2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5 * The Regents of the University of California.
6 * Copyright (c) 2007-2008,2010,2014
7 * Swinburne University of Technology, Melbourne, Australia.
8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
9 * Copyright (c) 2010 The FreeBSD Foundation
10 * All rights reserved.
12 * This software was developed at the Centre for Advanced Internet
13 * Architectures, Swinburne University of Technology, by Lawrence Stewart, James
14 * Healy and David Hayes, made possible in part by a grant from the Cisco
15 * University Research Program Fund at Community Foundation Silicon Valley.
17 * Portions of this software were developed at the Centre for Advanced
18 * Internet Architectures, Swinburne University of Technology, Melbourne,
19 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
21 * Redistribution and use in source and binary forms, with or without
22 * modification, are permitted provided that the following conditions
24 * 1. Redistributions of source code must retain the above copyright
25 * notice, this list of conditions and the following disclaimer.
26 * 2. Redistributions in binary form must reproduce the above copyright
27 * notice, this list of conditions and the following disclaimer in the
28 * documentation and/or other materials provided with the distribution.
30 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
44 * This software was first released in 2007 by James Healy and Lawrence Stewart
45 * whilst working on the NewTCP research project at Swinburne University of
46 * Technology's Centre for Advanced Internet Architectures, Melbourne,
47 * Australia, which was made possible in part by a grant from the Cisco
48 * University Research Program Fund at Community Foundation Silicon Valley.
49 * More details are available at:
50 * http://caia.swin.edu.au/urp/newtcp/
52 * Dec 2014 garmitage@swin.edu.au
53 * Borrowed code fragments from cc_cdg.c to add modifiable beta
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
61 #include <sys/param.h>
62 #include <sys/kernel.h>
63 #include <sys/malloc.h>
64 #include <sys/module.h>
65 #include <sys/socket.h>
66 #include <sys/socketvar.h>
67 #include <sys/sysctl.h>
68 #include <sys/systm.h>
72 #include <netinet/tcp.h>
73 #include <netinet/tcp_seq.h>
74 #include <netinet/tcp_var.h>
75 #include <netinet/cc/cc.h>
76 #include <netinet/cc/cc_module.h>
77 #include <netinet/cc/cc_newreno.h>
79 static MALLOC_DEFINE(M_NEWRENO, "newreno data",
80 "newreno beta values");
82 static void newreno_cb_destroy(struct cc_var *ccv);
83 static void newreno_ack_received(struct cc_var *ccv, uint16_t type);
84 static void newreno_after_idle(struct cc_var *ccv);
85 static void newreno_cong_signal(struct cc_var *ccv, uint32_t type);
86 static void newreno_post_recovery(struct cc_var *ccv);
87 static int newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf);
89 VNET_DEFINE_STATIC(uint32_t, newreno_beta) = 50;
90 VNET_DEFINE_STATIC(uint32_t, newreno_beta_ecn) = 80;
91 #define V_newreno_beta VNET(newreno_beta)
92 #define V_newreno_beta_ecn VNET(newreno_beta_ecn)
94 struct cc_algo newreno_cc_algo = {
96 .cb_destroy = newreno_cb_destroy,
97 .ack_received = newreno_ack_received,
98 .after_idle = newreno_after_idle,
99 .cong_signal = newreno_cong_signal,
100 .post_recovery = newreno_post_recovery,
101 .ctl_output = newreno_ctl_output,
109 static inline struct newreno *
110 newreno_malloc(struct cc_var *ccv)
112 struct newreno *nreno;
114 nreno = malloc(sizeof(struct newreno), M_NEWRENO, M_NOWAIT);
116 /* NB: nreno is not zeroed, so initialise all fields. */
117 nreno->beta = V_newreno_beta;
118 nreno->beta_ecn = V_newreno_beta_ecn;
119 ccv->cc_data = nreno;
126 newreno_cb_destroy(struct cc_var *ccv)
128 free(ccv->cc_data, M_NEWRENO);
132 newreno_ack_received(struct cc_var *ccv, uint16_t type)
134 if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) &&
135 (ccv->flags & CCF_CWND_LIMITED)) {
136 u_int cw = CCV(ccv, snd_cwnd);
137 u_int incr = CCV(ccv, t_maxseg);
140 * Regular in-order ACK, open the congestion window.
141 * Method depends on which congestion control state we're
142 * in (slow start or cong avoid) and if ABC (RFC 3465) is
145 * slow start: cwnd <= ssthresh
146 * cong avoid: cwnd > ssthresh
148 * slow start and ABC (RFC 3465):
149 * Grow cwnd exponentially by the amount of data
150 * ACKed capping the max increment per ACK to
151 * (abc_l_var * maxseg) bytes.
153 * slow start without ABC (RFC 5681):
154 * Grow cwnd exponentially by maxseg per ACK.
156 * cong avoid and ABC (RFC 3465):
157 * Grow cwnd linearly by maxseg per RTT for each
158 * cwnd worth of ACKed data.
160 * cong avoid without ABC (RFC 5681):
161 * Grow cwnd linearly by approximately maxseg per RTT using
162 * maxseg^2 / cwnd per ACK as the increment.
163 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to
164 * avoid capping cwnd.
166 if (cw > CCV(ccv, snd_ssthresh)) {
167 if (V_tcp_do_rfc3465) {
168 if (ccv->flags & CCF_ABC_SENTAWND)
169 ccv->flags &= ~CCF_ABC_SENTAWND;
173 incr = max((incr * incr / cw), 1);
174 } else if (V_tcp_do_rfc3465) {
176 * In slow-start with ABC enabled and no RTO in sight?
177 * (Must not use abc_l_var > 1 if slow starting after
178 * an RTO. On RTO, snd_nxt = snd_una, so the
179 * snd_nxt == snd_max check is sufficient to
182 * XXXLAS: Find a way to signal SS after RTO that
183 * doesn't rely on tcpcb vars.
185 if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max))
186 incr = min(ccv->bytes_this_ack,
187 ccv->nsegs * V_tcp_abc_l_var *
190 incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg));
192 /* ABC is on by default, so incr equals 0 frequently. */
194 CCV(ccv, snd_cwnd) = min(cw + incr,
195 TCP_MAXWIN << CCV(ccv, snd_scale));
200 newreno_after_idle(struct cc_var *ccv)
205 * If we've been idle for more than one retransmit timeout the old
206 * congestion window is no longer current and we have to reduce it to
207 * the restart window before we can transmit again.
209 * The restart window is the initial window or the last CWND, whichever
212 * This is done to prevent us from flooding the path with a full CWND at
213 * wirespeed, overloading router and switch buffers along the way.
215 * See RFC5681 Section 4.1. "Restarting Idle Connections".
217 rw = tcp_compute_initwnd(tcp_maxseg(ccv->ccvc.tcp));
219 CCV(ccv, snd_cwnd) = min(rw, CCV(ccv, snd_cwnd));
223 * Perform any necessary tasks before we enter congestion recovery.
226 newreno_cong_signal(struct cc_var *ccv, uint32_t type)
228 struct newreno *nreno;
229 uint32_t beta, beta_ecn, cwin, factor;
232 cwin = CCV(ccv, snd_cwnd);
233 mss = CCV(ccv, t_maxseg);
234 nreno = ccv->cc_data;
235 beta = (nreno == NULL) ? V_newreno_beta : nreno->beta;
236 beta_ecn = (nreno == NULL) ? V_newreno_beta_ecn : nreno->beta_ecn;
237 if (V_cc_do_abe && type == CC_ECN)
242 /* Catch algos which mistakenly leak private signal types. */
243 KASSERT((type & CC_SIGPRIVMASK) == 0,
244 ("%s: congestion signal type 0x%08x is private\n", __func__, type));
246 cwin = max(((uint64_t)cwin * (uint64_t)factor) / (100ULL * (uint64_t)mss),
251 if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) {
252 if (IN_CONGRECOVERY(CCV(ccv, t_flags) &&
253 V_cc_do_abe && V_cc_abe_frlossreduce)) {
254 CCV(ccv, snd_ssthresh) =
255 ((uint64_t)CCV(ccv, snd_ssthresh) *
257 (100ULL * (uint64_t)beta_ecn);
259 if (!IN_CONGRECOVERY(CCV(ccv, t_flags)))
260 CCV(ccv, snd_ssthresh) = cwin;
261 ENTER_RECOVERY(CCV(ccv, t_flags));
265 if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
266 CCV(ccv, snd_ssthresh) = cwin;
267 CCV(ccv, snd_cwnd) = cwin;
268 ENTER_CONGRECOVERY(CCV(ccv, t_flags));
275 * Perform any necessary tasks before we exit congestion recovery.
278 newreno_post_recovery(struct cc_var *ccv)
282 if (IN_FASTRECOVERY(CCV(ccv, t_flags))) {
284 * Fast recovery will conclude after returning from this
285 * function. Window inflation should have left us with
286 * approximately snd_ssthresh outstanding data. But in case we
287 * would be inclined to send a burst, better to do it via the
288 * slow start mechanism.
290 * XXXLAS: Find a way to do this without needing curack
292 if (V_tcp_do_rfc6675_pipe)
293 pipe = tcp_compute_pipe(ccv->ccvc.tcp);
295 pipe = CCV(ccv, snd_max) - ccv->curack;
297 if (pipe < CCV(ccv, snd_ssthresh))
298 CCV(ccv, snd_cwnd) = pipe + CCV(ccv, t_maxseg);
300 CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
305 newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf)
307 struct newreno *nreno;
308 struct cc_newreno_opts *opt;
310 if (sopt->sopt_valsize != sizeof(struct cc_newreno_opts))
313 nreno = ccv->cc_data;
316 switch (sopt->sopt_dir) {
318 /* We cannot set without cc_data memory. */
320 nreno = newreno_malloc(ccv);
325 case CC_NEWRENO_BETA:
326 nreno->beta = opt->val;
328 case CC_NEWRENO_BETA_ECN:
331 nreno->beta_ecn = opt->val;
334 return (ENOPROTOOPT);
339 case CC_NEWRENO_BETA:
340 opt->val = (nreno == NULL) ?
341 V_newreno_beta : nreno->beta;
343 case CC_NEWRENO_BETA_ECN:
344 opt->val = (nreno == NULL) ?
345 V_newreno_beta_ecn : nreno->beta_ecn;
348 return (ENOPROTOOPT);
359 newreno_beta_handler(SYSCTL_HANDLER_ARGS)
364 new = *(uint32_t *)arg1;
365 error = sysctl_handle_int(oidp, &new, 0, req);
366 if (error == 0 && req->newptr != NULL ) {
367 if (arg1 == &VNET_NAME(newreno_beta_ecn) && !V_cc_do_abe)
369 else if (new == 0 || new > 100)
372 *(uint32_t *)arg1 = new;
378 SYSCTL_DECL(_net_inet_tcp_cc_newreno);
379 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, newreno, CTLFLAG_RW, NULL,
380 "New Reno related settings");
382 SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta,
383 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW,
384 &VNET_NAME(newreno_beta), 3, &newreno_beta_handler, "IU",
385 "New Reno beta, specified as number between 1 and 100");
387 SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta_ecn,
388 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW,
389 &VNET_NAME(newreno_beta_ecn), 3, &newreno_beta_handler, "IU",
390 "New Reno beta ecn, specified as number between 1 and 100");
392 DECLARE_CC_MODULE(newreno, &newreno_cc_algo);