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 #define CAST_PTR_INT(X) (*((int*)(X)))
84 static void newreno_cb_destroy(struct cc_var *ccv);
85 static void newreno_ack_received(struct cc_var *ccv, uint16_t type);
86 static void newreno_after_idle(struct cc_var *ccv);
87 static void newreno_cong_signal(struct cc_var *ccv, uint32_t type);
88 static void newreno_post_recovery(struct cc_var *ccv);
89 static int newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf);
91 VNET_DEFINE_STATIC(uint32_t, newreno_beta) = 50;
92 VNET_DEFINE_STATIC(uint32_t, newreno_beta_ecn) = 80;
93 #define V_newreno_beta VNET(newreno_beta)
94 #define V_newreno_beta_ecn VNET(newreno_beta_ecn)
96 struct cc_algo newreno_cc_algo = {
98 .cb_destroy = newreno_cb_destroy,
99 .ack_received = newreno_ack_received,
100 .after_idle = newreno_after_idle,
101 .cong_signal = newreno_cong_signal,
102 .post_recovery = newreno_post_recovery,
103 .ctl_output = newreno_ctl_output,
111 static inline struct newreno *
112 newreno_malloc(struct cc_var *ccv)
114 struct newreno *nreno;
116 nreno = malloc(sizeof(struct newreno), M_NEWRENO, M_NOWAIT);
118 /* NB: nreno is not zeroed, so initialise all fields. */
119 nreno->beta = V_newreno_beta;
120 nreno->beta_ecn = V_newreno_beta_ecn;
121 ccv->cc_data = nreno;
128 newreno_cb_destroy(struct cc_var *ccv)
130 free(ccv->cc_data, M_NEWRENO);
134 newreno_ack_received(struct cc_var *ccv, uint16_t type)
136 if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) &&
137 (ccv->flags & CCF_CWND_LIMITED)) {
138 u_int cw = CCV(ccv, snd_cwnd);
139 u_int incr = CCV(ccv, t_maxseg);
142 * Regular in-order ACK, open the congestion window.
143 * Method depends on which congestion control state we're
144 * in (slow start or cong avoid) and if ABC (RFC 3465) is
147 * slow start: cwnd <= ssthresh
148 * cong avoid: cwnd > ssthresh
150 * slow start and ABC (RFC 3465):
151 * Grow cwnd exponentially by the amount of data
152 * ACKed capping the max increment per ACK to
153 * (abc_l_var * maxseg) bytes.
155 * slow start without ABC (RFC 5681):
156 * Grow cwnd exponentially by maxseg per ACK.
158 * cong avoid and ABC (RFC 3465):
159 * Grow cwnd linearly by maxseg per RTT for each
160 * cwnd worth of ACKed data.
162 * cong avoid without ABC (RFC 5681):
163 * Grow cwnd linearly by approximately maxseg per RTT using
164 * maxseg^2 / cwnd per ACK as the increment.
165 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to
166 * avoid capping cwnd.
168 if (cw > CCV(ccv, snd_ssthresh)) {
169 if (V_tcp_do_rfc3465) {
170 if (ccv->flags & CCF_ABC_SENTAWND)
171 ccv->flags &= ~CCF_ABC_SENTAWND;
175 incr = max((incr * incr / cw), 1);
176 } else if (V_tcp_do_rfc3465) {
178 * In slow-start with ABC enabled and no RTO in sight?
179 * (Must not use abc_l_var > 1 if slow starting after
180 * an RTO. On RTO, snd_nxt = snd_una, so the
181 * snd_nxt == snd_max check is sufficient to
184 * XXXLAS: Find a way to signal SS after RTO that
185 * doesn't rely on tcpcb vars.
187 if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max))
188 incr = min(ccv->bytes_this_ack,
189 ccv->nsegs * V_tcp_abc_l_var *
192 incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg));
194 /* ABC is on by default, so incr equals 0 frequently. */
196 CCV(ccv, snd_cwnd) = min(cw + incr,
197 TCP_MAXWIN << CCV(ccv, snd_scale));
202 newreno_after_idle(struct cc_var *ccv)
207 * If we've been idle for more than one retransmit timeout the old
208 * congestion window is no longer current and we have to reduce it to
209 * the restart window before we can transmit again.
211 * The restart window is the initial window or the last CWND, whichever
214 * This is done to prevent us from flooding the path with a full CWND at
215 * wirespeed, overloading router and switch buffers along the way.
217 * See RFC5681 Section 4.1. "Restarting Idle Connections".
219 if (V_tcp_do_rfc3390)
220 rw = min(4 * CCV(ccv, t_maxseg),
221 max(2 * CCV(ccv, t_maxseg), 4380));
223 rw = CCV(ccv, t_maxseg) * 2;
225 CCV(ccv, snd_cwnd) = min(rw, CCV(ccv, snd_cwnd));
229 * Perform any necessary tasks before we enter congestion recovery.
232 newreno_cong_signal(struct cc_var *ccv, uint32_t type)
234 struct newreno *nreno;
235 uint32_t beta, beta_ecn, cwin, factor;
238 cwin = CCV(ccv, snd_cwnd);
239 mss = CCV(ccv, t_maxseg);
240 nreno = ccv->cc_data;
241 beta = (nreno == NULL) ? V_newreno_beta : nreno->beta;
242 beta_ecn = (nreno == NULL) ? V_newreno_beta_ecn : nreno->beta_ecn;
243 if (V_cc_do_abe && type == CC_ECN)
248 /* Catch algos which mistakenly leak private signal types. */
249 KASSERT((type & CC_SIGPRIVMASK) == 0,
250 ("%s: congestion signal type 0x%08x is private\n", __func__, type));
252 cwin = max(((uint64_t)cwin * (uint64_t)factor) / (100ULL * (uint64_t)mss),
257 if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) {
258 if (IN_CONGRECOVERY(CCV(ccv, t_flags) &&
259 V_cc_do_abe && V_cc_abe_frlossreduce)) {
260 CCV(ccv, snd_ssthresh) =
261 ((uint64_t)CCV(ccv, snd_ssthresh) *
263 (100ULL * (uint64_t)beta_ecn);
265 if (!IN_CONGRECOVERY(CCV(ccv, t_flags)))
266 CCV(ccv, snd_ssthresh) = cwin;
267 ENTER_RECOVERY(CCV(ccv, t_flags));
271 if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) {
272 CCV(ccv, snd_ssthresh) = cwin;
273 CCV(ccv, snd_cwnd) = cwin;
274 ENTER_CONGRECOVERY(CCV(ccv, t_flags));
281 * Perform any necessary tasks before we exit congestion recovery.
284 newreno_post_recovery(struct cc_var *ccv)
288 if (IN_FASTRECOVERY(CCV(ccv, t_flags))) {
290 * Fast recovery will conclude after returning from this
291 * function. Window inflation should have left us with
292 * approximately snd_ssthresh outstanding data. But in case we
293 * would be inclined to send a burst, better to do it via the
294 * slow start mechanism.
296 * XXXLAS: Find a way to do this without needing curack
298 if (V_tcp_do_rfc6675_pipe)
299 pipe = tcp_compute_pipe(ccv->ccvc.tcp);
301 pipe = CCV(ccv, snd_max) - ccv->curack;
303 if (pipe < CCV(ccv, snd_ssthresh))
304 CCV(ccv, snd_cwnd) = pipe + CCV(ccv, t_maxseg);
306 CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
311 newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf)
313 struct newreno *nreno;
314 struct cc_newreno_opts *opt;
316 if (sopt->sopt_valsize != sizeof(struct cc_newreno_opts))
319 nreno = ccv->cc_data;
322 switch (sopt->sopt_dir) {
324 /* We cannot set without cc_data memory. */
326 nreno = newreno_malloc(ccv);
331 case CC_NEWRENO_BETA:
332 nreno->beta = opt->val;
334 case CC_NEWRENO_BETA_ECN:
337 nreno->beta_ecn = opt->val;
340 return (ENOPROTOOPT);
345 case CC_NEWRENO_BETA:
346 opt->val = (nreno == NULL) ?
347 V_newreno_beta : nreno->beta;
349 case CC_NEWRENO_BETA_ECN:
350 opt->val = (nreno == NULL) ?
351 V_newreno_beta_ecn : nreno->beta_ecn;
354 return (ENOPROTOOPT);
365 newreno_beta_handler(SYSCTL_HANDLER_ARGS)
368 if (req->newptr != NULL ) {
369 if (arg1 == &VNET_NAME(newreno_beta_ecn) && !V_cc_do_abe)
371 if (CAST_PTR_INT(req->newptr) <= 0 || CAST_PTR_INT(req->newptr) > 100)
375 return (sysctl_handle_int(oidp, arg1, arg2, req));
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);