?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[DCCP]: enable debug messages also for static builds
[linux-2.6/zen-sources.git] / net / dccp / ccids / ccid2.c
blob0fb0d66544a2a4b798538e992de7fa1f55cfe847
1 /*
2 * net/dccp/ccids/ccid2.c
3 *
4 * Copyright (c) 2005, 2006 Andrea Bittau <a.bittau@cs.ucl.ac.uk>
5 *
6 * Changes to meet Linux coding standards, and DCCP infrastructure fixes.
7 *
8 * Copyright (c) 2006 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 */
24
25 /*
26 * This implementation should follow RFC 4341
27 *
28 * BUGS:
29 * - sequence number wrapping
30 */
31
32 #include "../ccid.h"
33 #include "../dccp.h"
34 #include "ccid2.h"
35
36
37 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
38 static int ccid2_debug;
39 #define ccid2_pr_debug(format, a...) DCCP_PR_DEBUG(ccid2_debug, format, ##a)
40
41 static void ccid2_hc_tx_check_sanity(const struct ccid2_hc_tx_sock *hctx)
42 {
43 int len = 0;
44 int pipe = 0;
45 struct ccid2_seq *seqp = hctx->ccid2hctx_seqh;
46
47 /* there is data in the chain */
48 if (seqp != hctx->ccid2hctx_seqt) {
49 seqp = seqp->ccid2s_prev;
50 len++;
51 if (!seqp->ccid2s_acked)
52 pipe++;
53
54 while (seqp != hctx->ccid2hctx_seqt) {
55 struct ccid2_seq *prev = seqp->ccid2s_prev;
56
57 len++;
58 if (!prev->ccid2s_acked)
59 pipe++;
60
61 /* packets are sent sequentially */
62 BUG_ON(seqp->ccid2s_seq <= prev->ccid2s_seq);
63 BUG_ON(time_before(seqp->ccid2s_sent,
64 prev->ccid2s_sent));
65
66 seqp = prev;
67 }
68 }
69
70 BUG_ON(pipe != hctx->ccid2hctx_pipe);
71 ccid2_pr_debug("len of chain=%d\n", len);
72
73 do {
74 seqp = seqp->ccid2s_prev;
75 len++;
76 } while (seqp != hctx->ccid2hctx_seqh);
77
78 ccid2_pr_debug("total len=%d\n", len);
79 BUG_ON(len != hctx->ccid2hctx_seqbufc * CCID2_SEQBUF_LEN);
80 }
81 #else
82 #define ccid2_pr_debug(format, a...)
83 #define ccid2_hc_tx_check_sanity(hctx)
84 #endif
85
86 static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hctx, int num,
87 gfp_t gfp)
88 {
89 struct ccid2_seq *seqp;
90 int i;
91
92 /* check if we have space to preserve the pointer to the buffer */
93 if (hctx->ccid2hctx_seqbufc >= (sizeof(hctx->ccid2hctx_seqbuf) /
94 sizeof(struct ccid2_seq*)))
95 return -ENOMEM;
96
97 /* allocate buffer and initialize linked list */
98 seqp = kmalloc(sizeof(*seqp) * num, gfp);
99 if (seqp == NULL)
100 return -ENOMEM;
101
102 for (i = 0; i < (num - 1); i++) {
103 seqp[i].ccid2s_next = &seqp[i + 1];
104 seqp[i + 1].ccid2s_prev = &seqp[i];
105 }
106 seqp[num - 1].ccid2s_next = seqp;
107 seqp->ccid2s_prev = &seqp[num - 1];
108
109 /* This is the first allocation. Initiate the head and tail. */
110 if (hctx->ccid2hctx_seqbufc == 0)
111 hctx->ccid2hctx_seqh = hctx->ccid2hctx_seqt = seqp;
112 else {
113 /* link the existing list with the one we just created */
114 hctx->ccid2hctx_seqh->ccid2s_next = seqp;
115 seqp->ccid2s_prev = hctx->ccid2hctx_seqh;
116
117 hctx->ccid2hctx_seqt->ccid2s_prev = &seqp[num - 1];
118 seqp[num - 1].ccid2s_next = hctx->ccid2hctx_seqt;
119 }
120
121 /* store the original pointer to the buffer so we can free it */
122 hctx->ccid2hctx_seqbuf[hctx->ccid2hctx_seqbufc] = seqp;
123 hctx->ccid2hctx_seqbufc++;
124
125 return 0;
126 }
127
128 static int ccid2_hc_tx_send_packet(struct sock *sk,
129 struct sk_buff *skb, int len)
130 {
131 struct ccid2_hc_tx_sock *hctx;
132
133 switch (DCCP_SKB_CB(skb)->dccpd_type) {
134 case 0: /* XXX data packets from userland come through like this */
135 case DCCP_PKT_DATA:
136 case DCCP_PKT_DATAACK:
137 break;
138 /* No congestion control on other packets */
139 default:
140 return 0;
141 }
142
143 hctx = ccid2_hc_tx_sk(sk);
144
145 ccid2_pr_debug("pipe=%d cwnd=%d\n", hctx->ccid2hctx_pipe,
146 hctx->ccid2hctx_cwnd);
147
148 if (hctx->ccid2hctx_pipe < hctx->ccid2hctx_cwnd) {
149 /* OK we can send... make sure previous packet was sent off */
150 if (!hctx->ccid2hctx_sendwait) {
151 hctx->ccid2hctx_sendwait = 1;
152 return 0;
153 }
154 }
155
156 return 1; /* XXX CCID should dequeue when ready instead of polling */
157 }
158
159 static void ccid2_change_l_ack_ratio(struct sock *sk, int val)
160 {
161 struct dccp_sock *dp = dccp_sk(sk);
162 /*
163 * XXX I don't really agree with val != 2. If cwnd is 1, ack ratio
164 * should be 1... it shouldn't be allowed to become 2.
165 * -sorbo.
166 */
167 if (val != 2) {
168 const struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
169 int max = hctx->ccid2hctx_cwnd / 2;
170
171 /* round up */
172 if (hctx->ccid2hctx_cwnd & 1)
173 max++;
174
175 if (val > max)
176 val = max;
177 }
178
179 ccid2_pr_debug("changing local ack ratio to %d\n", val);
180 WARN_ON(val <= 0);
181 dp->dccps_l_ack_ratio = val;
182 }
183
184 static void ccid2_change_cwnd(struct ccid2_hc_tx_sock *hctx, int val)
185 {
186 if (val == 0)
187 val = 1;
188
189 /* XXX do we need to change ack ratio? */
190 ccid2_pr_debug("change cwnd to %d\n", val);
191
192 BUG_ON(val < 1);
193 hctx->ccid2hctx_cwnd = val;
194 }
195
196 static void ccid2_change_srtt(struct ccid2_hc_tx_sock *hctx, long val)
197 {
198 ccid2_pr_debug("change SRTT to %ld\n", val);
199 hctx->ccid2hctx_srtt = val;
200 }
201
202 static void ccid2_change_pipe(struct ccid2_hc_tx_sock *hctx, long val)
203 {
204 hctx->ccid2hctx_pipe = val;
205 }
206
207 static void ccid2_start_rto_timer(struct sock *sk);
208
209 static void ccid2_hc_tx_rto_expire(unsigned long data)
210 {
211 struct sock *sk = (struct sock *)data;
212 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
213 long s;
214
215 bh_lock_sock(sk);
216 if (sock_owned_by_user(sk)) {
217 sk_reset_timer(sk, &hctx->ccid2hctx_rtotimer,
218 jiffies + HZ / 5);
219 goto out;
220 }
221
222 ccid2_pr_debug("RTO_EXPIRE\n");
223
224 ccid2_hc_tx_check_sanity(hctx);
225
226 /* back-off timer */
227 hctx->ccid2hctx_rto <<= 1;
228
229 s = hctx->ccid2hctx_rto / HZ;
230 if (s > 60)
231 hctx->ccid2hctx_rto = 60 * HZ;
232
233 ccid2_start_rto_timer(sk);
234
235 /* adjust pipe, cwnd etc */
236 ccid2_change_pipe(hctx, 0);
237 hctx->ccid2hctx_ssthresh = hctx->ccid2hctx_cwnd >> 1;
238 if (hctx->ccid2hctx_ssthresh < 2)
239 hctx->ccid2hctx_ssthresh = 2;
240 ccid2_change_cwnd(hctx, 1);
241
242 /* clear state about stuff we sent */
243 hctx->ccid2hctx_seqt = hctx->ccid2hctx_seqh;
244 hctx->ccid2hctx_ssacks = 0;
245 hctx->ccid2hctx_acks = 0;
246 hctx->ccid2hctx_sent = 0;
247
248 /* clear ack ratio state. */
249 hctx->ccid2hctx_arsent = 0;
250 hctx->ccid2hctx_ackloss = 0;
251 hctx->ccid2hctx_rpseq = 0;
252 hctx->ccid2hctx_rpdupack = -1;
253 ccid2_change_l_ack_ratio(sk, 1);
254 ccid2_hc_tx_check_sanity(hctx);
255 out:
256 bh_unlock_sock(sk);
257 sock_put(sk);
258 }
259
260 static void ccid2_start_rto_timer(struct sock *sk)
261 {
262 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
263
264 ccid2_pr_debug("setting RTO timeout=%ld\n", hctx->ccid2hctx_rto);
265
266 BUG_ON(timer_pending(&hctx->ccid2hctx_rtotimer));
267 sk_reset_timer(sk, &hctx->ccid2hctx_rtotimer,
268 jiffies + hctx->ccid2hctx_rto);
269 }
270
271 static void ccid2_hc_tx_packet_sent(struct sock *sk, int more, int len)
272 {
273 struct dccp_sock *dp = dccp_sk(sk);
274 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
275 struct ccid2_seq *next;
276 u64 seq;
277
278 ccid2_hc_tx_check_sanity(hctx);
279
280 BUG_ON(!hctx->ccid2hctx_sendwait);
281 hctx->ccid2hctx_sendwait = 0;
282 ccid2_change_pipe(hctx, hctx->ccid2hctx_pipe + 1);
283 BUG_ON(hctx->ccid2hctx_pipe < 0);
284
285 /* There is an issue. What if another packet is sent between
286 * packet_send() and packet_sent(). Then the sequence number would be
287 * wrong.
288 * -sorbo.
289 */
290 seq = dp->dccps_gss;
291
292 hctx->ccid2hctx_seqh->ccid2s_seq = seq;
293 hctx->ccid2hctx_seqh->ccid2s_acked = 0;
294 hctx->ccid2hctx_seqh->ccid2s_sent = jiffies;
295
296 next = hctx->ccid2hctx_seqh->ccid2s_next;
297 /* check if we need to alloc more space */
298 if (next == hctx->ccid2hctx_seqt) {
299 int rc;
300
301 ccid2_pr_debug("allocating more space in history\n");
302 rc = ccid2_hc_tx_alloc_seq(hctx, CCID2_SEQBUF_LEN, GFP_KERNEL);
303 BUG_ON(rc); /* XXX what do we do? */
304
305 next = hctx->ccid2hctx_seqh->ccid2s_next;
306 BUG_ON(next == hctx->ccid2hctx_seqt);
307 }
308 hctx->ccid2hctx_seqh = next;
309
310 ccid2_pr_debug("cwnd=%d pipe=%d\n", hctx->ccid2hctx_cwnd,
311 hctx->ccid2hctx_pipe);
312
313 hctx->ccid2hctx_sent++;
314
315 /* Ack Ratio. Need to maintain a concept of how many windows we sent */
316 hctx->ccid2hctx_arsent++;
317 /* We had an ack loss in this window... */
318 if (hctx->ccid2hctx_ackloss) {
319 if (hctx->ccid2hctx_arsent >= hctx->ccid2hctx_cwnd) {
320 hctx->ccid2hctx_arsent = 0;
321 hctx->ccid2hctx_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 = hctx->ccid2hctx_cwnd * hctx->ccid2hctx_cwnd / denom;
332
333 if (hctx->ccid2hctx_arsent >= denom) {
334 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
335 hctx->ccid2hctx_arsent = 0;
336 }
337 } else {
338 /* we can't increase ack ratio further [1] */
339 hctx->ccid2hctx_arsent = 0; /* or maybe set it to cwnd*/
340 }
341 }
342
343 /* setup RTO timer */
344 if (!timer_pending(&hctx->ccid2hctx_rtotimer))
345 ccid2_start_rto_timer(sk);
346
347 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
348 ccid2_pr_debug("pipe=%d\n", hctx->ccid2hctx_pipe);
349 ccid2_pr_debug("Sent: seq=%llu\n", (unsigned long long)seq);
350 do {
351 struct ccid2_seq *seqp = hctx->ccid2hctx_seqt;
352
353 while (seqp != hctx->ccid2hctx_seqh) {
354 ccid2_pr_debug("out seq=%llu acked=%d time=%lu\n",
355 (unsigned long long)seqp->ccid2s_seq,
356 seqp->ccid2s_acked, seqp->ccid2s_sent);
357 seqp = seqp->ccid2s_next;
358 }
359 } while (0);
360 ccid2_pr_debug("=========\n");
361 ccid2_hc_tx_check_sanity(hctx);
362 #endif
363 }
364
365 /* XXX Lame code duplication!
366 * returns -1 if none was found.
367 * else returns the next offset to use in the function call.
368 */
369 static int ccid2_ackvector(struct sock *sk, struct sk_buff *skb, int offset,
370 unsigned char **vec, unsigned char *veclen)
371 {
372 const struct dccp_hdr *dh = dccp_hdr(skb);
373 unsigned char *options = (unsigned char *)dh + dccp_hdr_len(skb);
374 unsigned char *opt_ptr;
375 const unsigned char *opt_end = (unsigned char *)dh +
376 (dh->dccph_doff * 4);
377 unsigned char opt, len;
378 unsigned char *value;
379
380 BUG_ON(offset < 0);
381 options += offset;
382 opt_ptr = options;
383 if (opt_ptr >= opt_end)
384 return -1;
385
386 while (opt_ptr != opt_end) {
387 opt = *opt_ptr++;
388 len = 0;
389 value = NULL;
390
391 /* Check if this isn't a single byte option */
392 if (opt > DCCPO_MAX_RESERVED) {
393 if (opt_ptr == opt_end)
394 goto out_invalid_option;
395
396 len = *opt_ptr++;
397 if (len < 3)
398 goto out_invalid_option;
399 /*
400 * Remove the type and len fields, leaving
401 * just the value size
402 */
403 len -= 2;
404 value = opt_ptr;
405 opt_ptr += len;
406
407 if (opt_ptr > opt_end)
408 goto out_invalid_option;
409 }
410
411 switch (opt) {
412 case DCCPO_ACK_VECTOR_0:
413 case DCCPO_ACK_VECTOR_1:
414 *vec = value;
415 *veclen = len;
416 return offset + (opt_ptr - options);
417 }
418 }
419
420 return -1;
421
422 out_invalid_option:
423 BUG_ON(1); /* should never happen... options were previously parsed ! */
424 return -1;
425 }
426
427 static void ccid2_hc_tx_kill_rto_timer(struct sock *sk)
428 {
429 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
430
431 sk_stop_timer(sk, &hctx->ccid2hctx_rtotimer);
432 ccid2_pr_debug("deleted RTO timer\n");
433 }
434
435 static inline void ccid2_new_ack(struct sock *sk,
436 struct ccid2_seq *seqp,
437 unsigned int *maxincr)
438 {
439 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
440
441 /* slow start */
442 if (hctx->ccid2hctx_cwnd < hctx->ccid2hctx_ssthresh) {
443 hctx->ccid2hctx_acks = 0;
444
445 /* We can increase cwnd at most maxincr [ack_ratio/2] */
446 if (*maxincr) {
447 /* increase every 2 acks */
448 hctx->ccid2hctx_ssacks++;
449 if (hctx->ccid2hctx_ssacks == 2) {
450 ccid2_change_cwnd(hctx, hctx->ccid2hctx_cwnd+1);
451 hctx->ccid2hctx_ssacks = 0;
452 *maxincr = *maxincr - 1;
453 }
454 } else {
455 /* increased cwnd enough for this single ack */
456 hctx->ccid2hctx_ssacks = 0;
457 }
458 } else {
459 hctx->ccid2hctx_ssacks = 0;
460 hctx->ccid2hctx_acks++;
461
462 if (hctx->ccid2hctx_acks >= hctx->ccid2hctx_cwnd) {
463 ccid2_change_cwnd(hctx, hctx->ccid2hctx_cwnd + 1);
464 hctx->ccid2hctx_acks = 0;
465 }
466 }
467
468 /* update RTO */
469 if (hctx->ccid2hctx_srtt == -1 ||
470 time_after(jiffies, hctx->ccid2hctx_lastrtt + hctx->ccid2hctx_srtt)) {
471 unsigned long r = (long)jiffies - (long)seqp->ccid2s_sent;
472 int s;
473
474 /* first measurement */
475 if (hctx->ccid2hctx_srtt == -1) {
476 ccid2_pr_debug("R: %lu Time=%lu seq=%llu\n",
477 r, jiffies,
478 (unsigned long long)seqp->ccid2s_seq);
479 ccid2_change_srtt(hctx, r);
480 hctx->ccid2hctx_rttvar = r >> 1;
481 } else {
482 /* RTTVAR */
483 long tmp = hctx->ccid2hctx_srtt - r;
484 long srtt;
485
486 if (tmp < 0)
487 tmp *= -1;
488
489 tmp >>= 2;
490 hctx->ccid2hctx_rttvar *= 3;
491 hctx->ccid2hctx_rttvar >>= 2;
492 hctx->ccid2hctx_rttvar += tmp;
493
494 /* SRTT */
495 srtt = hctx->ccid2hctx_srtt;
496 srtt *= 7;
497 srtt >>= 3;
498 tmp = r >> 3;
499 srtt += tmp;
500 ccid2_change_srtt(hctx, srtt);
501 }
502 s = hctx->ccid2hctx_rttvar << 2;
503 /* clock granularity is 1 when based on jiffies */
504 if (!s)
505 s = 1;
506 hctx->ccid2hctx_rto = hctx->ccid2hctx_srtt + s;
507
508 /* must be at least a second */
509 s = hctx->ccid2hctx_rto / HZ;
510 /* DCCP doesn't require this [but I like it cuz my code sux] */
511 #if 1
512 if (s < 1)
513 hctx->ccid2hctx_rto = HZ;
514 #endif
515 /* max 60 seconds */
516 if (s > 60)
517 hctx->ccid2hctx_rto = HZ * 60;
518
519 hctx->ccid2hctx_lastrtt = jiffies;
520
521 ccid2_pr_debug("srtt: %ld rttvar: %ld rto: %ld (HZ=%d) R=%lu\n",
522 hctx->ccid2hctx_srtt, hctx->ccid2hctx_rttvar,
523 hctx->ccid2hctx_rto, HZ, r);
524 hctx->ccid2hctx_sent = 0;
525 }
526
527 /* we got a new ack, so re-start RTO timer */
528 ccid2_hc_tx_kill_rto_timer(sk);
529 ccid2_start_rto_timer(sk);
530 }
531
532 static void ccid2_hc_tx_dec_pipe(struct sock *sk)
533 {
534 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
535
536 ccid2_change_pipe(hctx, hctx->ccid2hctx_pipe-1);
537 BUG_ON(hctx->ccid2hctx_pipe < 0);
538
539 if (hctx->ccid2hctx_pipe == 0)
540 ccid2_hc_tx_kill_rto_timer(sk);
541 }
542
543 static void ccid2_congestion_event(struct ccid2_hc_tx_sock *hctx,
544 struct ccid2_seq *seqp)
545 {
546 if (time_before(seqp->ccid2s_sent, hctx->ccid2hctx_last_cong)) {
547 ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
548 return;
549 }
550
551 hctx->ccid2hctx_last_cong = jiffies;
552
553 ccid2_change_cwnd(hctx, hctx->ccid2hctx_cwnd >> 1);
554 hctx->ccid2hctx_ssthresh = hctx->ccid2hctx_cwnd;
555 if (hctx->ccid2hctx_ssthresh < 2)
556 hctx->ccid2hctx_ssthresh = 2;
557 }
558
559 static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
560 {
561 struct dccp_sock *dp = dccp_sk(sk);
562 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
563 u64 ackno, seqno;
564 struct ccid2_seq *seqp;
565 unsigned char *vector;
566 unsigned char veclen;
567 int offset = 0;
568 int done = 0;
569 unsigned int maxincr = 0;
570
571 ccid2_hc_tx_check_sanity(hctx);
572 /* check reverse path congestion */
573 seqno = DCCP_SKB_CB(skb)->dccpd_seq;
574
575 /* XXX this whole "algorithm" is broken. Need to fix it to keep track
576 * of the seqnos of the dupacks so that rpseq and rpdupack are correct
577 * -sorbo.
578 */
579 /* need to bootstrap */
580 if (hctx->ccid2hctx_rpdupack == -1) {
581 hctx->ccid2hctx_rpdupack = 0;
582 hctx->ccid2hctx_rpseq = seqno;
583 } else {
584 /* check if packet is consecutive */
585 if ((hctx->ccid2hctx_rpseq + 1) == seqno)
586 hctx->ccid2hctx_rpseq++;
587 /* it's a later packet */
588 else if (after48(seqno, hctx->ccid2hctx_rpseq)) {
589 hctx->ccid2hctx_rpdupack++;
590
591 /* check if we got enough dupacks */
592 if (hctx->ccid2hctx_rpdupack >=
593 hctx->ccid2hctx_numdupack) {
594 hctx->ccid2hctx_rpdupack = -1; /* XXX lame */
595 hctx->ccid2hctx_rpseq = 0;
596
597 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio << 1);
598 }
599 }
600 }
601
602 /* check forward path congestion */
603 /* still didn't send out new data packets */
604 if (hctx->ccid2hctx_seqh == hctx->ccid2hctx_seqt)
605 return;
606
607 switch (DCCP_SKB_CB(skb)->dccpd_type) {
608 case DCCP_PKT_ACK:
609 case DCCP_PKT_DATAACK:
610 break;
611 default:
612 return;
613 }
614
615 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
616 if (after48(ackno, hctx->ccid2hctx_high_ack))
617 hctx->ccid2hctx_high_ack = ackno;
618
619 seqp = hctx->ccid2hctx_seqt;
620 while (before48(seqp->ccid2s_seq, ackno)) {
621 seqp = seqp->ccid2s_next;
622 if (seqp == hctx->ccid2hctx_seqh) {
623 seqp = hctx->ccid2hctx_seqh->ccid2s_prev;
624 break;
625 }
626 }
627
628 /* If in slow-start, cwnd can increase at most Ack Ratio / 2 packets for
629 * this single ack. I round up.
630 * -sorbo.
631 */
632 maxincr = dp->dccps_l_ack_ratio >> 1;
633 maxincr++;
634
635 /* go through all ack vectors */
636 while ((offset = ccid2_ackvector(sk, skb, offset,
637 &vector, &veclen)) != -1) {
638 /* go through this ack vector */
639 while (veclen--) {
640 const u8 rl = *vector & DCCP_ACKVEC_LEN_MASK;
641 u64 ackno_end_rl;
642
643 dccp_set_seqno(&ackno_end_rl, ackno - rl);
644 ccid2_pr_debug("ackvec start:%llu end:%llu\n",
645 (unsigned long long)ackno,
646 (unsigned long long)ackno_end_rl);
647 /* if the seqno we are analyzing is larger than the
648 * current ackno, then move towards the tail of our
649 * seqnos.
650 */
651 while (after48(seqp->ccid2s_seq, ackno)) {
652 if (seqp == hctx->ccid2hctx_seqt) {
653 done = 1;
654 break;
655 }
656 seqp = seqp->ccid2s_prev;
657 }
658 if (done)
659 break;
660
661 /* check all seqnos in the range of the vector
662 * run length
663 */
664 while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
665 const u8 state = *vector &
666 DCCP_ACKVEC_STATE_MASK;
667
668 /* new packet received or marked */
669 if (state != DCCP_ACKVEC_STATE_NOT_RECEIVED &&
670 !seqp->ccid2s_acked) {
671 if (state ==
672 DCCP_ACKVEC_STATE_ECN_MARKED) {
673 ccid2_congestion_event(hctx,
674 seqp);
675 } else
676 ccid2_new_ack(sk, seqp,
677 &maxincr);
678
679 seqp->ccid2s_acked = 1;
680 ccid2_pr_debug("Got ack for %llu\n",
681 (unsigned long long)seqp->ccid2s_seq);
682 ccid2_hc_tx_dec_pipe(sk);
683 }
684 if (seqp == hctx->ccid2hctx_seqt) {
685 done = 1;
686 break;
687 }
688 seqp = seqp->ccid2s_next;
689 }
690 if (done)
691 break;
692
693
694 dccp_set_seqno(&ackno, ackno_end_rl - 1);
695 vector++;
696 }
697 if (done)
698 break;
699 }
700
701 /* The state about what is acked should be correct now
702 * Check for NUMDUPACK
703 */
704 seqp = hctx->ccid2hctx_seqt;
705 while (before48(seqp->ccid2s_seq, hctx->ccid2hctx_high_ack)) {
706 seqp = seqp->ccid2s_next;
707 if (seqp == hctx->ccid2hctx_seqh) {
708 seqp = hctx->ccid2hctx_seqh->ccid2s_prev;
709 break;
710 }
711 }
712 done = 0;
713 while (1) {
714 if (seqp->ccid2s_acked) {
715 done++;
716 if (done == hctx->ccid2hctx_numdupack)
717 break;
718 }
719 if (seqp == hctx->ccid2hctx_seqt)
720 break;
721 seqp = seqp->ccid2s_prev;
722 }
723
724 /* If there are at least 3 acknowledgements, anything unacknowledged
725 * below the last sequence number is considered lost
726 */
727 if (done == hctx->ccid2hctx_numdupack) {
728 struct ccid2_seq *last_acked = seqp;
729
730 /* check for lost packets */
731 while (1) {
732 if (!seqp->ccid2s_acked) {
733 ccid2_pr_debug("Packet lost: %llu\n",
734 (unsigned long long)seqp->ccid2s_seq);
735 /* XXX need to traverse from tail -> head in
736 * order to detect multiple congestion events in
737 * one ack vector.
738 */
739 ccid2_congestion_event(hctx, seqp);
740 ccid2_hc_tx_dec_pipe(sk);
741 }
742 if (seqp == hctx->ccid2hctx_seqt)
743 break;
744 seqp = seqp->ccid2s_prev;
745 }
746
747 hctx->ccid2hctx_seqt = last_acked;
748 }
749
750 /* trim acked packets in tail */
751 while (hctx->ccid2hctx_seqt != hctx->ccid2hctx_seqh) {
752 if (!hctx->ccid2hctx_seqt->ccid2s_acked)
753 break;
754
755 hctx->ccid2hctx_seqt = hctx->ccid2hctx_seqt->ccid2s_next;
756 }
757
758 ccid2_hc_tx_check_sanity(hctx);
759 }
760
761 static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
762 {
763 struct ccid2_hc_tx_sock *hctx = ccid_priv(ccid);
764
765 ccid2_change_cwnd(hctx, 1);
766 /* Initialize ssthresh to infinity. This means that we will exit the
767 * initial slow-start after the first packet loss. This is what we
768 * want.
769 */
770 hctx->ccid2hctx_ssthresh = ~0;
771 hctx->ccid2hctx_numdupack = 3;
772 hctx->ccid2hctx_seqbufc = 0;
773
774 /* XXX init ~ to window size... */
775 if (ccid2_hc_tx_alloc_seq(hctx, CCID2_SEQBUF_LEN, GFP_ATOMIC) != 0)
776 return -ENOMEM;
777
778 hctx->ccid2hctx_sent = 0;
779 hctx->ccid2hctx_rto = 3 * HZ;
780 ccid2_change_srtt(hctx, -1);
781 hctx->ccid2hctx_rttvar = -1;
782 hctx->ccid2hctx_lastrtt = 0;
783 hctx->ccid2hctx_rpdupack = -1;
784 hctx->ccid2hctx_last_cong = jiffies;
785 hctx->ccid2hctx_high_ack = 0;
786
787 hctx->ccid2hctx_rtotimer.function = &ccid2_hc_tx_rto_expire;
788 hctx->ccid2hctx_rtotimer.data = (unsigned long)sk;
789 init_timer(&hctx->ccid2hctx_rtotimer);
790
791 ccid2_hc_tx_check_sanity(hctx);
792 return 0;
793 }
794
795 static void ccid2_hc_tx_exit(struct sock *sk)
796 {
797 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
798 int i;
799
800 ccid2_hc_tx_kill_rto_timer(sk);
801
802 for (i = 0; i < hctx->ccid2hctx_seqbufc; i++)
803 kfree(hctx->ccid2hctx_seqbuf[i]);
804 hctx->ccid2hctx_seqbufc = 0;
805 }
806
807 static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
808 {
809 const struct dccp_sock *dp = dccp_sk(sk);
810 struct ccid2_hc_rx_sock *hcrx = ccid2_hc_rx_sk(sk);
811
812 switch (DCCP_SKB_CB(skb)->dccpd_type) {
813 case DCCP_PKT_DATA:
814 case DCCP_PKT_DATAACK:
815 hcrx->ccid2hcrx_data++;
816 if (hcrx->ccid2hcrx_data >= dp->dccps_r_ack_ratio) {
817 dccp_send_ack(sk);
818 hcrx->ccid2hcrx_data = 0;
819 }
820 break;
821 }
822 }
823
824 static struct ccid_operations ccid2 = {
825 .ccid_id = DCCPC_CCID2,
826 .ccid_name = "ccid2",
827 .ccid_owner = THIS_MODULE,
828 .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock),
829 .ccid_hc_tx_init = ccid2_hc_tx_init,
830 .ccid_hc_tx_exit = ccid2_hc_tx_exit,
831 .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet,
832 .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent,
833 .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv,
834 .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock),
835 .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv,
836 };
837
838 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
839 module_param(ccid2_debug, int, 0444);
840 MODULE_PARM_DESC(ccid2_debug, "Enable debug messages");
841 #endif
842
843 static __init int ccid2_module_init(void)
844 {
845 return ccid_register(&ccid2);
846 }
847 module_init(ccid2_module_init);
848
849 static __exit void ccid2_module_exit(void)
850 {
851 ccid_unregister(&ccid2);
852 }
853 module_exit(ccid2_module_exit);
854
855 MODULE_AUTHOR("Andrea Bittau <a.bittau@cs.ucl.ac.uk>");
856 MODULE_DESCRIPTION("DCCP TCP-Like (CCID2) CCID");
857 MODULE_LICENSE("GPL");
858 MODULE_ALIAS("net-dccp-ccid-2");
Stable & featurefull patchset based on Linus's git branch