Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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
32static int 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
38static 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
79static 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
86static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
87{
88 struct dccp_sock *dp = dccp_sk(sk);
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 if (val > DCCPF_ACK_RATIO_MAX)
102 val = DCCPF_ACK_RATIO_MAX;
103
104 if (val == dp->dccps_l_ack_ratio)
105 return;
106
107 ccid2_pr_debug("changing local ack ratio to %u\n", val);
108 dp->dccps_l_ack_ratio = val;
109}
110
111static void ccid2_hc_tx_rto_expire(unsigned long data)
112{
113 struct sock *sk = (struct sock *)data;
114 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
115 const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
116
117 bh_lock_sock(sk);
118 if (sock_owned_by_user(sk)) {
119 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5);
120 goto out;
121 }
122
123 ccid2_pr_debug("RTO_EXPIRE\n");
124
125 /* back-off timer */
126 hc->tx_rto <<= 1;
127 if (hc->tx_rto > DCCP_RTO_MAX)
128 hc->tx_rto = DCCP_RTO_MAX;
129
130 /* adjust pipe, cwnd etc */
131 hc->tx_ssthresh = hc->tx_cwnd / 2;
132 if (hc->tx_ssthresh < 2)
133 hc->tx_ssthresh = 2;
134 hc->tx_cwnd = 1;
135 hc->tx_pipe = 0;
136
137 /* clear state about stuff we sent */
138 hc->tx_seqt = hc->tx_seqh;
139 hc->tx_packets_acked = 0;
140
141 /* clear ack ratio state. */
142 hc->tx_rpseq = 0;
143 hc->tx_rpdupack = -1;
144 ccid2_change_l_ack_ratio(sk, 1);
145
146 /* if we were blocked before, we may now send cwnd=1 packet */
147 if (sender_was_blocked)
148 tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
149 /* restart backed-off timer */
150 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
151out:
152 bh_unlock_sock(sk);
153 sock_put(sk);
154}
155
156static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
157{
158 struct dccp_sock *dp = dccp_sk(sk);
159 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
160 struct ccid2_seq *next;
161
162 hc->tx_pipe++;
163
164 hc->tx_seqh->ccid2s_seq = dp->dccps_gss;
165 hc->tx_seqh->ccid2s_acked = 0;
166 hc->tx_seqh->ccid2s_sent = ccid2_time_stamp;
167
168 next = hc->tx_seqh->ccid2s_next;
169 /* check if we need to alloc more space */
170 if (next == hc->tx_seqt) {
171 if (ccid2_hc_tx_alloc_seq(hc)) {
172 DCCP_CRIT("packet history - out of memory!");
173 /* FIXME: find a more graceful way to bail out */
174 return;
175 }
176 next = hc->tx_seqh->ccid2s_next;
177 BUG_ON(next == hc->tx_seqt);
178 }
179 hc->tx_seqh = next;
180
181 ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe);
182
183 /*
184 * FIXME: The code below is broken and the variables have been removed
185 * from the socket struct. The `ackloss' variable was always set to 0,
186 * and with arsent there are several problems:
187 * (i) it doesn't just count the number of Acks, but all sent packets;
188 * (ii) it is expressed in # of packets, not # of windows, so the
189 * comparison below uses the wrong formula: Appendix A of RFC 4341
190 * comes up with the number K = cwnd / (R^2 - R) of consecutive windows
191 * of data with no lost or marked Ack packets. If arsent were the # of
192 * consecutive Acks received without loss, then Ack Ratio needs to be
193 * decreased by 1 when
194 * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2)
195 * where cwnd / R is the number of Acks received per window of data
196 * (cf. RFC 4341, App. A). The problems are that
197 * - arsent counts other packets as well;
198 * - the comparison uses a formula different from RFC 4341;
199 * - computing a cubic/quadratic equation each time is too complicated.
200 * Hence a different algorithm is needed.
201 */
202#if 0
203 /* Ack Ratio. Need to maintain a concept of how many windows we sent */
204 hc->tx_arsent++;
205 /* We had an ack loss in this window... */
206 if (hc->tx_ackloss) {
207 if (hc->tx_arsent >= hc->tx_cwnd) {
208 hc->tx_arsent = 0;
209 hc->tx_ackloss = 0;
210 }
211 } else {
212 /* No acks lost up to now... */
213 /* decrease ack ratio if enough packets were sent */
214 if (dp->dccps_l_ack_ratio > 1) {
215 /* XXX don't calculate denominator each time */
216 int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
217 dp->dccps_l_ack_ratio;
218
219 denom = hc->tx_cwnd * hc->tx_cwnd / denom;
220
221 if (hc->tx_arsent >= denom) {
222 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
223 hc->tx_arsent = 0;
224 }
225 } else {
226 /* we can't increase ack ratio further [1] */
227 hc->tx_arsent = 0; /* or maybe set it to cwnd*/
228 }
229 }
230#endif
231
232 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
233
234#ifdef CONFIG_IP_DCCP_CCID2_DEBUG
235 do {
236 struct ccid2_seq *seqp = hc->tx_seqt;
237
238 while (seqp != hc->tx_seqh) {
239 ccid2_pr_debug("out seq=%llu acked=%d time=%u\n",
240 (unsigned long long)seqp->ccid2s_seq,
241 seqp->ccid2s_acked, seqp->ccid2s_sent);
242 seqp = seqp->ccid2s_next;
243 }
244 } while (0);
245 ccid2_pr_debug("=========\n");
246#endif
247}
248
249/**
250 * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
251 * This code is almost identical with TCP's tcp_rtt_estimator(), since
252 * - it has a higher sampling frequency (recommended by RFC 1323),
253 * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
254 * - it is simple (cf. more complex proposals such as Eifel timer or research
255 * which suggests that the gain should be set according to window size),
256 * - in tests it was found to work well with CCID2 [gerrit].
257 */
258static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
259{
260 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
261 long m = mrtt ? : 1;
262
263 if (hc->tx_srtt == 0) {
264 /* First measurement m */
265 hc->tx_srtt = m << 3;
266 hc->tx_mdev = m << 1;
267
268 hc->tx_mdev_max = max(hc->tx_mdev, tcp_rto_min(sk));
269 hc->tx_rttvar = hc->tx_mdev_max;
270
271 hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss;
272 } else {
273 /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
274 m -= (hc->tx_srtt >> 3);
275 hc->tx_srtt += m;
276
277 /* Similarly, update scaled mdev with regard to |m| */
278 if (m < 0) {
279 m = -m;
280 m -= (hc->tx_mdev >> 2);
281 /*
282 * This neutralises RTO increase when RTT < SRTT - mdev
283 * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
284 * in Linux TCP", USENIX 2002, pp. 49-62).
285 */
286 if (m > 0)
287 m >>= 3;
288 } else {
289 m -= (hc->tx_mdev >> 2);
290 }
291 hc->tx_mdev += m;
292
293 if (hc->tx_mdev > hc->tx_mdev_max) {
294 hc->tx_mdev_max = hc->tx_mdev;
295 if (hc->tx_mdev_max > hc->tx_rttvar)
296 hc->tx_rttvar = hc->tx_mdev_max;
297 }
298
299 /*
300 * Decay RTTVAR at most once per flight, exploiting that
301 * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2)
302 * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1)
303 * GAR is a useful bound for FlightSize = pipe.
304 * AWL is probably too low here, as it over-estimates pipe.
305 */
306 if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) {
307 if (hc->tx_mdev_max < hc->tx_rttvar)
308 hc->tx_rttvar -= (hc->tx_rttvar -
309 hc->tx_mdev_max) >> 2;
310 hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss;
311 hc->tx_mdev_max = tcp_rto_min(sk);
312 }
313 }
314
315 /*
316 * Set RTO from SRTT and RTTVAR
317 * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms.
318 * This agrees with RFC 4341, 5:
319 * "Because DCCP does not retransmit data, DCCP does not require
320 * TCP's recommended minimum timeout of one second".
321 */
322 hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar;
323
324 if (hc->tx_rto > DCCP_RTO_MAX)
325 hc->tx_rto = DCCP_RTO_MAX;
326}
327
328static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
329 unsigned int *maxincr)
330{
331 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
332
333 if (hc->tx_cwnd < hc->tx_ssthresh) {
334 if (*maxincr > 0 && ++hc->tx_packets_acked == 2) {
335 hc->tx_cwnd += 1;
336 *maxincr -= 1;
337 hc->tx_packets_acked = 0;
338 }
339 } else if (++hc->tx_packets_acked >= hc->tx_cwnd) {
340 hc->tx_cwnd += 1;
341 hc->tx_packets_acked = 0;
342 }
343 /*
344 * FIXME: RTT is sampled several times per acknowledgment (for each
345 * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
346 * This causes the RTT to be over-estimated, since the older entries
347 * in the Ack Vector have earlier sending times.
348 * The cleanest solution is to not use the ccid2s_sent field at all
349 * and instead use DCCP timestamps: requires changes in other places.
350 */
351 ccid2_rtt_estimator(sk, ccid2_time_stamp - seqp->ccid2s_sent);
352}
353
354static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
355{
356 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
357
358 if ((s32)(seqp->ccid2s_sent - hc->tx_last_cong) < 0) {
359 ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
360 return;
361 }
362
363 hc->tx_last_cong = ccid2_time_stamp;
364
365 hc->tx_cwnd = hc->tx_cwnd / 2 ? : 1U;
366 hc->tx_ssthresh = max(hc->tx_cwnd, 2U);
367
368 /* Avoid spurious timeouts resulting from Ack Ratio > cwnd */
369 if (dccp_sk(sk)->dccps_l_ack_ratio > hc->tx_cwnd)
370 ccid2_change_l_ack_ratio(sk, hc->tx_cwnd);
371}
372
373static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
374 u8 option, u8 *optval, u8 optlen)
375{
376 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
377
378 switch (option) {
379 case DCCPO_ACK_VECTOR_0:
380 case DCCPO_ACK_VECTOR_1:
381 return dccp_ackvec_parsed_add(&hc->tx_av_chunks, optval, optlen,
382 option - DCCPO_ACK_VECTOR_0);
383 }
384 return 0;
385}
386
387static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
388{
389 struct dccp_sock *dp = dccp_sk(sk);
390 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
391 const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
392 struct dccp_ackvec_parsed *avp;
393 u64 ackno, seqno;
394 struct ccid2_seq *seqp;
395 int done = 0;
396 unsigned int maxincr = 0;
397
398 /* check reverse path congestion */
399 seqno = DCCP_SKB_CB(skb)->dccpd_seq;
400
401 /* XXX this whole "algorithm" is broken. Need to fix it to keep track
402 * of the seqnos of the dupacks so that rpseq and rpdupack are correct
403 * -sorbo.
404 */
405 /* need to bootstrap */
406 if (hc->tx_rpdupack == -1) {
407 hc->tx_rpdupack = 0;
408 hc->tx_rpseq = seqno;
409 } else {
410 /* check if packet is consecutive */
411 if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1)
412 hc->tx_rpseq = seqno;
413 /* it's a later packet */
414 else if (after48(seqno, hc->tx_rpseq)) {
415 hc->tx_rpdupack++;
416
417 /* check if we got enough dupacks */
418 if (hc->tx_rpdupack >= NUMDUPACK) {
419 hc->tx_rpdupack = -1; /* XXX lame */
420 hc->tx_rpseq = 0;
421
422 ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
423 }
424 }
425 }
426
427 /* check forward path congestion */
428 if (dccp_packet_without_ack(skb))
429 return;
430
431 /* still didn't send out new data packets */
432 if (hc->tx_seqh == hc->tx_seqt)
433 goto done;
434
435 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
436 if (after48(ackno, hc->tx_high_ack))
437 hc->tx_high_ack = ackno;
438
439 seqp = hc->tx_seqt;
440 while (before48(seqp->ccid2s_seq, ackno)) {
441 seqp = seqp->ccid2s_next;
442 if (seqp == hc->tx_seqh) {
443 seqp = hc->tx_seqh->ccid2s_prev;
444 break;
445 }
446 }
447
448 /*
449 * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
450 * packets per acknowledgement. Rounding up avoids that cwnd is not
451 * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
452 */
453 if (hc->tx_cwnd < hc->tx_ssthresh)
454 maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
455
456 /* go through all ack vectors */
457 list_for_each_entry(avp, &hc->tx_av_chunks, node) {
458 /* go through this ack vector */
459 for (; avp->len--; avp->vec++) {
460 u64 ackno_end_rl = SUB48(ackno,
461 dccp_ackvec_runlen(avp->vec));
462
463 ccid2_pr_debug("ackvec %llu |%u,%u|\n",
464 (unsigned long long)ackno,
465 dccp_ackvec_state(avp->vec) >> 6,
466 dccp_ackvec_runlen(avp->vec));
467 /* if the seqno we are analyzing is larger than the
468 * current ackno, then move towards the tail of our
469 * seqnos.
470 */
471 while (after48(seqp->ccid2s_seq, ackno)) {
472 if (seqp == hc->tx_seqt) {
473 done = 1;
474 break;
475 }
476 seqp = seqp->ccid2s_prev;
477 }
478 if (done)
479 break;
480
481 /* check all seqnos in the range of the vector
482 * run length
483 */
484 while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
485 const u8 state = dccp_ackvec_state(avp->vec);
486
487 /* new packet received or marked */
488 if (state != DCCPAV_NOT_RECEIVED &&
489 !seqp->ccid2s_acked) {
490 if (state == DCCPAV_ECN_MARKED)
491 ccid2_congestion_event(sk,
492 seqp);
493 else
494 ccid2_new_ack(sk, seqp,
495 &maxincr);
496
497 seqp->ccid2s_acked = 1;
498 ccid2_pr_debug("Got ack for %llu\n",
499 (unsigned long long)seqp->ccid2s_seq);
500 hc->tx_pipe--;
501 }
502 if (seqp == hc->tx_seqt) {
503 done = 1;
504 break;
505 }
506 seqp = seqp->ccid2s_prev;
507 }
508 if (done)
509 break;
510
511 ackno = SUB48(ackno_end_rl, 1);
512 }
513 if (done)
514 break;
515 }
516
517 /* The state about what is acked should be correct now
518 * Check for NUMDUPACK
519 */
520 seqp = hc->tx_seqt;
521 while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) {
522 seqp = seqp->ccid2s_next;
523 if (seqp == hc->tx_seqh) {
524 seqp = hc->tx_seqh->ccid2s_prev;
525 break;
526 }
527 }
528 done = 0;
529 while (1) {
530 if (seqp->ccid2s_acked) {
531 done++;
532 if (done == NUMDUPACK)
533 break;
534 }
535 if (seqp == hc->tx_seqt)
536 break;
537 seqp = seqp->ccid2s_prev;
538 }
539
540 /* If there are at least 3 acknowledgements, anything unacknowledged
541 * below the last sequence number is considered lost
542 */
543 if (done == NUMDUPACK) {
544 struct ccid2_seq *last_acked = seqp;
545
546 /* check for lost packets */
547 while (1) {
548 if (!seqp->ccid2s_acked) {
549 ccid2_pr_debug("Packet lost: %llu\n",
550 (unsigned long long)seqp->ccid2s_seq);
551 /* XXX need to traverse from tail -> head in
552 * order to detect multiple congestion events in
553 * one ack vector.
554 */
555 ccid2_congestion_event(sk, seqp);
556 hc->tx_pipe--;
557 }
558 if (seqp == hc->tx_seqt)
559 break;
560 seqp = seqp->ccid2s_prev;
561 }
562
563 hc->tx_seqt = last_acked;
564 }
565
566 /* trim acked packets in tail */
567 while (hc->tx_seqt != hc->tx_seqh) {
568 if (!hc->tx_seqt->ccid2s_acked)
569 break;
570
571 hc->tx_seqt = hc->tx_seqt->ccid2s_next;
572 }
573
574 /* restart RTO timer if not all outstanding data has been acked */
575 if (hc->tx_pipe == 0)
576 sk_stop_timer(sk, &hc->tx_rtotimer);
577 else
578 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
579done:
580 /* check if incoming Acks allow pending packets to be sent */
581 if (sender_was_blocked && !ccid2_cwnd_network_limited(hc))
582 tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
583 dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
584}
585
586/*
587 * Convert RFC 3390 larger initial window into an equivalent number of packets.
588 * This is based on the numbers specified in RFC 5681, 3.1.
589 */
590static inline u32 rfc3390_bytes_to_packets(const u32 smss)
591{
592 return smss <= 1095 ? 4 : (smss > 2190 ? 2 : 3);
593}
594
595static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
596{
597 struct ccid2_hc_tx_sock *hc = ccid_priv(ccid);
598 struct dccp_sock *dp = dccp_sk(sk);
599 u32 max_ratio;
600
601 /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
602 hc->tx_ssthresh = ~0U;
603
604 /* Use larger initial windows (RFC 4341, section 5). */
605 hc->tx_cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache);
606
607 /* Make sure that Ack Ratio is enabled and within bounds. */
608 max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2);
609 if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
610 dp->dccps_l_ack_ratio = max_ratio;
611
612 /* XXX init ~ to window size... */
613 if (ccid2_hc_tx_alloc_seq(hc))
614 return -ENOMEM;
615
616 hc->tx_rto = DCCP_TIMEOUT_INIT;
617 hc->tx_rpdupack = -1;
618 hc->tx_last_cong = ccid2_time_stamp;
619 setup_timer(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire,
620 (unsigned long)sk);
621 INIT_LIST_HEAD(&hc->tx_av_chunks);
622 return 0;
623}
624
625static void ccid2_hc_tx_exit(struct sock *sk)
626{
627 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
628 int i;
629
630 sk_stop_timer(sk, &hc->tx_rtotimer);
631
632 for (i = 0; i < hc->tx_seqbufc; i++)
633 kfree(hc->tx_seqbuf[i]);
634 hc->tx_seqbufc = 0;
635}
636
637static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
638{
639 const struct dccp_sock *dp = dccp_sk(sk);
640 struct ccid2_hc_rx_sock *hc = ccid2_hc_rx_sk(sk);
641
642 switch (DCCP_SKB_CB(skb)->dccpd_type) {
643 case DCCP_PKT_DATA:
644 case DCCP_PKT_DATAACK:
645 hc->rx_data++;
646 if (hc->rx_data >= dp->dccps_r_ack_ratio) {
647 dccp_send_ack(sk);
648 hc->rx_data = 0;
649 }
650 break;
651 }
652}
653
654struct ccid_operations ccid2_ops = {
655 .ccid_id = DCCPC_CCID2,
656 .ccid_name = "TCP-like",
657 .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock),
658 .ccid_hc_tx_init = ccid2_hc_tx_init,
659 .ccid_hc_tx_exit = ccid2_hc_tx_exit,
660 .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet,
661 .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent,
662 .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
663 .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv,
664 .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock),
665 .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv,
666};
667
668#ifdef CONFIG_IP_DCCP_CCID2_DEBUG
669module_param(ccid2_debug, bool, 0644);
670MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages");
671#endif