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 $ 34.\" 35.Dd February 14, 1995 36.Dt TCP 4 37.Os 38.Sh NAME 39.Nm tcp 40.Nd Internet Transmission Control Protocol 41.Sh SYNOPSIS 42.In sys/types.h 43.In sys/socket.h 44.In netinet/in.h 45.Ft int 46.Fn socket AF_INET SOCK_STREAM 0 47.Sh DESCRIPTION 48The 49.Tn TCP 50protocol provides reliable, flow-controlled, two-way 51transmission of data. It is a byte-stream protocol used to 52support the 53.Dv SOCK_STREAM 54abstraction. TCP uses the standard 55Internet address format and, in addition, provides a per-host 56collection of 57.Dq port addresses . 58Thus, each address is composed 59of an Internet address specifying the host and network, with 60a specific 61.Tn TCP 62port on the host identifying the peer entity. 63.Pp 64Sockets utilizing the tcp protocol are either 65.Dq active 66or 67.Dq passive . 68Active sockets initiate connections to passive 69sockets. By default 70.Tn TCP 71sockets are created active; to create a 72passive socket the 73.Xr listen 2 74system call must be used 75after binding the socket with the 76.Xr bind 2 77system call. Only 78passive sockets may use the 79.Xr accept 2 80call to accept incoming connections. Only active sockets may 81use the 82.Xr connect 2 83call to initiate connections. 84.Tn TCP 85also supports a more datagram-like mode, called Transaction 86.Tn TCP , 87which is described in 88.Xr ttcp 4 . 89.Pp 90Passive sockets may 91.Dq underspecify 92their location to match 93incoming connection requests from multiple networks. This 94technique, termed 95.Dq wildcard addressing , 96allows a single 97server to provide service to clients on multiple networks. 98To create a socket which listens on all networks, the Internet 99address 100.Dv INADDR_ANY 101must be bound. The 102.Tn TCP 103port may still be specified 104at this time; if the port is not specified the system will assign one. 105Once a connection has been established the socket's address is 106fixed by the peer entity's location. The address assigned the 107socket is the address associated with the network interface 108through which packets are being transmitted and received. Normally 109this address corresponds to the peer entity's network. 110.Pp 111.Tn TCP 112supports a number of socket options which can be set with 113.Xr setsockopt 2 114and tested with 115.Xr getsockopt 2 : 116.Bl -tag -width TCP_NODELAYx 117.It Dv TCP_NODELAY 118Under most circumstances, 119.Tn TCP 120sends data when it is presented; 121when outstanding data has not yet been acknowledged, it gathers 122small amounts of output to be sent in a single packet once 123an acknowledgement is received. 124For a small number of clients, such as window systems 125that send a stream of mouse events which receive no replies, 126this packetization may cause significant delays. 127The boolean option 128.Dv TCP_NODELAY 129defeats this algorithm. 130.It Dv TCP_MAXSEG 131By default, a sender\- and receiver-TCP 132will negotiate among themselves to determine the maximum segment size 133to be used for each connection. The 134.Dv TCP_MAXSEG 135option allows the user to determine the result of this negotiation, 136and to reduce it if desired. 137.It Dv TCP_NOOPT 138.Tn TCP 139usually sends a number of options in each packet, corresponding to 140various 141.Tn TCP 142extensions which are provided in this implementation. The boolean 143option 144.Dv TCP_NOOPT 145is provided to disable 146.Tn TCP 147option use on a per-connection basis. 148.It Dv TCP_NOPUSH 149By convention, the sender-TCP 150will set the 151.Dq push 152bit and begin transmission immediately (if permitted) at the end of 153every user call to 154.Xr write 2 155or 156.Xr writev 2 . 157The 158.Dv TCP_NOPUSH 159option is provided to allow servers to easily make use of Transaction 160TCP (see 161.Xr ttcp 4 ) . 162When the option is set to a non-zero value, 163.Tn TCP 164will delay sending any data at all until either the socket is closed, 165or the internal send buffer is filled. 166.El 167.Pp 168The option level for the 169.Xr setsockopt 2 170call is the protocol number for 171.Tn TCP , 172available from 173.Xr getprotobyname 3 , 174or 175.Dv IPPROTO_TCP . 176All options are declared in 177.Aq Pa netinet/tcp.h . 178.Pp 179Options at the 180.Tn IP 181transport level may be used with 182.Tn TCP ; 183see 184.Xr ip 4 . 185Incoming connection requests that are source-routed are noted, 186and the reverse source route is used in responding. 187.Sh MIB VARIABLES 188The 189.Nm 190protocol implements a number of variables in the 191.Li net.inet 192branch of the 193.Xr sysctl 3 194MIB. 195.Bl -tag -width TCPCTL_DO_RFC1644 196.It Dv TCPCTL_DO_RFC1323 197.Pq tcp.rfc1323 198Implement the window scaling and timestamp options of RFC 1323 199(default true). 200.It Dv TCPCTL_DO_RFC1644 201.Pq tcp.rfc1644 202Implement Transaction 203.Tn TCP , 204as described in RFC 1644. 205.It Dv TCPCTL_MSSDFLT 206.Pq tcp.mssdflt 207The default value used for the maximum segment size 208.Pq Dq MSS 209when no advice to the contrary is received from MSS negotiation. 210.It Dv TCPCTL_SENDSPACE 211.Pq tcp.sendspace 212Maximum TCP send window. 213.It Dv TCPCTL_RECVSPACE 214.Pq tcp.recvspace 215Maximum TCP receive window. 216.It tcp.log_in_vain 217Log any connection attempts to ports where there is not a socket 218accepting connections. 219The value of 1 limits the logging to SYN (connection establishment) 220packets only. 221That of 2 results in any TCP packets to closed ports being logged. 222Any value unlisted above disables the logging 223(default is 0, i.e., the logging is disabled). 224.It tcp.slowstart_flightsize 225The number of packets allowed to be in-flight during the 226.Tn TCP 227slow-start phase on a non-local network. 228.It tcp.local_slowstart_flightsize 229The number of packets allowed to be in-flight during the 230.Tn TCP 231slow-start phase to local machines in the same subnet. 232.It tcp.msl 233The Maximum Segment Lifetime for a packet. 234.It tcp.keepinit 235Timeout for new, non-established TCP connections. 236.It tcp.keepidle 237Amount of time the connection should be idle before keepalive 238probes (if enabled) are sent. 239.It tcp.keepintvl 240The interval between keepalive probes sent to remote machines. 241After 242.Dv TCPTV_KEEPCNT 243(default 8) probes are sent, with no response, the connection is dropped. 244.It tcp.always_keepalive 245Assume that 246.Dv SO_KEEPALIVE 247is set on all 248.Tn TCP 249connections, the kernel will 250periodically send a packet to the remote host to verify the connection 251is still up. 252.It tcp.icmp_may_rst 253Certain 254.Tn ICMP 255unreachable messages may abort connections in 256.Tn SYN-SENT 257state. 258.It tcp.do_tcpdrain 259Flush packets in the 260.Tn TCP 261reassembly queue if the system is low on mbufs. 262.It tcp.blackhole 263If enabled, disable sending of RST when a connection is attempted 264to a port where there is not a socket accepting connections. 265See 266.Xr blackhole 4 . 267.It tcp.delayed_ack 268Delay ACK to try and piggyback it onto a data packet. 269.It tcp.delacktime 270Maximum amount of time before a delayed ACK is sent. 271.It tcp.newreno 272Enable TCP NewReno Fast Recovery algorithm, 273as described in RFC 2582. 274.It tcp.path_mtu_discovery 275Enable Path MTU Discovery 276.It tcp.tcbhashsize 277Size of the 278.Tn TCP 279control-block hashtable 280(read-only). 281This may be tuned using the kernel option 282.Dv TCBHASHSIZE 283or by setting 284.Va net.inet.tcp.tcbhashsize 285in the 286.Xr loader 8 . 287.It tcp.pcbcount 288Number of active process control blocks 289(read-only). 290.It tcp.syncookies 291Determines whether or not syn cookies should be generated for 292outbound syn-ack packets. Syn cookies are a great help during 293syn flood attacks, and are enabled by default. 294.It tcp.isn_reseed_interval 295The interval (in seconds) specifying how often the secret data used in 296RFC 1948 initial sequence number calculations should be reseeded. 297By default, this variable is set to zero, indicating that 298no reseeding will occur. 299Reseeding should not be necessary, and will break 300.Dv TIME_WAIT 301recycling for a few minutes. 302.It tcp.inet.tcp.rexmit_{min,slop} 303Adjust the retransmit timer calculation for TCP. The slop is 304typically added to the raw calculation to take into account 305occasional variances that the SRTT (smoothed round trip time) 306is unable to accomodate, while the minimum specifies an 307absolute minimum. While a number of TCP RFCs suggest a 1 308second minimum these RFCs tend to focus on streaming behavior 309and fail to deal with the fact that a 1 second minimum has severe 310detrimental effects over lossy interactive connections, such 311as a 802.11b wireless link, and over very fast but lossy 312connections for those cases not covered by the fast retransmit 313code. For this reason we suggest changing the slop to 200ms and 314setting the minimum to something out of the way, like 20ms, 315which gives you an effective minimum of 200ms (similar to Linux). 316.It tcp.inflight_enable 317Enable 318.Tn TCP 319bandwidth delay product limiting. An attempt will be made to calculate 320the bandwidth delay product for each individual TCP connection and limit 321the amount of inflight data being transmitted to avoid building up 322unnecessary packets in the network. This option is recommended if you 323are serving a lot of data over connections with high bandwidth-delay 324products, such as modems, GigE links, and fast long-haul WANs, and/or 325you have configured your machine to accomodate large TCP windows. In such 326situations, without this option, you may experience high interactive 327latencies or packet loss due to the overloading of intermediate routers 328and switches. Note that bandwidth delay product limiting only effects 329the transmit side of a TCP connection. 330.It tcp.inflight_debug 331Enable debugging for the bandwidth delay product algorithm. This may 332default to on (1) so if you enable the algorithm you should probably also 333disable debugging by setting this variable to 0. 334.It tcp.inflight_min 335This puts an lower bound on the bandwidth delay product window, in bytes. 336A value of 1024 is typically used for debugging. 6000-16000 is more typical 337in a production installation. Setting this value too low may result in 338slow ramp-up times for bursty connections. Setting this value too high 339effectively disables the algorithm. 340.It tcp.inflight_max 341This puts an upper bound on the bandwidth delay product window, in bytes. 342This value should not generally be modified but may be used to set a 343global per-connection limit on queued data, potentially allowing you to 344intentionally set a less then optimum limit to smooth data flow over a 345network while still being able to specify huge internal TCP buffers. 346.It tcp.inflight_stab 347The bandwidth delay product algorithm requires a slightly larger window 348then it otherwise calculates for stability. This parameter determines the 349extra window in maximal packets / 10. The default value of 20 represents 3502 maximal packets. Reducing this value is not recommended but you may 351come across a situation with very slow links where the ping time 352reduction of the default inflight code is not sufficient. If this case 353occurs you should first try reducing tcp.inflight_min and, if that does not 354work, reduce both tcp.inflight_min and tcp.inflight_stab, trying values of 35515, 10, or 5 for the latter. Never use a value less then 5. Reducing 356tcp.inflight_stab can lead to upwards of a 20% underutilization of the link 357as well as reducing the algorithm's ability to adapt to changing 358situations and should only be done as a last resort. 359.El 360.Sh ERRORS 361A socket operation may fail with one of the following errors returned: 362.Bl -tag -width Er 363.It Bq Er EISCONN 364when trying to establish a connection on a socket which 365already has one; 366.It Bq Er ENOBUFS 367when the system runs out of memory for 368an internal data structure; 369.It Bq Er ETIMEDOUT 370when a connection was dropped 371due to excessive retransmissions; 372.It Bq Er ECONNRESET 373when the remote peer 374forces the connection to be closed; 375.It Bq Er ECONNREFUSED 376when the remote 377peer actively refuses connection establishment (usually because 378no process is listening to the port); 379.It Bq Er EADDRINUSE 380when an attempt 381is made to create a socket with a port which has already been 382allocated; 383.It Bq Er EADDRNOTAVAIL 384when an attempt is made to create a 385socket with a network address for which no network interface 386exists. 387.It Bq Er EAFNOSUPPORT 388when an attempt is made to bind or connect a socket to a multicast 389address. 390.El 391.Sh SEE ALSO 392.Xr getsockopt 2 , 393.Xr socket 2 , 394.Xr sysctl 3 , 395.Xr blackhole 4 , 396.Xr inet 4 , 397.Xr intro 4 , 398.Xr ip 4 , 399.Xr ttcp 4 400.Rs 401.%A V. Jacobson 402.%A R. Braden 403.%A D. Borman 404.%T "TCP Extensions for High Performance" 405.%O RFC 1323 406.Re 407.Rs 408.%A R. Braden 409.%T "T/TCP \- TCP Extensions for Transactions" 410.%O RFC 1644 411.Re 412.Sh HISTORY 413The 414.Nm 415protocol appeared in 416.Bx 4.2 . 417The RFC 1323 extensions for window scaling and timestamps were added 418in 419.Bx 4.4 . 420