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 dccp_tasklet_schedule(struct sock *sk) 118 { 130 { 119 struct tasklet_struct *t = &dccp_sk(sk 131 struct tasklet_struct *t = &dccp_sk(sk)->dccps_xmitlet; 120 132 121 if (!test_and_set_bit(TASKLET_STATE_SC 133 if (!test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) { 122 sock_hold(sk); 134 sock_hold(sk); 123 __tasklet_schedule(t); 135 __tasklet_schedule(t); 124 } 136 } 125 } 137 } 126 138 127 static void ccid2_hc_tx_rto_expire(struct time 139 static void ccid2_hc_tx_rto_expire(struct timer_list *t) 128 { 140 { 129 struct ccid2_hc_tx_sock *hc = from_tim 141 struct ccid2_hc_tx_sock *hc = from_timer(hc, t, tx_rtotimer); 130 struct sock *sk = hc->sk; 142 struct sock *sk = hc->sk; 131 const bool sender_was_blocked = ccid2_ 143 const bool sender_was_blocked = ccid2_cwnd_network_limited(hc); 132 144 133 bh_lock_sock(sk); 145 bh_lock_sock(sk); 134 if (sock_owned_by_user(sk)) { 146 if (sock_owned_by_user(sk)) { 135 sk_reset_timer(sk, &hc->tx_rto 147 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + HZ / 5); 136 goto out; 148 goto out; 137 } 149 } 138 150 139 ccid2_pr_debug("RTO_EXPIRE\n"); 151 ccid2_pr_debug("RTO_EXPIRE\n"); 140 152 141 if (sk->sk_state == DCCP_CLOSED) 153 if (sk->sk_state == DCCP_CLOSED) 142 goto out; 154 goto out; 143 155 144 /* back-off timer */ 156 /* back-off timer */ 145 hc->tx_rto <<= 1; 157 hc->tx_rto <<= 1; 146 if (hc->tx_rto > DCCP_RTO_MAX) 158 if (hc->tx_rto > DCCP_RTO_MAX) 147 hc->tx_rto = DCCP_RTO_MAX; 159 hc->tx_rto = DCCP_RTO_MAX; 148 160 149 /* adjust pipe, cwnd etc */ 161 /* adjust pipe, cwnd etc */ 150 hc->tx_ssthresh = hc->tx_cwnd / 2; 162 hc->tx_ssthresh = hc->tx_cwnd / 2; 151 if (hc->tx_ssthresh < 2) 163 if (hc->tx_ssthresh < 2) 152 hc->tx_ssthresh = 2; 164 hc->tx_ssthresh = 2; 153 hc->tx_cwnd = 1; 165 hc->tx_cwnd = 1; 154 hc->tx_pipe = 0; 166 hc->tx_pipe = 0; 155 167 156 /* clear state about stuff we sent */ 168 /* clear state about stuff we sent */ 157 hc->tx_seqt = hc->tx_seqh; 169 hc->tx_seqt = hc->tx_seqh; 158 hc->tx_packets_acked = 0; 170 hc->tx_packets_acked = 0; 159 171 160 /* clear ack ratio state. */ 172 /* clear ack ratio state. */ 161 hc->tx_rpseq = 0; 173 hc->tx_rpseq = 0; 162 hc->tx_rpdupack = -1; 174 hc->tx_rpdupack = -1; 163 ccid2_change_l_ack_ratio(sk, 1); 175 ccid2_change_l_ack_ratio(sk, 1); 164 176 165 /* if we were blocked before, we may n 177 /* if we were blocked before, we may now send cwnd=1 packet */ 166 if (sender_was_blocked) 178 if (sender_was_blocked) 167 dccp_tasklet_schedule(sk); 179 dccp_tasklet_schedule(sk); 168 /* restart backed-off timer */ 180 /* restart backed-off timer */ 169 sk_reset_timer(sk, &hc->tx_rtotimer, j 181 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto); 170 out: 182 out: 171 bh_unlock_sock(sk); 183 bh_unlock_sock(sk); 172 sock_put(sk); 184 sock_put(sk); 173 } 185 } 174 186 175 /* 187 /* 176 * Congestion window validation (RFC 2861 188 * Congestion window validation (RFC 2861). 177 */ 189 */ 178 static bool ccid2_do_cwv = true; 190 static bool ccid2_do_cwv = true; 179 module_param(ccid2_do_cwv, bool, 0644); 191 module_param(ccid2_do_cwv, bool, 0644); 180 MODULE_PARM_DESC(ccid2_do_cwv, "Perform RFC286 192 MODULE_PARM_DESC(ccid2_do_cwv, "Perform RFC2861 Congestion Window Validation"); 181 193 182 /** 194 /** 183 * ccid2_update_used_window - Track how much 195 * 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 196 * 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 197 * packets may be in flight, to set the local Sequence Window value accordingly 189 * (RFC 4340, 7.5.2). The CWV mechanism is exp 198 * (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 199 * maximum-used window. We use an EWMA low-pass filter to filter out noise. 191 */ 200 */ 192 static void ccid2_update_used_window(struct cc 201 static void ccid2_update_used_window(struct ccid2_hc_tx_sock *hc, u32 new_wnd) 193 { 202 { 194 hc->tx_expected_wnd = (3 * hc->tx_expe 203 hc->tx_expected_wnd = (3 * hc->tx_expected_wnd + new_wnd) / 4; 195 } 204 } 196 205 197 /* This borrows the code of tcp_cwnd_applicati 206 /* This borrows the code of tcp_cwnd_application_limited() */ 198 static void ccid2_cwnd_application_limited(str 207 static void ccid2_cwnd_application_limited(struct sock *sk, const u32 now) 199 { 208 { 200 struct ccid2_hc_tx_sock *hc = ccid2_hc 209 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); 201 /* don't reduce cwnd below the initial 210 /* don't reduce cwnd below the initial window (IW) */ 202 u32 init_win = rfc3390_bytes_to_packet 211 u32 init_win = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache), 203 win_used = max(hc->tx_cwnd_used, i 212 win_used = max(hc->tx_cwnd_used, init_win); 204 213 205 if (win_used < hc->tx_cwnd) { 214 if (win_used < hc->tx_cwnd) { 206 hc->tx_ssthresh = max(hc->tx_s 215 hc->tx_ssthresh = max(hc->tx_ssthresh, 207 (hc->tx_c 216 (hc->tx_cwnd >> 1) + (hc->tx_cwnd >> 2)); 208 hc->tx_cwnd = (hc->tx_cwnd + w 217 hc->tx_cwnd = (hc->tx_cwnd + win_used) >> 1; 209 } 218 } 210 hc->tx_cwnd_used = 0; 219 hc->tx_cwnd_used = 0; 211 hc->tx_cwnd_stamp = now; 220 hc->tx_cwnd_stamp = now; 212 221 213 ccid2_check_l_ack_ratio(sk); 222 ccid2_check_l_ack_ratio(sk); 214 } 223 } 215 224 216 /* This borrows the code of tcp_cwnd_restart() 225 /* This borrows the code of tcp_cwnd_restart() */ 217 static void ccid2_cwnd_restart(struct sock *sk 226 static void ccid2_cwnd_restart(struct sock *sk, const u32 now) 218 { 227 { 219 struct ccid2_hc_tx_sock *hc = ccid2_hc 228 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); 220 u32 cwnd = hc->tx_cwnd, restart_cwnd, 229 u32 cwnd = hc->tx_cwnd, restart_cwnd, 221 iwnd = rfc3390_bytes_to_packets(dc 230 iwnd = rfc3390_bytes_to_packets(dccp_sk(sk)->dccps_mss_cache); 222 s32 delta = now - hc->tx_lsndtime; << 223 231 224 hc->tx_ssthresh = max(hc->tx_ssthresh, 232 hc->tx_ssthresh = max(hc->tx_ssthresh, (cwnd >> 1) + (cwnd >> 2)); 225 233 226 /* don't reduce cwnd below the initial 234 /* don't reduce cwnd below the initial window (IW) */ 227 restart_cwnd = min(cwnd, iwnd); 235 restart_cwnd = min(cwnd, iwnd); 228 !! 236 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); 237 hc->tx_cwnd = max(cwnd, restart_cwnd); >> 238 232 hc->tx_cwnd_stamp = now; 239 hc->tx_cwnd_stamp = now; 233 hc->tx_cwnd_used = 0; 240 hc->tx_cwnd_used = 0; 234 241 235 ccid2_check_l_ack_ratio(sk); 242 ccid2_check_l_ack_ratio(sk); 236 } 243 } 237 244 238 static void ccid2_hc_tx_packet_sent(struct soc 245 static void ccid2_hc_tx_packet_sent(struct sock *sk, unsigned int len) 239 { 246 { 240 struct dccp_sock *dp = dccp_sk(sk); 247 struct dccp_sock *dp = dccp_sk(sk); 241 struct ccid2_hc_tx_sock *hc = ccid2_hc 248 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); 242 const u32 now = ccid2_jiffies32; 249 const u32 now = ccid2_jiffies32; 243 struct ccid2_seq *next; 250 struct ccid2_seq *next; 244 251 245 /* slow-start after idle periods (RFC 252 /* slow-start after idle periods (RFC 2581, RFC 2861) */ 246 if (ccid2_do_cwv && !hc->tx_pipe && 253 if (ccid2_do_cwv && !hc->tx_pipe && 247 (s32)(now - hc->tx_lsndtime) >= hc 254 (s32)(now - hc->tx_lsndtime) >= hc->tx_rto) 248 ccid2_cwnd_restart(sk, now); 255 ccid2_cwnd_restart(sk, now); 249 256 250 hc->tx_lsndtime = now; 257 hc->tx_lsndtime = now; 251 hc->tx_pipe += 1; 258 hc->tx_pipe += 1; 252 259 253 /* see whether cwnd was fully used (RF 260 /* see whether cwnd was fully used (RFC 2861), update expected window */ 254 if (ccid2_cwnd_network_limited(hc)) { 261 if (ccid2_cwnd_network_limited(hc)) { 255 ccid2_update_used_window(hc, h 262 ccid2_update_used_window(hc, hc->tx_cwnd); 256 hc->tx_cwnd_used = 0; 263 hc->tx_cwnd_used = 0; 257 hc->tx_cwnd_stamp = now; 264 hc->tx_cwnd_stamp = now; 258 } else { 265 } else { 259 if (hc->tx_pipe > hc->tx_cwnd_ 266 if (hc->tx_pipe > hc->tx_cwnd_used) 260 hc->tx_cwnd_used = hc- 267 hc->tx_cwnd_used = hc->tx_pipe; 261 268 262 ccid2_update_used_window(hc, h 269 ccid2_update_used_window(hc, hc->tx_cwnd_used); 263 270 264 if (ccid2_do_cwv && (s32)(now 271 if (ccid2_do_cwv && (s32)(now - hc->tx_cwnd_stamp) >= hc->tx_rto) 265 ccid2_cwnd_application 272 ccid2_cwnd_application_limited(sk, now); 266 } 273 } 267 274 268 hc->tx_seqh->ccid2s_seq = dp->dccps_ 275 hc->tx_seqh->ccid2s_seq = dp->dccps_gss; 269 hc->tx_seqh->ccid2s_acked = 0; 276 hc->tx_seqh->ccid2s_acked = 0; 270 hc->tx_seqh->ccid2s_sent = now; 277 hc->tx_seqh->ccid2s_sent = now; 271 278 272 next = hc->tx_seqh->ccid2s_next; 279 next = hc->tx_seqh->ccid2s_next; 273 /* check if we need to alloc more spac 280 /* check if we need to alloc more space */ 274 if (next == hc->tx_seqt) { 281 if (next == hc->tx_seqt) { 275 if (ccid2_hc_tx_alloc_seq(hc)) 282 if (ccid2_hc_tx_alloc_seq(hc)) { 276 DCCP_CRIT("packet hist 283 DCCP_CRIT("packet history - out of memory!"); 277 /* FIXME: find a more 284 /* FIXME: find a more graceful way to bail out */ 278 return; 285 return; 279 } 286 } 280 next = hc->tx_seqh->ccid2s_nex 287 next = hc->tx_seqh->ccid2s_next; 281 BUG_ON(next == hc->tx_seqt); 288 BUG_ON(next == hc->tx_seqt); 282 } 289 } 283 hc->tx_seqh = next; 290 hc->tx_seqh = next; 284 291 285 ccid2_pr_debug("cwnd=%d pipe=%d\n", hc 292 ccid2_pr_debug("cwnd=%d pipe=%d\n", hc->tx_cwnd, hc->tx_pipe); 286 293 287 /* 294 /* 288 * FIXME: The code below is broken and 295 * FIXME: The code below is broken and the variables have been removed 289 * from the socket struct. The `acklos 296 * from the socket struct. The `ackloss' variable was always set to 0, 290 * and with arsent there are several p 297 * and with arsent there are several problems: 291 * (i) it doesn't just count the numb 298 * (i) it doesn't just count the number of Acks, but all sent packets; 292 * (ii) it is expressed in # of packe 299 * (ii) it is expressed in # of packets, not # of windows, so the 293 * comparison below uses the wrong fo 300 * comparison below uses the wrong formula: Appendix A of RFC 4341 294 * comes up with the number K = cwnd 301 * comes up with the number K = cwnd / (R^2 - R) of consecutive windows 295 * of data with no lost or marked Ack 302 * of data with no lost or marked Ack packets. If arsent were the # of 296 * consecutive Acks received without 303 * consecutive Acks received without loss, then Ack Ratio needs to be 297 * decreased by 1 when 304 * decreased by 1 when 298 * arsent >= K * cwnd / R 305 * arsent >= K * cwnd / R = cwnd^2 / (R^3 - R^2) 299 * where cwnd / R is the number of Ac 306 * where cwnd / R is the number of Acks received per window of data 300 * (cf. RFC 4341, App. A). The proble 307 * (cf. RFC 4341, App. A). The problems are that 301 * - arsent counts other packets as w 308 * - arsent counts other packets as well; 302 * - the comparison uses a formula di 309 * - the comparison uses a formula different from RFC 4341; 303 * - computing a cubic/quadratic equa 310 * - computing a cubic/quadratic equation each time is too complicated. 304 * Hence a different algorithm is nee 311 * Hence a different algorithm is needed. 305 */ 312 */ 306 #if 0 313 #if 0 307 /* Ack Ratio. Need to maintain a conc 314 /* Ack Ratio. Need to maintain a concept of how many windows we sent */ 308 hc->tx_arsent++; 315 hc->tx_arsent++; 309 /* We had an ack loss in this window.. 316 /* We had an ack loss in this window... */ 310 if (hc->tx_ackloss) { 317 if (hc->tx_ackloss) { 311 if (hc->tx_arsent >= hc->tx_cw 318 if (hc->tx_arsent >= hc->tx_cwnd) { 312 hc->tx_arsent = 0; 319 hc->tx_arsent = 0; 313 hc->tx_ackloss = 0; 320 hc->tx_ackloss = 0; 314 } 321 } 315 } else { 322 } else { 316 /* No acks lost up to now... * 323 /* No acks lost up to now... */ 317 /* decrease ack ratio if enoug 324 /* decrease ack ratio if enough packets were sent */ 318 if (dp->dccps_l_ack_ratio > 1) 325 if (dp->dccps_l_ack_ratio > 1) { 319 /* XXX don't calculate 326 /* XXX don't calculate denominator each time */ 320 int denom = dp->dccps_ 327 int denom = dp->dccps_l_ack_ratio * dp->dccps_l_ack_ratio - 321 dp->dccps_ 328 dp->dccps_l_ack_ratio; 322 329 323 denom = hc->tx_cwnd * 330 denom = hc->tx_cwnd * hc->tx_cwnd / denom; 324 331 325 if (hc->tx_arsent >= d 332 if (hc->tx_arsent >= denom) { 326 ccid2_change_l 333 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio - 1); 327 hc->tx_arsent 334 hc->tx_arsent = 0; 328 } 335 } 329 } else { 336 } else { 330 /* we can't increase a 337 /* we can't increase ack ratio further [1] */ 331 hc->tx_arsent = 0; /* 338 hc->tx_arsent = 0; /* or maybe set it to cwnd*/ 332 } 339 } 333 } 340 } 334 #endif 341 #endif 335 342 336 sk_reset_timer(sk, &hc->tx_rtotimer, j 343 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto); 337 344 338 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG 345 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG 339 do { 346 do { 340 struct ccid2_seq *seqp = hc->t 347 struct ccid2_seq *seqp = hc->tx_seqt; 341 348 342 while (seqp != hc->tx_seqh) { 349 while (seqp != hc->tx_seqh) { 343 ccid2_pr_debug("out se 350 ccid2_pr_debug("out seq=%llu acked=%d time=%u\n", 344 (unsign 351 (unsigned long long)seqp->ccid2s_seq, 345 seqp->c 352 seqp->ccid2s_acked, seqp->ccid2s_sent); 346 seqp = seqp->ccid2s_ne 353 seqp = seqp->ccid2s_next; 347 } 354 } 348 } while (0); 355 } while (0); 349 ccid2_pr_debug("=========\n"); 356 ccid2_pr_debug("=========\n"); 350 #endif 357 #endif 351 } 358 } 352 359 353 /** 360 /** 354 * ccid2_rtt_estimator - Sample RTT and comput 361 * 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 362 * This code is almost identical with TCP's tcp_rtt_estimator(), since 359 * - it has a higher sampling frequency (recom 363 * - it has a higher sampling frequency (recommended by RFC 1323), 360 * - the RTO does not collapse into RTT due to 364 * - the RTO does not collapse into RTT due to RTTVAR going towards zero, 361 * - it is simple (cf. more complex proposals 365 * - it is simple (cf. more complex proposals such as Eifel timer or research 362 * which suggests that the gain should be se 366 * which suggests that the gain should be set according to window size), 363 * - in tests it was found to work well with C 367 * - in tests it was found to work well with CCID2 [gerrit]. 364 */ 368 */ 365 static void ccid2_rtt_estimator(struct sock *s 369 static void ccid2_rtt_estimator(struct sock *sk, const long mrtt) 366 { 370 { 367 struct ccid2_hc_tx_sock *hc = ccid2_hc 371 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); 368 long m = mrtt ? : 1; 372 long m = mrtt ? : 1; 369 373 370 if (hc->tx_srtt == 0) { 374 if (hc->tx_srtt == 0) { 371 /* First measurement m */ 375 /* First measurement m */ 372 hc->tx_srtt = m << 3; 376 hc->tx_srtt = m << 3; 373 hc->tx_mdev = m << 1; 377 hc->tx_mdev = m << 1; 374 378 375 hc->tx_mdev_max = max(hc->tx_m 379 hc->tx_mdev_max = max(hc->tx_mdev, tcp_rto_min(sk)); 376 hc->tx_rttvar = hc->tx_mdev_ 380 hc->tx_rttvar = hc->tx_mdev_max; 377 381 378 hc->tx_rtt_seq = dccp_sk(sk)- 382 hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss; 379 } else { 383 } else { 380 /* Update scaled SRTT as SRTT 384 /* Update scaled SRTT as SRTT += 1/8 * (m - SRTT) */ 381 m -= (hc->tx_srtt >> 3); 385 m -= (hc->tx_srtt >> 3); 382 hc->tx_srtt += m; 386 hc->tx_srtt += m; 383 387 384 /* Similarly, update scaled md 388 /* Similarly, update scaled mdev with regard to |m| */ 385 if (m < 0) { 389 if (m < 0) { 386 m = -m; 390 m = -m; 387 m -= (hc->tx_mdev >> 2 391 m -= (hc->tx_mdev >> 2); 388 /* 392 /* 389 * This neutralises RT 393 * This neutralises RTO increase when RTT < SRTT - mdev 390 * (see P. Sarolahti, 394 * (see P. Sarolahti, A. Kuznetsov,"Congestion Control 391 * in Linux TCP", USEN 395 * in Linux TCP", USENIX 2002, pp. 49-62). 392 */ 396 */ 393 if (m > 0) 397 if (m > 0) 394 m >>= 3; 398 m >>= 3; 395 } else { 399 } else { 396 m -= (hc->tx_mdev >> 2 400 m -= (hc->tx_mdev >> 2); 397 } 401 } 398 hc->tx_mdev += m; 402 hc->tx_mdev += m; 399 403 400 if (hc->tx_mdev > hc->tx_mdev_ 404 if (hc->tx_mdev > hc->tx_mdev_max) { 401 hc->tx_mdev_max = hc-> 405 hc->tx_mdev_max = hc->tx_mdev; 402 if (hc->tx_mdev_max > 406 if (hc->tx_mdev_max > hc->tx_rttvar) 403 hc->tx_rttvar 407 hc->tx_rttvar = hc->tx_mdev_max; 404 } 408 } 405 409 406 /* 410 /* 407 * Decay RTTVAR at most once p 411 * Decay RTTVAR at most once per flight, exploiting that 408 * 1) pipe <= cwnd <= Sequenc 412 * 1) pipe <= cwnd <= Sequence_Window = W (RFC 4340, 7.5.2) 409 * 2) AWL = GSS-W+1 <= GAR <= 413 * 2) AWL = GSS-W+1 <= GAR <= GSS (RFC 4340, 7.5.1) 410 * GAR is a useful bound for F 414 * GAR is a useful bound for FlightSize = pipe. 411 * AWL is probably too low her 415 * AWL is probably too low here, as it over-estimates pipe. 412 */ 416 */ 413 if (after48(dccp_sk(sk)->dccps 417 if (after48(dccp_sk(sk)->dccps_gar, hc->tx_rtt_seq)) { 414 if (hc->tx_mdev_max < 418 if (hc->tx_mdev_max < hc->tx_rttvar) 415 hc->tx_rttvar 419 hc->tx_rttvar -= (hc->tx_rttvar - 416 420 hc->tx_mdev_max) >> 2; 417 hc->tx_rtt_seq = dccp 421 hc->tx_rtt_seq = dccp_sk(sk)->dccps_gss; 418 hc->tx_mdev_max = tcp_ 422 hc->tx_mdev_max = tcp_rto_min(sk); 419 } 423 } 420 } 424 } 421 425 422 /* 426 /* 423 * Set RTO from SRTT and RTTVAR 427 * Set RTO from SRTT and RTTVAR 424 * As in TCP, 4 * RTTVAR >= TCP_RTO_MI 428 * As in TCP, 4 * RTTVAR >= TCP_RTO_MIN, giving a minimum RTO of 200 ms. 425 * This agrees with RFC 4341, 5: 429 * This agrees with RFC 4341, 5: 426 * "Because DCCP does not retrans 430 * "Because DCCP does not retransmit data, DCCP does not require 427 * TCP's recommended minimum tim 431 * TCP's recommended minimum timeout of one second". 428 */ 432 */ 429 hc->tx_rto = (hc->tx_srtt >> 3) + hc-> 433 hc->tx_rto = (hc->tx_srtt >> 3) + hc->tx_rttvar; 430 434 431 if (hc->tx_rto > DCCP_RTO_MAX) 435 if (hc->tx_rto > DCCP_RTO_MAX) 432 hc->tx_rto = DCCP_RTO_MAX; 436 hc->tx_rto = DCCP_RTO_MAX; 433 } 437 } 434 438 435 static void ccid2_new_ack(struct sock *sk, str 439 static void ccid2_new_ack(struct sock *sk, struct ccid2_seq *seqp, 436 unsigned int *maxinc 440 unsigned int *maxincr) 437 { 441 { 438 struct ccid2_hc_tx_sock *hc = ccid2_hc 442 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); 439 struct dccp_sock *dp = dccp_sk(sk); 443 struct dccp_sock *dp = dccp_sk(sk); 440 int r_seq_used = hc->tx_cwnd / dp->dcc 444 int r_seq_used = hc->tx_cwnd / dp->dccps_l_ack_ratio; 441 445 442 if (hc->tx_cwnd < dp->dccps_l_seq_win 446 if (hc->tx_cwnd < dp->dccps_l_seq_win && 443 r_seq_used < dp->dccps_r_seq_win) 447 r_seq_used < dp->dccps_r_seq_win) { 444 if (hc->tx_cwnd < hc->tx_ssthr 448 if (hc->tx_cwnd < hc->tx_ssthresh) { 445 if (*maxincr > 0 && ++ 449 if (*maxincr > 0 && ++hc->tx_packets_acked >= 2) { 446 hc->tx_cwnd += 450 hc->tx_cwnd += 1; 447 *maxincr -= 451 *maxincr -= 1; 448 hc->tx_packets 452 hc->tx_packets_acked = 0; 449 } 453 } 450 } else if (++hc->tx_packets_ac 454 } else if (++hc->tx_packets_acked >= hc->tx_cwnd) { 451 hc->tx_cwnd += 1; 455 hc->tx_cwnd += 1; 452 hc->tx_packets_acked = 456 hc->tx_packets_acked = 0; 453 } 457 } 454 } 458 } 455 459 456 /* 460 /* 457 * Adjust the local sequence window an 461 * Adjust the local sequence window and the ack ratio to allow about 458 * 5 times the number of packets in th 462 * 5 times the number of packets in the network (RFC 4340 7.5.2) 459 */ 463 */ 460 if (r_seq_used * CCID2_WIN_CHANGE_FACT 464 if (r_seq_used * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_r_seq_win) 461 ccid2_change_l_ack_ratio(sk, d 465 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio * 2); 462 else if (r_seq_used * CCID2_WIN_CHANGE 466 else if (r_seq_used * CCID2_WIN_CHANGE_FACTOR < dp->dccps_r_seq_win/2) 463 ccid2_change_l_ack_ratio(sk, d 467 ccid2_change_l_ack_ratio(sk, dp->dccps_l_ack_ratio / 2 ? : 1U); 464 468 465 if (hc->tx_cwnd * CCID2_WIN_CHANGE_FAC 469 if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR >= dp->dccps_l_seq_win) 466 ccid2_change_l_seq_window(sk, 470 ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win * 2); 467 else if (hc->tx_cwnd * CCID2_WIN_CHANG 471 else if (hc->tx_cwnd * CCID2_WIN_CHANGE_FACTOR < dp->dccps_l_seq_win/2) 468 ccid2_change_l_seq_window(sk, 472 ccid2_change_l_seq_window(sk, dp->dccps_l_seq_win / 2); 469 473 470 /* 474 /* 471 * FIXME: RTT is sampled several times 475 * FIXME: RTT is sampled several times per acknowledgment (for each 472 * entry in the Ack Vector), instead o 476 * entry in the Ack Vector), instead of once per Ack (as in TCP SACK). 473 * This causes the RTT to be over-esti 477 * This causes the RTT to be over-estimated, since the older entries 474 * in the Ack Vector have earlier send 478 * in the Ack Vector have earlier sending times. 475 * The cleanest solution is to not use 479 * The cleanest solution is to not use the ccid2s_sent field at all 476 * and instead use DCCP timestamps: re 480 * and instead use DCCP timestamps: requires changes in other places. 477 */ 481 */ 478 ccid2_rtt_estimator(sk, ccid2_jiffies3 482 ccid2_rtt_estimator(sk, ccid2_jiffies32 - seqp->ccid2s_sent); 479 } 483 } 480 484 481 static void ccid2_congestion_event(struct sock 485 static void ccid2_congestion_event(struct sock *sk, struct ccid2_seq *seqp) 482 { 486 { 483 struct ccid2_hc_tx_sock *hc = ccid2_hc 487 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); 484 488 485 if ((s32)(seqp->ccid2s_sent - hc->tx_l 489 if ((s32)(seqp->ccid2s_sent - hc->tx_last_cong) < 0) { 486 ccid2_pr_debug("Multiple losse 490 ccid2_pr_debug("Multiple losses in an RTT---treating as one\n"); 487 return; 491 return; 488 } 492 } 489 493 490 hc->tx_last_cong = ccid2_jiffies32; 494 hc->tx_last_cong = ccid2_jiffies32; 491 495 492 hc->tx_cwnd = hc->tx_cwnd / 2 ? : 496 hc->tx_cwnd = hc->tx_cwnd / 2 ? : 1U; 493 hc->tx_ssthresh = max(hc->tx_cwnd, 2U 497 hc->tx_ssthresh = max(hc->tx_cwnd, 2U); 494 498 495 ccid2_check_l_ack_ratio(sk); 499 ccid2_check_l_ack_ratio(sk); 496 } 500 } 497 501 498 static int ccid2_hc_tx_parse_options(struct so 502 static int ccid2_hc_tx_parse_options(struct sock *sk, u8 packet_type, 499 u8 option 503 u8 option, u8 *optval, u8 optlen) 500 { 504 { 501 struct ccid2_hc_tx_sock *hc = ccid2_hc 505 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); 502 506 503 switch (option) { 507 switch (option) { 504 case DCCPO_ACK_VECTOR_0: 508 case DCCPO_ACK_VECTOR_0: 505 case DCCPO_ACK_VECTOR_1: 509 case DCCPO_ACK_VECTOR_1: 506 return dccp_ackvec_parsed_add( 510 return dccp_ackvec_parsed_add(&hc->tx_av_chunks, optval, optlen, 507 511 option - DCCPO_ACK_VECTOR_0); 508 } 512 } 509 return 0; 513 return 0; 510 } 514 } 511 515 512 static void ccid2_hc_tx_packet_recv(struct soc 516 static void ccid2_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb) 513 { 517 { 514 struct dccp_sock *dp = dccp_sk(sk); 518 struct dccp_sock *dp = dccp_sk(sk); 515 struct ccid2_hc_tx_sock *hc = ccid2_hc 519 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); 516 const bool sender_was_blocked = ccid2_ 520 const bool sender_was_blocked = ccid2_cwnd_network_limited(hc); 517 struct dccp_ackvec_parsed *avp; 521 struct dccp_ackvec_parsed *avp; 518 u64 ackno, seqno; 522 u64 ackno, seqno; 519 struct ccid2_seq *seqp; 523 struct ccid2_seq *seqp; 520 int done = 0; 524 int done = 0; 521 unsigned int maxincr = 0; 525 unsigned int maxincr = 0; 522 526 523 /* check reverse path congestion */ 527 /* check reverse path congestion */ 524 seqno = DCCP_SKB_CB(skb)->dccpd_seq; 528 seqno = DCCP_SKB_CB(skb)->dccpd_seq; 525 529 526 /* XXX this whole "algorithm" is broke 530 /* XXX this whole "algorithm" is broken. Need to fix it to keep track 527 * of the seqnos of the dupacks so tha 531 * of the seqnos of the dupacks so that rpseq and rpdupack are correct 528 * -sorbo. 532 * -sorbo. 529 */ 533 */ 530 /* need to bootstrap */ 534 /* need to bootstrap */ 531 if (hc->tx_rpdupack == -1) { 535 if (hc->tx_rpdupack == -1) { 532 hc->tx_rpdupack = 0; 536 hc->tx_rpdupack = 0; 533 hc->tx_rpseq = seqno; 537 hc->tx_rpseq = seqno; 534 } else { 538 } else { 535 /* check if packet is consecut 539 /* check if packet is consecutive */ 536 if (dccp_delta_seqno(hc->tx_rp 540 if (dccp_delta_seqno(hc->tx_rpseq, seqno) == 1) 537 hc->tx_rpseq = seqno; 541 hc->tx_rpseq = seqno; 538 /* it's a later packet */ 542 /* it's a later packet */ 539 else if (after48(seqno, hc->tx 543 else if (after48(seqno, hc->tx_rpseq)) { 540 hc->tx_rpdupack++; 544 hc->tx_rpdupack++; 541 545 542 /* check if we got eno 546 /* check if we got enough dupacks */ 543 if (hc->tx_rpdupack >= 547 if (hc->tx_rpdupack >= NUMDUPACK) { 544 hc->tx_rpdupac 548 hc->tx_rpdupack = -1; /* XXX lame */ 545 hc->tx_rpseq 549 hc->tx_rpseq = 0; 546 #ifdef __CCID2_COPES_GRACEFULLY_WITH_ACK_CONGE 550 #ifdef __CCID2_COPES_GRACEFULLY_WITH_ACK_CONGESTION_CONTROL__ 547 /* 551 /* 548 * FIXME: Ack 552 * FIXME: Ack Congestion Control is broken; in 549 * the current 553 * the current state instabilities occurred with 550 * Ack Ratios 554 * Ack Ratios greater than 1; causing hang-ups 551 * and long RT 555 * and long RTO timeouts. This needs to be fixed 552 * before open 556 * before opening up dynamic changes. -- gerrit 553 */ 557 */ 554 ccid2_change_l 558 ccid2_change_l_ack_ratio(sk, 2 * dp->dccps_l_ack_ratio); 555 #endif 559 #endif 556 } 560 } 557 } 561 } 558 } 562 } 559 563 560 /* check forward path congestion */ 564 /* check forward path congestion */ 561 if (dccp_packet_without_ack(skb)) 565 if (dccp_packet_without_ack(skb)) 562 return; 566 return; 563 567 564 /* still didn't send out new data pack 568 /* still didn't send out new data packets */ 565 if (hc->tx_seqh == hc->tx_seqt) 569 if (hc->tx_seqh == hc->tx_seqt) 566 goto done; 570 goto done; 567 571 568 ackno = DCCP_SKB_CB(skb)->dccpd_ack_se 572 ackno = DCCP_SKB_CB(skb)->dccpd_ack_seq; 569 if (after48(ackno, hc->tx_high_ack)) 573 if (after48(ackno, hc->tx_high_ack)) 570 hc->tx_high_ack = ackno; 574 hc->tx_high_ack = ackno; 571 575 572 seqp = hc->tx_seqt; 576 seqp = hc->tx_seqt; 573 while (before48(seqp->ccid2s_seq, ackn 577 while (before48(seqp->ccid2s_seq, ackno)) { 574 seqp = seqp->ccid2s_next; 578 seqp = seqp->ccid2s_next; 575 if (seqp == hc->tx_seqh) { 579 if (seqp == hc->tx_seqh) { 576 seqp = hc->tx_seqh->cc 580 seqp = hc->tx_seqh->ccid2s_prev; 577 break; 581 break; 578 } 582 } 579 } 583 } 580 584 581 /* 585 /* 582 * In slow-start, cwnd can increase up 586 * In slow-start, cwnd can increase up to a maximum of Ack Ratio/2 583 * packets per acknowledgement. Roundi 587 * packets per acknowledgement. Rounding up avoids that cwnd is not 584 * advanced when Ack Ratio is 1 and gi 588 * advanced when Ack Ratio is 1 and gives a slight edge otherwise. 585 */ 589 */ 586 if (hc->tx_cwnd < hc->tx_ssthresh) 590 if (hc->tx_cwnd < hc->tx_ssthresh) 587 maxincr = DIV_ROUND_UP(dp->dcc 591 maxincr = DIV_ROUND_UP(dp->dccps_l_ack_ratio, 2); 588 592 589 /* go through all ack vectors */ 593 /* go through all ack vectors */ 590 list_for_each_entry(avp, &hc->tx_av_ch 594 list_for_each_entry(avp, &hc->tx_av_chunks, node) { 591 /* go through this ack vector 595 /* go through this ack vector */ 592 for (; avp->len--; avp->vec++) 596 for (; avp->len--; avp->vec++) { 593 u64 ackno_end_rl = SUB 597 u64 ackno_end_rl = SUB48(ackno, 594 598 dccp_ackvec_runlen(avp->vec)); 595 599 596 ccid2_pr_debug("ackvec 600 ccid2_pr_debug("ackvec %llu |%u,%u|\n", 597 (unsign 601 (unsigned long long)ackno, 598 dccp_ac 602 dccp_ackvec_state(avp->vec) >> 6, 599 dccp_ac 603 dccp_ackvec_runlen(avp->vec)); 600 /* if the seqno we are 604 /* if the seqno we are analyzing is larger than the 601 * current ackno, then 605 * current ackno, then move towards the tail of our 602 * seqnos. 606 * seqnos. 603 */ 607 */ 604 while (after48(seqp->c 608 while (after48(seqp->ccid2s_seq, ackno)) { 605 if (seqp == hc 609 if (seqp == hc->tx_seqt) { 606 done = 610 done = 1; 607 break; 611 break; 608 } 612 } 609 seqp = seqp->c 613 seqp = seqp->ccid2s_prev; 610 } 614 } 611 if (done) 615 if (done) 612 break; 616 break; 613 617 614 /* check all seqnos in 618 /* check all seqnos in the range of the vector 615 * run length 619 * run length 616 */ 620 */ 617 while (between48(seqp- 621 while (between48(seqp->ccid2s_seq,ackno_end_rl,ackno)) { 618 const u8 state 622 const u8 state = dccp_ackvec_state(avp->vec); 619 623 620 /* new packet 624 /* new packet received or marked */ 621 if (state != D 625 if (state != DCCPAV_NOT_RECEIVED && 622 !seqp->cci 626 !seqp->ccid2s_acked) { 623 if (st 627 if (state == DCCPAV_ECN_MARKED) 624 628 ccid2_congestion_event(sk, 625 629 seqp); 626 else 630 else 627 631 ccid2_new_ack(sk, seqp, 628 632 &maxincr); 629 633 630 seqp-> 634 seqp->ccid2s_acked = 1; 631 ccid2_ 635 ccid2_pr_debug("Got ack for %llu\n", 632 636 (unsigned long long)seqp->ccid2s_seq); 633 hc->tx 637 hc->tx_pipe--; 634 } 638 } 635 if (seqp == hc 639 if (seqp == hc->tx_seqt) { 636 done = 640 done = 1; 637 break; 641 break; 638 } 642 } 639 seqp = seqp->c 643 seqp = seqp->ccid2s_prev; 640 } 644 } 641 if (done) 645 if (done) 642 break; 646 break; 643 647 644 ackno = SUB48(ackno_en 648 ackno = SUB48(ackno_end_rl, 1); 645 } 649 } 646 if (done) 650 if (done) 647 break; 651 break; 648 } 652 } 649 653 650 /* The state about what is acked shoul 654 /* The state about what is acked should be correct now 651 * Check for NUMDUPACK 655 * Check for NUMDUPACK 652 */ 656 */ 653 seqp = hc->tx_seqt; 657 seqp = hc->tx_seqt; 654 while (before48(seqp->ccid2s_seq, hc-> 658 while (before48(seqp->ccid2s_seq, hc->tx_high_ack)) { 655 seqp = seqp->ccid2s_next; 659 seqp = seqp->ccid2s_next; 656 if (seqp == hc->tx_seqh) { 660 if (seqp == hc->tx_seqh) { 657 seqp = hc->tx_seqh->cc 661 seqp = hc->tx_seqh->ccid2s_prev; 658 break; 662 break; 659 } 663 } 660 } 664 } 661 done = 0; 665 done = 0; 662 while (1) { 666 while (1) { 663 if (seqp->ccid2s_acked) { 667 if (seqp->ccid2s_acked) { 664 done++; 668 done++; 665 if (done == NUMDUPACK) 669 if (done == NUMDUPACK) 666 break; 670 break; 667 } 671 } 668 if (seqp == hc->tx_seqt) 672 if (seqp == hc->tx_seqt) 669 break; 673 break; 670 seqp = seqp->ccid2s_prev; 674 seqp = seqp->ccid2s_prev; 671 } 675 } 672 676 673 /* If there are at least 3 acknowledge 677 /* If there are at least 3 acknowledgements, anything unacknowledged 674 * below the last sequence number is c 678 * below the last sequence number is considered lost 675 */ 679 */ 676 if (done == NUMDUPACK) { 680 if (done == NUMDUPACK) { 677 struct ccid2_seq *last_acked = 681 struct ccid2_seq *last_acked = seqp; 678 682 679 /* check for lost packets */ 683 /* check for lost packets */ 680 while (1) { 684 while (1) { 681 if (!seqp->ccid2s_acke 685 if (!seqp->ccid2s_acked) { 682 ccid2_pr_debug 686 ccid2_pr_debug("Packet lost: %llu\n", 683 687 (unsigned long long)seqp->ccid2s_seq); 684 /* XXX need to 688 /* XXX need to traverse from tail -> head in 685 * order to de 689 * order to detect multiple congestion events in 686 * one ack vec 690 * one ack vector. 687 */ 691 */ 688 ccid2_congesti 692 ccid2_congestion_event(sk, seqp); 689 hc->tx_pipe--; 693 hc->tx_pipe--; 690 } 694 } 691 if (seqp == hc->tx_seq 695 if (seqp == hc->tx_seqt) 692 break; 696 break; 693 seqp = seqp->ccid2s_pr 697 seqp = seqp->ccid2s_prev; 694 } 698 } 695 699 696 hc->tx_seqt = last_acked; 700 hc->tx_seqt = last_acked; 697 } 701 } 698 702 699 /* trim acked packets in tail */ 703 /* trim acked packets in tail */ 700 while (hc->tx_seqt != hc->tx_seqh) { 704 while (hc->tx_seqt != hc->tx_seqh) { 701 if (!hc->tx_seqt->ccid2s_acked 705 if (!hc->tx_seqt->ccid2s_acked) 702 break; 706 break; 703 707 704 hc->tx_seqt = hc->tx_seqt->cci 708 hc->tx_seqt = hc->tx_seqt->ccid2s_next; 705 } 709 } 706 710 707 /* restart RTO timer if not all outsta 711 /* restart RTO timer if not all outstanding data has been acked */ 708 if (hc->tx_pipe == 0) 712 if (hc->tx_pipe == 0) 709 sk_stop_timer(sk, &hc->tx_rtot 713 sk_stop_timer(sk, &hc->tx_rtotimer); 710 else 714 else 711 sk_reset_timer(sk, &hc->tx_rto 715 sk_reset_timer(sk, &hc->tx_rtotimer, jiffies + hc->tx_rto); 712 done: 716 done: 713 /* check if incoming Acks allow pendin 717 /* check if incoming Acks allow pending packets to be sent */ 714 if (sender_was_blocked && !ccid2_cwnd_ 718 if (sender_was_blocked && !ccid2_cwnd_network_limited(hc)) 715 dccp_tasklet_schedule(sk); 719 dccp_tasklet_schedule(sk); 716 dccp_ackvec_parsed_cleanup(&hc->tx_av_ 720 dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks); 717 } 721 } 718 722 719 static int ccid2_hc_tx_init(struct ccid *ccid, 723 static int ccid2_hc_tx_init(struct ccid *ccid, struct sock *sk) 720 { 724 { 721 struct ccid2_hc_tx_sock *hc = ccid_pri 725 struct ccid2_hc_tx_sock *hc = ccid_priv(ccid); 722 struct dccp_sock *dp = dccp_sk(sk); 726 struct dccp_sock *dp = dccp_sk(sk); 723 u32 max_ratio; 727 u32 max_ratio; 724 728 725 /* RFC 4341, 5: initialise ssthresh to 729 /* RFC 4341, 5: initialise ssthresh to arbitrarily high (max) value */ 726 hc->tx_ssthresh = ~0U; 730 hc->tx_ssthresh = ~0U; 727 731 728 /* Use larger initial windows (RFC 434 732 /* Use larger initial windows (RFC 4341, section 5). */ 729 hc->tx_cwnd = rfc3390_bytes_to_packets 733 hc->tx_cwnd = rfc3390_bytes_to_packets(dp->dccps_mss_cache); 730 hc->tx_expected_wnd = hc->tx_cwnd; 734 hc->tx_expected_wnd = hc->tx_cwnd; 731 735 732 /* Make sure that Ack Ratio is enabled 736 /* Make sure that Ack Ratio is enabled and within bounds. */ 733 max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 737 max_ratio = DIV_ROUND_UP(hc->tx_cwnd, 2); 734 if (dp->dccps_l_ack_ratio == 0 || dp-> 738 if (dp->dccps_l_ack_ratio == 0 || dp->dccps_l_ack_ratio > max_ratio) 735 dp->dccps_l_ack_ratio = max_ra 739 dp->dccps_l_ack_ratio = max_ratio; 736 740 737 /* XXX init ~ to window size... */ 741 /* XXX init ~ to window size... */ 738 if (ccid2_hc_tx_alloc_seq(hc)) 742 if (ccid2_hc_tx_alloc_seq(hc)) 739 return -ENOMEM; 743 return -ENOMEM; 740 744 741 hc->tx_rto = DCCP_TIMEOUT_INIT; 745 hc->tx_rto = DCCP_TIMEOUT_INIT; 742 hc->tx_rpdupack = -1; 746 hc->tx_rpdupack = -1; 743 hc->tx_last_cong = hc->tx_lsndtime = h 747 hc->tx_last_cong = hc->tx_lsndtime = hc->tx_cwnd_stamp = ccid2_jiffies32; 744 hc->tx_cwnd_used = 0; 748 hc->tx_cwnd_used = 0; 745 hc->sk = sk; 749 hc->sk = sk; 746 timer_setup(&hc->tx_rtotimer, ccid2_hc 750 timer_setup(&hc->tx_rtotimer, ccid2_hc_tx_rto_expire, 0); 747 INIT_LIST_HEAD(&hc->tx_av_chunks); 751 INIT_LIST_HEAD(&hc->tx_av_chunks); 748 return 0; 752 return 0; 749 } 753 } 750 754 751 static void ccid2_hc_tx_exit(struct sock *sk) 755 static void ccid2_hc_tx_exit(struct sock *sk) 752 { 756 { 753 struct ccid2_hc_tx_sock *hc = ccid2_hc 757 struct ccid2_hc_tx_sock *hc = ccid2_hc_tx_sk(sk); 754 int i; 758 int i; 755 759 756 sk_stop_timer(sk, &hc->tx_rtotimer); 760 sk_stop_timer(sk, &hc->tx_rtotimer); 757 761 758 for (i = 0; i < hc->tx_seqbufc; i++) 762 for (i = 0; i < hc->tx_seqbufc; i++) 759 kfree(hc->tx_seqbuf[i]); 763 kfree(hc->tx_seqbuf[i]); 760 hc->tx_seqbufc = 0; 764 hc->tx_seqbufc = 0; 761 dccp_ackvec_parsed_cleanup(&hc->tx_av_ 765 dccp_ackvec_parsed_cleanup(&hc->tx_av_chunks); 762 } 766 } 763 767 764 static void ccid2_hc_rx_packet_recv(struct soc 768 static void ccid2_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb) 765 { 769 { 766 struct ccid2_hc_rx_sock *hc = ccid2_hc 770 struct ccid2_hc_rx_sock *hc = ccid2_hc_rx_sk(sk); 767 771 768 if (!dccp_data_packet(skb)) 772 if (!dccp_data_packet(skb)) 769 return; 773 return; 770 774 771 if (++hc->rx_num_data_pkts >= dccp_sk( 775 if (++hc->rx_num_data_pkts >= dccp_sk(sk)->dccps_r_ack_ratio) { 772 dccp_send_ack(sk); 776 dccp_send_ack(sk); 773 hc->rx_num_data_pkts = 0; 777 hc->rx_num_data_pkts = 0; 774 } 778 } 775 } 779 } 776 780 777 struct ccid_operations ccid2_ops = { 781 struct ccid_operations ccid2_ops = { 778 .ccid_id = DCCPC_CCID 782 .ccid_id = DCCPC_CCID2, 779 .ccid_name = "TCP-like" 783 .ccid_name = "TCP-like", 780 .ccid_hc_tx_obj_size = sizeof(str 784 .ccid_hc_tx_obj_size = sizeof(struct ccid2_hc_tx_sock), 781 .ccid_hc_tx_init = ccid2_hc_t 785 .ccid_hc_tx_init = ccid2_hc_tx_init, 782 .ccid_hc_tx_exit = ccid2_hc_t 786 .ccid_hc_tx_exit = ccid2_hc_tx_exit, 783 .ccid_hc_tx_send_packet = ccid2_hc_t 787 .ccid_hc_tx_send_packet = ccid2_hc_tx_send_packet, 784 .ccid_hc_tx_packet_sent = ccid2_hc_t 788 .ccid_hc_tx_packet_sent = ccid2_hc_tx_packet_sent, 785 .ccid_hc_tx_parse_options = ccid2_hc_t 789 .ccid_hc_tx_parse_options = ccid2_hc_tx_parse_options, 786 .ccid_hc_tx_packet_recv = ccid2_hc_t 790 .ccid_hc_tx_packet_recv = ccid2_hc_tx_packet_recv, 787 .ccid_hc_rx_obj_size = sizeof(str 791 .ccid_hc_rx_obj_size = sizeof(struct ccid2_hc_rx_sock), 788 .ccid_hc_rx_packet_recv = ccid2_hc_r 792 .ccid_hc_rx_packet_recv = ccid2_hc_rx_packet_recv, 789 }; 793 }; 790 794 791 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG 795 #ifdef CONFIG_IP_DCCP_CCID2_DEBUG 792 module_param(ccid2_debug, bool, 0644); 796 module_param(ccid2_debug, bool, 0644); 793 MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 d 797 MODULE_PARM_DESC(ccid2_debug, "Enable CCID-2 debug messages"); 794 #endif 798 #endif 795 799
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.