1 /*- 2 * Copyright (c) 2009-2010 3 * Swinburne University of Technology, Melbourne, Australia 4 * Copyright (c) 2010 Lawrence Stewart <lstewart@freebsd.org> 5 * Copyright (c) 2010-2011 The FreeBSD Foundation 6 * All rights reserved. 7 * 8 * This software was developed at the Centre for Advanced Internet 9 * Architectures, Swinburne University of Technology, by David Hayes and 10 * Lawrence Stewart, made possible in part by a grant from the Cisco University 11 * Research Program Fund at Community Foundation Silicon Valley. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, Melbourne, Australia by 15 * David Hayes under sponsorship from the FreeBSD Foundation. 16 * 17 * Redistribution and use in source and binary forms, with or without 18 * modification, are permitted provided that the following conditions 19 * are met: 20 * 1. Redistributions of source code must retain the above copyright 21 * notice, this list of conditions and the following disclaimer. 22 * 2. Redistributions in binary form must reproduce the above copyright 23 * notice, this list of conditions and the following disclaimer in the 24 * documentation and/or other materials provided with the distribution. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 */ 38 39 /* 40 * An implementation of the Vegas congestion control algorithm for FreeBSD, 41 * based on L. S. Brakmo and L. L. Peterson, "TCP Vegas: end to end congestion 42 * avoidance on a global internet", IEEE J. Sel. Areas Commun., vol. 13, no. 8, 43 * pp. 1465-1480, Oct. 1995. The original Vegas duplicate ack policy has not 44 * been implemented, since clock ticks are not as coarse as they were (i.e. 45 * 500ms) when Vegas was designed. Also, packets are timed once per RTT as in 46 * the original paper. 47 * 48 * Originally released as part of the NewTCP research project at Swinburne 49 * University of Technology's Centre for Advanced Internet Architectures, 50 * Melbourne, Australia, which was made possible in part by a grant from the 51 * Cisco University Research Program Fund at Community Foundation Silicon 52 * Valley. More details are available at: 53 * http://caia.swin.edu.au/urp/newtcp/ 54 */ 55 56 #include <sys/cdefs.h> 57 __FBSDID("$FreeBSD$"); 58 59 #include <sys/param.h> 60 #include <sys/kernel.h> 61 #include <sys/khelp.h> 62 #include <sys/malloc.h> 63 #include <sys/module.h> 64 #include <sys/queue.h> 65 #include <sys/socket.h> 66 #include <sys/socketvar.h> 67 #include <sys/sysctl.h> 68 #include <sys/systm.h> 69 70 #include <net/if.h> 71 #include <net/vnet.h> 72 73 #include <netinet/cc.h> 74 #include <netinet/tcp_seq.h> 75 #include <netinet/tcp_timer.h> 76 #include <netinet/tcp_var.h> 77 78 #include <netinet/cc/cc_module.h> 79 80 #include <netinet/khelp/h_ertt.h> 81 82 #define CAST_PTR_INT(X) (*((int*)(X))) 83 84 /* 85 * Private signal type for rate based congestion signal. 86 * See <netinet/cc.h> for appropriate bit-range to use for private signals. 87 */ 88 #define CC_VEGAS_RATE 0x01000000 89 90 static void vegas_ack_received(struct cc_var *ccv, uint16_t ack_type); 91 static void vegas_cb_destroy(struct cc_var *ccv); 92 static int vegas_cb_init(struct cc_var *ccv); 93 static void vegas_cong_signal(struct cc_var *ccv, uint32_t signal_type); 94 static void vegas_conn_init(struct cc_var *ccv); 95 static int vegas_mod_init(void); 96 97 struct vegas { 98 int slow_start_toggle; 99 }; 100 101 static int32_t ertt_id; 102 103 static VNET_DEFINE(uint32_t, vegas_alpha) = 1; 104 static VNET_DEFINE(uint32_t, vegas_beta) = 3; 105 #define V_vegas_alpha VNET(vegas_alpha) 106 #define V_vegas_beta VNET(vegas_beta) 107 108 static MALLOC_DEFINE(M_VEGAS, "vegas data", 109 "Per connection data required for the Vegas congestion control algorithm"); 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 }; 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 measurment data 124 * has been used. 125 */ 126 static void 127 vegas_ack_received(struct cc_var *ccv, uint16_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, 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_algo.ack_received(ccv, ack_type); 168 } 169 170 static void 171 vegas_cb_destroy(struct cc_var *ccv) 172 { 173 174 if (ccv->cc_data != NULL) 175 free(ccv->cc_data, M_VEGAS); 176 } 177 178 static int 179 vegas_cb_init(struct cc_var *ccv) 180 { 181 struct vegas *vegas_data; 182 183 vegas_data = malloc(sizeof(struct vegas), M_VEGAS, M_NOWAIT); 184 185 if (vegas_data == NULL) 186 return (ENOMEM); 187 188 vegas_data->slow_start_toggle = 1; 189 ccv->cc_data = vegas_data; 190 191 return (0); 192 } 193 194 /* 195 * If congestion has been triggered triggered by the Vegas measured rates, it is 196 * handled here, otherwise it falls back to newreno's congestion handling. 197 */ 198 static void 199 vegas_cong_signal(struct cc_var *ccv, uint32_t signal_type) 200 { 201 struct vegas *vegas_data; 202 int presignalrecov; 203 204 vegas_data = ccv->cc_data; 205 206 if (IN_RECOVERY(CCV(ccv, t_flags))) 207 presignalrecov = 1; 208 else 209 presignalrecov = 0; 210 211 switch(signal_type) { 212 case CC_VEGAS_RATE: 213 if (!IN_RECOVERY(CCV(ccv, t_flags))) { 214 CCV(ccv, snd_cwnd) = max(2 * CCV(ccv, t_maxseg), 215 CCV(ccv, snd_cwnd) - CCV(ccv, t_maxseg)); 216 if (CCV(ccv, snd_cwnd) < CCV(ccv, snd_ssthresh)) 217 /* Exit slow start. */ 218 CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); 219 } 220 break; 221 222 default: 223 newreno_cc_algo.cong_signal(ccv, signal_type); 224 } 225 226 if (IN_RECOVERY(CCV(ccv, t_flags)) && !presignalrecov) 227 vegas_data->slow_start_toggle = 228 (CCV(ccv, snd_cwnd) < CCV(ccv, snd_ssthresh)) ? 1 : 0; 229 } 230 231 static void 232 vegas_conn_init(struct cc_var *ccv) 233 { 234 struct vegas *vegas_data; 235 236 vegas_data = ccv->cc_data; 237 vegas_data->slow_start_toggle = 1; 238 } 239 240 static int 241 vegas_mod_init(void) 242 { 243 244 ertt_id = khelp_get_id("ertt"); 245 if (ertt_id <= 0) { 246 printf("%s: h_ertt module not found\n", __func__); 247 return (ENOENT); 248 } 249 250 vegas_cc_algo.after_idle = newreno_cc_algo.after_idle; 251 vegas_cc_algo.post_recovery = newreno_cc_algo.post_recovery; 252 253 return (0); 254 } 255 256 static int 257 vegas_alpha_handler(SYSCTL_HANDLER_ARGS) 258 { 259 int error; 260 uint32_t new; 261 262 new = V_vegas_alpha; 263 error = sysctl_handle_int(oidp, &new, 0, req); 264 if (error == 0 && req->newptr != NULL) { 265 if (CAST_PTR_INT(req->newptr) < 1 || 266 CAST_PTR_INT(req->newptr) > V_vegas_beta) 267 error = EINVAL; 268 else 269 V_vegas_alpha = new; 270 } 271 272 return (error); 273 } 274 275 static int 276 vegas_beta_handler(SYSCTL_HANDLER_ARGS) 277 { 278 int error; 279 uint32_t new; 280 281 new = V_vegas_beta; 282 error = sysctl_handle_int(oidp, &new, 0, req); 283 if (error == 0 && req->newptr != NULL) { 284 if (CAST_PTR_INT(req->newptr) < 1 || 285 CAST_PTR_INT(req->newptr) < V_vegas_alpha) 286 error = EINVAL; 287 else 288 V_vegas_beta = new; 289 } 290 291 return (error); 292 } 293 294 SYSCTL_DECL(_net_inet_tcp_cc_vegas); 295 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, vegas, CTLFLAG_RW, NULL, 296 "Vegas related settings"); 297 298 SYSCTL_PROC(_net_inet_tcp_cc_vegas, OID_AUTO, alpha, 299 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 300 &VNET_NAME(vegas_alpha), 1, &vegas_alpha_handler, "IU", 301 "vegas alpha, specified as number of \"buffers\" (0 < alpha < beta)"); 302 303 SYSCTL_PROC(_net_inet_tcp_cc_vegas, OID_AUTO, beta, 304 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 305 &VNET_NAME(vegas_beta), 3, &vegas_beta_handler, "IU", 306 "vegas beta, specified as number of \"buffers\" (0 < alpha < beta)"); 307 308 DECLARE_CC_MODULE(vegas, &vegas_cc_algo); 309 MODULE_DEPEND(vegas, ertt, 1, 1, 1); 310