1/*
2 * Copyright (c) 2000-2014 Apple Computer, 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, 1993
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_timer.h 8.1 (Berkeley) 6/10/93
61 * $FreeBSD: src/sys/netinet/tcp_timer.h,v 1.18 1999/12/29 04:41:03 peter Exp $
62 */
63
64#ifndef _NETINET_TCP_TIMER_H_
65#define _NETINET_TCP_TIMER_H_
66#include <sys/appleapiopts.h>
67
68#ifdef BSD_KERNEL_PRIVATE
69#include <kern/thread_call.h>
70#endif /* BSD_KERNEL_PRIVATE */
71
72/* Keep the external definition the same for binary compatibility */
73#define TCPT_NTIMERS_EXT 4
74
75/*
76 * Definitions of the TCP timers.
77 *
78 * The TCPT_PTO timer is used for probing for a tail loss in a send window.
79 * If this probe gets acknowledged using SACK, it will allow the connection
80 * to enter fast-recovery instead of hitting a retransmit timeout. A probe
81 * timeout will send the last unacknowledged segment to generate more acks
82 * with SACK information which can be used for fast-retransmiting the lost
83 * packets. This will fire in the order of 10ms.
84 *
85 * The TCPT_REXMT timer is used to force retransmissions.
86 * The TCP has the TCPT_REXMT timer set whenever segments
87 * have been sent for which ACKs are expected but not yet
88 * received. If an ACK is received which advances tp->snd_una,
89 * then the retransmit timer is cleared (if there are no more
90 * outstanding segments) or reset to the base value (if there
91 * are more ACKs expected). Whenever the retransmit timer goes off,
92 * we retransmit one unacknowledged segment, and do a backoff
93 * on the retransmit timer.
94 *
95 * The TCPT_DELACK timer is used for transmitting delayed acknowledgements
96 * if an acknowledgement was delayed in anticipation of a new segment.
97 *
98 * The TCPT_PERSIST timer is used to keep window size information
99 * flowing even if the window goes shut. If all previous transmissions
100 * have been acknowledged(so that there are no retransmissions in progress),
101 * and the window is too small to bother sending anything, then we start
102 * the TCPT_PERSIST timer. When it expires, if the window is nonzero,
103 * we go to transmit state. Otherwise, at intervals send a single byte
104 * into the peer's window to force him to update our window information.
105 * We do this at most as often as TCPT_PERSMIN time intervals,
106 * but no more frequently than the current estimate of round-trip
107 * packet time. The TCPT_PERSIST timer is cleared whenever we receive
108 * a window update from the peer.
109 *
110 * The TCPT_KEEP timer is used to keep connections alive. If an
111 * connection is idle (no segments received) for TCPTV_KEEP_INIT amount
112 * of time, but not yet established, then we drop the connection.
113 * Once the connection is established, if the connection is idle for
114 * TCPTV_KEEP_IDLE time (and keepalives have been enabled on the socket),
115 * we begin to probe the connection. We force the peer to send us a
116 * segment by sending:
117 * <SEQ=SND.UNA-1><ACK=RCV.NXT><CTL=ACK>
118 * This segment is (deliberately) outside the window, and should elicit
119 * an ack segment in response from the peer. If, despite the TCPT_KEEP
120 * initiated segments we cannot elicit a response from a peer in
121 * TCPT_MAXIDLE amount of time probing, then we drop the connection.
122 *
123 * The TCPT_2MSL timer is used for keeping the conenction in Time-wait state
124 * before fully closing it so that the connection 4-tuple can be reused.
125 */
126#ifdef BSD_KERNEL_PRIVATE
127
128#define TCPT_PTO 0 /* Probe timeout */
129#define TCPT_DELAYFR 1 /* Delay recovery if there is reordering */
130#define TCPT_REXMT 2 /* retransmit */
131#define TCPT_DELACK 3 /* delayed ack */
132#define TCPT_PERSIST 4 /* retransmit persistence */
133#define TCPT_KEEP 5 /* keep alive */
134#define TCPT_2MSL 6 /* 2*msl quiet time timer */
135#if MPTCP
136#define TCPT_JACK_RXMT 7 /* retransmit timer for join ack */
137#define TCPT_MAX 7
138#else /* MPTCP */
139#define TCPT_MAX 6
140#endif /* !MPTCP */
141
142#define TCPT_NONE (TCPT_MAX + 1)
143#define TCPT_NTIMERS (TCPT_MAX + 1)
144
145/* External definitions */
146#define TCPT_REXMT_EXT 0
147#define TCPT_PERSIST_EXT 1
148#define TCPT_KEEP_EXT 2
149#define TCPT_2MSL_EXT 3
150#define TCPT_DELACK_EXT 4
151
152#else /* !BSD_KERNEL_PRIVATE */
153#define TCPT_REXMT 0 /* retransmit */
154#define TCPT_PERSIST 1 /* retransmit persistence */
155#define TCPT_KEEP 2 /* keep alive */
156#define TCPT_2MSL 3 /* 2*msl quiet time timer */
157#define TCPT_DELACK 4 /* delayed ack timer */
158#if MPTCP
159#define TCPT_JACK_RXMT 5 /* retransmit timer for join ack */
160#define TCPT_MAX 5
161#else /* MPTCP */
162#define TCPT_MAX 4
163#endif /* !MPTCP */
164#define TCPT_NONE (TCPT_MAX + 1)
165#define TCPT_NTIMERS (TCPT_MAX + 1)
166
167#endif /* BSD_KERNEL_PRIVATE */
168
169#ifdef BSD_KERNEL_PRIVATE
170/*
171 * Time constants.
172 */
173#define TCPTV_MSL ( 15*TCP_RETRANSHZ) /* max seg lifetime */
174#define TCPTV_SRTTBASE 0 /* base roundtrip time; if 0, no idea yet */
175#define TCPTV_RTOBASE ( 1*TCP_RETRANSHZ) /* assumed RTO if no info */
176#define TCPTV_SRTTDFLT ( 1*TCP_RETRANSHZ) /* assumed RTT if no info */
177#define TCPTV_PERSMIN ( 5*TCP_RETRANSHZ) /* retransmit persistence */
178#define TCPTV_PERSMAX ( 60*TCP_RETRANSHZ) /* maximum persist interval */
179
180extern int tcptv_persmin_val;
181
182#define TCPTV_KEEP_INIT ( 75*TCP_RETRANSHZ) /* connect keep alive */
183#define TCPTV_KEEP_IDLE (120*60*TCP_RETRANSHZ) /* time before probing */
184#define TCPTV_KEEPINTVL ( 75*TCP_RETRANSHZ) /* default probe interval */
185#define TCPTV_KEEPCNT 8 /* max probes before drop */
186
187#define TCPTV_REXMTMAX ( 64*TCP_RETRANSHZ ) /* max REXMT value */
188#define TCPTV_REXMTMIN ( TCP_RETRANSHZ/33 ) /* min REXMT for non-local connections */
189
190/*
191 * Window for counting received bytes to see if ack-stretching
192 * can start (default 100 ms)
193 */
194#define TCPTV_UNACKWIN ( TCP_RETRANSHZ/10 )
195
196/* Receiver idle time, avoid ack-stretching after this idle time */
197#define TCPTV_MAXRCVIDLE (TCP_RETRANSHZ/5 )
198
199/*
200 * No ack stretching during slow-start, until we see some packets.
201 * By the time the receiver gets 512 packets, the senders cwnd
202 * should open by a few hundred packets consdering the
203 * slow-start progression.
204 */
205#define TCP_RCV_SS_PKTCOUNT 512
206
207/* Receiver idle time, for rcv socket buffer resizing */
208#define TCPTV_RCVBUFIDLE (TCP_RETRANSHZ/2)
209#define TCPTV_TWTRUNC 8 /* RTO factor to truncate TW */
210
211#define TCP_LINGERTIME 120 /* linger at most 2 minutes */
212
213#define TCP_MAXRXTSHIFT 12 /* maximum retransmits */
214
215#ifdef TCPTIMERS
216static char *tcptimers[] =
217 { "REXMT", "PERSIST", "KEEP", "2MSL" , "DELACK"};
218#endif /* TCPTIMERS */
219
220/*
221 * Persist, keep, 2msl and MPTCP's join-ack timer as slow timers which can
222 * be coalesced at a higher granularity (500 ms).
223 *
224 * Rexmt and delayed ack timers are considered as fast timers which run
225 * in the order of 100ms.
226 *
227 * Probe timeout is a quick timer which will run in the order of 10ms.
228 */
229#define IS_TIMER_HZ_500MS(i) ((i) >= TCPT_PERSIST)
230#define IS_TIMER_HZ_100MS(i) ((i) >= TCPT_REXMT && (i) < TCPT_PERSIST)
231#define IS_TIMER_HZ_10MS(i) ((i) < TCPT_REXMT)
232
233struct tcptimerlist;
234
235struct tcptimerentry {
236 LIST_ENTRY(tcptimerentry) le; /* links for timer list */
237 uint32_t timer_start; /* tcp clock when the timer was started */
238 uint16_t index; /* index of lowest timer that needs to run first */
239 uint16_t mode; /* Bit-wise OR of timers that are active */
240 uint32_t runtime; /* deadline at which the first timer has to fire */
241};
242
243LIST_HEAD(timerlisthead, tcptimerentry);
244
245struct tcptimerlist {
246 struct timerlisthead lhead; /* head of the list */
247 lck_mtx_t *mtx; /* lock to protect the list */
248 lck_attr_t *mtx_attr; /* mutex attributes */
249 lck_grp_t *mtx_grp; /* mutex group definition */
250 lck_grp_attr_t *mtx_grp_attr; /* mutex group attributes */
251 thread_call_t call; /* call entry */
252 uint32_t runtime; /* time at which this list is going to run */
253 uint32_t schedtime; /* time at which this list was scheduled */
254 uint32_t entries; /* Number of entries on the list */
255 uint32_t maxentries; /* Max number of entries at any time */
256
257 /* Set desired mode when timer list running */
258 boolean_t running; /* Set when timer list is being processed */
259 boolean_t scheduled; /* set when the timer is scheduled */
260#define TCP_TIMERLIST_10MS_MODE 0x1
261#define TCP_TIMERLIST_100MS_MODE 0x2
262#define TCP_TIMERLIST_500MS_MODE 0x4
263 uint32_t mode; /* Current mode of the timer */
264 uint32_t pref_mode; /* Preferred mode set by a connection */
265 uint32_t pref_offset; /* Preferred offset set by a connection */
266 uint32_t idleruns; /* Number of times the list has been idle in fast mode */
267 struct tcptimerentry *next_te; /* next timer entry pointer to process */
268 u_int16_t probe_if_index; /* Interface index that needs to send probes */
269
270};
271
272/* number of idle runs allowed for TCP timer list in fast or quick modes */
273#define TCP_FASTMODE_IDLERUN_MAX 10
274
275/*
276 * Minimum retransmit timeout is set to 30ms. We add a slop of
277 * 200 ms to the retransmit value to account for processing
278 * variance and delayed ack. This extra 200ms will help to avoid
279 * spurious retransmits by taking into consideration the receivers
280 * that wait for delayed ack timer instead of generating an ack
281 * for every two packets.
282 *
283 * On a local link, the minimum retransmit timeout is 100ms and
284 * variance is set to 0. This will make the sender a little bit more
285 * aggressive on local link. When the connection is not established yet,
286 * there is no need to add an extra 200ms to retransmit timeout because
287 * the initial value is high (1s) and delayed ack is not a problem in
288 * that case.
289 */
290#define TCPTV_REXMTSLOP ( TCP_RETRANSHZ/5 ) /* extra 200 ms slop */
291
292/* macro to decide when retransmit slop (described above) should be added */
293#define TCP_ADD_REXMTSLOP(tp) (tp->t_state >= TCPS_ESTABLISHED)
294
295#define TCPT_RANGESET(tv, value, tvmin, tvmax, addslop) do { \
296 (tv) = ((addslop) ? tcp_rexmt_slop : 0) + (value); \
297 if ((uint32_t)(tv) < (uint32_t)(tvmin)) \
298 (tv) = (tvmin); \
299 else if ((uint32_t)(tv) > (uint32_t)(tvmax)) \
300 (tv) = (tvmax); \
301} while(0)
302
303#define TCP_CONN_KEEPIDLE(tp) \
304 ((tp)->t_keepidle && \
305 ((tp)->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) ? \
306 (tp)->t_keepidle : tcp_keepidle)
307#define TCP_CONN_KEEPINIT(tp) \
308 (((tp)->t_keepinit > 0) ? (tp)->t_keepinit : tcp_keepinit)
309#define TCP_CONN_KEEPCNT(tp) \
310 (((tp)->t_keepcnt > 0) ? (tp)->t_keepcnt : tcp_keepcnt)
311#define TCP_CONN_KEEPINTVL(tp) \
312 (((tp)->t_keepintvl > 0) ? (tp)->t_keepintvl : tcp_keepintvl)
313#define TCP_CONN_MAXIDLE(tp) \
314 (TCP_CONN_KEEPCNT(tp) * TCP_CONN_KEEPINTVL(tp))
315
316#define TCP_IDLETIMEOUT(tp) \
317 (((TCP_ADD_REXMTSLOP(tp)) ? 0 : tcp_rexmt_slop) + tp->t_rxtcur)
318
319TAILQ_HEAD(tcptailq, tcpcb);
320
321extern int tcp_keepinit; /* time to establish connection */
322extern int tcp_keepidle; /* time before keepalive probes begin */
323extern int tcp_keepintvl; /* time between keepalive probes */
324extern int tcp_keepcnt; /* number of keepalives */
325extern int tcp_delack; /* delayed ack timer */
326extern int tcp_maxpersistidle;
327extern int tcp_msl;
328extern int tcp_ttl; /* time to live for TCP segs */
329extern int tcp_backoff[];
330extern int tcp_rexmt_slop;
331extern u_int32_t tcp_max_persist_timeout; /* Maximum persistence for Zero Window Probes */
332
333#define OFFSET_FROM_START(tp, off) ((tcp_now + (off)) - (tp)->tentry.timer_start)
334
335#endif /* BSD_KERNEL_PRIVATE */
336#endif /* !_NETINET_TCP_TIMER_H_ */
337
338