TCP - Internet Transmission Control Protocol
s = socket(AF_INET, SOCK_STREAM, 0);
The TCP protocol provides reliable, flow-controlled, two-way
transmission of data. It is a byte-stream protocol used to support
the SOCK_STREAM socket type. TCP constructs virtual circuits between
peer entities. A virtual circuit consists of remote Internet
addresses, remote ports, local Internet addresses and local ports. IP
uses the Internet addresses to direct messages between hosts, and the
port numbers to identify a TCP entity at a particular host.
Sockets using TCP are either active or passive. connect() creates
active sockets, which initiate connections to passive sockets (see
connect(2)). To create a passive socket, use the listen() system call
after binding the socket with the bind() system call (see listen(2)
and bind(2)). Only passive sockets can use the accept() call to
accept incoming connections (see accept(2)).
Passive sockets can underspecify their location to match incoming
connection requests from multiple networks. This technique, called
wildcard addressing, allows a single server to provide service to
clients on multiple networks. To create a socket that listens on all
networks, the Internet address INADDR_ANY must be bound. The TCP port
can still be specified even if wildcard addressing is being used. If
the port is specified as zero, the system assigns a port.
Once accept() has a rendezvous with a connect request, a virtual
circuit is established between peer entities. bind() supplies the
local port and local Internet address and accept() gathers the remote
port and remote Internet address from the peer requesting the
The system supports four socket options: TCP_MAXSEG, TCP_NODELAY,
TCP_ABORT_THRESHOLD, and TCP_CONN_ABORT_THRESHOLD (defined in the
include file <<<<netinet/tcp.h>>>>). TCP_MAXSEG option can only be used
with getsockopt(), while TCP_NODELAY, TCP_ABORT_THRESHOLD, and
TCP_CONN_ABORT_THRESHOLD can be set with setsockopt() and tested with
getsockopt() (see getsockopt(2)). These four options require level to
be set to IPPROTO_TCP in the getsockopt/setsockopt call.
TCP_MAXSEG (non-boolean option) lets an application
to receive the current segment size of
the TCP SOCK_STREAM socket. The current
Hewlett-Packard Company - 1 - HP-UX Release 11i: November 2000
segment size will be returned in optval.
TCP_NODELAY (boolean option) causes small amounts of
output to be sent immediately.
TCP_ABORT_THRESHOLD (non-boolean option) sets the second
threshold timer for the connections that
are in ESTABLISHED state. The option
value is the threshold time in
When it must retransmit packets because
a timer has expired, TCP first compares
the total time it has waited against the
two thresholds, as described in RFC
1122, 188.8.131.52. If it has waited longer
than the second threshold (R2), TCP
terminates the connection. The default
value for this option is the current
value of the ndd tunable parameter
tcp_ip_abort_interval. Refer to ndd(1M)
online help for details on the
tcp_ip_abort_interval default value.
TCP_CONN_ABORT_THRESHOLD (non-boolean option) sets the second
threshold timer during connection
establishment. The option value is the
threshold time in milliseconds.
This option is the same as
TCP_ABORT_THRESHOLD, except that this
value is used during connection
establishment. When it must retransmit
the SYN packet because a timer has
expired, TCP first compares the total
time it has waited against the two
thresholds. If it has waited longer than
the second threshold, TCP terminates the
connection. The default value for this
option is the current value of the ndd
tunable tcp_ip_abort_cinterval. See
ndd(1M) online help for details on the
tcp_ip_abort_cinterval default value.
If TCP_NODELAY is set, the system sends small amounts of output
immediately rather than gathering them into a single packet after an
acknowledgement is received. If TCP_NODELAY is not set, the system
sends data when it is presented, if there is no outstanding
unacknowledged data. If there is outstanding unacknowledged data, the
system gathers small amounts of data to be sent in a single packet
once an acknowledgement is received. For clients such as window
Hewlett-Packard Company - 2 - HP-UX Release 11i: November 2000
managers that send a stream of mouse events which receive no replies,
this packetization may cause significant delays. The TCP_NODELAY
option can be used to avoid this situation. Note, however, that
setting the TCP_NODELAY option may result in a large number of small
packets being sent over the network.
By default, TCP_NODELAY is not set when a socket is created.
The option level to use for accessing the TCP option with the
setsockopt() or getsockopt() calls is the protocol number for TCP
which is available from getprotobyname() (see getprotoent(3N)).
If the SO_KEEPALIVE socket option is enabled on an established TCP
connection and the connection has been idle for two hours, TCP sends a
packet to the remote socket, expecting the remote TCP to acknowledge
that the connection is still active. If the remote TCP does not
respond in a timely manner, TCP continues to send keepalive packets
according to its normal retransmission algorithm. If the remote TCP
does not respond within a particular time limit, TCP drops the
connection. The next socket system call (for example, recv()) returns
an error, and errno is set to ETIMEDOUT. See getsockopt(2) for
details on enabling SO_KEEPALIVE.
The default send and receives buffer size is 32768 bytes (see WARNINGS
below). The send and receive buffer sizes for TCP stream sockets can
be altered by using the SO_SNDBUF and SO_RCVBUF options of the
setsockopt() system call or the XTI_SNDBUF and XTI_RCVBUF options of
the t_optmgmt() system call. Refer to getsockopt(2) or t_optmgmt(3)
The maximum transmit buffer size for a TCP stream socket is 2147483647
bytes. The maximum receive buffer size for a TCP stream socket is
1073725440 bytes. These maximum values can be lowered using the ndd
variables tcp_xmit_hiwater_max and tcp_recv_hiwater_max.
One of the following errors may be returned in errno if a socket
operation fails. For a more detailed list of errors, see the man
pages for specific system calls.
[EISCONN] The socket is already connected.
[ENOBUFS] No buffer space is available for an
internal data structure.
[ETIMEDOUT] Connection dropped due to excessive
[ECONNRESET] The connection was forcibly closed by
the peer socket.
Hewlett-Packard Company - 3 - HP-UX Release 11i: November 2000
[ECONNREFUSED] Remote peer actively refuses connection
establishment (usually because no
process is listening to the port).
[EADDRINUSE] The specified address is already in use.
[EADDRNOTAVAIL] The specified address is not available
on this machine.
The default socket buffer size might increase without notice in a
future release or patch. Therefore, if an application calls
setsockopt() with SO_RCVBUF, it should do so before calling listen(),
or it should first call getsockopt() with SO_RCVBUF and ensure that
the intended new receive buffer size is not less than the current
buffer size. These programming conventions are consistent with TCP
protocol restrictions against reducing the TCP receive window after a
connection has been established.
The socket interfaces to TCP were developed by the University of
getsockopt(2), socket(2), send(2), recv(2), t_open(3), t_optmgmt(3),
socket(7), inet(7F), ndd(1M).
RFC 793 Transmission Control Protocol
RFC 1122 Requirements for Internet hosts
RFC 1323 TCP Extensions for High Performance
RFC 1878 Variable Length Subnet Table for IPv4
RFC 2018 TCP Selective Acknowledgement Options
RFC 2414 Increasing TCP's Initial Window
RFC 2582 NewReno Modifications to TCP's Fast Recovery Algorithm
Hewlett-Packard Company - 4 - HP-UX Release 11i: November 2000