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 
ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock * hc)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(CCID2_SEQBUF_LEN * sizeof(struct ccid2_seq), gfp_any());
50 	if (seqp == NULL)
51 		return -ENOMEM;
52 
53 	for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) {
54 		seqp[i].ccid2s_next = &seqp[i + 1];
55 		seqp[i + 1].ccid2s_prev = &seqp[i];
56 	}
57 	seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp;
58 	seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
59 
60 	/* This is the first allocation.  Initiate the head and tail.  */
61 	if (hc->tx_seqbufc == 0)
62 		hc->tx_seqh = hc->tx_seqt = seqp;
63 	else {
64 		/* link the existing list with the one we just created */
65 		hc->tx_seqh->ccid2s_next = seqp;
66 		seqp->ccid2s_prev = hc->tx_seqh;
67 
68 		hc->tx_seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
69 		seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hc->tx_seqt;
70 	}
71 
72 	/* store the original pointer to the buffer so we can free it */
73 	hc->tx_seqbuf[hc->tx_seqbufc] = seqp;
74 	hc->tx_seqbufc++;
75 
76 	return 0;
77 }
78 
ccid2_hc_tx_send_packet(struct sock * sk,struct sk_buff * skb)79 static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
80 {
81 	if (ccid2_cwnd_network_limited(ccid2_hc_tx_sk(sk)))
82 		return CCID_PACKET_WILL_DEQUEUE_LATER;
83 	return CCID_PACKET_SEND_AT_ONCE;
84 }
85 
ccid2_change_l_ack_ratio(struct sock * sk,u32 val)86 static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
87 {
88 	u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->tx_cwnd, 2);
89 
90 	/*
91 	 * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from
92 	 * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always
93 	 * acceptable since this causes starvation/deadlock whenever cwnd < 2.
94 	 * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled).
95 	 */
96 	if (val == 0 || val > max_ratio) {
97 		DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio);
98 		val = max_ratio;
99 	}
100 	dccp_feat_signal_nn_change(sk, DCCPF_ACK_RATIO,
101 				   min_t(u32, val, DCCPF_ACK_RATIO_MAX));
102 }
103 
ccid2_check_l_ack_ratio(struct sock * sk)104 static void ccid2_check_l_ack_ratio(struct sock *sk)
105 {
106 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
107 
108 	/*
109 	 * After a loss, idle period, application limited period, or RTO we
110 	 * need to check that the ack ratio is still less than the congestion
111 	 * window. Otherwise, we will send an entire congestion window of
112 	 * packets and got no response because we haven't sent ack ratio
113 	 * packets yet.
114 	 * If the ack ratio does need to be reduced, we reduce it to half of
115 	 * the congestion window (or 1 if that's zero) instead of to the
116 	 * congestion window. This prevents problems if one ack is lost.
117 	 */
118 	if (dccp_feat_nn_get(sk, DCCPF_ACK_RATIO) > hc->tx_cwnd)
119 		ccid2_change_l_ack_ratio(sk, hc->tx_cwnd/2 ? : 1U);
120 }
121 
ccid2_change_l_seq_window(struct sock * sk,u64 val)122 static void ccid2_change_l_seq_window(struct sock *sk, u64 val)
123 {
124 	dccp_feat_signal_nn_change(sk, DCCPF_SEQUENCE_WINDOW,
125 				   clamp_val(val, DCCPF_SEQ_WMIN,
126 						  DCCPF_SEQ_WMAX));
127 }
128 
ccid2_hc_tx_rto_expire(unsigned long data)129 static void ccid2_hc_tx_rto_expire(unsigned long data)
130 {
131 	struct sock *sk = (struct sock *)data;
132 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
133 	const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
134 
135 	bh_lock_sock(sk);
136 	if (sock_owned_by_user(sk)) {
137 		sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5);
138 		goto out;
139 	}
140 
141 	ccid2_pr_debug("RTO_EXPIRE\n");
142 
143 	/* back-off timer */
144 	hc->tx_rto <<= 1;
145 	if (hc->tx_rto > DCCP_RTO_MAX)
146 		hc->tx_rto = DCCP_RTO_MAX;
147 
148 	/* adjust pipe, cwnd etc */
149 	hc->tx_ssthresh = hc->tx_cwnd / 2;
150 	if (hc->tx_ssthresh < 2)
151 		hc->tx_ssthresh = 2;
152 	hc->tx_cwnd	= 1;
153 	hc->tx_pipe	= 0;
154 
155 	/* clear state about stuff we sent */
156 	hc->tx_seqt = hc->tx_seqh;
157 	hc->tx_packets_acked = 0;
158 
159 	/* clear ack ratio state. */
160 	hc->tx_rpseq    = 0;
161 	hc->tx_rpdupack = -1;
162 	ccid2_change_l_ack_ratio(sk, 1);
163 
164 	/* if we were blocked before, we may now send cwnd=1 packet */
165 	if (sender_was_blocked)
166 		tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
167 	/* restart backed-off timer */
168 	sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
169 out:
170 	bh_unlock_sock(sk);
171 	sock_put(sk);
172 }
173 
174 /*
175  *	Congestion window validation (RFC 2861).
176  */
177 static bool ccid2_do_cwv = true;
178 module_param(ccid2_do_cwv, bool, 0644);
179 MODULE_PARM_DESC(ccid2_do_cwv, "Perform RFC2861 Congestion Window Validation");
180 
181 /**
182  * ccid2_update_used_window  -  Track how much of cwnd is actually used
183  * This is done in addition to CWV. The sender needs to have an idea of how many
184  * packets may be in flight, to set the local Sequence Window value accordingly
185  * (RFC 4340, 7.5.2). The CWV mechanism is exploited to keep track of the
186  * maximum-used window. We use an EWMA low-pass filter to filter out noise.
187  */
ccid2_update_used_window(struct ccid2_hc_tx_sock * hc,u32 new_wnd)188 static void ccid2_update_used_window(struct ccid2_hc_tx_sock *hc, u32 new_wnd)
189 {
190 	hc->tx_expected_wnd = (3 * hc->tx_expected_wnd + new_wnd) / 4;
191 }
192 
193 /* This borrows the code of tcp_cwnd_application_limited() */
ccid2_cwnd_application_limited(struct sock * sk,const u32 now)194 static void ccid2_cwnd_application_limited(struct sock *sk, const u32 now)
195 {
196 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
197 	/* don't reduce cwnd below the initial window (IW) */
198 	u32 init_win = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache),
199 	    win_used = max(hc->tx_cwnd_used, init_win);
200 
201 	if (win_used < hc->tx_cwnd) {
202 		hc->tx_ssthresh = max(hc->tx_ssthresh,
203 				     (hc->tx_cwnd >> 1) + (hc->tx_cwnd >> 2));
204 		hc->tx_cwnd = (hc->tx_cwnd + win_used) >> 1;
205 	}
206 	hc->tx_cwnd_used  = 0;
207 	hc->tx_cwnd_stamp = now;
208 
209 	ccid2_check_l_ack_ratio(sk);
210 }
211 
212 /* This borrows the code of tcp_cwnd_restart() */
ccid2_cwnd_restart(struct sock * sk,const u32 now)213 static void ccid2_cwnd_restart(struct sock *sk, const u32 now)
214 {
215 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
216 	u32 cwnd = hc->tx_cwnd, restart_cwnd,
217 	    iwnd = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache);
218 
219 	hc->tx_ssthresh = max(hc->tx_ssthresh, (cwnd >> 1) + (cwnd >> 2));
220 
221 	/* don't reduce cwnd below the initial window (IW) */
222 	restart_cwnd = min(cwnd, iwnd);
223 	cwnd >>= (now - hc->tx_lsndtime) / hc->tx_rto;
224 	hc->tx_cwnd = max(cwnd, restart_cwnd);
225 
226 	hc->tx_cwnd_stamp = now;
227 	hc->tx_cwnd_used  = 0;
228 
229 	ccid2_check_l_ack_ratio(sk);
230 }
231 
ccid2_hc_tx_packet_sent(struct sock * sk,unsigned int len)232 static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
233 {
234 	struct dccp_sock *dp = dccp_sk(sk);
235 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
236 	const u32 now = ccid2_time_stamp;
237 	struct ccid2_seq *next;
238 
239 	/* slow-start after idle periods (RFC 2581, RFC 2861) */
240 	if (ccid2_do_cwv && !hc->tx_pipe &&
241 	    (s32)(now - hc->tx_lsndtime) >= hc->tx_rto)
242 		ccid2_cwnd_restart(sk, now);
243 
244 	hc->tx_lsndtime = now;
245 	hc->tx_pipe    += 1;
246 
247 	/* see whether cwnd was fully used (RFC 2861), update expected window */
248 	if (ccid2_cwnd_network_limited(hc)) {
249 		ccid2_update_used_window(hc, hc->tx_cwnd);
250 		hc->tx_cwnd_used  = 0;
251 		hc->tx_cwnd_stamp = now;
252 	} else {
253 		if (hc->tx_pipe > hc->tx_cwnd_used)
254 			hc->tx_cwnd_used = hc->tx_pipe;
255 
256 		ccid2_update_used_window(hc, hc->tx_cwnd_used);
257 
258 		if (ccid2_do_cwv && (s32)(now - hc->tx_cwnd_stamp) >= hc->tx_rto)
259 			ccid2_cwnd_application_limited(sk, now);
260 	}
261 
262 	hc->tx_seqh->ccid2s_seq   = dp->dccps_gss;
263 	hc->tx_seqh->ccid2s_acked = 0;
264 	hc->tx_seqh->ccid2s_sent  = now;
265 
266 	next = hc->tx_seqh->ccid2s_next;
267 	/* check if we need to alloc more space */
268 	if (next == hc->tx_seqt) {
269 		if (ccid2_hc_tx_alloc_seq(hc)) {
270 			DCCP_CRIT("packet history - out of memory!");
271 			/* FIXME: find a more graceful way to bail out */
272 			return;
273 		}
274 		next = hc->tx_seqh->ccid2s_next;
275 		BUG_ON(next == hc->tx_seqt);
276 	}
277 	hc->tx_seqh = next;
278 
279 	ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe);
280 
281 	/*
282 	 * FIXME: The code below is broken and the variables have been removed
283 	 * from the socket struct. The `ackloss' variable was always set to 0,
284 	 * and with arsent there are several problems:
285 	 *  (i) it doesn't just count the number of Acks, but all sent packets;
286 	 *  (ii) it is expressed in # of packets, not # of windows, so the
287 	 *  comparison below uses the wrong formula: Appendix A of RFC 4341
288 	 *  comes up with the number K = cwnd / (R^2 - R) of consecutive windows
289 	 *  of data with no lost or marked Ack packets. If arsent were the # of
290 	 *  consecutive Acks received without loss, then Ack Ratio needs to be
291 	 *  decreased by 1 when
292 	 *	      arsent >=  K * cwnd / R  =  cwnd^2 / (R^3 - R^2)
293 	 *  where cwnd / R is the number of Acks received per window of data
294 	 *  (cf. RFC 4341, App. A). The problems are that
295 	 *  - arsent counts other packets as well;
296 	 *  - the comparison uses a formula different from RFC 4341;
297 	 *  - computing a cubic/quadratic equation each time is too complicated.
298 	 *  Hence a different algorithm is needed.
299 	 */
300 #if 0
301 	/* Ack Ratio.  Need to maintain a concept of how many windows we sent */
302 	hc->tx_arsent++;
303 	/* We had an ack loss in this window... */
304 	if (hc->tx_ackloss) {
305 		if (hc->tx_arsent >= hc->tx_cwnd) {
306 			hc->tx_arsent  = 0;
307 			hc->tx_ackloss = 0;
308 		}
309 	} else {
310 		/* No acks lost up to now... */
311 		/* decrease ack ratio if enough packets were sent */
312 		if (dp->dccps_l_ack_ratio > 1) {
313 			/* XXX don't calculate denominator each time */
314 			int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
315 				    dp->dccps_l_ack_ratio;
316 
317 			denom = hc->tx_cwnd * hc->tx_cwnd / denom;
318 
319 			if (hc->tx_arsent >= denom) {
320 				ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
321 				hc->tx_arsent = 0;
322 			}
323 		} else {
324 			/* we can't increase ack ratio further [1] */
325 			hc->tx_arsent = 0; /* or maybe set it to cwnd*/
326 		}
327 	}
328 #endif
329 
330 	sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
331 
332 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
333 	do {
334 		struct ccid2_seq *seqp = hc->tx_seqt;
335 
336 		while (seqp != hc->tx_seqh) {
337 			ccid2_pr_debug("out seq=%llu acked=%d time=%u\n",
338 				       (unsigned long long)seqp->ccid2s_seq,
339 				       seqp->ccid2s_acked, seqp->ccid2s_sent);
340 			seqp = seqp->ccid2s_next;
341 		}
342 	} while (0);
343 	ccid2_pr_debug("=========\n");
344 #endif
345 }
346 
347 /**
348  * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
349  * This code is almost identical with TCP's tcp_rtt_estimator(), since
350  * - it has a higher sampling frequency (recommended by RFC 1323),
351  * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
352  * - it is simple (cf. more complex proposals such as Eifel timer or research
353  *   which suggests that the gain should be set according to window size),
354  * - in tests it was found to work well with CCID2 [gerrit].
355  */
ccid2_rtt_estimator(struct sock * sk,const long mrtt)356 static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
357 {
358 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
359 	long m = mrtt ? : 1;
360 
361 	if (hc->tx_srtt == 0) {
362 		/* First measurement m */
363 		hc->tx_srtt = m << 3;
364 		hc->tx_mdev = m << 1;
365 
366 		hc->tx_mdev_max = max(hc->tx_mdev, tcp_rto_min(sk));
367 		hc->tx_rttvar   = hc->tx_mdev_max;
368 
369 		hc->tx_rtt_seq  = dccp_sk(sk)->dccps_gss;
370 	} else {
371 		/* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
372 		m -= (hc->tx_srtt >> 3);
373 		hc->tx_srtt += m;
374 
375 		/* Similarly, update scaled mdev with regard to |m| */
376 		if (m < 0) {
377 			m = -m;
378 			m -= (hc->tx_mdev >> 2);
379 			/*
380 			 * This neutralises RTO increase when RTT < SRTT - mdev
381 			 * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
382 			 * in Linux TCP", USENIX 2002, pp. 49-62).
383 			 */
384 			if (m > 0)
385 				m >>= 3;
386 		} else {
387 			m -= (hc->tx_mdev >> 2);
388 		}
389 		hc->tx_mdev += m;
390 
391 		if (hc->tx_mdev > hc->tx_mdev_max) {
392 			hc->tx_mdev_max = hc->tx_mdev;
393 			if (hc->tx_mdev_max > hc->tx_rttvar)
394 				hc->tx_rttvar = hc->tx_mdev_max;
395 		}
396 
397 		/*
398 		 * Decay RTTVAR at most once per flight, exploiting that
399 		 *  1) pipe <= cwnd <= Sequence_Window = W  (RFC 4340, 7.5.2)
400 		 *  2) AWL = GSS-W+1 <= GAR <= GSS          (RFC 4340, 7.5.1)
401 		 * GAR is a useful bound for FlightSize = pipe.
402 		 * AWL is probably too low here, as it over-estimates pipe.
403 		 */
404 		if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) {
405 			if (hc->tx_mdev_max < hc->tx_rttvar)
406 				hc->tx_rttvar -= (hc->tx_rttvar -
407 						  hc->tx_mdev_max) >> 2;
408 			hc->tx_rtt_seq  = dccp_sk(sk)->dccps_gss;
409 			hc->tx_mdev_max = tcp_rto_min(sk);
410 		}
411 	}
412 
413 	/*
414 	 * Set RTO from SRTT and RTTVAR
415 	 * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms.
416 	 * This agrees with RFC 4341, 5:
417 	 *	"Because DCCP does not retransmit data, DCCP does not require
418 	 *	 TCP's recommended minimum timeout of one second".
419 	 */
420 	hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar;
421 
422 	if (hc->tx_rto > DCCP_RTO_MAX)
423 		hc->tx_rto = DCCP_RTO_MAX;
424 }
425 
ccid2_new_ack(struct sock * sk,struct ccid2_seq * seqp,unsigned int * maxincr)426 static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
427 			  unsigned int *maxincr)
428 {
429 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
430 	struct dccp_sock *dp = dccp_sk(sk);
431 	int r_seq_used = hc->tx_cwnd / dp->dccps_l_ack_ratio;
432 
433 	if (hc->tx_cwnd < dp->dccps_l_seq_win &&
434 	    r_seq_used < dp->dccps_r_seq_win) {
435 		if (hc->tx_cwnd < hc->tx_ssthresh) {
436 			if (*maxincr > 0 && ++hc->tx_packets_acked >= 2) {
437 				hc->tx_cwnd += 1;
438 				*maxincr    -= 1;
439 				hc->tx_packets_acked = 0;
440 			}
441 		} else if (++hc->tx_packets_acked >= hc->tx_cwnd) {
442 			hc->tx_cwnd += 1;
443 			hc->tx_packets_acked = 0;
444 		}
445 	}
446 
447 	/*
448 	 * Adjust the local sequence window and the ack ratio to allow about
449 	 * 5 times the number of packets in the network (RFC 4340 7.5.2)
450 	 */
451 	if (r_seq_used * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_r_seq_win)
452 		ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio * 2);
453 	else if (r_seq_used * CCID2_WIN_CHANGE_FACTOR < dp->dccps_r_seq_win/2)
454 		ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio / 2 ? : 1U);
455 
456 	if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_l_seq_win)
457 		ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win * 2);
458 	else if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR < dp->dccps_l_seq_win/2)
459 		ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win / 2);
460 
461 	/*
462 	 * FIXME: RTT is sampled several times per acknowledgment (for each
463 	 * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
464 	 * This causes the RTT to be over-estimated, since the older entries
465 	 * in the Ack Vector have earlier sending times.
466 	 * The cleanest solution is to not use the ccid2s_sent field at all
467 	 * and instead use DCCP timestamps: requires changes in other places.
468 	 */
469 	ccid2_rtt_estimator(sk, ccid2_time_stamp - seqp->ccid2s_sent);
470 }
471 
ccid2_congestion_event(struct sock * sk,struct ccid2_seq * seqp)472 static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
473 {
474 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
475 
476 	if ((s32)(seqp->ccid2s_sent - hc->tx_last_cong) < 0) {
477 		ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
478 		return;
479 	}
480 
481 	hc->tx_last_cong = ccid2_time_stamp;
482 
483 	hc->tx_cwnd      = hc->tx_cwnd / 2 ? : 1U;
484 	hc->tx_ssthresh  = max(hc->tx_cwnd, 2U);
485 
486 	ccid2_check_l_ack_ratio(sk);
487 }
488 
ccid2_hc_tx_parse_options(struct sock * sk,u8 packet_type,u8 option,u8 * optval,u8 optlen)489 static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
490 				     u8 option, u8 *optval, u8 optlen)
491 {
492 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
493 
494 	switch (option) {
495 	case DCCPO_ACK_VECTOR_0:
496 	case DCCPO_ACK_VECTOR_1:
497 		return dccp_ackvec_parsed_add(&hc->tx_av_chunks, optval, optlen,
498 					      option - DCCPO_ACK_VECTOR_0);
499 	}
500 	return 0;
501 }
502 
ccid2_hc_tx_packet_recv(struct sock * sk,struct sk_buff * skb)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 *hc = ccid2_hc_tx_sk(sk);
507 	const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
508 	struct dccp_ackvec_parsed *avp;
509 	u64 ackno, seqno;
510 	struct ccid2_seq *seqp;
511 	int done = 0;
512 	unsigned int maxincr = 0;
513 
514 	/* check reverse path congestion */
515 	seqno = DCCP_SKB_CB(skb)->dccpd_seq;
516 
517 	/* XXX this whole "algorithm" is broken.  Need to fix it to keep track
518 	 * of the seqnos of the dupacks so that rpseq and rpdupack are correct
519 	 * -sorbo.
520 	 */
521 	/* need to bootstrap */
522 	if (hc->tx_rpdupack == -1) {
523 		hc->tx_rpdupack = 0;
524 		hc->tx_rpseq    = seqno;
525 	} else {
526 		/* check if packet is consecutive */
527 		if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1)
528 			hc->tx_rpseq = seqno;
529 		/* it's a later packet */
530 		else if (after48(seqno, hc->tx_rpseq)) {
531 			hc->tx_rpdupack++;
532 
533 			/* check if we got enough dupacks */
534 			if (hc->tx_rpdupack >= NUMDUPACK) {
535 				hc->tx_rpdupack = -1; /* XXX lame */
536 				hc->tx_rpseq    = 0;
537 #ifdef __CCID2_COPES_GRACEFULLY_WITH_ACK_CONGESTION_CONTROL__
538 				/*
539 				 * FIXME: Ack Congestion Control is broken; in
540 				 * the current state instabilities occurred with
541 				 * Ack Ratios greater than 1; causing hang-ups
542 				 * and long RTO timeouts. This needs to be fixed
543 				 * before opening up dynamic changes. -- gerrit
544 				 */
545 				ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
546 #endif
547 			}
548 		}
549 	}
550 
551 	/* check forward path congestion */
552 	if (dccp_packet_without_ack(skb))
553 		return;
554 
555 	/* still didn't send out new data packets */
556 	if (hc->tx_seqh == hc->tx_seqt)
557 		goto done;
558 
559 	ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
560 	if (after48(ackno, hc->tx_high_ack))
561 		hc->tx_high_ack = ackno;
562 
563 	seqp = hc->tx_seqt;
564 	while (before48(seqp->ccid2s_seq, ackno)) {
565 		seqp = seqp->ccid2s_next;
566 		if (seqp == hc->tx_seqh) {
567 			seqp = hc->tx_seqh->ccid2s_prev;
568 			break;
569 		}
570 	}
571 
572 	/*
573 	 * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
574 	 * packets per acknowledgement. Rounding up avoids that cwnd is not
575 	 * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
576 	 */
577 	if (hc->tx_cwnd < hc->tx_ssthresh)
578 		maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
579 
580 	/* go through all ack vectors */
581 	list_for_each_entry(avp, &hc->tx_av_chunks, node) {
582 		/* go through this ack vector */
583 		for (; avp->len--; avp->vec++) {
584 			u64 ackno_end_rl = SUB48(ackno,
585 						 dccp_ackvec_runlen(avp->vec));
586 
587 			ccid2_pr_debug("ackvec %llu |%u,%u|\n",
588 				       (unsigned long long)ackno,
589 				       dccp_ackvec_state(avp->vec) >> 6,
590 				       dccp_ackvec_runlen(avp->vec));
591 			/* if the seqno we are analyzing is larger than the
592 			 * current ackno, then move towards the tail of our
593 			 * seqnos.
594 			 */
595 			while (after48(seqp->ccid2s_seq, ackno)) {
596 				if (seqp == hc->tx_seqt) {
597 					done = 1;
598 					break;
599 				}
600 				seqp = seqp->ccid2s_prev;
601 			}
602 			if (done)
603 				break;
604 
605 			/* check all seqnos in the range of the vector
606 			 * run length
607 			 */
608 			while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
609 				const u8 state = dccp_ackvec_state(avp->vec);
610 
611 				/* new packet received or marked */
612 				if (state != DCCPAV_NOT_RECEIVED &&
613 				    !seqp->ccid2s_acked) {
614 					if (state == DCCPAV_ECN_MARKED)
615 						ccid2_congestion_event(sk,
616 								       seqp);
617 					else
618 						ccid2_new_ack(sk, seqp,
619 							      &maxincr);
620 
621 					seqp->ccid2s_acked = 1;
622 					ccid2_pr_debug("Got ack for %llu\n",
623 						       (unsigned long long)seqp->ccid2s_seq);
624 					hc->tx_pipe--;
625 				}
626 				if (seqp == hc->tx_seqt) {
627 					done = 1;
628 					break;
629 				}
630 				seqp = seqp->ccid2s_prev;
631 			}
632 			if (done)
633 				break;
634 
635 			ackno = SUB48(ackno_end_rl, 1);
636 		}
637 		if (done)
638 			break;
639 	}
640 
641 	/* The state about what is acked should be correct now
642 	 * Check for NUMDUPACK
643 	 */
644 	seqp = hc->tx_seqt;
645 	while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) {
646 		seqp = seqp->ccid2s_next;
647 		if (seqp == hc->tx_seqh) {
648 			seqp = hc->tx_seqh->ccid2s_prev;
649 			break;
650 		}
651 	}
652 	done = 0;
653 	while (1) {
654 		if (seqp->ccid2s_acked) {
655 			done++;
656 			if (done == NUMDUPACK)
657 				break;
658 		}
659 		if (seqp == hc->tx_seqt)
660 			break;
661 		seqp = seqp->ccid2s_prev;
662 	}
663 
664 	/* If there are at least 3 acknowledgements, anything unacknowledged
665 	 * below the last sequence number is considered lost
666 	 */
667 	if (done == NUMDUPACK) {
668 		struct ccid2_seq *last_acked = seqp;
669 
670 		/* check for lost packets */
671 		while (1) {
672 			if (!seqp->ccid2s_acked) {
673 				ccid2_pr_debug("Packet lost: %llu\n",
674 					       (unsigned long long)seqp->ccid2s_seq);
675 				/* XXX need to traverse from tail -> head in
676 				 * order to detect multiple congestion events in
677 				 * one ack vector.
678 				 */
679 				ccid2_congestion_event(sk, seqp);
680 				hc->tx_pipe--;
681 			}
682 			if (seqp == hc->tx_seqt)
683 				break;
684 			seqp = seqp->ccid2s_prev;
685 		}
686 
687 		hc->tx_seqt = last_acked;
688 	}
689 
690 	/* trim acked packets in tail */
691 	while (hc->tx_seqt != hc->tx_seqh) {
692 		if (!hc->tx_seqt->ccid2s_acked)
693 			break;
694 
695 		hc->tx_seqt = hc->tx_seqt->ccid2s_next;
696 	}
697 
698 	/* restart RTO timer if not all outstanding data has been acked */
699 	if (hc->tx_pipe == 0)
700 		sk_stop_timer(sk, &hc->tx_rtotimer);
701 	else
702 		sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
703 done:
704 	/* check if incoming Acks allow pending packets to be sent */
705 	if (sender_was_blocked && !ccid2_cwnd_network_limited(hc))
706 		tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
707 	dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
708 }
709 
ccid2_hc_tx_init(struct ccid * ccid,struct sock * sk)710 static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
711 {
712 	struct ccid2_hc_tx_sock *hc = ccid_priv(ccid);
713 	struct dccp_sock *dp = dccp_sk(sk);
714 	u32 max_ratio;
715 
716 	/* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
717 	hc->tx_ssthresh = ~0U;
718 
719 	/* Use larger initial windows (RFC 4341, section 5). */
720 	hc->tx_cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache);
721 	hc->tx_expected_wnd = hc->tx_cwnd;
722 
723 	/* Make sure that Ack Ratio is enabled and within bounds. */
724 	max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2);
725 	if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
726 		dp->dccps_l_ack_ratio = max_ratio;
727 
728 	/* XXX init ~ to window size... */
729 	if (ccid2_hc_tx_alloc_seq(hc))
730 		return -ENOMEM;
731 
732 	hc->tx_rto	 = DCCP_TIMEOUT_INIT;
733 	hc->tx_rpdupack  = -1;
734 	hc->tx_last_cong = hc->tx_lsndtime = hc->tx_cwnd_stamp = ccid2_time_stamp;
735 	hc->tx_cwnd_used = 0;
736 	setup_timer(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire,
737 			(unsigned long)sk);
738 	INIT_LIST_HEAD(&hc->tx_av_chunks);
739 	return 0;
740 }
741 
ccid2_hc_tx_exit(struct sock * sk)742 static void ccid2_hc_tx_exit(struct sock *sk)
743 {
744 	struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
745 	int i;
746 
747 	sk_stop_timer(sk, &hc->tx_rtotimer);
748 
749 	for (i = 0; i < hc->tx_seqbufc; i++)
750 		kfree(hc->tx_seqbuf[i]);
751 	hc->tx_seqbufc = 0;
752 }
753 
ccid2_hc_rx_packet_recv(struct sock * sk,struct sk_buff * skb)754 static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
755 {
756 	struct ccid2_hc_rx_sock *hc = ccid2_hc_rx_sk(sk);
757 
758 	if (!dccp_data_packet(skb))
759 		return;
760 
761 	if (++hc->rx_num_data_pkts >= dccp_sk(sk)->dccps_r_ack_ratio) {
762 		dccp_send_ack(sk);
763 		hc->rx_num_data_pkts = 0;
764 	}
765 }
766 
767 struct ccid_operations ccid2_ops = {
768 	.ccid_id		  = DCCPC_CCID2,
769 	.ccid_name		  = "TCP-like",
770 	.ccid_hc_tx_obj_size	  = sizeof(struct ccid2_hc_tx_sock),
771 	.ccid_hc_tx_init	  = ccid2_hc_tx_init,
772 	.ccid_hc_tx_exit	  = ccid2_hc_tx_exit,
773 	.ccid_hc_tx_send_packet	  = ccid2_hc_tx_send_packet,
774 	.ccid_hc_tx_packet_sent	  = ccid2_hc_tx_packet_sent,
775 	.ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
776 	.ccid_hc_tx_packet_recv	  = ccid2_hc_tx_packet_recv,
777 	.ccid_hc_rx_obj_size	  = sizeof(struct ccid2_hc_rx_sock),
778 	.ccid_hc_rx_packet_recv	  = ccid2_hc_rx_packet_recv,
779 };
780 
781 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
782 module_param(ccid2_debug, bool, 0644);
783 MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages");
784 #endif
785