blob: e96d5e810039a5bcdf3a0503534e7d5ebf596047 [file] [log] [blame]
Andrea Bittau2a91aa32006-03-20 17:41:47 -08001/*
Andrea Bittau2a91aa32006-03-20 17:41:47 -08002 * Copyright (c) 2005, 2006 Andrea Bittau <a.bittau@cs.ucl.ac.uk>
3 *
4 * Changes to meet Linux coding standards, and DCCP infrastructure fixes.
5 *
6 * Copyright (c) 2006 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 */
22
23/*
Gerrit Renker0e64e942006-10-24 16:17:51 -070024 * This implementation should follow RFC 4341
Andrea Bittau2a91aa32006-03-20 17:41:47 -080025 */
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090026#include <linux/slab.h>
Gerrit Renkere8ef9672008-11-12 00:43:40 -080027#include "../feat.h"
Andrea Bittau2a91aa32006-03-20 17:41:47 -080028#include "ccid2.h"
29
Gerrit Renker84116712006-11-20 18:26:03 -020030
31#ifdef CONFIG_IP_DCCP_CCID2_DEBUG
Andrea Bittau2a91aa32006-03-20 17:41:47 -080032static int ccid2_debug;
Gerrit Renker84116712006-11-20 18:26:03 -020033#define ccid2_pr_debug(format, a...) DCCP_PR_DEBUG(ccid2_debug, format, ##a)
Andrea Bittau2a91aa32006-03-20 17:41:47 -080034#else
Gerrit Renker84116712006-11-20 18:26:03 -020035#define ccid2_pr_debug(format, a...)
Andrea Bittau2a91aa32006-03-20 17:41:47 -080036#endif
37
Gerrit Renker77d2dd92009-10-05 00:53:12 +000038static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hc)
Andrea Bittau07978aa2006-09-19 13:13:37 -070039{
40 struct ccid2_seq *seqp;
41 int i;
42
43 /* check if we have space to preserve the pointer to the buffer */
Gerrit Renker77d2dd92009-10-05 00:53:12 +000044 if (hc->tx_seqbufc >= (sizeof(hc->tx_seqbuf) /
45 sizeof(struct ccid2_seq *)))
Andrea Bittau07978aa2006-09-19 13:13:37 -070046 return -ENOMEM;
47
48 /* allocate buffer and initialize linked list */
Gerrit Renkercd1f7d32007-10-04 14:41:00 -070049 seqp = kmalloc(CCID2_SEQBUF_LEN * sizeof(struct ccid2_seq), gfp_any());
Andrea Bittau07978aa2006-09-19 13:13:37 -070050 if (seqp == NULL)
51 return -ENOMEM;
52
Gerrit Renkercd1f7d32007-10-04 14:41:00 -070053 for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) {
Andrea Bittau07978aa2006-09-19 13:13:37 -070054 seqp[i].ccid2s_next = &seqp[i + 1];
55 seqp[i + 1].ccid2s_prev = &seqp[i];
56 }
Gerrit Renkercd1f7d32007-10-04 14:41:00 -070057 seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp;
58 seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
Andrea Bittau07978aa2006-09-19 13:13:37 -070059
60 /* This is the first allocation. Initiate the head and tail. */
Gerrit Renker77d2dd92009-10-05 00:53:12 +000061 if (hc->tx_seqbufc == 0)
62 hc->tx_seqh = hc->tx_seqt = seqp;
Andrea Bittau07978aa2006-09-19 13:13:37 -070063 else {
64 /* link the existing list with the one we just created */
Gerrit Renker77d2dd92009-10-05 00:53:12 +000065 hc->tx_seqh->ccid2s_next = seqp;
66 seqp->ccid2s_prev = hc->tx_seqh;
Andrea Bittau07978aa2006-09-19 13:13:37 -070067
Gerrit Renker77d2dd92009-10-05 00:53:12 +000068 hc->tx_seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
69 seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hc->tx_seqt;
Andrea Bittau07978aa2006-09-19 13:13:37 -070070 }
71
72 /* store the original pointer to the buffer so we can free it */
Gerrit Renker77d2dd92009-10-05 00:53:12 +000073 hc->tx_seqbuf[hc->tx_seqbufc] = seqp;
74 hc->tx_seqbufc++;
Andrea Bittau07978aa2006-09-19 13:13:37 -070075
76 return 0;
77}
78
Gerrit Renker6b57c932006-11-28 19:55:06 -020079static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
Andrea Bittau2a91aa32006-03-20 17:41:47 -080080{
Gerrit Renker1c0e0a02010-10-27 19:16:28 +000081 if (ccid2_cwnd_network_limited(ccid2_hc_tx_sk(sk)))
82 return CCID_PACKET_WILL_DEQUEUE_LATER;
83 return CCID_PACKET_SEND_AT_ONCE;
Andrea Bittau2a91aa32006-03-20 17:41:47 -080084}
85
Gerrit Renkerdf054e12007-11-24 21:32:53 -020086static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
Andrea Bittau2a91aa32006-03-20 17:41:47 -080087{
88 struct dccp_sock *dp = dccp_sk(sk);
Gerrit Renkerb1c00fe2009-10-05 00:53:10 +000089 u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->tx_cwnd, 2);
Gerrit Renkerd50ad162007-11-24 21:40:24 -020090
Andrea Bittau2a91aa32006-03-20 17:41:47 -080091 /*
Gerrit Renkerd50ad162007-11-24 21:40:24 -020092 * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from
93 * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always
94 * acceptable since this causes starvation/deadlock whenever cwnd < 2.
95 * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled).
Andrea Bittau2a91aa32006-03-20 17:41:47 -080096 */
Gerrit Renkerd50ad162007-11-24 21:40:24 -020097 if (val == 0 || val > max_ratio) {
98 DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio);
99 val = max_ratio;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800100 }
Gerrit Renkere8ef9672008-11-12 00:43:40 -0800101 if (val > DCCPF_ACK_RATIO_MAX)
102 val = DCCPF_ACK_RATIO_MAX;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800103
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200104 if (val == dp->dccps_l_ack_ratio)
105 return;
106
Gerrit Renkerdf054e12007-11-24 21:32:53 -0200107 ccid2_pr_debug("changing local ack ratio to %u\n", val);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800108 dp->dccps_l_ack_ratio = val;
109}
110
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800111static void ccid2_hc_tx_rto_expire(unsigned long data)
112{
113 struct sock *sk = (struct sock *)data;
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000114 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
Gerrit Renker1c0e0a02010-10-27 19:16:28 +0000115 const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800116
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800117 bh_lock_sock(sk);
118 if (sock_owned_by_user(sk)) {
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000119 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800120 goto out;
121 }
122
123 ccid2_pr_debug("RTO_EXPIRE\n");
124
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800125 /* back-off timer */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000126 hc->tx_rto <<= 1;
Gerrit Renker231cc2a2010-08-22 19:41:40 +0000127 if (hc->tx_rto > DCCP_RTO_MAX)
128 hc->tx_rto = DCCP_RTO_MAX;
Gerrit Renker410e27a2008-09-09 13:27:22 +0200129
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800130 /* adjust pipe, cwnd etc */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000131 hc->tx_ssthresh = hc->tx_cwnd / 2;
132 if (hc->tx_ssthresh < 2)
133 hc->tx_ssthresh = 2;
Gerrit Renker67b67e32010-08-22 19:41:36 +0000134 hc->tx_cwnd = 1;
135 hc->tx_pipe = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800136
137 /* clear state about stuff we sent */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000138 hc->tx_seqt = hc->tx_seqh;
139 hc->tx_packets_acked = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800140
141 /* clear ack ratio state. */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000142 hc->tx_rpseq = 0;
143 hc->tx_rpdupack = -1;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800144 ccid2_change_l_ack_ratio(sk, 1);
Gerrit Renker1c0e0a02010-10-27 19:16:28 +0000145
146 /* if we were blocked before, we may now send cwnd=1 packet */
147 if (sender_was_blocked)
148 tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
149 /* restart backed-off timer */
150 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800151out:
152 bh_unlock_sock(sk);
Andrea Bittau77ff72d2006-03-20 17:57:52 -0800153 sock_put(sk);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800154}
155
Gerrit Renkerbaf9e782010-10-11 20:37:38 +0200156static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800157{
158 struct dccp_sock *dp = dccp_sk(sk);
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000159 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700160 struct ccid2_seq *next;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800161
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000162 hc->tx_pipe++;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800163
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000164 hc->tx_seqh->ccid2s_seq = dp->dccps_gss;
165 hc->tx_seqh->ccid2s_acked = 0;
Gerrit Renkerd82b6f82010-08-29 19:23:10 +0000166 hc->tx_seqh->ccid2s_sent = ccid2_time_stamp;
Andrea Bittau07978aa2006-09-19 13:13:37 -0700167
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000168 next = hc->tx_seqh->ccid2s_next;
Andrea Bittau07978aa2006-09-19 13:13:37 -0700169 /* check if we need to alloc more space */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000170 if (next == hc->tx_seqt) {
171 if (ccid2_hc_tx_alloc_seq(hc)) {
Gerrit Renker7d9e8932007-10-04 14:41:26 -0700172 DCCP_CRIT("packet history - out of memory!");
173 /* FIXME: find a more graceful way to bail out */
174 return;
175 }
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000176 next = hc->tx_seqh->ccid2s_next;
177 BUG_ON(next == hc->tx_seqt);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700178 }
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000179 hc->tx_seqh = next;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800180
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000181 ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800182
Gerrit Renker900bfed2007-11-24 21:58:33 -0200183 /*
184 * FIXME: The code below is broken and the variables have been removed
185 * from the socket struct. The `ackloss' variable was always set to 0,
186 * and with arsent there are several problems:
187 * (i) it doesn't just count the number of Acks, but all sent packets;
188 * (ii) it is expressed in # of packets, not # of windows, so the
189 * comparison below uses the wrong formula: Appendix A of RFC 4341
190 * comes up with the number K = cwnd / (R^2 - R) of consecutive windows
191 * of data with no lost or marked Ack packets. If arsent were the # of
192 * consecutive Acks received without loss, then Ack Ratio needs to be
193 * decreased by 1 when
194 * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2)
195 * where cwnd / R is the number of Acks received per window of data
196 * (cf. RFC 4341, App. A). The problems are that
197 * - arsent counts other packets as well;
198 * - the comparison uses a formula different from RFC 4341;
199 * - computing a cubic/quadratic equation each time is too complicated.
200 * Hence a different algorithm is needed.
201 */
202#if 0
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800203 /* Ack Ratio. Need to maintain a concept of how many windows we sent */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000204 hc->tx_arsent++;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800205 /* We had an ack loss in this window... */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000206 if (hc->tx_ackloss) {
207 if (hc->tx_arsent >= hc->tx_cwnd) {
208 hc->tx_arsent = 0;
209 hc->tx_ackloss = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800210 }
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800211 } else {
212 /* No acks lost up to now... */
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800213 /* decrease ack ratio if enough packets were sent */
214 if (dp->dccps_l_ack_ratio > 1) {
215 /* XXX don't calculate denominator each time */
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800216 int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
217 dp->dccps_l_ack_ratio;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800218
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000219 denom = hc->tx_cwnd * hc->tx_cwnd / denom;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800220
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000221 if (hc->tx_arsent >= denom) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800222 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000223 hc->tx_arsent = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800224 }
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800225 } else {
226 /* we can't increase ack ratio further [1] */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000227 hc->tx_arsent = 0; /* or maybe set it to cwnd*/
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800228 }
229 }
Gerrit Renker900bfed2007-11-24 21:58:33 -0200230#endif
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800231
Gerrit Renkerd26eeb02010-08-29 19:23:11 +0000232 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800233
Andrea Bittau8d424f62006-09-19 13:12:44 -0700234#ifdef CONFIG_IP_DCCP_CCID2_DEBUG
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800235 do {
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000236 struct ccid2_seq *seqp = hc->tx_seqt;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800237
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000238 while (seqp != hc->tx_seqh) {
Gerrit Renkerd82b6f82010-08-29 19:23:10 +0000239 ccid2_pr_debug("out seq=%llu acked=%d time=%u\n",
Arnaldo Carvalho de Melo8109b022006-12-10 16:01:18 -0200240 (unsigned long long)seqp->ccid2s_seq,
Randy Dunlap234af482006-10-29 16:03:30 -0800241 seqp->ccid2s_acked, seqp->ccid2s_sent);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800242 seqp = seqp->ccid2s_next;
243 }
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800244 } while (0);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800245 ccid2_pr_debug("=========\n");
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800246#endif
247}
248
Gerrit Renker231cc2a2010-08-22 19:41:40 +0000249/**
250 * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
251 * This code is almost identical with TCP's tcp_rtt_estimator(), since
252 * - it has a higher sampling frequency (recommended by RFC 1323),
253 * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
254 * - it is simple (cf. more complex proposals such as Eifel timer or research
255 * which suggests that the gain should be set according to window size),
256 * - in tests it was found to work well with CCID2 [gerrit].
257 */
258static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
259{
260 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
261 long m = mrtt ? : 1;
262
263 if (hc->tx_srtt == 0) {
264 /* First measurement m */
265 hc->tx_srtt = m << 3;
266 hc->tx_mdev = m << 1;
267
Gerrit Renker4886fca2010-08-29 19:23:13 +0000268 hc->tx_mdev_max = max(hc->tx_mdev, tcp_rto_min(sk));
Gerrit Renker231cc2a2010-08-22 19:41:40 +0000269 hc->tx_rttvar = hc->tx_mdev_max;
Gerrit Renker4886fca2010-08-29 19:23:13 +0000270
Gerrit Renker231cc2a2010-08-22 19:41:40 +0000271 hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss;
272 } else {
273 /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
274 m -= (hc->tx_srtt >> 3);
275 hc->tx_srtt += m;
276
277 /* Similarly, update scaled mdev with regard to |m| */
278 if (m < 0) {
279 m = -m;
280 m -= (hc->tx_mdev >> 2);
281 /*
282 * This neutralises RTO increase when RTT < SRTT - mdev
283 * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
284 * in Linux TCP", USENIX 2002, pp. 49-62).
285 */
286 if (m > 0)
287 m >>= 3;
288 } else {
289 m -= (hc->tx_mdev >> 2);
290 }
291 hc->tx_mdev += m;
292
293 if (hc->tx_mdev > hc->tx_mdev_max) {
294 hc->tx_mdev_max = hc->tx_mdev;
295 if (hc->tx_mdev_max > hc->tx_rttvar)
296 hc->tx_rttvar = hc->tx_mdev_max;
297 }
298
299 /*
300 * Decay RTTVAR at most once per flight, exploiting that
301 * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2)
302 * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1)
303 * GAR is a useful bound for FlightSize = pipe.
304 * AWL is probably too low here, as it over-estimates pipe.
305 */
306 if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) {
307 if (hc->tx_mdev_max < hc->tx_rttvar)
308 hc->tx_rttvar -= (hc->tx_rttvar -
309 hc->tx_mdev_max) >> 2;
310 hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss;
Gerrit Renker4886fca2010-08-29 19:23:13 +0000311 hc->tx_mdev_max = tcp_rto_min(sk);
Gerrit Renker231cc2a2010-08-22 19:41:40 +0000312 }
313 }
314
315 /*
316 * Set RTO from SRTT and RTTVAR
317 * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms.
318 * This agrees with RFC 4341, 5:
319 * "Because DCCP does not retransmit data, DCCP does not require
320 * TCP's recommended minimum timeout of one second".
321 */
322 hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar;
323
324 if (hc->tx_rto > DCCP_RTO_MAX)
325 hc->tx_rto = DCCP_RTO_MAX;
326}
327
328static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
329 unsigned int *maxincr)
Gerrit Renker410e27a2008-09-09 13:27:22 +0200330{
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000331 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
Gerrit Renker410e27a2008-09-09 13:27:22 +0200332
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000333 if (hc->tx_cwnd < hc->tx_ssthresh) {
334 if (*maxincr > 0 && ++hc->tx_packets_acked == 2) {
335 hc->tx_cwnd += 1;
336 *maxincr -= 1;
337 hc->tx_packets_acked = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800338 }
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000339 } else if (++hc->tx_packets_acked >= hc->tx_cwnd) {
340 hc->tx_cwnd += 1;
341 hc->tx_packets_acked = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800342 }
Gerrit Renker231cc2a2010-08-22 19:41:40 +0000343 /*
344 * FIXME: RTT is sampled several times per acknowledgment (for each
345 * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
346 * This causes the RTT to be over-estimated, since the older entries
347 * in the Ack Vector have earlier sending times.
348 * The cleanest solution is to not use the ccid2s_sent field at all
349 * and instead use DCCP timestamps: requires changes in other places.
350 */
Gerrit Renkerd82b6f82010-08-29 19:23:10 +0000351 ccid2_rtt_estimator(sk, ccid2_time_stamp - seqp->ccid2s_sent);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800352}
353
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200354static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
Andrea Bittau374bcf32006-09-19 13:14:43 -0700355{
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000356 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200357
Gerrit Renkerd82b6f82010-08-29 19:23:10 +0000358 if ((s32)(seqp->ccid2s_sent - hc->tx_last_cong) < 0) {
Andrea Bittau374bcf32006-09-19 13:14:43 -0700359 ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
360 return;
361 }
362
Gerrit Renkerd82b6f82010-08-29 19:23:10 +0000363 hc->tx_last_cong = ccid2_time_stamp;
Andrea Bittau374bcf32006-09-19 13:14:43 -0700364
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000365 hc->tx_cwnd = hc->tx_cwnd / 2 ? : 1U;
366 hc->tx_ssthresh = max(hc->tx_cwnd, 2U);
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200367
368 /* Avoid spurious timeouts resulting from Ack Ratio > cwnd */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000369 if (dccp_sk(sk)->dccps_l_ack_ratio > hc->tx_cwnd)
370 ccid2_change_l_ack_ratio(sk, hc->tx_cwnd);
Gerrit Renkerc8bf4622008-09-04 07:30:19 +0200371}
372
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100373static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
374 u8 option, u8 *optval, u8 optlen)
375{
376 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
377
378 switch (option) {
379 case DCCPO_ACK_VECTOR_0:
380 case DCCPO_ACK_VECTOR_1:
381 return dccp_ackvec_parsed_add(&hc->tx_av_chunks, optval, optlen,
382 option - DCCPO_ACK_VECTOR_0);
383 }
384 return 0;
385}
386
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800387static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
388{
389 struct dccp_sock *dp = dccp_sk(sk);
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000390 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
Gerrit Renker1c0e0a02010-10-27 19:16:28 +0000391 const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100392 struct dccp_ackvec_parsed *avp;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800393 u64 ackno, seqno;
394 struct ccid2_seq *seqp;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800395 int done = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800396 unsigned int maxincr = 0;
397
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800398 /* check reverse path congestion */
399 seqno = DCCP_SKB_CB(skb)->dccpd_seq;
400
401 /* XXX this whole "algorithm" is broken. Need to fix it to keep track
402 * of the seqnos of the dupacks so that rpseq and rpdupack are correct
403 * -sorbo.
404 */
405 /* need to bootstrap */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000406 if (hc->tx_rpdupack == -1) {
407 hc->tx_rpdupack = 0;
408 hc->tx_rpseq = seqno;
Arnaldo Carvalho de Meloc0c736d2006-03-20 22:05:37 -0800409 } else {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800410 /* check if packet is consecutive */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000411 if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1)
412 hc->tx_rpseq = seqno;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800413 /* it's a later packet */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000414 else if (after48(seqno, hc->tx_rpseq)) {
415 hc->tx_rpdupack++;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800416
417 /* check if we got enough dupacks */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000418 if (hc->tx_rpdupack >= NUMDUPACK) {
419 hc->tx_rpdupack = -1; /* XXX lame */
420 hc->tx_rpseq = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800421
Gerrit Renkerdf054e12007-11-24 21:32:53 -0200422 ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800423 }
424 }
425 }
426
427 /* check forward path congestion */
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100428 if (dccp_packet_without_ack(skb))
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800429 return;
430
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100431 /* still didn't send out new data packets */
432 if (hc->tx_seqh == hc->tx_seqt)
433 goto done;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800434
435 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000436 if (after48(ackno, hc->tx_high_ack))
437 hc->tx_high_ack = ackno;
Andrea Bittau32aac182006-11-16 14:28:40 -0200438
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000439 seqp = hc->tx_seqt;
Andrea Bittau32aac182006-11-16 14:28:40 -0200440 while (before48(seqp->ccid2s_seq, ackno)) {
441 seqp = seqp->ccid2s_next;
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000442 if (seqp == hc->tx_seqh) {
443 seqp = hc->tx_seqh->ccid2s_prev;
Andrea Bittau32aac182006-11-16 14:28:40 -0200444 break;
445 }
446 }
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800447
Gerrit Renkera3020022007-11-24 22:10:29 -0200448 /*
449 * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
450 * packets per acknowledgement. Rounding up avoids that cwnd is not
451 * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800452 */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000453 if (hc->tx_cwnd < hc->tx_ssthresh)
Gerrit Renkera3020022007-11-24 22:10:29 -0200454 maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800455
456 /* go through all ack vectors */
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100457 list_for_each_entry(avp, &hc->tx_av_chunks, node) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800458 /* go through this ack vector */
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100459 for (; avp->len--; avp->vec++) {
460 u64 ackno_end_rl = SUB48(ackno,
461 dccp_ackvec_runlen(avp->vec));
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800462
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100463 ccid2_pr_debug("ackvec %llu |%u,%u|\n",
Randy Dunlap234af482006-10-29 16:03:30 -0800464 (unsigned long long)ackno,
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100465 dccp_ackvec_state(avp->vec) >> 6,
466 dccp_ackvec_runlen(avp->vec));
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800467 /* if the seqno we are analyzing is larger than the
468 * current ackno, then move towards the tail of our
469 * seqnos.
470 */
471 while (after48(seqp->ccid2s_seq, ackno)) {
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000472 if (seqp == hc->tx_seqt) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800473 done = 1;
474 break;
475 }
476 seqp = seqp->ccid2s_prev;
477 }
478 if (done)
479 break;
480
481 /* check all seqnos in the range of the vector
482 * run length
483 */
484 while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100485 const u8 state = dccp_ackvec_state(avp->vec);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800486
487 /* new packet received or marked */
Gerrit Renkerf17a37c2010-11-10 21:20:07 +0100488 if (state != DCCPAV_NOT_RECEIVED &&
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800489 !seqp->ccid2s_acked) {
Gerrit Renkerf17a37c2010-11-10 21:20:07 +0100490 if (state == DCCPAV_ECN_MARKED)
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200491 ccid2_congestion_event(sk,
Andrea Bittau374bcf32006-09-19 13:14:43 -0700492 seqp);
Gerrit Renkerf17a37c2010-11-10 21:20:07 +0100493 else
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800494 ccid2_new_ack(sk, seqp,
495 &maxincr);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800496
497 seqp->ccid2s_acked = 1;
498 ccid2_pr_debug("Got ack for %llu\n",
Randy Dunlap234af482006-10-29 16:03:30 -0800499 (unsigned long long)seqp->ccid2s_seq);
Gerrit Renkerc38c92a2010-08-22 19:41:39 +0000500 hc->tx_pipe--;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800501 }
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000502 if (seqp == hc->tx_seqt) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800503 done = 1;
504 break;
505 }
Gerrit Renker3de54892007-11-24 20:37:48 -0200506 seqp = seqp->ccid2s_prev;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800507 }
508 if (done)
509 break;
510
Gerrit Renkercfbbeab2007-11-24 20:43:59 -0200511 ackno = SUB48(ackno_end_rl, 1);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800512 }
513 if (done)
514 break;
515 }
516
517 /* The state about what is acked should be correct now
518 * Check for NUMDUPACK
519 */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000520 seqp = hc->tx_seqt;
521 while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) {
Andrea Bittau32aac182006-11-16 14:28:40 -0200522 seqp = seqp->ccid2s_next;
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000523 if (seqp == hc->tx_seqh) {
524 seqp = hc->tx_seqh->ccid2s_prev;
Andrea Bittau32aac182006-11-16 14:28:40 -0200525 break;
526 }
527 }
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800528 done = 0;
529 while (1) {
530 if (seqp->ccid2s_acked) {
531 done++;
Gerrit Renker63df18a2007-11-24 22:04:35 -0200532 if (done == NUMDUPACK)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800533 break;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800534 }
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000535 if (seqp == hc->tx_seqt)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800536 break;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800537 seqp = seqp->ccid2s_prev;
538 }
539
540 /* If there are at least 3 acknowledgements, anything unacknowledged
541 * below the last sequence number is considered lost
542 */
Gerrit Renker63df18a2007-11-24 22:04:35 -0200543 if (done == NUMDUPACK) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800544 struct ccid2_seq *last_acked = seqp;
545
546 /* check for lost packets */
547 while (1) {
548 if (!seqp->ccid2s_acked) {
Andrea Bittau374bcf32006-09-19 13:14:43 -0700549 ccid2_pr_debug("Packet lost: %llu\n",
Randy Dunlap234af482006-10-29 16:03:30 -0800550 (unsigned long long)seqp->ccid2s_seq);
Andrea Bittau374bcf32006-09-19 13:14:43 -0700551 /* XXX need to traverse from tail -> head in
552 * order to detect multiple congestion events in
553 * one ack vector.
554 */
Gerrit Renkerd50ad162007-11-24 21:40:24 -0200555 ccid2_congestion_event(sk, seqp);
Gerrit Renkerc38c92a2010-08-22 19:41:39 +0000556 hc->tx_pipe--;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800557 }
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000558 if (seqp == hc->tx_seqt)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800559 break;
560 seqp = seqp->ccid2s_prev;
561 }
562
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000563 hc->tx_seqt = last_acked;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800564 }
565
566 /* trim acked packets in tail */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000567 while (hc->tx_seqt != hc->tx_seqh) {
568 if (!hc->tx_seqt->ccid2s_acked)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800569 break;
570
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000571 hc->tx_seqt = hc->tx_seqt->ccid2s_next;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800572 }
Gerrit Renkerc38c92a2010-08-22 19:41:39 +0000573
574 /* restart RTO timer if not all outstanding data has been acked */
575 if (hc->tx_pipe == 0)
576 sk_stop_timer(sk, &hc->tx_rtotimer);
577 else
578 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100579done:
Gerrit Renker1c0e0a02010-10-27 19:16:28 +0000580 /* check if incoming Acks allow pending packets to be sent */
581 if (sender_was_blocked && !ccid2_cwnd_network_limited(hc))
582 tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100583 dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800584}
585
Arnaldo Carvalho de Melo91f0ebf2006-03-20 19:21:44 -0800586static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800587{
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000588 struct ccid2_hc_tx_sock *hc = ccid_priv(ccid);
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200589 struct dccp_sock *dp = dccp_sk(sk);
590 u32 max_ratio;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800591
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200592 /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000593 hc->tx_ssthresh = ~0U;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800594
Gerrit Renker22b71c82010-08-29 19:23:12 +0000595 /* Use larger initial windows (RFC 4341, section 5). */
596 hc->tx_cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache);
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200597
598 /* Make sure that Ack Ratio is enabled and within bounds. */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000599 max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2);
Gerrit Renkerb00d2bb2007-11-24 21:44:30 -0200600 if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
601 dp->dccps_l_ack_ratio = max_ratio;
602
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800603 /* XXX init ~ to window size... */
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000604 if (ccid2_hc_tx_alloc_seq(hc))
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800605 return -ENOMEM;
Arnaldo Carvalho de Melo91f0ebf2006-03-20 19:21:44 -0800606
Gerrit Renker231cc2a2010-08-22 19:41:40 +0000607 hc->tx_rto = DCCP_TIMEOUT_INIT;
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000608 hc->tx_rpdupack = -1;
Gerrit Renkerd82b6f82010-08-29 19:23:10 +0000609 hc->tx_last_cong = ccid2_time_stamp;
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000610 setup_timer(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire,
Gerrit Renker410e27a2008-09-09 13:27:22 +0200611 (unsigned long)sk);
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100612 INIT_LIST_HEAD(&hc->tx_av_chunks);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800613 return 0;
614}
615
616static void ccid2_hc_tx_exit(struct sock *sk)
617{
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000618 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700619 int i;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800620
Gerrit Renkerd26eeb02010-08-29 19:23:11 +0000621 sk_stop_timer(sk, &hc->tx_rtotimer);
Andrea Bittau07978aa2006-09-19 13:13:37 -0700622
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000623 for (i = 0; i < hc->tx_seqbufc; i++)
624 kfree(hc->tx_seqbuf[i]);
625 hc->tx_seqbufc = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800626}
627
628static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
629{
630 const struct dccp_sock *dp = dccp_sk(sk);
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000631 struct ccid2_hc_rx_sock *hc = ccid2_hc_rx_sk(sk);
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800632
633 switch (DCCP_SKB_CB(skb)->dccpd_type) {
634 case DCCP_PKT_DATA:
635 case DCCP_PKT_DATAACK:
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000636 hc->rx_data++;
637 if (hc->rx_data >= dp->dccps_r_ack_ratio) {
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800638 dccp_send_ack(sk);
Gerrit Renker77d2dd92009-10-05 00:53:12 +0000639 hc->rx_data = 0;
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800640 }
641 break;
642 }
643}
644
Gerrit Renkerddebc972009-01-04 21:42:53 -0800645struct ccid_operations ccid2_ops = {
Gerrit Renker7e87fe82010-11-14 17:26:13 +0100646 .ccid_id = DCCPC_CCID2,
647 .ccid_name = "TCP-like",
648 .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock),
649 .ccid_hc_tx_init = ccid2_hc_tx_init,
650 .ccid_hc_tx_exit = ccid2_hc_tx_exit,
651 .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet,
652 .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent,
653 .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
654 .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv,
655 .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock),
656 .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv,
Andrea Bittau2a91aa32006-03-20 17:41:47 -0800657};
658
Gerrit Renker84116712006-11-20 18:26:03 -0200659#ifdef CONFIG_IP_DCCP_CCID2_DEBUG
Gerrit Renker43264992008-08-23 13:28:27 +0200660module_param(ccid2_debug, bool, 0644);
Gerrit Renkerddebc972009-01-04 21:42:53 -0800661MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages");
Gerrit Renker84116712006-11-20 18:26:03 -0200662#endif