Functions for communicating with sockets using the SCTP protocol.

This module provides functions for communicating with sockets using the SCTP protocol. The implementation assumes that the OS kernel supports SCTP (RFC 2960) through the user-level Sockets API Extensions.

During development, this implementation was tested on:

  • Linux Fedora Core 5.0 (kernel 2.6.15-2054 or later is needed)
  • Solaris 10, 11

During OTP adaptation it was tested on:

  • SUSE Linux Enterprise Server 10 (x86_64) kernel, with lksctp-tools-1.0.6
  • Briefly on Solaris 10
  • SUSE Linux Enterprise Server 10 Service Pack 1 (x86_64) kernel with lksctp-tools-1.0.7
  • FreeBSD 8.2

This module was written for one-to-many style sockets (type seqpacket). With the addition of peeloff/2, one-to-one style sockets (type stream) were introduced.

Record definitions for this module can be found using:


These record definitions use the "new" spelling 'adaptation', not the deprecated 'adaption', regardless of which spelling the underlying C API uses.

An opaque term returned in, for example, #sctp_paddr_change{}, which identifies an association for an SCTP socket. The term is opaque except for the special value 0, which has a meaning such as "the whole endpoint" or "all future associations".

One of the SCTP Socket Options.

Socket identifier returned from open/*.


Socket = sctp_socket()
Assoc = #sctp_assoc_change{}

Abnormally terminates the association specified by Assoc, without flushing of unsent data. The socket itself remains open. Other associations opened on this socket are still valid, and the socket can be used in new associations.


Socket = sctp_socket()

Closes the socket and all associations on it. The unsent data is flushed as in eof/2. The close/1 call is blocking or otherwise depending of the value of the linger socket option. If close does not linger or linger time-out expires, the call returns and the data is flushed in the background.


Socket = sctp_socket()
Opts = [Opt :: option()]

Same as connect(Socket, SockAddr, Opts, infinity).


Socket = sctp_socket()
Opts = [Opt :: option()]
Timeout = timeout()

This is conceptually the same as connect/5, only with the difference that we use a socket address, socket:sockaddr_in() or socket:sockaddr_in6() instead of an address (inet:ip_address() or inet:hostname()) and port-number.


Socket = sctp_socket()
Opts = [Opt :: option()]
Timeout = timeout()

Establishes a new association for socket Socket, with the peer (SCTP server socket) specified by Addr and Port. Timeout, is expressed in milliseconds. A socket can be associated with multiple peers.


Using a value of Timeout less than the maximum time taken by the OS to establish an association (around 4.5 minutes if the default values from RFC 4960 are used), can result in inconsistent or incorrect return values. This is especially relevant for associations sharing the same Socket (that is, source address and port), as the controlling process blocks until connect/* returns. connect_init/* provides an alternative without this limitation.

The result of connect/* is an #sctp_assoc_change{} event that contains, in particular, the new Association ID:

      state             = atom(),
      error             = integer(),
      outbound_streams  = integer(),
      inbound_streams   = integer(),
      assoc_id          = assoc_id()

The number of outbound and inbound streams can be set by giving an sctp_initmsg option to connect as in:

connect(Socket, Ip, Port>,

All options Opt are set on the socket before the association is attempted. If an option record has undefined field values, the options record is first read from the socket for those values. In effect, Opt option records only define field values to change before connecting.

The returned outbound_streams and inbound_streams are the stream numbers on the socket. These can be different from the requested values (OutStreams and MaxInStreams, respectively) if the peer requires lower values.

state can have the following values:


Association is successfully established. This indicates a successful completion of connect.


The association cannot be established (connect/* failure).

Other states do not normally occur in the output from connect/*. Rather, they can occur in #sctp_assoc_change{} events received instead of data in recv/* calls. All of them indicate losing the association because of various error conditions, and are listed here for the sake of completeness:


Field error can provide more detailed diagnostics. The error field value can be converted into a string using error_string/1.


Socket = sctp_socket()
Opts = [option()]
Timeout = timeout()

Initiates a new association for socket Socket, with the peer (SCTP server socket) specified by Addr and Port.

The fundamental difference between this API and connect/* is that the return value is that of the underlying OS connect(2) system call. If ok is returned, the result of the association establishment is received by the calling process as an #sctp_assoc_change{} event. The calling process must be prepared to receive this, or poll for it using recv/*, depending on the value of the active option.

The parameters are as described in connect/*, except the Timeout value.

The timer associated with Timeout only supervises IP resolution of Addr.


Socket = sctp_socket()
Assoc = #sctp_assoc_change{}
Reason = term()

Gracefully terminates the association specified by Assoc, with flushing of all unsent data. The socket itself remains open. Other associations opened on this socket are still valid. The socket can be used in new associations.


ErrorNumber = integer()

Translates an SCTP error number from, for example, #sctp_remote_error{} or #sctp_send_failed{} into an explanatory string, or one of the atoms ok for no error or undefined for an unrecognized error.


Socket = sctp_socket()
Backlog = integer()
Reason = term()

Sets up a socket to listen on the IP address and port number it is bound to.

For type seqpacket, sockets (the default) IsServer must be true or false. In contrast to TCP, there is no listening queue length in SCTP. If IsServer is true, the socket accepts new associations, that is, it becomes an SCTP server socket.

For type stream, sockets Backlog define the backlog queue length just like in TCP.


Opts = [Opt]
Opt =
    {ifaddr, IP | SockAddr} |
    {ip, IP} |
    {port, Port} |
    inet:address_family() |
    {type, SockType} |
    {netns, file:filename_all()} |
    {bind_to_device, binary()} |
IP = inet:ip_address() | any | loopback
SockType = seqpacket | stream
Socket = sctp_socket()

Creates an SCTP socket and binds it to the local addresses specified by all {ip,IP} (or synonymously {ifaddr,IP}) options (this feature is called SCTP multi-homing). The default IP and Port are any and 0, meaning bind to all local addresses on any free port.

It is also possible to use {ifaddr, SockAddr}, in which case it takes precedence over the ip and port options. These options can however be used to update the address and port of ifaddr (if they occur after ifaddr in the options list), although this is not recommended.

Other options:


Sets up the socket for IPv6.


Sets up the socket for IPv4. This is the default.

A default set of socket options is used. In particular, the socket is opened in binary and passive mode, with SockType seqpacket, and with reasonably large kernel and driver buffers.

If the socket is in passive mode data can be received through the recv/1,2 calls.

If the socket is in active mode data received data is delivered to the controlling process as messages:

{sctp, Socket, FromIP, FromPort, {AncData, Data}}

See recv/1,2 for a description of the message fields.


This message format unfortunately differs slightly from the gen_udp message format with ancillary data, and from the recv/1,2 return tuple format.


Socket = sctp_socket()
Assoc = #sctp_assoc_change{} | assoc_id()
NewSocket = sctp_socket()
Reason = term()

Branches off an existing association Assoc in a socket Socket of type seqpacket (one-to-many style) into a new socket NewSocket of type stream (one-to-one style).

The existing association argument Assoc can be either a #sctp_assoc_change{} record as returned from, for example, recv/*, connect/*, or from a listening socket in active mode. It can also be just the field assoc_id integer from such a record.


Socket = sctp_socket()
Timeout = timeout()
AncData = [#sctp_sndrcvinfo{} | inet:ancillary_data()]
Data =
    binary() |
    string() |
    #sctp_sndrcvinfo{} |
    #sctp_assoc_change{} |
    #sctp_paddr_change{} |
Reason =
    inet:posix() |
    #sctp_send_failed{} |
    #sctp_paddr_change{} |
    #sctp_pdapi_event{} |
    #sctp_remote_error{} |

Receives the Data message from any association of the socket. If the receive times out, {error,timeout} is returned. The default time-out is infinity. FromIP and FromPort indicate the address of the sender.

AncData is a list of ancillary data items that can be received along with the main Data. This list can be empty, or contain a single #sctp_sndrcvinfo{} record if receiving of such ancillary data is enabled (see option sctp_events). It is enabled by default, as such ancillary data provides an easy way of determining the association and stream over which the message is received. (An alternative way is to get the association ID from FromIP and FromPort using socket option sctp_get_peer_addr_info, but this does still not produce the stream number).

AncData may also contain ancillary data from the socket options recvtos, recvtclass or recvttl, if that is supported by the platform for the socket.

The Data received can be a binary() or a list() of bytes (integers in the range 0 through 255) depending on the socket mode, or an SCTP event.

Possible SCTP events:

  • #sctp_sndrcvinfo{}
  • #sctp_assoc_change{}
  • #sctp_paddr_change{
          addr      = {ip_address(),port()},
          state     = atom(),
          error     = integer(),
          assoc_id  = assoc_id()

    Indicates change of the status of the IP address of the peer specified by addr within association assoc_id. Possible values of state (mostly self-explanatory) include:


    In case of an error (for example, addr_unreachable), field error provides more diagnostics. In such cases, event #sctp_paddr_change{} is automatically converted into an error term returned by recv. The error field value can be converted into a string using error_string/1.

  • #sctp_send_failed{
          flags     = true | false,
          error     = integer(),
          info      = #sctp_sndrcvinfo{},
          assoc_id  = assoc_id()
          data      = binary()

    The sender can receive this event if a send operation fails.


    A Boolean specifying if the data has been transmitted over the wire.


    Provides extended diagnostics, use error_string/1.


    The original #sctp_sndrcvinfo{} record used in the failed send/*.


    The whole original data chunk attempted to be sent.

    In the current implementation of the Erlang/SCTP binding, this event is internally converted into an error term returned by recv/*.

  • #sctp_adaptation_event{
          adaptation_ind = integer(),
          assoc_id       = assoc_id()

    Delivered when a peer sends an adaptation layer indication parameter (configured through option sctp_adaptation_layer). Notice that with the current implementation of the Erlang/SCTP binding, this event is disabled by default.

  • #sctp_pdapi_event{
          indication = sctp_partial_delivery_aborted,
          assoc_id   = assoc_id()

    A partial delivery failure. In the current implementation of the Erlang/SCTP binding, this event is internally converted into an error term returned by recv/*.


Socket = sctp_socket()
SndRcvInfo = #sctp_sndrcvinfo{}
Reason = term()

Sends the Data message with all sending parameters from a #sctp_sndrcvinfo{} record. This way, the user can specify the PPID (passed to the remote end) and context (passed to the local SCTP layer), which can be used, for example, for error identification. However, such a fine level of user control is rarely required. The function send/4 is sufficient for most applications.

The set of admissible SCTP socket options is by construction orthogonal to the sets of TCP, UDP, and generic inet options. Only options listed here are allowed for SCTP sockets. Options can be set on the socket using open/1,2 or inet:setopts/2, retrieved using inet:getopts/2. Options can be changed when calling connect/4,5.

{mode, list|binary} or just list or binary

Determines the type of data returned from recv/1,2.

{active, true|false|once|N}
  • If false (passive mode, the default), the caller must do an explicit recv call to retrieve the available data from the socket.

  • If true|once|N (active modes) received data or events are sent to the owning process. See open/0..2 for the message format.

  • If true (full active mode) there is no flow control.


    Note that this can cause the message queue to overflow causing for example the virtual machine to run out of memory and crash.

  • If once, only one message is automatically placed in the message queue, and after that the mode is automatically reset to passive. This provides flow control and the possibility for the receiver to listen for its incoming SCTP data interleaved with other inter-process messages.

  • If active is specified as an integer N in the range -32768 to 32767 (inclusive), that number is added to the socket's counting of data messages to be delivered to the controlling process. If the result of the addition is negative, the count is set to 0. Once the count reaches 0, either through the delivery of messages or by being explicitly set with inet:setopts/2, the socket mode is automatically reset to passive ({active, false}). When a socket in this active mode transitions to passive mode, the message {sctp_passive, Socket} is sent to the controlling process to notify it that if it wants to receive more data messages from the socket, it must call inet:setopts/2 to set the socket back into an active mode.

{tos, integer()}

Sets the Type-Of-Service field on the IP datagrams that are sent, to the specified value. This effectively determines a prioritization policy for the outbound packets. The acceptable values are system-dependent.

{priority, integer()}

A protocol-independent equivalent of tos above. Setting priority implies setting tos as well.

{dontroute, true|false}

Defaults to false. If true, the kernel does not send packets through any gateway, only sends them to directly connected hosts.

{reuseaddr, true|false}

Defaults to false. If true, the local binding address {IP,Port} of the socket can be reused immediately. No waiting in state CLOSE_WAIT is performed (can be required for high-throughput servers).

{sndbuf, integer()}

The size, in bytes, of the OS kernel send buffer for this socket. Sending errors would occur for datagrams larger than val(sndbuf). Setting this option also adjusts the size of the driver buffer (see buffer above).

{recbuf, integer()}

The size, in bytes, of the OS kernel receive buffer for this socket. Sending errors would occur for datagrams larger than val(recbuf). Setting this option also adjusts the size of the driver buffer (see buffer above).

{sctp_module, module()}

Overrides which callback module is used. Defaults to inet_sctp for IPv4 and inet6_sctp for IPv6.

{sctp_rtoinfo, #sctp_rtoinfo{}}
      assoc_id = assoc_id(),
      initial  = integer(),
      max      = integer(),
      min      = integer()

Determines retransmission time-out parameters, in milliseconds, for the association(s) specified by assoc_id.

assoc_id = 0 (default) indicates the whole endpoint. See RFC 2960 and Sockets API Extensions for SCTP for the exact semantics of the field values.

{sctp_associnfo, #sctp_assocparams{}}
      assoc_id                 = assoc_id(),
      asocmaxrxt               = integer(),
      number_peer_destinations = integer(),
      peer_rwnd                = integer(),
      local_rwnd               = integer(),
      cookie_life              = integer()

Determines association parameters for the association(s) specified by assoc_id.

assoc_id = 0 (default) indicates the whole endpoint. See Sockets API Extensions for SCTP for the discussion of their semantics. Rarely used.

{sctp_initmsg, #sctp_initmsg{}}
     num_ostreams   = integer(),
     max_instreams  = integer(),
     max_attempts   = integer(),
     max_init_timeo = integer()

Determines the default parameters that this socket tries to negotiate with its peer while establishing an association with it. Is to be set after open/* but before the first connect/*. #sctp_initmsg{} can also be used as ancillary data with the first call of send/* to a new peer (when a new association is created).

Number of outbound streams
Maximum number of inbound streams
Maximum retransmissions while establishing an association
Time-out, in milliseconds, for establishing an association
{sctp_autoclose, integer() >= 0}

Determines the time, in seconds, after which an idle association is automatically closed. 0 means that the association is never automatically closed.

{sctp_nodelay, true|false}

Turns on|off the Nagle algorithm for merging small packets into larger ones. This improves throughput at the expense of latency.

{sctp_disable_fragments, true|false}

If true, induces an error on an attempt to send a message larger than the current PMTU size (which would require fragmentation/reassembling). Notice that message fragmentation does not affect the logical atomicity of its delivery; this option is provided for performance reasons only.

{sctp_i_want_mapped_v4_addr, true|false}

Turns on|off automatic mapping of IPv4 addresses into IPv6 ones (if the socket address family is AF_INET6).

{sctp_maxseg, integer()}

Determines the maximum chunk size if message fragmentation is used. If 0, the chunk size is limited by the Path MTU only.

{sctp_primary_addr, #sctp_prim{}}
      assoc_id = assoc_id(),
      addr     = {IP, Port}
 IP = ip_address()
 Port = port_number()

For the association specified by assoc_id, {IP,Port} must be one of the peer addresses. This option determines that the specified address is treated by the local SCTP stack as the primary address of the peer.

{sctp_set_peer_primary_addr, #sctp_setpeerprim{}}
      assoc_id = assoc_id(),
      addr     = {IP, Port}
 IP = ip_address()
 Port = port_number()

When set, informs the peer to use {IP, Port} as the primary address of the local endpoint for the association specified by assoc_id.

{sctp_adaptation_layer, #sctp_setadaptation{}}
      adaptation_ind = integer()

When set, requests that the local endpoint uses the value specified by adaptation_ind as the Adaptation Indication parameter for establishing new associations. For details, see RFC 2960 and Sockets API Extenstions for SCTP.

{sctp_peer_addr_params, #sctp_paddrparams{}}
      assoc_id   = assoc_id(),
      address    = {IP, Port},
      hbinterval = integer(),
      pathmaxrxt = integer(),
      pathmtu    = integer(),
      sackdelay  = integer(),
      flags      = list()
IP = ip_address()
Port = port_number()

Determines various per-address parameters for the association specified by assoc_id and the peer address address (the SCTP protocol supports multi-homing, so more than one address can correspond to a specified association).


Heartbeat interval, in milliseconds


Maximum number of retransmissions before this address is considered unreachable (and an alternative address is selected)


Fixed Path MTU, if automatic discovery is disabled (see flags below)


Delay, in milliseconds, for SAC messages (if the delay is enabled, see flags below)


The following flags are available:

Enables heartbeat
Disables heartbeat
Initiates heartbeat immediately
Enables automatic Path MTU discovery
Disables automatic Path MTU discovery
Enables SAC delay
Disables SAC delay
{sctp_default_send_param, #sctp_sndrcvinfo{}}
      stream     = integer(),
      ssn        = integer(),
      flags      = list(),
      ppid       = integer(),
      context    = integer(),
      timetolive = integer(),
      tsn        = integer(),
      cumtsn     = integer(),
      assoc_id   = assoc_id()

#sctp_sndrcvinfo{} is used both in this socket option, and as ancillary data while sending or receiving SCTP messages. When set as an option, it provides default values for subsequent send calls on the association specified by assoc_id.

assoc_id = 0 (default) indicates the whole endpoint.

The following fields typically must be specified by the sender:


Stream number (0-base) within the association to send the messages through;


The following flags are recognised:

The message is to be sent unordered
The address specified in send overwrites the primary peer address
Aborts the current association without flushing any unsent data
Gracefully shuts down the current association, with flushing of unsent data

Other fields are rarely used. For complete information, see RFC 2960 and Sockets API Extensions for SCTP.

{sctp_events, #sctp_event_subscribe{}}
        data_io_event          = true | false,
        association_event      = true | false,
        address_event          = true | false,
        send_failure_event     = true | false,
        peer_error_event       = true | false,
        shutdown_event         = true | false,
        partial_delivery_event = true | false,
        adaptation_layer_event = true | false

This option determines which SCTP Events are to be received (through recv/*) along with the data. The only exception is data_io_event, which enables or disables receiving of #sctp_sndrcvinfo{} ancillary data, not events. By default, all flags except adaptation_layer_event are enabled, although sctp_data_io_event and association_event are used by the driver itself and not exported to the user level.

{sctp_delayed_ack_time, #sctp_assoc_value{}}
      assoc_id    = assoc_id(),
      assoc_value = integer()

Rarely used. Determines the ACK time (specified by assoc_value, in milliseconds) for the specified association or the whole endpoint if assoc_value = 0 (default).

{sctp_status, #sctp_status{}}
      assoc_id            = assoc_id(),
      state               = atom(),
      rwnd                = integer(),
      unackdata           = integer(),
      penddata            = integer(),
      instrms             = integer(),
      outstrms            = integer(),
      fragmentation_point = integer(),
      primary             = #sctp_paddrinfo{}

This option is read-only. It determines the status of the SCTP association specified by assoc_id. The following are the possible values of state (the state designations are mostly self-explanatory):

Default. Means that no other state is active.

Semantics of the other fields:

Current receiver window size of the association
Number of unacked data chunks
Number of data chunks pending receipt
Number of inbound streams
Number of outbound streams
Message size at which SCTP fragmentation occurs
Information on the current primary peer address (see below for the format of #sctp_paddrinfo{})
{sctp_get_peer_addr_info, #sctp_paddrinfo{}}
      assoc_id  = assoc_id(),
      address   = {IP, Port},
      state     = inactive | active | unconfirmed,
      cwnd      = integer(),
      srtt      = integer(),
      rto       = integer(),
      mtu       = integer()
IP = ip_address()
Port = port_number()

This option is read-only. It determines the parameters specific to the peer address specified by address within the association specified by assoc_id. Field address fmust be set by the caller; all other fields are filled in on return. If assoc_id = 0 (default), the address is automatically translated into the corresponding association ID. This option is rarely used. For the semantics of all fields, see RFC 2960 and Sockets API Extensions for SCTP.

Example of an Erlang SCTP server that receives SCTP messages and prints them on the standard output:



server() ->
    server(any, 2006).

server([Host,Port]) when is_list(Host), is_list(Port) ->
    {ok, #hostent{h_addr_list = [IP|_]}} = inet:gethostbyname(Host),
    io:format("~w -> ~w~n", [Host, IP]),
    server([IP, list_to_integer(Port)]).

server(IP, Port) when is_tuple(IP) orelse IP == any orelse IP == loopback,
                      is_integer(Port) ->
    {ok,S} = gen_sctp:open(Port, [{recbuf,65536}, {ip,IP}]),
    io:format("Listening on ~w:~w. ~w~n", [IP,Port,S]),
    ok     = gen_sctp:listen(S, true),

server_loop(S) ->
    case gen_sctp:recv(S) of
    {error, Error} ->
        io:format("SCTP RECV ERROR: ~p~n", [Error]);
    Data ->
        io:format("Received: ~p~n", [Data])

Example of an Erlang SCTP client interacting with the above server. Notice that in this example the client creates an association with the server with 5 outbound streams. Therefore, sending of "Test 0" over stream 0 succeeds, but sending of "Test 5" over stream 5 fails. The client then aborts the association, which results in that the corresponding event is received on the server side.


-export([client/0, client/1, client/2]).

client() ->

client([Host]) ->
    client(Host, 2006);
client([Host, Port]) when is_list(Host), is_list(Port) ->

client(Host, Port) when is_integer(Port) ->
    {ok,S}     = gen_sctp:open(),
    {ok,Assoc} = gen_sctp:connect
        (S, Host, Port, [{sctp_initmsg,#sctp_initmsg{num_ostreams=5}}]),
    io:format("Connection Successful, Assoc=~p~n", [Assoc]),

    io:write(gen_sctp:send(S, Assoc, 0, <<"Test 0">>)),
    io:write(gen_sctp:send(S, Assoc, 5, <<"Test 5">>)),
    io:write(gen_sctp:abort(S, Assoc)),


A simple Erlang SCTP client that uses the connect_init API:



client(Peer1, Port1, Peer2, Port2)
  when is_tuple(Peer1), is_integer(Port1), is_tuple(Peer2), is_integer(Port2) ->
    {ok,S}     = gen_sctp:open(),
    SctpInitMsgOpt = {sctp_initmsg,#sctp_initmsg{num_ostreams=5}},
    ActiveOpt = {active, true},
    Opts = [SctpInitMsgOpt, ActiveOpt],
    ok = gen_sctp:connect(S, Peer1, Port1, Opts),
    ok = gen_sctp:connect(S, Peer2, Port2, Opts),
    io:format("Connections initiated~n", []),
    client_loop(S, Peer1, Port1, undefined, Peer2, Port2, undefined).

client_loop(S, Peer1, Port1, AssocId1, Peer2, Port2, AssocId2) ->
        {sctp, S, Peer1, Port1, {_Anc, SAC}}
          when is_record(SAC, sctp_assoc_change), AssocId1 == undefined ->
            io:format("Association 1 connect result: ~p. AssocId: ~p~n",
            client_loop(S, Peer1, Port1, SAC#sctp_assoc_change.assoc_id,
                        Peer2, Port2, AssocId2);

        {sctp, S, Peer2, Port2, {_Anc, SAC}}
          when is_record(SAC, sctp_assoc_change), AssocId2 == undefined ->
            io:format("Association 2 connect result: ~p. AssocId: ~p~n",
                      [SAC#sctp_assoc_change.state, SAC#sctp_assoc_change.assoc_id]),
            client_loop(S, Peer1, Port1, AssocId1, Peer2, Port2,

        {sctp, S, Peer1, Port1, Data} ->
            io:format("Association 1: received ~p~n", [Data]),
            client_loop(S, Peer1, Port1, AssocId1,
                        Peer2, Port2, AssocId2);

        {sctp, S, Peer2, Port2, Data} ->
            io:format("Association 2: received ~p~n", [Data]),
            client_loop(S, Peer1, Port1, AssocId1,
                        Peer2, Port2, AssocId2);

        Other ->
            io:format("Other ~p~n", [Other]),
            client_loop(S, Peer1, Port1, AssocId1,
                        Peer2, Port2, AssocId2)

    after 5000 ->

gen_tcp(3), gen_udp(3), inet(3), RFC 2960 (Stream Control Transmission Protocol), Sockets API Extensions for SCTP