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 <[email protected]>
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
88 static int ertt_id;
89
90 VNET_DEFINE_STATIC(uint32_t, hd_qthresh) = 20;
91 VNET_DEFINE_STATIC(uint32_t, hd_qmin) = 5;
92 VNET_DEFINE_STATIC(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
should_backoff(int qdly,int maxqdly)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
hd_ack_received(struct cc_var * ccv,uint16_t ack_type)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
hd_mod_init(void)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
hd_pmax_handler(SYSCTL_HANDLER_ARGS)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 (new == 0 || new > 100)
190 error = EINVAL;
191 else
192 V_hd_pmax = new;
193 }
194
195 return (error);
196 }
197
198 static int
hd_qmin_handler(SYSCTL_HANDLER_ARGS)199 hd_qmin_handler(SYSCTL_HANDLER_ARGS)
200 {
201 int error;
202 uint32_t new;
203
204 new = V_hd_qmin;
205 error = sysctl_handle_int(oidp, &new, 0, req);
206 if (error == 0 && req->newptr != NULL) {
207 if (new > V_hd_qthresh)
208 error = EINVAL;
209 else
210 V_hd_qmin = new;
211 }
212
213 return (error);
214 }
215
216 static int
hd_qthresh_handler(SYSCTL_HANDLER_ARGS)217 hd_qthresh_handler(SYSCTL_HANDLER_ARGS)
218 {
219 int error;
220 uint32_t new;
221
222 new = V_hd_qthresh;
223 error = sysctl_handle_int(oidp, &new, 0, req);
224 if (error == 0 && req->newptr != NULL) {
225 if (new == 0 || new < V_hd_qmin)
226 error = EINVAL;
227 else
228 V_hd_qthresh = new;
229 }
230
231 return (error);
232 }
233
234 SYSCTL_DECL(_net_inet_tcp_cc_hd);
235 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, hd, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
236 "Hamilton delay-based congestion control related settings");
237
238 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, queue_threshold,
239 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
240 &VNET_NAME(hd_qthresh), 20, &hd_qthresh_handler, "IU",
241 "queueing congestion threshold (qth) in ticks");
242
243 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, pmax,
244 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
245 &VNET_NAME(hd_pmax), 5, &hd_pmax_handler, "IU",
246 "per packet maximum backoff probability as a percentage");
247
248 SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, queue_min,
249 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
250 &VNET_NAME(hd_qmin), 5, &hd_qmin_handler, "IU",
251 "minimum queueing delay threshold (qmin) in ticks");
252
253 DECLARE_CC_MODULE(hd, &hd_cc_algo);
254 MODULE_VERSION(hd, 1);
255 MODULE_DEPEND(hd, ertt, 1, 1, 1);
256