Kernel

Reference Manual

Version 7.0

Table of Contents

inet

Module

inet

Module Summary

Access to TCP/IP protocols.

Description

This module provides access to TCP/IP protocols.

See also ERTS User's Guide: Inet Configuration for more information about how to configure an Erlang runtime system for IP communication.

The following two Kernel configuration parameters affect the behavior of all sockets opened on an Erlang node:

  • inet_default_connect_options can contain a list of default options used for all sockets returned when doing connect.

  • inet_default_listen_options can contain a list of default options used when issuing a listen call.

When accept is issued, the values of the listening socket options are inherited. No such application variable is therefore needed for accept.

Using the Kernel configuration parameters above, one can set default options for all TCP sockets on a node, but use this with care. Options such as {delay_send,true} can be specified in this way. The following is an example of starting an Erlang node with all sockets using delayed send:

$ erl -sname test -kernel \
inet_default_connect_options '[{delay_send,true}]' \
inet_default_listen_options '[{delay_send,true}]'

Notice that default option {active, true} cannot be changed, for internal reasons.

Addresses as inputs to functions can be either a string or a tuple. For example, the IP address 150.236.20.73 can be passed to gethostbyaddr/1, either as string "150.236.20.73" or as tuple {150, 236, 20, 73}.

IPv4 address examples:

Address          ip_address()
-------          ------------
127.0.0.1        {127,0,0,1}
192.168.42.2     {192,168,42,2}

IPv6 address examples:

Address          ip_address()
-------          ------------
::1             {0,0,0,0,0,0,0,1}
::192.168.42.2  {0,0,0,0,0,0,(192 bsl 8) bor 168,(42 bsl 8) bor 2}
::FFFF:192.168.42.2
                {0,0,0,0,0,16#FFFF,(192 bsl 8) bor 168,(42 bsl 8) bor 2}
3ffe:b80:1f8d:2:204:acff:fe17:bf38
                {16#3ffe,16#b80,16#1f8d,16#2,16#204,16#acff,16#fe17,16#bf38}
fe80::204:acff:fe17:bf38
                {16#fe80,0,0,0,0,16#204,16#acff,16#fe17,16#bf38}

Function parse_address/1 can be useful:

1> inet:parse_address("192.168.42.2").
{ok,{192,168,42,2}}
2> inet:parse_address("::FFFF:192.168.42.2").
{ok,{0,0,0,0,0,65535,49320,10754}}

Data Types

hostent() =
    #hostent{h_name = inet:hostname(),
             h_aliases = [inet:hostname()],
             h_addrtype = inet | inet6,
             h_length = integer() >= 0,
             h_addr_list = [inet:ip_address()]}

The record is defined in the Kernel include file "inet.hrl".

Add the following directive to the module:

-include_lib("kernel/include/inet.hrl").

hostname() = atom() | string()
ip4_address() = {0..255, 0..255, 0..255, 0..255}
ip6_address() =
    {0..65535,
     0..65535,
     0..65535,
     0..65535,
     0..65535,
     0..65535,
     0..65535,
     0..65535}

port_number() = 0..65535

A general address format on the form {Family, Destination} where Family is an atom such as local and the format of Destination depends on Family, and is a complete address (for example an IP address including port number).

inet_address() =
    {inet, {ip4_address() | any | loopback, port_number()}}

Warning

This address format is for now experimental and for completeness to make all address families have a {Family, Destination} representation.

inet6_address() =
    {inet6, {ip6_address() | any | loopback, port_number()}}

Warning

This address format is for now experimental and for completeness to make all address families have a {Family, Destination} representation.

local_address() = {local, File :: binary() | string()}

This address family only works on Unix-like systems.

File is normally a file pathname in a local filesystem. It is limited in length by the operating system, traditionally to 108 bytes.

A binary() is passed as is to the operating system, but a string() is encoded according to the system filename encoding mode.

Other addresses are possible, for example Linux implements "Abstract Addresses". See the documentation for Unix Domain Sockets on your system, normally unix in manual section 7.

In most API functions where you can use this address family the port number must be 0.

socket_address() =
    ip_address() | any | loopback | local_address()

returned_non_ip_address() =
    {local, binary()} | {unspec, <<>>} | {undefined, any()}

Addresses besides ip_address() ones that are returned from socket API functions. See in particular local_address(). The unspec family corresponds to AF_UNSPEC and can occur if the other side has no socket address. The undefined family can only occur in the unlikely event of an address family that the VM does not recognize.

ancillary_data() =
    [{tos, byte()} | {tclass, byte()} | {ttl, byte()}]

Ancillary data received with the data packet, read with the socket option pktoptions from a TCP socket, or to set in a call to gen_udp:send/4 or gen_udp:send/5.

The value(s) correspond to the currently active socket options recvtos, recvtclass and recvttl, or for a single send operation the option(s) to override the currently active socket option(s).

getifaddrs_ifopts() =
    [Ifopt ::
         {flags,
          Flags ::
              [up | broadcast | loopback | pointtopoint |
               running | multicast]} |
         {addr, Addr :: ip_address()} |
         {netmask, Netmask :: ip_address()} |
         {broadaddr, Broadaddr :: ip_address()} |
         {dstaddr, Dstaddr :: ip_address()} |
         {hwaddr, Hwaddr :: [byte()]}]

Interface address description list returned from getifaddrs/0,1 for a named interface, translated from the returned data of the POSIX API function getaddrinfo().

Hwaddr is hardware dependent, for example, on Ethernet interfaces it is the 6-byte Ethernet address (MAC address (EUI-48 address)).

The tuples {addr,Addr}, {netmask,Netmask}, and possibly {broadaddr,Broadaddr} or {dstaddr,Dstaddr} are repeated in the list if the interface has got multiple addresses. An interface may have multiple {flag,_} tuples for example if it has different flags for different address families. Multiple {hwaddr,Hwaddr} tuples is hard to say anything definite about, though. The tuple {flag,Flags} is mandatory, all others are optional.

Do not rely too much on the order of Flags atoms or the Ifopt tuples. There are however some rules:

  • A {flag,_} tuple applies to all other tuples that follow.

  • Immediately after {addr,_} follows {netmask,_}.

  • Immediately thereafter may {broadaddr,_} follow if broadcast is member of Flags, or {dstaddr,_} if pointtopoint is member of Flags. Both {dstaddr,_} and {broadaddr,_} does not occur for the same {addr,_}.

  • Any {netmask,_}, {broadaddr,_}, or {dstaddr,_} tuples that follow an {addr,Addr} tuple concerns the address Addr.

The tuple {hwaddr,_} is not returned on Solaris, as the hardware address historically belongs to the link layer and it is not returned by the Solaris API function getaddrinfo().

Warning

On Windows, the data is fetched from different OS API functions, so the Netmask and Broadaddr values may be calculated, just as some Flags values.

posix() =
    eaddrinuse | eaddrnotavail | eafnosupport | ealready |
    econnaborted | econnrefused | econnreset | edestaddrreq |
    ehostdown | ehostunreach | einprogress | eisconn | emsgsize |
    enetdown | enetunreach | enopkg | enoprotoopt | enotconn |
    enotty | enotsock | eproto | eprotonosupport | eprototype |
    esocktnosupport | etimedout | ewouldblock | exbadport |
    exbadseq |
    file:posix()

An atom that is named from the POSIX error codes used in Unix, and in the runtime libraries of most C compilers. See section POSIX Error Codes.

address_family() = inet | inet6 | local
socket_protocol() = tcp | udp | sctp
stat_option() =
    recv_cnt | recv_max | recv_avg | recv_oct | recv_dvi |
    send_cnt | send_max | send_avg | send_oct | send_pend

Exports

close(Socket) -> ok

Types

Socket = socket()

Closes a socket of any type.

format_error(Reason) -> string()

Types

Reason = posix() | system_limit

Returns a diagnostic error string. For possible POSIX values and corresponding strings, see section POSIX Error Codes.

get_rc() ->
          [{Par :: atom(), Val :: any()} |
           {Par :: atom(), Val1 :: any(), Val2 :: any()}]

Returns the state of the Inet configuration database in form of a list of recorded configuration parameters. For more information, see ERTS User's Guide: Inet Configuration.

Only actual parameters with other than default values are returned, for example not directives that specify other sources for configuration parameters nor directives that clear parameters.

getaddr(Host, Family) -> {ok, Address} | {error, posix()}

Types

Address = ip_address()

Returns the IP address for Host as a tuple of integers. Host can be an IP address, a single hostname, or a fully qualified hostname.

getaddrs(Host, Family) -> {ok, Addresses} | {error, posix()}

Types

Addresses = [ip_address()]

Returns a list of all IP addresses for Host. Host can be an IP address, a single hostname, or a fully qualified hostname.

gethostbyaddr(Address) -> {ok, Hostent} | {error, posix()}

Types

Address = string() | ip_address()
Hostent = hostent()

Returns a hostent record for the host with the specified address.

gethostbyname(Hostname) -> {ok, Hostent} | {error, posix()}

Types

Hostname = hostname()
Hostent = hostent()

Returns a hostent record for the host with the specified hostname.

If resolver option inet6 is true, an IPv6 address is looked up.

gethostbyname(Hostname, Family) ->
                 {ok, Hostent} | {error, posix()}

Types

Hostname = hostname()
Hostent = hostent()

Returns a hostent record for the host with the specified name, restricted to the specified address family.

gethostname() -> {ok, Hostname}

Types

Hostname = string()

Returns the local hostname. Never fails.

getifaddrs() ->
              {ok,
               [{Ifname :: string(),
                 Ifopts :: getifaddrs_ifopts()}]} |
              {error, posix()}
OTP R14B01

Returns a list of 2-tuples containing interface names and the interfaces' addresses. Ifname is a Unicode string and Ifopts is a list of interface address description tuples.

The interface address description tuples are documented under the type of the Ifopts value.

OTP 21.2

Types

Opts = [{netns, Namespace}]
Ifname = string()
Posix = posix()

The same as getifaddrs/0 but the Option {netns, Namespace} sets a network namespace for the OS call, on platforms that supports that feature.

See the socket option {netns, Namespace} under setopts/2.

getopts(Socket, Options) -> {ok, OptionValues} | {error, posix()}

Types

Socket = socket()
Options = [socket_getopt()]

Gets one or more options for a socket. For a list of available options, see setopts/2. See also the description for the type gen_tcp:pktoptions_value() .

The number of elements in the returned OptionValues list does not necessarily correspond to the number of options asked for. If the operating system fails to support an option, it is left out in the returned list. An error tuple is returned only when getting options for the socket is impossible (that is, the socket is closed or the buffer size in a raw request is too large). This behavior is kept for backward compatibility reasons.

A raw option request RawOptReq = {raw, Protocol, OptionNum, ValueSpec} can be used to get information about socket options not (explicitly) supported by the emulator. The use of raw socket options makes the code non-portable, but allows the Erlang programmer to take advantage of unusual features present on a particular platform.

RawOptReq consists of tag raw followed by the protocol level, the option number, and either a binary or the size, in bytes, of the buffer in which the option value is to be stored. A binary is to be used when the underlying getsockopt requires input in the argument field. In this case, the binary size is to correspond to the required buffer size of the return value. The supplied values in a RawOptReq correspond to the second, third, and fourth/fifth parameters to the getsockopt call in the C socket API. The value stored in the buffer is returned as a binary ValueBin, where all values are coded in the native endianess.

Asking for and inspecting raw socket options require low-level information about the current operating system and TCP stack.

Example:

Consider a Linux machine where option TCP_INFO can be used to collect TCP statistics for a socket. Assume you are interested in field tcpi_sacked of struct tcp_info filled in when asking for TCP_INFO. To be able to access this information, you need to know the following:

  • The numeric value of protocol level IPPROTO_TCP
  • The numeric value of option TCP_INFO
  • The size of struct tcp_info
  • The size and offset of the specific field

By inspecting the headers or writing a small C program, it is found that IPPROTO_TCP is 6, TCP_INFO is 11, the structure size is 92 (bytes), the offset of tcpi_sacked is 28 bytes, and the value is a 32-bit integer. The following code can be used to retrieve the value:

get_tcpi_sacked(Sock) ->
    {ok,[{raw,_,_,Info}]} = inet:getopts(Sock,[{raw,6,11,92}]),
    <<_:28/binary,TcpiSacked:32/native,_/binary>> = Info,
    TcpiSacked.

Preferably, you would check the machine type, the operating system, and the Kernel version before executing anything similar to this code.

getstat(Socket) -> {ok, OptionValues} | {error, posix()}
getstat(Socket, Options) -> {ok, OptionValues} | {error, posix()}

Types

Socket = socket()
Options = [stat_option()]
OptionValues = [{stat_option(), integer()}]
stat_option() =
    recv_cnt | recv_max | recv_avg | recv_oct | recv_dvi |
    send_cnt | send_max | send_avg | send_oct | send_pend

Gets one or more statistic options for a socket.

getstat(Socket) is equivalent to getstat(Socket, [recv_avg, recv_cnt, recv_dvi, recv_max, recv_oct, send_avg, send_cnt, send_pend, send_max, send_oct]).

The following options are available:

recv_avg

Average size of packets, in bytes, received by the socket.

recv_cnt

Number of packets received by the socket.

recv_dvi

Average packet size deviation, in bytes, received by the socket.

recv_max

Size of the largest packet, in bytes, received by the socket.

recv_oct

Number of bytes received by the socket.

send_avg

Average size of packets, in bytes, sent from the socket.

send_cnt

Number of packets sent from the socket.

send_pend

Number of bytes waiting to be sent by the socket.

send_max

Size of the largest packet, in bytes, sent from the socket.

send_oct

Number of bytes sent from the socket.

i() -> ok
OTP 21.0
i(Proto :: socket_protocol()) -> ok
OTP 21.0
i(X1 :: socket_protocol(), Fs :: [atom()]) -> ok
OTP 21.0

Lists all TCP, UDP and SCTP sockets, including those that the Erlang runtime system uses as well as those created by the application.

The following options are available:

port

The internal index of the port.

module

The callback module of the socket.

recv

Number of bytes received by the socket.

sent

Number of bytes sent from the socket.

owner

The socket owner process.

local_address

The local address of the socket.

foreign_address

The address and port of the other end of the connection.

state

The connection state.

type

STREAM or DGRAM or SEQPACKET.

ntoa(IpAddress) -> Address | {error, einval}
OTP R16B02

Types

Address = string()
IpAddress = ip_address()

Parses an ip_address() and returns an IPv4 or IPv6 address string.

parse_address(Address) -> {ok, IPAddress} | {error, einval}
OTP R16B

Types

Address = string()
IPAddress = ip_address()

Parses an IPv4 or IPv6 address string and returns an ip4_address() or ip6_address(). Accepts a shortened IPv4 address string.

parse_ipv4_address(Address) -> {ok, IPv4Address} | {error, einval}
OTP R16B

Types

Address = string()
IPv4Address = ip_address()

Parses an IPv4 address string and returns an ip4_address(). Accepts a shortened IPv4 address string.

parse_ipv4strict_address(Address) ->
                            {ok, IPv4Address} | {error, einval}
OTP R16B

Types

Address = string()
IPv4Address = ip_address()

Parses an IPv4 address string containing four fields, that is, not shortened, and returns an ip4_address().

parse_ipv6_address(Address) -> {ok, IPv6Address} | {error, einval}
OTP R16B

Types

Address = string()
IPv6Address = ip_address()

Parses an IPv6 address string and returns an ip6_address(). If an IPv4 address string is specified, an IPv4-mapped IPv6 address is returned.

parse_ipv6strict_address(Address) ->
                            {ok, IPv6Address} | {error, einval}
OTP R16B

Types

Address = string()
IPv6Address = ip_address()

Parses an IPv6 address string and returns an ip6_address(). Does not accept IPv4 addresses.

ipv4_mapped_ipv6_address(X1 :: ip_address()) -> ip_address()
OTP 21.0

Convert an IPv4 address to an IPv4-mapped IPv6 address or the reverse. When converting from an IPv6 address all but the 2 low words are ignored so this function also works on some other types of addresses than IPv4-mapped.

parse_strict_address(Address) -> {ok, IPAddress} | {error, einval}
OTP R16B

Types

Address = string()
IPAddress = ip_address()

Parses an IPv4 or IPv6 address string and returns an ip4_address() or ip6_address(). Does not accept a shortened IPv4 address string.

peername(Socket :: socket()) ->
            {ok,
             {ip_address(), port_number()} |
             returned_non_ip_address()} |
            {error, posix()}

Returns the address and port for the other end of a connection.

Notice that for SCTP sockets, this function returns only one of the peer addresses of the socket. Function peernames/1,2 returns all.

peernames(Socket :: socket()) ->
             {ok,
              [{ip_address(), port_number()} |
               returned_non_ip_address()]} |
             {error, posix()}
OTP R16B03

Equivalent to peernames(Socket, 0).

Notice that the behavior of this function for an SCTP one-to-many style socket is not defined by the SCTP Sockets API Extensions.

peernames(Socket, Assoc) ->
             {ok, [{Address, Port}]} | {error, posix()}
OTP R16B03

Types

Socket = socket()
Assoc = #sctp_assoc_change{} | gen_sctp:assoc_id()
Address = ip_address()
Port = integer() >= 0

Returns a list of all address/port number pairs for the other end of an association Assoc of a socket.

This function can return multiple addresses for multihomed sockets, such as SCTP sockets. For other sockets it returns a one-element list.

Notice that parameter Assoc is by the SCTP Sockets API Extensions defined to be ignored for one-to-one style sockets. What the special value 0 means, hence its behavior for one-to-many style sockets, is unfortunately undefined.

port(Socket) -> {ok, Port} | {error, any()}

Types

Socket = socket()

Returns the local port number for a socket.

setopts(Socket, Options) -> ok | {error, posix()}

Types

Socket = socket()
Options = [socket_setopt()]

Sets one or more options for a socket.

The following options are available:

{active, true | false | once | N}

If the value is true, which is the default, everything received from the socket is sent as messages to the receiving process.

If the value is false (passive mode), the process must explicitly receive incoming data by calling gen_tcp:recv/2,3, gen_udp:recv/2,3, or gen_sctp:recv/1,2 (depending on the type of socket).

If the value is once ({active, once}), one data message from the socket is sent to the process. To receive one more message, setopts/2 must be called again with option {active, once}.

If the value is an integer N in the range -32768 to 32767 (inclusive), the value is added to the socket's count of data messages sent to the controlling process. A socket's default message count is 0. If a negative value is specified, and its magnitude is equal to or greater than the socket's current message count, the socket's message count is set to 0. Once the socket's message count reaches 0, either because of sending received data messages to the process or by being explicitly set, the process is then notified by a special message, specific to the type of socket, that the socket has entered passive mode. Once the socket enters passive mode, to receive more messages setopts/2 must be called again to set the socket back into an active mode.

When using {active, once} or {active, N}, the socket changes behavior automatically when data is received. This can be confusing in combination with connection-oriented sockets (that is, gen_tcp), as a socket with {active, false} behavior reports closing differently than a socket with {active, true} behavior. To simplify programming, a socket where the peer closed, and this is detected while in {active, false} mode, still generates message {tcp_closed,Socket} when set to {active, once}, {active, true}, or {active, N} mode. It is therefore safe to assume that message {tcp_closed,Socket}, possibly followed by socket port termination (depending on option exit_on_close) eventually appears when a socket changes back and forth between {active, true} and {active, false} mode. However, when peer closing is detected it is all up to the underlying TCP/IP stack and protocol.

Notice that {active, true} mode provides no flow control; a fast sender can easily overflow the receiver with incoming messages. The same is true for {active, N} mode, while the message count is greater than zero.

Use active mode only if your high-level protocol provides its own flow control (for example, acknowledging received messages) or the amount of data exchanged is small. {active, false} mode, use of the {active, once} mode, or {active, N} mode with values of N appropriate for the application provides flow control. The other side cannot send faster than the receiver can read.

{broadcast, Boolean} (UDP sockets)

Enables/disables permission to send broadcasts.

{buffer, Size}

The size of the user-level buffer used by the driver. Not to be confused with options sndbuf and recbuf, which correspond to the Kernel socket buffers. For TCP it is recommended to have val(buffer) >= val(recbuf) to avoid performance issues because of unnecessary copying. For UDP the same recommendation applies, but the max should not be larger than the MTU of the network path. val(buffer) is automatically set to the above maximum when recbuf is set. However, as the size set for recbuf usually become larger, you are encouraged to use getopts/2 to analyze the behavior of your operating system.

Note that this is also the maximum amount of data that can be received from a single recv call. If you are using higher than normal MTU consider setting buffer higher.

{delay_send, Boolean}

Normally, when an Erlang process sends to a socket, the driver tries to send the data immediately. If that fails, the driver uses any means available to queue up the message to be sent whenever the operating system says it can handle it. Setting {delay_send, true} makes all messages queue up. The messages sent to the network are then larger but fewer. The option affects the scheduling of send requests versus Erlang processes instead of changing any real property of the socket. The option is implementation-specific. Defaults to false.

{deliver, port | term}

When {active, true}, data is delivered on the form port : {S, {data, [H1,..Hsz | Data]}} or term : {tcp, S, [H1..Hsz | Data]}.

{dontroute, Boolean}

Enables/disables routing bypass for outgoing messages.

{exit_on_close, Boolean}

This option is set to true by default.

The only reason to set it to false is if you want to continue sending data to the socket after a close is detected, for example, if the peer uses gen_tcp:shutdown/2 to shut down the write side.

{header, Size}

This option is only meaningful if option binary was specified when the socket was created. If option header is specified, the first Size number bytes of data received from the socket are elements of a list, and the remaining data is a binary specified as the tail of the same list. For example, if Size == 2, the data received matches [Byte1,Byte2|Binary].

{high_msgq_watermark, Size}

The socket message queue is set to a busy state when the amount of data on the message queue reaches this limit. Notice that this limit only concerns data that has not yet reached the ERTS internal socket implementation. Defaults to 8 kB.

Senders of data to the socket are suspended if either the socket message queue is busy or the socket itself is busy.

For more information, see options low_msgq_watermark, high_watermark, and low_watermark.

Notice that distribution sockets disable the use of high_msgq_watermark and low_msgq_watermark. Instead use the distribution buffer busy limit, which is a similar feature.

{high_watermark, Size} (TCP/IP sockets)

The socket is set to a busy state when the amount of data queued internally by the ERTS socket implementation reaches this limit. Defaults to 8 kB.

Senders of data to the socket are suspended if either the socket message queue is busy or the socket itself is busy.

For more information, see options low_watermark, high_msgq_watermark, and low_msqg_watermark.

{ipv6_v6only, Boolean}

Restricts the socket to use only IPv6, prohibiting any IPv4 connections. This is only applicable for IPv6 sockets (option inet6).

On most platforms this option must be set on the socket before associating it to an address. It is therefore only reasonable to specify it when creating the socket and not to use it when calling function (setopts/2) containing this description.

The behavior of a socket with this option set to true is the only portable one. The original idea when IPv6 was new of using IPv6 for all traffic is now not recommended by FreeBSD (you can use {ipv6_v6only,false} to override the recommended system default value), forbidden by OpenBSD (the supported GENERIC kernel), and impossible on Windows (which has separate IPv4 and IPv6 protocol stacks). Most Linux distros still have a system default value of false. This policy shift among operating systems to separate IPv6 from IPv4 traffic has evolved, as it gradually proved hard and complicated to get a dual stack implementation correct and secure.

On some platforms, the only allowed value for this option is true, for example, OpenBSD and Windows. Trying to set this option to false, when creating the socket, fails in this case.

Setting this option on platforms where it does not exist is ignored. Getting this option with getopts/2 returns no value, that is, the returned list does not contain an {ipv6_v6only,_} tuple. On Windows, the option does not exist, but it is emulated as a read-only option with value true.

Therefore, setting this option to true when creating a socket never fails, except possibly on a platform where you have customized the kernel to only allow false, which can be doable (but awkward) on, for example, OpenBSD.

If you read back the option value using getopts/2 and get no value, the option does not exist in the host operating system. The behavior of both an IPv6 and an IPv4 socket listening on the same port, and for an IPv6 socket getting IPv4 traffic is then no longer predictable.

{keepalive, Boolean}(TCP/IP sockets)

Enables/disables periodic transmission on a connected socket when no other data is exchanged. If the other end does not respond, the connection is considered broken and an error message is sent to the controlling process. Defaults to disabled.

{linger, {true|false, Seconds}}

Determines the time-out, in seconds, for flushing unsent data in the close/1 socket call.

The first component is if linger is enabled, the second component is the flushing time-out, in seconds. There are 3 alternatives:

{false, _}

close/1 or shutdown/2 returns immediately, not waiting for data to be flushed, with closing happening in the background.

{true, 0}

Aborts the connection when it is closed. Discards any data still remaining in the send buffers and sends RST to the peer.

This avoids TCP's TIME_WAIT state, but leaves open the possibility that another "incarnation" of this connection being created.

{true, Time} when Time > 0

close/1 or shutdown/2 will not return until all queued messages for the socket have been successfully sent or the linger timeout (Time) has been reached.

{low_msgq_watermark, Size}

If the socket message queue is in a busy state, the socket message queue is set in a not busy state when the amount of data queued in the message queue falls below this limit. Notice that this limit only concerns data that has not yet reached the ERTS internal socket implementation. Defaults to 4 kB.

Senders that are suspended because of either a busy message queue or a busy socket are resumed when the socket message queue and the socket are not busy.

For more information, see options high_msgq_watermark, high_watermark, and low_watermark.

Notice that distribution sockets disable the use of high_msgq_watermark and low_msgq_watermark. Instead they use the distribution buffer busy limit, which is a similar feature.

{low_watermark, Size} (TCP/IP sockets)

If the socket is in a busy state, the socket is set in a not busy state when the amount of data queued internally by the ERTS socket implementation falls below this limit. Defaults to 4 kB.

Senders that are suspended because of a busy message queue or a busy socket are resumed when the socket message queue and the socket are not busy.

For more information, see options high_watermark, high_msgq_watermark, and low_msgq_watermark.

{mode, Mode :: binary | list}

Received Packet is delivered as defined by Mode.

{netns, Namespace :: file:filename_all()}

Sets a network namespace for the socket. Parameter Namespace is a filename defining the namespace, for example, "/var/run/netns/example", typically created by command ip netns add example. This option must be used in a function call that creates a socket, that is, gen_tcp:connect/3,4, gen_tcp:listen/2, gen_udp:open/1,2 or gen_sctp:open/0,1,2, and also getifaddrs/1.

This option uses the Linux-specific syscall setns(), such as in Linux kernel 3.0 or later, and therefore only exists when the runtime system is compiled for such an operating system.

The virtual machine also needs elevated privileges, either running as superuser or (for Linux) having capability CAP_SYS_ADMIN according to the documentation for setns(2). However, during testing also CAP_SYS_PTRACE and CAP_DAC_READ_SEARCH have proven to be necessary.

Example:

setcap cap_sys_admin,cap_sys_ptrace,cap_dac_read_search+epi beam.smp

Notice that the filesystem containing the virtual machine executable (beam.smp in the example) must be local, mounted without flag nosetuid, support extended attributes, and the kernel must support file capabilities. All this runs out of the box on at least Ubuntu 12.04 LTS, except that SCTP sockets appear to not support network namespaces.

Namespace is a filename and is encoded and decoded as discussed in module file, with the following exceptions:

  • Emulator flag +fnu is ignored.

  • getopts/2 for this option returns a binary for the filename if the stored filename cannot be decoded. This is only to occur if you set the option using a binary that cannot be decoded with the emulator's filename encoding: file:native_name_encoding/0.

{bind_to_device, Ifname :: binary()}

Binds a socket to a specific network interface. This option must be used in a function call that creates a socket, that is, gen_tcp:connect/3,4, gen_tcp:listen/2, gen_udp:open/1,2, or gen_sctp:open/0,1,2.

Unlike getifaddrs/0, Ifname is encoded a binary. In the unlikely case that a system is using non-7-bit-ASCII characters in network device names, special care has to be taken when encoding this argument.

This option uses the Linux-specific socket option SO_BINDTODEVICE, such as in Linux kernel 2.0.30 or later, and therefore only exists when the runtime system is compiled for such an operating system.

Before Linux 3.8, this socket option could be set, but could not retrieved with getopts/2. Since Linux 3.8, it is readable.

The virtual machine also needs elevated privileges, either running as superuser or (for Linux) having capability CAP_NET_RAW.

The primary use case for this option is to bind sockets into Linux VRF instances.

list

Received Packet is delivered as a list.

binary

Received Packet is delivered as a binary.

{nodelay, Boolean}(TCP/IP sockets)

If Boolean == true, option TCP_NODELAY is turned on for the socket, which means that also small amounts of data are sent immediately.

{nopush, Boolean}(TCP/IP sockets)

This translates to TCP_NOPUSH on BSD and to TCP_CORK on Linux.

If Boolean == true, the corresponding option is turned on for the socket, which means that small amounts of data are accumulated until a full MSS-worth of data is available or this option is turned off.

Note that while TCP_NOPUSH socket option is available on OSX, its semantics is very different (e.g., unsetting it does not cause immediate send of accumulated data). Hence, nopush option is intentionally ignored on OSX.

{packet, PacketType}(TCP/IP sockets)

Defines the type of packets to use for a socket. Possible values:

raw | 0

No packaging is done.

1 | 2 | 4

Packets consist of a header specifying the number of bytes in the packet, followed by that number of bytes. The header length can be one, two, or four bytes, and containing an unsigned integer in big-endian byte order. Each send operation generates the header, and the header is stripped off on each receive operation.

The 4-byte header is limited to 2Gb.

asn1 | cdr | sunrm | fcgi | tpkt | line

These packet types only have effect on receiving. When sending a packet, it is the responsibility of the application to supply a correct header. On receiving, however, one message is sent to the controlling process for each complete packet received, and, similarly, each call to gen_tcp:recv/2,3 returns one complete packet. The header is not stripped off.

The meanings of the packet types are as follows:

  • asn1 - ASN.1 BER
  • sunrm - Sun's RPC encoding
  • cdr - CORBA (GIOP 1.1)
  • fcgi - Fast CGI
  • tpkt - TPKT format [RFC1006]
  • line - Line mode, a packet is a line-terminated with newline, lines longer than the receive buffer are truncated
http | http_bin

The Hypertext Transfer Protocol. The packets are returned with the format according to HttpPacket described in erlang:decode_packet/3 in ERTS. A socket in passive mode returns {ok, HttpPacket} from gen_tcp:recv while an active socket sends messages like {http, Socket, HttpPacket}.

httph | httph_bin

These two types are often not needed, as the socket automatically switches from http/http_bin to httph/httph_bin internally after the first line is read. However, there can be occasions when they are useful, such as parsing trailers from chunked encoding.

{packet_size, Integer}(TCP/IP sockets)

Sets the maximum allowed length of the packet body. If the packet header indicates that the length of the packet is longer than the maximum allowed length, the packet is considered invalid. The same occurs if the packet header is too large for the socket receive buffer.

For line-oriented protocols (line, http*), option packet_size also guarantees that lines up to the indicated length are accepted and not considered invalid because of internal buffer limitations.

{line_delimiter, Char}(TCP/IP sockets)

Sets the line delimiting character for line-oriented protocols (line). Defaults to $\n.

{raw, Protocol, OptionNum, ValueBin}

See below.

{read_packets, Integer}(UDP sockets)

Sets the maximum number of UDP packets to read without intervention from the socket when data is available. When this many packets have been read and delivered to the destination process, new packets are not read until a new notification of available data has arrived. Defaults to 5. If this parameter is set too high, the system can become unresponsive because of UDP packet flooding.

{recbuf, Size}

The minimum size of the receive buffer to use for the socket. You are encouraged to use getopts/2 to retrieve the size set by your operating system.

{recvtclass, Boolean}

If set to true activates returning the received TCLASS value on platforms that implements the protocol IPPROTO_IPV6 option IPV6_RECVTCLASS or IPV6_2292RECVTCLASS for the socket. The value is returned as a {tclass,TCLASS} tuple regardless of if the platform returns an IPV6_TCLASS or an IPV6_RECVTCLASS CMSG value.

For packet oriented sockets that supports receiving ancillary data with the payload data (gen_udp and gen_sctp), the TCLASS value is returned in an extended return tuple contained in an ancillary data list. For stream oriented sockets (gen_tcp) the only way to get the TCLASS value is if the platform supports the pktoptions option.

{recvtos, Boolean}

If set to true activates returning the received TOS value on platforms that implements the protocol IPPROTO_IP option IP_RECVTOS for the socket. The value is returned as a {tos,TOS} tuple regardless of if the platform returns an IP_TOS or an IP_RECVTOS CMSG value.

For packet oriented sockets that supports receiving ancillary data with the payload data (gen_udp and gen_sctp), the TOS value is returned in an extended return tuple contained in an ancillary data list. For stream oriented sockets (gen_tcp) the only way to get the TOS value is if the platform supports the pktoptions option.

{recvttl, Boolean}

If set to true activates returning the received TTL value on platforms that implements the protocol IPPROTO_IP option IP_RECVTTL for the socket. The value is returned as a {ttl,TTL} tuple regardless of if the platform returns an IP_TTL or an IP_RECVTTL CMSG value.

For packet oriented sockets that supports receiving ancillary data with the payload data (gen_udp and gen_sctp), the TTL value is returned in an extended return tuple contained in an ancillary data list. For stream oriented sockets (gen_tcp) the only way to get the TTL value is if the platform supports the pktoptions option.

{reuseaddr, Boolean}

Allows or disallows local reuse of port numbers. By default, reuse is disallowed.

{send_timeout, Integer}

Only allowed for connection-oriented sockets.

Specifies a longest time to wait for a send operation to be accepted by the underlying TCP stack. When the limit is exceeded, the send operation returns {error,timeout}. How much of a packet that got sent is unknown; the socket is therefore to be closed whenever a time-out has occurred (see send_timeout_close below). Defaults to infinity.

{send_timeout_close, Boolean}

Only allowed for connection-oriented sockets.

Used together with send_timeout to specify whether the socket is to be automatically closed when the send operation returns {error,timeout}. The recommended setting is true, which automatically closes the socket. Defaults to false because of backward compatibility.

{show_econnreset, Boolean}(TCP/IP sockets)

When this option is set to false, which is default, an RST received from the TCP peer is treated as a normal close (as though an FIN was sent). A caller to gen_tcp:recv/2 gets {error, closed}. In active mode, the controlling process receives a {tcp_closed, Socket} message, indicating that the peer has closed the connection.

Setting this option to true allows you to distinguish between a connection that was closed normally, and one that was aborted (intentionally or unintentionally) by the TCP peer. A call to gen_tcp:recv/2 returns {error, econnreset}. In active mode, the controlling process receives a {tcp_error, Socket, econnreset} message before the usual {tcp_closed, Socket}, as is the case for any other socket error. Calls to gen_tcp:send/2 also returns {error, econnreset} when it is detected that a TCP peer has sent an RST.

A connected socket returned from gen_tcp:accept/1 inherits the show_econnreset setting from the listening socket.

{sndbuf, Size}

The minimum size of the send buffer to use for the socket. You are encouraged to use getopts/2, to retrieve the size set by your operating system.

{priority, Integer}

Sets the SO_PRIORITY socket level option on platforms where this is implemented. The behavior and allowed range varies between different systems. The option is ignored on platforms where it is not implemented. Use with caution.

{tos, Integer}

Sets IP_TOS IP level options on platforms where this is implemented. The behavior and allowed range varies between different systems. The option is ignored on platforms where it is not implemented. Use with caution.

{tclass, Integer}

Sets IPV6_TCLASS IP level options on platforms where this is implemented. The behavior and allowed range varies between different systems. The option is ignored on platforms where it is not implemented. Use with caution.

In addition to these options, raw option specifications can be used. The raw options are specified as a tuple of arity four, beginning with tag raw, followed by the protocol level, the option number, and the option value specified as a binary. This corresponds to the second, third, and fourth arguments to the setsockopt call in the C socket API. The option value must be coded in the native endianess of the platform and, if a structure is required, must follow the structure alignment conventions on the specific platform.

Using raw socket options requires detailed knowledge about the current operating system and TCP stack.

Example:

This example concerns the use of raw options. Consider a Linux system where you want to set option TCP_LINGER2 on protocol level IPPROTO_TCP in the stack. You know that on this particular system it defaults to 60 (seconds), but you want to lower it to 30 for a particular socket. Option TCP_LINGER2 is not explicitly supported by inet, but you know that the protocol level translates to number 6, the option number to number 8, and the value is to be specified as a 32-bit integer. You can use this code line to set the option for the socket named Sock:

inet:setopts(Sock,[{raw,6,8,<<30:32/native>>}]),

As many options are silently discarded by the stack if they are specified out of range; it can be a good idea to check that a raw option is accepted. The following code places the value in variable TcpLinger2:

{ok,[{raw,6,8,<<TcpLinger2:32/native>>}]}=inet:getopts(Sock,[{raw,6,8,4}]),

Code such as these examples is inherently non-portable, even different versions of the same OS on the same platform can respond differently to this kind of option manipulation. Use with care.

Notice that the default options for TCP/IP sockets can be changed with the Kernel configuration parameters mentioned in the beginning of this manual page.

sockname(Socket :: socket()) ->
            {ok,
             {ip_address(), port_number()} |
             returned_non_ip_address()} |
            {error, posix()}

Returns the local address and port number for a socket.

Notice that for SCTP sockets this function returns only one of the socket addresses. Function socknames/1,2 returns all.

socknames(Socket :: socket()) ->
             {ok,
              [{ip_address(), port_number()} |
               returned_non_ip_address()]} |
             {error, posix()}
OTP R16B03

Equivalent to socknames(Socket, 0).

socknames(Socket, Assoc) ->
             {ok, [{Address, Port}]} | {error, posix()}
OTP R16B03

Types

Socket = socket()
Assoc = #sctp_assoc_change{} | gen_sctp:assoc_id()
Address = ip_address()
Port = integer() >= 0

Returns a list of all local address/port number pairs for a socket for the specified association Assoc.

This function can return multiple addresses for multihomed sockets, such as SCTP sockets. For other sockets it returns a one-element list.

Notice that parameter Assoc is by the SCTP Sockets API Extensions defined to be ignored for one-to-one style sockets. For one-to-many style sockets, the special value 0 is defined to mean that the returned addresses must be without any particular association. How different SCTP implementations interpret this varies somewhat.

POSIX Error Codes

  • e2big - Too long argument list
  • eacces - Permission denied
  • eaddrinuse - Address already in use
  • eaddrnotavail - Cannot assign requested address
  • eadv - Advertise error
  • eafnosupport - Address family not supported by protocol family
  • eagain - Resource temporarily unavailable
  • ealign - EALIGN
  • ealready - Operation already in progress
  • ebade - Bad exchange descriptor
  • ebadf - Bad file number
  • ebadfd - File descriptor in bad state
  • ebadmsg - Not a data message
  • ebadr - Bad request descriptor
  • ebadrpc - Bad RPC structure
  • ebadrqc - Bad request code
  • ebadslt - Invalid slot
  • ebfont - Bad font file format
  • ebusy - File busy
  • echild - No children
  • echrng - Channel number out of range
  • ecomm - Communication error on send
  • econnaborted - Software caused connection abort
  • econnrefused - Connection refused
  • econnreset - Connection reset by peer
  • edeadlk - Resource deadlock avoided
  • edeadlock - Resource deadlock avoided
  • edestaddrreq - Destination address required
  • edirty - Mounting a dirty fs without force
  • edom - Math argument out of range
  • edotdot - Cross mount point
  • edquot - Disk quota exceeded
  • eduppkg - Duplicate package name
  • eexist - File already exists
  • efault - Bad address in system call argument
  • efbig - File too large
  • ehostdown - Host is down
  • ehostunreach - Host is unreachable
  • eidrm - Identifier removed
  • einit - Initialization error
  • einprogress - Operation now in progress
  • eintr - Interrupted system call
  • einval - Invalid argument
  • eio - I/O error
  • eisconn - Socket is already connected
  • eisdir - Illegal operation on a directory
  • eisnam - Is a named file
  • el2hlt - Level 2 halted
  • el2nsync - Level 2 not synchronized
  • el3hlt - Level 3 halted
  • el3rst - Level 3 reset
  • elbin - ELBIN
  • elibacc - Cannot access a needed shared library
  • elibbad - Accessing a corrupted shared library
  • elibexec - Cannot exec a shared library directly
  • elibmax - Attempting to link in more shared libraries than system limit
  • elibscn - .lib section in a.out corrupted
  • elnrng - Link number out of range
  • eloop - Too many levels of symbolic links
  • emfile - Too many open files
  • emlink - Too many links
  • emsgsize - Message too long
  • emultihop - Multihop attempted
  • enametoolong - Filename too long
  • enavail - Unavailable
  • enet - ENET
  • enetdown - Network is down
  • enetreset - Network dropped connection on reset
  • enetunreach - Network is unreachable
  • enfile - File table overflow
  • enoano - Anode table overflow
  • enobufs - No buffer space available
  • enocsi - No CSI structure available
  • enodata - No data available
  • enodev - No such device
  • enoent - No such file or directory
  • enoexec - Exec format error
  • enolck - No locks available
  • enolink - Link has been severed
  • enomem - Not enough memory
  • enomsg - No message of desired type
  • enonet - Machine is not on the network
  • enopkg - Package not installed
  • enoprotoopt - Bad protocol option
  • enospc - No space left on device
  • enosr - Out of stream resources or not a stream device
  • enosym - Unresolved symbol name
  • enosys - Function not implemented
  • enotblk - Block device required
  • enotconn - Socket is not connected
  • enotdir - Not a directory
  • enotempty - Directory not empty
  • enotnam - Not a named file
  • enotsock - Socket operation on non-socket
  • enotsup - Operation not supported
  • enotty - Inappropriate device for ioctl
  • enotuniq - Name not unique on network
  • enxio - No such device or address
  • eopnotsupp - Operation not supported on socket
  • eperm - Not owner
  • epfnosupport - Protocol family not supported
  • epipe - Broken pipe
  • eproclim - Too many processes
  • eprocunavail - Bad procedure for program
  • eprogmismatch - Wrong program version
  • eprogunavail - RPC program unavailable
  • eproto - Protocol error
  • eprotonosupport - Protocol not supported
  • eprototype - Wrong protocol type for socket
  • erange - Math result unrepresentable
  • erefused - EREFUSED
  • eremchg - Remote address changed
  • eremdev - Remote device
  • eremote - Pathname hit remote filesystem
  • eremoteio - Remote I/O error
  • eremoterelease - EREMOTERELEASE
  • erofs - Read-only filesystem
  • erpcmismatch - Wrong RPC version
  • erremote - Object is remote
  • eshutdown - Cannot send after socket shutdown
  • esocktnosupport - Socket type not supported
  • espipe - Invalid seek
  • esrch - No such process
  • esrmnt - Srmount error
  • estale - Stale remote file handle
  • esuccess - Error 0
  • etime - Timer expired
  • etimedout - Connection timed out
  • etoomanyrefs - Too many references
  • etxtbsy - Text file or pseudo-device busy
  • euclean - Structure needs cleaning
  • eunatch - Protocol driver not attached
  • eusers - Too many users
  • eversion - Version mismatch
  • ewouldblock - Operation would block
  • exdev - Cross-domain link
  • exfull - Message tables full
  • nxdomain - Hostname or domain name cannot be found