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 Hamilton Institute's delay-based congestion control 41 * algorithm for FreeBSD, based on "A strategy for fair coexistence of loss and 42 * delay-based congestion control algorithms," by L. Budzisz, R. Stanojevic, R. 43 * Shorten, and F. Baker, IEEE Commun. Lett., vol. 13, no. 7, pp. 555--557, Jul. 44 * 2009. 45 * 46 * Originally released as part of the NewTCP research project at Swinburne 47 * University of Technology's Centre for Advanced Internet Architectures, 48 * Melbourne, Australia, which was made possible in part by a grant from the 49 * Cisco University Research Program Fund at Community Foundation Silicon 50 * Valley. More details are available at: 51 * http://caia.swin.edu.au/urp/newtcp/ 52 */ 53 54 #include <sys/cdefs.h> 55 __FBSDID("$FreeBSD$"); 56 57 #include <sys/param.h> 58 #include <sys/kernel.h> 59 #include <sys/khelp.h> 60 #include <sys/limits.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 <netinet/tcp.h> 72 #include <netinet/tcp_seq.h> 73 #include <netinet/tcp_timer.h> 74 #include <netinet/tcp_var.h> 75 #include <netinet/cc/cc.h> 76 #include <netinet/cc/cc_module.h> 77 78 #include <netinet/khelp/h_ertt.h> 79 80 #define CAST_PTR_INT(X) (*((int*)(X))) 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 88 static int ertt_id; 89 90 static VNET_DEFINE(uint32_t, hd_qthresh) = 20; 91 static VNET_DEFINE(uint32_t, hd_qmin) = 5; 92 static VNET_DEFINE(uint32_t, hd_pmax) = 5; 93 #define V_hd_qthresh VNET(hd_qthresh) 94 #define V_hd_qmin VNET(hd_qmin) 95 #define V_hd_pmax VNET(hd_pmax) 96 97 struct cc_algo hd_cc_algo = { 98 .name = "hd", 99 .ack_received = hd_ack_received, 100 .mod_init = hd_mod_init 101 }; 102 103 /* 104 * Hamilton backoff function. Returns 1 if we should backoff or 0 otherwise. 105 */ 106 static __inline int 107 should_backoff(int qdly, int maxqdly) 108 { 109 unsigned long p; 110 111 if (qdly < V_hd_qthresh) { 112 p = (((RANDOM_MAX / 100) * V_hd_pmax) / 113 (V_hd_qthresh - V_hd_qmin)) * (qdly - V_hd_qmin); 114 } else { 115 if (qdly > V_hd_qthresh) 116 p = (((RANDOM_MAX / 100) * V_hd_pmax) / 117 (maxqdly - V_hd_qthresh)) * (maxqdly - qdly); 118 else 119 p = (RANDOM_MAX / 100) * V_hd_pmax; 120 } 121 122 return (random() < p); 123 } 124 125 /* 126 * If the ack type is CC_ACK, and the inferred queueing delay is greater than 127 * the Qmin threshold, cwnd is reduced probabilistically. When backing off due 128 * to delay, HD behaves like NewReno when an ECN signal is received. HD behaves 129 * as NewReno in all other circumstances. 130 */ 131 static void 132 hd_ack_received(struct cc_var *ccv, uint16_t ack_type) 133 { 134 struct ertt *e_t; 135 int qdly; 136 137 if (ack_type == CC_ACK) { 138 e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); 139 140 if (e_t->rtt && e_t->minrtt && V_hd_qthresh > 0) { 141 qdly = e_t->rtt - e_t->minrtt; 142 143 if (qdly > V_hd_qmin && 144 !IN_RECOVERY(CCV(ccv, t_flags))) { 145 /* Probabilistic backoff of cwnd. */ 146 if (should_backoff(qdly, 147 e_t->maxrtt - e_t->minrtt)) { 148 /* 149 * Update cwnd and ssthresh update to 150 * half cwnd and behave like an ECN (ie 151 * not a packet loss). 152 */ 153 newreno_cc_algo.cong_signal(ccv, 154 CC_ECN); 155 return; 156 } 157 } 158 } 159 } 160 newreno_cc_algo.ack_received(ccv, ack_type); /* As for NewReno. */ 161 } 162 163 static int 164 hd_mod_init(void) 165 { 166 167 ertt_id = khelp_get_id("ertt"); 168 if (ertt_id <= 0) { 169 printf("%s: h_ertt module not found\n", __func__); 170 return (ENOENT); 171 } 172 173 hd_cc_algo.after_idle = newreno_cc_algo.after_idle; 174 hd_cc_algo.cong_signal = newreno_cc_algo.cong_signal; 175 hd_cc_algo.post_recovery = newreno_cc_algo.post_recovery; 176 177 return (0); 178 } 179 180 static int 181 hd_pmax_handler(SYSCTL_HANDLER_ARGS) 182 { 183 int error; 184 uint32_t new; 185 186 new = V_hd_pmax; 187 error = sysctl_handle_int(oidp, &new, 0, req); 188 if (error == 0 && req->newptr != NULL) { 189 if (CAST_PTR_INT(req->newptr) == 0 || 190 CAST_PTR_INT(req->newptr) > 100) 191 error = EINVAL; 192 else 193 V_hd_pmax = new; 194 } 195 196 return (error); 197 } 198 199 static int 200 hd_qmin_handler(SYSCTL_HANDLER_ARGS) 201 { 202 int error; 203 uint32_t new; 204 205 new = V_hd_qmin; 206 error = sysctl_handle_int(oidp, &new, 0, req); 207 if (error == 0 && req->newptr != NULL) { 208 if (CAST_PTR_INT(req->newptr) > V_hd_qthresh) 209 error = EINVAL; 210 else 211 V_hd_qmin = new; 212 } 213 214 return (error); 215 } 216 217 static int 218 hd_qthresh_handler(SYSCTL_HANDLER_ARGS) 219 { 220 int error; 221 uint32_t new; 222 223 new = V_hd_qthresh; 224 error = sysctl_handle_int(oidp, &new, 0, req); 225 if (error == 0 && req->newptr != NULL) { 226 if (CAST_PTR_INT(req->newptr) < 1 || 227 CAST_PTR_INT(req->newptr) < V_hd_qmin) 228 error = EINVAL; 229 else 230 V_hd_qthresh = new; 231 } 232 233 return (error); 234 } 235 236 SYSCTL_DECL(_net_inet_tcp_cc_hd); 237 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, hd, CTLFLAG_RW, NULL, 238 "Hamilton delay-based congestion control related settings"); 239 240 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, queue_threshold, 241 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(hd_qthresh), 20, 242 &hd_qthresh_handler, "IU", "queueing congestion threshold (qth) in ticks"); 243 244 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, pmax, 245 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(hd_pmax), 5, 246 &hd_pmax_handler, "IU", 247 "per packet maximum backoff probability as a percentage"); 248 249 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, queue_min, 250 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(hd_qmin), 5, 251 &hd_qmin_handler, "IU", "minimum queueing delay threshold (qmin) in ticks"); 252 253 DECLARE_CC_MODULE(hd, &hd_cc_algo); 254 MODULE_DEPEND(hd, ertt, 1, 1, 1); 255