- Uniformly use .In for header file references.
[dragonfly.git] / share / man / man4 / tcp.4
CommitLineData
984263bc
MD
1.\" Copyright (c) 1983, 1991, 1993
2.\" The Regents of the University of California. All rights reserved.
3.\"
4.\" Redistribution and use in source and binary forms, with or without
5.\" modification, are permitted provided that the following conditions
6.\" are met:
7.\" 1. Redistributions of source code must retain the above copyright
8.\" notice, this list of conditions and the following disclaimer.
9.\" 2. Redistributions in binary form must reproduce the above copyright
10.\" notice, this list of conditions and the following disclaimer in the
11.\" documentation and/or other materials provided with the distribution.
12.\" 3. All advertising materials mentioning features or use of this software
13.\" must display the following acknowledgement:
14.\" This product includes software developed by the University of
15.\" California, Berkeley and its contributors.
16.\" 4. Neither the name of the University nor the names of its contributors
17.\" may be used to endorse or promote products derived from this software
18.\" without specific prior written permission.
19.\"
20.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23.\" ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30.\" SUCH DAMAGE.
31.\"
32.\" From: @(#)tcp.4 8.1 (Berkeley) 6/5/93
33.\" $FreeBSD: src/share/man/man4/tcp.4,v 1.11.2.14 2002/12/29 16:35:38 schweikh Exp $
44cb301e 34.\" $DragonFly: src/share/man/man4/tcp.4,v 1.6 2006/05/26 19:39:39 swildner Exp $
984263bc
MD
35.\"
36.Dd February 14, 1995
37.Dt TCP 4
38.Os
39.Sh NAME
40.Nm tcp
41.Nd Internet Transmission Control Protocol
42.Sh SYNOPSIS
43.In sys/types.h
44.In sys/socket.h
45.In netinet/in.h
46.Ft int
47.Fn socket AF_INET SOCK_STREAM 0
48.Sh DESCRIPTION
49The
50.Tn TCP
51protocol provides reliable, flow-controlled, two-way
52transmission of data. It is a byte-stream protocol used to
53support the
54.Dv SOCK_STREAM
55abstraction. TCP uses the standard
56Internet address format and, in addition, provides a per-host
57collection of
58.Dq port addresses .
59Thus, each address is composed
60of an Internet address specifying the host and network, with
61a specific
62.Tn TCP
63port on the host identifying the peer entity.
64.Pp
65Sockets utilizing the tcp protocol are either
66.Dq active
67or
68.Dq passive .
69Active sockets initiate connections to passive
70sockets. By default
71.Tn TCP
72sockets are created active; to create a
73passive socket the
74.Xr listen 2
75system call must be used
76after binding the socket with the
77.Xr bind 2
78system call. Only
79passive sockets may use the
80.Xr accept 2
81call to accept incoming connections. Only active sockets may
82use the
83.Xr connect 2
84call to initiate connections.
85.Tn TCP
86also supports a more datagram-like mode, called Transaction
87.Tn TCP ,
88which is described in
89.Xr ttcp 4 .
90.Pp
91Passive sockets may
92.Dq underspecify
93their location to match
94incoming connection requests from multiple networks. This
95technique, termed
96.Dq wildcard addressing ,
97allows a single
98server to provide service to clients on multiple networks.
99To create a socket which listens on all networks, the Internet
100address
101.Dv INADDR_ANY
102must be bound. The
103.Tn TCP
104port may still be specified
105at this time; if the port is not specified the system will assign one.
106Once a connection has been established the socket's address is
107fixed by the peer entity's location. The address assigned the
108socket is the address associated with the network interface
109through which packets are being transmitted and received. Normally
110this address corresponds to the peer entity's network.
111.Pp
112.Tn TCP
113supports a number of socket options which can be set with
114.Xr setsockopt 2
115and tested with
116.Xr getsockopt 2 :
117.Bl -tag -width TCP_NODELAYx
118.It Dv TCP_NODELAY
119Under most circumstances,
120.Tn TCP
121sends data when it is presented;
122when outstanding data has not yet been acknowledged, it gathers
123small amounts of output to be sent in a single packet once
124an acknowledgement is received.
125For a small number of clients, such as window systems
126that send a stream of mouse events which receive no replies,
127this packetization may cause significant delays.
128The boolean option
129.Dv TCP_NODELAY
130defeats this algorithm.
131.It Dv TCP_MAXSEG
132By default, a sender\- and receiver-TCP
133will negotiate among themselves to determine the maximum segment size
134to be used for each connection. The
135.Dv TCP_MAXSEG
136option allows the user to determine the result of this negotiation,
137and to reduce it if desired.
138.It Dv TCP_NOOPT
139.Tn TCP
140usually sends a number of options in each packet, corresponding to
141various
142.Tn TCP
143extensions which are provided in this implementation. The boolean
144option
145.Dv TCP_NOOPT
146is provided to disable
147.Tn TCP
148option use on a per-connection basis.
149.It Dv TCP_NOPUSH
150By convention, the sender-TCP
151will set the
152.Dq push
153bit and begin transmission immediately (if permitted) at the end of
154every user call to
155.Xr write 2
156or
157.Xr writev 2 .
158The
159.Dv TCP_NOPUSH
160option is provided to allow servers to easily make use of Transaction
161TCP (see
162.Xr ttcp 4 ) .
163When the option is set to a non-zero value,
164.Tn TCP
165will delay sending any data at all until either the socket is closed,
166or the internal send buffer is filled.
167.El
168.Pp
169The option level for the
170.Xr setsockopt 2
171call is the protocol number for
172.Tn TCP ,
173available from
174.Xr getprotobyname 3 ,
175or
176.Dv IPPROTO_TCP .
177All options are declared in
44cb301e 178.In netinet/tcp.h .
984263bc
MD
179.Pp
180Options at the
181.Tn IP
182transport level may be used with
183.Tn TCP ;
184see
185.Xr ip 4 .
186Incoming connection requests that are source-routed are noted,
187and the reverse source route is used in responding.
188.Sh MIB VARIABLES
189The
190.Nm
191protocol implements a number of variables in the
192.Li net.inet
193branch of the
194.Xr sysctl 3
195MIB.
196.Bl -tag -width TCPCTL_DO_RFC1644
197.It Dv TCPCTL_DO_RFC1323
198.Pq tcp.rfc1323
199Implement the window scaling and timestamp options of RFC 1323
200(default true).
201.It Dv TCPCTL_DO_RFC1644
202.Pq tcp.rfc1644
203Implement Transaction
204.Tn TCP ,
205as described in RFC 1644.
206.It Dv TCPCTL_MSSDFLT
207.Pq tcp.mssdflt
208The default value used for the maximum segment size
209.Pq Dq MSS
210when no advice to the contrary is received from MSS negotiation.
211.It Dv TCPCTL_SENDSPACE
212.Pq tcp.sendspace
213Maximum TCP send window.
214.It Dv TCPCTL_RECVSPACE
215.Pq tcp.recvspace
216Maximum TCP receive window.
217.It tcp.log_in_vain
218Log any connection attempts to ports where there is not a socket
219accepting connections.
220The value of 1 limits the logging to SYN (connection establishment)
221packets only.
222That of 2 results in any TCP packets to closed ports being logged.
223Any value unlisted above disables the logging
224(default is 0, i.e., the logging is disabled).
984263bc
MD
225.It tcp.msl
226The Maximum Segment Lifetime for a packet.
227.It tcp.keepinit
228Timeout for new, non-established TCP connections.
229.It tcp.keepidle
230Amount of time the connection should be idle before keepalive
231probes (if enabled) are sent.
232.It tcp.keepintvl
233The interval between keepalive probes sent to remote machines.
234After
235.Dv TCPTV_KEEPCNT
236(default 8) probes are sent, with no response, the connection is dropped.
237.It tcp.always_keepalive
238Assume that
239.Dv SO_KEEPALIVE
240is set on all
241.Tn TCP
242connections, the kernel will
243periodically send a packet to the remote host to verify the connection
244is still up.
245.It tcp.icmp_may_rst
246Certain
247.Tn ICMP
248unreachable messages may abort connections in
249.Tn SYN-SENT
250state.
251.It tcp.do_tcpdrain
252Flush packets in the
253.Tn TCP
254reassembly queue if the system is low on mbufs.
255.It tcp.blackhole
256If enabled, disable sending of RST when a connection is attempted
257to a port where there is not a socket accepting connections.
258See
259.Xr blackhole 4 .
260.It tcp.delayed_ack
261Delay ACK to try and piggyback it onto a data packet.
262.It tcp.delacktime
263Maximum amount of time before a delayed ACK is sent.
264.It tcp.newreno
265Enable TCP NewReno Fast Recovery algorithm,
266as described in RFC 2582.
267.It tcp.path_mtu_discovery
7c3b84db
JH
268Enables Path MTU Discovery. PMTU Discovery is helpful for avoiding
269IP fragmentation when tranferring lots of data to the same client.
270For web servers, where most of the connections are short and to
271different clients, PMTU Discovery actually hurts performance due
272to unnecessary retransmissions. Turn this on only if most of your
273TCP connections are long transfers or are repeatedly to the same
274set of clients.
984263bc
MD
275.It tcp.tcbhashsize
276Size of the
277.Tn TCP
278control-block hashtable
279(read-only).
280This may be tuned using the kernel option
281.Dv TCBHASHSIZE
282or by setting
283.Va net.inet.tcp.tcbhashsize
284in the
285.Xr loader 8 .
286.It tcp.pcbcount
287Number of active process control blocks
288(read-only).
289.It tcp.syncookies
290Determines whether or not syn cookies should be generated for
291outbound syn-ack packets. Syn cookies are a great help during
292syn flood attacks, and are enabled by default.
293.It tcp.isn_reseed_interval
294The interval (in seconds) specifying how often the secret data used in
295RFC 1948 initial sequence number calculations should be reseeded.
296By default, this variable is set to zero, indicating that
297no reseeding will occur.
298Reseeding should not be necessary, and will break
299.Dv TIME_WAIT
300recycling for a few minutes.
301.It tcp.inet.tcp.rexmit_{min,slop}
302Adjust the retransmit timer calculation for TCP. The slop is
303typically added to the raw calculation to take into account
304occasional variances that the SRTT (smoothed round trip time)
305is unable to accomodate, while the minimum specifies an
306absolute minimum. While a number of TCP RFCs suggest a 1
307second minimum these RFCs tend to focus on streaming behavior
308and fail to deal with the fact that a 1 second minimum has severe
309detrimental effects over lossy interactive connections, such
310as a 802.11b wireless link, and over very fast but lossy
311connections for those cases not covered by the fast retransmit
312code. For this reason we suggest changing the slop to 200ms and
313setting the minimum to something out of the way, like 20ms,
314which gives you an effective minimum of 200ms (similar to Linux).
315.It tcp.inflight_enable
316Enable
317.Tn TCP
318bandwidth delay product limiting. An attempt will be made to calculate
319the bandwidth delay product for each individual TCP connection and limit
1bf4b486 320the amount of inflight data being transmitted to avoid building up
984263bc
MD
321unnecessary packets in the network. This option is recommended if you
322are serving a lot of data over connections with high bandwidth-delay
323products, such as modems, GigE links, and fast long-haul WANs, and/or
324you have configured your machine to accomodate large TCP windows. In such
325situations, without this option, you may experience high interactive
326latencies or packet loss due to the overloading of intermediate routers
327and switches. Note that bandwidth delay product limiting only effects
328the transmit side of a TCP connection.
329.It tcp.inflight_debug
330Enable debugging for the bandwidth delay product algorithm. This may
331default to on (1) so if you enable the algorithm you should probably also
332disable debugging by setting this variable to 0.
333.It tcp.inflight_min
334This puts an lower bound on the bandwidth delay product window, in bytes.
335A value of 1024 is typically used for debugging. 6000-16000 is more typical
336in a production installation. Setting this value too low may result in
337slow ramp-up times for bursty connections. Setting this value too high
338effectively disables the algorithm.
339.It tcp.inflight_max
340This puts an upper bound on the bandwidth delay product window, in bytes.
341This value should not generally be modified but may be used to set a
342global per-connection limit on queued data, potentially allowing you to
343intentionally set a less then optimum limit to smooth data flow over a
344network while still being able to specify huge internal TCP buffers.
345.It tcp.inflight_stab
346The bandwidth delay product algorithm requires a slightly larger window
347then it otherwise calculates for stability. This parameter determines the
1bf4b486 348extra window in maximal packets / 10. The default value of 20 represents
984263bc
MD
3492 maximal packets. Reducing this value is not recommended but you may
350come across a situation with very slow links where the ping time
351reduction of the default inflight code is not sufficient. If this case
352occurs you should first try reducing tcp.inflight_min and, if that does not
353work, reduce both tcp.inflight_min and tcp.inflight_stab, trying values of
35415, 10, or 5 for the latter. Never use a value less then 5. Reducing
355tcp.inflight_stab can lead to upwards of a 20% underutilization of the link
356as well as reducing the algorithm's ability to adapt to changing
357situations and should only be done as a last resort.
358.El
359.Sh ERRORS
360A socket operation may fail with one of the following errors returned:
361.Bl -tag -width Er
362.It Bq Er EISCONN
363when trying to establish a connection on a socket which
364already has one;
365.It Bq Er ENOBUFS
366when the system runs out of memory for
367an internal data structure;
368.It Bq Er ETIMEDOUT
369when a connection was dropped
370due to excessive retransmissions;
371.It Bq Er ECONNRESET
372when the remote peer
373forces the connection to be closed;
374.It Bq Er ECONNREFUSED
375when the remote
376peer actively refuses connection establishment (usually because
377no process is listening to the port);
378.It Bq Er EADDRINUSE
379when an attempt
380is made to create a socket with a port which has already been
381allocated;
382.It Bq Er EADDRNOTAVAIL
383when an attempt is made to create a
384socket with a network address for which no network interface
385exists.
386.It Bq Er EAFNOSUPPORT
387when an attempt is made to bind or connect a socket to a multicast
388address.
389.El
390.Sh SEE ALSO
391.Xr getsockopt 2 ,
392.Xr socket 2 ,
393.Xr sysctl 3 ,
394.Xr blackhole 4 ,
395.Xr inet 4 ,
396.Xr intro 4 ,
397.Xr ip 4 ,
398.Xr ttcp 4
399.Rs
400.%A V. Jacobson
401.%A R. Braden
402.%A D. Borman
403.%T "TCP Extensions for High Performance"
404.%O RFC 1323
405.Re
406.Rs
407.%A R. Braden
408.%T "T/TCP \- TCP Extensions for Transactions"
409.%O RFC 1644
410.Re
411.Sh HISTORY
412The
413.Nm
414protocol appeared in
415.Bx 4.2 .
416The RFC 1323 extensions for window scaling and timestamps were added
417in
418.Bx 4.4 .