1 /*
2 * Copyright (c) 2010-2014 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 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/kernel.h>
66 #include <sys/protosw.h>
67 #include <sys/socketvar.h>
68
69 #include <net/route.h>
70 #include <netinet/in.h>
71 #include <netinet/in_systm.h>
72 #include <netinet/ip.h>
73
74 #include <netinet/ip6.h>
75 #include <netinet/ip_var.h>
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_fsm.h>
78 #include <netinet/tcp_timer.h>
79 #include <netinet/tcp_var.h>
80 #include <netinet/tcpip.h>
81 #include <netinet/tcp_cc.h>
82 #include <libkern/OSAtomic.h>
83
84 int tcp_newreno_init(struct tcpcb *tp);
85 int tcp_newreno_cleanup(struct tcpcb *tp);
86 void tcp_newreno_cwnd_init_or_reset(struct tcpcb *tp);
87 void tcp_newreno_congestion_avd(struct tcpcb *tp, struct tcphdr *th);
88 void tcp_newreno_ack_rcvd(struct tcpcb *tp, struct tcphdr *th);
89 void tcp_newreno_pre_fr(struct tcpcb *tp);
90 void tcp_newreno_post_fr(struct tcpcb *tp, struct tcphdr *th);
91 void tcp_newreno_after_idle(struct tcpcb *tp);
92 void tcp_newreno_after_timeout(struct tcpcb *tp);
93 int tcp_newreno_delay_ack(struct tcpcb *tp, struct tcphdr *th);
94 void tcp_newreno_switch_cc(struct tcpcb *tp);
95
96 struct tcp_cc_algo tcp_cc_newreno = {
97 .name = "newreno",
98 .init = tcp_newreno_init,
99 .cleanup = tcp_newreno_cleanup,
100 .cwnd_init = tcp_newreno_cwnd_init_or_reset,
101 .congestion_avd = tcp_newreno_congestion_avd,
102 .ack_rcvd = tcp_newreno_ack_rcvd,
103 .pre_fr = tcp_newreno_pre_fr,
104 .post_fr = tcp_newreno_post_fr,
105 .after_idle = tcp_newreno_cwnd_init_or_reset,
106 .after_timeout = tcp_newreno_after_timeout,
107 .delay_ack = tcp_newreno_delay_ack,
108 .switch_to = tcp_newreno_switch_cc
109 };
110
111 int
tcp_newreno_init(struct tcpcb * tp)112 tcp_newreno_init(struct tcpcb *tp)
113 {
114 #pragma unused(tp)
115 OSIncrementAtomic((volatile SInt32 *)&tcp_cc_newreno.num_sockets);
116 return 0;
117 }
118
119 int
tcp_newreno_cleanup(struct tcpcb * tp)120 tcp_newreno_cleanup(struct tcpcb *tp)
121 {
122 #pragma unused(tp)
123 OSDecrementAtomic((volatile SInt32 *)&tcp_cc_newreno.num_sockets);
124 return 0;
125 }
126
127 /* Initialize the congestion window for a connection or
128 * handles connections that have been idle for
129 * some time. In this state, no acks are
130 * expected to clock out any data we send --
131 * slow start to get ack "clock" running again.
132 *
133 * Set the slow-start flight size depending on whether
134 * this is a local network or not.
135 */
136 void
tcp_newreno_cwnd_init_or_reset(struct tcpcb * tp)137 tcp_newreno_cwnd_init_or_reset(struct tcpcb *tp)
138 {
139 tcp_cc_cwnd_init_or_reset(tp);
140 }
141
142
143 /* Function to handle an in-sequence ack during congestion avoidance phase.
144 * This will get called from header prediction code.
145 */
146 void
tcp_newreno_congestion_avd(struct tcpcb * tp,struct tcphdr * th)147 tcp_newreno_congestion_avd(struct tcpcb *tp, struct tcphdr *th)
148 {
149 uint32_t acked = 0;
150 acked = BYTES_ACKED(th, tp);
151 /*
152 * Grow the congestion window, if the
153 * connection is cwnd bound.
154 */
155 if (tp->snd_cwnd < tp->snd_wnd) {
156 tp->t_bytes_acked += acked;
157 if (tp->t_bytes_acked > tp->snd_cwnd) {
158 tp->t_bytes_acked -= tp->snd_cwnd;
159 tp->snd_cwnd += tp->t_maxseg;
160 }
161 }
162 }
163 /* Function to process an ack.
164 */
165 void
tcp_newreno_ack_rcvd(struct tcpcb * tp,struct tcphdr * th)166 tcp_newreno_ack_rcvd(struct tcpcb *tp, struct tcphdr *th)
167 {
168 /*
169 * RFC 3465 - Appropriate Byte Counting.
170 *
171 * If the window is currently less than ssthresh,
172 * open the window by the number of bytes ACKed by
173 * the last ACK, however clamp the window increase
174 * to an upper limit "L".
175 *
176 * In congestion avoidance phase, open the window by
177 * one segment each time "bytes_acked" grows to be
178 * greater than or equal to the congestion window.
179 */
180
181 uint32_t cw = tp->snd_cwnd;
182 uint32_t incr = tp->t_maxseg;
183 uint32_t acked = 0;
184
185 acked = BYTES_ACKED(th, tp);
186 if (cw >= tp->snd_ssthresh) {
187 tp->t_bytes_acked += acked;
188 if (tp->t_bytes_acked >= cw) {
189 /* Time to increase the window. */
190 tp->t_bytes_acked -= cw;
191 } else {
192 /* No need to increase yet. */
193 incr = 0;
194 }
195 } else {
196 /*
197 * If the user explicitly enables RFC3465
198 * use 2*SMSS for the "L" param. Otherwise
199 * use the more conservative 1*SMSS.
200 *
201 * (See RFC 3465 2.3 Choosing the Limit)
202 */
203 uint32_t abc_lim;
204 abc_lim = (tp->snd_nxt == tp->snd_max) ? incr * 2 : incr;
205
206 incr = ulmin(acked, abc_lim);
207 }
208 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN << tp->snd_scale);
209 }
210
211 void
tcp_newreno_pre_fr(struct tcpcb * tp)212 tcp_newreno_pre_fr(struct tcpcb *tp)
213 {
214 uint32_t win;
215
216 win = min(tp->snd_wnd, tp->snd_cwnd) /
217 2 / tp->t_maxseg;
218 if (win < 2) {
219 win = 2;
220 }
221 tp->snd_ssthresh = win * tp->t_maxseg;
222 tcp_cc_resize_sndbuf(tp);
223 }
224
225 void
tcp_newreno_post_fr(struct tcpcb * tp,struct tcphdr * th)226 tcp_newreno_post_fr(struct tcpcb *tp, struct tcphdr *th)
227 {
228 int32_t ss;
229
230 if (th) {
231 ss = tp->snd_max - th->th_ack;
232 } else {
233 ss = tp->snd_max - tp->snd_una;
234 }
235
236 /*
237 * Complete ack. Inflate the congestion window to
238 * ssthresh and exit fast recovery.
239 *
240 * Window inflation should have left us with approx.
241 * snd_ssthresh outstanding data. But in case we
242 * would be inclined to send a burst, better to do
243 * it via the slow start mechanism.
244 *
245 * If the flight size is zero, then make congestion
246 * window to be worth at least 2 segments to avoid
247 * delayed acknowledgement (draft-ietf-tcpm-rfc3782-bis-05).
248 */
249 if (ss < (int32_t)tp->snd_ssthresh) {
250 tp->snd_cwnd = max(ss, tp->t_maxseg) + tp->t_maxseg;
251 } else {
252 tp->snd_cwnd = tp->snd_ssthresh;
253 }
254 tp->t_bytes_acked = 0;
255 }
256
257 /* Function to change the congestion window when the retransmit
258 * timer fires.
259 */
260 void
tcp_newreno_after_timeout(struct tcpcb * tp)261 tcp_newreno_after_timeout(struct tcpcb *tp)
262 {
263 /*
264 * Close the congestion window down to one segment
265 * (we'll open it by one segment for each ack we get).
266 * Since we probably have a window's worth of unacked
267 * data accumulated, this "slow start" keeps us from
268 * dumping all that data as back-to-back packets (which
269 * might overwhelm an intermediate gateway).
270 *
271 * There are two phases to the opening: Initially we
272 * open by one mss on each ack. This makes the window
273 * size increase exponentially with time. If the
274 * window is larger than the path can handle, this
275 * exponential growth results in dropped packet(s)
276 * almost immediately. To get more time between
277 * drops but still "push" the network to take advantage
278 * of improving conditions, we switch from exponential
279 * to linear window opening at some threshhold size.
280 * For a threshhold, we use half the current window
281 * size, truncated to a multiple of the mss.
282 *
283 * (the minimum cwnd that will give us exponential
284 * growth is 2 mss. We don't allow the threshhold
285 * to go below this.)
286 */
287 if (tp->t_state >= TCPS_ESTABLISHED) {
288 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
289 if (win < 2) {
290 win = 2;
291 }
292 tp->snd_ssthresh = win * tp->t_maxseg;
293
294 tp->snd_cwnd = tp->t_maxseg;
295 tcp_cc_resize_sndbuf(tp);
296 }
297 }
298
299 /*
300 * Indicate whether this ack should be delayed.
301 * We can delay the ack if:
302 * - delayed acks are enabled and set to 1, same as when value is set to 2.
303 * We kept this for binary compatibility.
304 * - delayed acks are enabled and set to 2, will "ack every other packet"
305 * - if our last ack wasn't a 0-sized window.
306 * - if the peer hasn't sent us a TH_PUSH data packet (this solves 3649245).
307 * If TH_PUSH is set, take this as a clue that we need to ACK
308 * with no delay. This helps higher level protocols who won't send
309 * us more data even if the window is open because their
310 * last "segment" hasn't been ACKed
311 * - delayed acks are enabled and set to 3, will do "streaming detection"
312 * (see the comment in tcp_input.c) and
313 * - if we receive more than "maxseg_unacked" full packets in the last 100ms
314 * - if the connection is not in slow-start or idle or loss/recovery states
315 * - if those criteria aren't met, it will ack every other packet.
316 */
317
318 int
tcp_newreno_delay_ack(struct tcpcb * tp,struct tcphdr * th)319 tcp_newreno_delay_ack(struct tcpcb *tp, struct tcphdr *th)
320 {
321 return tcp_cc_delay_ack(tp, th);
322 }
323
324 /* Switch to newreno from a different CC. If the connection is in
325 * congestion avoidance state, it can continue to use the current
326 * congestion window because it is going to be conservative. But
327 * if the connection is in slow-start, we will halve the congestion
328 * window and let newreno work from there.
329 */
330 void
tcp_newreno_switch_cc(struct tcpcb * tp)331 tcp_newreno_switch_cc(struct tcpcb *tp)
332 {
333 uint32_t cwnd = min(tp->snd_wnd, tp->snd_cwnd);
334 if (tp->snd_cwnd >= tp->snd_ssthresh) {
335 cwnd = cwnd / tp->t_maxseg;
336 } else {
337 cwnd = cwnd / 2 / tp->t_maxseg;
338 }
339 tp->snd_cwnd = max(tcp_initial_cwnd(tp), cwnd * tp->t_maxseg);
340
341 /* Start counting bytes for RFC 3465 again */
342 tp->t_bytes_acked = 0;
343
344 OSIncrementAtomic((volatile SInt32 *)&tcp_cc_newreno.num_sockets);
345 }
346