Net

Source Code: lib/net.js

The node:net module provides an asynchronous network API for creating stream-based TCP or IPC servers (net.createServer()) and clients (net.createConnection()).

It can be accessed using:

The node:net module supports IPC with named pipes on Windows, and Unix domain sockets on other operating systems.

net.connect(), net.createConnection(), server.listen(), and socket.connect() take a path parameter to identify IPC endpoints.

On Unix, the local domain is also known as the Unix domain. The path is a file system pathname. It gets truncated to an OS-dependent length of sizeof(sockaddr_un.sun_path) - 1. Typical values are 107 bytes on Linux and 103 bytes on macOS. If a Node.js API abstraction creates the Unix domain socket, it will unlink the Unix domain socket as well. For example, net.createServer() may create a Unix domain socket and server.close() will unlink it. But if a user creates the Unix domain socket outside of these abstractions, the user will need to remove it. The same applies when a Node.js API creates a Unix domain socket but the program then crashes. In short, a Unix domain socket will be visible in the file system and will persist until unlinked.

On Windows, the local domain is implemented using a named pipe. The path must refer to an entry in \\?\pipe\ or \\.\pipe\. Any characters are permitted, but the latter may do some processing of pipe names, such as resolving .. sequences. Despite how it might look, the pipe namespace is flat. Pipes will not persist. They are removed when the last reference to them is closed. Unlike Unix domain sockets, Windows will close and remove the pipe when the owning process exits.

JavaScript string escaping requires paths to be specified with extra backslash escaping such as:

The BlockList object can be used with some network APIs to specify rules for disabling inbound or outbound access to specific IP addresses, IP ranges, or IP subnets.

Adds a rule to block the given IP address.

Adds a rule to block a range of IP addresses from start (inclusive) to end (inclusive).

Adds a rule to block a range of IP addresses specified as a subnet mask.

Returns true if the given IP address matches any of the rules added to the BlockList.

The list of rules added to the blocklist.

This class is used to create a TCP or IPC server.

net.Server is an EventEmitter with the following events:

Emitted when the server closes. If connections exist, this event is not emitted until all connections are ended.

Emitted when a new connection is made. socket is an instance of net.Socket.

Emitted when an error occurs. Unlike net.Socket, the 'close' event will not be emitted directly following this event unless server.close() is manually called. See the example in discussion of server.listen().

Emitted when the server has been bound after calling server.listen().

When the number of connections reaches the threshold of server.maxConnections, the server will drop new connections and emit 'drop' event instead. If it is a TCP server, the argument is as follows, otherwise the argument is undefined.

The family property now returns a string instead of a number.

The family property now returns a number instead of a string.

Added in: v0.1.90

Returns the bound address, the address family name, and port of the server as reported by the operating system if listening on an IP socket (useful to find which port was assigned when getting an OS-assigned address): { port: 12346, family: 'IPv4', address: '127.0.0.1' }.

For a server listening on a pipe or Unix domain socket, the name is returned as a string.

server.address() returns null before the 'listening' event has been emitted or after calling server.close().

Stops the server from accepting new connections and keeps existing connections. This function is asynchronous, the server is finally closed when all connections are ended and the server emits a 'close' event. The optional callback will be called once the 'close' event occurs. Unlike that event, it will be called with an Error as its only argument if the server was not open when it was closed.

Asynchronously get the number of concurrent connections on the server. Works when sockets were sent to forks.

Callback should take two arguments err and count.

Start a server listening for connections. A net.Server can be a TCP or an IPC server depending on what it listens to.

Possible signatures:

This function is asynchronous. When the server starts listening, the 'listening' event will be emitted. The last parameter callback will be added as a listener for the 'listening' event.

All listen() methods can take a backlog parameter to specify the maximum length of the queue of pending connections. The actual length will be determined by the OS through sysctl settings such as tcp_max_syn_backlog and somaxconn on Linux. The default value of this parameter is 511 (not 512).

All net.Socket are set to SO_REUSEADDR (see socket(7) for details).

The server.listen() method can be called again if and only if there was an error during the first server.listen() call or server.close() has been called. Otherwise, an ERR_SERVER_ALREADY_LISTEN error will be thrown.

One of the most common errors raised when listening is EADDRINUSE. This happens when another server is already listening on the requested port/path/handle. One way to handle this would be to retry after a certain amount of time:

Start a server listening for connections on a given handle that has already been bound to a port, a Unix domain socket, or a Windows named pipe.

The handle object can be either a server, a socket (anything with an underlying _handle member), or an object with an fd member that is a valid file descriptor.

Listening on a file descriptor is not supported on Windows.

AbortSignal support was added.

The ipv6Only option is supported.

Added in: v0.11.14

If port is specified, it behaves the same as server.listen([port[, host[, backlog]]][, callback]). Otherwise, if path is specified, it behaves the same as server.listen(path[, backlog][, callback]). If none of them is specified, an error will be thrown.

If exclusive is false (default), then cluster workers will use the same underlying handle, allowing connection handling duties to be shared. When exclusive is true, the handle is not shared, and attempted port sharing results in an error. An example which listens on an exclusive port is shown below.

When exclusive is true and the underlying handle is shared, it is possible that several workers query a handle with different backlogs. In this case, the first backlog passed to the master process will be used.

Starting an IPC server as root may cause the server path to be inaccessible for unprivileged users. Using readableAll and writableAll will make the server accessible for all users.

If the signal option is enabled, calling .abort() on the corresponding AbortController is similar to calling .close() on the server:

Start an IPC server listening for connections on the given path.

Start a TCP server listening for connections on the given port and host.

If port is omitted or is 0, the operating system will assign an arbitrary unused port, which can be retrieved by using server.address().port after the 'listening' event has been emitted.

If host is omitted, the server will accept connections on the unspecified IPv6 address (::) when IPv6 is available, or the unspecified IPv4 address (0.0.0.0) otherwise.

In most operating systems, listening to the unspecified IPv6 address (::) may cause the net.Server to also listen on the unspecified IPv4 address (0.0.0.0).

Set this property to reject connections when the server's connection count gets high.

It is not recommended to use this option once a socket has been sent to a child with child_process.fork().

Opposite of unref(), calling ref() on a previously unrefed server will not let the program exit if it's the only server left (the default behavior). If the server is refed calling ref() again will have no effect.

Calling unref() on a server will allow the program to exit if this is the only active server in the event system. If the server is already unrefed calling unref() again will have no effect.

This class is an abstraction of a TCP socket or a streaming IPC endpoint (uses named pipes on Windows, and Unix domain sockets otherwise). It is also an EventEmitter.

A net.Socket can be created by the user and used directly to interact with a server. For example, it is returned by net.createConnection(), so the user can use it to talk to the server.

It can also be created by Node.js and passed to the user when a connection is received. For example, it is passed to the listeners of a 'connection' event emitted on a net.Server, so the user can use it to interact with the client.

AbortSignal support was added.

Added in: v0.3.4

Creates a new socket object.

The newly created socket can be either a TCP socket or a streaming IPC endpoint, depending on what it connect() to.

Emitted once the socket is fully closed. The argument hadError is a boolean which says if the socket was closed due to a transmission error.

Emitted when a socket connection is successfully established. See net.createConnection().

Emitted when data is received. The argument data will be a Buffer or String. Encoding of data is set by socket.setEncoding().

The data will be lost if there is no listener when a Socket emits a 'data' event.

Emitted when the write buffer becomes empty. Can be used to throttle uploads.

See also: the return values of socket.write().

Emitted when the other end of the socket signals the end of transmission, thus ending the readable side of the socket.

By default (allowHalfOpen is false) the socket will send an end of transmission packet back and destroy its file descriptor once it has written out its pending write queue. However, if allowHalfOpen is set to true, the socket will not automatically end() its writable side, allowing the user to write arbitrary amounts of data. The user must call end() explicitly to close the connection (i.e. sending a FIN packet back).

Emitted when an error occurs. The 'close' event will be called directly following this event.

The host parameter is supported now.

Added in: v0.11.3

Emitted after resolving the host name but before connecting. Not applicable to Unix sockets.

Emitted when a socket is ready to be used.

Triggered immediately after 'connect'.

Emitted if the socket times out from inactivity. This is only to notify that the socket has been idle. The user must manually close the connection.

See also: socket.setTimeout().

The family property now returns a string instead of a number.

The family property now returns a number instead of a string.

Added in: v0.1.90

Returns the bound address, the address family name and port of the socket as reported by the operating system: { port: 12346, family: 'IPv4', address: '127.0.0.1' }

This property is only present if the family autoselection algorithm is enabled in socket.connect(options) and it is an array of the addresses that have been attempted.

Each address is a string in the form of $IP:$PORT. If the connection was successful, then the last address is the one that the socket is currently connected to.

This property shows the number of characters buffered for writing. The buffer may contain strings whose length after encoding is not yet known. So this number is only an approximation of the number of bytes in the buffer.

net.Socket has the property that socket.write() always works. This is to help users get up and running quickly. The computer cannot always keep up with the amount of data that is written to a socket. The network connection simply might be too slow. Node.js will internally queue up the data written to a socket and send it out over the wire when it is possible.

The consequence of this internal buffering is that memory may grow. Users who experience large or growing bufferSize should attempt to "throttle" the data flows in their program with socket.pause() and socket.resume().

The amount of received bytes.

The amount of bytes sent.

Initiate a connection on a given socket.

Possible signatures:

This function is asynchronous. When the connection is established, the 'connect' event will be emitted. If there is a problem connecting, instead of a 'connect' event, an 'error' event will be emitted with the error passed to the 'error' listener. The last parameter connectListener, if supplied, will be added as a listener for the 'connect' event once.

This function should only be used for reconnecting a socket after 'close' has been emitted or otherwise it may lead to undefined behavior.

The default value for autoSelectFamily option can be changed at runtime using setDefaultAutoSelectFamily or via the command line option --enable-network-family-autoselection.

Added the autoSelectFamily option.

The noDelay, keepAlive, and keepAliveInitialDelay options are supported now.

Added onread option.

The hints option defaults to 0 in all cases now. Previously, in the absence of the family option it would default to dns.ADDRCONFIG | dns.V4MAPPED.

The hints option is supported now.

Added in: v0.1.90

Initiate a connection on a given socket. Normally this method is not needed, the socket should be created and opened with net.createConnection(). Use this only when implementing a custom Socket.

For TCP connections, available options are:

For IPC connections, available options are:

For both types, available options include:

Following is an example of a client using the onread option:

Initiate an IPC connection on the given socket.

Alias to socket.connect(options[, connectListener]) called with { path: path } as options.

Initiate a TCP connection on the given socket.

Alias to socket.connect(options[, connectListener]) called with {port: port, host: host} as options.

If true, socket.connect(options[, connectListener]) was called and has not yet finished. It will stay true until the socket becomes connected, then it is set to false and the 'connect' event is emitted. Note that the socket.connect(options[, connectListener]) callback is a listener for the 'connect' event.

Ensures that no more I/O activity happens on this socket. Destroys the stream and closes the connection.

See writable.destroy() for further details.

See writable.destroyed for further details.

Destroys the socket after all data is written. If the 'finish' event was already emitted the socket is destroyed immediately. If the socket is still writable it implicitly calls socket.end().

Half-closes the socket. i.e., it sends a FIN packet. It is possible the server will still send some data.

See writable.end() for further details.

The string representation of the local IP address the remote client is connecting on. For example, in a server listening on '0.0.0.0', if a client connects on '192.168.1.1', the value of socket.localAddress would be '192.168.1.1'.

The numeric representation of the local port. For example, 80 or 21.

The string representation of the local IP family. 'IPv4' or 'IPv6'.

Pauses the reading of data. That is, 'data' events will not be emitted. Useful to throttle back an upload.

This is true if the socket is not connected yet, either because .connect() has not yet been called or because it is still in the process of connecting (see socket.connecting).

Opposite of unref(), calling ref() on a previously unrefed socket will not let the program exit if it's the only socket left (the default behavior). If the socket is refed calling ref again will have no effect.

The string representation of the remote IP address. For example, '74.125.127.100' or '2001:4860:a005::68'. Value may be undefined if the socket is destroyed (for example, if the client disconnected).

The string representation of the remote IP family. 'IPv4' or 'IPv6'.

The numeric representation of the remote port. For example, 80 or 21.

Close the TCP connection by sending an RST packet and destroy the stream. If this TCP socket is in connecting status, it will send an RST packet and destroy this TCP socket once it is connected. Otherwise, it will call socket.destroy with an ERR_SOCKET_CLOSED Error. If this is not a TCP socket (for example, a pipe), calling this method will immediately throw an ERR_INVALID_HANDLE_TYPE Error.

Resumes reading after a call to socket.pause().

Set the encoding for the socket as a Readable Stream. See readable.setEncoding() for more information.

New defaults for TCP_KEEPCNT and TCP_KEEPINTVL socket options were added.

Added in: v0.1.92

Enable/disable keep-alive functionality, and optionally set the initial delay before the first keepalive probe is sent on an idle socket.

Set initialDelay (in milliseconds) to set the delay between the last data packet received and the first keepalive probe. Setting 0 for initialDelay will leave the value unchanged from the default (or previous) setting.

Enabling the keep-alive functionality will set the following socket options:

Enable/disable the use of Nagle's algorithm.

When a TCP connection is created, it will have Nagle's algorithm enabled.

Nagle's algorithm delays data before it is sent via the network. It attempts to optimize throughput at the expense of latency.

Passing true for noDelay or not passing an argument will disable Nagle's algorithm for the socket. Passing false for noDelay will enable Nagle's algorithm.

Passing an invalid callback to the callback argument now throws ERR_INVALID_ARG_TYPE instead of ERR_INVALID_CALLBACK.

Added in: v0.1.90

Sets the socket to timeout after timeout milliseconds of inactivity on the socket. By default net.Socket do not have a timeout.

When an idle timeout is triggered the socket will receive a 'timeout' event but the connection will not be severed. The user must manually call socket.end() or socket.destroy() to end the connection.

If timeout is 0, then the existing idle timeout is disabled.

The optional callback parameter will be added as a one-time listener for the 'timeout' event.

The socket timeout in milliseconds as set by socket.setTimeout(). It is undefined if a timeout has not been set.

Calling unref() on a socket will allow the program to exit if this is the only active socket in the event system. If the socket is already unrefed calling unref() again will have no effect.

Sends data on the socket. The second parameter specifies the encoding in the case of a string. It defaults to UTF8 encoding.

Returns true if the entire data was flushed successfully to the kernel buffer. Returns false if all or part of the data was queued in user memory. 'drain' will be emitted when the buffer is again free.

The optional callback parameter will be executed when the data is finally written out, which may not be immediately.

See Writable stream write() method for more information.

This property represents the state of the connection as a string.

Aliases to net.createConnection().

Possible signatures:

Alias to net.createConnection(options[, connectListener]).

Alias to net.createConnection(path[, connectListener]).

Alias to net.createConnection(port[, host][, connectListener]).

A factory function, which creates a new net.Socket, immediately initiates connection with socket.connect(), then returns the net.Socket that starts the connection.

When the connection is established, a 'connect' event will be emitted on the returned socket. The last parameter connectListener, if supplied, will be added as a listener for the 'connect' event once.

Possible signatures:

The net.connect() function is an alias to this function.

For available options, see new net.Socket([options]) and socket.connect(options[, connectListener]).

Additional options:

Following is an example of a client of the echo server described in the net.createServer() section:

To connect on the socket /tmp/echo.sock:

Initiates an IPC connection.

This function creates a new net.Socket with all options set to default, immediately initiates connection with socket.connect(path[, connectListener]), then returns the net.Socket that starts the connection.

Initiates a TCP connection.

This function creates a new net.Socket with all options set to default, immediately initiates connection with socket.connect(port[, host][, connectListener]), then returns the net.Socket that starts the connection.

Sets the default value of the autoSelectFamily聽option of socket.connect(options).

Gets the current default value of the autoSelectFamily聽option of socket.connect(options).

The noDelay, keepAlive, and keepAliveInitialDelay options are supported now.

Added in: v0.5.0

options

connectionListener Automatically set as a listener for the 'connection' event.

Returns:

Creates a new TCP or IPC server.

If allowHalfOpen is set to true, when the other end of the socket signals the end of transmission, the server will only send back the end of transmission when socket.end() is explicitly called. For example, in the context of TCP, when a FIN packed is received, a FIN packed is sent back only when socket.end() is explicitly called. Until then the connection is half-closed (non-readable but still writable). See 'end' event and RFC 1122 (section 4.2.2.13) for more information.

If pauseOnConnect is set to true, then the socket associated with each incoming connection will be paused, and no data will be read from its handle. This allows connections to be passed between processes without any data being read by the original process. To begin reading data from a paused socket, call socket.resume().

The server can be a TCP server or an IPC server, depending on what it listen() to.

Here is an example of a TCP echo server which listens for connections on port 8124:

Test this by using telnet:

To listen on the socket /tmp/echo.sock:

Use nc to connect to a Unix domain socket server:

Returns 6 if input is an IPv6 address. Returns 4 if input is an IPv4 address in dot-decimal notation with no leading zeroes. Otherwise, returns 0.

Returns true if input is an IPv4 address in dot-decimal notation with no leading zeroes. Otherwise, returns false.

Returns true if input is an IPv6 address. Otherwise, returns false.