1 /*
2 * Copyright (c) 2010-2021 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /*
29 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
30 * The Regents of the University of California. All rights reserved.
31 *
32 * Redistribution and use in source and binary forms, with or without
33 * modification, are permitted provided that the following conditions
34 * are met:
35 * 1. Redistributions of source code must retain the above copyright
36 * notice, this list of conditions and the following disclaimer.
37 * 2. Redistributions in binary form must reproduce the above copyright
38 * notice, this list of conditions and the following disclaimer in the
39 * documentation and/or other materials provided with the distribution.
40 * 3. All advertising materials mentioning features or use of this software
41 * must display the following acknowledgement:
42 * This product includes software developed by the University of
43 * California, Berkeley and its contributors.
44 * 4. Neither the name of the University nor the names of its contributors
45 * may be used to endorse or promote products derived from this software
46 * without specific prior written permission.
47 *
48 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58 * SUCH DAMAGE.
59 *
60 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
61 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.16 2001/08/22 00:59:12 silby Exp $
62 */
63
64 #include "tcp_includes.h"
65
66 #include <sys/param.h>
67 #include <sys/kernel.h>
68 #include <sys/socketvar.h>
69
70 #include <net/route.h>
71 #include <netinet/in.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/ip.h>
74 #include <netinet/ip6.h>
75 #include <netinet/ip_var.h>
76
77 int tcp_newreno_init(struct tcpcb *tp);
78 int tcp_newreno_cleanup(struct tcpcb *tp);
79 void tcp_newreno_cwnd_init_or_reset(struct tcpcb *tp);
80 void tcp_newreno_congestion_avd(struct tcpcb *tp, struct tcphdr *th);
81 void tcp_newreno_ack_rcvd(struct tcpcb *tp, struct tcphdr *th);
82 void tcp_newreno_pre_fr(struct tcpcb *tp);
83 void tcp_newreno_post_fr(struct tcpcb *tp, struct tcphdr *th);
84 void tcp_newreno_after_idle(struct tcpcb *tp);
85 void tcp_newreno_after_timeout(struct tcpcb *tp);
86 int tcp_newreno_delay_ack(struct tcpcb *tp, struct tcphdr *th);
87 void tcp_newreno_switch_cc(struct tcpcb *tp);
88
89 struct tcp_cc_algo tcp_cc_newreno = {
90 .name = "newreno",
91 .init = tcp_newreno_init,
92 .cleanup = tcp_newreno_cleanup,
93 .cwnd_init = tcp_newreno_cwnd_init_or_reset,
94 .congestion_avd = tcp_newreno_congestion_avd,
95 .ack_rcvd = tcp_newreno_ack_rcvd,
96 .pre_fr = tcp_newreno_pre_fr,
97 .post_fr = tcp_newreno_post_fr,
98 .after_idle = tcp_newreno_cwnd_init_or_reset,
99 .after_timeout = tcp_newreno_after_timeout,
100 .delay_ack = tcp_newreno_delay_ack,
101 .switch_to = tcp_newreno_switch_cc
102 };
103
104 int
tcp_newreno_init(struct tcpcb * tp)105 tcp_newreno_init(struct tcpcb *tp)
106 {
107 #pragma unused(tp)
108 os_atomic_inc(&tcp_cc_newreno.num_sockets, relaxed);
109 return 0;
110 }
111
112 int
tcp_newreno_cleanup(struct tcpcb * tp)113 tcp_newreno_cleanup(struct tcpcb *tp)
114 {
115 #pragma unused(tp)
116 os_atomic_dec(&tcp_cc_newreno.num_sockets, relaxed);
117 return 0;
118 }
119
120 /* Initialize the congestion window for a connection or
121 * handles connections that have been idle for
122 * some time. In this state, no acks are
123 * expected to clock out any data we send --
124 * slow start to get ack "clock" running again.
125 *
126 * Set the slow-start flight size depending on whether
127 * this is a local network or not.
128 */
129 void
tcp_newreno_cwnd_init_or_reset(struct tcpcb * tp)130 tcp_newreno_cwnd_init_or_reset(struct tcpcb *tp)
131 {
132 tcp_cc_cwnd_init_or_reset(tp);
133
134 tcp_update_pacer_state(tp);
135 }
136
137
138 /* Function to handle an in-sequence ack during congestion avoidance phase.
139 * This will get called from header prediction code.
140 */
141 void
tcp_newreno_congestion_avd(struct tcpcb * tp,struct tcphdr * th)142 tcp_newreno_congestion_avd(struct tcpcb *tp, struct tcphdr *th)
143 {
144 uint32_t acked = 0;
145 acked = BYTES_ACKED(th, tp);
146 /*
147 * Grow the congestion window, if the
148 * connection is cwnd bound.
149 */
150 if (tp->snd_cwnd < tp->snd_wnd) {
151 tp->t_bytes_acked += acked;
152 if (tp->t_bytes_acked > tp->snd_cwnd) {
153 tp->t_bytes_acked -= tp->snd_cwnd;
154 tp->snd_cwnd += tp->t_maxseg;
155 }
156 }
157
158 tcp_update_pacer_state(tp);
159 }
160 /* Function to process an ack.
161 */
162 void
tcp_newreno_ack_rcvd(struct tcpcb * tp,struct tcphdr * th)163 tcp_newreno_ack_rcvd(struct tcpcb *tp, struct tcphdr *th)
164 {
165 /*
166 * RFC 3465 - Appropriate Byte Counting.
167 *
168 * If the window is currently less than ssthresh,
169 * open the window by the number of bytes ACKed by
170 * the last ACK, however clamp the window increase
171 * to an upper limit "L".
172 *
173 * In congestion avoidance phase, open the window by
174 * one segment each time "bytes_acked" grows to be
175 * greater than or equal to the congestion window.
176 */
177
178 uint32_t cw = tp->snd_cwnd;
179 uint32_t incr = tp->t_maxseg;
180 uint32_t acked = 0;
181
182 acked = BYTES_ACKED(th, tp);
183 if (cw >= tp->snd_ssthresh) {
184 tp->t_bytes_acked += acked;
185 if (tp->t_bytes_acked >= cw) {
186 /* Time to increase the window. */
187 tp->t_bytes_acked -= cw;
188 } else {
189 /* No need to increase yet. */
190 incr = 0;
191 }
192 } else {
193 /*
194 * If the user explicitly enables RFC3465
195 * use 2*SMSS for the "L" param. Otherwise
196 * use the more conservative 1*SMSS.
197 *
198 * (See RFC 3465 2.3 Choosing the Limit)
199 */
200 uint32_t abc_lim;
201 abc_lim = (tp->snd_nxt == tp->snd_max) ? incr * 2 : incr;
202
203 incr = ulmin(acked, abc_lim);
204 }
205 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN << tp->snd_scale);
206
207 tcp_update_pacer_state(tp);
208 }
209
210 void
tcp_newreno_pre_fr(struct tcpcb * tp)211 tcp_newreno_pre_fr(struct tcpcb *tp)
212 {
213 uint32_t win;
214
215 win = min(tp->snd_wnd, tp->snd_cwnd) /
216 2 / tp->t_maxseg;
217 if (win < 2) {
218 win = 2;
219 }
220 tp->snd_ssthresh = win * tp->t_maxseg;
221 tcp_cc_resize_sndbuf(tp);
222 }
223
224 void
tcp_newreno_post_fr(struct tcpcb * tp,struct tcphdr * th)225 tcp_newreno_post_fr(struct tcpcb *tp, struct tcphdr *th)
226 {
227 int32_t ss;
228
229 if (th) {
230 ss = tp->snd_max - th->th_ack;
231 } else {
232 ss = tp->snd_max - tp->snd_una;
233 }
234
235 /*
236 * Complete ack. Inflate the congestion window to
237 * ssthresh and exit fast recovery.
238 *
239 * Window inflation should have left us with approx.
240 * snd_ssthresh outstanding data. But in case we
241 * would be inclined to send a burst, better to do
242 * it via the slow start mechanism.
243 *
244 * If the flight size is zero, then make congestion
245 * window to be worth at least 2 segments to avoid
246 * delayed acknowledgement (draft-ietf-tcpm-rfc3782-bis-05).
247 */
248 if (ss < (int32_t)tp->snd_ssthresh) {
249 tp->snd_cwnd = max(ss, tp->t_maxseg) + tp->t_maxseg;
250 } else {
251 tp->snd_cwnd = tp->snd_ssthresh;
252 }
253 tp->t_bytes_acked = 0;
254
255 tcp_update_pacer_state(tp);
256 }
257
258 /* Function to change the congestion window when the retransmit
259 * timer fires.
260 */
261 void
tcp_newreno_after_timeout(struct tcpcb * tp)262 tcp_newreno_after_timeout(struct tcpcb *tp)
263 {
264 /*
265 * Close the congestion window down to one segment
266 * (we'll open it by one segment for each ack we get).
267 * Since we probably have a window's worth of unacked
268 * data accumulated, this "slow start" keeps us from
269 * dumping all that data as back-to-back packets (which
270 * might overwhelm an intermediate gateway).
271 *
272 * There are two phases to the opening: Initially we
273 * open by one mss on each ack. This makes the window
274 * size increase exponentially with time. If the
275 * window is larger than the path can handle, this
276 * exponential growth results in dropped packet(s)
277 * almost immediately. To get more time between
278 * drops but still "push" the network to take advantage
279 * of improving conditions, we switch from exponential
280 * to linear window opening at some threshhold size.
281 * For a threshhold, we use half the current window
282 * size, truncated to a multiple of the mss.
283 *
284 * (the minimum cwnd that will give us exponential
285 * growth is 2 mss. We don't allow the threshhold
286 * to go below this.)
287 */
288 if (tp->t_state >= TCPS_ESTABLISHED) {
289 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
290 if (win < 2) {
291 win = 2;
292 }
293 tp->snd_ssthresh = win * tp->t_maxseg;
294
295 tp->snd_cwnd = tp->t_maxseg;
296 tcp_cc_resize_sndbuf(tp);
297
298 tcp_update_pacer_state(tp);
299 }
300 }
301
302 /*
303 * Indicate whether this ack should be delayed.
304 * We can delay the ack if:
305 * - delayed acks are enabled and set to 1, same as when value is set to 2.
306 * We kept this for binary compatibility.
307 * - delayed acks are enabled and set to 2, will "ack every other packet"
308 * - if our last ack wasn't a 0-sized window.
309 * - if the peer hasn't sent us a TH_PUSH data packet (this solves 3649245).
310 * If TH_PUSH is set, take this as a clue that we need to ACK
311 * with no delay. This helps higher level protocols who won't send
312 * us more data even if the window is open because their
313 * last "segment" hasn't been ACKed
314 * - delayed acks are enabled and set to 3, will do "streaming detection"
315 * (see the comment in tcp_input.c) and
316 * - if we receive more than "maxseg_unacked" full packets in the last 100ms
317 * - if the connection is not in slow-start or idle or loss/recovery states
318 * - if those criteria aren't met, it will ack every other packet.
319 */
320
321 int
tcp_newreno_delay_ack(struct tcpcb * tp,struct tcphdr * th)322 tcp_newreno_delay_ack(struct tcpcb *tp, struct tcphdr *th)
323 {
324 return tcp_cc_delay_ack(tp, th);
325 }
326
327 /* Switch to newreno from a different CC. If the connection is in
328 * congestion avoidance state, it can continue to use the current
329 * congestion window because it is going to be conservative. But
330 * if the connection is in slow-start, we will halve the congestion
331 * window and let newreno work from there.
332 */
333 void
tcp_newreno_switch_cc(struct tcpcb * tp)334 tcp_newreno_switch_cc(struct tcpcb *tp)
335 {
336 uint32_t cwnd = min(tp->snd_wnd, tp->snd_cwnd);
337 if (tp->snd_cwnd >= tp->snd_ssthresh) {
338 cwnd = cwnd / tp->t_maxseg;
339 } else {
340 cwnd = cwnd / 2 / tp->t_maxseg;
341 }
342 tp->snd_cwnd = max(tcp_initial_cwnd(tp), cwnd * tp->t_maxseg);
343
344 /* Start counting bytes for RFC 3465 again */
345 tp->t_bytes_acked = 0;
346
347 os_atomic_inc(&tcp_cc_newreno.num_sockets, relaxed);
348 }
349