1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2009-2010 5 * Swinburne University of Technology, Melbourne, Australia 6 * Copyright (c) 2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010-2011 The FreeBSD Foundation 8 * All rights reserved. 9 * 10 * This software was developed at the Centre for Advanced Internet 11 * Architectures, Swinburne University of Technology, by David Hayes and 12 * Lawrence Stewart, made possible in part by a grant from the Cisco University 13 * Research Program Fund at Community Foundation Silicon Valley. 14 * 15 * Portions of this software were developed at the Centre for Advanced Internet 16 * Architectures, Swinburne University of Technology, Melbourne, Australia by 17 * David Hayes under sponsorship from the FreeBSD Foundation. 18 * 19 * Redistribution and use in source and binary forms, with or without 20 * modification, are permitted provided that the following conditions 21 * are met: 22 * 1. Redistributions of source code must retain the above copyright 23 * notice, this list of conditions and the following disclaimer. 24 * 2. Redistributions in binary form must reproduce the above copyright 25 * notice, this list of conditions and the following disclaimer in the 26 * documentation and/or other materials provided with the distribution. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 31 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 38 * SUCH DAMAGE. 39 */ 40 41 /* 42 * An implementation of the Vegas congestion control algorithm for FreeBSD, 43 * based on L. S. Brakmo and L. L. Peterson, "TCP Vegas: end to end congestion 44 * avoidance on a global internet", IEEE J. Sel. Areas Commun., vol. 13, no. 8, 45 * pp. 1465-1480, Oct. 1995. The original Vegas duplicate ack policy has not 46 * been implemented, since clock ticks are not as coarse as they were (i.e. 47 * 500ms) when Vegas was designed. Also, packets are timed once per RTT as in 48 * the original paper. 49 * 50 * Originally released as part of the NewTCP research project at Swinburne 51 * University of Technology's Centre for Advanced Internet Architectures, 52 * Melbourne, Australia, which was made possible in part by a grant from the 53 * Cisco University Research Program Fund at Community Foundation Silicon 54 * Valley. More details are available at: 55 * http://caia.swin.edu.au/urp/newtcp/ 56 */ 57 58 #include <sys/param.h> 59 #include <sys/kernel.h> 60 #include <sys/khelp.h> 61 #include <sys/malloc.h> 62 #include <sys/module.h> 63 #include <sys/queue.h> 64 #include <sys/socket.h> 65 #include <sys/socketvar.h> 66 #include <sys/sysctl.h> 67 #include <sys/systm.h> 68 69 #include <net/vnet.h> 70 71 #include <net/route.h> 72 #include <net/route/nhop.h> 73 74 #include <netinet/in_pcb.h> 75 #include <netinet/tcp.h> 76 #include <netinet/tcp_timer.h> 77 #include <netinet/tcp_var.h> 78 #include <netinet/cc/cc.h> 79 #include <netinet/cc/cc_module.h> 80 81 #include <netinet/khelp/h_ertt.h> 82 83 /* 84 * Private signal type for rate based congestion signal. 85 * See <netinet/cc.h> for appropriate bit-range to use for private signals. 86 */ 87 #define CC_VEGAS_RATE 0x04000000 88 89 static void vegas_ack_received(struct cc_var *ccv, ccsignal_t ack_type); 90 static void vegas_cb_destroy(struct cc_var *ccv); 91 static int vegas_cb_init(struct cc_var *ccv, void *ptr); 92 static void vegas_cong_signal(struct cc_var *ccv, ccsignal_t signal_type); 93 static void vegas_conn_init(struct cc_var *ccv); 94 static int vegas_mod_init(void); 95 static size_t vegas_data_sz(void); 96 97 struct vegas { 98 int slow_start_toggle; 99 }; 100 101 static int32_t ertt_id; 102 103 VNET_DEFINE_STATIC(uint32_t, vegas_alpha) = 1; 104 VNET_DEFINE_STATIC(uint32_t, vegas_beta) = 3; 105 #define V_vegas_alpha VNET(vegas_alpha) 106 #define V_vegas_beta VNET(vegas_beta) 107 108 struct cc_algo vegas_cc_algo = { 109 .name = "vegas", 110 .ack_received = vegas_ack_received, 111 .cb_destroy = vegas_cb_destroy, 112 .cb_init = vegas_cb_init, 113 .cong_signal = vegas_cong_signal, 114 .conn_init = vegas_conn_init, 115 .mod_init = vegas_mod_init, 116 .cc_data_sz = vegas_data_sz, 117 .after_idle = newreno_cc_after_idle, 118 .post_recovery = newreno_cc_post_recovery, 119 }; 120 121 /* 122 * The vegas window adjustment is done once every RTT, as indicated by the 123 * ERTT_NEW_MEASUREMENT flag. This flag is reset once the new measurement data 124 * has been used. 125 */ 126 static void 127 vegas_ack_received(struct cc_var *ccv, ccsignal_t ack_type) 128 { 129 struct ertt *e_t; 130 struct vegas *vegas_data; 131 long actual_tx_rate, expected_tx_rate, ndiff; 132 133 e_t = khelp_get_osd(&CCV(ccv, t_osd), ertt_id); 134 vegas_data = ccv->cc_data; 135 136 if (e_t->flags & ERTT_NEW_MEASUREMENT) { /* Once per RTT. */ 137 if (e_t->minrtt && e_t->markedpkt_rtt) { 138 expected_tx_rate = e_t->marked_snd_cwnd / e_t->minrtt; 139 actual_tx_rate = e_t->bytes_tx_in_marked_rtt / 140 e_t->markedpkt_rtt; 141 ndiff = (expected_tx_rate - actual_tx_rate) * 142 e_t->minrtt / CCV(ccv, t_maxseg); 143 144 if (ndiff < V_vegas_alpha) { 145 if (CCV(ccv, snd_cwnd) <= 146 CCV(ccv, snd_ssthresh)) { 147 vegas_data->slow_start_toggle = 148 vegas_data->slow_start_toggle ? 149 0 : 1; 150 } else { 151 vegas_data->slow_start_toggle = 0; 152 CCV(ccv, snd_cwnd) = 153 min(CCV(ccv, snd_cwnd) + 154 CCV(ccv, t_maxseg), 155 TCP_MAXWIN << CCV(ccv, snd_scale)); 156 } 157 } else if (ndiff > V_vegas_beta) { 158 /* Rate-based congestion. */ 159 vegas_cong_signal(ccv, CC_VEGAS_RATE); 160 vegas_data->slow_start_toggle = 0; 161 } 162 } 163 e_t->flags &= ~ERTT_NEW_MEASUREMENT; 164 } 165 166 if (vegas_data->slow_start_toggle) 167 newreno_cc_ack_received(ccv, ack_type); 168 } 169 170 static void 171 vegas_cb_destroy(struct cc_var *ccv) 172 { 173 free(ccv->cc_data, M_CC_MEM); 174 } 175 176 static size_t 177 vegas_data_sz(void) 178 { 179 return (sizeof(struct vegas)); 180 } 181 182 static int 183 vegas_cb_init(struct cc_var *ccv, void *ptr) 184 { 185 struct vegas *vegas_data; 186 187 INP_WLOCK_ASSERT(tptoinpcb(ccv->tp)); 188 if (ptr == NULL) { 189 vegas_data = malloc(sizeof(struct vegas), M_CC_MEM, M_NOWAIT); 190 if (vegas_data == NULL) 191 return (ENOMEM); 192 } else 193 vegas_data = ptr; 194 195 vegas_data->slow_start_toggle = 1; 196 ccv->cc_data = vegas_data; 197 198 return (0); 199 } 200 201 /* 202 * If congestion has been triggered triggered by the Vegas measured rates, it is 203 * handled here, otherwise it falls back to newreno's congestion handling. 204 */ 205 static void 206 vegas_cong_signal(struct cc_var *ccv, ccsignal_t signal_type) 207 { 208 struct vegas *vegas_data; 209 int presignalrecov; 210 211 vegas_data = ccv->cc_data; 212 213 if (IN_RECOVERY(CCV(ccv, t_flags))) 214 presignalrecov = 1; 215 else 216 presignalrecov = 0; 217 218 switch((int)signal_type) { 219 case CC_VEGAS_RATE: 220 if (!IN_RECOVERY(CCV(ccv, t_flags))) { 221 CCV(ccv, snd_cwnd) = max(2 * CCV(ccv, t_maxseg), 222 CCV(ccv, snd_cwnd) - CCV(ccv, t_maxseg)); 223 if (CCV(ccv, snd_cwnd) < CCV(ccv, snd_ssthresh)) 224 /* Exit slow start. */ 225 CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); 226 } 227 break; 228 229 default: 230 newreno_cc_cong_signal(ccv, signal_type); 231 break; 232 } 233 234 if (IN_RECOVERY(CCV(ccv, t_flags)) && !presignalrecov) 235 vegas_data->slow_start_toggle = 236 (CCV(ccv, snd_cwnd) < CCV(ccv, snd_ssthresh)) ? 1 : 0; 237 } 238 239 static void 240 vegas_conn_init(struct cc_var *ccv) 241 { 242 struct vegas *vegas_data; 243 244 vegas_data = ccv->cc_data; 245 vegas_data->slow_start_toggle = 1; 246 } 247 248 static int 249 vegas_mod_init(void) 250 { 251 ertt_id = khelp_get_id("ertt"); 252 if (ertt_id <= 0) { 253 printf("%s: h_ertt module not found\n", __func__); 254 return (ENOENT); 255 } 256 return (0); 257 } 258 259 static int 260 vegas_alpha_handler(SYSCTL_HANDLER_ARGS) 261 { 262 int error; 263 uint32_t new; 264 265 new = V_vegas_alpha; 266 error = sysctl_handle_int(oidp, &new, 0, req); 267 if (error == 0 && req->newptr != NULL) { 268 if (new == 0 || new > V_vegas_beta) 269 error = EINVAL; 270 else 271 V_vegas_alpha = new; 272 } 273 274 return (error); 275 } 276 277 static int 278 vegas_beta_handler(SYSCTL_HANDLER_ARGS) 279 { 280 int error; 281 uint32_t new; 282 283 new = V_vegas_beta; 284 error = sysctl_handle_int(oidp, &new, 0, req); 285 if (error == 0 && req->newptr != NULL) { 286 if (new == 0 || new < V_vegas_alpha) 287 error = EINVAL; 288 else 289 V_vegas_beta = new; 290 } 291 292 return (error); 293 } 294 295 SYSCTL_DECL(_net_inet_tcp_cc_vegas); 296 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, vegas, 297 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 298 "Vegas related settings"); 299 300 SYSCTL_PROC(_net_inet_tcp_cc_vegas, OID_AUTO, alpha, 301 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 302 &VNET_NAME(vegas_alpha), 1, &vegas_alpha_handler, "IU", 303 "vegas alpha, specified as number of \"buffers\" (0 < alpha < beta)"); 304 305 SYSCTL_PROC(_net_inet_tcp_cc_vegas, OID_AUTO, beta, 306 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 307 &VNET_NAME(vegas_beta), 3, &vegas_beta_handler, "IU", 308 "vegas beta, specified as number of \"buffers\" (0 < alpha < beta)"); 309 310 DECLARE_CC_MODULE(vegas, &vegas_cc_algo); 311 MODULE_VERSION(vegas, 2); 312 MODULE_DEPEND(vegas, ertt, 1, 1, 1); 313