blob: fa713227c66f10bb6be9518d0accbecea2a28257 [file] [log] [blame]
Andrea Bittau2a91aa32006-03-20 17:41:47 -08001/*
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/*
Gerrit Renker0e64e942006-10-24 16:17:51 -070026 * This implementation should follow RFC 4341
Andrea Bittau2a91aa32006-03-20 17:41:47 -080027 */
Gerrit Renker86349c82008-09-04 07:30:19 +020028#include "../feat.h"
Andrea Bittau2a91aa32006-03-20 17:41:47 -080029#include "../ccid.h"
30#include "../dccp.h"
31#include "ccid2.h"
32
Gerrit Renker84116712006-11-20 18:26:03 -020033
34#ifdef CONFIG_IP_DCCP_CCID2_DEBUG
Andrea Bittau2a91aa32006-03-20 17:41:47 -080035static int ccid2_debug;
Gerrit Renker84116712006-11-20 18:26:03 -020036#define ccid2_pr_debug(format, a...) DCCP_PR_DEBUG(ccid2_debug, format, ##a)
Andrea Bittau2a91aa32006-03-20 17:41:47 -080037#else
Gerrit Renker84116712006-11-20 18:26:03 -020038#define ccid2_pr_debug(format, a...)
Andrea Bittau2a91aa32006-03-20 17:41:47 -080039#endif
40
Gerrit Renkercd1f7d32007-10-04 14:41:00 -070041static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hctx)
Andrea Bittau07978aa2006-09-19 13:13:37 -070042{
43 struct ccid2_seq *seqp;
44 int i;
45
46 /* check if we have space to preserve the pointer to the buffer */
Gerrit Renker1fb87502008-09-04 07:30:19 +020047 if (hctx->seqbufc >= sizeof(hctx->seqbuf) / sizeof(struct ccid2_seq *))
Andrea Bittau07978aa2006-09-19 13:13:37 -070048 return -ENOMEM;
49
50 /* allocate buffer and initialize linked list */
Gerrit Renkercd1f7d32007-10-04 14:41:00 -070051 seqp = kmalloc(CCID2_SEQBUF_LEN * sizeof(struct ccid2_seq), gfp_any());
Andrea Bittau07978aa2006-09-19 13:13:37 -070052 if (seqp == NULL)
53 return -ENOMEM;
54
Gerrit Renkercd1f7d32007-10-04 14:41:00 -070055 for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) {
Andrea Bittau07978aa2006-09-19 13:13:37 -070056 seqp[i].ccid2s_next = &seqp[i + 1];
57 seqp[i + 1].ccid2s_prev = &seqp[i];
58 }
Gerrit Renkercd1f7d32007-10-04 14:41:00 -070059 seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp;
60 seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
Andrea Bittau07978aa2006-09-19 13:13:37 -070061
62 /* This is the first allocation. Initiate the head and tail. */
Gerrit Renker1fb87502008-09-04 07:30:19 +020063 if (hctx->seqbufc == 0)
64 hctx->seqh = hctx->seqt = seqp;
Andrea Bittau07978aa2006-09-19 13:13:37 -070065 else {
66 /* link the existing list with the one we just created */
Gerrit Renker1fb87502008-09-04 07:30:19 +020067 hctx->seqh->ccid2s_next = seqp;
68 seqp->ccid2s_prev = hctx->seqh;
Andrea Bittau07978aa2006-09-19 13:13:37 -070069
Gerrit Renker1fb87502008-09-04 07:30:19 +020070 hctx->seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
71 seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hctx->seqt;
Andrea Bittau07978aa2006-09-19 13:13:37 -070072 }
73
74 /* store the original pointer to the buffer so we can free it */
Gerrit Renker1fb87502008-09-04 07:30:19 +020075 hctx->seqbuf[hctx->seqbufc] = seqp;
76 hctx->seqbufc++;
Andrea Bittau07978aa2006-09-19 13:13:37 -070077
78 return 0;
79}
80
Gerrit Renker6b57c932006-11-28 19:55:06 -020081static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
Andrea Bittau2a91aa32006-03-20 17:41:47 -080082{
Gerrit Renker83337da2008-09-04 07:30:19 +020083 if (ccid2_cwnd_network_limited(ccid2_hc_tx_sk(sk)))
84 return CCID_PACKET_WILL_DEQUEUE_LATER;
85 return CCID_PACKET_SEND_AT_ONCE;
Andrea Bittau2a91aa32006-03-20 17:41:47 -080086}
87
Gerrit Renkerdf054e12007-11-24 21:32:53 -020088static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
Andrea Bittau2a91aa32006-03-20 17:41:47 -080089{
90 struct dccp_sock *dp = dccp_sk(sk);
Gerrit Renker1fb87502008-09-04 07:30:19 +020091 u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->cwnd, 2);
Gerrit Renkerd50ad162007-11-24 21:40:24 -020092
Andrea Bittau2a91aa32006-03-20 17:41:47 -080093 /*
Gerrit Renkerd50ad162007-11-24 21:40:24 -020094 * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from
95 * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always
96 * acceptable since this causes starvation/deadlock whenever cwnd < 2.
97 * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled).
Andrea Bittau2a91aa32006-03-20 17:41:47 -080098 */
Gerrit Renkerd50ad162007-11-24 21:40:24 -020099 if (val == 0 || val > max_ratio) {
100 DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio);
101 val = max_ratio;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800102 }
Gerrit Renker86349c82008-09-04 07:30:19 +0200103 if (val > DCCPF_ACK_RATIO_MAX)
104 val = DCCPF_ACK_RATIO_MAX;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800105
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200106 if (val == dp->dccps_l_ack_ratio)
107 return;
108
Gerrit Renkerdf054e12007-11-24 21:32:53 -0200109 ccid2_pr_debug("changing local ack ratio to %u\n", val);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800110 dp->dccps_l_ack_ratio = val;
111}
112
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800113static void ccid2_hc_tx_rto_expire(unsigned long data)
114{
115 struct sock *sk = (struct sock *)data;
116 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
Gerrit Renker83337da2008-09-04 07:30:19 +0200117 const bool sender_was_blocked = ccid2_cwnd_network_limited(hctx);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800118
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800119 bh_lock_sock(sk);
120 if (sock_owned_by_user(sk)) {
Gerrit Renker1fb87502008-09-04 07:30:19 +0200121 sk_reset_timer(sk, &hctx->rtotimer, jiffies + HZ / 5);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800122 goto out;
123 }
124
125 ccid2_pr_debug("RTO_EXPIRE\n");
126
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800127 /* back-off timer */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200128 hctx->rto <<= 1;
Gerrit Renker14355622008-09-04 07:30:19 +0200129 if (hctx->rto > DCCP_RTO_MAX)
130 hctx->rto = DCCP_RTO_MAX;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800131
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800132 /* adjust pipe, cwnd etc */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200133 hctx->ssthresh = hctx->cwnd / 2;
134 if (hctx->ssthresh < 2)
135 hctx->ssthresh = 2;
136 hctx->cwnd = 1;
137 hctx->pipe = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800138
139 /* clear state about stuff we sent */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200140 hctx->seqt = hctx->seqh;
141 hctx->packets_acked = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800142
143 /* clear ack ratio state. */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200144 hctx->rpseq = 0;
145 hctx->rpdupack = -1;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800146 ccid2_change_l_ack_ratio(sk, 1);
Gerrit Renker83337da2008-09-04 07:30:19 +0200147
148 /* if we were blocked before, we may now send cwnd=1 packet */
149 if (sender_was_blocked)
150 tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
Gerrit Renker20bbd0f2008-09-04 07:30:19 +0200151 /* restart backed-off timer */
152 sk_reset_timer(sk, &hctx->rtotimer, jiffies + hctx->rto);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800153out:
154 bh_unlock_sock(sk);
Andrea Bittau77ff72d2006-03-20 17:57:52 -0800155 sock_put(sk);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800156}
157
Gerrit Renkerc506d912008-09-04 07:30:19 +0200158static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800159{
160 struct dccp_sock *dp = dccp_sk(sk);
161 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700162 struct ccid2_seq *next;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800163
Gerrit Renker1fb87502008-09-04 07:30:19 +0200164 hctx->pipe++;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800165
Gerrit Renker1fb87502008-09-04 07:30:19 +0200166 hctx->seqh->ccid2s_seq = dp->dccps_gss;
167 hctx->seqh->ccid2s_acked = 0;
168 hctx->seqh->ccid2s_sent = jiffies;
Andrea Bittau07978aa2006-09-19 13:13:37 -0700169
Gerrit Renker1fb87502008-09-04 07:30:19 +0200170 next = hctx->seqh->ccid2s_next;
Andrea Bittau07978aa2006-09-19 13:13:37 -0700171 /* check if we need to alloc more space */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200172 if (next == hctx->seqt) {
Gerrit Renker7d9e8932007-10-04 14:41:26 -0700173 if (ccid2_hc_tx_alloc_seq(hctx)) {
174 DCCP_CRIT("packet history - out of memory!");
175 /* FIXME: find a more graceful way to bail out */
176 return;
177 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200178 next = hctx->seqh->ccid2s_next;
179 BUG_ON(next == hctx->seqt);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700180 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200181 hctx->seqh = next;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800182
Gerrit Renker1fb87502008-09-04 07:30:19 +0200183 ccid2_pr_debug("cwnd=%d pipe=%d\n", hctx->cwnd, hctx->pipe);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800184
Gerrit Renker900bfed2007-11-24 21:58:33 -0200185 /*
186 * FIXME: The code below is broken and the variables have been removed
187 * from the socket struct. The `ackloss' variable was always set to 0,
188 * and with arsent there are several problems:
189 * (i) it doesn't just count the number of Acks, but all sent packets;
190 * (ii) it is expressed in # of packets, not # of windows, so the
191 * comparison below uses the wrong formula: Appendix A of RFC 4341
192 * comes up with the number K = cwnd / (R^2 - R) of consecutive windows
193 * of data with no lost or marked Ack packets. If arsent were the # of
194 * consecutive Acks received without loss, then Ack Ratio needs to be
195 * decreased by 1 when
196 * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2)
197 * where cwnd / R is the number of Acks received per window of data
198 * (cf. RFC 4341, App. A). The problems are that
199 * - arsent counts other packets as well;
200 * - the comparison uses a formula different from RFC 4341;
201 * - computing a cubic/quadratic equation each time is too complicated.
202 * Hence a different algorithm is needed.
203 */
204#if 0
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800205 /* Ack Ratio. Need to maintain a concept of how many windows we sent */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200206 hctx->arsent++;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800207 /* We had an ack loss in this window... */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200208 if (hctx->ackloss) {
209 if (hctx->arsent >= hctx->cwnd) {
210 hctx->arsent = 0;
211 hctx->ackloss = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800212 }
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800213 } else {
214 /* No acks lost up to now... */
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800215 /* decrease ack ratio if enough packets were sent */
216 if (dp->dccps_l_ack_ratio > 1) {
217 /* XXX don't calculate denominator each time */
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800218 int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
219 dp->dccps_l_ack_ratio;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800220
Gerrit Renker1fb87502008-09-04 07:30:19 +0200221 denom = hctx->cwnd * hctx->cwnd / denom;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800222
Gerrit Renker1fb87502008-09-04 07:30:19 +0200223 if (hctx->arsent >= denom) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800224 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
Gerrit Renker1fb87502008-09-04 07:30:19 +0200225 hctx->arsent = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800226 }
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800227 } else {
228 /* we can't increase ack ratio further [1] */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200229 hctx->arsent = 0; /* or maybe set it to cwnd*/
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800230 }
231 }
Gerrit Renker900bfed2007-11-24 21:58:33 -0200232#endif
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800233
234 /* setup RTO timer */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200235 if (!timer_pending(&hctx->rtotimer))
Gerrit Renker20bbd0f2008-09-04 07:30:19 +0200236 sk_reset_timer(sk, &hctx->rtotimer, jiffies + hctx->rto);
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800237
Andrea Bittau8d424f62006-09-19 13:12:44 -0700238#ifdef CONFIG_IP_DCCP_CCID2_DEBUG
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800239 do {
Gerrit Renker1fb87502008-09-04 07:30:19 +0200240 struct ccid2_seq *seqp = hctx->seqt;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800241
Gerrit Renker1fb87502008-09-04 07:30:19 +0200242 while (seqp != hctx->seqh) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800243 ccid2_pr_debug("out seq=%llu acked=%d time=%lu\n",
Arnaldo Carvalho de Melo8109b022006-12-10 16:01:18 -0200244 (unsigned long long)seqp->ccid2s_seq,
Randy Dunlap234af482006-10-29 16:03:30 -0800245 seqp->ccid2s_acked, seqp->ccid2s_sent);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800246 seqp = seqp->ccid2s_next;
247 }
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800248 } while (0);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800249 ccid2_pr_debug("=========\n");
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800250#endif
251}
252
Gerrit Renker14355622008-09-04 07:30:19 +0200253/**
254 * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
255 * This code is almost identical with TCP's tcp_rtt_estimator(), since
256 * - it has a higher sampling frequency (recommended by RFC 1323),
257 * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
258 * - it is simple (cf. more complex proposals such as Eifel timer or research
259 * which suggests that the gain should be set according to window size),
260 * - in tests it was found to work well with CCID2 [gerrit].
261 */
262static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
263{
264 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
265 long m = mrtt ? : 1;
266
267 if (hctx->srtt == 0) {
268 /* First measurement m */
269 hctx->srtt = m << 3;
270 hctx->mdev = m << 1;
271
272 hctx->mdev_max = max(TCP_RTO_MIN, hctx->mdev);
273 hctx->rttvar = hctx->mdev_max;
274 hctx->rtt_seq = dccp_sk(sk)->dccps_gss;
275 } else {
276 /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
277 m -= (hctx->srtt >> 3);
278 hctx->srtt += m;
279
280 /* Similarly, update scaled mdev with regard to |m| */
281 if (m < 0) {
282 m = -m;
283 m -= (hctx->mdev >> 2);
284 /*
285 * This neutralises RTO increase when RTT < SRTT - mdev
286 * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
287 * in Linux TCP", USENIX 2002, pp. 49-62).
288 */
289 if (m > 0)
290 m >>= 3;
291 } else {
292 m -= (hctx->mdev >> 2);
293 }
294 hctx->mdev += m;
295
296 if (hctx->mdev > hctx->mdev_max) {
297 hctx->mdev_max = hctx->mdev;
298 if (hctx->mdev_max > hctx->rttvar)
299 hctx->rttvar = hctx->mdev_max;
300 }
301
302 /*
303 * Decay RTTVAR at most once per flight, exploiting that
304 * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2)
305 * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1)
306 * GAR is a useful bound for FlightSize = pipe, AWL is probably
307 * too low as it over-estimates pipe.
308 */
309 if (after48(dccp_sk(sk)->dccps_gar, hctx->rtt_seq)) {
310 if (hctx->mdev_max < hctx->rttvar)
311 hctx->rttvar -= (hctx->rttvar -
312 hctx->mdev_max) >> 2;
313 hctx->rtt_seq = dccp_sk(sk)->dccps_gss;
314 hctx->mdev_max = TCP_RTO_MIN;
315 }
316 }
317
318 /*
319 * Set RTO from SRTT and RTTVAR
320 * Clock granularity is ignored since the minimum error for RTTVAR is
321 * clamped to 50msec (corresponding to HZ=20). This leads to a minimum
322 * RTO of 200msec. This agrees with TCP and RFC 4341, 5.: "Because DCCP
323 * does not retransmit data, DCCP does not require TCP's recommended
324 * minimum timeout of one second".
325 */
326 hctx->rto = (hctx->srtt >> 3) + hctx->rttvar;
327
328 if (hctx->rto > DCCP_RTO_MAX)
329 hctx->rto = DCCP_RTO_MAX;
330}
331
332static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
333 unsigned int *maxincr)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800334{
335 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
336
Gerrit Renker1fb87502008-09-04 07:30:19 +0200337 if (hctx->cwnd < hctx->ssthresh) {
338 if (*maxincr > 0 && ++hctx->packets_acked == 2) {
339 hctx->cwnd += 1;
340 *maxincr -= 1;
341 hctx->packets_acked = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800342 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200343 } else if (++hctx->packets_acked >= hctx->cwnd) {
344 hctx->cwnd += 1;
345 hctx->packets_acked = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800346 }
Gerrit Renker14355622008-09-04 07:30:19 +0200347 /*
348 * FIXME: RTT is sampled several times per acknowledgment (for each
349 * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
350 * This causes the RTT to be over-estimated, since the older entries
351 * in the Ack Vector have earlier sending times.
352 * The cleanest solution is to not use the ccid2s_sent field at all
353 * and instead use DCCP timestamps - need to be resolved at some time.
354 */
355 ccid2_rtt_estimator(sk, jiffies - seqp->ccid2s_sent);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800356}
357
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200358static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
Andrea Bittau374bcf32006-09-19 13:14:43 -0700359{
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200360 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
361
Gerrit Renker1fb87502008-09-04 07:30:19 +0200362 if (time_before(seqp->ccid2s_sent, hctx->last_cong)) {
Andrea Bittau374bcf32006-09-19 13:14:43 -0700363 ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
364 return;
365 }
366
Gerrit Renker1fb87502008-09-04 07:30:19 +0200367 hctx->last_cong = jiffies;
Andrea Bittau374bcf32006-09-19 13:14:43 -0700368
Gerrit Renker1fb87502008-09-04 07:30:19 +0200369 hctx->cwnd = hctx->cwnd / 2 ? : 1U;
370 hctx->ssthresh = max(hctx->cwnd, 2U);
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200371
372 /* Avoid spurious timeouts resulting from Ack Ratio > cwnd */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200373 if (dccp_sk(sk)->dccps_l_ack_ratio > hctx->cwnd)
374 ccid2_change_l_ack_ratio(sk, hctx->cwnd);
Andrea Bittau374bcf32006-09-19 13:14:43 -0700375}
376
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200377static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
378 u8 option, u8 *optval, u8 optlen)
379{
380 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
381
382 switch (option) {
383 case DCCPO_ACK_VECTOR_0:
384 case DCCPO_ACK_VECTOR_1:
385 return dccp_ackvec_parsed_add(&hctx->av_chunks, optval, optlen,
386 option - DCCPO_ACK_VECTOR_0);
387 }
388 return 0;
389}
390
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800391static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
392{
393 struct dccp_sock *dp = dccp_sk(sk);
394 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
Gerrit Renker83337da2008-09-04 07:30:19 +0200395 const bool sender_was_blocked = ccid2_cwnd_network_limited(hctx);
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200396 struct dccp_ackvec_parsed *avp;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800397 u64 ackno, seqno;
398 struct ccid2_seq *seqp;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800399 int done = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800400 unsigned int maxincr = 0;
401
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800402 /* check reverse path congestion */
403 seqno = DCCP_SKB_CB(skb)->dccpd_seq;
404
405 /* XXX this whole "algorithm" is broken. Need to fix it to keep track
406 * of the seqnos of the dupacks so that rpseq and rpdupack are correct
407 * -sorbo.
408 */
409 /* need to bootstrap */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200410 if (hctx->rpdupack == -1) {
411 hctx->rpdupack = 0;
412 hctx->rpseq = seqno;
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800413 } else {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800414 /* check if packet is consecutive */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200415 if (dccp_delta_seqno(hctx->rpseq, seqno) == 1)
416 hctx->rpseq = seqno;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800417 /* it's a later packet */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200418 else if (after48(seqno, hctx->rpseq)) {
419 hctx->rpdupack++;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800420
421 /* check if we got enough dupacks */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200422 if (hctx->rpdupack >= NUMDUPACK) {
423 hctx->rpdupack = -1; /* XXX lame */
424 hctx->rpseq = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800425
Gerrit Renkerdf054e12007-11-24 21:32:53 -0200426 ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800427 }
428 }
429 }
430
431 /* check forward path congestion */
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200432 if (dccp_packet_without_ack(skb))
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800433 return;
434
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200435 /* still didn't send out new data packets */
436 if (hctx->seqh == hctx->seqt)
437 goto done;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800438
439 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
Gerrit Renker1fb87502008-09-04 07:30:19 +0200440 if (after48(ackno, hctx->high_ack))
441 hctx->high_ack = ackno;
Andrea Bittau32aac182006-11-16 14:28:40 -0200442
Gerrit Renker1fb87502008-09-04 07:30:19 +0200443 seqp = hctx->seqt;
Andrea Bittau32aac182006-11-16 14:28:40 -0200444 while (before48(seqp->ccid2s_seq, ackno)) {
445 seqp = seqp->ccid2s_next;
Gerrit Renker1fb87502008-09-04 07:30:19 +0200446 if (seqp == hctx->seqh) {
447 seqp = hctx->seqh->ccid2s_prev;
Andrea Bittau32aac182006-11-16 14:28:40 -0200448 break;
449 }
450 }
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800451
Gerrit Renkera3020022007-11-24 22:10:29 -0200452 /*
453 * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
454 * packets per acknowledgement. Rounding up avoids that cwnd is not
455 * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800456 */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200457 if (hctx->cwnd < hctx->ssthresh)
Gerrit Renkera3020022007-11-24 22:10:29 -0200458 maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800459
460 /* go through all ack vectors */
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200461 list_for_each_entry(avp, &hctx->av_chunks, node) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800462 /* go through this ack vector */
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200463 for (; avp->len--; avp->vec++) {
464 u64 ackno_end_rl = SUB48(ackno,
465 dccp_ackvec_runlen(avp->vec));
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800466
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200467 ccid2_pr_debug("ackvec %llu |%u,%u|\n",
Randy Dunlap234af482006-10-29 16:03:30 -0800468 (unsigned long long)ackno,
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200469 dccp_ackvec_state(avp->vec) >> 6,
470 dccp_ackvec_runlen(avp->vec));
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800471 /* if the seqno we are analyzing is larger than the
472 * current ackno, then move towards the tail of our
473 * seqnos.
474 */
475 while (after48(seqp->ccid2s_seq, ackno)) {
Gerrit Renker1fb87502008-09-04 07:30:19 +0200476 if (seqp == hctx->seqt) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800477 done = 1;
478 break;
479 }
480 seqp = seqp->ccid2s_prev;
481 }
482 if (done)
483 break;
484
485 /* check all seqnos in the range of the vector
486 * run length
487 */
488 while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200489 const u8 state = dccp_ackvec_state(avp->vec);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800490
491 /* new packet received or marked */
Gerrit Renkerff49e272008-09-04 07:30:19 +0200492 if (state != DCCPAV_NOT_RECEIVED &&
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800493 !seqp->ccid2s_acked) {
Gerrit Renkerff49e272008-09-04 07:30:19 +0200494 if (state == DCCPAV_ECN_MARKED)
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200495 ccid2_congestion_event(sk,
Andrea Bittau374bcf32006-09-19 13:14:43 -0700496 seqp);
Gerrit Renkerff49e272008-09-04 07:30:19 +0200497 else
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800498 ccid2_new_ack(sk, seqp,
499 &maxincr);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800500
501 seqp->ccid2s_acked = 1;
502 ccid2_pr_debug("Got ack for %llu\n",
Randy Dunlap234af482006-10-29 16:03:30 -0800503 (unsigned long long)seqp->ccid2s_seq);
Gerrit Renkere9803c02008-09-04 07:30:19 +0200504 hctx->pipe--;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800505 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200506 if (seqp == hctx->seqt) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800507 done = 1;
508 break;
509 }
Gerrit Renker3de54892007-11-24 20:37:48 -0200510 seqp = seqp->ccid2s_prev;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800511 }
512 if (done)
513 break;
514
Gerrit Renkercfbbeab2007-11-24 20:43:59 -0200515 ackno = SUB48(ackno_end_rl, 1);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800516 }
517 if (done)
518 break;
519 }
520
521 /* The state about what is acked should be correct now
522 * Check for NUMDUPACK
523 */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200524 seqp = hctx->seqt;
525 while (before48(seqp->ccid2s_seq, hctx->high_ack)) {
Andrea Bittau32aac182006-11-16 14:28:40 -0200526 seqp = seqp->ccid2s_next;
Gerrit Renker1fb87502008-09-04 07:30:19 +0200527 if (seqp == hctx->seqh) {
528 seqp = hctx->seqh->ccid2s_prev;
Andrea Bittau32aac182006-11-16 14:28:40 -0200529 break;
530 }
531 }
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800532 done = 0;
533 while (1) {
534 if (seqp->ccid2s_acked) {
535 done++;
Gerrit Renker63df18a2007-11-24 22:04:35 -0200536 if (done == NUMDUPACK)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800537 break;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800538 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200539 if (seqp == hctx->seqt)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800540 break;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800541 seqp = seqp->ccid2s_prev;
542 }
543
544 /* If there are at least 3 acknowledgements, anything unacknowledged
545 * below the last sequence number is considered lost
546 */
Gerrit Renker63df18a2007-11-24 22:04:35 -0200547 if (done == NUMDUPACK) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800548 struct ccid2_seq *last_acked = seqp;
549
550 /* check for lost packets */
551 while (1) {
552 if (!seqp->ccid2s_acked) {
Andrea Bittau374bcf32006-09-19 13:14:43 -0700553 ccid2_pr_debug("Packet lost: %llu\n",
Randy Dunlap234af482006-10-29 16:03:30 -0800554 (unsigned long long)seqp->ccid2s_seq);
Andrea Bittau374bcf32006-09-19 13:14:43 -0700555 /* XXX need to traverse from tail -> head in
556 * order to detect multiple congestion events in
557 * one ack vector.
558 */
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200559 ccid2_congestion_event(sk, seqp);
Gerrit Renkere9803c02008-09-04 07:30:19 +0200560 hctx->pipe--;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800561 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200562 if (seqp == hctx->seqt)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800563 break;
564 seqp = seqp->ccid2s_prev;
565 }
566
Gerrit Renker1fb87502008-09-04 07:30:19 +0200567 hctx->seqt = last_acked;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800568 }
569
570 /* trim acked packets in tail */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200571 while (hctx->seqt != hctx->seqh) {
572 if (!hctx->seqt->ccid2s_acked)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800573 break;
574
Gerrit Renker1fb87502008-09-04 07:30:19 +0200575 hctx->seqt = hctx->seqt->ccid2s_next;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800576 }
577
Gerrit Renkere9803c02008-09-04 07:30:19 +0200578 /* restart RTO timer if not all outstanding data has been acked */
579 if (hctx->pipe == 0)
580 sk_stop_timer(sk, &hctx->rtotimer);
581 else
Gerrit Renker14355622008-09-04 07:30:19 +0200582 sk_reset_timer(sk, &hctx->rtotimer, jiffies + hctx->rto);
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200583done:
Gerrit Renker83337da2008-09-04 07:30:19 +0200584 /* check if incoming Acks allow pending packets to be sent */
585 if (sender_was_blocked && !ccid2_cwnd_network_limited(hctx))
586 tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200587 dccp_ackvec_parsed_cleanup(&hctx->av_chunks);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800588}
589
Arnaldo Carvalho de Melo91f0ebf2006-03-20 19:21:44 -0800590static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800591{
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +0900592 struct ccid2_hc_tx_sock *hctx = ccid_priv(ccid);
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200593 struct dccp_sock *dp = dccp_sk(sk);
594 u32 max_ratio;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800595
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200596 /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200597 hctx->ssthresh = ~0U;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800598
Gerrit Renker62248772008-09-04 07:30:19 +0200599 /* Use larger initial windows (RFC 3390, rfc2581bis) */
600 hctx->cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache);
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200601
602 /* Make sure that Ack Ratio is enabled and within bounds. */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200603 max_ratio = DIV_ROUND_UP(hctx->cwnd, 2);
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200604 if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
605 dp->dccps_l_ack_ratio = max_ratio;
606
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800607 /* XXX init ~ to window size... */
Gerrit Renkercd1f7d32007-10-04 14:41:00 -0700608 if (ccid2_hc_tx_alloc_seq(hctx))
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800609 return -ENOMEM;
Arnaldo Carvalho de Melo91f0ebf2006-03-20 19:21:44 -0800610
Gerrit Renker14355622008-09-04 07:30:19 +0200611 hctx->rto = DCCP_TIMEOUT_INIT;
Gerrit Renker1fb87502008-09-04 07:30:19 +0200612 hctx->rpdupack = -1;
613 hctx->last_cong = jiffies;
614 setup_timer(&hctx->rtotimer, ccid2_hc_tx_rto_expire, (unsigned long)sk);
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200615 INIT_LIST_HEAD(&hctx->av_chunks);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800616 return 0;
617}
618
619static void ccid2_hc_tx_exit(struct sock *sk)
620{
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +0900621 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700622 int i;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800623
Gerrit Renker20bbd0f2008-09-04 07:30:19 +0200624 sk_stop_timer(sk, &hctx->rtotimer);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700625
Gerrit Renker1fb87502008-09-04 07:30:19 +0200626 for (i = 0; i < hctx->seqbufc; i++)
627 kfree(hctx->seqbuf[i]);
628 hctx->seqbufc = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800629}
630
631static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
632{
633 const struct dccp_sock *dp = dccp_sk(sk);
634 struct ccid2_hc_rx_sock *hcrx = ccid2_hc_rx_sk(sk);
635
636 switch (DCCP_SKB_CB(skb)->dccpd_type) {
637 case DCCP_PKT_DATA:
638 case DCCP_PKT_DATAACK:
Gerrit Renker1fb87502008-09-04 07:30:19 +0200639 hcrx->data++;
640 if (hcrx->data >= dp->dccps_r_ack_ratio) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800641 dccp_send_ack(sk);
Gerrit Renker1fb87502008-09-04 07:30:19 +0200642 hcrx->data = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800643 }
644 break;
645 }
646}
647
Arnaldo Carvalho de Melo91f0ebf2006-03-20 19:21:44 -0800648static struct ccid_operations ccid2 = {
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200649 .ccid_id = DCCPC_CCID2,
650 .ccid_name = "TCP-like",
651 .ccid_owner = THIS_MODULE,
652 .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock),
653 .ccid_hc_tx_init = ccid2_hc_tx_init,
654 .ccid_hc_tx_exit = ccid2_hc_tx_exit,
655 .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet,
656 .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent,
657 .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
658 .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv,
659 .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock),
660 .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv,
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800661};
662
Gerrit Renker84116712006-11-20 18:26:03 -0200663#ifdef CONFIG_IP_DCCP_CCID2_DEBUG
Gerrit Renker43264992008-08-23 13:28:27 +0200664module_param(ccid2_debug, bool, 0644);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800665MODULE_PARM_DESC(ccid2_debug, "Enable debug messages");
Gerrit Renker84116712006-11-20 18:26:03 -0200666#endif
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800667
668static __init int ccid2_module_init(void)
669{
670 return ccid_register(&ccid2);
671}
672module_init(ccid2_module_init);
673
674static __exit void ccid2_module_exit(void)
675{
676 ccid_unregister(&ccid2);
677}
678module_exit(ccid2_module_exit);
679
680MODULE_AUTHOR("Andrea Bittau <a.bittau@cs.ucl.ac.uk>");
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800681MODULE_DESCRIPTION("DCCP TCP-Like (CCID2) CCID");
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800682MODULE_LICENSE("GPL");
683MODULE_ALIAS("net-dccp-ccid-2");