tcp(7)



TCP(7)                     Linux Programmer's Manual                    TCP(7)

NAME
       tcp - TCP protocol

SYNOPSIS
       #include <sys/socket.h>
       #include <netinet/in.h>
       #include <netinet/tcp.h>

       tcp_socket = socket(AF_INET, SOCK_STREAM, 0);

DESCRIPTION
       This  is  an  implementation  of  the  TCP protocol defined in RFC 793,
       RFC 1122 and RFC 2001 with the NewReno and SACK  extensions.   It  pro-
       vides  a  reliable, stream-oriented, full-duplex connection between two
       sockets on top of ip(7), for both v4 and v6 versions.   TCP  guarantees
       that the data arrives in order and retransmits lost packets.  It gener-
       ates and checks a per-packet checksum  to  catch  transmission  errors.
       TCP does not preserve record boundaries.

       A  newly  created  TCP socket has no remote or local address and is not
       fully specified.  To create an outgoing TCP connection  use  connect(2)
       to establish a connection to another TCP socket.  To receive new incom-
       ing connections, first bind(2) the socket to a local address  and  port
       and  then  call  listen(2)  to put the socket into the listening state.
       After that a new socket for each incoming connection  can  be  accepted
       using  accept(2).   A socket which has had accept(2) or connect(2) suc-
       cessfully called on it is fully specified and may transmit data.   Data
       cannot be transmitted on listening or not yet connected sockets.

       Linux supports RFC 1323 TCP high performance extensions.  These include
       Protection Against Wrapped Sequence Numbers (PAWS), Window Scaling  and
       Timestamps.   Window scaling allows the use of large (> 64 kB) TCP win-
       dows in order to support links with high latency or bandwidth.  To make
       use of them, the send and receive buffer sizes must be increased.  They
       can  be  set  globally   with   the   /proc/sys/net/ipv4/tcp_wmem   and
       /proc/sys/net/ipv4/tcp_rmem  files,  or  on individual sockets by using
       the SO_SNDBUF and SO_RCVBUF socket options with the setsockopt(2) call.

       The maximum sizes for socket buffers declared  via  the  SO_SNDBUF  and
       SO_RCVBUF    mechanisms    are   limited   by   the   values   in   the
       /proc/sys/net/core/rmem_max  and   /proc/sys/net/core/wmem_max   files.
       Note that TCP actually allocates twice the size of the buffer requested
       in the setsockopt(2) call, and so a succeeding getsockopt(2) call  will
       not  return  the  same size of buffer as requested in the setsockopt(2)
       call.  TCP uses the extra space for administrative purposes and  inter-
       nal  kernel  structures,  and  the /proc file values reflect the larger
       sizes compared to the actual TCP windows.  On  individual  connections,
       the socket buffer size must be set prior to the listen(2) or connect(2)
       calls in order to have it take effect.  See socket(7) for more informa-
       tion.

       TCP  supports  urgent data.  Urgent data is used to signal the receiver
       that some important message is part of the  data  stream  and  that  it
       should  be  processed as soon as possible.  To send urgent data specify
       the MSG_OOB option to send(2).  When urgent data is received, the  ker-
       nel sends a SIGURG signal to the process or process group that has been
       set as the socket "owner" using the SIOCSPGRP or FIOSETOWN  ioctls  (or
       the POSIX.1-specified fcntl(2) F_SETOWN operation).  When the SO_OOBIN-
       LINE socket option is enabled, urgent data is put into the normal  data
       stream  (a program can test for its location using the SIOCATMARK ioctl
       described below), otherwise it can be received only  when  the  MSG_OOB
       flag is set for recv(2) or recvmsg(2).

       When out-of-band data is present, select(2) indicates the file descrip-
       tor as having an exceptional condition and poll (2) indicates a POLLPRI
       event.

       Linux  2.4  introduced  a number of changes for improved throughput and
       scaling, as well as enhanced functionality.  Some of these features in-
       clude  support for zero-copy sendfile(2), Explicit Congestion Notifica-
       tion, new management of TIME_WAIT sockets,  keep-alive  socket  options
       and support for Duplicate SACK extensions.

   Address formats
       TCP  is built on top of IP (see ip(7)).  The address formats defined by
       ip(7) apply to TCP.  TCP supports  point-to-point  communication  only;
       broadcasting and multicasting are not supported.

   /proc interfaces
       System-wide  TCP parameter settings can be accessed by files in the di-
       rectory /proc/sys/net/ipv4/.  In addition,  most  IP  /proc  interfaces
       also  apply  to TCP; see ip(7).  Variables described as Boolean take an
       integer value, with a nonzero value ("true") meaning  that  the  corre-
       sponding option is enabled, and a zero value ("false") meaning that the
       option is disabled.

       tcp_abc (Integer; default: 0; Linux 2.6.15 to Linux 3.8)
              Control the Appropriate Byte Count (ABC), defined in  RFC  3465.
              ABC  is  a  way  of increasing the congestion window (cwnd) more
              slowly in response to partial acknowledgments.  Possible  values
              are:

              0  increase cwnd once per acknowledgment (no ABC)

              1  increase cwnd once per acknowledgment of full sized segment

              2  allow  increase  cwnd by two if acknowledgment is of two seg-
                 ments to compensate for delayed acknowledgments.

       tcp_abort_on_overflow (Boolean; default: disabled; since Linux 2.4)
              Enable resetting connections if the  listening  service  is  too
              slow  and  unable  to keep up and accept them.  It means that if
              overflow occurred due to a burst, the connection  will  recover.
              Enable  this option only if you are really sure that the listen-
              ing daemon cannot be tuned to accept  connections  faster.   En-
              abling this option can harm the clients of your server.

       tcp_adv_win_scale (integer; default: 2; since Linux 2.4)
              Count   buffering   overhead  as  bytes/2^tcp_adv_win_scale,  if
              tcp_adv_win_scale    is    greater    than    0;    or    bytes-
              bytes/2^(-tcp_adv_win_scale),  if tcp_adv_win_scale is less than
              or equal to zero.

              The socket receive buffer space is shared between  the  applica-
              tion  and  kernel.   TCP maintains part of the buffer as the TCP
              window, this is the size of the receive window advertised to the
              other  end.   The rest of the space is used as the "application"
              buffer, used to isolate the network from scheduling and applica-
              tion  latencies.   The  tcp_adv_win_scale default value of 2 im-
              plies that the space used for  the  application  buffer  is  one
              fourth that of the total.

       tcp_allowed_congestion_control  (String; default: see text; since Linux
       2.4.20)
              Show/set the congestion control algorithm choices  available  to
              unprivileged  processes  (see the description of the TCP_CONGES-
              TION socket option).  The items in the  list  are  separated  by
              white  space and terminated by a newline character.  The list is
              a subset of those  listed  in  tcp_available_congestion_control.
              The  default value for this list is "reno" plus the default set-
              ting of tcp_congestion_control.

       tcp_autocorking (Boolean; default: enabled; since Linux 3.14)
              If this option is enabled, the kernel tries  to  coalesce  small
              writes  (from consecutive write(2) and sendmsg(2) calls) as much
              as possible, in order to decrease the total number of sent pack-
              ets.   Coalescing  is  done if at least one prior packet for the
              flow is waiting in Qdisc queues or device transmit  queue.   Ap-
              plications  can  still  use the TCP_CORK socket option to obtain
              optimal behavior when they know how/when to uncork  their  sock-
              ets.

       tcp_available_congestion_control   (String;   read-only;   since  Linux
       2.4.20)
              Show a list of the congestion-control algorithms that are regis-
              tered.   The  items in the list are separated by white space and
              terminated by a newline character.  This list is a limiting  set
              for  the  list  in tcp_allowed_congestion_control.  More conges-
              tion-control algorithms may be available  as  modules,  but  not
              loaded.

       tcp_app_win (integer; default: 31; since Linux 2.4)
              This  variable  defines how many bytes of the TCP window are re-
              served for buffering overhead.

              A maximum of (window/2^tcp_app_win, mss) bytes in the window are
              reserved  for the application buffer.  A value of 0 implies that
              no amount is reserved.

       tcp_base_mss (Integer; default: 512; since Linux 2.6.17)
              The initial value of search_low to be used by the  packetization
              layer  Path  MTU discovery (MTU probing).  If MTU probing is en-
              abled, this is the initial MSS used by the connection.

       tcp_bic (Boolean; default: disabled; Linux 2.4.27/2.6.6 to 2.6.13)
              Enable BIC TCP  congestion  control  algorithm.   BIC-TCP  is  a
              sender-side-only change that ensures a linear RTT fairness under
              large windows while offering both scalability and  bounded  TCP-
              friendliness.  The protocol combines two schemes called additive
              increase and binary search increase.  When the congestion window
              is  large, additive increase with a large increment ensures lin-
              ear RTT fairness as well as good scalability.  Under small  con-
              gestion  windows,  binary search increase provides TCP friendli-
              ness.

       tcp_bic_low_window (integer; default: 14; Linux 2.4.27/2.6.6 to 2.6.13)
              Set the threshold window (in packets) where BIC  TCP  starts  to
              adjust  the congestion window.  Below this threshold BIC TCP be-
              haves the same as the default TCP Reno.

       tcp_bic_fast_convergence (Boolean; default: enabled; Linux 2.4.27/2.6.6
       to 2.6.13)
              Force  BIC  TCP to more quickly respond to changes in congestion
              window.  Allows two flows sharing the same  connection  to  con-
              verge more rapidly.

       tcp_congestion_control (String; default: see text; since Linux 2.4.13)
              Set  the default congestion-control algorithm to be used for new
              connections.  The algorithm "reno" is always available, but  ad-
              ditional choices may be available depending on kernel configura-
              tion.  The default value for this file is set as part of  kernel
              configuration.

       tcp_dma_copybreak (integer; default: 4096; since Linux 2.6.24)
              Lower  limit, in bytes, of the size of socket reads that will be
              offloaded to a DMA copy engine, if one is present in the  system
              and the kernel was configured with the CONFIG_NET_DMA option.

       tcp_dsack (Boolean; default: enabled; since Linux 2.4)
              Enable RFC 2883 TCP Duplicate SACK support.

       tcp_ecn (Integer; default: see below; since Linux 2.4)
              Enable RFC 3168 Explicit Congestion Notification.

              This file can have one of the following values:

              0      Disable  ECN.  Neither initiate nor accept ECN.  This was
                     the default up to and including Linux 2.6.30.

              1      Enable ECN when requested  by  incoming  connections  and
                     also request ECN on outgoing connection attempts.

              2      Enable ECN when requested by incoming connections, but do
                     not request ECN on outgoing connections.  This  value  is
                     supported, and is the default, since Linux 2.6.31.

              When  enabled,  connectivity  to  some destinations could be af-
              fected due to older, misbehaving middle boxes  along  the  path,
              causing  connections  to be dropped.  However, to facilitate and
              encourage deployment with option 1,  and  to  work  around  such
              buggy  equipment,  the  tcp_ecn_fallback  option has been intro-
              duced.

       tcp_ecn_fallback (Boolean; default: enabled; since Linux 4.1)
              Enable RFC 3168, Section 6.1.1.1. fallback.  When enabled,  out-
              going  ECN-setup  SYNs  that  time out within the normal SYN re-
              transmission timeout will be resent with CWR and ECE cleared.

       tcp_fack (Boolean; default: enabled; since Linux 2.2)
              Enable TCP Forward Acknowledgement support.

       tcp_fin_timeout (integer; default: 60; since Linux 2.2)
              This specifies how many seconds to wait for a final  FIN  packet
              before the socket is forcibly closed.  This is strictly a viola-
              tion of the TCP specification, but required to  prevent  denial-
              of-service attacks.  In Linux 2.2, the default value was 180.

       tcp_frto (integer; default: see below; since Linux 2.4.21/2.6)
              Enable F-RTO, an enhanced recovery algorithm for TCP retransmis-
              sion timeouts (RTOs).  It is particularly beneficial in wireless
              environments  where packet loss is typically due to random radio
              interference rather than intermediate  router  congestion.   See
              RFC 4138 for more details.

              This file can have one of the following values:

              0  Disabled.   This  was  the  default up to and including Linux
                 2.6.23.

              1  The basic version F-RTO algorithm is enabled.

              2  Enable SACK-enhanced F-RTO if flow uses SACK.  The basic ver-
                 sion can be used also when SACK is in use though in that case
                 scenario(s) exists  where  F-RTO  interacts  badly  with  the
                 packet  counting of the SACK-enabled TCP flow.  This value is
                 the default since Linux 2.6.24.

              Before Linux 2.6.22, this parameter was a  Boolean  value,  sup-
              porting just values 0 and 1 above.

       tcp_frto_response (integer; default: 0; since Linux 2.6.22)
              When  F-RTO  has  detected that a TCP retransmission timeout was
              spurious (i.e., the timeout would have been avoided had TCP  set
              a  longer  retransmission timeout), TCP has several options con-
              cerning what to do next.  Possible values are:

              0  Rate halving based; a smooth and conservative  response,  re-
                 sults  in  halved  congestion  window  (cwnd)  and slow-start
                 threshold (ssthresh) after one RTT.

              1  Very conservative  response;  not  recommended  because  even
                 though  being  valid,  it  interacts  poorly with the rest of
                 Linux TCP; halves cwnd and ssthresh immediately.

              2  Aggressive response; undoes congestion-control measures  that
                 are  now known to be unnecessary (ignoring the possibility of
                 a lost retransmission that would require TCP to be more  cau-
                 tious); cwnd and ssthresh are restored to the values prior to
                 timeout.

       tcp_keepalive_intvl (integer; default: 75; since Linux 2.4)
              The number of seconds between TCP keep-alive probes.

       tcp_keepalive_probes (integer; default: 9; since Linux 2.2)
              The maximum number of TCP keep-alive probes to send before  giv-
              ing  up  and  killing  the connection if no response is obtained
              from the other end.

       tcp_keepalive_time (integer; default: 7200; since Linux 2.2)
              The number of seconds a connection needs to be idle  before  TCP
              begins sending out keep-alive probes.  Keep-alives are sent only
              when the SO_KEEPALIVE socket option  is  enabled.   The  default
              value  is  7200 seconds (2 hours).  An idle connection is termi-
              nated after approximately an additional 11 minutes (9 probes  an
              interval of 75 seconds apart) when keep-alive is enabled.

              Note that underlying connection tracking mechanisms and applica-
              tion timeouts may be much shorter.

       tcp_low_latency (Boolean; default: disabled;  since  Linux  2.4.21/2.6;
       obsolete since Linux 4.14)
              If  enabled, the TCP stack makes decisions that prefer lower la-
              tency as opposed to higher throughput.  It this option  is  dis-
              abled,  then  higher  throughput is preferred.  An example of an
              application where this default should be changed would be a  Be-
              owulf  compute  cluster.   Since Linux 4.14, this file still ex-
              ists, but its value is ignored.

       tcp_max_orphans (integer; default: see below; since Linux 2.4)
              The maximum number of orphaned (not attached to  any  user  file
              handle)  TCP sockets allowed in the system.  When this number is
              exceeded, the orphaned connection is  reset  and  a  warning  is
              printed.   This  limit  exists only to prevent simple denial-of-
              service attacks.  Lowering this limit is not recommended.   Net-
              work  conditions might require you to increase the number of or-
              phans allowed, but note that each orphan can eat up to ~64 kB of
              unswappable  memory.   The default initial value is set equal to
              the kernel parameter NR_FILE.  This initial default is  adjusted
              depending on the memory in the system.

       tcp_max_syn_backlog (integer; default: see below; since Linux 2.2)
              The  maximum  number  of  queued  connection requests which have
              still  not  received  an  acknowledgement  from  the  connecting
              client.  If this number is exceeded, the kernel will begin drop-
              ping requests.  The default value of 256 is  increased  to  1024
              when the memory present in the system is adequate or greater (>=
              128 MB), and reduced to 128 for those systems with very low mem-
              ory (<= 32 MB).

              Prior to Linux 2.6.20, it was recommended that if this needed to
              be increased above 1024, the  size  of  the  SYNACK  hash  table
              (TCP_SYNQ_HSIZE) in include/net/tcp.h should be modified to keep

                  TCP_SYNQ_HSIZE * 16 <= tcp_max_syn_backlog

              and the kernel should be recompiled.  In Linux 2.6.20, the fixed
              sized TCP_SYNQ_HSIZE was removed in favor of dynamic sizing.

       tcp_max_tw_buckets (integer; default: see below; since Linux 2.4)
              The maximum number of sockets in TIME_WAIT state allowed in  the
              system.  This limit exists only to prevent simple denial-of-ser-
              vice attacks.  The default value of NR_FILE*2  is  adjusted  de-
              pending  on  the  memory  in  the system.  If this number is ex-
              ceeded, the socket is closed and a warning is printed.

       tcp_moderate_rcvbuf   (Boolean;   default:   enabled;    since    Linux
       2.4.17/2.6.7)
              If  enabled, TCP performs receive buffer auto-tuning, attempting
              to automatically size the buffer (no greater  than  tcp_rmem[2])
              to match the size required by the path for full throughput.

       tcp_mem (since Linux 2.4)
              This  is  a  vector of 3 integers: [low, pressure, high].  These
              bounds, measured in units of the system page size, are  used  by
              TCP  to  track its memory usage.  The defaults are calculated at
              boot time from the amount of available memory.   (TCP  can  only
              use  low  memory  for  this,  which  is  limited  to  around 900
              megabytes on 32-bit systems.  64-bit systems do not suffer  this
              limitation.)

              low    TCP  doesn't regulate its memory allocation when the num-
                     ber of pages it has allocated globally is below this num-
                     ber.

              pressure
                     When  the  amount of memory allocated by TCP exceeds this
                     number of pages, TCP moderates  its  memory  consumption.
                     This  memory  pressure state is exited once the number of
                     pages allocated falls below the low mark.

              high   The maximum number of pages, globally, that TCP will  al-
                     locate.  This value overrides any other limits imposed by
                     the kernel.

       tcp_mtu_probing (integer; default: 0; since Linux 2.6.17)
              This parameter controls TCP Packetization-Layer Path MTU Discov-
              ery.  The following values may be assigned to the file:

              0  Disabled

              1  Disabled by default, enabled when an ICMP black hole detected

              2  Always enabled, use initial MSS of tcp_base_mss.

       tcp_no_metrics_save (Boolean; default: disabled; since Linux 2.6.6)
              By  default,  TCP  saves various connection metrics in the route
              cache when the connection closes,  so  that  connections  estab-
              lished  in  the  near future can use these to set initial condi-
              tions.  Usually, this increases overall performance, but it  may
              sometimes cause performance degradation.  If tcp_no_metrics_save
              is enabled, TCP will not cache metrics on closing connections.

       tcp_orphan_retries (integer; default: 8; since Linux 2.4)
              The maximum number of attempts made to probe the other end of  a
              connection which has been closed by our end.

       tcp_reordering (integer; default: 3; since Linux 2.4)
              The  maximum  a  packet  can be reordered in a TCP packet stream
              without TCP assuming packet loss and going into slow start.   It
              is  not  advisable  to change this number.  This is a packet re-
              ordering detection metric designed to minimize unnecessary  back
              off  and retransmits provoked by reordering of packets on a con-
              nection.

       tcp_retrans_collapse (Boolean; default: enabled; since Linux 2.2)
              Try to send full-sized packets during retransmit.

       tcp_retries1 (integer; default: 3; since Linux 2.2)
              The number of times TCP will attempt to retransmit a  packet  on
              an  established connection normally, without the extra effort of
              getting the network layers involved.  Once we exceed this number
              of retransmits, we first have the network layer update the route
              if possible before each new retransmit.  The default is the  RFC
              specified minimum of 3.

       tcp_retries2 (integer; default: 15; since Linux 2.2)
              The maximum number of times a TCP packet is retransmitted in es-
              tablished state before giving up.   The  default  value  is  15,
              which  corresponds  to a duration of approximately between 13 to
              30  minutes,  depending  on  the  retransmission  timeout.   The
              RFC 1122  specified  minimum  limit  of 100 seconds is typically
              deemed too short.

       tcp_rfc1337 (Boolean; default: disabled; since Linux 2.2)
              Enable TCP behavior conformant with RFC 1337.  When disabled, if
              a  RST is received in TIME_WAIT state, we close the socket imme-
              diately without waiting for the end of the TIME_WAIT period.

       tcp_rmem (since Linux 2.4)
              This is a vector of 3 integers: [min, default, max].  These  pa-
              rameters  are used by TCP to regulate receive buffer sizes.  TCP
              dynamically adjusts the size of the receive buffer from the  de-
              faults  listed below, in the range of these values, depending on
              memory available in the system.

              min    minimum size of the  receive  buffer  used  by  each  TCP
                     socket.   The default value is the system page size.  (On
                     Linux  2.4,  the  default  value  is  4 kB,  lowered   to
                     PAGE_SIZE  bytes  in  low-memory systems.)  This value is
                     used to ensure that in memory pressure mode,  allocations
                     below  this size will still succeed.  This is not used to
                     bound the size  of  the  receive  buffer  declared  using
                     SO_RCVBUF on a socket.

              default
                     the  default size of the receive buffer for a TCP socket.
                     This value overwrites the  initial  default  buffer  size
                     from the generic global net.core.rmem_default defined for
                     all protocols.  The default value is  87380  bytes.   (On
                     Linux  2.4,  this  will be lowered to 43689 in low-memory
                     systems.)  If larger receive buffer  sizes  are  desired,
                     this  value  should be increased (to affect all sockets).
                     To  employ  large  TCP  windows,  the   net.ipv4.tcp_win-
                     dow_scaling must be enabled (default).

              max    the  maximum  size of the receive buffer used by each TCP
                     socket.   This  value  does  not  override   the   global
                     net.core.rmem_max.  This is not used to limit the size of
                     the receive buffer declared using SO_RCVBUF on a  socket.
                     The default value is calculated using the formula

                         max(87380, min(4 MB, tcp_mem[1]*PAGE_SIZE/128))

                     (On  Linux  2.4, the default is 87380*2 bytes, lowered to
                     87380 in low-memory systems).

       tcp_sack (Boolean; default: enabled; since Linux 2.2)
              Enable RFC 2018 TCP Selective Acknowledgements.

       tcp_slow_start_after_idle  (Boolean;  default:  enabled;  since   Linux
       2.6.18)
              If  enabled,  provide RFC 2861 behavior and time out the conges-
              tion window after an idle period.  An idle period is defined  as
              the current RTO (retransmission timeout).  If disabled, the con-
              gestion window will not be timed out after an idle period.

       tcp_stdurg (Boolean; default: disabled; since Linux 2.2)
              If this option is enabled, then use the RFC 1122  interpretation
              of  the TCP urgent-pointer field.  According to this interpreta-
              tion, the urgent pointer points to the last byte of urgent data.
              If  this  option is disabled, then use the BSD-compatible inter-
              pretation of the urgent pointer: the urgent  pointer  points  to
              the  first byte after the urgent data.  Enabling this option may
              lead to interoperability problems.

       tcp_syn_retries (integer; default: 6; since Linux 2.2)
              The maximum number of times initial SYNs for an active TCP  con-
              nection attempt will be retransmitted.  This value should not be
              higher than 255.  The default value is 6, which  corresponds  to
              retrying for up to approximately 127 seconds.  Before Linux 3.7,
              the default value was 5, which (in conjunction with  calculation
              based  on other kernel parameters) corresponded to approximately
              180 seconds.

       tcp_synack_retries (integer; default: 5; since Linux 2.2)
              The maximum number of times a SYN/ACK segment for a passive  TCP
              connection  will  be  retransmitted.   This number should not be
              higher than 255.

       tcp_syncookies (Boolean; since Linux 2.2)
              Enable TCP syncookies.  The kernel must be  compiled  with  CON-
              FIG_SYN_COOKIES.  Send out syncookies when the syn backlog queue
              of a socket overflows.  The syncookies feature attempts to  pro-
              tect a socket from a SYN flood attack.  This should be used as a
              last resort, if at all.  This is a violation of the  TCP  proto-
              col,  and  conflicts  with other areas of TCP such as TCP exten-
              sions.  It can cause problems for clients and relays.  It is not
              recommended  as a tuning mechanism for heavily loaded servers to
              help with overloaded or misconfigured  conditions.   For  recom-
              mended alternatives see tcp_max_syn_backlog, tcp_synack_retries,
              and tcp_abort_on_overflow.

       tcp_timestamps (integer; default: 1; since Linux 2.2)
              Set to one of the following values to enable or disable RFC 1323
              TCP timestamps:

              0  Disable timestamps.

              1  Enable timestamps as defined in RFC1323 and use random offset
                 for each connection rather than only using the current time.

              2  As for the value 1,  but  without  random  offsets.   Setting
                 tcp_timestamps to this value is meaningful since Linux 4.10.

       tcp_tso_win_divisor (integer; default: 3; since Linux 2.6.9)
              This parameter controls what percentage of the congestion window
              can be consumed by  a  single  TCP  Segmentation  Offload  (TSO)
              frame.   The  setting  of  this  parameter is a tradeoff between
              burstiness and building larger TSO frames.

       tcp_tw_recycle (Boolean; default: disabled; Linux 2.4 to 4.11)
              Enable fast recycling of TIME_WAIT sockets.  Enabling  this  op-
              tion  is  not recommended as the remote IP may not use monotoni-
              cally increasing timestamps (devices behind  NAT,  devices  with
              per-connection  timestamp offsets).  See RFC 1323 (PAWS) and RFC
              6191.

       tcp_tw_reuse (Boolean; default: disabled; since Linux 2.4.19/2.6)
              Allow to reuse TIME_WAIT sockets for new connections when it  is
              safe  from protocol viewpoint.  It should not be changed without
              advice/request of technical experts.

       tcp_vegas_cong_avoid (Boolean; default: disabled; Linux 2.2 to 2.6.13)
              Enable TCP Vegas congestion avoidance algorithm.  TCP Vegas is a
              sender-side-only  change  to  TCP  that anticipates the onset of
              congestion by estimating the bandwidth.  TCP Vegas  adjusts  the
              sending  rate  by  modifying  the  congestion window.  TCP Vegas
              should provide less packet loss, but it is not as aggressive  as
              TCP Reno.

       tcp_westwood (Boolean; default: disabled; Linux 2.4.26/2.6.3 to 2.6.13)
              Enable  TCP  Westwood+  congestion control algorithm.  TCP West-
              wood+ is a sender-side-only modification of the TCP Reno  proto-
              col  stack that optimizes the performance of TCP congestion con-
              trol.  It is based on end-to-end  bandwidth  estimation  to  set
              congestion  window  and  slow start threshold after a congestion
              episode.  Using this estimation, TCP Westwood+ adaptively sets a
              slow  start  threshold  and a congestion window which takes into
              account the bandwidth used at the  time  congestion  is  experi-
              enced.   TCP Westwood+ significantly increases fairness with re-
              spect to TCP Reno in wired networks and throughput over wireless
              links.

       tcp_window_scaling (Boolean; default: enabled; since Linux 2.2)
              Enable RFC 1323 TCP window scaling.  This feature allows the use
              of a large window (> 64 kB) on  a  TCP  connection,  should  the
              other  end support it.  Normally, the 16 bit window length field
              in the TCP header limits the window size to less than 64 kB.  If
              larger  windows  are desired, applications can increase the size
              of their socket buffers and the window scaling  option  will  be
              employed.  If tcp_window_scaling is disabled, TCP will not nego-
              tiate the use of window scaling with the other end  during  con-
              nection setup.

       tcp_wmem (since Linux 2.4)
              This  is a vector of 3 integers: [min, default, max].  These pa-
              rameters are used by TCP to regulate send buffer sizes.  TCP dy-
              namically  adjusts  the size of the send buffer from the default
              values listed below, in the range of these values, depending  on
              memory available.

              min    Minimum  size of the send buffer used by each TCP socket.
                     The default value is the system  page  size.   (On  Linux
                     2.4,  the  default value is 4 kB.)  This value is used to
                     ensure that in memory pressure  mode,  allocations  below
                     this  size will still succeed.  This is not used to bound
                     the size of the send buffer declared using SO_SNDBUF on a
                     socket.

              default
                     The  default  size  of  the send buffer for a TCP socket.
                     This value overwrites the  initial  default  buffer  size
                     from  the  generic global /proc/sys/net/core/wmem_default
                     defined for all protocols.  The default value  is  16 kB.
                     If  larger  send  buffer  sizes  are  desired, this value
                     should be increased (to affect all sockets).   To  employ
                     large   TCP   windows,   the  /proc/sys/net/ipv4/tcp_win-
                     dow_scaling must be set to a nonzero value (default).

              max    The maximum size of the send  buffer  used  by  each  TCP
                     socket.   This  value  does  not  override  the  value in
                     /proc/sys/net/core/wmem_max.  This is not used  to  limit
                     the size of the send buffer declared using SO_SNDBUF on a
                     socket.  The default value is calculated using  the  for-
                     mula

                         max(65536, min(4 MB, tcp_mem[1]*PAGE_SIZE/128))

                     (On Linux 2.4, the default value is 128 kB, lowered 64 kB
                     depending on low-memory systems.)

       tcp_workaround_signed_windows (Boolean; default: disabled; since  Linux
       2.6.26)
              If  enabled,  assume  that no receipt of a window-scaling option
              means that the remote TCP is broken and treats the window  as  a
              signed quantity.  If disabled, assume that the remote TCP is not
              broken even if we do not receive a window  scaling  option  from
              it.

   Socket options
       To  set  or get a TCP socket option, call getsockopt(2) to read or set-
       sockopt(2) to write the option with the option level  argument  set  to
       IPPROTO_TCP.   Unless  otherwise  noted, optval is a pointer to an int.
       In addition, most IPPROTO_IP socket options are valid on  TCP  sockets.
       For more information see ip(7).

       Following  is  a  list  of TCP-specific socket options.  For details of
       some other socket options that are also applicable for TCP sockets, see
       socket(7).

       TCP_CONGESTION (since Linux 2.6.13)
              The  argument  for  this option is a string.  This option allows
              the caller to set the TCP congestion  control  algorithm  to  be
              used,  on  a  per-socket  basis.  Unprivileged processes are re-
              stricted to choosing one of the algorithms  in  tcp_allowed_con-
              gestion_control   (described   above).    Privileged   processes
              (CAP_NET_ADMIN) can choose from any of the available congestion-
              control algorithms (see the description of tcp_available_conges-
              tion_control above).

       TCP_CORK (since Linux 2.2)
              If set, don't send  out  partial  frames.   All  queued  partial
              frames  are sent when the option is cleared again.  This is use-
              ful for prepending headers before calling  sendfile(2),  or  for
              throughput  optimization.   As currently implemented, there is a
              200 millisecond ceiling on the time for which output  is  corked
              by  TCP_CORK.   If  this ceiling is reached, then queued data is
              automatically transmitted.  This option  can  be  combined  with
              TCP_NODELAY  only since Linux 2.5.71.  This option should not be
              used in code intended to be portable.

       TCP_DEFER_ACCEPT (since Linux 2.4)
              Allow a listener to be awakened only when data  arrives  on  the
              socket.   Takes  an  integer value (seconds), this can bound the
              maximum number of attempts TCP will make to complete the connec-
              tion.   This  option  should  not be used in code intended to be
              portable.

       TCP_INFO (since Linux 2.4)
              Used to collect information about this socket.  The  kernel  re-
              turns  a  struct  tcp_info  as  defined  in  the  file  /usr/in-
              clude/linux/tcp.h.  This option should not be used in  code  in-
              tended to be portable.

       TCP_KEEPCNT (since Linux 2.4)
              The  maximum  number  of keepalive probes TCP should send before
              dropping the connection.  This option should not be used in code
              intended to be portable.

       TCP_KEEPIDLE (since Linux 2.4)
              The time (in seconds) the connection needs to remain idle before
              TCP starts  sending  keepalive  probes,  if  the  socket  option
              SO_KEEPALIVE  has  been  set on this socket.  This option should
              not be used in code intended to be portable.

       TCP_KEEPINTVL (since Linux 2.4)
              The time (in seconds) between individual keepalive probes.  This
              option should not be used in code intended to be portable.

       TCP_LINGER2 (since Linux 2.4)
              The  lifetime  of orphaned FIN_WAIT2 state sockets.  This option
              can be used to override the  system-wide  setting  in  the  file
              /proc/sys/net/ipv4/tcp_fin_timeout for this socket.  This is not
              to be confused with the socket(7) level option SO_LINGER.   This
              option should not be used in code intended to be portable.

       TCP_MAXSEG
              The maximum segment size for outgoing TCP packets.  In Linux 2.2
              and earlier, and in Linux 2.6.28 and later, if  this  option  is
              set  before  connection  establishment,  it also changes the MSS
              value announced to the other end in the initial packet.   Values
              greater  than  the (eventual) interface MTU have no effect.  TCP
              will also impose its minimum and maximum bounds over  the  value
              provided.

       TCP_NODELAY
              If  set,  disable the Nagle algorithm.  This means that segments
              are always sent as soon as possible, even if  there  is  only  a
              small  amount  of  data.   When  not set, data is buffered until
              there is a sufficient amount to send out, thereby  avoiding  the
              frequent  sending  of  small packets, which results in poor uti-
              lization of the network.  This option is overridden by TCP_CORK;
              however, setting this option forces an explicit flush of pending
              output, even if TCP_CORK is currently set.

       TCP_QUICKACK (since Linux 2.4.4)
              Enable quickack mode if set or disable quickack mode if cleared.
              In quickack mode, acks are sent immediately, rather than delayed
              if needed in accordance to normal TCP operation.  This  flag  is
              not  permanent,  it  only  enables  a switch to or from quickack
              mode.  Subsequent operation of the TCP protocol will once  again
              enter/leave  quickack  mode  depending on internal protocol pro-
              cessing and factors such as delayed ack timeouts  occurring  and
              data  transfer.  This option should not be used in code intended
              to be portable.

       TCP_SYNCNT (since Linux 2.4)
              Set the number of SYN retransmits that TCP  should  send  before
              aborting  the  attempt  to connect.  It cannot exceed 255.  This
              option should not be used in code intended to be portable.

       TCP_USER_TIMEOUT (since Linux 2.6.37)
              This option takes an unsigned int  as  an  argument.   When  the
              value is greater than 0, it specifies the maximum amount of time
              in milliseconds that transmitted data may remain  unacknowledged
              before  TCP will forcibly close the corresponding connection and
              return ETIMEDOUT to the application.  If  the  option  value  is
              specified as 0, TCP will use the system default.

              Increasing  user timeouts allows a TCP connection to survive ex-
              tended periods without end-to-end connectivity.  Decreasing user
              timeouts  allows  applications  to  "fail  fast", if so desired.
              Otherwise, failure may take up to 20 minutes  with  the  current
              system defaults in a normal WAN environment.

              This option can be set during any state of a TCP connection, but
              is effective only during the synchronized states of a connection
              (ESTABLISHED,  FIN-WAIT-1,  FIN-WAIT-2, CLOSE-WAIT, CLOSING, and
              LAST-ACK).   Moreover,  when  used  with   the   TCP   keepalive
              (SO_KEEPALIVE)  option, TCP_USER_TIMEOUT will override keepalive
              to determine when to close a connection due to  keepalive  fail-
              ure.

              The  option  has no effect on when TCP retransmits a packet, nor
              when a keepalive probe is sent.

              This option, like many others, will be inherited by  the  socket
              returned by accept(2), if it was set on the listening socket.

              Further  details  on  the  user  timeout feature can be found in
              RFC 793 and RFC 5482 ("TCP User Timeout Option").

       TCP_WINDOW_CLAMP (since Linux 2.4)
              Bound the size of the advertised window to this value.  The ker-
              nel  imposes  a  minimum size of SOCK_MIN_RCVBUF/2.  This option
              should not be used in code intended to be portable.

   Sockets API
       TCP provides limited support for out-of-band data, in the  form  of  (a
       single  byte  of)  urgent  data.   In Linux this means if the other end
       sends newer out-of-band data the older urgent data is inserted as  nor-
       mal  data  into  the  stream (even when SO_OOBINLINE is not set).  This
       differs from BSD-based stacks.

       Linux uses the BSD compatible  interpretation  of  the  urgent  pointer
       field by default.  This violates RFC 1122, but is required for interop-
       erability   with   other   stacks.    It    can    be    changed    via
       /proc/sys/net/ipv4/tcp_stdurg.

       It  is  possible to peek at out-of-band data using the recv(2) MSG_PEEK
       flag.

       Since version 2.4, Linux supports the use of MSG_TRUNC in the flags ar-
       gument  of  recv(2)  (and  recvmsg(2)).   This flag causes the received
       bytes of data to be discarded, rather than passed back in a caller-sup-
       plied  buffer.   Since Linux 2.4.4, MSG_TRUNC also has this effect when
       used in conjunction with MSG_OOB to receive out-of-band data.

   Ioctls
       The following ioctl(2) calls return information in value.  The  correct
       syntax is:

              int value;
              error = ioctl(tcp_socket, ioctl_type, &value);

       ioctl_type is one of the following:

       SIOCINQ
              Returns  the amount of queued unread data in the receive buffer.
              The socket must not be in LISTEN state, otherwise an error (EIN-
              VAL) is returned.  SIOCINQ is defined in <linux/sockios.h>.  Al-
              ternatively, you can use the  synonymous  FIONREAD,  defined  in
              <sys/ioctl.h>.

       SIOCATMARK
              Returns true (i.e., value is nonzero) if the inbound data stream
              is at the urgent mark.

              If the SO_OOBINLINE socket option is set, and SIOCATMARK returns
              true,  then the next read from the socket will return the urgent
              data.  If the SO_OOBINLINE socket option is not set, and SIOCAT-
              MARK  returns  true, then the next read from the socket will re-
              turn the bytes following the urgent data (to actually  read  the
              urgent data requires the recv(MSG_OOB) flag).

              Note  that a read never reads across the urgent mark.  If an ap-
              plication is informed of the presence of  urgent  data  via  se-
              lect(2)  (using the exceptfds argument) or through delivery of a
              SIGURG signal, then it can advance up to the mark using  a  loop
              which  repeatedly tests SIOCATMARK and performs a read (request-
              ing any number of bytes) as long as SIOCATMARK returns false.

       SIOCOUTQ
              Returns the amount of unsent data in the socket send queue.  The
              socket  must not be in LISTEN state, otherwise an error (EINVAL)
              is returned.  SIOCOUTQ is defined in <linux/sockios.h>.   Alter-
              natively,  you  can  use  the  synonymous  TIOCOUTQ,  defined in
              <sys/ioctl.h>.

   Error handling
       When a network error occurs, TCP tries to resend  the  packet.   If  it
       doesn't  succeed after some time, either ETIMEDOUT or the last received
       error on this connection is reported.

       Some applications require a quicker error notification.   This  can  be
       enabled  with the IPPROTO_IP level IP_RECVERR socket option.  When this
       option is enabled, all incoming errors are immediately  passed  to  the
       user  program.  Use this option with care -- it makes TCP less tolerant
       to routing changes and other normal network conditions.

ERRORS
       EAFNOTSUPPORT
              Passed socket address type in sin_family was not AF_INET.

       EPIPE  The other end closed the socket unexpectedly or a read  is  exe-
              cuted on a shut down socket.

       ETIMEDOUT
              The  other  end didn't acknowledge retransmitted data after some
              time.

       Any errors defined for ip(7) or the generic socket layer  may  also  be
       returned for TCP.

VERSIONS
       Support  for  Explicit  Congestion Notification, zero-copy sendfile(2),
       reordering support and some SACK extensions (DSACK) were introduced  in
       2.4.   Support for forward acknowledgement (FACK), TIME_WAIT recycling,
       and per-connection keepalive socket options were introduced in 2.3.

BUGS
       Not all errors are documented.

       IPv6 is not described.

SEE ALSO
       accept(2), bind(2), connect(2), getsockopt(2),  listen(2),  recvmsg(2),
       sendfile(2), sendmsg(2), socket(2), ip(7), socket(7)

       The kernel source file Documentation/networking/ip-sysctl.txt.

       RFC 793 for the TCP specification.
       RFC 1122  for the TCP requirements and a description of the Nagle algo-
       rithm.
       RFC 1323 for TCP timestamp and window scaling options.
       RFC 1337 for a description of TIME_WAIT assassination hazards.
       RFC 3168 for a description of Explicit Congestion Notification.
       RFC 2581 for TCP congestion control algorithms.
       RFC 2018 and RFC 2883 for SACK and extensions to SACK.

COLOPHON
       This page is part of release 5.07 of the Linux  man-pages  project.   A
       description  of  the project, information about reporting bugs, and the
       latest    version    of    this    page,    can     be     found     at
       https://www.kernel.org/doc/man-pages/.

Linux                             2020-06-09                            TCP(7)

Man(1) output converted with man2html
list of all man pages