1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 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/cdefs.h> 59 __FBSDID("$FreeBSD$"); 60 61 #include <sys/param.h> 62 #include <sys/kernel.h> 63 #include <sys/khelp.h> 64 #include <sys/malloc.h> 65 #include <sys/module.h> 66 #include <sys/queue.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/sysctl.h> 70 #include <sys/systm.h> 71 72 #include <net/vnet.h> 73 74 #include <net/route.h> 75 #include <net/route/nhop.h> 76 77 #include <netinet/in_pcb.h> 78 #include <netinet/tcp.h> 79 #include <netinet/tcp_timer.h> 80 #include <netinet/tcp_var.h> 81 #include <netinet/cc/cc.h> 82 #include <netinet/cc/cc_module.h> 83 84 #include <netinet/khelp/h_ertt.h> 85 86 /* 87 * Private signal type for rate based congestion signal. 88 * See <netinet/cc.h> for appropriate bit-range to use for private signals. 89 */ 90 #define CC_VEGAS_RATE 0x01000000 91 92 static void vegas_ack_received(struct cc_var *ccv, uint16_t ack_type); 93 static void vegas_cb_destroy(struct cc_var *ccv); 94 static int vegas_cb_init(struct cc_var *ccv, void *ptr); 95 static void vegas_cong_signal(struct cc_var *ccv, uint32_t signal_type); 96 static void vegas_conn_init(struct cc_var *ccv); 97 static int vegas_mod_init(void); 98 static size_t vegas_data_sz(void); 99 100 struct vegas { 101 int slow_start_toggle; 102 }; 103 104 static int32_t ertt_id; 105 106 VNET_DEFINE_STATIC(uint32_t, vegas_alpha) = 1; 107 VNET_DEFINE_STATIC(uint32_t, vegas_beta) = 3; 108 #define V_vegas_alpha VNET(vegas_alpha) 109 #define V_vegas_beta VNET(vegas_beta) 110 111 struct cc_algo vegas_cc_algo = { 112 .name = "vegas", 113 .ack_received = vegas_ack_received, 114 .cb_destroy = vegas_cb_destroy, 115 .cb_init = vegas_cb_init, 116 .cong_signal = vegas_cong_signal, 117 .conn_init = vegas_conn_init, 118 .mod_init = vegas_mod_init, 119 .cc_data_sz = vegas_data_sz, 120 .after_idle = newreno_cc_after_idle, 121 .post_recovery = newreno_cc_post_recovery, 122 }; 123 124 /* 125 * The vegas window adjustment is done once every RTT, as indicated by the 126 * ERTT_NEW_MEASUREMENT flag. This flag is reset once the new measurement data 127 * has been used. 128 */ 129 static void 130 vegas_ack_received(struct cc_var *ccv, uint16_t ack_type) 131 { 132 struct ertt *e_t; 133 struct vegas *vegas_data; 134 long actual_tx_rate, expected_tx_rate, ndiff; 135 136 e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); 137 vegas_data = ccv->cc_data; 138 139 if (e_t->flags & ERTT_NEW_MEASUREMENT) { /* Once per RTT. */ 140 if (e_t->minrtt && e_t->markedpkt_rtt) { 141 expected_tx_rate = e_t->marked_snd_cwnd / e_t->minrtt; 142 actual_tx_rate = e_t->bytes_tx_in_marked_rtt / 143 e_t->markedpkt_rtt; 144 ndiff = (expected_tx_rate - actual_tx_rate) * 145 e_t->minrtt / CCV(ccv, t_maxseg); 146 147 if (ndiff < V_vegas_alpha) { 148 if (CCV(ccv, snd_cwnd) <= 149 CCV(ccv, snd_ssthresh)) { 150 vegas_data->slow_start_toggle = 151 vegas_data->slow_start_toggle ? 152 0 : 1; 153 } else { 154 vegas_data->slow_start_toggle = 0; 155 CCV(ccv, snd_cwnd) = 156 min(CCV(ccv, snd_cwnd) + 157 CCV(ccv, t_maxseg), 158 TCP_MAXWIN << CCV(ccv, snd_scale)); 159 } 160 } else if (ndiff > V_vegas_beta) { 161 /* Rate-based congestion. */ 162 vegas_cong_signal(ccv, CC_VEGAS_RATE); 163 vegas_data->slow_start_toggle = 0; 164 } 165 } 166 e_t->flags &= ~ERTT_NEW_MEASUREMENT; 167 } 168 169 if (vegas_data->slow_start_toggle) 170 newreno_cc_ack_received(ccv, ack_type); 171 } 172 173 static void 174 vegas_cb_destroy(struct cc_var *ccv) 175 { 176 free(ccv->cc_data, M_CC_MEM); 177 } 178 179 static size_t 180 vegas_data_sz(void) 181 { 182 return (sizeof(struct vegas)); 183 } 184 185 static int 186 vegas_cb_init(struct cc_var *ccv, void *ptr) 187 { 188 struct vegas *vegas_data; 189 190 INP_WLOCK_ASSERT(ccv->ccvc.tcp->t_inpcb); 191 if (ptr == NULL) { 192 vegas_data = malloc(sizeof(struct vegas), M_CC_MEM, M_NOWAIT); 193 if (vegas_data == NULL) 194 return (ENOMEM); 195 } else 196 vegas_data = ptr; 197 198 vegas_data->slow_start_toggle = 1; 199 ccv->cc_data = vegas_data; 200 201 return (0); 202 } 203 204 /* 205 * If congestion has been triggered triggered by the Vegas measured rates, it is 206 * handled here, otherwise it falls back to newreno's congestion handling. 207 */ 208 static void 209 vegas_cong_signal(struct cc_var *ccv, uint32_t signal_type) 210 { 211 struct vegas *vegas_data; 212 int presignalrecov; 213 214 vegas_data = ccv->cc_data; 215 216 if (IN_RECOVERY(CCV(ccv, t_flags))) 217 presignalrecov = 1; 218 else 219 presignalrecov = 0; 220 221 switch(signal_type) { 222 case CC_VEGAS_RATE: 223 if (!IN_RECOVERY(CCV(ccv, t_flags))) { 224 CCV(ccv, snd_cwnd) = max(2 * CCV(ccv, t_maxseg), 225 CCV(ccv, snd_cwnd) - CCV(ccv, t_maxseg)); 226 if (CCV(ccv, snd_cwnd) < CCV(ccv, snd_ssthresh)) 227 /* Exit slow start. */ 228 CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); 229 } 230 break; 231 232 default: 233 newreno_cc_cong_signal(ccv, signal_type); 234 } 235 236 if (IN_RECOVERY(CCV(ccv, t_flags)) && !presignalrecov) 237 vegas_data->slow_start_toggle = 238 (CCV(ccv, snd_cwnd) < CCV(ccv, snd_ssthresh)) ? 1 : 0; 239 } 240 241 static void 242 vegas_conn_init(struct cc_var *ccv) 243 { 244 struct vegas *vegas_data; 245 246 vegas_data = ccv->cc_data; 247 vegas_data->slow_start_toggle = 1; 248 } 249 250 static int 251 vegas_mod_init(void) 252 { 253 ertt_id = khelp_get_id("ertt"); 254 if (ertt_id <= 0) { 255 printf("%s: h_ertt module not found\n", __func__); 256 return (ENOENT); 257 } 258 return (0); 259 } 260 261 static int 262 vegas_alpha_handler(SYSCTL_HANDLER_ARGS) 263 { 264 int error; 265 uint32_t new; 266 267 new = V_vegas_alpha; 268 error = sysctl_handle_int(oidp, &new, 0, req); 269 if (error == 0 && req->newptr != NULL) { 270 if (new == 0 || new > V_vegas_beta) 271 error = EINVAL; 272 else 273 V_vegas_alpha = new; 274 } 275 276 return (error); 277 } 278 279 static int 280 vegas_beta_handler(SYSCTL_HANDLER_ARGS) 281 { 282 int error; 283 uint32_t new; 284 285 new = V_vegas_beta; 286 error = sysctl_handle_int(oidp, &new, 0, req); 287 if (error == 0 && req->newptr != NULL) { 288 if (new == 0 || new < V_vegas_alpha) 289 error = EINVAL; 290 else 291 V_vegas_beta = new; 292 } 293 294 return (error); 295 } 296 297 SYSCTL_DECL(_net_inet_tcp_cc_vegas); 298 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, vegas, 299 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 300 "Vegas related settings"); 301 302 SYSCTL_PROC(_net_inet_tcp_cc_vegas, OID_AUTO, alpha, 303 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 304 &VNET_NAME(vegas_alpha), 1, &vegas_alpha_handler, "IU", 305 "vegas alpha, specified as number of \"buffers\" (0 < alpha < beta)"); 306 307 SYSCTL_PROC(_net_inet_tcp_cc_vegas, OID_AUTO, beta, 308 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 309 &VNET_NAME(vegas_beta), 3, &vegas_beta_handler, "IU", 310 "vegas beta, specified as number of \"buffers\" (0 < alpha < beta)"); 311 312 DECLARE_CC_MODULE(vegas, &vegas_cc_algo); 313 MODULE_VERSION(vegas, 2); 314 MODULE_DEPEND(vegas, ertt, 1, 1, 1); 315