
385 	struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
386 
387 	sk_stop_timer(sk, &hctx->ccid2hctx_rtotimer);
388 	ccid2_pr_debug("deleted RTO timer\n");
389 }
390 
391 static inline void ccid2_new_ack(struct sock *sk,
392 				 struct ccid2_seq *seqp,
393 				 unsigned int *maxincr)
394 {
395 	struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
396 
397 	if (hctx->ccid2hctx_cwnd < hctx->ccid2hctx_ssthresh) {
398 		if (*maxincr > 0 && ++hctx->ccid2hctx_packets_acked == 2) {
399 			hctx->ccid2hctx_cwnd += 1;
400 			*maxincr	     -= 1;
401 			hctx->ccid2hctx_packets_acked = 0;
402 		}
403 	} else if (++hctx->ccid2hctx_packets_acked >= hctx->ccid2hctx_cwnd) {
404 			hctx->ccid2hctx_cwnd += 1;
405 			hctx->ccid2hctx_packets_acked = 0;
406 	}
407 
408 	/* update RTO */
409 	if (hctx->ccid2hctx_srtt == -1 ||
410 	    time_after(jiffies, hctx->ccid2hctx_lastrtt + hctx->ccid2hctx_srtt)) {
411 		unsigned long r = (long)jiffies - (long)seqp->ccid2s_sent;
412 		int s;
413 
414 		/* first measurement */
415 		if (hctx->ccid2hctx_srtt == -1) {
416 			ccid2_pr_debug("R: %lu Time=%lu seq=%llu\n",
417 				       r, jiffies,
418 				       (unsigned long long)seqp->ccid2s_seq);
419 			ccid2_change_srtt(hctx, r);
420 			hctx->ccid2hctx_rttvar = r >> 1;
421 		} else {
422 			/* RTTVAR */
423 			long tmp = hctx->ccid2hctx_srtt - r;
424 			long srtt;
425 
426 			if (tmp < 0)
427 				tmp *= -1;
428 
429 			tmp >>= 2;
430 			hctx->ccid2hctx_rttvar *= 3;
431 			hctx->ccid2hctx_rttvar >>= 2;
432 			hctx->ccid2hctx_rttvar += tmp;
433 
434 			/* SRTT */
435 			srtt = hctx->ccid2hctx_srtt;
436 			srtt *= 7;
437 			srtt >>= 3;
438 			tmp = r >> 3;
439 			srtt += tmp;
440 			ccid2_change_srtt(hctx, srtt);
441 		}
442 		s = hctx->ccid2hctx_rttvar << 2;
443 		/* clock granularity is 1 when based on jiffies */
444 		if (!s)
445 			s = 1;
446 		hctx->ccid2hctx_rto = hctx->ccid2hctx_srtt + s;
447 
448 		/* must be at least a second */
449 		s = hctx->ccid2hctx_rto / HZ;
450 		/* DCCP doesn't require this [but I like it cuz my code sux] */
451 #if 1
452 		if (s < 1)
453 			hctx->ccid2hctx_rto = HZ;
454 #endif
455 		/* max 60 seconds */
456 		if (s > 60)
457 			hctx->ccid2hctx_rto = HZ * 60;
458 
459 		hctx->ccid2hctx_lastrtt = jiffies;
460 
461 		ccid2_pr_debug("srtt: %ld rttvar: %ld rto: %ld (HZ=%d) R=%lu\n",
462 			       hctx->ccid2hctx_srtt, hctx->ccid2hctx_rttvar,
463 			       hctx->ccid2hctx_rto, HZ, r);
464 	}
465 
466 	/* we got a new ack, so re-start RTO timer */
467 	ccid2_hc_tx_kill_rto_timer(sk);
468 	ccid2_start_rto_timer(sk);
469 }
470 
471 static void ccid2_hc_tx_dec_pipe(struct sock *sk)
472 {
473 	struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
474 
475 	if (hctx->ccid2hctx_pipe == 0)
476 		DCCP_BUG("pipe == 0");
477 	else
478 		hctx->ccid2hctx_pipe--;
479 
480 	if (hctx->ccid2hctx_pipe == 0)
481 		ccid2_hc_tx_kill_rto_timer(sk);
482 }
483 
484 static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
485 {
486 	struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
487 
488 	if (time_before(seqp->ccid2s_sent, hctx->ccid2hctx_last_cong)) {
489 		ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
490 		return;
491 	}
492 
493 	hctx->ccid2hctx_last_cong = jiffies;
494 
495 	hctx->ccid2hctx_cwnd     = hctx->ccid2hctx_cwnd / 2 ? : 1U;
496 	hctx->ccid2hctx_ssthresh = max(hctx->ccid2hctx_cwnd, 2U);
497 
498 	/* Avoid spurious timeouts resulting from Ack Ratio > cwnd */
499 	if (dccp_sk(sk)->dccps_l_ack_ratio > hctx->ccid2hctx_cwnd)
500 		ccid2_change_l_ack_ratio(sk, hctx->ccid2hctx_cwnd);
501 }
502 
503 static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
504 {
505 	struct dccp_sock *dp = dccp_sk(sk);
506 	struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
507 	u64 ackno, seqno;
508 	struct ccid2_seq *seqp;
509 	unsigned char *vector;
510 	unsigned char veclen;
511 	int offset = 0;
512 	int done = 0;
513 	unsigned int maxincr = 0;
514 
515 	ccid2_hc_tx_check_sanity(hctx);
516 	/* check reverse path congestion */
517 	seqno = DCCP_SKB_CB(skb)->dccpd_seq;
518 
519 	/* XXX this whole "algorithm" is broken.  Need to fix it to keep track
520 	 * of the seqnos of the dupacks so that rpseq and rpdupack are correct
521 	 * -sorbo.
522 	 */
523 	/* need to bootstrap */
524 	if (hctx->ccid2hctx_rpdupack == -1) {
525 		hctx->ccid2hctx_rpdupack = 0;
526 		hctx->ccid2hctx_rpseq = seqno;
527 	} else {
528 		/* check if packet is consecutive */
529 		if (dccp_delta_seqno(hctx->ccid2hctx_rpseq, seqno) == 1)
530 			hctx->ccid2hctx_rpseq = seqno;
531 		/* it's a later packet */
532 		else if (after48(seqno, hctx->ccid2hctx_rpseq)) {
533 			hctx->ccid2hctx_rpdupack++;
534 
535 			/* check if we got enough dupacks */
536 			if (hctx->ccid2hctx_rpdupack >= NUMDUPACK) {
537 				hctx->ccid2hctx_rpdupack = -1; /* XXX lame */
538 				hctx->ccid2hctx_rpseq = 0;
539 
540 				ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
541 			}
542 		}
543 	}
544 
545 	/* check forward path congestion */
546 	/* still didn't send out new data packets */
547 	if (hctx->ccid2hctx_seqh == hctx->ccid2hctx_seqt)
548 		return;
549 
550 	switch (DCCP_SKB_CB(skb)->dccpd_type) {
551 	case DCCP_PKT_ACK:
552 	case DCCP_PKT_DATAACK:
553 		break;
554 	default:
555 		return;
556 	}
557 
558 	ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
559 	if (after48(ackno, hctx->ccid2hctx_high_ack))
560 		hctx->ccid2hctx_high_ack = ackno;
561 
562 	seqp = hctx->ccid2hctx_seqt;
563 	while (before48(seqp->ccid2s_seq, ackno)) {
564 		seqp = seqp->ccid2s_next;
565 		if (seqp == hctx->ccid2hctx_seqh) {
566 			seqp = hctx->ccid2hctx_seqh->ccid2s_prev;
567 			break;
568 		}
569 	}
570 
571 	/*
572 	 * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
573 	 * packets per acknowledgement. Rounding up avoids that cwnd is not
574 	 * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
575 	 */
576 	if (hctx->ccid2hctx_cwnd < hctx->ccid2hctx_ssthresh)
577 		maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
578 
579 	/* go through all ack vectors */
580 	while ((offset = ccid2_ackvector(sk, skb, offset,
581 					 &vector, &veclen)) != -1) {
582 		/* go through this ack vector */
583 		while (veclen--) {
584 			const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK;
585 			u64 ackno_end_rl = SUB48(ackno, rl);
586 
587 			ccid2_pr_debug("ackvec start:%llu end:%llu\n",
588 				       (unsigned long long)ackno,
589 				       (unsigned long long)ackno_end_rl);
590 			/* if the seqno we are analyzing is larger than the
591 			 * current ackno, then move towards the tail of our
592 			 * seqnos.
593 			 */
594 			while (after48(seqp->ccid2s_seq, ackno)) {
595 				if (seqp == hctx->ccid2hctx_seqt) {
596 					done = 1;
597 					break;
598 				}
599 				seqp = seqp->ccid2s_prev;
600 			}
601 			if (done)
602 				break;
603 
604 			/* check all seqnos in the range of the vector
605 			 * run length
606 			 */
607 			while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
608 				const u8 state = *vector &
609 						 DCCP_ACKVEC_STATE_MASK;
610 
611 				/* new packet received or marked */
612 				if (state != DCCP_ACKVEC_STATE_NOT_RECEIVED &&
613 				    !seqp->ccid2s_acked) {
614 					if (state ==
615 					    DCCP_ACKVEC_STATE_ECN_MARKED) {
616 						ccid2_congestion_event(sk,
617 								       seqp);
618 					} else
619 						ccid2_new_ack(sk, seqp,
620 							      &maxincr);
621 
622 					seqp->ccid2s_acked = 1;
623 					ccid2_pr_debug("Got ack for %llu\n",
624 						       (unsigned long long)seqp->ccid2s_seq);
625 					ccid2_hc_tx_dec_pipe(sk);
626 				}
627 				if (seqp == hctx->ccid2hctx_seqt) {
628 					done = 1;
629 					break;
630 				}
631 				seqp = seqp->ccid2s_prev;
632 			}
633 			if (done)
634 				break;
635 
636 			ackno = SUB48(ackno_end_rl, 1);
637 			vector++;
638 		}
639 		if (done)
640 			break;
641 	}
642 
643 	/* The state about what is acked should be correct now
644 	 * Check for NUMDUPACK
645 	 */
646 	seqp = hctx->ccid2hctx_seqt;
647 	while (before48(seqp->ccid2s_seq, hctx->ccid2hctx_high_ack)) {
648 		seqp = seqp->ccid2s_next;
649 		if (seqp == hctx->ccid2hctx_seqh) {
650 			seqp = hctx->ccid2hctx_seqh->ccid2s_prev;
651 			break;
652 		}
653 	}
654 	done = 0;
655 	while (1) {
656 		if (seqp->ccid2s_acked) {
657 			done++;
658 			if (done == NUMDUPACK)
659 				break;
660 		}
661 		if (seqp == hctx->ccid2hctx_seqt)
662 			break;
663 		seqp = seqp->ccid2s_prev;
664 	}
665 
666 	/* If there are at least 3 acknowledgements, anything unacknowledged
667 	 * below the last sequence number is considered lost
668 	 */
669 	if (done == NUMDUPACK) {
670 		struct ccid2_seq *last_acked = seqp;
671 
672 		/* check for lost packets */
673 		while (1) {
674 			if (!seqp->ccid2s_acked) {
675 				ccid2_pr_debug("Packet lost: %llu\n",
676 					       (unsigned long long)seqp->ccid2s_seq);
677 				/* XXX need to traverse from tail -> head in
678 				 * order to detect multiple congestion events in
679 				 * one ack vector.
680 				 */
681 				ccid2_congestion_event(sk, seqp);
682 				ccid2_hc_tx_dec_pipe(sk);
683 			}
684 			if (seqp == hctx->ccid2hctx_seqt)
685 				break;
686 			seqp = seqp->ccid2s_prev;
687 		}
688 
689 		hctx->ccid2hctx_seqt = last_acked;
690 	}
691 
692 	/* trim acked packets in tail */
693 	while (hctx->ccid2hctx_seqt != hctx->ccid2hctx_seqh) {
694 		if (!hctx->ccid2hctx_seqt->ccid2s_acked)
695 			break;
696 
697 		hctx->ccid2hctx_seqt = hctx->ccid2hctx_seqt->ccid2s_next;
698 	}
699 
700 	ccid2_hc_tx_check_sanity(hctx);
701 }
702 
703 static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
704 {
705 	struct ccid2_hc_tx_sock *hctx = ccid_priv(ccid);
706 	struct dccp_sock *dp = dccp_sk(sk);
707 	u32 max_ratio;
708 
709 	/* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
710 	hctx->ccid2hctx_ssthresh  = ~0U;
711 
712 	/*
713 	 * RFC 4341, 5: "The cwnd parameter is initialized to at most four
714 	 * packets for new connections, following the rules from [RFC3390]".
715 	 * We need to convert the bytes of RFC3390 into the packets of RFC 4341.
716 	 */
717 	hctx->ccid2hctx_cwnd = clamp(4380U / dp->dccps_mss_cache, 2U, 4U);
718 
719 	/* Make sure that Ack Ratio is enabled and within bounds. */
720 	max_ratio = DIV_ROUND_UP(hctx->ccid2hctx_cwnd, 2);
721 	if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
722 		dp->dccps_l_ack_ratio = max_ratio;
723 
724 	/* XXX init ~ to window size... */
725 	if (ccid2_hc_tx_alloc_seq(hctx))
726 		return -ENOMEM;
727 
728 	hctx->ccid2hctx_rto	 = 3 * HZ;
729 	ccid2_change_srtt(hctx, -1);
730 	hctx->ccid2hctx_rttvar	 = -1;
731 	hctx->ccid2hctx_rpdupack = -1;
732 	hctx->ccid2hctx_last_cong = jiffies;
733 	setup_timer(&hctx->ccid2hctx_rtotimer, ccid2_hc_tx_rto_expire,
734 			(unsigned long)sk);
735 
736 	ccid2_hc_tx_check_sanity(hctx);
737 	return 0;
738 }
739 
740 static void ccid2_hc_tx_exit(struct sock *sk)
741 {
742 	struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
743 	int i;
744 
745 	ccid2_hc_tx_kill_rto_timer(sk);
746 
747 	for (i = 0; i < hctx->ccid2hctx_seqbufc; i++)
748 		kfree(hctx->ccid2hctx_seqbuf[i]);
749 	hctx->ccid2hctx_seqbufc = 0;
750 }
751 
752 static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
753 {
754 	const struct dccp_sock *dp = dccp_sk(sk);
755 	struct ccid2_hc_rx_sock *hcrx = ccid2_hc_rx_sk(sk);
756 
757 	switch (DCCP_SKB_CB(skb)->dccpd_type) {
758 	case DCCP_PKT_DATA:
759 	case DCCP_PKT_DATAACK:
760 		hcrx->ccid2hcrx_data++;
761 		if (hcrx->ccid2hcrx_data >= dp->dccps_r_ack_ratio) {
762 			dccp_send_ack(sk);
763 			hcrx->ccid2hcrx_data = 0;
764 		}
765 		break;
766 	}
767 }
768 
769 struct ccid_operations ccid2_ops = {
770 	.ccid_id		= DCCPC_CCID2,
771 	.ccid_name		= "TCP-like",
772 	.ccid_hc_tx_obj_size	= sizeof(struct ccid2_hc_tx_sock),
773 	.ccid_hc_tx_init	= ccid2_hc_tx_init,
774 	.ccid_hc_tx_exit	= ccid2_hc_tx_exit,
775 	.ccid_hc_tx_send_packet	= ccid2_hc_tx_send_packet,
776 	.ccid_hc_tx_packet_sent	= ccid2_hc_tx_packet_sent,
777 	.ccid_hc_tx_packet_recv	= ccid2_hc_tx_packet_recv,
778 	.ccid_hc_rx_obj_size	= sizeof(struct ccid2_hc_rx_sock),
779 	.ccid_hc_rx_packet_recv	= ccid2_hc_rx_packet_recv,
780 };
781 
782 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
783 module_param(ccid2_debug, bool, 0644);
784 MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages");
785 #endif
786