This is an implementation of the TCP protocol defined in RFC793, RFC1122
and RFC2001 with the NewReno extensions.
It implements a reliable stream oriented full duplex stream
between
two sockets. TCP ensures that packets are not reordered and retransmits them
when they are dropped. It generates and checks a per packet checksum to catch
transmission errors.
A fresh TCP socket has no remote or local address and is not fully specified.
To create an outgoing TCP connection the
connect(2)
function is called on the socket. To accept incoming connections
bind(2)
the socket first to a local address and port and then call
listen(2)
to allow the accepting of incoming connections. Then use
accept(2)
to get a new socket with the incoming connection. The listening socket
stays. After
accept(2)
or
connect(2)
a socket is fully specified.
Data may be only transferred on fully specified sockets.
When the initial connection request packet carries IP options and the
accept_source_routes sysctl is enabled all outgoing datagrams on this
connection will carry the reversed source route.
Linux 2.2 supports the RFC1323 TCP high performance extensions.
They include window scaling to support large windows and the timestamp option with
protection against wrapped sequence numbers (
PAWS
).
Large windows are needed for good performance over links with long latencies
or very high bandwidth. To use them the send and receive buffers have to be
increased from the default values. This can be either done globally using the
net.core.wmen_default
and
net.core.rmem_default
sysctls, or on a per socket basis using the
SO_SNDBUF
and
SO_RCVBUF
socket options. The maximum receive buffer size settable on a socket
is limited by the global
net.core.rmem_max
and
net.core.wmem_max
sysctls. See
socket(4)
for more information.
TCP supports urgent data. Urgent data is used to signal the receiver
that some important message is part of the data stream and that is should
be processed as soon as possible.
To send urgent data specify the
MSG_OOB
option to
sendfile(2).
When urgent data is received the kernel sends an SIGURG signal to the
reading process or the process or process group that has been set for the socket
using the
FIOCSPGRP
or
FIOCSETOWN
ioctls. When the
SO_OOBINLINE
socket option is enabled urgent data is put into the normal data stream
(and can be tested for by the
SIOCATMARK
ioctl),
otherwise it can be only received when the
MSG_OOB
flag is set for
sendmsg(2).NotethatLinuxperdefaultusestheBSDcompatible
interpretation of the urgent pointer field, see the
tcp_stdurg
sysctl below.
ADDRESS FORMATS
TCP is built on top of IP (see
ip(4)).
The address formats defined by
ip(4)
apply to TCP. TCP only supports
point-to-point communication; broadcasting and multicasting are not supported.
SYSCTLS
These sysctls can be accessed by the
/proc/sys/net/ipv4/*
files or with the
sysctl(2)
interface. In addition, most IP sysctls also apply to TCP; see
ip(4).
tcp_window_scaling
Enable RFC1323 TCP window scaling.
tcp_sack
Enable RFC2018 TCP Selective Acknowledgements.
tcp_timestamps
Enable RFC1323 TCP timestamps.
tcp_fin_timeout
How many seconds to wait for a final FIN packet before the socket is forcibly closed.
This is strictly a violation of the TCP specification, but required to prevent
denial-of-service attacks.
tcp_keepalive_probes
Maximum TCP keep-alive probes to send before giving up. Keep-alives are only
send when the
SO_KEEPALIVE
socket option is enabled.
tcp_keepalive_time
How often keep-alives are sent on a connection. Defined in seconds. Default
is 2 hours.
tcp_max_ka_probes
How many keep-alive probes are sent per slow timer run. To prevent
bursts, this value should not be set too high.
tcp_stdurg
Enable the strict RFC793 interpretation of the TCP urgent-pointer field.
The default is to use the BSD-compatible interpretation of the urgent-pointer,
pointing to the first byte after the urgent data. The RFC793 interpretation
is to have it point to the last byte of urgent data. Enabling this option
may lead to interoperatibility problems.
tcp_syncookies
Enable TCP syncookies. The kernel must be compiled with
CONFIG_SYN_COOKIES.
They defend against a particular TCP denial-of-service attack. Note that the
concept of a socket backlog is abandoned; this means the peer may not receive
reliable error messages from an overloaded server with syncookies enabled.
tcp_max_syn_backlog
Length of the per-socket backlog queue. As of Linux 2.2, the backlog specified
in
listen(2)
only specifies the length of the backlog queue of already established sockets.
The maximum queue of sockets not yet established (in
SYN_RECV
state)
per listen socket is set by this sysctl. When more connection requests arrive,
Linux starts to drop packets. When syncookies, are enabled the packets are still
answered and the maximum queue is effectively ignored.
tcp_retries1
Defines how many times an answer to a TCP connection request is retransmited before
giving up.
tcp_retries2
Defines how many times a TCP packet is retransmitted in established state
before giving up.
tcp_syn_retries
Defines how many times to try to send an initial SYN packet to a remote
host before giving up and returns an error. Must be below 255.
This is only the timeout for outgoing connections; for incoming
connections the number of retransmits is defined by
tcp_retries1.tcp_retries1.
tcp_retrans_collapse
Try to send full-sized packets during retransmit. This is used to work around
TCP bugs in some stacks.
SOCKET OPTIONS
To set or get a TCP socket option, call
getsockopt(2)
to read or
setsockopt(2)
to write the option with the socket family argument set to
SOL_TCP.
In addition,
most
SOL_IP
socket options are valid on TCP sockets. For more information see
ip(4).
TCP_NODELAY
Turn the Nagle algorithm off. This means that packets are always sent as soon
as possible and no unnecessary delays are introduced, at the cost of more
packets in the network. Expects an integer boolean flag.
TCP_MAXSEG
Set or receive the maximum segment size for outgoing TCP packets. 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 interface MTU are ignored and have no effect.
TCP_CORK
If enabled don't send out partial frames.
All queued partial frames are sent when the option is cleared again.
This is useful for prepending headers
before calling
sendfile(2),
or for throughput optimization. This option cannot be combined with
TCP_NODELAY.
IOCTLS
These ioctls can be accessed using
ioctl(2).
The correct syntax is:
int value;error = ioctl(tcp_socket, ioctl_type, &value);
FIONREAD
Returns the amount of queued unread data in the receive buffer. Argument
is a pointer to an integer.
SIOCATMARK
Returns true when the all urgent data has been already received by the user
program.
This is used together with
SO_OOBINLINE.
Argument is an pointer to an integer for the test result.
TIOCOUTQ
Returns the amount of unsent data in the socket send queue. Argument is an integer.
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
SOL_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.
When the other end closes the socket without doing a proper closing handshake, a
SIGPIPE
signal is raised and EPIPE is returned. This can be prevented by the
MSG_NOSIGNAL
flag.
ERRORS
EPIPE
The other end closed the socket unexpectedly.
ETIMEDOUT
The other end didn't acknowledge retransmitted data after some time.
EAFNOTSUPPORT
Passed socket address type in
sin_family
was not
AF_INET.
Any errors defined for
ip(4)
or the generic socket layer may also be returned for TCP.
BUGS
Not all errors are documented.
IPv6 is not described.
Transparent proxy options are not described.
VERSIONS
The sysctls are new in Linux 2.2.
IP_RECVERR
and
MSG_NOSIGNAL
are a new
feature in Linux 2.2.
TCP_CORK
is new in 2.2.
