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 #define CAST_PTR_INT(X) (*((int*)(X))) 83 84 /* Largest possible number returned by random(). */ 85 #define RANDOM_MAX INT_MAX 86 87 static void hd_ack_received(struct cc_var *ccv, uint16_t ack_type); 88 static int hd_mod_init(void); 89 90 static int ertt_id; 91 92 VNET_DEFINE_STATIC(uint32_t, hd_qthresh) = 20; 93 VNET_DEFINE_STATIC(uint32_t, hd_qmin) = 5; 94 VNET_DEFINE_STATIC(uint32_t, hd_pmax) = 5; 95 #define V_hd_qthresh VNET(hd_qthresh) 96 #define V_hd_qmin VNET(hd_qmin) 97 #define V_hd_pmax VNET(hd_pmax) 98 99 struct cc_algo hd_cc_algo = { 100 .name = "hd", 101 .ack_received = hd_ack_received, 102 .mod_init = hd_mod_init 103 }; 104 105 /* 106 * Hamilton backoff function. Returns 1 if we should backoff or 0 otherwise. 107 */ 108 static __inline int 109 should_backoff(int qdly, int maxqdly) 110 { 111 unsigned long p; 112 113 if (qdly < V_hd_qthresh) { 114 p = (((RANDOM_MAX / 100) * V_hd_pmax) / 115 (V_hd_qthresh - V_hd_qmin)) * (qdly - V_hd_qmin); 116 } else { 117 if (qdly > V_hd_qthresh) 118 p = (((RANDOM_MAX / 100) * V_hd_pmax) / 119 (maxqdly - V_hd_qthresh)) * (maxqdly - qdly); 120 else 121 p = (RANDOM_MAX / 100) * V_hd_pmax; 122 } 123 124 return (random() < p); 125 } 126 127 /* 128 * If the ack type is CC_ACK, and the inferred queueing delay is greater than 129 * the Qmin threshold, cwnd is reduced probabilistically. When backing off due 130 * to delay, HD behaves like NewReno when an ECN signal is received. HD behaves 131 * as NewReno in all other circumstances. 132 */ 133 static void 134 hd_ack_received(struct cc_var *ccv, uint16_t ack_type) 135 { 136 struct ertt *e_t; 137 int qdly; 138 139 if (ack_type == CC_ACK) { 140 e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); 141 142 if (e_t->rtt && e_t->minrtt && V_hd_qthresh > 0) { 143 qdly = e_t->rtt - e_t->minrtt; 144 145 if (qdly > V_hd_qmin && 146 !IN_RECOVERY(CCV(ccv, t_flags))) { 147 /* Probabilistic backoff of cwnd. */ 148 if (should_backoff(qdly, 149 e_t->maxrtt - e_t->minrtt)) { 150 /* 151 * Update cwnd and ssthresh update to 152 * half cwnd and behave like an ECN (ie 153 * not a packet loss). 154 */ 155 newreno_cc_algo.cong_signal(ccv, 156 CC_ECN); 157 return; 158 } 159 } 160 } 161 } 162 newreno_cc_algo.ack_received(ccv, ack_type); /* As for NewReno. */ 163 } 164 165 static int 166 hd_mod_init(void) 167 { 168 169 ertt_id = khelp_get_id("ertt"); 170 if (ertt_id <= 0) { 171 printf("%s: h_ertt module not found\n", __func__); 172 return (ENOENT); 173 } 174 175 hd_cc_algo.after_idle = newreno_cc_algo.after_idle; 176 hd_cc_algo.cong_signal = newreno_cc_algo.cong_signal; 177 hd_cc_algo.post_recovery = newreno_cc_algo.post_recovery; 178 179 return (0); 180 } 181 182 static int 183 hd_pmax_handler(SYSCTL_HANDLER_ARGS) 184 { 185 int error; 186 uint32_t new; 187 188 new = V_hd_pmax; 189 error = sysctl_handle_int(oidp, &new, 0, req); 190 if (error == 0 && req->newptr != NULL) { 191 if (CAST_PTR_INT(req->newptr) == 0 || 192 CAST_PTR_INT(req->newptr) > 100) 193 error = EINVAL; 194 else 195 V_hd_pmax = new; 196 } 197 198 return (error); 199 } 200 201 static int 202 hd_qmin_handler(SYSCTL_HANDLER_ARGS) 203 { 204 int error; 205 uint32_t new; 206 207 new = V_hd_qmin; 208 error = sysctl_handle_int(oidp, &new, 0, req); 209 if (error == 0 && req->newptr != NULL) { 210 if (CAST_PTR_INT(req->newptr) > V_hd_qthresh) 211 error = EINVAL; 212 else 213 V_hd_qmin = new; 214 } 215 216 return (error); 217 } 218 219 static int 220 hd_qthresh_handler(SYSCTL_HANDLER_ARGS) 221 { 222 int error; 223 uint32_t new; 224 225 new = V_hd_qthresh; 226 error = sysctl_handle_int(oidp, &new, 0, req); 227 if (error == 0 && req->newptr != NULL) { 228 if (CAST_PTR_INT(req->newptr) < 1 || 229 CAST_PTR_INT(req->newptr) < V_hd_qmin) 230 error = EINVAL; 231 else 232 V_hd_qthresh = new; 233 } 234 235 return (error); 236 } 237 238 SYSCTL_DECL(_net_inet_tcp_cc_hd); 239 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, hd, CTLFLAG_RW, NULL, 240 "Hamilton delay-based congestion control related settings"); 241 242 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, queue_threshold, 243 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(hd_qthresh), 20, 244 &hd_qthresh_handler, "IU", "queueing congestion threshold (qth) in ticks"); 245 246 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, pmax, 247 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(hd_pmax), 5, 248 &hd_pmax_handler, "IU", 249 "per packet maximum backoff probability as a percentage"); 250 251 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, queue_min, 252 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(hd_qmin), 5, 253 &hd_qmin_handler, "IU", "minimum queueing delay threshold (qmin) in ticks"); 254 255 DECLARE_CC_MODULE(hd, &hd_cc_algo); 256 MODULE_DEPEND(hd, ertt, 1, 1, 1); 257