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 Hamilton Institute's delay-based congestion control 43 * algorithm for FreeBSD, based on "A strategy for fair coexistence of loss and 44 * delay-based congestion control algorithms," by L. Budzisz, R. Stanojevic, R. 45 * Shorten, and F. Baker, IEEE Commun. Lett., vol. 13, no. 7, pp. 555--557, Jul. 46 * 2009. 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/limits.h> 63 #include <sys/malloc.h> 64 #include <sys/module.h> 65 #include <sys/queue.h> 66 #include <sys/socket.h> 67 #include <sys/socketvar.h> 68 #include <sys/sysctl.h> 69 #include <sys/systm.h> 70 71 #include <net/vnet.h> 72 73 #include <netinet/tcp.h> 74 #include <netinet/tcp_seq.h> 75 #include <netinet/tcp_timer.h> 76 #include <netinet/tcp_var.h> 77 #include <netinet/cc/cc.h> 78 #include <netinet/cc/cc_module.h> 79 80 #include <netinet/khelp/h_ertt.h> 81 82 /* Largest possible number returned by random(). */ 83 #define RANDOM_MAX INT_MAX 84 85 static void hd_ack_received(struct cc_var *ccv, uint16_t ack_type); 86 static int hd_mod_init(void); 87 static size_t hd_data_sz(void); 88 89 static int ertt_id; 90 91 VNET_DEFINE_STATIC(uint32_t, hd_qthresh) = 20; 92 VNET_DEFINE_STATIC(uint32_t, hd_qmin) = 5; 93 VNET_DEFINE_STATIC(uint32_t, hd_pmax) = 5; 94 #define V_hd_qthresh VNET(hd_qthresh) 95 #define V_hd_qmin VNET(hd_qmin) 96 #define V_hd_pmax VNET(hd_pmax) 97 98 struct cc_algo hd_cc_algo = { 99 .name = "hd", 100 .ack_received = hd_ack_received, 101 .mod_init = hd_mod_init, 102 .cc_data_sz = hd_data_sz, 103 .after_idle = newreno_cc_after_idle, 104 .cong_signal = newreno_cc_cong_signal, 105 .post_recovery = newreno_cc_post_recovery, 106 }; 107 108 static size_t 109 hd_data_sz(void) 110 { 111 return (0); 112 } 113 114 /* 115 * Hamilton backoff function. Returns 1 if we should backoff or 0 otherwise. 116 */ 117 static __inline int 118 should_backoff(int qdly, int maxqdly) 119 { 120 unsigned long p; 121 122 if (qdly < V_hd_qthresh) { 123 p = (((RANDOM_MAX / 100) * V_hd_pmax) / 124 (V_hd_qthresh - V_hd_qmin)) * (qdly - V_hd_qmin); 125 } else { 126 if (qdly > V_hd_qthresh) 127 p = (((RANDOM_MAX / 100) * V_hd_pmax) / 128 (maxqdly - V_hd_qthresh)) * (maxqdly - qdly); 129 else 130 p = (RANDOM_MAX / 100) * V_hd_pmax; 131 } 132 133 return (random() < p); 134 } 135 136 /* 137 * If the ack type is CC_ACK, and the inferred queueing delay is greater than 138 * the Qmin threshold, cwnd is reduced probabilistically. When backing off due 139 * to delay, HD behaves like NewReno when an ECN signal is received. HD behaves 140 * as NewReno in all other circumstances. 141 */ 142 static void 143 hd_ack_received(struct cc_var *ccv, uint16_t ack_type) 144 { 145 struct ertt *e_t; 146 int qdly; 147 148 if (ack_type == CC_ACK) { 149 e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); 150 151 if (e_t->rtt && e_t->minrtt && V_hd_qthresh > 0) { 152 qdly = e_t->rtt - e_t->minrtt; 153 154 if (qdly > V_hd_qmin && 155 !IN_RECOVERY(CCV(ccv, t_flags))) { 156 /* Probabilistic backoff of cwnd. */ 157 if (should_backoff(qdly, 158 e_t->maxrtt - e_t->minrtt)) { 159 /* 160 * Update cwnd and ssthresh update to 161 * half cwnd and behave like an ECN (ie 162 * not a packet loss). 163 */ 164 newreno_cc_cong_signal(ccv, 165 CC_ECN); 166 return; 167 } 168 } 169 } 170 } 171 newreno_cc_ack_received(ccv, ack_type); 172 } 173 174 static int 175 hd_mod_init(void) 176 { 177 178 ertt_id = khelp_get_id("ertt"); 179 if (ertt_id <= 0) { 180 printf("%s: h_ertt module not found\n", __func__); 181 return (ENOENT); 182 } 183 return (0); 184 } 185 186 static int 187 hd_pmax_handler(SYSCTL_HANDLER_ARGS) 188 { 189 int error; 190 uint32_t new; 191 192 new = V_hd_pmax; 193 error = sysctl_handle_int(oidp, &new, 0, req); 194 if (error == 0 && req->newptr != NULL) { 195 if (new == 0 || new > 100) 196 error = EINVAL; 197 else 198 V_hd_pmax = new; 199 } 200 201 return (error); 202 } 203 204 static int 205 hd_qmin_handler(SYSCTL_HANDLER_ARGS) 206 { 207 int error; 208 uint32_t new; 209 210 new = V_hd_qmin; 211 error = sysctl_handle_int(oidp, &new, 0, req); 212 if (error == 0 && req->newptr != NULL) { 213 if (new > V_hd_qthresh) 214 error = EINVAL; 215 else 216 V_hd_qmin = new; 217 } 218 219 return (error); 220 } 221 222 static int 223 hd_qthresh_handler(SYSCTL_HANDLER_ARGS) 224 { 225 int error; 226 uint32_t new; 227 228 new = V_hd_qthresh; 229 error = sysctl_handle_int(oidp, &new, 0, req); 230 if (error == 0 && req->newptr != NULL) { 231 if (new == 0 || new < V_hd_qmin) 232 error = EINVAL; 233 else 234 V_hd_qthresh = new; 235 } 236 237 return (error); 238 } 239 240 SYSCTL_DECL(_net_inet_tcp_cc_hd); 241 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, hd, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 242 "Hamilton delay-based congestion control related settings"); 243 244 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, queue_threshold, 245 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 246 &VNET_NAME(hd_qthresh), 20, &hd_qthresh_handler, "IU", 247 "queueing congestion threshold (qth) in ticks"); 248 249 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, pmax, 250 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 251 &VNET_NAME(hd_pmax), 5, &hd_pmax_handler, "IU", 252 "per packet maximum backoff probability as a percentage"); 253 254 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, queue_min, 255 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 256 &VNET_NAME(hd_qmin), 5, &hd_qmin_handler, "IU", 257 "minimum queueing delay threshold (qmin) in ticks"); 258 259 DECLARE_CC_MODULE(hd, &hd_cc_algo); 260 MODULE_VERSION(hd, 2); 261 MODULE_DEPEND(hd, ertt, 1, 1, 1); 262