1 // SPDX-License-Identifier: GPL-2.0-or-later <<
2 /* 1 /*
3 * Copyright (c) 2005, 2006 Andrea Bittau <a. 2 * Copyright (c) 2005, 2006 Andrea Bittau <a.bittau@cs.ucl.ac.uk>
4 * 3 *
5 * Changes to meet Linux coding standards, an 4 * Changes to meet Linux coding standards, and DCCP infrastructure fixes.
6 * 5 *
7 * Copyright (c) 2006 Arnaldo Carvalho de Mel 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.
8 */ 21 */
9 22
10 /* 23 /*
11 * This implementation should follow RFC 4341 24 * This implementation should follow RFC 4341
12 */ 25 */
13 #include <linux/slab.h> 26 #include <linux/slab.h>
14 #include "../feat.h" 27 #include "../feat.h"
15 #include "ccid2.h" 28 #include "ccid2.h"
16 29
17 30
18 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG 31 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
19 static bool ccid2_debug; 32 static bool ccid2_debug;
20 #define ccid2_pr_debug(format, a...) DCCP_P 33 #define ccid2_pr_debug(format, a...) DCCP_PR_DEBUG(ccid2_debug, format, ##a)
21 #else 34 #else
22 #define ccid2_pr_debug(format, a...) 35 #define ccid2_pr_debug(format, a...)
23 #endif 36 #endif
24 37
25 static int ccid2_hc_tx_alloc_seq(struct ccid2_ 38 static int ccid2_hc_tx_alloc_seq(struct ccid2_hc_tx_sock *hc)
26 { 39 {
27 struct ccid2_seq *seqp; 40 struct ccid2_seq *seqp;
28 int i; 41 int i;
29 42
30 /* check if we have space to preserve 43 /* check if we have space to preserve the pointer to the buffer */
31 if (hc->tx_seqbufc >= (sizeof(hc->tx_s 44 if (hc->tx_seqbufc >= (sizeof(hc->tx_seqbuf) /
32 sizeof(struct c 45 sizeof(struct ccid2_seq *)))
33 return -ENOMEM; 46 return -ENOMEM;
34 47
35 /* allocate buffer and initialize link 48 /* allocate buffer and initialize linked list */
36 seqp = kmalloc_array(CCID2_SEQBUF_LEN, !! 49 seqp = kmalloc(CCID2_SEQBUF_LEN * sizeof(struct ccid2_seq), gfp_any());
37 gfp_any()); <<
38 if (seqp == NULL) 50 if (seqp == NULL)
39 return -ENOMEM; 51 return -ENOMEM;
40 52
41 for (i = 0; i < (CCID2_SEQBUF_LEN - 1) 53 for (i = 0; i < (CCID2_SEQBUF_LEN - 1); i++) {
42 seqp[i].ccid2s_next = &seqp[i 54 seqp[i].ccid2s_next = &seqp[i + 1];
43 seqp[i + 1].ccid2s_prev = &seq 55 seqp[i + 1].ccid2s_prev = &seqp[i];
44 } 56 }
45 seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next 57 seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = seqp;
46 seqp->ccid2s_prev = &seqp[CCID2_SEQBUF 58 seqp->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
47 59
48 /* This is the first allocation. Init 60 /* This is the first allocation. Initiate the head and tail. */
49 if (hc->tx_seqbufc == 0) 61 if (hc->tx_seqbufc == 0)
50 hc->tx_seqh = hc->tx_seqt = se 62 hc->tx_seqh = hc->tx_seqt = seqp;
51 else { 63 else {
52 /* link the existing list with 64 /* link the existing list with the one we just created */
53 hc->tx_seqh->ccid2s_next = seq 65 hc->tx_seqh->ccid2s_next = seqp;
54 seqp->ccid2s_prev = hc->tx_seq 66 seqp->ccid2s_prev = hc->tx_seqh;
55 67
56 hc->tx_seqt->ccid2s_prev = &se 68 hc->tx_seqt->ccid2s_prev = &seqp[CCID2_SEQBUF_LEN - 1];
57 seqp[CCID2_SEQBUF_LEN - 1].cci 69 seqp[CCID2_SEQBUF_LEN - 1].ccid2s_next = hc->tx_seqt;
58 } 70 }
59 71
60 /* store the original pointer to the b 72 /* store the original pointer to the buffer so we can free it */
61 hc->tx_seqbuf[hc->tx_seqbufc] = seqp; 73 hc->tx_seqbuf[hc->tx_seqbufc] = seqp;
62 hc->tx_seqbufc++; 74 hc->tx_seqbufc++;
63 75
64 return 0; 76 return 0;
65 } 77 }
66 78
67 static int ccid2_hc_tx_send_packet(struct sock 79 static int ccid2_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
68 { 80 {
69 if (ccid2_cwnd_network_limited(ccid2_h 81 if (ccid2_cwnd_network_limited(ccid2_hc_tx_sk(sk)))
70 return CCID_PACKET_WILL_DEQUEU 82 return CCID_PACKET_WILL_DEQUEUE_LATER;
71 return CCID_PACKET_SEND_AT_ONCE; 83 return CCID_PACKET_SEND_AT_ONCE;
72 } 84 }
73 85
74 static void ccid2_change_l_ack_ratio(struct so 86 static void ccid2_change_l_ack_ratio(struct sock *sk, u32 val)
75 { 87 {
76 u32 max_ratio = DIV_ROUND_UP(ccid2_hc_ 88 u32 max_ratio = DIV_ROUND_UP(ccid2_hc_tx_sk(sk)->tx_cwnd, 2);
77 89
78 /* 90 /*
79 * Ensure that Ack Ratio does not exce 91 * Ensure that Ack Ratio does not exceed ceil(cwnd/2), which is (2) from
80 * RFC 4341, 6.1.2. We ignore the stat 92 * RFC 4341, 6.1.2. We ignore the statement that Ack Ratio 2 is always
81 * acceptable since this causes starva 93 * acceptable since this causes starvation/deadlock whenever cwnd < 2.
82 * The same problem arises when Ack Ra 94 * The same problem arises when Ack Ratio is 0 (ie. Ack Ratio disabled).
83 */ 95 */
84 if (val == 0 || val > max_ratio) { 96 if (val == 0 || val > max_ratio) {
85 DCCP_WARN("Limiting Ack Ratio 97 DCCP_WARN("Limiting Ack Ratio (%u) to %u\n", val, max_ratio);
86 val = max_ratio; 98 val = max_ratio;
87 } 99 }
88 dccp_feat_signal_nn_change(sk, DCCPF_A 100 dccp_feat_signal_nn_change(sk, DCCPF_ACK_RATIO,
89 min_t(u32, 101 min_t(u32, val, DCCPF_ACK_RATIO_MAX));
90 } 102 }
91 103
92 static void ccid2_check_l_ack_ratio(struct soc 104 static void ccid2_check_l_ack_ratio(struct sock *sk)
93 { 105 {
94 struct ccid2_hc_tx_sock *hc = ccid2_hc 106 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
95 107
96 /* 108 /*
97 * After a loss, idle period, applicat 109 * After a loss, idle period, application limited period, or RTO we
98 * need to check that the ack ratio is 110 * need to check that the ack ratio is still less than the congestion
99 * window. Otherwise, we will send an 111 * window. Otherwise, we will send an entire congestion window of
100 * packets and got no response because 112 * packets and got no response because we haven't sent ack ratio
101 * packets yet. 113 * packets yet.
102 * If the ack ratio does need to be re 114 * If the ack ratio does need to be reduced, we reduce it to half of
103 * the congestion window (or 1 if that 115 * the congestion window (or 1 if that's zero) instead of to the
104 * congestion window. This prevents pr 116 * congestion window. This prevents problems if one ack is lost.
105 */ 117 */
106 if (dccp_feat_nn_get(sk, DCCPF_ACK_RAT 118 if (dccp_feat_nn_get(sk, DCCPF_ACK_RATIO) > hc->tx_cwnd)
107 ccid2_change_l_ack_ratio(sk, h 119 ccid2_change_l_ack_ratio(sk, hc->tx_cwnd/2 ? : 1U);
108 } 120 }
109 121
110 static void ccid2_change_l_seq_window(struct s 122 static void ccid2_change_l_seq_window(struct sock *sk, u64 val)
111 { 123 {
112 dccp_feat_signal_nn_change(sk, DCCPF_S 124 dccp_feat_signal_nn_change(sk, DCCPF_SEQUENCE_WINDOW,
113 clamp_val(v 125 clamp_val(val, DCCPF_SEQ_WMIN,
114 126 DCCPF_SEQ_WMAX));
115 } 127 }
116 128
117 static void dccp_tasklet_schedule(struct sock !! 129 static void ccid2_hc_tx_rto_expire(unsigned long data)
118 { <<
119 struct tasklet_struct *t = &dccp_sk(sk <<
120 <<
121 if (!test_and_set_bit(TASKLET_STATE_SC <<
122 sock_hold(sk); <<
123 __tasklet_schedule(t); <<
124 } <<
125 } <<
126 <<
127 static void ccid2_hc_tx_rto_expire(struct time <<
128 { 130 {
129 struct ccid2_hc_tx_sock *hc = from_tim !! 131 struct sock *sk = (struct sock *)data;
130 struct sock *sk = hc->sk; !! 132 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
131 const bool sender_was_blocked = ccid2_ 133 const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
132 134
133 bh_lock_sock(sk); 135 bh_lock_sock(sk);
134 if (sock_owned_by_user(sk)) { 136 if (sock_owned_by_user(sk)) {
135 sk_reset_timer(sk, &hc->tx_rto 137 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5);
136 goto out; 138 goto out;
137 } 139 }
138 140
139 ccid2_pr_debug("RTO_EXPIRE\n"); 141 ccid2_pr_debug("RTO_EXPIRE\n");
140 142
141 if (sk->sk_state == DCCP_CLOSED) <<
142 goto out; <<
143 <<
144 /* back-off timer */ 143 /* back-off timer */
145 hc->tx_rto <<= 1; 144 hc->tx_rto <<= 1;
146 if (hc->tx_rto > DCCP_RTO_MAX) 145 if (hc->tx_rto > DCCP_RTO_MAX)
147 hc->tx_rto = DCCP_RTO_MAX; 146 hc->tx_rto = DCCP_RTO_MAX;
148 147
149 /* adjust pipe, cwnd etc */ 148 /* adjust pipe, cwnd etc */
150 hc->tx_ssthresh = hc->tx_cwnd / 2; 149 hc->tx_ssthresh = hc->tx_cwnd / 2;
151 if (hc->tx_ssthresh < 2) 150 if (hc->tx_ssthresh < 2)
152 hc->tx_ssthresh = 2; 151 hc->tx_ssthresh = 2;
153 hc->tx_cwnd = 1; 152 hc->tx_cwnd = 1;
154 hc->tx_pipe = 0; 153 hc->tx_pipe = 0;
155 154
156 /* clear state about stuff we sent */ 155 /* clear state about stuff we sent */
157 hc->tx_seqt = hc->tx_seqh; 156 hc->tx_seqt = hc->tx_seqh;
158 hc->tx_packets_acked = 0; 157 hc->tx_packets_acked = 0;
159 158
160 /* clear ack ratio state. */ 159 /* clear ack ratio state. */
161 hc->tx_rpseq = 0; 160 hc->tx_rpseq = 0;
162 hc->tx_rpdupack = -1; 161 hc->tx_rpdupack = -1;
163 ccid2_change_l_ack_ratio(sk, 1); 162 ccid2_change_l_ack_ratio(sk, 1);
164 163
165 /* if we were blocked before, we may n 164 /* if we were blocked before, we may now send cwnd=1 packet */
166 if (sender_was_blocked) 165 if (sender_was_blocked)
167 dccp_tasklet_schedule(sk); !! 166 tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
168 /* restart backed-off timer */ 167 /* restart backed-off timer */
169 sk_reset_timer(sk, &hc->tx_rtotimer, j 168 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
170 out: 169 out:
171 bh_unlock_sock(sk); 170 bh_unlock_sock(sk);
172 sock_put(sk); 171 sock_put(sk);
173 } 172 }
174 173
175 /* 174 /*
176 * Congestion window validation (RFC 2861 175 * Congestion window validation (RFC 2861).
177 */ 176 */
178 static bool ccid2_do_cwv = true; 177 static bool ccid2_do_cwv = true;
179 module_param(ccid2_do_cwv, bool, 0644); 178 module_param(ccid2_do_cwv, bool, 0644);
180 MODULE_PARM_DESC(ccid2_do_cwv, "Perform RFC286 179 MODULE_PARM_DESC(ccid2_do_cwv, "Perform RFC2861 Congestion Window Validation");
181 180
182 /** 181 /**
183 * ccid2_update_used_window - Track how much 182 * ccid2_update_used_window - Track how much of cwnd is actually used
184 * @hc: socket to update window <<
185 * @new_wnd: new window values to add into the <<
186 * <<
187 * This is done in addition to CWV. The sender 183 * This is done in addition to CWV. The sender needs to have an idea of how many
188 * packets may be in flight, to set the local 184 * packets may be in flight, to set the local Sequence Window value accordingly
189 * (RFC 4340, 7.5.2). The CWV mechanism is exp 185 * (RFC 4340, 7.5.2). The CWV mechanism is exploited to keep track of the
190 * maximum-used window. We use an EWMA low-pas 186 * maximum-used window. We use an EWMA low-pass filter to filter out noise.
191 */ 187 */
192 static void ccid2_update_used_window(struct cc 188 static void ccid2_update_used_window(struct ccid2_hc_tx_sock *hc, u32 new_wnd)
193 { 189 {
194 hc->tx_expected_wnd = (3 * hc->tx_expe 190 hc->tx_expected_wnd = (3 * hc->tx_expected_wnd + new_wnd) / 4;
195 } 191 }
196 192
197 /* This borrows the code of tcp_cwnd_applicati 193 /* This borrows the code of tcp_cwnd_application_limited() */
198 static void ccid2_cwnd_application_limited(str 194 static void ccid2_cwnd_application_limited(struct sock *sk, const u32 now)
199 { 195 {
200 struct ccid2_hc_tx_sock *hc = ccid2_hc 196 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
201 /* don't reduce cwnd below the initial 197 /* don't reduce cwnd below the initial window (IW) */
202 u32 init_win = rfc3390_bytes_to_packet 198 u32 init_win = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache),
203 win_used = max(hc->tx_cwnd_used, i 199 win_used = max(hc->tx_cwnd_used, init_win);
204 200
205 if (win_used < hc->tx_cwnd) { 201 if (win_used < hc->tx_cwnd) {
206 hc->tx_ssthresh = max(hc->tx_s 202 hc->tx_ssthresh = max(hc->tx_ssthresh,
207 (hc->tx_c 203 (hc->tx_cwnd >> 1) + (hc->tx_cwnd >> 2));
208 hc->tx_cwnd = (hc->tx_cwnd + w 204 hc->tx_cwnd = (hc->tx_cwnd + win_used) >> 1;
209 } 205 }
210 hc->tx_cwnd_used = 0; 206 hc->tx_cwnd_used = 0;
211 hc->tx_cwnd_stamp = now; 207 hc->tx_cwnd_stamp = now;
212 208
213 ccid2_check_l_ack_ratio(sk); 209 ccid2_check_l_ack_ratio(sk);
214 } 210 }
215 211
216 /* This borrows the code of tcp_cwnd_restart() 212 /* This borrows the code of tcp_cwnd_restart() */
217 static void ccid2_cwnd_restart(struct sock *sk 213 static void ccid2_cwnd_restart(struct sock *sk, const u32 now)
218 { 214 {
219 struct ccid2_hc_tx_sock *hc = ccid2_hc 215 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
220 u32 cwnd = hc->tx_cwnd, restart_cwnd, 216 u32 cwnd = hc->tx_cwnd, restart_cwnd,
221 iwnd = rfc3390_bytes_to_packets(dc 217 iwnd = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache);
222 s32 delta = now - hc->tx_lsndtime; <<
223 218
224 hc->tx_ssthresh = max(hc->tx_ssthresh, 219 hc->tx_ssthresh = max(hc->tx_ssthresh, (cwnd >> 1) + (cwnd >> 2));
225 220
226 /* don't reduce cwnd below the initial 221 /* don't reduce cwnd below the initial window (IW) */
227 restart_cwnd = min(cwnd, iwnd); 222 restart_cwnd = min(cwnd, iwnd);
228 !! 223 cwnd >>= (now - hc->tx_lsndtime) / hc->tx_rto;
229 while ((delta -= hc->tx_rto) >= 0 && c <<
230 cwnd >>= 1; <<
231 hc->tx_cwnd = max(cwnd, restart_cwnd); 224 hc->tx_cwnd = max(cwnd, restart_cwnd);
>> 225
232 hc->tx_cwnd_stamp = now; 226 hc->tx_cwnd_stamp = now;
233 hc->tx_cwnd_used = 0; 227 hc->tx_cwnd_used = 0;
234 228
235 ccid2_check_l_ack_ratio(sk); 229 ccid2_check_l_ack_ratio(sk);
236 } 230 }
237 231
238 static void ccid2_hc_tx_packet_sent(struct soc 232 static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len)
239 { 233 {
240 struct dccp_sock *dp = dccp_sk(sk); 234 struct dccp_sock *dp = dccp_sk(sk);
241 struct ccid2_hc_tx_sock *hc = ccid2_hc 235 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
242 const u32 now = ccid2_jiffies32; 236 const u32 now = ccid2_jiffies32;
243 struct ccid2_seq *next; 237 struct ccid2_seq *next;
244 238
245 /* slow-start after idle periods (RFC 239 /* slow-start after idle periods (RFC 2581, RFC 2861) */
246 if (ccid2_do_cwv && !hc->tx_pipe && 240 if (ccid2_do_cwv && !hc->tx_pipe &&
247 (s32)(now - hc->tx_lsndtime) >= hc 241 (s32)(now - hc->tx_lsndtime) >= hc->tx_rto)
248 ccid2_cwnd_restart(sk, now); 242 ccid2_cwnd_restart(sk, now);
249 243
250 hc->tx_lsndtime = now; 244 hc->tx_lsndtime = now;
251 hc->tx_pipe += 1; 245 hc->tx_pipe += 1;
252 246
253 /* see whether cwnd was fully used (RF 247 /* see whether cwnd was fully used (RFC 2861), update expected window */
254 if (ccid2_cwnd_network_limited(hc)) { 248 if (ccid2_cwnd_network_limited(hc)) {
255 ccid2_update_used_window(hc, h 249 ccid2_update_used_window(hc, hc->tx_cwnd);
256 hc->tx_cwnd_used = 0; 250 hc->tx_cwnd_used = 0;
257 hc->tx_cwnd_stamp = now; 251 hc->tx_cwnd_stamp = now;
258 } else { 252 } else {
259 if (hc->tx_pipe > hc->tx_cwnd_ 253 if (hc->tx_pipe > hc->tx_cwnd_used)
260 hc->tx_cwnd_used = hc- 254 hc->tx_cwnd_used = hc->tx_pipe;
261 255
262 ccid2_update_used_window(hc, h 256 ccid2_update_used_window(hc, hc->tx_cwnd_used);
263 257
264 if (ccid2_do_cwv && (s32)(now 258 if (ccid2_do_cwv && (s32)(now - hc->tx_cwnd_stamp) >= hc->tx_rto)
265 ccid2_cwnd_application 259 ccid2_cwnd_application_limited(sk, now);
266 } 260 }
267 261
268 hc->tx_seqh->ccid2s_seq = dp->dccps_ 262 hc->tx_seqh->ccid2s_seq = dp->dccps_gss;
269 hc->tx_seqh->ccid2s_acked = 0; 263 hc->tx_seqh->ccid2s_acked = 0;
270 hc->tx_seqh->ccid2s_sent = now; 264 hc->tx_seqh->ccid2s_sent = now;
271 265
272 next = hc->tx_seqh->ccid2s_next; 266 next = hc->tx_seqh->ccid2s_next;
273 /* check if we need to alloc more spac 267 /* check if we need to alloc more space */
274 if (next == hc->tx_seqt) { 268 if (next == hc->tx_seqt) {
275 if (ccid2_hc_tx_alloc_seq(hc)) 269 if (ccid2_hc_tx_alloc_seq(hc)) {
276 DCCP_CRIT("packet hist 270 DCCP_CRIT("packet history - out of memory!");
277 /* FIXME: find a more 271 /* FIXME: find a more graceful way to bail out */
278 return; 272 return;
279 } 273 }
280 next = hc->tx_seqh->ccid2s_nex 274 next = hc->tx_seqh->ccid2s_next;
281 BUG_ON(next == hc->tx_seqt); 275 BUG_ON(next == hc->tx_seqt);
282 } 276 }
283 hc->tx_seqh = next; 277 hc->tx_seqh = next;
284 278
285 ccid2_pr_debug("cwnd=%d pipe=%d\n", hc 279 ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe);
286 280
287 /* 281 /*
288 * FIXME: The code below is broken and 282 * FIXME: The code below is broken and the variables have been removed
289 * from the socket struct. The `acklos 283 * from the socket struct. The `ackloss' variable was always set to 0,
290 * and with arsent there are several p 284 * and with arsent there are several problems:
291 * (i) it doesn't just count the numb 285 * (i) it doesn't just count the number of Acks, but all sent packets;
292 * (ii) it is expressed in # of packe 286 * (ii) it is expressed in # of packets, not # of windows, so the
293 * comparison below uses the wrong fo 287 * comparison below uses the wrong formula: Appendix A of RFC 4341
294 * comes up with the number K = cwnd 288 * comes up with the number K = cwnd / (R^2 - R) of consecutive windows
295 * of data with no lost or marked Ack 289 * of data with no lost or marked Ack packets. If arsent were the # of
296 * consecutive Acks received without 290 * consecutive Acks received without loss, then Ack Ratio needs to be
297 * decreased by 1 when 291 * decreased by 1 when
298 * arsent >= K * cwnd / R 292 * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2)
299 * where cwnd / R is the number of Ac 293 * where cwnd / R is the number of Acks received per window of data
300 * (cf. RFC 4341, App. A). The proble 294 * (cf. RFC 4341, App. A). The problems are that
301 * - arsent counts other packets as w 295 * - arsent counts other packets as well;
302 * - the comparison uses a formula di 296 * - the comparison uses a formula different from RFC 4341;
303 * - computing a cubic/quadratic equa 297 * - computing a cubic/quadratic equation each time is too complicated.
304 * Hence a different algorithm is nee 298 * Hence a different algorithm is needed.
305 */ 299 */
306 #if 0 300 #if 0
307 /* Ack Ratio. Need to maintain a conc 301 /* Ack Ratio. Need to maintain a concept of how many windows we sent */
308 hc->tx_arsent++; 302 hc->tx_arsent++;
309 /* We had an ack loss in this window.. 303 /* We had an ack loss in this window... */
310 if (hc->tx_ackloss) { 304 if (hc->tx_ackloss) {
311 if (hc->tx_arsent >= hc->tx_cw 305 if (hc->tx_arsent >= hc->tx_cwnd) {
312 hc->tx_arsent = 0; 306 hc->tx_arsent = 0;
313 hc->tx_ackloss = 0; 307 hc->tx_ackloss = 0;
314 } 308 }
315 } else { 309 } else {
316 /* No acks lost up to now... * 310 /* No acks lost up to now... */
317 /* decrease ack ratio if enoug 311 /* decrease ack ratio if enough packets were sent */
318 if (dp->dccps_l_ack_ratio > 1) 312 if (dp->dccps_l_ack_ratio > 1) {
319 /* XXX don't calculate 313 /* XXX don't calculate denominator each time */
320 int denom = dp->dccps_ 314 int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio -
321 dp->dccps_ 315 dp->dccps_l_ack_ratio;
322 316
323 denom = hc->tx_cwnd * 317 denom = hc->tx_cwnd * hc->tx_cwnd / denom;
324 318
325 if (hc->tx_arsent >= d 319 if (hc->tx_arsent >= denom) {
326 ccid2_change_l 320 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1);
327 hc->tx_arsent 321 hc->tx_arsent = 0;
328 } 322 }
329 } else { 323 } else {
330 /* we can't increase a 324 /* we can't increase ack ratio further [1] */
331 hc->tx_arsent = 0; /* 325 hc->tx_arsent = 0; /* or maybe set it to cwnd*/
332 } 326 }
333 } 327 }
334 #endif 328 #endif
335 329
336 sk_reset_timer(sk, &hc->tx_rtotimer, j 330 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
337 331
338 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG 332 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
339 do { 333 do {
340 struct ccid2_seq *seqp = hc->t 334 struct ccid2_seq *seqp = hc->tx_seqt;
341 335
342 while (seqp != hc->tx_seqh) { 336 while (seqp != hc->tx_seqh) {
343 ccid2_pr_debug("out se 337 ccid2_pr_debug("out seq=%llu acked=%d time=%u\n",
344 (unsign 338 (unsigned long long)seqp->ccid2s_seq,
345 seqp->c 339 seqp->ccid2s_acked, seqp->ccid2s_sent);
346 seqp = seqp->ccid2s_ne 340 seqp = seqp->ccid2s_next;
347 } 341 }
348 } while (0); 342 } while (0);
349 ccid2_pr_debug("=========\n"); 343 ccid2_pr_debug("=========\n");
350 #endif 344 #endif
351 } 345 }
352 346
353 /** 347 /**
354 * ccid2_rtt_estimator - Sample RTT and comput 348 * ccid2_rtt_estimator - Sample RTT and compute RTO using RFC2988 algorithm
355 * @sk: socket to perform estimator on <<
356 * @mrtt: measured RTT <<
357 * <<
358 * This code is almost identical with TCP's tc 349 * This code is almost identical with TCP's tcp_rtt_estimator(), since
359 * - it has a higher sampling frequency (recom 350 * - it has a higher sampling frequency (recommended by RFC 1323),
360 * - the RTO does not collapse into RTT due to 351 * - the RTO does not collapse into RTT due to RTTVAR going towards zero,
361 * - it is simple (cf. more complex proposals 352 * - it is simple (cf. more complex proposals such as Eifel timer or research
362 * which suggests that the gain should be se 353 * which suggests that the gain should be set according to window size),
363 * - in tests it was found to work well with C 354 * - in tests it was found to work well with CCID2 [gerrit].
364 */ 355 */
365 static void ccid2_rtt_estimator(struct sock *s 356 static void ccid2_rtt_estimator(struct sock *sk, const long mrtt)
366 { 357 {
367 struct ccid2_hc_tx_sock *hc = ccid2_hc 358 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
368 long m = mrtt ? : 1; 359 long m = mrtt ? : 1;
369 360
370 if (hc->tx_srtt == 0) { 361 if (hc->tx_srtt == 0) {
371 /* First measurement m */ 362 /* First measurement m */
372 hc->tx_srtt = m << 3; 363 hc->tx_srtt = m << 3;
373 hc->tx_mdev = m << 1; 364 hc->tx_mdev = m << 1;
374 365
375 hc->tx_mdev_max = max(hc->tx_m 366 hc->tx_mdev_max = max(hc->tx_mdev, tcp_rto_min(sk));
376 hc->tx_rttvar = hc->tx_mdev_ 367 hc->tx_rttvar = hc->tx_mdev_max;
377 368
378 hc->tx_rtt_seq = dccp_sk(sk)- 369 hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss;
379 } else { 370 } else {
380 /* Update scaled SRTT as SRTT 371 /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */
381 m -= (hc->tx_srtt >> 3); 372 m -= (hc->tx_srtt >> 3);
382 hc->tx_srtt += m; 373 hc->tx_srtt += m;
383 374
384 /* Similarly, update scaled md 375 /* Similarly, update scaled mdev with regard to |m| */
385 if (m < 0) { 376 if (m < 0) {
386 m = -m; 377 m = -m;
387 m -= (hc->tx_mdev >> 2 378 m -= (hc->tx_mdev >> 2);
388 /* 379 /*
389 * This neutralises RT 380 * This neutralises RTO increase when RTT < SRTT - mdev
390 * (see P. Sarolahti, 381 * (see P. Sarolahti, A. Kuznetsov,"Congestion Control
391 * in Linux TCP", USEN 382 * in Linux TCP", USENIX 2002, pp. 49-62).
392 */ 383 */
393 if (m > 0) 384 if (m > 0)
394 m >>= 3; 385 m >>= 3;
395 } else { 386 } else {
396 m -= (hc->tx_mdev >> 2 387 m -= (hc->tx_mdev >> 2);
397 } 388 }
398 hc->tx_mdev += m; 389 hc->tx_mdev += m;
399 390
400 if (hc->tx_mdev > hc->tx_mdev_ 391 if (hc->tx_mdev > hc->tx_mdev_max) {
401 hc->tx_mdev_max = hc-> 392 hc->tx_mdev_max = hc->tx_mdev;
402 if (hc->tx_mdev_max > 393 if (hc->tx_mdev_max > hc->tx_rttvar)
403 hc->tx_rttvar 394 hc->tx_rttvar = hc->tx_mdev_max;
404 } 395 }
405 396
406 /* 397 /*
407 * Decay RTTVAR at most once p 398 * Decay RTTVAR at most once per flight, exploiting that
408 * 1) pipe <= cwnd <= Sequenc 399 * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2)
409 * 2) AWL = GSS-W+1 <= GAR <= 400 * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1)
410 * GAR is a useful bound for F 401 * GAR is a useful bound for FlightSize = pipe.
411 * AWL is probably too low her 402 * AWL is probably too low here, as it over-estimates pipe.
412 */ 403 */
413 if (after48(dccp_sk(sk)->dccps 404 if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) {
414 if (hc->tx_mdev_max < 405 if (hc->tx_mdev_max < hc->tx_rttvar)
415 hc->tx_rttvar 406 hc->tx_rttvar -= (hc->tx_rttvar -
416 407 hc->tx_mdev_max) >> 2;
417 hc->tx_rtt_seq = dccp 408 hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss;
418 hc->tx_mdev_max = tcp_ 409 hc->tx_mdev_max = tcp_rto_min(sk);
419 } 410 }
420 } 411 }
421 412
422 /* 413 /*
423 * Set RTO from SRTT and RTTVAR 414 * Set RTO from SRTT and RTTVAR
424 * As in TCP, 4 * RTTVAR >= TCP_RTO_MI 415 * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms.
425 * This agrees with RFC 4341, 5: 416 * This agrees with RFC 4341, 5:
426 * "Because DCCP does not retrans 417 * "Because DCCP does not retransmit data, DCCP does not require
427 * TCP's recommended minimum tim 418 * TCP's recommended minimum timeout of one second".
428 */ 419 */
429 hc->tx_rto = (hc->tx_srtt >> 3) + hc-> 420 hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar;
430 421
431 if (hc->tx_rto > DCCP_RTO_MAX) 422 if (hc->tx_rto > DCCP_RTO_MAX)
432 hc->tx_rto = DCCP_RTO_MAX; 423 hc->tx_rto = DCCP_RTO_MAX;
433 } 424 }
434 425
435 static void ccid2_new_ack(struct sock *sk, str 426 static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp,
436 unsigned int *maxinc 427 unsigned int *maxincr)
437 { 428 {
438 struct ccid2_hc_tx_sock *hc = ccid2_hc 429 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
439 struct dccp_sock *dp = dccp_sk(sk); 430 struct dccp_sock *dp = dccp_sk(sk);
440 int r_seq_used = hc->tx_cwnd / dp->dcc 431 int r_seq_used = hc->tx_cwnd / dp->dccps_l_ack_ratio;
441 432
442 if (hc->tx_cwnd < dp->dccps_l_seq_win 433 if (hc->tx_cwnd < dp->dccps_l_seq_win &&
443 r_seq_used < dp->dccps_r_seq_win) 434 r_seq_used < dp->dccps_r_seq_win) {
444 if (hc->tx_cwnd < hc->tx_ssthr 435 if (hc->tx_cwnd < hc->tx_ssthresh) {
445 if (*maxincr > 0 && ++ 436 if (*maxincr > 0 && ++hc->tx_packets_acked >= 2) {
446 hc->tx_cwnd += 437 hc->tx_cwnd += 1;
447 *maxincr -= 438 *maxincr -= 1;
448 hc->tx_packets 439 hc->tx_packets_acked = 0;
449 } 440 }
450 } else if (++hc->tx_packets_ac 441 } else if (++hc->tx_packets_acked >= hc->tx_cwnd) {
451 hc->tx_cwnd += 1; 442 hc->tx_cwnd += 1;
452 hc->tx_packets_acked = 443 hc->tx_packets_acked = 0;
453 } 444 }
454 } 445 }
455 446
456 /* 447 /*
457 * Adjust the local sequence window an 448 * Adjust the local sequence window and the ack ratio to allow about
458 * 5 times the number of packets in th 449 * 5 times the number of packets in the network (RFC 4340 7.5.2)
459 */ 450 */
460 if (r_seq_used * CCID2_WIN_CHANGE_FACT 451 if (r_seq_used * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_r_seq_win)
461 ccid2_change_l_ack_ratio(sk, d 452 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio * 2);
462 else if (r_seq_used * CCID2_WIN_CHANGE 453 else if (r_seq_used * CCID2_WIN_CHANGE_FACTOR < dp->dccps_r_seq_win/2)
463 ccid2_change_l_ack_ratio(sk, d 454 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio / 2 ? : 1U);
464 455
465 if (hc->tx_cwnd * CCID2_WIN_CHANGE_FAC 456 if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_l_seq_win)
466 ccid2_change_l_seq_window(sk, 457 ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win * 2);
467 else if (hc->tx_cwnd * CCID2_WIN_CHANG 458 else if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR < dp->dccps_l_seq_win/2)
468 ccid2_change_l_seq_window(sk, 459 ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win / 2);
469 460
470 /* 461 /*
471 * FIXME: RTT is sampled several times 462 * FIXME: RTT is sampled several times per acknowledgment (for each
472 * entry in the Ack Vector), instead o 463 * entry in the Ack Vector), instead of once per Ack (as in TCP SACK).
473 * This causes the RTT to be over-esti 464 * This causes the RTT to be over-estimated, since the older entries
474 * in the Ack Vector have earlier send 465 * in the Ack Vector have earlier sending times.
475 * The cleanest solution is to not use 466 * The cleanest solution is to not use the ccid2s_sent field at all
476 * and instead use DCCP timestamps: re 467 * and instead use DCCP timestamps: requires changes in other places.
477 */ 468 */
478 ccid2_rtt_estimator(sk, ccid2_jiffies3 469 ccid2_rtt_estimator(sk, ccid2_jiffies32 - seqp->ccid2s_sent);
479 } 470 }
480 471
481 static void ccid2_congestion_event(struct sock 472 static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp)
482 { 473 {
483 struct ccid2_hc_tx_sock *hc = ccid2_hc 474 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
484 475
485 if ((s32)(seqp->ccid2s_sent - hc->tx_l 476 if ((s32)(seqp->ccid2s_sent - hc->tx_last_cong) < 0) {
486 ccid2_pr_debug("Multiple losse 477 ccid2_pr_debug("Multiple losses in an RTT---treating as one\n");
487 return; 478 return;
488 } 479 }
489 480
490 hc->tx_last_cong = ccid2_jiffies32; 481 hc->tx_last_cong = ccid2_jiffies32;
491 482
492 hc->tx_cwnd = hc->tx_cwnd / 2 ? : 483 hc->tx_cwnd = hc->tx_cwnd / 2 ? : 1U;
493 hc->tx_ssthresh = max(hc->tx_cwnd, 2U 484 hc->tx_ssthresh = max(hc->tx_cwnd, 2U);
494 485
495 ccid2_check_l_ack_ratio(sk); 486 ccid2_check_l_ack_ratio(sk);
496 } 487 }
497 488
498 static int ccid2_hc_tx_parse_options(struct so 489 static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type,
499 u8 option 490 u8 option, u8 *optval, u8 optlen)
500 { 491 {
501 struct ccid2_hc_tx_sock *hc = ccid2_hc 492 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
502 493
503 switch (option) { 494 switch (option) {
504 case DCCPO_ACK_VECTOR_0: 495 case DCCPO_ACK_VECTOR_0:
505 case DCCPO_ACK_VECTOR_1: 496 case DCCPO_ACK_VECTOR_1:
506 return dccp_ackvec_parsed_add( 497 return dccp_ackvec_parsed_add(&hc->tx_av_chunks, optval, optlen,
507 498 option - DCCPO_ACK_VECTOR_0);
508 } 499 }
509 return 0; 500 return 0;
510 } 501 }
511 502
512 static void ccid2_hc_tx_packet_recv(struct soc 503 static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
513 { 504 {
514 struct dccp_sock *dp = dccp_sk(sk); 505 struct dccp_sock *dp = dccp_sk(sk);
515 struct ccid2_hc_tx_sock *hc = ccid2_hc 506 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
516 const bool sender_was_blocked = ccid2_ 507 const bool sender_was_blocked = ccid2_cwnd_network_limited(hc);
517 struct dccp_ackvec_parsed *avp; 508 struct dccp_ackvec_parsed *avp;
518 u64 ackno, seqno; 509 u64 ackno, seqno;
519 struct ccid2_seq *seqp; 510 struct ccid2_seq *seqp;
520 int done = 0; 511 int done = 0;
521 unsigned int maxincr = 0; 512 unsigned int maxincr = 0;
522 513
523 /* check reverse path congestion */ 514 /* check reverse path congestion */
524 seqno = DCCP_SKB_CB(skb)->dccpd_seq; 515 seqno = DCCP_SKB_CB(skb)->dccpd_seq;
525 516
526 /* XXX this whole "algorithm" is broke 517 /* XXX this whole "algorithm" is broken. Need to fix it to keep track
527 * of the seqnos of the dupacks so tha 518 * of the seqnos of the dupacks so that rpseq and rpdupack are correct
528 * -sorbo. 519 * -sorbo.
529 */ 520 */
530 /* need to bootstrap */ 521 /* need to bootstrap */
531 if (hc->tx_rpdupack == -1) { 522 if (hc->tx_rpdupack == -1) {
532 hc->tx_rpdupack = 0; 523 hc->tx_rpdupack = 0;
533 hc->tx_rpseq = seqno; 524 hc->tx_rpseq = seqno;
534 } else { 525 } else {
535 /* check if packet is consecut 526 /* check if packet is consecutive */
536 if (dccp_delta_seqno(hc->tx_rp 527 if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1)
537 hc->tx_rpseq = seqno; 528 hc->tx_rpseq = seqno;
538 /* it's a later packet */ 529 /* it's a later packet */
539 else if (after48(seqno, hc->tx 530 else if (after48(seqno, hc->tx_rpseq)) {
540 hc->tx_rpdupack++; 531 hc->tx_rpdupack++;
541 532
542 /* check if we got eno 533 /* check if we got enough dupacks */
543 if (hc->tx_rpdupack >= 534 if (hc->tx_rpdupack >= NUMDUPACK) {
544 hc->tx_rpdupac 535 hc->tx_rpdupack = -1; /* XXX lame */
545 hc->tx_rpseq 536 hc->tx_rpseq = 0;
546 #ifdef __CCID2_COPES_GRACEFULLY_WITH_ACK_CONGE 537 #ifdef __CCID2_COPES_GRACEFULLY_WITH_ACK_CONGESTION_CONTROL__
547 /* 538 /*
548 * FIXME: Ack 539 * FIXME: Ack Congestion Control is broken; in
549 * the current 540 * the current state instabilities occurred with
550 * Ack Ratios 541 * Ack Ratios greater than 1; causing hang-ups
551 * and long RT 542 * and long RTO timeouts. This needs to be fixed
552 * before open 543 * before opening up dynamic changes. -- gerrit
553 */ 544 */
554 ccid2_change_l 545 ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio);
555 #endif 546 #endif
556 } 547 }
557 } 548 }
558 } 549 }
559 550
560 /* check forward path congestion */ 551 /* check forward path congestion */
561 if (dccp_packet_without_ack(skb)) 552 if (dccp_packet_without_ack(skb))
562 return; 553 return;
563 554
564 /* still didn't send out new data pack 555 /* still didn't send out new data packets */
565 if (hc->tx_seqh == hc->tx_seqt) 556 if (hc->tx_seqh == hc->tx_seqt)
566 goto done; 557 goto done;
567 558
568 ackno = DCCP_SKB_CB(skb)->dccpd_ack_se 559 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq;
569 if (after48(ackno, hc->tx_high_ack)) 560 if (after48(ackno, hc->tx_high_ack))
570 hc->tx_high_ack = ackno; 561 hc->tx_high_ack = ackno;
571 562
572 seqp = hc->tx_seqt; 563 seqp = hc->tx_seqt;
573 while (before48(seqp->ccid2s_seq, ackn 564 while (before48(seqp->ccid2s_seq, ackno)) {
574 seqp = seqp->ccid2s_next; 565 seqp = seqp->ccid2s_next;
575 if (seqp == hc->tx_seqh) { 566 if (seqp == hc->tx_seqh) {
576 seqp = hc->tx_seqh->cc 567 seqp = hc->tx_seqh->ccid2s_prev;
577 break; 568 break;
578 } 569 }
579 } 570 }
580 571
581 /* 572 /*
582 * In slow-start, cwnd can increase up 573 * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2
583 * packets per acknowledgement. Roundi 574 * packets per acknowledgement. Rounding up avoids that cwnd is not
584 * advanced when Ack Ratio is 1 and gi 575 * advanced when Ack Ratio is 1 and gives a slight edge otherwise.
585 */ 576 */
586 if (hc->tx_cwnd < hc->tx_ssthresh) 577 if (hc->tx_cwnd < hc->tx_ssthresh)
587 maxincr = DIV_ROUND_UP(dp->dcc 578 maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2);
588 579
589 /* go through all ack vectors */ 580 /* go through all ack vectors */
590 list_for_each_entry(avp, &hc->tx_av_ch 581 list_for_each_entry(avp, &hc->tx_av_chunks, node) {
591 /* go through this ack vector 582 /* go through this ack vector */
592 for (; avp->len--; avp->vec++) 583 for (; avp->len--; avp->vec++) {
593 u64 ackno_end_rl = SUB 584 u64 ackno_end_rl = SUB48(ackno,
594 585 dccp_ackvec_runlen(avp->vec));
595 586
596 ccid2_pr_debug("ackvec 587 ccid2_pr_debug("ackvec %llu |%u,%u|\n",
597 (unsign 588 (unsigned long long)ackno,
598 dccp_ac 589 dccp_ackvec_state(avp->vec) >> 6,
599 dccp_ac 590 dccp_ackvec_runlen(avp->vec));
600 /* if the seqno we are 591 /* if the seqno we are analyzing is larger than the
601 * current ackno, then 592 * current ackno, then move towards the tail of our
602 * seqnos. 593 * seqnos.
603 */ 594 */
604 while (after48(seqp->c 595 while (after48(seqp->ccid2s_seq, ackno)) {
605 if (seqp == hc 596 if (seqp == hc->tx_seqt) {
606 done = 597 done = 1;
607 break; 598 break;
608 } 599 }
609 seqp = seqp->c 600 seqp = seqp->ccid2s_prev;
610 } 601 }
611 if (done) 602 if (done)
612 break; 603 break;
613 604
614 /* check all seqnos in 605 /* check all seqnos in the range of the vector
615 * run length 606 * run length
616 */ 607 */
617 while (between48(seqp- 608 while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) {
618 const u8 state 609 const u8 state = dccp_ackvec_state(avp->vec);
619 610
620 /* new packet 611 /* new packet received or marked */
621 if (state != D 612 if (state != DCCPAV_NOT_RECEIVED &&
622 !seqp->cci 613 !seqp->ccid2s_acked) {
623 if (st 614 if (state == DCCPAV_ECN_MARKED)
624 615 ccid2_congestion_event(sk,
625 616 seqp);
626 else 617 else
627 618 ccid2_new_ack(sk, seqp,
628 619 &maxincr);
629 620
630 seqp-> 621 seqp->ccid2s_acked = 1;
631 ccid2_ 622 ccid2_pr_debug("Got ack for %llu\n",
632 623 (unsigned long long)seqp->ccid2s_seq);
633 hc->tx 624 hc->tx_pipe--;
634 } 625 }
635 if (seqp == hc 626 if (seqp == hc->tx_seqt) {
636 done = 627 done = 1;
637 break; 628 break;
638 } 629 }
639 seqp = seqp->c 630 seqp = seqp->ccid2s_prev;
640 } 631 }
641 if (done) 632 if (done)
642 break; 633 break;
643 634
644 ackno = SUB48(ackno_en 635 ackno = SUB48(ackno_end_rl, 1);
645 } 636 }
646 if (done) 637 if (done)
647 break; 638 break;
648 } 639 }
649 640
650 /* The state about what is acked shoul 641 /* The state about what is acked should be correct now
651 * Check for NUMDUPACK 642 * Check for NUMDUPACK
652 */ 643 */
653 seqp = hc->tx_seqt; 644 seqp = hc->tx_seqt;
654 while (before48(seqp->ccid2s_seq, hc-> 645 while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) {
655 seqp = seqp->ccid2s_next; 646 seqp = seqp->ccid2s_next;
656 if (seqp == hc->tx_seqh) { 647 if (seqp == hc->tx_seqh) {
657 seqp = hc->tx_seqh->cc 648 seqp = hc->tx_seqh->ccid2s_prev;
658 break; 649 break;
659 } 650 }
660 } 651 }
661 done = 0; 652 done = 0;
662 while (1) { 653 while (1) {
663 if (seqp->ccid2s_acked) { 654 if (seqp->ccid2s_acked) {
664 done++; 655 done++;
665 if (done == NUMDUPACK) 656 if (done == NUMDUPACK)
666 break; 657 break;
667 } 658 }
668 if (seqp == hc->tx_seqt) 659 if (seqp == hc->tx_seqt)
669 break; 660 break;
670 seqp = seqp->ccid2s_prev; 661 seqp = seqp->ccid2s_prev;
671 } 662 }
672 663
673 /* If there are at least 3 acknowledge 664 /* If there are at least 3 acknowledgements, anything unacknowledged
674 * below the last sequence number is c 665 * below the last sequence number is considered lost
675 */ 666 */
676 if (done == NUMDUPACK) { 667 if (done == NUMDUPACK) {
677 struct ccid2_seq *last_acked = 668 struct ccid2_seq *last_acked = seqp;
678 669
679 /* check for lost packets */ 670 /* check for lost packets */
680 while (1) { 671 while (1) {
681 if (!seqp->ccid2s_acke 672 if (!seqp->ccid2s_acked) {
682 ccid2_pr_debug 673 ccid2_pr_debug("Packet lost: %llu\n",
683 674 (unsigned long long)seqp->ccid2s_seq);
684 /* XXX need to 675 /* XXX need to traverse from tail -> head in
685 * order to de 676 * order to detect multiple congestion events in
686 * one ack vec 677 * one ack vector.
687 */ 678 */
688 ccid2_congesti 679 ccid2_congestion_event(sk, seqp);
689 hc->tx_pipe--; 680 hc->tx_pipe--;
690 } 681 }
691 if (seqp == hc->tx_seq 682 if (seqp == hc->tx_seqt)
692 break; 683 break;
693 seqp = seqp->ccid2s_pr 684 seqp = seqp->ccid2s_prev;
694 } 685 }
695 686
696 hc->tx_seqt = last_acked; 687 hc->tx_seqt = last_acked;
697 } 688 }
698 689
699 /* trim acked packets in tail */ 690 /* trim acked packets in tail */
700 while (hc->tx_seqt != hc->tx_seqh) { 691 while (hc->tx_seqt != hc->tx_seqh) {
701 if (!hc->tx_seqt->ccid2s_acked 692 if (!hc->tx_seqt->ccid2s_acked)
702 break; 693 break;
703 694
704 hc->tx_seqt = hc->tx_seqt->cci 695 hc->tx_seqt = hc->tx_seqt->ccid2s_next;
705 } 696 }
706 697
707 /* restart RTO timer if not all outsta 698 /* restart RTO timer if not all outstanding data has been acked */
708 if (hc->tx_pipe == 0) 699 if (hc->tx_pipe == 0)
709 sk_stop_timer(sk, &hc->tx_rtot 700 sk_stop_timer(sk, &hc->tx_rtotimer);
710 else 701 else
711 sk_reset_timer(sk, &hc->tx_rto 702 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto);
712 done: 703 done:
713 /* check if incoming Acks allow pendin 704 /* check if incoming Acks allow pending packets to be sent */
714 if (sender_was_blocked && !ccid2_cwnd_ 705 if (sender_was_blocked && !ccid2_cwnd_network_limited(hc))
715 dccp_tasklet_schedule(sk); !! 706 tasklet_schedule(&dccp_sk(sk)->dccps_xmitlet);
716 dccp_ackvec_parsed_cleanup(&hc->tx_av_ 707 dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
717 } 708 }
718 709
719 static int ccid2_hc_tx_init(struct ccid *ccid, 710 static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk)
720 { 711 {
721 struct ccid2_hc_tx_sock *hc = ccid_pri 712 struct ccid2_hc_tx_sock *hc = ccid_priv(ccid);
722 struct dccp_sock *dp = dccp_sk(sk); 713 struct dccp_sock *dp = dccp_sk(sk);
723 u32 max_ratio; 714 u32 max_ratio;
724 715
725 /* RFC 4341, 5: initialise ssthresh to 716 /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */
726 hc->tx_ssthresh = ~0U; 717 hc->tx_ssthresh = ~0U;
727 718
728 /* Use larger initial windows (RFC 434 719 /* Use larger initial windows (RFC 4341, section 5). */
729 hc->tx_cwnd = rfc3390_bytes_to_packets 720 hc->tx_cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache);
730 hc->tx_expected_wnd = hc->tx_cwnd; 721 hc->tx_expected_wnd = hc->tx_cwnd;
731 722
732 /* Make sure that Ack Ratio is enabled 723 /* Make sure that Ack Ratio is enabled and within bounds. */
733 max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 724 max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2);
734 if (dp->dccps_l_ack_ratio == 0 || dp-> 725 if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio)
735 dp->dccps_l_ack_ratio = max_ra 726 dp->dccps_l_ack_ratio = max_ratio;
736 727
737 /* XXX init ~ to window size... */ 728 /* XXX init ~ to window size... */
738 if (ccid2_hc_tx_alloc_seq(hc)) 729 if (ccid2_hc_tx_alloc_seq(hc))
739 return -ENOMEM; 730 return -ENOMEM;
740 731
741 hc->tx_rto = DCCP_TIMEOUT_INIT; 732 hc->tx_rto = DCCP_TIMEOUT_INIT;
742 hc->tx_rpdupack = -1; 733 hc->tx_rpdupack = -1;
743 hc->tx_last_cong = hc->tx_lsndtime = h 734 hc->tx_last_cong = hc->tx_lsndtime = hc->tx_cwnd_stamp = ccid2_jiffies32;
744 hc->tx_cwnd_used = 0; 735 hc->tx_cwnd_used = 0;
745 hc->sk = sk; !! 736 setup_timer(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire,
746 timer_setup(&hc->tx_rtotimer, ccid2_hc !! 737 (unsigned long)sk);
747 INIT_LIST_HEAD(&hc->tx_av_chunks); 738 INIT_LIST_HEAD(&hc->tx_av_chunks);
748 return 0; 739 return 0;
749 } 740 }
750 741
751 static void ccid2_hc_tx_exit(struct sock *sk) 742 static void ccid2_hc_tx_exit(struct sock *sk)
752 { 743 {
753 struct ccid2_hc_tx_sock *hc = ccid2_hc 744 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk);
754 int i; 745 int i;
755 746
756 sk_stop_timer(sk, &hc->tx_rtotimer); 747 sk_stop_timer(sk, &hc->tx_rtotimer);
757 748
758 for (i = 0; i < hc->tx_seqbufc; i++) 749 for (i = 0; i < hc->tx_seqbufc; i++)
759 kfree(hc->tx_seqbuf[i]); 750 kfree(hc->tx_seqbuf[i]);
760 hc->tx_seqbufc = 0; 751 hc->tx_seqbufc = 0;
761 dccp_ackvec_parsed_cleanup(&hc->tx_av_ 752 dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks);
762 } 753 }
763 754
764 static void ccid2_hc_rx_packet_recv(struct soc 755 static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
765 { 756 {
766 struct ccid2_hc_rx_sock *hc = ccid2_hc 757 struct ccid2_hc_rx_sock *hc = ccid2_hc_rx_sk(sk);
767 758
768 if (!dccp_data_packet(skb)) 759 if (!dccp_data_packet(skb))
769 return; 760 return;
770 761
771 if (++hc->rx_num_data_pkts >= dccp_sk( 762 if (++hc->rx_num_data_pkts >= dccp_sk(sk)->dccps_r_ack_ratio) {
772 dccp_send_ack(sk); 763 dccp_send_ack(sk);
773 hc->rx_num_data_pkts = 0; 764 hc->rx_num_data_pkts = 0;
774 } 765 }
775 } 766 }
776 767
777 struct ccid_operations ccid2_ops = { 768 struct ccid_operations ccid2_ops = {
778 .ccid_id = DCCPC_CCID 769 .ccid_id = DCCPC_CCID2,
779 .ccid_name = "TCP-like" 770 .ccid_name = "TCP-like",
780 .ccid_hc_tx_obj_size = sizeof(str 771 .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock),
781 .ccid_hc_tx_init = ccid2_hc_t 772 .ccid_hc_tx_init = ccid2_hc_tx_init,
782 .ccid_hc_tx_exit = ccid2_hc_t 773 .ccid_hc_tx_exit = ccid2_hc_tx_exit,
783 .ccid_hc_tx_send_packet = ccid2_hc_t 774 .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet,
784 .ccid_hc_tx_packet_sent = ccid2_hc_t 775 .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent,
785 .ccid_hc_tx_parse_options = ccid2_hc_t 776 .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options,
786 .ccid_hc_tx_packet_recv = ccid2_hc_t 777 .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv,
787 .ccid_hc_rx_obj_size = sizeof(str 778 .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock),
788 .ccid_hc_rx_packet_recv = ccid2_hc_r 779 .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv,
789 }; 780 };
790 781
791 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG 782 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG
792 module_param(ccid2_debug, bool, 0644); 783 module_param(ccid2_debug, bool, 0644);
793 MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 d 784 MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages");
794 #endif 785 #endif
795 786
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.