blob: 22753fd986988a66c2ed0c2df7b46ef7d7dc5b7b [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_start_rto_timer(struct sock *sk);
114
115static void ccid2_hc_tx_rto_expire(unsigned long data)
116{
117 struct sock *sk = (struct sock *)data;
118 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
Gerrit Renker83337da2008-09-04 07:30:19 +0200119 const bool sender_was_blocked = ccid2_cwnd_network_limited(hctx);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800120
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800121 bh_lock_sock(sk);
122 if (sock_owned_by_user(sk)) {
Gerrit Renker1fb87502008-09-04 07:30:19 +0200123 sk_reset_timer(sk, &hctx->rtotimer, jiffies + HZ / 5);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800124 goto out;
125 }
126
127 ccid2_pr_debug("RTO_EXPIRE\n");
128
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800129 /* back-off timer */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200130 hctx->rto <<= 1;
Gerrit Renker14355622008-09-04 07:30:19 +0200131 if (hctx->rto > DCCP_RTO_MAX)
132 hctx->rto = DCCP_RTO_MAX;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800133
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800134 /* adjust pipe, cwnd etc */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200135 hctx->ssthresh = hctx->cwnd / 2;
136 if (hctx->ssthresh < 2)
137 hctx->ssthresh = 2;
138 hctx->cwnd = 1;
139 hctx->pipe = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800140
141 /* clear state about stuff we sent */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200142 hctx->seqt = hctx->seqh;
143 hctx->packets_acked = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800144
145 /* clear ack ratio state. */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200146 hctx->rpseq = 0;
147 hctx->rpdupack = -1;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800148 ccid2_change_l_ack_ratio(sk, 1);
Gerrit Renker83337da2008-09-04 07:30:19 +0200149
150 /* if we were blocked before, we may now send cwnd=1 packet */
151 if (sender_was_blocked)
152 tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
153 ccid2_start_rto_timer(sk);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800154out:
155 bh_unlock_sock(sk);
Andrea Bittau77ff72d2006-03-20 17:57:52 -0800156 sock_put(sk);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800157}
158
159static void ccid2_start_rto_timer(struct sock *sk)
160{
161 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
162
Gerrit Renker1fb87502008-09-04 07:30:19 +0200163 ccid2_pr_debug("setting RTO timeout=%ld\n", hctx->rto);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800164
Gerrit Renker1fb87502008-09-04 07:30:19 +0200165 BUG_ON(timer_pending(&hctx->rtotimer));
166 sk_reset_timer(sk, &hctx->rtotimer,
167 jiffies + hctx->rto);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800168}
169
Gerrit Renkerc506d912008-09-04 07:30:19 +0200170static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800171{
172 struct dccp_sock *dp = dccp_sk(sk);
173 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700174 struct ccid2_seq *next;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800175
Gerrit Renker1fb87502008-09-04 07:30:19 +0200176 hctx->pipe++;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800177
Gerrit Renker1fb87502008-09-04 07:30:19 +0200178 hctx->seqh->ccid2s_seq = dp->dccps_gss;
179 hctx->seqh->ccid2s_acked = 0;
180 hctx->seqh->ccid2s_sent = jiffies;
Andrea Bittau07978aa2006-09-19 13:13:37 -0700181
Gerrit Renker1fb87502008-09-04 07:30:19 +0200182 next = hctx->seqh->ccid2s_next;
Andrea Bittau07978aa2006-09-19 13:13:37 -0700183 /* check if we need to alloc more space */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200184 if (next == hctx->seqt) {
Gerrit Renker7d9e8932007-10-04 14:41:26 -0700185 if (ccid2_hc_tx_alloc_seq(hctx)) {
186 DCCP_CRIT("packet history - out of memory!");
187 /* FIXME: find a more graceful way to bail out */
188 return;
189 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200190 next = hctx->seqh->ccid2s_next;
191 BUG_ON(next == hctx->seqt);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700192 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200193 hctx->seqh = next;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800194
Gerrit Renker1fb87502008-09-04 07:30:19 +0200195 ccid2_pr_debug("cwnd=%d pipe=%d\n", hctx->cwnd, hctx->pipe);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800196
Gerrit Renker900bfed2007-11-24 21:58:33 -0200197 /*
198 * FIXME: The code below is broken and the variables have been removed
199 * from the socket struct. The `ackloss' variable was always set to 0,
200 * and with arsent there are several problems:
201 * (i) it doesn't just count the number of Acks, but all sent packets;
202 * (ii) it is expressed in # of packets, not # of windows, so the
203 * comparison below uses the wrong formula: Appendix A of RFC 4341
204 * comes up with the number K = cwnd / (R^2 - R) of consecutive windows
205 * of data with no lost or marked Ack packets. If arsent were the # of
206 * consecutive Acks received without loss, then Ack Ratio needs to be
207 * decreased by 1 when
208 * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2)
209 * where cwnd / R is the number of Acks received per window of data
210 * (cf. RFC 4341, App. A). The problems are that
211 * - arsent counts other packets as well;
212 * - the comparison uses a formula different from RFC 4341;
213 * - computing a cubic/quadratic equation each time is too complicated.
214 * Hence a different algorithm is needed.
215 */
216#if 0
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800217 /* Ack Ratio. Need to maintain a concept of how many windows we sent */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200218 hctx->arsent++;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800219 /* We had an ack loss in this window... */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200220 if (hctx->ackloss) {
221 if (hctx->arsent >= hctx->cwnd) {
222 hctx->arsent = 0;
223 hctx->ackloss = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800224 }
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800225 } else {
226 /* No acks lost up to now... */
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800227 /* decrease ack ratio if enough packets were sent */
228 if (dp->dccps_l_ack_ratio > 1) {
229 /* XXX don't calculate denominator each time */
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800230 int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
231 dp->dccps_l_ack_ratio;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800232
Gerrit Renker1fb87502008-09-04 07:30:19 +0200233 denom = hctx->cwnd * hctx->cwnd / denom;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800234
Gerrit Renker1fb87502008-09-04 07:30:19 +0200235 if (hctx->arsent >= denom) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800236 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
Gerrit Renker1fb87502008-09-04 07:30:19 +0200237 hctx->arsent = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800238 }
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800239 } else {
240 /* we can't increase ack ratio further [1] */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200241 hctx->arsent = 0; /* or maybe set it to cwnd*/
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800242 }
243 }
Gerrit Renker900bfed2007-11-24 21:58:33 -0200244#endif
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800245
246 /* setup RTO timer */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200247 if (!timer_pending(&hctx->rtotimer))
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800248 ccid2_start_rto_timer(sk);
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800249
Andrea Bittau8d424f62006-09-19 13:12:44 -0700250#ifdef CONFIG_IP_DCCP_CCID2_DEBUG
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800251 do {
Gerrit Renker1fb87502008-09-04 07:30:19 +0200252 struct ccid2_seq *seqp = hctx->seqt;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800253
Gerrit Renker1fb87502008-09-04 07:30:19 +0200254 while (seqp != hctx->seqh) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800255 ccid2_pr_debug("out seq=%llu acked=%d time=%lu\n",
Arnaldo Carvalho de Melo8109b022006-12-10 16:01:18 -0200256 (unsigned long long)seqp->ccid2s_seq,
Randy Dunlap234af482006-10-29 16:03:30 -0800257 seqp->ccid2s_acked, seqp->ccid2s_sent);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800258 seqp = seqp->ccid2s_next;
259 }
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800260 } while (0);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800261 ccid2_pr_debug("=========\n");
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800262#endif
263}
264
Andrea Bittau77ff72d2006-03-20 17:57:52 -0800265static void ccid2_hc_tx_kill_rto_timer(struct sock *sk)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800266{
Andrea Bittau77ff72d2006-03-20 17:57:52 -0800267 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
268
Gerrit Renker1fb87502008-09-04 07:30:19 +0200269 sk_stop_timer(sk, &hctx->rtotimer);
Andrea Bittau77ff72d2006-03-20 17:57:52 -0800270 ccid2_pr_debug("deleted RTO timer\n");
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800271}
272
Gerrit Renker14355622008-09-04 07:30:19 +0200273/**
274 * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
275 * This code is almost identical with TCP's tcp_rtt_estimator(), since
276 * - it has a higher sampling frequency (recommended by RFC 1323),
277 * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
278 * - it is simple (cf. more complex proposals such as Eifel timer or research
279 * which suggests that the gain should be set according to window size),
280 * - in tests it was found to work well with CCID2 [gerrit].
281 */
282static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
283{
284 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
285 long m = mrtt ? : 1;
286
287 if (hctx->srtt == 0) {
288 /* First measurement m */
289 hctx->srtt = m << 3;
290 hctx->mdev = m << 1;
291
292 hctx->mdev_max = max(TCP_RTO_MIN, hctx->mdev);
293 hctx->rttvar = hctx->mdev_max;
294 hctx->rtt_seq = dccp_sk(sk)->dccps_gss;
295 } else {
296 /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
297 m -= (hctx->srtt >> 3);
298 hctx->srtt += m;
299
300 /* Similarly, update scaled mdev with regard to |m| */
301 if (m < 0) {
302 m = -m;
303 m -= (hctx->mdev >> 2);
304 /*
305 * This neutralises RTO increase when RTT < SRTT - mdev
306 * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
307 * in Linux TCP", USENIX 2002, pp. 49-62).
308 */
309 if (m > 0)
310 m >>= 3;
311 } else {
312 m -= (hctx->mdev >> 2);
313 }
314 hctx->mdev += m;
315
316 if (hctx->mdev > hctx->mdev_max) {
317 hctx->mdev_max = hctx->mdev;
318 if (hctx->mdev_max > hctx->rttvar)
319 hctx->rttvar = hctx->mdev_max;
320 }
321
322 /*
323 * Decay RTTVAR at most once per flight, exploiting that
324 * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2)
325 * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1)
326 * GAR is a useful bound for FlightSize = pipe, AWL is probably
327 * too low as it over-estimates pipe.
328 */
329 if (after48(dccp_sk(sk)->dccps_gar, hctx->rtt_seq)) {
330 if (hctx->mdev_max < hctx->rttvar)
331 hctx->rttvar -= (hctx->rttvar -
332 hctx->mdev_max) >> 2;
333 hctx->rtt_seq = dccp_sk(sk)->dccps_gss;
334 hctx->mdev_max = TCP_RTO_MIN;
335 }
336 }
337
338 /*
339 * Set RTO from SRTT and RTTVAR
340 * Clock granularity is ignored since the minimum error for RTTVAR is
341 * clamped to 50msec (corresponding to HZ=20). This leads to a minimum
342 * RTO of 200msec. This agrees with TCP and RFC 4341, 5.: "Because DCCP
343 * does not retransmit data, DCCP does not require TCP's recommended
344 * minimum timeout of one second".
345 */
346 hctx->rto = (hctx->srtt >> 3) + hctx->rttvar;
347
348 if (hctx->rto > DCCP_RTO_MAX)
349 hctx->rto = DCCP_RTO_MAX;
350}
351
352static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
353 unsigned int *maxincr)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800354{
355 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
356
Gerrit Renker1fb87502008-09-04 07:30:19 +0200357 if (hctx->cwnd < hctx->ssthresh) {
358 if (*maxincr > 0 && ++hctx->packets_acked == 2) {
359 hctx->cwnd += 1;
360 *maxincr -= 1;
361 hctx->packets_acked = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800362 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200363 } else if (++hctx->packets_acked >= hctx->cwnd) {
364 hctx->cwnd += 1;
365 hctx->packets_acked = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800366 }
Gerrit Renker14355622008-09-04 07:30:19 +0200367 /*
368 * FIXME: RTT is sampled several times per acknowledgment (for each
369 * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
370 * This causes the RTT to be over-estimated, since the older entries
371 * in the Ack Vector have earlier sending times.
372 * The cleanest solution is to not use the ccid2s_sent field at all
373 * and instead use DCCP timestamps - need to be resolved at some time.
374 */
375 ccid2_rtt_estimator(sk, jiffies - seqp->ccid2s_sent);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800376}
377
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200378static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
Andrea Bittau374bcf32006-09-19 13:14:43 -0700379{
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200380 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
381
Gerrit Renker1fb87502008-09-04 07:30:19 +0200382 if (time_before(seqp->ccid2s_sent, hctx->last_cong)) {
Andrea Bittau374bcf32006-09-19 13:14:43 -0700383 ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
384 return;
385 }
386
Gerrit Renker1fb87502008-09-04 07:30:19 +0200387 hctx->last_cong = jiffies;
Andrea Bittau374bcf32006-09-19 13:14:43 -0700388
Gerrit Renker1fb87502008-09-04 07:30:19 +0200389 hctx->cwnd = hctx->cwnd / 2 ? : 1U;
390 hctx->ssthresh = max(hctx->cwnd, 2U);
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200391
392 /* Avoid spurious timeouts resulting from Ack Ratio > cwnd */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200393 if (dccp_sk(sk)->dccps_l_ack_ratio > hctx->cwnd)
394 ccid2_change_l_ack_ratio(sk, hctx->cwnd);
Andrea Bittau374bcf32006-09-19 13:14:43 -0700395}
396
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200397static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
398 u8 option, u8 *optval, u8 optlen)
399{
400 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
401
402 switch (option) {
403 case DCCPO_ACK_VECTOR_0:
404 case DCCPO_ACK_VECTOR_1:
405 return dccp_ackvec_parsed_add(&hctx->av_chunks, optval, optlen,
406 option - DCCPO_ACK_VECTOR_0);
407 }
408 return 0;
409}
410
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800411static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
412{
413 struct dccp_sock *dp = dccp_sk(sk);
414 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
Gerrit Renker83337da2008-09-04 07:30:19 +0200415 const bool sender_was_blocked = ccid2_cwnd_network_limited(hctx);
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200416 struct dccp_ackvec_parsed *avp;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800417 u64 ackno, seqno;
418 struct ccid2_seq *seqp;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800419 int done = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800420 unsigned int maxincr = 0;
421
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800422 /* check reverse path congestion */
423 seqno = DCCP_SKB_CB(skb)->dccpd_seq;
424
425 /* XXX this whole "algorithm" is broken. Need to fix it to keep track
426 * of the seqnos of the dupacks so that rpseq and rpdupack are correct
427 * -sorbo.
428 */
429 /* need to bootstrap */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200430 if (hctx->rpdupack == -1) {
431 hctx->rpdupack = 0;
432 hctx->rpseq = seqno;
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800433 } else {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800434 /* check if packet is consecutive */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200435 if (dccp_delta_seqno(hctx->rpseq, seqno) == 1)
436 hctx->rpseq = seqno;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800437 /* it's a later packet */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200438 else if (after48(seqno, hctx->rpseq)) {
439 hctx->rpdupack++;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800440
441 /* check if we got enough dupacks */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200442 if (hctx->rpdupack >= NUMDUPACK) {
443 hctx->rpdupack = -1; /* XXX lame */
444 hctx->rpseq = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800445
Gerrit Renkerdf054e12007-11-24 21:32:53 -0200446 ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800447 }
448 }
449 }
450
451 /* check forward path congestion */
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200452 if (dccp_packet_without_ack(skb))
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800453 return;
454
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200455 /* still didn't send out new data packets */
456 if (hctx->seqh == hctx->seqt)
457 goto done;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800458
459 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
Gerrit Renker1fb87502008-09-04 07:30:19 +0200460 if (after48(ackno, hctx->high_ack))
461 hctx->high_ack = ackno;
Andrea Bittau32aac182006-11-16 14:28:40 -0200462
Gerrit Renker1fb87502008-09-04 07:30:19 +0200463 seqp = hctx->seqt;
Andrea Bittau32aac182006-11-16 14:28:40 -0200464 while (before48(seqp->ccid2s_seq, ackno)) {
465 seqp = seqp->ccid2s_next;
Gerrit Renker1fb87502008-09-04 07:30:19 +0200466 if (seqp == hctx->seqh) {
467 seqp = hctx->seqh->ccid2s_prev;
Andrea Bittau32aac182006-11-16 14:28:40 -0200468 break;
469 }
470 }
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800471
Gerrit Renkera3020022007-11-24 22:10:29 -0200472 /*
473 * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
474 * packets per acknowledgement. Rounding up avoids that cwnd is not
475 * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800476 */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200477 if (hctx->cwnd < hctx->ssthresh)
Gerrit Renkera3020022007-11-24 22:10:29 -0200478 maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800479
480 /* go through all ack vectors */
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200481 list_for_each_entry(avp, &hctx->av_chunks, node) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800482 /* go through this ack vector */
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200483 for (; avp->len--; avp->vec++) {
484 u64 ackno_end_rl = SUB48(ackno,
485 dccp_ackvec_runlen(avp->vec));
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800486
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200487 ccid2_pr_debug("ackvec %llu |%u,%u|\n",
Randy Dunlap234af482006-10-29 16:03:30 -0800488 (unsigned long long)ackno,
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200489 dccp_ackvec_state(avp->vec) >> 6,
490 dccp_ackvec_runlen(avp->vec));
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800491 /* if the seqno we are analyzing is larger than the
492 * current ackno, then move towards the tail of our
493 * seqnos.
494 */
495 while (after48(seqp->ccid2s_seq, ackno)) {
Gerrit Renker1fb87502008-09-04 07:30:19 +0200496 if (seqp == hctx->seqt) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800497 done = 1;
498 break;
499 }
500 seqp = seqp->ccid2s_prev;
501 }
502 if (done)
503 break;
504
505 /* check all seqnos in the range of the vector
506 * run length
507 */
508 while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200509 const u8 state = dccp_ackvec_state(avp->vec);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800510
511 /* new packet received or marked */
Gerrit Renkerff49e272008-09-04 07:30:19 +0200512 if (state != DCCPAV_NOT_RECEIVED &&
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800513 !seqp->ccid2s_acked) {
Gerrit Renkerff49e272008-09-04 07:30:19 +0200514 if (state == DCCPAV_ECN_MARKED)
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200515 ccid2_congestion_event(sk,
Andrea Bittau374bcf32006-09-19 13:14:43 -0700516 seqp);
Gerrit Renkerff49e272008-09-04 07:30:19 +0200517 else
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800518 ccid2_new_ack(sk, seqp,
519 &maxincr);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800520
521 seqp->ccid2s_acked = 1;
522 ccid2_pr_debug("Got ack for %llu\n",
Randy Dunlap234af482006-10-29 16:03:30 -0800523 (unsigned long long)seqp->ccid2s_seq);
Gerrit Renkere9803c02008-09-04 07:30:19 +0200524 hctx->pipe--;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800525 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200526 if (seqp == hctx->seqt) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800527 done = 1;
528 break;
529 }
Gerrit Renker3de54892007-11-24 20:37:48 -0200530 seqp = seqp->ccid2s_prev;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800531 }
532 if (done)
533 break;
534
Gerrit Renkercfbbeab2007-11-24 20:43:59 -0200535 ackno = SUB48(ackno_end_rl, 1);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800536 }
537 if (done)
538 break;
539 }
540
541 /* The state about what is acked should be correct now
542 * Check for NUMDUPACK
543 */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200544 seqp = hctx->seqt;
545 while (before48(seqp->ccid2s_seq, hctx->high_ack)) {
Andrea Bittau32aac182006-11-16 14:28:40 -0200546 seqp = seqp->ccid2s_next;
Gerrit Renker1fb87502008-09-04 07:30:19 +0200547 if (seqp == hctx->seqh) {
548 seqp = hctx->seqh->ccid2s_prev;
Andrea Bittau32aac182006-11-16 14:28:40 -0200549 break;
550 }
551 }
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800552 done = 0;
553 while (1) {
554 if (seqp->ccid2s_acked) {
555 done++;
Gerrit Renker63df18a2007-11-24 22:04:35 -0200556 if (done == NUMDUPACK)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800557 break;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800558 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200559 if (seqp == hctx->seqt)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800560 break;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800561 seqp = seqp->ccid2s_prev;
562 }
563
564 /* If there are at least 3 acknowledgements, anything unacknowledged
565 * below the last sequence number is considered lost
566 */
Gerrit Renker63df18a2007-11-24 22:04:35 -0200567 if (done == NUMDUPACK) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800568 struct ccid2_seq *last_acked = seqp;
569
570 /* check for lost packets */
571 while (1) {
572 if (!seqp->ccid2s_acked) {
Andrea Bittau374bcf32006-09-19 13:14:43 -0700573 ccid2_pr_debug("Packet lost: %llu\n",
Randy Dunlap234af482006-10-29 16:03:30 -0800574 (unsigned long long)seqp->ccid2s_seq);
Andrea Bittau374bcf32006-09-19 13:14:43 -0700575 /* XXX need to traverse from tail -> head in
576 * order to detect multiple congestion events in
577 * one ack vector.
578 */
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200579 ccid2_congestion_event(sk, seqp);
Gerrit Renkere9803c02008-09-04 07:30:19 +0200580 hctx->pipe--;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800581 }
Gerrit Renker1fb87502008-09-04 07:30:19 +0200582 if (seqp == hctx->seqt)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800583 break;
584 seqp = seqp->ccid2s_prev;
585 }
586
Gerrit Renker1fb87502008-09-04 07:30:19 +0200587 hctx->seqt = last_acked;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800588 }
589
590 /* trim acked packets in tail */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200591 while (hctx->seqt != hctx->seqh) {
592 if (!hctx->seqt->ccid2s_acked)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800593 break;
594
Gerrit Renker1fb87502008-09-04 07:30:19 +0200595 hctx->seqt = hctx->seqt->ccid2s_next;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800596 }
597
Gerrit Renkere9803c02008-09-04 07:30:19 +0200598 /* restart RTO timer if not all outstanding data has been acked */
599 if (hctx->pipe == 0)
600 sk_stop_timer(sk, &hctx->rtotimer);
601 else
Gerrit Renker14355622008-09-04 07:30:19 +0200602 sk_reset_timer(sk, &hctx->rtotimer, jiffies + hctx->rto);
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200603done:
Gerrit Renker83337da2008-09-04 07:30:19 +0200604 /* check if incoming Acks allow pending packets to be sent */
605 if (sender_was_blocked && !ccid2_cwnd_network_limited(hctx))
606 tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200607 dccp_ackvec_parsed_cleanup(&hctx->av_chunks);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800608}
609
Arnaldo Carvalho de Melo91f0ebf2006-03-20 19:21:44 -0800610static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800611{
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +0900612 struct ccid2_hc_tx_sock *hctx = ccid_priv(ccid);
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200613 struct dccp_sock *dp = dccp_sk(sk);
614 u32 max_ratio;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800615
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200616 /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200617 hctx->ssthresh = ~0U;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800618
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200619 /*
620 * RFC 4341, 5: "The cwnd parameter is initialized to at most four
621 * packets for new connections, following the rules from [RFC3390]".
622 * We need to convert the bytes of RFC3390 into the packets of RFC 4341.
623 */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200624 hctx->cwnd = clamp(4380U / dp->dccps_mss_cache, 2U, 4U);
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200625
626 /* Make sure that Ack Ratio is enabled and within bounds. */
Gerrit Renker1fb87502008-09-04 07:30:19 +0200627 max_ratio = DIV_ROUND_UP(hctx->cwnd, 2);
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200628 if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
629 dp->dccps_l_ack_ratio = max_ratio;
630
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800631 /* XXX init ~ to window size... */
Gerrit Renkercd1f7d32007-10-04 14:41:00 -0700632 if (ccid2_hc_tx_alloc_seq(hctx))
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800633 return -ENOMEM;
Arnaldo Carvalho de Melo91f0ebf2006-03-20 19:21:44 -0800634
Gerrit Renker14355622008-09-04 07:30:19 +0200635 hctx->rto = DCCP_TIMEOUT_INIT;
Gerrit Renker1fb87502008-09-04 07:30:19 +0200636 hctx->rpdupack = -1;
637 hctx->last_cong = jiffies;
638 setup_timer(&hctx->rtotimer, ccid2_hc_tx_rto_expire, (unsigned long)sk);
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200639 INIT_LIST_HEAD(&hctx->av_chunks);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800640 return 0;
641}
642
643static void ccid2_hc_tx_exit(struct sock *sk)
644{
YOSHIFUJI Hideakic9eaf172007-02-09 23:24:38 +0900645 struct ccid2_hc_tx_sock *hctx = ccid2_hc_tx_sk(sk);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700646 int i;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800647
Andrea Bittau77ff72d2006-03-20 17:57:52 -0800648 ccid2_hc_tx_kill_rto_timer(sk);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700649
Gerrit Renker1fb87502008-09-04 07:30:19 +0200650 for (i = 0; i < hctx->seqbufc; i++)
651 kfree(hctx->seqbuf[i]);
652 hctx->seqbufc = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800653}
654
655static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
656{
657 const struct dccp_sock *dp = dccp_sk(sk);
658 struct ccid2_hc_rx_sock *hcrx = ccid2_hc_rx_sk(sk);
659
660 switch (DCCP_SKB_CB(skb)->dccpd_type) {
661 case DCCP_PKT_DATA:
662 case DCCP_PKT_DATAACK:
Gerrit Renker1fb87502008-09-04 07:30:19 +0200663 hcrx->data++;
664 if (hcrx->data >= dp->dccps_r_ack_ratio) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800665 dccp_send_ack(sk);
Gerrit Renker1fb87502008-09-04 07:30:19 +0200666 hcrx->data = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800667 }
668 break;
669 }
670}
671
Arnaldo Carvalho de Melo91f0ebf2006-03-20 19:21:44 -0800672static struct ccid_operations ccid2 = {
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200673 .ccid_id = DCCPC_CCID2,
674 .ccid_name = "TCP-like",
675 .ccid_owner = THIS_MODULE,
676 .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock),
677 .ccid_hc_tx_init = ccid2_hc_tx_init,
678 .ccid_hc_tx_exit = ccid2_hc_tx_exit,
679 .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet,
680 .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent,
681 .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
682 .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv,
683 .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock),
684 .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv,
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800685};
686
Gerrit Renker84116712006-11-20 18:26:03 -0200687#ifdef CONFIG_IP_DCCP_CCID2_DEBUG
Gerrit Renker43264992008-08-23 13:28:27 +0200688module_param(ccid2_debug, bool, 0644);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800689MODULE_PARM_DESC(ccid2_debug, "Enable debug messages");
Gerrit Renker84116712006-11-20 18:26:03 -0200690#endif
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800691
692static __init int ccid2_module_init(void)
693{
694 return ccid_register(&ccid2);
695}
696module_init(ccid2_module_init);
697
698static __exit void ccid2_module_exit(void)
699{
700 ccid_unregister(&ccid2);
701}
702module_exit(ccid2_module_exit);
703
704MODULE_AUTHOR("Andrea Bittau <a.bittau@cs.ucl.ac.uk>");
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800705MODULE_DESCRIPTION("DCCP TCP-Like (CCID2) CCID");
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800706MODULE_LICENSE("GPL");
707MODULE_ALIAS("net-dccp-ccid-2");