<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE chapter SYSTEM "chapter.dtd">
<chapter>
<header>
<copyright>
<year>2000</year><year>2018</year>
<holder>Ericsson AB. All Rights Reserved.</holder>
</copyright>
<legalnotice>
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
</legalnotice>
<title>How to Implement an Alternative Carrier for the Erlang Distribution
</title>
<prepared>Patrik Nyblom</prepared>
<responsible></responsible>
<docno></docno>
<approved></approved>
<checked></checked>
<date>2000-10-17</date>
<rev>PA2</rev>
<file>alt_dist.xml</file>
</header>
<p>This section describes how to implement an alternative carrier
protocol for the Erlang distribution. The distribution is normally
carried by TCP/IP. Here is explained a method for replacing TCP/IP
with another protocol.</p>
<p>The section is a step-by-step explanation of the
<c><![CDATA[uds_dist]]></c> example application (in the
Kernel application <c><![CDATA[examples]]></c> directory). The
<c><![CDATA[uds_dist]]></c> application implements distribution over Unix
domain sockets and is written for the Sun Solaris 2 operating environment.
The mechanisms are however general and apply to any operating system Erlang
runs on. The reason the C code is not made portable, is simply
readability.</p>
<section>
<title>Introduction</title>
<p>To implement a new carrier for the Erlang distribution, the main
steps are as follows.</p>
<note><p>
As of ERTS version 10.0 support for distribution controller
processes has been introduced. That is, the traffic over a
distribution channel can be managed by a process instead of
only by a port. This makes it possible to implement large
parts of the logic in Erlang code, and you perhaps do not
even need a new driver for the protocol. One example could
be Erlang distribution over UDP using <c>gen_udp</c> (your
Erlang code will of course have to take care of retransmissions,
etc in this example). That is, depending on what you want
to do you perhaps do not need to implement a driver at all
and can then skip the driver related sections below.
The <c>gen_tcp_dist</c> example described in the
<seealso marker="#distribution_module">Distribution
Module</seealso> section utilize distribution controller
processes and can be worth having a look at if you want to
use distribution controller processes.
</p></note>
<section>
<title>Writing an Erlang Driver</title>
<p>First, the protocol must be available to the Erlang machine, which
involves writing an Erlang driver. A port program cannot be used,
an Erlang driver is required. Erlang drivers can be:</p>
<list type="bulleted">
<item>
<p>Statically linked to the emulator, which can be an alternative
when using the open source distribution of Erlang, or</p>
</item>
<item>
<p>Dynamically loaded into the Erlang machines address space,
which is the only alternative if a precompiled version of
Erlang is to be used</p>
</item>
</list>
<p>Writing an Erlang driver is not easy. The driver is written
as some callback functions called by the Erlang emulator when
data is sent to the driver, or the driver has any data available on
a file descriptor. As the driver callback routines execute in the main
thread of the Erlang machine, the callback functions can perform
no blocking activity whatsoever. The callbacks are only to set up
file descriptors for waiting and/or read/write available data. All
I/O must be non-blocking. Driver callbacks are however executed
in sequence, why a global state can safely be updated within the
routines.</p>
</section>
<section>
<title>Writing an Erlang Interface for the Driver</title>
<p>When the driver is implemented, one would preferably write an
Erlang interface for the driver to be able to test the
functionality of the driver separately. This interface can then
be used by the distribution module, which will cover the details of
the protocol from the <c><![CDATA[net_kernel]]></c>.</p>
<p>The easiest path
is to mimic the <c><![CDATA[inet]]></c> and <c><![CDATA[inet_tcp]]></c>
interfaces, but not much
functionality in those modules needs to be implemented. In the
example application, only a few of the usual interfaces are
implemented, and they are much simplified.</p>
</section>
<section>
<title>Writing a Distribution Module</title>
<p>When the protocol is available to Erlang through a driver and an
Erlang interface module, a distribution module can be written.
The distribution module is a module with well-defined callbacks,
much like a <c><![CDATA[gen_server]]></c> (there is no compiler support
for checking the callbacks, though). This module implements:</p>
<list type="bulleted">
<item>The details of finding other nodes (that is, talking to
<c>epmd</c> or something similar)</item>
<item>Creating a listen port (or similar)</item>
<item>Connecting to other nodes</item>
<item>Performing the handshakes/cookie verification</item>
</list>
<p>There is however a utility module, <c><![CDATA[dist_util]]></c>, which
does most of the hard work of handling handshakes, cookies, timers,
and ticking. Using <c><![CDATA[dist_util]]></c> makes implementing a
distribution module much easier and that is done in
the example application.</p>
</section>
<section>
<title>Creating Boot Scripts</title>
<p>The last step is to create boot scripts to make the protocol
implementation available at boot time. The implementation can be
debugged by starting the distribution when all the system is
running, but in a real system the distribution is to start very
early, why a boot script and some command-line parameters are
necessary.</p>
<p>This step also implies that the Erlang code in the
interface and distribution modules is written in such a way that
it can be run in the startup phase. In particular, there can be no
calls to the <c><![CDATA[application]]></c> module or to any modules
not loaded at boot time. That is, only <c><![CDATA[Kernel]]></c>,
<c><![CDATA[STDLIB]]></c>, and the application itself can be used.</p>
</section>
</section>
<section>
<marker id="distribution_module"/>
<title>Distribution Module</title>
<p>
The distribution module expose an API that <c>net_kernel</c> call
in order to manage connections to other nodes. The module name
should have the suffix <c>_dist</c>.
</p>
<p>
The module needs to create some kind of listening entity (process
or port) and an acceptor process that accepts incoming connections
using the listening entity. For each connection, the module at least
needs to create one connection supervisor process, which also is
responsible for the handshake when setting up the connection, and
a distribution controller (process or port) responsible for
transport of data over the connection. The distribution controller
and the connection supervisor process should be linked together
so both of them are cleaned up when the connection is taken down.
</p>
<p>
Note that there need to be exactly one distribution controller
per connection. A process or port can only be distribution
controller for one connection. The registration as distribution
controller cannot be undone. It will stick until the distribution
controller terminates. The distribution controller should not
ignore exit signals. It is allowed to trap exits, but it should
then voluntarily terminate when an exit signal is received.
</p>
<p>
An example implementation of a distribution module can be found
in
<url href="gen_tcp_dist.erl">$ERL_TOP/lib/kernel/examples/gen_tcp_dist/src/gen_tcp_dist.erl</url>.
It implements the distribution over TCP/IP using the <c>gen_tcp</c>
API with distribution controllers implemented by processes. This
instead of using port distribution controllers as the ordinary TCP/IP
distribution uses.
</p>
<section>
<marker id="distribution_module_exported_callback_functions"/>
<title>Exported Callback Functions</title>
<p>
The following functions are mandatory:
</p>
<taglist>
<tag><marker id="listen"/><c>listen(Name) -></c><br/> <c>{ok, {Listen, Address, Creation}} | {error, Error} </c></tag>
<item>
<p>
<c>listen/1</c> is called once in order to listen for incoming
connection requests. The call is made when the distribution is brought
up. The argument <c>Name</c> is the part of the node name before
the <c>@</c> sign in the full node name. It can be either an atom or a
string.
</p>
<p>
The return value consists of a <c>Listen</c> handle (which is
later passed to the <seealso marker="#accept"><c>accept/1</c></seealso>
callback), <c>Address</c> which is a <c>#net_address{}</c> record
with information about the address for the node (the
<c>#net_address{}</c> record is defined in
<c>kernel/include/net_address.hrl</c>), and <c>Creation</c> which
(currently) is an integer <c>1</c>, <c>2</c>, or <c>3</c>.
</p>
<p>
If <seealso marker="erts:epmd"><c>epmd</c></seealso> is to be used
for node discovery, you typically want to use the (unfortunately
undocumented) <c>erl_epmd</c> module (part of the <c>kernel</c>
application) in order to register the listen port with <c>epmd</c>
and retrieve <c>Creation</c> to use.
</p>
</item>
<tag><marker id="accept"/><c>accept(Listen) -></c><br/> <c>AcceptorPid</c></tag>
<item>
<p>
<c>accept/1</c> should spawn a process that accepts connections. This
process should preferably execute on <c>max</c> priority. The process
identifier of this process should be returned.
</p>
<p>
The <c>Listen</c> argument will be the same as the <c>Listen</c> handle
part of the return value of the
<seealso marker="#listen"><c>listen/1</c></seealso> callback above.
<c>accept/1</c> is called only once when the distribution protocol is
started.
</p>
<p>
The caller of this function is a representative for <c>net_kernel</c>
(this may or may not be the process registered as <c>net_kernel</c>)
and is in this document identified as <c>Kernel</c>.
When a connection has been accepted by the acceptor process, it needs
to inform <c>Kernel</c> about the accepted connection. This is done by
passing a message on the form:
</p>
<code type="none"><![CDATA[Kernel ! {accept, AcceptorPid, DistController, Family, Proto}]]></code>
<p>
<c>DistController</c> is either the process or port identifier
of the distribution controller for the connection. The
distribution controller should be created by the acceptor
processes when a new connection is accepted. Its job is to
dispatch traffic on the connection.
</p>
<c>Kernel</c> responds with one of the following messages:
<taglist>
<tag><c>{Kernel, controller, SupervisorPid}</c></tag>
<item>
<p>
The request was accepted and <c>SupervisorPid</c> is the
process identifier of the connection supervisor process
(which is created in the
<seealso marker="#accept_connection"><c>accept_connection/5</c></seealso>
callback).
</p>
</item>
<tag><c>{Kernel, unsupported_protocol}</c></tag>
<item>
<p>
The request was rejected. This is a fatal error. The acceptor
process should terminate.
</p>
</item>
</taglist>
<p>
When an accept sequence has been completed the acceptor process
is expected to continue accepting further requests.
</p>
</item>
<tag><marker id="accept_connection"/><c>accept_connection(AcceptorPid, DistCtrl, MyNode, Allowed, SetupTime) -></c><br/> <c>ConnectionSupervisorPid</c></tag>
<item>
<p>
<c>accept_connection/5</c> should spawn a process that will
perform the Erlang distribution handshake for the connection.
If the handshake successfully completes it should continue to
function as a connection supervisor. This process
should preferably execute on <c>max</c> priority.
</p>
<p>The arguments:</p>
<taglist>
<tag><c>AcceptorPid</c></tag>
<item>
<p>
Process identifier of the process created by the
<seealso marker="#accept"><c>accept/1</c></seealso>
callback.
</p>
</item>
<tag><c>DistCtrl</c></tag>
<item>
<p>The identifier of the distribution controller identifier
created by the acceptor process. To be passed along to
<c>dist_util:handshake_other_started(HsData)</c>.
</p>
</item>
<tag><c>MyNode</c></tag>
<item>
<p>
Node name of this node. To be passed along to
<c>dist_util:handshake_other_started(HsData)</c>.
</p>
</item>
<tag><c>Allowed</c></tag>
<item>
<p>
To be passed along to
<c>dist_util:handshake_other_started(HsData)</c>.
</p>
</item>
<tag><c>SetupTime</c></tag>
<item>
<p>
Time used for creating a setup timer by a
call to <c>dist_util:start_timer(SetupTime)</c>.
The timer should be passed along to
<c>dist_util:handshake_other_started(HsData)</c>.
</p>
</item>
</taglist>
<p>
The created process should provide callbacks and other
information needed for the handshake in a
<seealso marker="#hs_data_record"><c>#hs_data{}</c></seealso>
record and call <c>dist_util:handshake_other_started(HsData)</c>
with this record.
</p>
<p>
<c>dist_util:handshake_other_started(HsData)</c> will perform
the handshake and if the handshake successfully completes this
process will then continue in a connection supervisor loop
as long as the connection is up.
</p>
</item>
<tag><marker id="setup"/><c>setup(Node, Type, MyNode, LongOrShortNames, SetupTime) -></c><br/> <c>ConnectionSupervisorPid</c></tag>
<item>
<p>
<c>setup/5</c> should spawn a process that connects to
<c>Node</c>. When connection has been established it should
perform the Erlang distribution handshake for the connection.
If the handshake successfully completes it should continue to
function as a connection supervisor. This process
should preferably execute on <c>max</c> priority.
</p>
<p>The arguments:</p>
<taglist>
<tag><c>Node</c></tag>
<item>
<p>
Node name of remote node. To be passed along to
<c>dist_util:handshake_we_started(HsData)</c>.
</p>
</item>
<tag><c>Type</c></tag>
<item>
<p>
Connection type. To be passed along to
<c>dist_util:handshake_we_started(HsData)</c>.
</p>
</item>
<tag><c>MyNode</c></tag>
<item>
<p>
Node name of this node. To be passed along to
<c>dist_util:handshake_we_started(HsData)</c>.
</p>
</item>
<tag><c>LongOrShortNames</c></tag>
<item>
<p>
Either the atom <c>longnames</c> or
the atom <c>shortnames</c> indicating
whether long or short names is used.
</p>
</item>
<tag><c>SetupTime</c></tag>
<item>
<p>
Time used for creating a setup timer by a
call to <c>dist_util:start_timer(SetupTime)</c>.
The timer should be passed along to
<c>dist_util:handshake_we_started(HsData)</c>.
</p>
</item>
</taglist>
<p>
The caller of this function is a representative for <c>net_kernel</c>
(this may or may not be the process registered as <c>net_kernel</c>)
and is in this document identified as <c>Kernel</c>.
</p>
<p>
This function should, besides spawning the connection supervisor,
also create a distribution controller. The distribution
controller is either a process or a port which is responsible
for dispatching traffic.
</p>
<p>
The created process should provide callbacks and other
information needed for the handshake in a
<seealso marker="#hs_data_record"><c>#hs_data{}</c></seealso>
record and call <c>dist_util:handshake_we_started(HsData)</c>
with this record.
</p>
<p>
<c>dist_util:handshake_we_started(HsData)</c> will perform
the handshake and the handshake successfully completes this
process will then continue in a connection supervisor loop
as long as the connection is up.
</p>
</item>
<tag><marker id="close"/><c>close(Listen) -></c><br/> <c>void()</c></tag>
<item><p>
Called in order to close the <c>Listen</c> handle
that originally was passed from the
<seealso marker="#listen"><c>listen/1</c></seealso> callback.
</p></item>
<tag><marker id="select"/><c>select(NodeName) -></c><br/> <c>boolean()</c></tag>
<item>
<p>Return <c>true</c> if the host name part
of the <c>NodeName</c> is valid for use
with this protocol; otherwise, <c>false</c>.
</p>
</item>
</taglist>
<p>
There are also two optional functions that may be
exported:
</p>
<taglist>
<tag><marker id="select"/><c>setopts(Listen, Opts) -></c><br/> <c>ok | {error, Error}</c></tag>
<item>
<p>
The argument <c>Listen</c> is the handle originally passed
from the
<seealso marker="#listen"><c>listen/1</c></seealso> callback.
The argument <c>Opts</c> is a list of options to set on future
connections.
</p>
</item>
<tag><marker id="select"/><c>getopts(Listen, Opts) -></c><br/> <c>{ok, OptionValues} | {error, Error}</c></tag>
<item>
<p>
The argument <c>Listen</c> is the handle originally passed
from the
<seealso marker="#listen"><c>listen/1</c></seealso> callback.
The argument <c>Opts</c> is a list of options to read for future
connections.
</p>
</item>
</taglist>
</section>
<section>
<marker id="hs_data_record"/>
<title>The #hs_data{} Record</title>
<p>
The <c>dist_util:handshake_we_started/1</c> and
<c>dist_util:handshake_other_started/1</c> functions
takes a <c>#hs_data{}</c> record as argument. There
are quite a lot of fields in this record that you
need to set. The record is defined in
<c>kernel/include/dist_util.hrl</c>. Not documented
fields should not be set, i.e., should be left as
<c>undefined</c>.
</p>
<p>
The following <c>#hs_data{}</c> record fields need
to be set unless otherwise stated:</p>
<taglist>
<tag><marker id="hs_data_kernel_pid"/><c>kernel_pid</c></tag>
<item>
<p>
Process identifier of the <c>Kernel</c> process. That is,
the process that called either
<seealso marker="#setup"><c>setup/5</c></seealso> or
<seealso marker="#accept_connection"><c>accept_connection/5</c></seealso>.
</p>
</item>
<tag><marker id="hs_data_other_node"/><c>other_node</c></tag>
<item>
<p>Name of the other node. This field is only
mandatory when this node initiates the connection.
That is, when connection is set up via
<seealso marker="#setup"><c>setup/5</c></seealso>.
</p>
</item>
<tag><marker id="hs_data_this_node"/><c>this_node</c></tag>
<item>
<p>
The node name of this node.
</p>
</item>
<tag><marker id="hs_data_socket"/><c>socket</c></tag>
<item>
<p>
The identifier of the distribution controller.
</p>
</item>
<tag><marker id="hs_data_timer"/><c>timer</c></tag>
<item>
<p>
The timer created using <c>dist_util:start_timer/1</c>.
</p>
</item>
<tag><marker id="hs_data_allowed"/><c>allowed</c></tag>
<item>
<p>Information passed as <c>Allowed</c> to
<c>accept_connection/5</c>. This field is only
mandatory when the remote node initiated the
connection. That is, when the connection is set
up via
<seealso marker="#accept_connection"><c>accept_connection/5</c></seealso>.
</p>
</item>
<tag><marker id="hs_data_f_send"/><c>f_send</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrlr, Data) -> ok | {error, Error}]]></code>
<p>
where <c>DistCtrlr</c> is the identifier of
the distribution controller and <c>Data</c>
is io data to pass to the other side.
</p>
<p>Only used during handshake phase.</p>
</item>
<tag><marker id="hs_data_f_recv"/><c>f_recv</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrlr, Length) -> {ok, Packet} | {error, Reason}]]></code>
<p>
where <c>DistCtrlr</c> is the identifier of the distribution
controller.
If <c>Length</c> is <c>0</c>, all available bytes should be
returned. If <c>Length > 0</c>, exactly <c>Length</c> bytes
should be returned, or an error; possibly discarding less
than <c>Length</c> bytes of data when the connection is
closed from the other side.
It is used for passive receive of data from the
other end.
</p>
<p>Only used during handshake phase.</p>
</item>
<tag><marker id="hs_data_f_setopts_pre_nodeup"/><c>f_setopts_pre_nodeup</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrlr) -> ok | {error, Error}]]></code>
<p>
where <c>DistCtrlr</c> is the identifier of
the distribution controller. Called just
before the distribution channel is taken up
for normal traffic.
</p>
<p>Only used during handshake phase.</p>
</item>
<tag><marker id="hs_data_f_setopts_post_nodeup"/><c>f_setopts_post_nodeup</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrlr) -> ok | {error, Error}]]></code>
<p>
where <c>DistCtrlr</c> is the identifier of
the distribution controller. Called just
after distribution channel has been taken
up for normal traffic.
</p>
<p>Only used during handshake phase.</p>
</item>
<tag><marker id="hs_data_f_getll"/><c>f_getll</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrlr) -> ID]]></code>
<p>
where <c>DistCtrlr</c> is the identifier of
the distribution controller and <c>ID</c> is
the identifier of the low level entity that
handles the connection (often <c>DistCtrlr</c>
itself).
</p>
<p>Only used during handshake phase.</p>
</item>
<tag><marker id="hs_data_f_address"/><c>f_address</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrlr, Node) -> NetAddress]]></code>
<p>
where <c>DistCtrlr</c> is the identifier of
the distribution controller, <c>Node</c>
is the node name of the node on the other end,
and <c>NetAddress</c> is a <c>#net_address{}</c>
record with information about the address
for the <c>Node</c> on the other end of the
connection. The <c>#net_address{}</c> record
is defined in
<c>kernel/include/net_address.hrl</c>.
</p>
<p>Only used during handshake phase.</p>
</item>
<tag><marker id="hs_data_mf_tick"/><c>mf_tick</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrlr) -> void()]]></code>
<p>
where <c>DistCtrlr</c> is the identifier
of the distribution controller. This
function should send information over
the connection that is not interpreted
by the other end while increasing the
statistics of received packets on the
other end. This is usually implemented by
sending an empty packet.
</p>
<note><p>
It is of vital importance that this operation
does not block the caller for a long time.
This since it is called from the connection
supervisor.
</p></note>
<p>Used when connection is up.</p>
</item>
<tag><marker id="hs_data_mf_getstat"/><c>mf_getstat</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrlr) -> {ok, Received, Sent, PendSend}]]></code>
<p>
where <c>DistCtrlr</c> is the identifier
of the distribution controller, <c>Received</c>
is received packets, <c>Sent</c> is
sent packets, and <c>PendSend</c> is
amount of data in queue to be sent
(typically in bytes, but <c>dist_util</c>
only checks whether the value is non-zero
to know there is data in queue)
or a <c>boolean()</c> indicating whether
there are packets in queue to be sent.
</p>
<note><p>
It is of vital importance that this operation
does not block the caller for a long time.
This since it is called from the connection
supervisor.
</p></note>
<p>Used when connection is up.</p>
</item>
<tag><marker id="hs_data_request_type"/><c>request_type</c></tag>
<item>
<p>
The request <c>Type</c> as passed to
<seealso marker="#setup"><c>setup/5</c></seealso>.
This is only mandatory when the connection has
been initiated by this node. That is, the connection
is set up via <c>setup/5</c>.
</p>
</item>
<tag><marker id="hs_data_mf_setopts"/><c>mf_setopts</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrl, Opts) -> ok | {error, Error}]]></code>
<p>
where <c>DistCtrlr</c> is the identifier
of the distribution controller and <c>Opts</c>
is a list of options to set on the connection.
</p>
<p>This function is optional. Used when connection is up.</p>
</item>
<tag><marker id="hs_data_mf_getopts"/><c>mf_getopts</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrl, Opts) -> {ok, OptionValues} | {error, Error}]]></code>
<p>
where <c>DistCtrlr</c> is the identifier
of the distribution controller and <c>Opts</c>
is a list of options to read for the connection.
</p>
<p>This function is optional. Used when connection is up.</p>
</item>
<tag><marker id="hs_data_f_handshake_complete"/><c>f_handshake_complete</c></tag>
<item>
<p>
A fun with the following signature:
</p>
<code type="none"><![CDATA[fun (DistCtrlr, Node, DHandle) -> void()]]></code>
<p>
where <c>DistCtrlr</c> is the identifier
of the distribution controller, <c>Node</c> is
the node name of the node connected at the other
end, and <c>DHandle</c> is a distribution handle
needed by a distribution controller process when
calling the following BIFs:
</p>
<list>
<item><p><seealso marker="erts:erlang#dist_ctrl_get_data/1"><c>erlang:dist_ctrl_get_data/1</c></seealso></p></item>
<item><p><seealso marker="erts:erlang#dist_ctrl_get_data_notification/1"><c>erlang:dist_ctrl_get_data_notification/1</c></seealso></p></item>
<item><p><seealso marker="erts:erlang#dist_ctrl_input_handler/2"><c>erlang:dist_ctrl_input_handler/2</c></seealso></p></item>
<item><p><seealso marker="erts:erlang#dist_ctrl_put_data/2"><c>erlang:dist_ctrl_put_data/2</c></seealso></p></item>
</list>
<p>
This function is called when the handshake has
completed and the distribution channel is up.
The distribution controller can begin dispatching
traffic over the channel. This function is optional.
</p>
<p>Only used during handshake phase.</p>
</item>
<tag><marker id="hs_data_add_flags"/><c>add_flags</c></tag>
<item>
<p>
<seealso marker="erl_dist_protocol#dflags">Distribution flags</seealso>
to add to the connection. Currently all (non obsolete) flags will
automatically be enabled.
</p>
<p>
This flag field is optional.
</p>
</item>
<tag><marker id="hs_data_reject_flags"/><c>reject_flags</c></tag>
<item>
<p>
<seealso marker="erl_dist_protocol#dflags">Distribution flags</seealso>
to reject. Currently the following distribution flags can be rejected:
</p>
<taglist>
<tag><c>DFLAG_DIST_HDR_ATOM_CACHE</c></tag>
<item>Do not use atom cache over this connection.</item>
</taglist>
<p>Use function <c>dist_util:strict_order_flags/0</c> to get all flags
for features that require strict order delivery.</p>
<p>
This flag field is optional.
</p>
</item>
<tag><marker id="hs_data_require_flags"/><c>require_flags</c></tag>
<item>
<p>
Require these <seealso marker="erl_dist_protocol#dflags">distribution
flags</seealso> to be used. The connection will be aborted during the
handshake if the other end does not use them.
</p>
<p>
This flag field is optional.
</p>
</item>
</taglist>
</section>
<section>
<marker id="distribution_data_delivery"/>
<title>Distribution Data Delivery</title>
<p>
When using the default configuration, the data to pass
over a connection needs to be delivered as is
to the node on the receiving end in the <em>exact same
order</em>, with no loss of data what so ever, as sent
from the sending node.
</p>
<p>
The data delivery order can be relaxed by disabling
features that require strict ordering. This is done by
passing the
<seealso marker="erl_dist_protocol#dflags">distribution flags</seealso>
returned by <c>dist_util:strict_order_flags/0</c> in the
<seealso marker="alt_dist#hs_data_reject_flags"><c>reject_flags</c></seealso>
field of the <seealso marker="#hs_data_record"><c>#hs_data{}</c></seealso>
record used when setting up the connection. When relaxed
ordering is used, only the order of signals with the same
sender/receiver pair has to be preserved.
However, note that disabling the features that require
strict ordering may have a negative impact on performance,
throughput, and/or latency.
</p>
</section>
<section>
<marker id="enable_your_distribution_module"/>
<title>Enable Your Distribution Module</title>
<p>For <c>net_kernel</c> to find out which distribution module to use,
the <c>erl</c> command-line argument <c>-proto_dist</c> is used. It
is followed by one or more distribution module names, with suffix
"_dist" removed. That is, <c>gen_tcp_dist</c> as a distribution module
is specified as <c>-proto_dist gen_tcp</c>.</p>
<p>If no <c>epmd</c> (TCP port mapper daemon) is used, also command-line
option <c>-no_epmd</c> is to be specified, which makes
Erlang skip the <c>epmd</c> startup, both as an OS process and as an
Erlang ditto.</p>
</section>
</section>
<section>
<title>The Driver</title>
<note>
<p>This section was written a long time ago. Most of it is still
valid, but some things have changed since then. Some updates have
been made to the documentation of the driver presented here,
but more can be done and is planned for the future.
The reader is encouraged to read the
<seealso marker="erl_driver"><c>erl_driver</c></seealso> and
<seealso marker="driver_entry"><c>driver_entry</c></seealso>
documentation also.</p>
</note>
<p>Although Erlang drivers in general can be beyond the scope of this
section, a brief introduction seems to be in place.</p>
<section>
<title>Drivers in General</title>
<p>An Erlang driver is a native code module written in C (or
assembler), which serves as an interface for some special operating
system service. This is a general mechanism that is used
throughout the Erlang emulator for all kinds of I/O. An Erlang
driver can be dynamically linked (or loaded) to the Erlang
emulator at runtime by using the <c><![CDATA[erl_ddll]]></c> Erlang
module. Some of the drivers in OTP are however statically linked
to the runtime system, but that is more an optimization than a
necessity.</p>
<p>The driver data types and the functions available to the driver
writer are defined in header file <c><![CDATA[erl_driver.h]]></c>
seated in Erlang's include directory. See the
<seealso marker="erts:erl_driver">erl_driver</seealso> documentation
for details of which functions are available.</p>
<p>When writing a driver to make a communications protocol available
to Erlang, one should know just about everything worth knowing
about that particular protocol. All operation must be
non-blocking and all possible situations are to be accounted for in
the driver. A non-stable driver will affect and/or crash the
whole Erlang runtime system.</p>
<p>The emulator calls the driver in the following situations:</p>
<list type="bulleted">
<item>
<p>When the driver is loaded. This callback must have a special
name and inform the emulator of what callbacks are to be used
by returning a pointer to a <c><![CDATA[ErlDrvEntry]]></c> struct,
which is to be properly filled in (see below).</p>
</item>
<item>
<p>When a port to the driver is opened (by a
<c><![CDATA[open_port]]></c> call from Erlang). This routine is to
set up internal data structures and return an opaque data entity of
the type <c><![CDATA[ErlDrvData]]></c>, which is a data type large
enough to hold a pointer.
The pointer returned by this function is the first
argument to all other callbacks concerning this particular
port. It is usually called the port handle. The emulator only
stores the handle and does never try to interpret it, why it can
be virtually anything (anything not larger than a pointer
that is) and can point to anything if it is a pointer. Usually
this pointer refers to a structure holding information about
the particular port, as it does in the example.</p>
</item>
<item>
<p>When an Erlang process sends data to the port. The data
arrives as a buffer of bytes, the interpretation is not defined,
but is up to the implementor. This callback returns nothing to the
caller, answers are sent to the caller as messages (using a
routine called <c><![CDATA[driver_output]]></c> available to all
drivers). There is also a way to talk in a synchronous way to
drivers, described below. There can be an additional callback
function for handling data that is fragmented (sent in a deep
io-list). That interface gets the data in a form suitable for
Unix <c><![CDATA[writev]]></c> rather than in a single buffer.
There is no need for a distribution driver to implement such a
callback, so we will not.</p>
</item>
<item>
<p>When a file descriptor is signaled for input. This callback
is called when the emulator detects input on a file descriptor
that the driver has marked for monitoring by using the interface
<c><![CDATA[driver_select]]></c>. The mechanism of driver select
makes it possible to read non-blocking from file descriptors by
calling <c><![CDATA[driver_select]]></c> when reading is needed, and
then do the reading in this callback (when reading is possible).
The typical scenario is that <c><![CDATA[driver_select]]></c> is
called when an Erlang process orders a read operation, and that
this routine sends the answer when data is available on the file
descriptor.</p>
</item>
<item>
<p>When a file descriptor is signaled for output. This callback
is called in a similar way as the previous, but when writing to a
file descriptor is possible. The usual scenario is that Erlang
orders writing on a file descriptor and that the driver calls
<c><![CDATA[driver_select]]></c>. When the descriptor is ready for
output, this callback is called and the driver can try to send the
output. Queuing can be involved in such operations, and there are
convenient queue routines available to the driver writer to use.</p>
</item>
<item>
<p>When a port is closed, either by an Erlang process or by the
driver calling one of the <c><![CDATA[driver_failure_XXX]]></c>
routines. This routine is to clean up everything connected to one
particular port. When other callbacks call a
<c><![CDATA[driver_failure_XXX]]></c> routine, this routine is
immediately called. The callback routine issuing the error can
make no more use of the data structures for the port, as this
routine surely has freed all associated data and closed all file
descriptors. If the queue utility available to driver writer is
used, this routine is however <em>not</em> called until the
queue is empty.</p>
</item>
<item>
<p>When an Erlang process calls
<seealso marker="erlang#port_control/3">
<c>erlang:port_control/3</c></seealso>,
which is a synchronous interface to drivers. The control interface
is used to set driver options, change states of ports, and so on.
This interface is used a lot in the example.</p>
</item>
<item>
<p>When a timer expires. The driver can set timers with the function
<c><![CDATA[driver_set_timer]]></c>. When such timers expire, a
specific callback function is called. No timers are used in
the example.</p>
</item>
<item>
<p>When the whole driver is unloaded. Every resource allocated
by the driver is to be freed.</p>
</item>
</list>
</section>
<section>
<title>The Data Structures of the Distribution Driver</title>
<p>The driver used for Erlang distribution is to implement a
reliable, order maintaining, variable length packet-oriented
protocol. All error correction, resending and such need to be
implemented in the driver or by the underlying communications
protocol. If the protocol is stream-oriented (as is the case with
both TCP/IP and our streamed Unix domain sockets), some mechanism
for packaging is needed. We will use the simple method of having a
header of four bytes containing the length of the package in a
big-endian 32-bit integer. As Unix domain sockets only can be used
between processes on the same machine, we do not need to
code the integer in some special endianess, but we will do it anyway
because in most situation you need to do it. Unix domain
sockets are reliable and order maintaining, so we do not need to
implement resends and such in the driver.</p>
<p>We start writing the example Unix domain sockets driver by
declaring prototypes and filling in a static <c>ErlDrvEntry</c>
structure:</p>
<code type="none"><![CDATA[
( 1) #include <stdio.h>
( 2) #include <stdlib.h>
( 3) #include <string.h>
( 4) #include <unistd.h>
( 5) #include <errno.h>
( 6) #include <sys/types.h>
( 7) #include <sys/stat.h>
( 8) #include <sys/socket.h>
( 9) #include <sys/un.h>
(10) #include <fcntl.h>
(11) #define HAVE_UIO_H
(12) #include "erl_driver.h"
(13) /*
(14) ** Interface routines
(15) */
(16) static ErlDrvData uds_start(ErlDrvPort port, char *buff);
(17) static void uds_stop(ErlDrvData handle);
(18) static void uds_command(ErlDrvData handle, char *buff, int bufflen);
(19) static void uds_input(ErlDrvData handle, ErlDrvEvent event);
(20) static void uds_output(ErlDrvData handle, ErlDrvEvent event);
(21) static void uds_finish(void);
(22) static int uds_control(ErlDrvData handle, unsigned int command,
(23) char* buf, int count, char** res, int res_size);
(24) /* The driver entry */
(25) static ErlDrvEntry uds_driver_entry = {
(26) NULL, /* init, N/A */
(27) uds_start, /* start, called when port is opened */
(28) uds_stop, /* stop, called when port is closed */
(29) uds_command, /* output, called when erlang has sent */
(30) uds_input, /* ready_input, called when input
(31) descriptor ready */
(32) uds_output, /* ready_output, called when output
(33) descriptor ready */
(34) "uds_drv", /* char *driver_name, the argument
(35) to open_port */
(36) uds_finish, /* finish, called when unloaded */
(37) NULL, /* void * that is not used (BC) */
(38) uds_control, /* control, port_control callback */
(39) NULL, /* timeout, called on timeouts */
(40) NULL, /* outputv, vector output interface */
(41) NULL, /* ready_async callback */
(42) NULL, /* flush callback */
(43) NULL, /* call callback */
(44) NULL, /* event callback */
(45) ERL_DRV_EXTENDED_MARKER, /* Extended driver interface marker */
(46) ERL_DRV_EXTENDED_MAJOR_VERSION, /* Major version number */
(47) ERL_DRV_EXTENDED_MINOR_VERSION, /* Minor version number */
(48) ERL_DRV_FLAG_SOFT_BUSY, /* Driver flags. Soft busy flag is
(49) required for distribution drivers */
(50) NULL, /* Reserved for internal use */
(51) NULL, /* process_exit callback */
(52) NULL /* stop_select callback */
(53) };]]></code>
<p>On line 1-10 the OS headers needed for the driver are included.
As this driver is written for Solaris, we know that the
header <c><![CDATA[uio.h]]></c> exists. So the preprocessor variable
<c><![CDATA[HAVE_UIO_H]]></c> can be defined before
<c><![CDATA[erl_driver.h]]></c> is included on line 12.
The definition of <c><![CDATA[HAVE_UIO_H]]></c> will make the
I/O vectors used in Erlang's driver queues to correspond to the
operating systems ditto, which is very convenient.</p>
<p>On line 16-23 the different callback functions are declared ("forward
declarations").</p>
<p>The driver structure is similar for statically linked-in
drivers and dynamically loaded. However, some of the fields
are to be left empty (that is, initialized to NULL) in the
different types of drivers. The first field (the <c><![CDATA[init]]></c>
function pointer) is always left blank in a dynamically loaded
driver, see line 26. <c>NULL</c> on line 37
is always to be there, the field is no longer used and is
retained for backward compatibility. No timers are used in this
driver, why no callback for timers is needed. The <c>outputv</c> field
(line 40) can be used to implement an interface similar to
Unix <c><![CDATA[writev]]></c> for output. The Erlang runtime
system could previously not use <c>outputv</c> for the
distribution, but it can as from ERTS 5.7.2.
As this driver was written before ERTS 5.7.2 it does
not use the <c>outputv</c> callback. Using the <c>outputv</c>
callback is preferred, as it reduces copying of data. (We
will however use scatter/gather I/O internally in the driver.)</p>
<p>As from ERTS 5.5.3 the driver interface was extended with
version control and the possibility to pass capability information.
Capability flags are present on line 48. As from ERTS 5.7.4 flag
<seealso marker="driver_entry#driver_flags">
<c>ERL_DRV_FLAG_SOFT_BUSY</c></seealso> is required for drivers that
are to be used by the distribution. The soft busy flag implies that the
driver can handle calls to the <c>output</c> and <c>outputv</c>
callbacks although it has marked itself as busy. This has always been a
requirement on drivers used by the distribution, but no capability
information has been available about this previously. For more
information. see <seealso marker="erl_driver#set_busy_port">
<c>erl_driver:set_busy_port()</c></seealso>).</p>
<p>This driver was written before the runtime system had SMP support.
The driver will still function in the runtime system with SMP support,
but performance will suffer from lock contention on the driver lock
used for the driver. This can be alleviated by reviewing and perhaps
rewriting the code so that each instance of the driver safely can
execute in parallel. When instances safely can execute in parallel, it
is safe to enable instance-specific locking on the driver. This is done
by passing <seealso marker="driver_entry#driver_flags">
<c>ERL_DRV_FLAG_USE_PORT_LOCKING</c></seealso> as a driver flag. This
is left as an exercise for the reader.</p>
<p>Thus, the defined callbacks are as follows:</p>
<taglist>
<tag><c>uds_start</c></tag>
<item>
<p>Must initiate data for a port. We do not create any sockets
here, only initialize data structures.</p>
</item>
<tag><c>uds_stop</c></tag>
<item>
<p>Called when a port is closed.</p>
</item>
<tag><c>uds_command</c></tag>
<item>
<p>Handles messages from Erlang. The
messages can either be plain data to be sent or more subtle
instructions to the driver. This function is here mostly for
data pumping.</p>
</item>
<tag><c>uds_input</c></tag>
<item>
<p>Called when there is something to read from a socket.</p>
</item>
<tag><c>uds_output</c></tag>
<item>
<p>Called when it is possible to write to a socket.</p>
</item>
<tag><c>uds_finish</c></tag>
<item>
<p>Called when the driver is unloaded. A distribution driver will
never be unloaded, but we include this for completeness. To be
able to clean up after oneself is always a good thing.</p>
</item>
<tag><c>uds_control</c></tag>
<item>
<p>The <seealso marker="erlang#port_control/3">
<c>erlang:port_control/3</c></seealso> callback, which is
used a lot in this implementation.</p>
</item>
</taglist>
<p>The ports implemented by this driver operate in two major modes,
named <c>command</c> and <c>data</c>. In <c>command</c> mode,
only passive reading and writing (like
<c>gen_tcp:recv</c>/<c>gen_tcp:send</c>) can be done. The port is in
this mode during the distribution handshake. When the connection is up,
the port is switched to <c>data</c> mode and all data is immediately
read and passed further to the Erlang emulator. In <c>data</c>
mode, no data arriving to <c>uds_command</c> is interpreted, only
packaged and sent out on the socket. The <c>uds_control</c> callback
does the switching between those two modes.</p>
<p>While <c><![CDATA[net_kernel]]></c> informs different subsystems
that the connection is coming up, the port is to accept data to send.
However, the port should not receive any data, to avoid that data
arrives from another node before every kernel subsystem is prepared
to handle it. A third mode, named <c>intermediate</c>, is used for this
intermediate stage.</p>
<p>An enum is defined for the different types of ports:</p>
<code type="none"><![CDATA[
( 1) typedef enum {
( 2) portTypeUnknown, /* An uninitialized port */
( 3) portTypeListener, /* A listening port/socket */
( 4) portTypeAcceptor, /* An intermediate stage when accepting
( 5) on a listen port */
( 6) portTypeConnector, /* An intermediate stage when connecting */
( 7) portTypeCommand, /* A connected open port in command mode */
( 8) portTypeIntermediate, /* A connected open port in special
( 9) half active mode */
(10) portTypeData /* A connected open port in data mode */
(11) } PortType; ]]></code>
<p>The different types are as follows:</p>
<taglist>
<tag><c>portTypeUnknown</c></tag>
<item>
<p>The type a port has when it is opened, but
not bound to any file descriptor.</p>
</item>
<tag><c>portTypeListener</c></tag>
<item>
<p>A port that is connected to a listen socket. This port does not
do much, no data pumping is done on this socket, but read data is
available when one is trying to do an accept on the port.</p>
</item>
<tag><c>portTypeAcceptor</c></tag>
<item>
<p>This port is to represent the result of an accept operation. It is
created when one wants to accept from a listen socket, and it is
converted to a <c>portTypeCommand</c> when the accept succeeds.</p>
</item>
<tag><c>portTypeConnector</c></tag>
<item>
<p>Very similar to <c>portTypeAcceptor</c>, an
intermediate stage between the request for a connect operation and
that the socket is connected to an accepting ditto in the
other end. When the sockets are connected, the port
switches type to <c>portTypeCommand</c>.</p>
</item>
<tag><c>portTypeCommand</c></tag>
<item>
<p>A connected socket (or accepted socket) in <c>command</c> mode
mentioned earlier.</p>
</item>
<tag><c>portTypeIntermediate</c></tag>
<item>
<p>The intermediate stage for a connected socket.
There is to be no processing of input for this socket.</p>
</item>
<tag><c>portTypeData</c></tag>
<item>
<p>The mode where data is pumped through the port and the
<c>uds_command</c> routine regards every call as a call where
sending is wanted. In this mode, all input available is read and
sent to Erlang when it arrives on the socket, much like in the
active mode of a <c><![CDATA[gen_tcp]]></c> socket.</p>
</item>
</taglist>
<p>We study the state that is needed for the ports. Notice
that not all fields are used for all types of ports. Some space
could be saved by using unions, but that would clutter the
code with multiple indirections, so here is used one struct for
all types of ports, for readability:</p>
<code type="none"><![CDATA[
( 1) typedef unsigned char Byte;
( 2) typedef unsigned int Word;
( 3) typedef struct uds_data {
( 4) int fd; /* File descriptor */
( 5) ErlDrvPort port; /* The port identifier */
( 6) int lockfd; /* The file descriptor for a lock file in
( 7) case of listen sockets */
( 8) Byte creation; /* The creation serial derived from the
( 9) lock file */
(10) PortType type; /* Type of port */
(11) char *name; /* Short name of socket for unlink */
(12) Word sent; /* Bytes sent */
(13) Word received; /* Bytes received */
(14) struct uds_data *partner; /* The partner in an accept/listen pair */
(15) struct uds_data *next; /* Next structure in list */
(16) /* The input buffer and its data */
(17) int buffer_size; /* The allocated size of the input buffer */
(18) int buffer_pos; /* Current position in input buffer */
(19) int header_pos; /* Where the current header is in the
(20) input buffer */
(21) Byte *buffer; /* The actual input buffer */
(22) } UdsData; ]]></code>
<p>This structure is used for all types of ports although some
fields are useless for some types. The least memory consuming
solution would be to arrange this structure as a union of
structures. However, the multiple indirections in the code to
access a field in such a structure would clutter the code too
much for an example.</p>
<p>The fields in the structure are as follows:</p>
<taglist>
<tag><c>fd</c></tag>
<item>
<p>The file descriptor of the socket associated with the port.</p>
</item>
<tag><c>port</c></tag>
<item>
<p>The port identifier for the port that this structure
corresponds to. It is needed for most <c><![CDATA[driver_XXX]]></c>
calls from the driver back to the emulator.</p>
</item>
<tag><c>lockfd</c></tag>
<item>
<p>If the socket is a listen socket, we use a separate
(regular) file for two purposes:</p>
<list type="bulleted">
<item>
<p>We want a locking mechanism that gives no race
conditions, to be sure if another Erlang
node uses the listen socket name we require or if the
file is only left there from a previous (crashed) session.</p>
</item>
<item>
<p>We store the <c>creation</c> serial number in the
file. The <c>creation</c> is a number that is to
change between different instances of different Erlang
emulators with the same name, so that process
identifiers from one emulator do not become valid when sent
to a new emulator with the same distribution name. The
creation can be from 0 through 3 (two bits) and is stored
in every process identifier sent to another node.</p>
<p>In a system with TCP-based distribution, this data is
kept in the <em>Erlang port mapper daemon</em>
(<c><![CDATA[epmd]]></c>), which is contacted when a distributed
node starts. The lock file and a convention for the UDS
listen socket's name remove the need for
<c><![CDATA[epmd]]></c> when using this distribution module. UDS
is always restricted to one host, why avoiding a port
mapper is easy.</p>
</item>
</list>
</item>
<tag><c>creation</c></tag>
<item>
<p>The creation number for a listen socket, which is
calculated as (the value found in the lock-file + 1) rem 4.
This creation value is also written back into the
lock file, so that the next invocation of the emulator
finds our value in the file.</p>
</item>
<tag><c>type</c></tag>
<item>
<p>The current type/state of the port, which can be one
of the values declared above.</p>
</item>
<tag><c>name</c></tag>
<item>
<p>The name of the socket file (the path prefix removed),
which allows for deletion (<c><![CDATA[unlink]]></c>) when the
socket is closed.</p>
</item>
<tag><c>sent</c></tag>
<item>
<p>How many bytes that have been sent over the
socket. This can wrap, but that is no problem for the
distribution, as the Erlang distribution is only interested in
if this value has changed. (The Erlang
<c>net_kernel</c> <c>ticker</c> uses this value by calling the
driver to fetch it, which is done through the
<seealso marker="erlang#port_control/3">
<c>erlang:port_control/3</c></seealso> routine.)</p>
</item>
<tag><c>received</c></tag>
<item>
<p>How many bytes that are read (received) from the
socket, used in similar ways as <c><![CDATA[sent]]></c>.</p>
</item>
<tag><c>partner</c></tag>
<item>
<p>A pointer to another port structure, which is
either the listen port from which this port is accepting a
connection or conversely. The "partner relation"
is always bidirectional.</p>
</item>
<tag><c>next</c></tag>
<item>
<p>Pointer to next structure in a linked list of all
port structures. This list is used when accepting
connections and when the driver is unloaded.</p>
</item>
<tag><c>buffer_size</c>, <c>buffer_pos</c>, <c>header_pos</c>,
<c>buffer</c></tag>
<item>
<p>Data for input buffering. For details about the input buffering,
see the source code in directory <c>kernel/examples</c>. That
certainly goes beyond the scope of this section.</p>
</item>
</taglist>
</section>
<section>
<title>Selected Parts of the Distribution Driver Implementation</title>
<p>The implemenation of the distribution driver is not completely
covered here, details about buffering and other things
unrelated to driver writing are not explained. Likewise are
some peculiarities of the UDS protocol not explained in
detail. The chosen protocol is not important.</p>
<p>Prototypes for the driver callback routines can be found in
the <c><![CDATA[erl_driver.h]]></c> header file.</p>
<p>The driver initialization routine is (usually) declared with a
macro to make the driver easier to port between different
operating systems (and flavors of systems). This is the only
routine that must have a well-defined name. All other
callbacks are reached through the driver structure. The macro
to use is named <c><![CDATA[DRIVER_INIT]]></c> and takes the driver name
as parameter:</p>
<code type="none"><![CDATA[
(1) /* Beginning of linked list of ports */
(2) static UdsData *first_data;
(3) DRIVER_INIT(uds_drv)
(4) {
(5) first_data = NULL;
(6) return &uds_driver_entry;
(7) } ]]></code>
<p>The routine initializes the single global data structure and
returns a pointer to the driver entry. The routine is called
when <c><![CDATA[erl_ddll:load_driver]]></c> is called from Erlang.</p>
<p>The <c><![CDATA[uds_start]]></c> routine is called when a port is
opened from Erlang. In this case, we only allocate a structure and
initialize it. Creating the actual socket is left to the
<c><![CDATA[uds_command]]></c> routine.</p>
<code type="none"><![CDATA[
( 1) static ErlDrvData uds_start(ErlDrvPort port, char *buff)
( 2) {
( 3) UdsData *ud;
( 4)
( 5) ud = ALLOC(sizeof(UdsData));
( 6) ud->fd = -1;
( 7) ud->lockfd = -1;
( 8) ud->creation = 0;
( 9) ud->port = port;
(10) ud->type = portTypeUnknown;
(11) ud->name = NULL;
(12) ud->buffer_size = 0;
(13) ud->buffer_pos = 0;
(14) ud->header_pos = 0;
(15) ud->buffer = NULL;
(16) ud->sent = 0;
(17) ud->received = 0;
(18) ud->partner = NULL;
(19) ud->next = first_data;
(20) first_data = ud;
(21)
(22) return((ErlDrvData) ud);
(23) } ]]></code>
<p>Every data item is initialized, so that no problems arise
when a newly created port is closed (without there being any
corresponding socket). This routine is called when
<c><![CDATA[open_port({spawn, "uds_drv"},[])]]></c> is called from
Erlang.</p>
<p>The <c><![CDATA[uds_command]]></c> routine is the routine called when
an Erlang process sends data to the port. This routine handles all
asynchronous commands when the port is in <c>command</c> mode and
the sending of all data when the port is in <c>data</c> mode:</p>
<code type="none"><![CDATA[
( 1) static void uds_command(ErlDrvData handle, char *buff, int bufflen)
( 2) {
( 3) UdsData *ud = (UdsData *) handle;
( 4) if (ud->type == portTypeData || ud->type == portTypeIntermediate) {
( 5) DEBUGF(("Passive do_send %d",bufflen));
( 6) do_send(ud, buff + 1, bufflen - 1); /* XXX */
( 7) return;
( 8) }
( 9) if (bufflen == 0) {
(10) return;
(11) }
(12) switch (*buff) {
(13) case 'L':
(14) if (ud->type != portTypeUnknown) {
(15) driver_failure_posix(ud->port, ENOTSUP);
(16) return;
(17) }
(18) uds_command_listen(ud,buff,bufflen);
(19) return;
(20) case 'A':
(21) if (ud->type != portTypeUnknown) {
(22) driver_failure_posix(ud->port, ENOTSUP);
(23) return;
(24) }
(25) uds_command_accept(ud,buff,bufflen);
(26) return;
(27) case 'C':
(28) if (ud->type != portTypeUnknown) {
(29) driver_failure_posix(ud->port, ENOTSUP);
(30) return;
(31) }
(32) uds_command_connect(ud,buff,bufflen);
(33) return;
(34) case 'S':
(35) if (ud->type != portTypeCommand) {
(36) driver_failure_posix(ud->port, ENOTSUP);
(37) return;
(38) }
(39) do_send(ud, buff + 1, bufflen - 1);
(40) return;
(41) case 'R':
(42) if (ud->type != portTypeCommand) {
(43) driver_failure_posix(ud->port, ENOTSUP);
(44) return;
(45) }
(46) do_recv(ud);
(47) return;
(48) default:
(49) return;
(50) }
(51) } ]]></code>
<p>The command routine takes three parameters; the handle returned for
the port by <c><![CDATA[uds_start]]></c>, which is a pointer
to the internal port structure, the data buffer, and the length
of the data buffer. The buffer is the data sent from Erlang
(a list of bytes) converted to an C array (of bytes).</p>
<p>If Erlang sends, for example, the list <c><![CDATA[[$a,$b,$c]]]></c>
to the port, the <c><![CDATA[bufflen]]></c> variable is
<c><![CDATA[3]]></c> and the <c><![CDATA[buff]]></c> variable contains
<c><![CDATA[{'a','b','c'}]]></c> (no
<c>NULL</c> termination). Usually the first byte is used as an
opcode, which is the case in this driver too (at least when the
port is in <c>command</c> mode). The opcodes are defined as follows:</p>
<taglist>
<tag><c>'L'<socket name></c></tag>
<item>
<p>Creates and listens on socket with the specified name.</p>
</item>
<tag><c>'A'<listen number as 32-bit big-endian></c></tag>
<item>
<p>Accepts from the listen socket identified by the specified
identification number. The identification number is retrieved with
the <c>uds_control</c> routine.</p>
</item>
<tag><c>'C'<socket name></c></tag>
<item>
<p>Connects to the socket named <socket name>.</p>
</item>
<tag><c>'S'<data></c></tag>
<item>
<p>Sends the data <data> on the
connected/accepted socket (in <c>command</c> mode). The sending is
acknowledged when the data has left this process.</p>
</item>
<tag><c>'R'</c></tag>
<item>
<p>Receives one packet of data.</p>
</item>
</taglist>
<p>"One packet of data" in command <c>'R'</c> can be explained
as follows. This driver always sends data packaged with a 4
byte header containing a big-endian 32-bit integer that
represents the length of the data in the packet. There is no
need for different packet sizes or some kind of streamed
mode, as this driver is for the distribution only.
Why is the header word coded explicitly in big-endian when a UDS
socket is local to the host? It is good practice when writing a
distribution driver, as distribution in practice usually crosses
the host boundaries.</p>
<p>On line 4-8 is handled the case where the port is in <c>data</c> mode
or <c>intermediate</c> mode and the remaining routine handles the
different commands. The routine uses the
<c><![CDATA[driver_failure_posix()]]></c> routine to report errors
(see, for example, line 15). Notice that the failure routines make
a call to the <c><![CDATA[uds_stop]]></c> routine, which will
remove the internal port data. The handle (and the casted handle
<c><![CDATA[ud]]></c>) is therefore <em>invalid pointers</em> after a
<c><![CDATA[driver_failure]]></c> call and we should <em>return
immediately</em>. The runtime system will send exit signals to all
linked processes.</p>
<p>The <c>uds_input</c> routine is called when data is available on a
file descriptor previously passed to the
<c><![CDATA[driver_select]]></c> routine. This occurs typically when
a read command is issued and no data is available. The
<c><![CDATA[do_recv]]></c> routine is as follows:</p>
<code type="none"><![CDATA[
( 1) static void do_recv(UdsData *ud)
( 2) {
( 3) int res;
( 4) char *ibuf;
( 5) for(;;) {
( 6) if ((res = buffered_read_package(ud,&ibuf)) < 0) {
( 7) if (res == NORMAL_READ_FAILURE) {
( 8) driver_select(ud->port, (ErlDrvEvent) ud->fd, DO_READ, 1);
( 9) } else {
(10) driver_failure_eof(ud->port);
(11) }
(12) return;
(13) }
(14) /* Got a package */
(15) if (ud->type == portTypeCommand) {
(16) ibuf[-1] = 'R'; /* There is always room for a single byte
(17) opcode before the actual buffer
(18) (where the packet header was) */
(19) driver_output(ud->port,ibuf - 1, res + 1);
(20) driver_select(ud->port, (ErlDrvEvent) ud->fd, DO_READ,0);
(21) return;
(22) } else {
(23) ibuf[-1] = DIST_MAGIC_RECV_TAG; /* XXX */
(24) driver_output(ud->port,ibuf - 1, res + 1);
(25) driver_select(ud->port, (ErlDrvEvent) ud->fd, DO_READ,1);
(26) }
(27) }
(28) } ]]></code>
<p>The routine tries to read data until a packet is read or the
<c><![CDATA[buffered_read_package]]></c> routine returns a
<c><![CDATA[NORMAL_READ_FAILURE]]></c> (an internally defined constant
for the module, which means that the read operation resulted in an
<c><![CDATA[EWOULDBLOCK]]></c>). If the port is in <c>command</c> mode,
the reading stops when one package is read. If the port is in
<c>data</c> mode, the reading continues until the socket buffer is empty
(read failure). If no more data can be read and more is wanted (which
is always the case when the socket is in <c>data</c> mode),
<c>driver_select</c> is called to make the <c><![CDATA[uds_input]]></c>
callback be called when more data is available for reading.</p>
<p>When the port is in <c>data</c> mode, all data is sent to Erlang in a
format that suits the distribution. In fact, the raw data will
never reach any Erlang process, but will be
translated/interpreted by the emulator itself and then
delivered in the correct format to the correct processes. In
the current emulator version, received data is to be tagged
with a single byte of 100. That is what the macro
<c><![CDATA[DIST_MAGIC_RECV_TAG]]></c> is defined to. The tagging of
data in the distribution can be changed in the future.</p>
<p>The <c><![CDATA[uds_input]]></c> routine handles other input events
(like non-blocking <c><![CDATA[accept]]></c>), but most importantly
handle
data arriving at the socket by calling <c><![CDATA[do_recv]]></c>:</p>
<code type="none"><![CDATA[
( 1) static void uds_input(ErlDrvData handle, ErlDrvEvent event)
( 2) {
( 3) UdsData *ud = (UdsData *) handle;
( 4) if (ud->type == portTypeListener) {
( 5) UdsData *ad = ud->partner;
( 6) struct sockaddr_un peer;
( 7) int pl = sizeof(struct sockaddr_un);
( 8) int fd;
( 9) if ((fd = accept(ud->fd, (struct sockaddr *) &peer, &pl)) < 0) {
(10) if (errno != EWOULDBLOCK) {
(11) driver_failure_posix(ud->port, errno);
(12) return;
(13) }
(14) return;
(15) }
(16) SET_NONBLOCKING(fd);
(17) ad->fd = fd;
(18) ad->partner = NULL;
(19) ad->type = portTypeCommand;
(20) ud->partner = NULL;
(21) driver_select(ud->port, (ErlDrvEvent) ud->fd, DO_READ, 0);
(22) driver_output(ad->port, "Aok",3);
(23) return;
(24) }
(25) do_recv(ud);
(26) } ]]></code>
<p>The important line is the last line in the function: the
<c><![CDATA[do_read]]></c> routine is called to handle new input.
The remaining function handles input on a listen socket, which means
that it is to be possible to do an accept on the
socket, which is also recognized as a read event.</p>
<p>The output mechanisms are similar to the input.
The <c><![CDATA[do_send]]></c> routine is as follows:</p>
<code type="none"><![CDATA[
( 1) static void do_send(UdsData *ud, char *buff, int bufflen)
( 2) {
( 3) char header[4];
( 4) int written;
( 5) SysIOVec iov[2];
( 6) ErlIOVec eio;
( 7) ErlDrvBinary *binv[] = {NULL,NULL};
( 8) put_packet_length(header, bufflen);
( 9) iov[0].iov_base = (char *) header;
(10) iov[0].iov_len = 4;
(11) iov[1].iov_base = buff;
(12) iov[1].iov_len = bufflen;
(13) eio.iov = iov;
(14) eio.binv = binv;
(15) eio.vsize = 2;
(16) eio.size = bufflen + 4;
(17) written = 0;
(18) if (driver_sizeq(ud->port) == 0) {
(19) if ((written = writev(ud->fd, iov, 2)) == eio.size) {
(20) ud->sent += written;
(21) if (ud->type == portTypeCommand) {
(22) driver_output(ud->port, "Sok", 3);
(23) }
(24) return;
(25) } else if (written < 0) {
(26) if (errno != EWOULDBLOCK) {
(27) driver_failure_eof(ud->port);
(28) return;
(29) } else {
(30) written = 0;
(31) }
(32) } else {
(33) ud->sent += written;
(34) }
(35) /* Enqueue remaining */
(36) }
(37) driver_enqv(ud->port, &eio, written);
(38) send_out_queue(ud);
(39) } ]]></code>
<p>This driver uses the <c><![CDATA[writev]]></c> system call to send data
onto the socket. A combination of <c>writev</c> and the driver output
queues is very convenient. An <c>ErlIOVec</c> structure
contains a <c>SysIOVec</c> (which is equivalent to the
<c><![CDATA[struct iovec]]></c> structure defined in
<c><![CDATA[uio.h]]></c>. The
<c>ErlIOVec</c> also contains an array of <c>ErlDrvBinary</c>
pointers, of the same length as the number of buffers in the
I/O vector itself. One can use this to allocate the binaries
for the queue "manually" in the driver, but here
the binary array is filled with <c>NULL</c> values (line 7).
The runtime system then allocates its own buffers when
<c><![CDATA[driver_enqv]]></c> is called (line 37).</p>
<p>The routine builds an I/O vector containing the header bytes
and the buffer (the opcode has been removed and the buffer
length decreased by the output routine). If the queue is
empty, we write the data directly to the socket (or at
least try to). If any data is left, it is stored in the queue
and then we try to send the queue (line 38). An acknowledgement
is sent when the message is delivered completely (line 22). The
<c><![CDATA[send_out_queue]]></c> sends acknowledgements if the sending
is completed there. If the port is in <c>command</c> mode, the Erlang
code serializes the send operations so that only one packet
can be waiting for delivery at a time. Therefore the acknowledgement
can be sent whenever the queue is empty.</p>
<p>The <c><![CDATA[send_out_queue]]></c> routine is as follows:</p>
<code type="none"><![CDATA[
( 1) static int send_out_queue(UdsData *ud)
( 2) {
( 3) for(;;) {
( 4) int vlen;
( 5) SysIOVec *tmp = driver_peekq(ud->port, &vlen);
( 6) int wrote;
( 7) if (tmp == NULL) {
( 8) driver_select(ud->port, (ErlDrvEvent) ud->fd, DO_WRITE, 0);
( 9) if (ud->type == portTypeCommand) {
(10) driver_output(ud->port, "Sok", 3);
(11) }
(12) return 0;
(13) }
(14) if (vlen > IO_VECTOR_MAX) {
(15) vlen = IO_VECTOR_MAX;
(16) }
(17) if ((wrote = writev(ud->fd, tmp, vlen)) < 0) {
(18) if (errno == EWOULDBLOCK) {
(19) driver_select(ud->port, (ErlDrvEvent) ud->fd,
(20) DO_WRITE, 1);
(21) return 0;
(22) } else {
(23) driver_failure_eof(ud->port);
(24) return -1;
(25) }
(26) }
(27) driver_deq(ud->port, wrote);
(28) ud->sent += wrote;
(29) }
(30) } ]]></code>
<p>We simply pick out an I/O vector from the queue
(which is the whole queue as a <c>SysIOVec</c>). If the I/O
vector is too long (<c>IO_VECTOR_MAX</c> is defined to 16), the vector
length is decreased (line 15), otherwise the <c><![CDATA[writev]]></c>
call (line 17) fails. Writing is tried and anything written is dequeued
(line 27).
If the write fails with <c><![CDATA[EWOULDBLOCK]]></c> (notice that all
sockets are in non-blocking mode), <c><![CDATA[driver_select]]></c> is
called to make the <c><![CDATA[uds_output]]></c> routine be called when
there is space to write again.</p>
<p>We continue trying to write until the queue is empty or
the writing blocks.</p>
<p>The routine above is called from the <c><![CDATA[uds_output]]></c>
routine:</p>
<code type="none"><![CDATA[
( 1) static void uds_output(ErlDrvData handle, ErlDrvEvent event)
( 2) {
( 3) UdsData *ud = (UdsData *) handle;
( 4) if (ud->type == portTypeConnector) {
( 5) ud->type = portTypeCommand;
( 6) driver_select(ud->port, (ErlDrvEvent) ud->fd, DO_WRITE, 0);
( 7) driver_output(ud->port, "Cok",3);
( 8) return;
( 9) }
(10) send_out_queue(ud);
(11) } ]]></code>
<p>The routine is simple: it first handles the fact that the
output select will concern a socket in the business of
connecting (and the connecting blocked). If the socket is in
a connected state, it simply sends the output queue. This
routine is called when it is possible to write to a socket
where we have an output queue, so there is no question what to
do.</p>
<p>The driver implements a control interface, which is a
synchronous interface called when Erlang calls
<seealso marker="erlang#port_control/3">
<c>erlang:port_control/3</c></seealso>. Only this interface
can control the driver when it is in <c>data</c> mode. It can
be called with the following opcodes:</p>
<taglist>
<tag><c>'C'</c></tag>
<item>
<p>Sets port in <c>command</c> mode.</p>
</item>
<tag><c>'I'</c></tag>
<item>
<p>Sets port in <c>intermediate</c> mode.</p>
</item>
<tag><c>'D'</c></tag>
<item>
<p>Sets port in <c>data</c> mode.</p>
</item>
<tag><c>'N'</c></tag>
<item>
<p>Gets identification number for listen port. This
identification number is used in an accept command to the
driver. It is returned as a big-endian 32-bit integer, which
is the file identifier for the listen socket.</p>
</item>
<tag><c>'S'</c></tag>
<item>
<p>Gets statistics, which is the number of bytes received,
the number of bytes sent, and the number of bytes pending in
the output queue. This data is used when the distribution
checks that a connection is alive (ticking). The statistics
is returned as three 32-bit big-endian integers.</p>
</item>
<tag><c>'T'</c></tag>
<item>
<p>Sends a tick message, which is a packet of length 0.
Ticking is done when the port is in <c>data</c> mode, so the
command for sending data cannot be used (besides it ignores
zero length packages in <c>command</c> mode). This is used by the
ticker to send dummy data when no other traffic is present.</p>
<p><em>Note:</em> It is important that the interface for
sending ticks is not blocking. This implementation uses
<seealso marker="erlang#port_control/3">
<c>erlang:port_control/3</c></seealso>, which does not block the
caller. If <c>erlang:port_command</c> is used, use
<seealso marker="erlang#port_command/3">
<c>erlang:port_command/3</c></seealso> and pass <c>[force]</c> as
option list; otherwise the caller can be blocked indefinitely
on a busy port and prevent the system from taking down a
connection that is not functioning.</p>
</item>
<tag><c>'R'</c></tag>
<item>
<p>Gets creation number of a listen socket, which is used to
dig out the number stored in the lock file to differentiate
between invocations of Erlang nodes with the same name.</p>
</item>
</taglist>
<p>The control interface gets a buffer to return its value in,
but is free to allocate its own buffer if the provided one is
too small. The <c><![CDATA[uds_control]]></c> code is as follows:</p>
<code type="none"><![CDATA[
( 1) static int uds_control(ErlDrvData handle, unsigned int command,
( 2) char* buf, int count, char** res, int res_size)
( 3) {
( 4) /* Local macro to ensure large enough buffer. */
( 5) #define ENSURE(N) \
( 6) do { \
( 7) if (res_size < N) { \
( 8) *res = ALLOC(N); \
( 9) } \
(10) } while(0)
(11) UdsData *ud = (UdsData *) handle;
(12) switch (command) {
(13) case 'S':
(14) {
(15) ENSURE(13);
(16) **res = 0;
(17) put_packet_length((*res) + 1, ud->received);
(18) put_packet_length((*res) + 5, ud->sent);
(19) put_packet_length((*res) + 9, driver_sizeq(ud->port));
(20) return 13;
(21) }
(22) case 'C':
(23) if (ud->type < portTypeCommand) {
(24) return report_control_error(res, res_size, "einval");
(25) }
(26) ud->type = portTypeCommand;
(27) driver_select(ud->port, (ErlDrvEvent) ud->fd, DO_READ, 0);
(28) ENSURE(1);
(29) **res = 0;
(30) return 1;
(31) case 'I':
(32) if (ud->type < portTypeCommand) {
(33) return report_control_error(res, res_size, "einval");
(34) }
(35) ud->type = portTypeIntermediate;
(36) driver_select(ud->port, (ErlDrvEvent) ud->fd, DO_READ, 0);
(37) ENSURE(1);
(38) **res = 0;
(39) return 1;
(40) case 'D':
(41) if (ud->type < portTypeCommand) {
(42) return report_control_error(res, res_size, "einval");
(43) }
(44) ud->type = portTypeData;
(45) do_recv(ud);
(46) ENSURE(1);
(47) **res = 0;
(48) return 1;
(49) case 'N':
(50) if (ud->type != portTypeListener) {
(51) return report_control_error(res, res_size, "einval");
(52) }
(53) ENSURE(5);
(54) (*res)[0] = 0;
(55) put_packet_length((*res) + 1, ud->fd);
(56) return 5;
(57) case 'T': /* tick */
(58) if (ud->type != portTypeData) {
(59) return report_control_error(res, res_size, "einval");
(60) }
(61) do_send(ud,"",0);
(62) ENSURE(1);
(63) **res = 0;
(64) return 1;
(65) case 'R':
(66) if (ud->type != portTypeListener) {
(67) return report_control_error(res, res_size, "einval");
(68) }
(69) ENSURE(2);
(70) (*res)[0] = 0;
(71) (*res)[1] = ud->creation;
(72) return 2;
(73) default:
(74) return report_control_error(res, res_size, "einval");
(75) }
(76) #undef ENSURE
(77) } ]]></code>
<p>The macro <c><![CDATA[ENSURE]]></c> (line 5-10) is used to ensure that
the buffer is large enough for the answer. We switch on the command and
take actions. We always have read select active on a port in <c>data</c>
mode (achieved by calling <c><![CDATA[do_recv]]></c> on line 45), but
we turn off read selection in <c>intermediate</c> and <c>command</c>
modes (line 27 and 36).</p>
<p>The rest of the driver is more or less UDS-specific and not of
general interest.</p>
</section>
</section>
<section>
<title>Putting It All Together</title>
<p>To test the distribution, the <c><![CDATA[net_kernel:start/1]]></c>
function can be used. It is useful, as it starts the distribution on a
running system, where tracing/debugging can be performed.
The <c><![CDATA[net_kernel:start/1]]></c> routine takes a
list as its single argument. The list first element in the list is to be
the node name (without the "@hostname") as an atom. The second (and
last) element is to be one of the atoms <c><![CDATA[shortnames]]></c> or
<c><![CDATA[longnames]]></c>. In the example case,
<c><![CDATA[shortnames]]></c> is preferred.</p>
<p>For <c>net_kernel</c> to find out which distribution module to use,
command-line argument <c><![CDATA[-proto_dist]]></c> is used. It
is followed by one or more distribution module names, with suffix
"_dist" removed, that is, <c>uds_dist</c> as a distribution module
is specified as <c><![CDATA[-proto_dist uds]]></c>.</p>
<p>If no <c>epmd</c> (TCP port mapper daemon) is used, also command-line
option <c><![CDATA[-no_epmd]]></c> is to be specified, which makes
Erlang skip the <c>epmd</c> startup, both as an OS process and as an
Erlang ditto.</p>
<p>The path to the directory where the distribution modules reside
must be known at boot. This can be achieved either by
specifying <c><![CDATA[-pa <path>]]></c> on the command line or by
building a boot script containing the applications used for your
distribution protocol. (In the <c>uds_dist</c> protocol, only the
<c>uds_dist</c> application needs to be added to the script.)</p>
<p>The distribution starts at boot if all the above is
specified and an <c><![CDATA[-sname <name>]]></c> flag is present at the
command line.</p>
<p><em>Example 1:</em></p>
<pre>
$ <input>erl -pa $ERL_TOP/lib/kernel/examples/uds_dist/ebin -proto_dist uds -no_epmd</input>
Erlang (BEAM) emulator version 5.0
Eshell V5.0 (abort with ^G)
1> <input>net_kernel:start([bing,shortnames]).</input>
{ok,<0.30.0>}
(bing@hador)2></pre>
<p><em>Example 2:</em></p>
<pre>
$ <input>erl -pa $ERL_TOP/lib/kernel/examples/uds_dist/ebin -proto_dist uds \ </input>
<input> -no_epmd -sname bong</input>
Erlang (BEAM) emulator version 5.0
Eshell V5.0 (abort with ^G)
(bong@hador)1></pre>
<p>The <c>ERL_FLAGS</c> environment variable can be used to store the
complicated parameters in:</p>
<pre>
$ <input>ERL_FLAGS=-pa $ERL_TOP/lib/kernel/examples/uds_dist/ebin \ </input>
<input> -proto_dist uds -no_epmd</input>
$ <input>export ERL_FLAGS</input>
$ <input>erl -sname bang</input>
Erlang (BEAM) emulator version 5.0
Eshell V5.0 (abort with ^G)
(bang@hador)1></pre>
<p><c><![CDATA[ERL_FLAGS]]></c> should not include the node name.</p>
</section>
</chapter>
