xref: /linux/net/dccp/ccids/ccid2.c (revision 791d3ef2e11100449837dc0b6fe884e60ca3a484)
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 <linux/slab.h>
27 #include "../feat.h"
28 #include "ccid2.h"
29 
30 
31 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
32 static bool ccid2_debug;
33 #define ccid2_pr_debug(format, a...)	DCCP_PR_DEBUG(ccid2_debug, format, ##a)
34 #else
35 #define ccid2_pr_debug(format, a...)
36 #endif
37 
38 static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hc)
39 {
40 	struct ccid2_seq *seqp;
41 	int i;
42 
43 	/* check if we have space to preserve the pointer to the buffer */
44 	if (hc->tx_seqbufc >= (sizeof(hc->tx_seqbuf) /
45 			       sizeof(struct ccid2_seq *)))
46 		return -ENOMEM;
47 
48 	/* allocate buffer and initialize linked list */
49 	seqp = kmalloc_array(CCID2_SEQBUF_LEN, sizeof(struct ccid2_seq),
50 			     gfp_any());
51 	if (seqp == NULL)
52 		return -ENOMEM;
53 
54 	for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) {
55 		seqp[i].ccid2s_next = &seqp[i + 1];
56 		seqp[i + 1].ccid2s_prev = &seqp[i];
57 	}
58 	seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp;
59 	seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
60 
61 	/* This is the first allocation.  Initiate the head and tail.  */
62 	if (hc->tx_seqbufc == 0)
63 		hc->tx_seqh = hc->tx_seqt = seqp;
64 	else {
65 		/* link the existing list with the one we just created */
66 		hc->tx_seqh->ccid2s_next = seqp;
67 		seqp->ccid2s_prev = hc->tx_seqh;
68 
69 		hc->tx_seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
70 		seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hc->tx_seqt;
71 	}
72 
73 	/* store the original pointer to the buffer so we can free it */
74 	hc->tx_seqbuf[hc->tx_seqbufc] = seqp;
75 	hc->tx_seqbufc++;
76 
77 	return 0;
78 }
79 
80 static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
81 {
82 	if (ccid2_cwnd_network_limited(ccid2_hc_tx_sk(sk)))
83 		return CCID_PACKET_WILL_DEQUEUE_LATER;
84 	return CCID_PACKET_SEND_AT_ONCE;
85 }
86 
87 static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
88 {
89 	u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->tx_cwnd, 2);
90 
91 	/*
92 	 * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from
93 	 * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always
94 	 * acceptable since this causes starvation/deadlock whenever cwnd < 2.
95 	 * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled).
96 	 */
97 	if (val == 0 || val > max_ratio) {
98 		DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio);
99 		val = max_ratio;
100 	}
101 	dccp_feat_signal_nn_change(sk, DCCPF_ACK_RATIO,
102 				   min_t(u32, val, DCCPF_ACK_RATIO_MAX));
103 }
104 
105 static void ccid2_check_l_ack_ratio(struct sock *sk)
106 {
107 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
108 
109 	/*
110 	 * After a loss, idle period, application limited period, or RTO we
111 	 * need to check that the ack ratio is still less than the congestion
112 	 * window. Otherwise, we will send an entire congestion window of
113 	 * packets and got no response because we haven't sent ack ratio
114 	 * packets yet.
115 	 * If the ack ratio does need to be reduced, we reduce it to half of
116 	 * the congestion window (or 1 if that's zero) instead of to the
117 	 * congestion window. This prevents problems if one ack is lost.
118 	 */
119 	if (dccp_feat_nn_get(sk, DCCPF_ACK_RATIO) > hc->tx_cwnd)
120 		ccid2_change_l_ack_ratio(sk, hc->tx_cwnd/2 ? : 1U);
121 }
122 
123 static void ccid2_change_l_seq_window(struct sock *sk, u64 val)
124 {
125 	dccp_feat_signal_nn_change(sk, DCCPF_SEQUENCE_WINDOW,
126 				   clamp_val(val, DCCPF_SEQ_WMIN,
127 						  DCCPF_SEQ_WMAX));
128 }
129 
130 static void dccp_tasklet_schedule(struct sock *sk)
131 {
132 	struct tasklet_struct *t = &dccp_sk(sk)->dccps_xmitlet;
133 
134 	if (!test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
135 		sock_hold(sk);
136 		__tasklet_schedule(t);
137 	}
138 }
139 
140 static void ccid2_hc_tx_rto_expire(struct timer_list *t)
141 {
142 	struct ccid2_hc_tx_sock *hc = from_timer(hc, t, tx_rtotimer);
143 	struct sock *sk = hc->sk;
144 	const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
145 
146 	bh_lock_sock(sk);
147 	if (sock_owned_by_user(sk)) {
148 		sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5);
149 		goto out;
150 	}
151 
152 	ccid2_pr_debug("RTO_EXPIRE\n");
153 
154 	if (sk->sk_state == DCCP_CLOSED)
155 		goto out;
156 
157 	/* back-off timer */
158 	hc->tx_rto <<= 1;
159 	if (hc->tx_rto > DCCP_RTO_MAX)
160 		hc->tx_rto = DCCP_RTO_MAX;
161 
162 	/* adjust pipe, cwnd etc */
163 	hc->tx_ssthresh = hc->tx_cwnd / 2;
164 	if (hc->tx_ssthresh < 2)
165 		hc->tx_ssthresh = 2;
166 	hc->tx_cwnd	= 1;
167 	hc->tx_pipe	= 0;
168 
169 	/* clear state about stuff we sent */
170 	hc->tx_seqt = hc->tx_seqh;
171 	hc->tx_packets_acked = 0;
172 
173 	/* clear ack ratio state. */
174 	hc->tx_rpseq    = 0;
175 	hc->tx_rpdupack = -1;
176 	ccid2_change_l_ack_ratio(sk, 1);
177 
178 	/* if we were blocked before, we may now send cwnd=1 packet */
179 	if (sender_was_blocked)
180 		dccp_tasklet_schedule(sk);
181 	/* restart backed-off timer */
182 	sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
183 out:
184 	bh_unlock_sock(sk);
185 	sock_put(sk);
186 }
187 
188 /*
189  *	Congestion window validation (RFC 2861).
190  */
191 static bool ccid2_do_cwv = true;
192 module_param(ccid2_do_cwv, bool, 0644);
193 MODULE_PARM_DESC(ccid2_do_cwv, "Perform RFC2861 Congestion Window Validation");
194 
195 /**
196  * ccid2_update_used_window  -  Track how much of cwnd is actually used
197  * This is done in addition to CWV. The sender needs to have an idea of how many
198  * packets may be in flight, to set the local Sequence Window value accordingly
199  * (RFC 4340, 7.5.2). The CWV mechanism is exploited to keep track of the
200  * maximum-used window. We use an EWMA low-pass filter to filter out noise.
201  */
202 static void ccid2_update_used_window(struct ccid2_hc_tx_sock *hc, u32 new_wnd)
203 {
204 	hc->tx_expected_wnd = (3 * hc->tx_expected_wnd + new_wnd) / 4;
205 }
206 
207 /* This borrows the code of tcp_cwnd_application_limited() */
208 static void ccid2_cwnd_application_limited(struct sock *sk, const u32 now)
209 {
210 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
211 	/* don't reduce cwnd below the initial window (IW) */
212 	u32 init_win = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache),
213 	    win_used = max(hc->tx_cwnd_used, init_win);
214 
215 	if (win_used < hc->tx_cwnd) {
216 		hc->tx_ssthresh = max(hc->tx_ssthresh,
217 				     (hc->tx_cwnd >> 1) + (hc->tx_cwnd >> 2));
218 		hc->tx_cwnd = (hc->tx_cwnd + win_used) >> 1;
219 	}
220 	hc->tx_cwnd_used  = 0;
221 	hc->tx_cwnd_stamp = now;
222 
223 	ccid2_check_l_ack_ratio(sk);
224 }
225 
226 /* This borrows the code of tcp_cwnd_restart() */
227 static void ccid2_cwnd_restart(struct sock *sk, const u32 now)
228 {
229 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
230 	u32 cwnd = hc->tx_cwnd, restart_cwnd,
231 	    iwnd = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache);
232 
233 	hc->tx_ssthresh = max(hc->tx_ssthresh, (cwnd >> 1) + (cwnd >> 2));
234 
235 	/* don't reduce cwnd below the initial window (IW) */
236 	restart_cwnd = min(cwnd, iwnd);
237 	cwnd >>= (now - hc->tx_lsndtime) / hc->tx_rto;
238 	hc->tx_cwnd = max(cwnd, restart_cwnd);
239 
240 	hc->tx_cwnd_stamp = now;
241 	hc->tx_cwnd_used  = 0;
242 
243 	ccid2_check_l_ack_ratio(sk);
244 }
245 
246 static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
247 {
248 	struct dccp_sock *dp = dccp_sk(sk);
249 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
250 	const u32 now = ccid2_jiffies32;
251 	struct ccid2_seq *next;
252 
253 	/* slow-start after idle periods (RFC 2581, RFC 2861) */
254 	if (ccid2_do_cwv && !hc->tx_pipe &&
255 	    (s32)(now - hc->tx_lsndtime) >= hc->tx_rto)
256 		ccid2_cwnd_restart(sk, now);
257 
258 	hc->tx_lsndtime = now;
259 	hc->tx_pipe    += 1;
260 
261 	/* see whether cwnd was fully used (RFC 2861), update expected window */
262 	if (ccid2_cwnd_network_limited(hc)) {
263 		ccid2_update_used_window(hc, hc->tx_cwnd);
264 		hc->tx_cwnd_used  = 0;
265 		hc->tx_cwnd_stamp = now;
266 	} else {
267 		if (hc->tx_pipe > hc->tx_cwnd_used)
268 			hc->tx_cwnd_used = hc->tx_pipe;
269 
270 		ccid2_update_used_window(hc, hc->tx_cwnd_used);
271 
272 		if (ccid2_do_cwv && (s32)(now - hc->tx_cwnd_stamp) >= hc->tx_rto)
273 			ccid2_cwnd_application_limited(sk, now);
274 	}
275 
276 	hc->tx_seqh->ccid2s_seq   = dp->dccps_gss;
277 	hc->tx_seqh->ccid2s_acked = 0;
278 	hc->tx_seqh->ccid2s_sent  = now;
279 
280 	next = hc->tx_seqh->ccid2s_next;
281 	/* check if we need to alloc more space */
282 	if (next == hc->tx_seqt) {
283 		if (ccid2_hc_tx_alloc_seq(hc)) {
284 			DCCP_CRIT("packet history - out of memory!");
285 			/* FIXME: find a more graceful way to bail out */
286 			return;
287 		}
288 		next = hc->tx_seqh->ccid2s_next;
289 		BUG_ON(next == hc->tx_seqt);
290 	}
291 	hc->tx_seqh = next;
292 
293 	ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe);
294 
295 	/*
296 	 * FIXME: The code below is broken and the variables have been removed
297 	 * from the socket struct. The `ackloss' variable was always set to 0,
298 	 * and with arsent there are several problems:
299 	 *  (i) it doesn't just count the number of Acks, but all sent packets;
300 	 *  (ii) it is expressed in # of packets, not # of windows, so the
301 	 *  comparison below uses the wrong formula: Appendix A of RFC 4341
302 	 *  comes up with the number K = cwnd / (R^2 - R) of consecutive windows
303 	 *  of data with no lost or marked Ack packets. If arsent were the # of
304 	 *  consecutive Acks received without loss, then Ack Ratio needs to be
305 	 *  decreased by 1 when
306 	 *	      arsent >=  K * cwnd / R  =  cwnd^2 / (R^3 - R^2)
307 	 *  where cwnd / R is the number of Acks received per window of data
308 	 *  (cf. RFC 4341, App. A). The problems are that
309 	 *  - arsent counts other packets as well;
310 	 *  - the comparison uses a formula different from RFC 4341;
311 	 *  - computing a cubic/quadratic equation each time is too complicated.
312 	 *  Hence a different algorithm is needed.
313 	 */
314 #if 0
315 	/* Ack Ratio.  Need to maintain a concept of how many windows we sent */
316 	hc->tx_arsent++;
317 	/* We had an ack loss in this window... */
318 	if (hc->tx_ackloss) {
319 		if (hc->tx_arsent >= hc->tx_cwnd) {
320 			hc->tx_arsent  = 0;
321 			hc->tx_ackloss = 0;
322 		}
323 	} else {
324 		/* No acks lost up to now... */
325 		/* decrease ack ratio if enough packets were sent */
326 		if (dp->dccps_l_ack_ratio > 1) {
327 			/* XXX don't calculate denominator each time */
328 			int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
329 				    dp->dccps_l_ack_ratio;
330 
331 			denom = hc->tx_cwnd * hc->tx_cwnd / denom;
332 
333 			if (hc->tx_arsent >= denom) {
334 				ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
335 				hc->tx_arsent = 0;
336 			}
337 		} else {
338 			/* we can't increase ack ratio further [1] */
339 			hc->tx_arsent = 0; /* or maybe set it to cwnd*/
340 		}
341 	}
342 #endif
343 
344 	sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
345 
346 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
347 	do {
348 		struct ccid2_seq *seqp = hc->tx_seqt;
349 
350 		while (seqp != hc->tx_seqh) {
351 			ccid2_pr_debug("out seq=%llu acked=%d time=%u\n",
352 				       (unsigned long long)seqp->ccid2s_seq,
353 				       seqp->ccid2s_acked, seqp->ccid2s_sent);
354 			seqp = seqp->ccid2s_next;
355 		}
356 	} while (0);
357 	ccid2_pr_debug("=========\n");
358 #endif
359 }
360 
361 /**
362  * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
363  * This code is almost identical with TCP's tcp_rtt_estimator(), since
364  * - it has a higher sampling frequency (recommended by RFC 1323),
365  * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
366  * - it is simple (cf. more complex proposals such as Eifel timer or research
367  *   which suggests that the gain should be set according to window size),
368  * - in tests it was found to work well with CCID2 [gerrit].
369  */
370 static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
371 {
372 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
373 	long m = mrtt ? : 1;
374 
375 	if (hc->tx_srtt == 0) {
376 		/* First measurement m */
377 		hc->tx_srtt = m << 3;
378 		hc->tx_mdev = m << 1;
379 
380 		hc->tx_mdev_max = max(hc->tx_mdev, tcp_rto_min(sk));
381 		hc->tx_rttvar   = hc->tx_mdev_max;
382 
383 		hc->tx_rtt_seq  = dccp_sk(sk)->dccps_gss;
384 	} else {
385 		/* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
386 		m -= (hc->tx_srtt >> 3);
387 		hc->tx_srtt += m;
388 
389 		/* Similarly, update scaled mdev with regard to |m| */
390 		if (m < 0) {
391 			m = -m;
392 			m -= (hc->tx_mdev >> 2);
393 			/*
394 			 * This neutralises RTO increase when RTT < SRTT - mdev
395 			 * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
396 			 * in Linux TCP", USENIX 2002, pp. 49-62).
397 			 */
398 			if (m > 0)
399 				m >>= 3;
400 		} else {
401 			m -= (hc->tx_mdev >> 2);
402 		}
403 		hc->tx_mdev += m;
404 
405 		if (hc->tx_mdev > hc->tx_mdev_max) {
406 			hc->tx_mdev_max = hc->tx_mdev;
407 			if (hc->tx_mdev_max > hc->tx_rttvar)
408 				hc->tx_rttvar = hc->tx_mdev_max;
409 		}
410 
411 		/*
412 		 * Decay RTTVAR at most once per flight, exploiting that
413 		 *  1) pipe <= cwnd <= Sequence_Window = W  (RFC 4340, 7.5.2)
414 		 *  2) AWL = GSS-W+1 <= GAR <= GSS          (RFC 4340, 7.5.1)
415 		 * GAR is a useful bound for FlightSize = pipe.
416 		 * AWL is probably too low here, as it over-estimates pipe.
417 		 */
418 		if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) {
419 			if (hc->tx_mdev_max < hc->tx_rttvar)
420 				hc->tx_rttvar -= (hc->tx_rttvar -
421 						  hc->tx_mdev_max) >> 2;
422 			hc->tx_rtt_seq  = dccp_sk(sk)->dccps_gss;
423 			hc->tx_mdev_max = tcp_rto_min(sk);
424 		}
425 	}
426 
427 	/*
428 	 * Set RTO from SRTT and RTTVAR
429 	 * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms.
430 	 * This agrees with RFC 4341, 5:
431 	 *	"Because DCCP does not retransmit data, DCCP does not require
432 	 *	 TCP's recommended minimum timeout of one second".
433 	 */
434 	hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar;
435 
436 	if (hc->tx_rto > DCCP_RTO_MAX)
437 		hc->tx_rto = DCCP_RTO_MAX;
438 }
439 
440 static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
441 			  unsigned int *maxincr)
442 {
443 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
444 	struct dccp_sock *dp = dccp_sk(sk);
445 	int r_seq_used = hc->tx_cwnd / dp->dccps_l_ack_ratio;
446 
447 	if (hc->tx_cwnd < dp->dccps_l_seq_win &&
448 	    r_seq_used < dp->dccps_r_seq_win) {
449 		if (hc->tx_cwnd < hc->tx_ssthresh) {
450 			if (*maxincr > 0 && ++hc->tx_packets_acked >= 2) {
451 				hc->tx_cwnd += 1;
452 				*maxincr    -= 1;
453 				hc->tx_packets_acked = 0;
454 			}
455 		} else if (++hc->tx_packets_acked >= hc->tx_cwnd) {
456 			hc->tx_cwnd += 1;
457 			hc->tx_packets_acked = 0;
458 		}
459 	}
460 
461 	/*
462 	 * Adjust the local sequence window and the ack ratio to allow about
463 	 * 5 times the number of packets in the network (RFC 4340 7.5.2)
464 	 */
465 	if (r_seq_used * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_r_seq_win)
466 		ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio * 2);
467 	else if (r_seq_used * CCID2_WIN_CHANGE_FACTOR < dp->dccps_r_seq_win/2)
468 		ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio / 2 ? : 1U);
469 
470 	if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_l_seq_win)
471 		ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win * 2);
472 	else if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR < dp->dccps_l_seq_win/2)
473 		ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win / 2);
474 
475 	/*
476 	 * FIXME: RTT is sampled several times per acknowledgment (for each
477 	 * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
478 	 * This causes the RTT to be over-estimated, since the older entries
479 	 * in the Ack Vector have earlier sending times.
480 	 * The cleanest solution is to not use the ccid2s_sent field at all
481 	 * and instead use DCCP timestamps: requires changes in other places.
482 	 */
483 	ccid2_rtt_estimator(sk, ccid2_jiffies32 - seqp->ccid2s_sent);
484 }
485 
486 static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
487 {
488 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
489 
490 	if ((s32)(seqp->ccid2s_sent - hc->tx_last_cong) < 0) {
491 		ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
492 		return;
493 	}
494 
495 	hc->tx_last_cong = ccid2_jiffies32;
496 
497 	hc->tx_cwnd      = hc->tx_cwnd / 2 ? : 1U;
498 	hc->tx_ssthresh  = max(hc->tx_cwnd, 2U);
499 
500 	ccid2_check_l_ack_ratio(sk);
501 }
502 
503 static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
504 				     u8 option, u8 *optval, u8 optlen)
505 {
506 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
507 
508 	switch (option) {
509 	case DCCPO_ACK_VECTOR_0:
510 	case DCCPO_ACK_VECTOR_1:
511 		return dccp_ackvec_parsed_add(&hc->tx_av_chunks, optval, optlen,
512 					      option - DCCPO_ACK_VECTOR_0);
513 	}
514 	return 0;
515 }
516 
517 static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
518 {
519 	struct dccp_sock *dp = dccp_sk(sk);
520 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
521 	const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
522 	struct dccp_ackvec_parsed *avp;
523 	u64 ackno, seqno;
524 	struct ccid2_seq *seqp;
525 	int done = 0;
526 	unsigned int maxincr = 0;
527 
528 	/* check reverse path congestion */
529 	seqno = DCCP_SKB_CB(skb)->dccpd_seq;
530 
531 	/* XXX this whole "algorithm" is broken.  Need to fix it to keep track
532 	 * of the seqnos of the dupacks so that rpseq and rpdupack are correct
533 	 * -sorbo.
534 	 */
535 	/* need to bootstrap */
536 	if (hc->tx_rpdupack == -1) {
537 		hc->tx_rpdupack = 0;
538 		hc->tx_rpseq    = seqno;
539 	} else {
540 		/* check if packet is consecutive */
541 		if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1)
542 			hc->tx_rpseq = seqno;
543 		/* it's a later packet */
544 		else if (after48(seqno, hc->tx_rpseq)) {
545 			hc->tx_rpdupack++;
546 
547 			/* check if we got enough dupacks */
548 			if (hc->tx_rpdupack >= NUMDUPACK) {
549 				hc->tx_rpdupack = -1; /* XXX lame */
550 				hc->tx_rpseq    = 0;
551 #ifdef __CCID2_COPES_GRACEFULLY_WITH_ACK_CONGESTION_CONTROL__
552 				/*
553 				 * FIXME: Ack Congestion Control is broken; in
554 				 * the current state instabilities occurred with
555 				 * Ack Ratios greater than 1; causing hang-ups
556 				 * and long RTO timeouts. This needs to be fixed
557 				 * before opening up dynamic changes. -- gerrit
558 				 */
559 				ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
560 #endif
561 			}
562 		}
563 	}
564 
565 	/* check forward path congestion */
566 	if (dccp_packet_without_ack(skb))
567 		return;
568 
569 	/* still didn't send out new data packets */
570 	if (hc->tx_seqh == hc->tx_seqt)
571 		goto done;
572 
573 	ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
574 	if (after48(ackno, hc->tx_high_ack))
575 		hc->tx_high_ack = ackno;
576 
577 	seqp = hc->tx_seqt;
578 	while (before48(seqp->ccid2s_seq, ackno)) {
579 		seqp = seqp->ccid2s_next;
580 		if (seqp == hc->tx_seqh) {
581 			seqp = hc->tx_seqh->ccid2s_prev;
582 			break;
583 		}
584 	}
585 
586 	/*
587 	 * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
588 	 * packets per acknowledgement. Rounding up avoids that cwnd is not
589 	 * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
590 	 */
591 	if (hc->tx_cwnd < hc->tx_ssthresh)
592 		maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
593 
594 	/* go through all ack vectors */
595 	list_for_each_entry(avp, &hc->tx_av_chunks, node) {
596 		/* go through this ack vector */
597 		for (; avp->len--; avp->vec++) {
598 			u64 ackno_end_rl = SUB48(ackno,
599 						 dccp_ackvec_runlen(avp->vec));
600 
601 			ccid2_pr_debug("ackvec %llu |%u,%u|\n",
602 				       (unsigned long long)ackno,
603 				       dccp_ackvec_state(avp->vec) >> 6,
604 				       dccp_ackvec_runlen(avp->vec));
605 			/* if the seqno we are analyzing is larger than the
606 			 * current ackno, then move towards the tail of our
607 			 * seqnos.
608 			 */
609 			while (after48(seqp->ccid2s_seq, ackno)) {
610 				if (seqp == hc->tx_seqt) {
611 					done = 1;
612 					break;
613 				}
614 				seqp = seqp->ccid2s_prev;
615 			}
616 			if (done)
617 				break;
618 
619 			/* check all seqnos in the range of the vector
620 			 * run length
621 			 */
622 			while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
623 				const u8 state = dccp_ackvec_state(avp->vec);
624 
625 				/* new packet received or marked */
626 				if (state != DCCPAV_NOT_RECEIVED &&
627 				    !seqp->ccid2s_acked) {
628 					if (state == DCCPAV_ECN_MARKED)
629 						ccid2_congestion_event(sk,
630 								       seqp);
631 					else
632 						ccid2_new_ack(sk, seqp,
633 							      &maxincr);
634 
635 					seqp->ccid2s_acked = 1;
636 					ccid2_pr_debug("Got ack for %llu\n",
637 						       (unsigned long long)seqp->ccid2s_seq);
638 					hc->tx_pipe--;
639 				}
640 				if (seqp == hc->tx_seqt) {
641 					done = 1;
642 					break;
643 				}
644 				seqp = seqp->ccid2s_prev;
645 			}
646 			if (done)
647 				break;
648 
649 			ackno = SUB48(ackno_end_rl, 1);
650 		}
651 		if (done)
652 			break;
653 	}
654 
655 	/* The state about what is acked should be correct now
656 	 * Check for NUMDUPACK
657 	 */
658 	seqp = hc->tx_seqt;
659 	while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) {
660 		seqp = seqp->ccid2s_next;
661 		if (seqp == hc->tx_seqh) {
662 			seqp = hc->tx_seqh->ccid2s_prev;
663 			break;
664 		}
665 	}
666 	done = 0;
667 	while (1) {
668 		if (seqp->ccid2s_acked) {
669 			done++;
670 			if (done == NUMDUPACK)
671 				break;
672 		}
673 		if (seqp == hc->tx_seqt)
674 			break;
675 		seqp = seqp->ccid2s_prev;
676 	}
677 
678 	/* If there are at least 3 acknowledgements, anything unacknowledged
679 	 * below the last sequence number is considered lost
680 	 */
681 	if (done == NUMDUPACK) {
682 		struct ccid2_seq *last_acked = seqp;
683 
684 		/* check for lost packets */
685 		while (1) {
686 			if (!seqp->ccid2s_acked) {
687 				ccid2_pr_debug("Packet lost: %llu\n",
688 					       (unsigned long long)seqp->ccid2s_seq);
689 				/* XXX need to traverse from tail -> head in
690 				 * order to detect multiple congestion events in
691 				 * one ack vector.
692 				 */
693 				ccid2_congestion_event(sk, seqp);
694 				hc->tx_pipe--;
695 			}
696 			if (seqp == hc->tx_seqt)
697 				break;
698 			seqp = seqp->ccid2s_prev;
699 		}
700 
701 		hc->tx_seqt = last_acked;
702 	}
703 
704 	/* trim acked packets in tail */
705 	while (hc->tx_seqt != hc->tx_seqh) {
706 		if (!hc->tx_seqt->ccid2s_acked)
707 			break;
708 
709 		hc->tx_seqt = hc->tx_seqt->ccid2s_next;
710 	}
711 
712 	/* restart RTO timer if not all outstanding data has been acked */
713 	if (hc->tx_pipe == 0)
714 		sk_stop_timer(sk, &hc->tx_rtotimer);
715 	else
716 		sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
717 done:
718 	/* check if incoming Acks allow pending packets to be sent */
719 	if (sender_was_blocked && !ccid2_cwnd_network_limited(hc))
720 		dccp_tasklet_schedule(sk);
721 	dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
722 }
723 
724 static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
725 {
726 	struct ccid2_hc_tx_sock *hc = ccid_priv(ccid);
727 	struct dccp_sock *dp = dccp_sk(sk);
728 	u32 max_ratio;
729 
730 	/* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
731 	hc->tx_ssthresh = ~0U;
732 
733 	/* Use larger initial windows (RFC 4341, section 5). */
734 	hc->tx_cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache);
735 	hc->tx_expected_wnd = hc->tx_cwnd;
736 
737 	/* Make sure that Ack Ratio is enabled and within bounds. */
738 	max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2);
739 	if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
740 		dp->dccps_l_ack_ratio = max_ratio;
741 
742 	/* XXX init ~ to window size... */
743 	if (ccid2_hc_tx_alloc_seq(hc))
744 		return -ENOMEM;
745 
746 	hc->tx_rto	 = DCCP_TIMEOUT_INIT;
747 	hc->tx_rpdupack  = -1;
748 	hc->tx_last_cong = hc->tx_lsndtime = hc->tx_cwnd_stamp = ccid2_jiffies32;
749 	hc->tx_cwnd_used = 0;
750 	hc->sk		 = sk;
751 	timer_setup(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire, 0);
752 	INIT_LIST_HEAD(&hc->tx_av_chunks);
753 	return 0;
754 }
755 
756 static void ccid2_hc_tx_exit(struct sock *sk)
757 {
758 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
759 	int i;
760 
761 	sk_stop_timer(sk, &hc->tx_rtotimer);
762 
763 	for (i = 0; i < hc->tx_seqbufc; i++)
764 		kfree(hc->tx_seqbuf[i]);
765 	hc->tx_seqbufc = 0;
766 	dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
767 }
768 
769 static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
770 {
771 	struct ccid2_hc_rx_sock *hc = ccid2_hc_rx_sk(sk);
772 
773 	if (!dccp_data_packet(skb))
774 		return;
775 
776 	if (++hc->rx_num_data_pkts >= dccp_sk(sk)->dccps_r_ack_ratio) {
777 		dccp_send_ack(sk);
778 		hc->rx_num_data_pkts = 0;
779 	}
780 }
781 
782 struct ccid_operations ccid2_ops = {
783 	.ccid_id		  = DCCPC_CCID2,
784 	.ccid_name		  = "TCP-like",
785 	.ccid_hc_tx_obj_size	  = sizeof(struct ccid2_hc_tx_sock),
786 	.ccid_hc_tx_init	  = ccid2_hc_tx_init,
787 	.ccid_hc_tx_exit	  = ccid2_hc_tx_exit,
788 	.ccid_hc_tx_send_packet	  = ccid2_hc_tx_send_packet,
789 	.ccid_hc_tx_packet_sent	  = ccid2_hc_tx_packet_sent,
790 	.ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
791 	.ccid_hc_tx_packet_recv	  = ccid2_hc_tx_packet_recv,
792 	.ccid_hc_rx_obj_size	  = sizeof(struct ccid2_hc_rx_sock),
793 	.ccid_hc_rx_packet_recv	  = ccid2_hc_rx_packet_recv,
794 };
795 
796 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
797 module_param(ccid2_debug, bool, 0644);
798 MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages");
799 #endif
800