path: root/nixos/doc/manual/configuration.xml

<chapter xmlns="http://docbook.org/ns/docbook"
         xmlns:xlink="http://www.w3.org/1999/xlink"
         xml:id="ch-configuration">

<title>Configuring NixOS</title>

<para>This chapter describes how to configure various aspects of a
NixOS machine through the configuration file
<filename>/etc/nixos/configuration.nix</filename>.  As described in
<xref linkend="sec-changing-config" />, changes to this file only take
effect after you run <command>nixos-rebuild</command>.</para>


<!--===============================================================-->

<section xml:id="sec-configuration-syntax"><title>Configuration syntax</title>

<section><title>The basics</title>

<para>The NixOS configuration file
<filename>/etc/nixos/configuration.nix</filename> is actually a
<emphasis>Nix expression</emphasis>, which is the Nix package
manager’s purely functional language for describing how to build
packages and configurations.  This means you have all the expressive
power of that language at your disposal, including the ability to
abstract over common patterns, which is very useful when managing
complex systems.  The syntax and semantics of the Nix language are
fully described in the <link
xlink:href="http://nixos.org/nix/manual/#chap-writing-nix-expressions">Nix
manual</link>, but here we give a short overview of the most important
constructs useful in NixOS configuration files.</para>

<para>The NixOS configuration file generally looks like this:

<programlisting>
{ config, pkgs, ... }:

{ <replaceable>option definitions</replaceable>
}
</programlisting>

The first line (<literal>{ config, pkgs, ... }:</literal>) denotes
that this is actually a function that takes at least the two arguments
 <varname>config</varname> and <varname>pkgs</varname>.  (These are
explained later.)  The function returns a <emphasis>set</emphasis> of
option definitions (<literal>{ <replaceable>...</replaceable> }</literal>).  These definitions have the
form <literal><replaceable>name</replaceable> =
<replaceable>value</replaceable></literal>, where
<replaceable>name</replaceable> is the name of an option and
<replaceable>value</replaceable> is its value.  For example,

<programlisting>
{ config, pkgs, ... }:

{ services.httpd.enable = true;
  services.httpd.adminAddr = "alice@example.org";
  services.httpd.documentRoot = "/webroot";
}
</programlisting>

defines a configuration with three option definitions that together
enable the Apache HTTP Server with <filename>/webroot</filename> as
the document root.</para>

<para>Sets can be nested, and in fact dots in option names are
shorthand for defining a set containing another set.  For instance,
<option>services.httpd.enable</option> defines a set named
<varname>services</varname> that contains a set named
<varname>httpd</varname>, which in turn contains an option definition
named <varname>enable</varname> with value <literal>true</literal>.
This means that the example above can also be written as:

<programlisting>
{ config, pkgs, ... }:

{ services = {
    httpd = {
      enable = true;
      adminAddr = "alice@example.org";
      documentRoot = "/webroot";
    };
  };
}
</programlisting>

which may be more convenient if you have lots of option definitions
that share the same prefix (such as
<literal>services.httpd</literal>).</para>

<para>NixOS checks your option definitions for correctness.  For
instance, if you try to define an option that doesn’t exist (that is,
doesn’t have a corresponding <emphasis>option declaration</emphasis>),
<command>nixos-rebuild</command> will give an error like:
<screen>
The option `services.httpd.enabl' defined in `/etc/nixos/configuration.nix' does not exist.
</screen>
Likewise, values in option definitions must have a correct type.  For
instance, <option>services.httpd.enable</option> must be a Boolean
(<literal>true</literal> or <literal>false</literal>).  Trying to give
it a value of another type, such as a string, will cause an error:
<screen>
The option value `services.httpd.enable' in `/etc/nixos/configuration.nix' is not a boolean.
</screen>

</para>

<para>Options have various types of values.  The most important are:

<variablelist>
  <varlistentry>
    <term>Strings</term>
    <listitem>
      <para>Strings are enclosed in double quotes, e.g.

<programlisting>
networking.hostName = "dexter";
</programlisting>

      Special characters can be escaped by prefixing them with a
      backslash (e.g. <literal>\"</literal>).</para>

      <para>Multi-line strings can be enclosed in <emphasis>double
      single quotes</emphasis>, e.g.

<programlisting>
networking.extraHosts =
  ''
    127.0.0.2 other-localhost
    10.0.0.1 server
  '';
</programlisting>

      The main difference is that preceding whitespace is
      automatically stripped from each line, and that characters like
      <literal>"</literal> and <literal>\</literal> are not special
      (making it more convenient for including things like shell
      code).</para>
    </listitem>
  </varlistentry>

  <varlistentry>
    <term>Booleans</term>
    <listitem>
      <para>These can be <literal>true</literal> or
      <literal>false</literal>, e.g.

<programlisting>
networking.firewall.enable = true;
networking.firewall.allowPing = false;
</programlisting>
      </para>
    </listitem>
  </varlistentry>

  <varlistentry>
    <term>Integers</term>
    <listitem>
      <para>For example,

<programlisting>
boot.kernel.sysctl."net.ipv4.tcp_keepalive_time" = 60;
</programlisting>

      (Note that here the attribute name
      <literal>net.ipv4.tcp_keepalive_time</literal> is enclosed in
      quotes to prevent it from being interpreted as a set named
      <literal>net</literal> containing a set named
      <literal>ipv4</literal>, and so on.  This is because it’s not a
      NixOS option but the literal name of a Linux kernel
      setting.)</para>
    </listitem>
  </varlistentry>

  <varlistentry>
    <term>Sets</term>
    <listitem>
      <para>Sets were introduced above.  They are name/value pairs
      enclosed in braces, as in the option definition

<programlisting>
fileSystems."/boot" =
  { device = "/dev/sda1";
    fsType = "ext4";
    options = "rw,data=ordered,relatime";
  };
</programlisting>
      </para>
    </listitem>
  </varlistentry>

  <varlistentry>
    <term>Lists</term>
    <listitem>
      <para>The important thing to note about lists is that list
      elements are separated by whitespace, like this:

<programlisting>
boot.kernelModules = [ "fuse" "kvm-intel" "coretemp" ];
</programlisting>

      List elements can be any other type, e.g. sets:

<programlisting>
swapDevices = [ { device = "/dev/disk/by-label/swap"; } ];
</programlisting>
      </para>
    </listitem>
  </varlistentry>

  <varlistentry>
    <term>Packages</term>
    <listitem>
      <para>Usually, the packages you need are already part of the Nix
      Packages collection, which is a set that can be accessed through
      the function argument <varname>pkgs</varname>.  Typical uses:

<programlisting>
environment.systemPackages =
  [ pkgs.thunderbird
    pkgs.emacs
  ];

postgresql.package = pkgs.postgresql90;
</programlisting>

      The latter option definition changes the default PostgreSQL
      package used by NixOS’s PostgreSQL service to 9.0.  For more
      information on packages, including how to add new ones, see
      <xref linkend="sec-custom-packages"/>.</para>
    </listitem>
  </varlistentry>

</variablelist>

</para>

</section>


<section><title>Abstractions</title>

<para>If you find yourself repeating yourself over and over, it’s time
to abstract.  Take, for instance, this Apache HTTP Server configuration:

<programlisting>
{
  services.httpd.virtualHosts =
    [ { hostName = "example.org";
        documentRoot = "/webroot";
        adminAddr = "alice@example.org";
        enableUserDir = true;
      }
      { hostName = "example.org";
        documentRoot = "/webroot";
        adminAddr = "alice@example.org";
        enableUserDir = true;
        enableSSL = true;
        sslServerCert = "/root/ssl-example-org.crt";
        sslServerKey = "/root/ssl-example-org.key";
      }
    ];
}
</programlisting>

It defines two virtual hosts with nearly identical configuration; the
only difference is that the second one has SSL enabled.  To prevent
this duplication, we can use a <literal>let</literal>:

<programlisting>
let
  exampleOrgCommon =
    { hostName = "example.org";
      documentRoot = "/webroot";
      adminAddr = "alice@example.org";
      enableUserDir = true;
    };
in
{
  services.httpd.virtualHosts =
    [ exampleOrgCommon
      (exampleOrgCommon // {
        enableSSL = true;
        sslServerCert = "/root/ssl-example-org.crt";
        sslServerKey = "/root/ssl-example-org.key";
      })
    ];
}
</programlisting>

The <literal>let exampleOrgCommon =
<replaceable>...</replaceable></literal> defines a variable named
<literal>exampleOrgCommon</literal>.  The <literal>//</literal>
operator merges two attribute sets, so the configuration of the second
virtual host is the set <literal>exampleOrgCommon</literal> extended
with the SSL options.</para>

<para>You can write a <literal>let</literal> wherever an expression is
allowed.  Thus, you also could have written:

<programlisting>
{
  services.httpd.virtualHosts =
    let exampleOrgCommon = <replaceable>...</replaceable>; in
    [ exampleOrgCommon
      (exampleOrgCommon // { <replaceable>...</replaceable> })
    ];
}
</programlisting>

but not <literal>{ let exampleOrgCommon =
<replaceable>...</replaceable>; in <replaceable>...</replaceable>;
}</literal> since attributes (as opposed to attribute values) are not
expressions.</para>

<para><emphasis>Functions</emphasis> provide another method of
abstraction.  For instance, suppose that we want to generate lots of
different virtual hosts, all with identical configuration except for
the host name.  This can be done as follows:

<programlisting>
{
  services.httpd.virtualHosts =
    let
      makeVirtualHost = name:
        { hostName = name;
          documentRoot = "/webroot";
          adminAddr = "alice@example.org";
        };
    in
      [ (makeVirtualHost "example.org")
        (makeVirtualHost "example.com")
        (makeVirtualHost "example.gov")
        (makeVirtualHost "example.nl")
      ];
}
</programlisting>

Here, <varname>makeVirtualHost</varname> is a function that takes a
single argument <literal>name</literal> and returns the configuration
for a virtual host.  That function is then called for several names to
produce the list of virtual host configurations.</para>

<para>We can further improve on this by using the function
<varname>map</varname>, which applies another function to every
element in a list:

<programlisting>
{
  services.httpd.virtualHosts =
    let
      makeVirtualHost = <replaceable>...</replaceable>;
    in map makeVirtualHost
      [ "example.org" "example.com" "example.gov" "example.nl" ];
}
</programlisting>

(The function <literal>map</literal> is called a
<emphasis>higher-order function</emphasis> because it takes another
function as an argument.)</para>

<para>What if you need more than one argument, for instance, if we
want to use a different <literal>documentRoot</literal> for each
virtual host?  Then we can make <varname>makeVirtualHost</varname> a
function that takes a <emphasis>set</emphasis> as its argument, like this:

<programlisting>
{
  services.httpd.virtualHosts =
    let
      makeVirtualHost = { name, root }:
        { hostName = name;
          documentRoot = root;
          adminAddr = "alice@example.org";
        };
    in map makeVirtualHost
      [ { name = "example.org"; root = "/sites/example.org"; }
        { name = "example.com"; root = "/sites/example.com"; }
        { name = "example.gov"; root = "/sites/example.gov"; }
        { name = "example.nl"; root = "/sites/example.nl"; }
      ];
}
</programlisting>

But in this case (where every root is a subdirectory of
<filename>/sites</filename> named after the virtual host), it would
have been shorter to define <varname>makeVirtualHost</varname> as
<programlisting>
makeVirtualHost = name:
  { hostName = name;
    documentRoot = "/sites/${name}";
    adminAddr = "alice@example.org";
  };
</programlisting>

Here, the construct
<literal>${<replaceable>...</replaceable>}</literal> allows the result
of an expression to be spliced into a string.</para>

</section>


<section><title>Modularity</title>

<para>The NixOS configuration mechanism is modular.  If your
<filename>configuration.nix</filename> becomes too big, you can split
it into multiple files.  Likewise, if you have multiple NixOS
configurations (e.g. for different computers) with some commonality,
you can move the common configuration into a shared file.</para>

<para>Modules have exactly the same syntax as
<filename>configuration.nix</filename>.  In fact,
<filename>configuration.nix</filename> is itself a module.  You can
use other modules by including them from
<filename>configuration.nix</filename>, e.g.:

<programlisting>
{ config, pkgs, ... }:

{ imports = [ ./vpn.nix ./kde.nix ];
  services.httpd.enable = true;
  environment.systemPackages = [ pkgs.emacs ];
  <replaceable>...</replaceable>
}
</programlisting>

Here, we include two modules from the same directory,
<filename>vpn.nix</filename> and <filename>kde.nix</filename>.  The
latter might look like this:

<programlisting>
{ config, pkgs, ... }:

{ services.xserver.enable = true;
  services.xserver.displayManager.kdm.enable = true;
  services.xserver.desktopManager.kde4.enable = true;
  environment.systemPackages = [ pkgs.kde4.kscreensaver ];
}
</programlisting>

Note that both <filename>configuration.nix</filename> and
<filename>kde.nix</filename> define the option
<option>environment.systemPackages</option>.  When multiple modules
define an option, NixOS will try to <emphasis>merge</emphasis> the
definitions.  In the case of
<option>environment.systemPackages</option>, that’s easy: the lists of
packages can simply be concatenated.  For other types of options, a
merge may not be possible: for instance, if two modules define
<option>services.httpd.adminAddr</option>,
<command>nixos-rebuild</command> will give an error:

<screen>
The unique option `services.httpd.adminAddr' is defined multiple times, in `/etc/nixos/httpd.nix' and `/etc/nixos/configuration.nix'.
</screen>

When that happens, it’s possible to force one definition take
precedence over the others:

<programlisting>
services.httpd.adminAddr = pkgs.lib.mkForce "bob@example.org";
</programlisting>

</para>

<para>When using multiple modules, you may need to access
configuration values defined in other modules.  This is what the
<varname>config</varname> function argument is for: it contains the
complete, merged system configuration.  That is,
<varname>config</varname> is the result of combining the
configurations returned by every module<footnote><para>If you’re
wondering how it’s possible that the (indirect)
<emphasis>result</emphasis> of a function is passed as an
<emphasis>input</emphasis> to that same function: that’s because Nix
is a “lazy” language — it only computes values when they are needed.
This works as long as no individual configuration value depends on
itself.</para></footnote>.  For example, here is a module that adds
some packages to <option>environment.systemPackages</option> only if
<option>services.xserver.enable</option> is set to
<literal>true</literal> somewhere else:

<programlisting>
{ config, pkgs, ... }:

{ environment.systemPackages =
    if config.services.xserver.enable then
      [ pkgs.firefox
        pkgs.thunderbird
      ]
    else
      [ ];
}
</programlisting>

</para>

<para>With multiple modules, it may not be obvious what the final
value of a configuration option is.  The command
<option>nixos-option</option> allows you to find out:

<screen>
$ nixos-option services.xserver.enable
true

$ nixos-option boot.kernelModules
[ "tun" "ipv6" "loop" <replaceable>...</replaceable> ]
</screen>

Interactive exploration of the configuration is possible using
<command
xlink:href="https://github.com/edolstra/nix-repl">nix-repl</command>,
a read-eval-print loop for Nix expressions.  It’s not installed by
default; run <literal>nix-env -i nix-repl</literal> to get it.  A
typical use:

<screen>
$ nix-repl '&lt;nixos>'

nix-repl> config.networking.hostName
"mandark"

nix-repl> map (x: x.hostName) config.services.httpd.virtualHosts
[ "example.org" "example.gov" ]
</screen>

</para>

</section>


<section><title>Syntax summary</title>

<para>Below is a summary of the most important syntactic constructs in
the Nix expression language.  It’s not complete.  In particular, there
are many other built-in functions.  See the <link
xlink:href="http://nixos.org/nix/manual/#chap-writing-nix-expressions">Nix
manual</link> for the rest.</para>

<informaltable frame='none'>
  <tgroup cols='2'>
    <colspec colname='c1' rowsep='1' colsep='1' />
    <colspec colname='c2' rowsep='1' />
    <thead>
      <row>
        <entry>Example</entry>
        <entry>Description</entry>
      </row>
    </thead>
    <tbody>

      <row>
        <entry namest="c1" nameend="c2"><emphasis>Basic values</emphasis></entry>
      </row>
      <row>
        <entry><literal>"Hello world"</literal></entry>
        <entry>A string</entry>
      </row>
      <row>
        <entry><literal>"${pkgs.bash}/bin/sh"</literal></entry>
        <entry>A string containing an expression (expands to <literal>"/nix/store/<replaceable>hash</replaceable>-bash-<replaceable>version</replaceable>/bin/sh"</literal>)</entry>
      </row>
      <row>
        <entry><literal>true</literal>, <literal>false</literal></entry>
        <entry>Booleans</entry>
      </row>
      <row>
        <entry><literal>123</literal></entry>
        <entry>An integer</entry>
      </row>
      <row>
        <entry><literal>./foo.png</literal></entry>
        <entry>A path (relative to the containing Nix expression)</entry>
      </row>

      <row>
        <entry namest="c1" nameend="c2"><emphasis>Compound values</emphasis></entry>
      </row>
      <row>
        <entry><literal>{ x = 1; y = 2; }</literal></entry>
        <entry>An set with attributes names <literal>x</literal> and <literal>y</literal></entry>
      </row>
      <row>
        <entry><literal>{ foo.bar = 1; }</literal></entry>
        <entry>A nested set, equivalent to <literal>{ foo = { bar = 1; }; }</literal></entry>
      </row>
      <row>
        <entry><literal>rec { x = "bla"; y = x + "bar"; }</literal></entry>
        <entry>A recursive set, equivalent to <literal>{ x = "foo"; y = "foobar"; }</literal></entry>
      </row>
      <row>
        <entry><literal>[ "foo" "bar" ]</literal></entry>
        <entry>A list with two elements</entry>
      </row>

      <row>
        <entry namest="c1" nameend="c2"><emphasis>Operators</emphasis></entry>
      </row>
      <row>
        <entry><literal>"foo" + "bar"</literal></entry>
        <entry>String concatenation</entry>
      </row>
      <row>
        <entry><literal>1 + 2</literal></entry>
        <entry>Integer addition</entry>
      </row>
      <row>
        <entry><literal>"foo" == "f" + "oo"</literal></entry>
        <entry>Equality test (evaluates to <literal>true</literal>)</entry>
      </row>
      <row>
        <entry><literal>"foo" != "bar"</literal></entry>
        <entry>Inequality test (evaluates to <literal>true</literal>)</entry>
      </row>
      <row>
        <entry><literal>!true</literal></entry>
        <entry>Boolean negation</entry>
      </row>
      <row>
        <entry><literal>{ x = 1; y = 2; }.x</literal></entry>
        <entry>Attribute selection (evaluates to <literal>1</literal>)</entry>
      </row>
      <row>
        <entry><literal>{ x = 1; y = 2; }.z or 3</literal></entry>
        <entry>Attribute selection with default (evaluates to <literal>3</literal>)</entry>
      </row>
      <row>
        <entry><literal>{ x = 1; y = 2; } // { z = 3; }</literal></entry>
        <entry>Merge two sets (attributes in the right-hand set taking precedence)</entry>
      </row>

      <row>
        <entry namest="c1" nameend="c2"><emphasis>Control structures</emphasis></entry>
      </row>
      <row>
        <entry><literal>if 1 + 1 == 2 then "yes!" else "no!"</literal></entry>
        <entry>Conditional expression</entry>
      </row>
      <row>
        <entry><literal>assert 1 + 1 == 2; "yes!"</literal></entry>
        <entry>Assertion check (evaluates to <literal>"yes!"</literal>)</entry>
      </row>
      <row>
        <entry><literal>let x = "foo"; y = "bar"; in x + y</literal></entry>
        <entry>Variable definition</entry>
      </row>

      <row>
        <entry namest="c1" nameend="c2"><emphasis>Functions (lambdas)</emphasis></entry>
      </row>
      <row>
        <entry><literal>x: x + 1</literal></entry>
        <entry>A function that expects an integer and returns it increased by 1</entry>
      </row>
      <row>
        <entry><literal>(x: x + 1) 100</literal></entry>
        <entry>A function call (evaluates to 101)</entry>
      </row>
      <row>
        <entry><literal>let inc = x: x + 1; in inc (inc (inc 100))</literal></entry>
        <entry>A function bound to a variable and subsequently called by name (evaluates to 103)</entry>
      </row>
      <row>
        <entry><literal>{ x, y }: x + y</literal></entry>
        <entry>A function that expects a set with required attributes
        <literal>x</literal> and <literal>y</literal> and concatenates
        them</entry>
      </row>
      <row>
        <entry><literal>{ x, y ? "bar" }: x + y</literal></entry>
        <entry>A function that expects a set with required attribute
        <literal>x</literal> and optional <literal>y</literal>, using
        <literal>"bar"</literal> as default value for
        <literal>y</literal></entry>
      </row>
      <row>
        <entry><literal>{ x, y, ... }: x + y</literal></entry>
        <entry>A function that expects a set with required attributes
        <literal>x</literal> and <literal>y</literal> and ignores any
        other attributes</entry>
      </row>
      <row>
        <entry><literal>{ x, y } @ args: x + y</literal></entry>
        <entry>A function that expects a set with required attributes
        <literal>x</literal> and <literal>y</literal>, and binds the
        whole set to <literal>args</literal></entry>
      </row>

      <row>
        <entry namest="c1" nameend="c2"><emphasis>Built-in functions</emphasis></entry>
      </row>
      <row>
        <entry><literal>import ./foo.nix</literal></entry>
        <entry>Load and return Nix expression in given file</entry>
      </row>
      <row>
        <entry><literal>map (x: x + x) [ 1 2 3 ]</literal></entry>
        <entry>Apply a function to every element of a list (evaluates to <literal>[ 2 4 6 ]</literal>)</entry>
      </row>
      <!--
      <row>
        <entry><literal>throw "Urgh"</literal></entry>
        <entry>Raise an error condition</entry>
      </row>
      -->

    </tbody>
  </tgroup>
</informaltable>

</section>


</section>


<!--===============================================================-->

<section><title>Package management</title>

<para>This section describes how to add additional packages to your
system.  NixOS has two distinct styles of package management:

<itemizedlist>

  <listitem><para><emphasis>Declarative</emphasis>, where you declare
  what packages you want in your
  <filename>configuration.nix</filename>.  Every time you run
  <command>nixos-rebuild</command>, NixOS will ensure that you get a
  consistent set of binaries corresponding to your
  specification.</para></listitem>

  <listitem><para><emphasis>Ad hoc</emphasis>, where you install,
  upgrade and uninstall packages via the <command>nix-env</command>
  command.  This style allows mixing packages from different Nixpkgs
  versions.  It’s the only choice for non-root
  users.</para></listitem>

</itemizedlist>

</para>

<para>The next two sections describe these two styles.</para>


<section><title>Declarative package management</title>

<para>With declarative package management, you specify which packages
you want on your system by setting the option
<option>environment.systemPackages</option>.  For instance, adding the
following line to <filename>configuration.nix</filename> enables the
Mozilla Thunderbird email application:

<programlisting>
environment.systemPackages = [ pkgs.thunderbird ];
</programlisting>

The effect of this specification is that the Thunderbird package from
Nixpkgs will be built or downloaded as part of the system when you run
<command>nixos-rebuild switch</command>.</para>

<para>You can get a list of the available packages as follows:
<screen>
$ nix-env -qaP '*' --description
nixos.pkgs.firefox   firefox-23.0   Mozilla Firefox - the browser, reloaded
<replaceable>...</replaceable>
</screen>

The first column in the output is the <emphasis>attribute
name</emphasis>, such as
<literal>nixos.pkgs.thunderbird</literal>. (The
<literal>nixos</literal> prefix allows distinguishing between
different channels that you might have.)</para>

<para>To “uninstall” a package, simply remove it from
<option>environment.systemPackages</option> and run
<command>nixos-rebuild switch</command>.</para>


<section xml:id="sec-customising-packages"><title>Customising packages</title>

<para>Some packages in Nixpkgs have options to enable or disable
optional functionality or change other aspects of the package.  For
instance, the Firefox wrapper package (which provides Firefox with a
set of plugins such as the Adobe Flash player) has an option to enable
the Google Talk plugin.  It can be set in
<filename>configuration.nix</filename> as follows:

<filename>
nixpkgs.config.firefox.enableGoogleTalkPlugin = true;
</filename>
</para>

<warning><para>Unfortunately, Nixpkgs currently lacks a way to query
available configuration options.</para></warning>

<para>Apart from high-level options, it’s possible to tweak a package
in almost arbitrary ways, such as changing or disabling dependencies
of a package.  For instance, the Emacs package in Nixpkgs by default
has a dependency on GTK+ 2.  If you want to build it against GTK+ 3,
you can specify that as follows:

<programlisting>
environment.systemPackages = [ (pkgs.emacs.override { gtk = pkgs.gtk3; }) ];
</programlisting>

The function <varname>override</varname> performs the call to the Nix
function that produces Emacs, with the original arguments amended by
the set of arguments specified by you.  So here the function argument
<varname>gtk</varname> gets the value <literal>pkgs.gtk3</literal>,
causing Emacs to depend on GTK+ 3.  (The parentheses are necessary
because in Nix, function application binds more weakly than list
construction, so without them,
<literal>environment.systemPackages</literal> would be a list with two
elements.)</para>

<para>Even greater customisation is possible using the function
<varname>overrideDerivation</varname>.  While the
<varname>override</varname> mechanism above overrides the arguments of
a package function, <varname>overrideDerivation</varname> allows
changing the <emphasis>result</emphasis> of the function.  This
permits changing any aspect of the package, such as the source code.
For instance, if you want to override the source code of Emacs, you
can say:

<programlisting>
environment.systemPackages =
  [ (pkgs.lib.overrideDerivation pkgs.emacs (attrs: {
      name = "emacs-25.0-pre";
      src = /path/to/my/emacs/tree;
    }))
  ];
</programlisting>

Here, <varname>overrideDerivation</varname> takes the Nix derivation
specified by <varname>pkgs.emacs</varname> and produces a new
derivation in which the original’s <literal>name</literal> and
<literal>src</literal> attribute have been replaced by the given
values.  The original attributes are accessible via
<varname>attrs</varname>.</para>

<para>The overrides shown above are not global.  They do not affect
the original package; other packages in Nixpkgs continue to depend on
the original rather than the customised package.  This means that if
another package in your system depends on the original package, you
end up with two instances of the package.  If you want to have
everything depend on your customised instance, you can apply a
<emphasis>global</emphasis> override as follows:

<screen>
nixpkgs.config.packageOverrides = pkgs:
  { emacs = pkgs.emacs.override { gtk = pkgs.gtk3; };
  };
</screen>

The effect of this definition is essentially equivalent to modifying
the <literal>emacs</literal> attribute in the Nixpkgs source tree.
Any package in Nixpkgs that depends on <literal>emacs</literal> will
be passed your customised instance.  (However, the value
<literal>pkgs.emacs</literal> in
<varname>nixpkgs.config.packageOverrides</varname> refers to the
original rather than overriden instance, to prevent an infinite
recursion.)</para>

</section>

<section xml:id="sec-custom-packages"><title>Adding custom packages</title>

<para>It’s possible that a package you need is not available in NixOS.
In that case, you can do two things.  First, you can clone the Nixpkgs
repository, add the package to your clone, and (optionally) submit a
patch or pull request to have it accepted into the main Nixpkgs
repository.  This is described in detail in the <link
xlink:href="http://nixos.org/nixpkgs/manual">Nixpkgs manual</link>.
In short, you clone Nixpkgs:

<screen>
$ git clone git://github.com/NixOS/nixpkgs.git
$ cd nixpkgs
</screen>

Then you write and test the package as described in the Nixpkgs
manual.  Finally, you add it to
<literal>environment.systemPackages</literal>, e.g.

<programlisting>
environment.systemPackages = [ pkgs.my-package ];
</programlisting>

and you run <command>nixos-rebuild</command>, specifying your own
Nixpkgs tree:

<screen>
$ nixos-rebuild switch -I nixpkgs=/path/to/my/nixpkgs</screen>

</para>

<para>The second possibility is to add the package outside of the
Nixpkgs tree.  For instance, here is how you specify a build of the
<link xlink:href="http://www.gnu.org/software/hello/">GNU Hello</link>
package directly in <filename>configuration.nix</filename>:

<programlisting>
environment.systemPackages =
  let
    my-hello = with pkgs; stdenv.mkDerivation rec {
      name = "hello-2.8";
      src = fetchurl {
        url = "mirror://gnu/hello/${name}.tar.gz";
        sha256 = "0wqd8sjmxfskrflaxywc7gqw7sfawrfvdxd9skxawzfgyy0pzdz6";
      };
    };
  in
  [ my-hello ];
</programlisting>

Of course, you can also move the definition of
<literal>my-hello</literal> into a separate Nix expression, e.g.
<programlisting>
environment.systemPackages = [ (import ./my-hello.nix) ];
</programlisting>
where <filename>my-hello.nix</filename> contains:
<programlisting>
with &lt;nixpkgs> {}; # bring all of Nixpkgs into scope

stdenv.mkDerivation rec {
  name = "hello-2.8";
  src = fetchurl {
    url = "mirror://gnu/hello/${name}.tar.gz";
    sha256 = "0wqd8sjmxfskrflaxywc7gqw7sfawrfvdxd9skxawzfgyy0pzdz6";
  };
}
</programlisting>

This allows testing the package easily:
<screen>
$ nix-build my-hello.nix
$ ./result/bin/hello
Hello, world!
</screen>

</para>

</section>

</section>


<section><title>Ad hoc package management</title>

<para>With the command <command>nix-env</command>, you can install and
uninstall packages from the command line.  For instance, to install
Mozilla Thunderbird:

<screen>
$ nix-env -iA nixos.pkgs.thunderbird</screen>

If you invoke this as root, the package is installed in the Nix
profile <filename>/nix/var/nix/profiles/default</filename> and visible
to all users of the system; otherwise, the package ends up in
<filename>/nix/var/nix/profiles/per-user/<replaceable>username</replaceable>/profile</filename>
and is not visible to other users.  The <option>-A</option> flag
specifies the package by its attribute name; without it, the package
is installed by matching against its package name
(e.g. <literal>thunderbird</literal>).  The latter is slower because
it requires matching against all available Nix packages, and is
ambiguous if there are multiple matching packages.</para>

<para>Packages come from the NixOS channel.  You typically upgrade a
package by updating to the latest version of the NixOS channel:
<screen>
$ nix-channel --update nixos
</screen>
and then running <literal>nix-env -i</literal> again.  Other packages
in the profile are <emphasis>not</emphasis> affected; this is the
crucial difference with the declarative style of package management,
where running <command>nixos-rebuild switch</command> causes all
packages to be updated to their current versions in the NixOS channel.
You can however upgrade all packages for which there is a newer
version by doing:
<screen>
$ nix-env -u '*'
</screen>
</para>

<para>A package can be uninstalled using the <option>-e</option>
flag:
<screen>
$ nix-env -e thunderbird
</screen>
</para>

<para>Finally, you can roll back an undesirable
<command>nix-env</command> action:
<screen>
$ nix-env --rollback
</screen>
</para>

<para><command>nix-env</command> has many more flags.  For details,
see the
<citerefentry><refentrytitle>nix-env</refentrytitle><manvolnum>1</manvolnum></citerefentry>
manpage or the Nix manual.</para>

</section>


</section>


<!--===============================================================-->

<section><title>User management</title>

<para>NixOS supports both declarative and imperative styles of user
management.  In the declarative style, users are specified in
<filename>configuration.nix</filename>.  For instance, the following
states that a user account named <literal>alice</literal> shall exist:

<programlisting>
users.extraUsers.alice =
  { createHome = true;
    home = "/home/alice";
    description = "Alice Foobar";
    extraGroups = [ "wheel" ];
    isSystemUser = false;
    useDefaultShell = true;
    openssh.authorizedKeys.keys = [ "ssh-dss AAAAB3Nza... alice@foobar" ];
  };
</programlisting>

Note that <literal>alice</literal> is a member of the
<literal>wheel</literal> group, which allows her to use
<command>sudo</command> to execute commands as
<literal>root</literal>.  Also note the SSH public key that allows
remote logins with the corresponding private key.  Users created in
this way do not have a password by default, so they cannot log in via
mechanisms that require a password.  However, you can use the
<command>passwd</command> program to set a password, which is retained
across invocations of <command>nixos-rebuild</command>.</para>

<para>A user ID (uid) is assigned automatically.  You can also specify
a uid manually by adding

<programlisting>
    uid = 1000;
</programlisting>

to the user specification.</para>

<para>Groups can be specified similarly.  The following states that a
group named <literal>students</literal> shall exist:

<programlisting>
users.extraGroups.students.gid = 1000;
</programlisting>

As with users, the group ID (gid) is optional and will be assigned
automatically if it’s missing.</para>

<warning><para>Currently declarative user management is not perfect:
<command>nixos-rebuild</command> does not know how to realise certain
configuration changes.  This includes removing a user or group, and
removing group membership from a user.</para></warning>

<para>In the imperative style, users and groups are managed by
commands such as <command>useradd</command>,
<command>groupmod</command> and so on.  For instance, to create a user
account named <literal>alice</literal>:

<screen>
$ useradd -m alice</screen>

The flag <option>-m</option> causes the creation of a home directory
for the new user, which is generally what you want.  The user does not
have an initial password and therefore cannot log in.  A password can
be set using the <command>passwd</command> utility:

<screen>
$ passwd alice
Enter new UNIX password: ***
Retype new UNIX password: ***
</screen>

A user can be deleted using <command>userdel</command>:

<screen>
$ userdel -r alice</screen>

The flag <option>-r</option> deletes the user’s home directory.
Accounts can be modified using <command>usermod</command>.  Unix
groups can be managed using <command>groupadd</command>,
<command>groupmod</command> and <command>groupdel</command>.</para>

</section>


<!--===============================================================-->

<section><title>File systems</title>

<para>You can define file systems using the
<option>fileSystems</option> configuration option.  For instance, the
following definition causes NixOS to mount the Ext4 file system on
device <filename>/dev/disk/by-label/data</filename> onto the mount
point <filename>/data</filename>:

<programlisting>
fileSystems."/data" =
  { device = "/dev/disk/by-label/data";
    fsType = "ext4";
  };
</programlisting>

Mount points are created automatically if they don’t already exist.
For <option>device</option>, it’s best to use the topology-independent
device aliases in <filename>/dev/disk/by-label</filename> and
<filename>/dev/disk/by-uuid</filename>, as these don’t change if the
topology changes (e.g. if a disk is moved to another IDE
controller).</para>

<para>You can usually omit the file system type
(<option>fsType</option>), since <command>mount</command> can usually
detect the type and load the necessary kernel module automatically.
However, if the file system is needed at early boot (in the initial
ramdisk) and is not <literal>ext2</literal>, <literal>ext3</literal>
or <literal>ext4</literal>, then it’s best to specify
<option>fsType</option> to ensure that the kernel module is
available.</para>

<section><title>LUKS-encrypted file systems</title>

<para>NixOS supports file systems that are encrypted using
<emphasis>LUKS</emphasis> (Linux Unified Key Setup).  For example,
here is how you create an encrypted Ext4 file system on the device
<filename>/dev/sda2</filename>:

<screen>
$ cryptsetup luksFormat /dev/sda2

WARNING!
========
This will overwrite data on /dev/sda2 irrevocably.

Are you sure? (Type uppercase yes): YES
Enter LUKS passphrase: ***
Verify passphrase: ***

$ cryptsetup luksOpen /dev/sda2 crypted
Enter passphrase for /dev/sda2: ***

$ mkfs.ext4 /dev/mapper/crypted
</screen>

To ensure that this file system is automatically mounted at boot time
as <filename>/</filename>, add the following to
<filename>configuration.nix</filename>:

<programlisting>
boot.initrd.luks.devices = [ { device = "/dev/sda2"; name = "crypted"; } ];
fileSystems."/".device = "/dev/mapper/crypted";
</programlisting>

</para>

</section>

</section>


<!--===============================================================-->

<section><title>X Window System</title>

<para>The X Window System (X11) provides the basis of NixOS’ graphical
user interface.  It can be enabled as follows:
<programlisting>
services.xserver.enable = true;
</programlisting>
The X server will automatically detect and use the appropriate video
driver from a set of X.org drivers (such as <literal>vesa</literal>
and <literal>intel</literal>).  You can also specify a driver
manually, e.g.
<programlisting>
services.xserver.videoDrivers = [ "r128" ];
</programlisting>
to enable X.org’s <literal>xf86-video-r128</literal> driver.</para>

<para>You also need to enable at least one desktop or window manager.
Otherwise, you can only log into a plain undecorated
<command>xterm</command> window.  Thus you should pick one or more of
the following lines:
<programlisting>
services.xserver.desktopManager.kde4.enable = true;
services.xserver.desktopManager.xfce.enable = true;
services.xserver.windowManager.xmonad.enable = true;
services.xserver.windowManager.twm.enable = true;
services.xserver.windowManager.icewm.enable = true;
</programlisting>
</para>

<para>NixOS’s default <emphasis>display manager</emphasis> (the
program that provides a graphical login prompt and manages the X
server) is SLiM.  You can select KDE’s <command>kdm</command> instead:
<programlisting>
services.xserver.displayManager.kdm.enable = true;
</programlisting>
</para>

<para>The X server is started automatically at boot time.  If you
don’t want this to happen, you can set:
<programlisting>
services.xserver.autorun = false;
</programlisting>
The X server can then be started manually:
<screen>
$ systemctl start display-manager.service
</screen>
</para>


<section><title>NVIDIA graphics cards</title>

<para>NVIDIA provides a proprietary driver for its graphics cards that
has better 3D performance than the X.org drivers.  It is not enabled
by default because it’s not free software.  You can enable it as follows:
<programlisting>
services.xserver.videoDrivers = [ "nvidia" ];
</programlisting>
You may need to reboot after enabling this driver to prevent a clash
with other kernel modules.</para>

<para>On 64-bit systems, if you want full acceleration for 32-bit
programs such as Wine, you should also set the following:
<programlisting>
service.xserver.driSupport32Bit = true;
</programlisting>
</para>

</section>


<section><title>Touchpads</title>

<para>Support for Synaptics touchpads (found in many laptops such as
the Dell Latitude series) can be enabled as follows:
<programlisting>
services.xserver.synaptics.enable = true;
</programlisting>
The driver has many options (see <xref linkend="ch-options"/>).  For
instance, the following enables two-finger scrolling:
<programlisting>
services.xserver.synaptics.twoFingerScroll = true;
</programlisting>
</para>

</section>


</section>


<!--===============================================================-->

<section><title>Networking</title>

<section><title>Secure shell access</title>

<para>Secure shell (SSH) access to your machine can be enabled by
setting:

<programlisting>
services.openssh.enable = true;
</programlisting>

By default, root logins using a password are disallowed.  They can be
disabled entirely by setting
<literal>services.openssh.permitRootLogin</literal> to
<literal>"no"</literal>.</para>

<para>You can declaratively specify authorised RSA/DSA public keys for
a user as follows:

<!-- FIXME: this might not work if the user is unmanaged. -->
<programlisting>
users.extraUsers.alice.openssh.authorizedKeys.keys =
  [ "ssh-dss AAAAB3NzaC1kc3MAAACBAPIkGWVEt4..." ];
</programlisting>

</para>

</section>


<section><title>IPv4 configuration</title>

<para>By default, NixOS uses DHCP (specifically,
<command>dhcpcd</command>) to automatically configure network
interfaces.  However, you can configure an interface manually as
follows:

<programlisting>
networking.interfaces.eth0 = { ipAddress = "192.168.1.2"; prefixLength = 24; };
</programlisting>

(The network prefix can also be specified using the option
<literal>subnetMask</literal>,
e.g. <literal>"255.255.255.0"</literal>, but this is deprecated.)
Typically you’ll also want to set a default gateway and set of name
servers:

<programlisting>
networking.defaultGateway = "192.168.1.1";
networking.nameservers = [ "8.8.8.8" ];
</programlisting>

</para>

<note><para>Statically configured interfaces are set up by the systemd
service
<replaceable>interface-name</replaceable><literal>-cfg.service</literal>.
The default gateway and name server configuration is performed by
<literal>network-setup.service</literal>.</para></note>

<para>The host name is set using <option>networking.hostName</option>:

<programlisting>
networking.hostName = "cartman";
</programlisting>

The default host name is <literal>nixos</literal>.  Set it to the
empty string (<literal>""</literal>) to allow the DHCP server to
provide the host name.</para>

</section>


<section><title>IPv6 configuration</title>

<para>IPv6 is enabled by default.  Stateless address autoconfiguration
is used to automatically assign IPv6 addresses to all interfaces.  You
can disable IPv6 support globally by setting:

<programlisting>
networking.enableIPv6 = false;
</programlisting>

</para>

</section>


<section><title>Firewall</title>

<para>NixOS has a simple stateful firewall that blocks incoming
connections and other unexpected packets.  The firewall applies to
both IPv4 and IPv6 traffic.  It can be enabled as follows:

<programlisting>
networking.firewall.enable = true;
</programlisting>

You can open specific TCP ports to the outside world:

<programlisting>
networking.firewall.allowedTCPPorts = [ 80 443 ];
</programlisting>

Note that TCP port 22 (ssh) is opened automatically if the SSH daemon
is enabled (<option>services.openssh.enable = true</option>).  UDP
ports can be opened through
<option>networking.firewall.allowedUDPPorts</option>.  Also of
interest is

<programlisting>
networking.firewall.allowPing = true;
</programlisting>

to allow the machine to respond to ping requests.  (ICMPv6 pings are
always allowed.)</para>

</section>


<section><title>Wireless networks</title>

<para>
NixOS will start wpa_supplicant for you if you enable this setting:

<programlisting>
networking.wireless.enable = true;
</programlisting>

NixOS currently does not generate wpa_supplicant's
configuration file, <literal>/etc/wpa_supplicant.conf</literal>. You should edit this file
yourself to define wireless networks, WPA keys and so on (see
wpa_supplicant.conf(5)).
</para>

<para>
If you are using WPA2 the <command>wpa_passphrase</command> tool might be useful
to generate the <literal>wpa_supplicant.conf</literal>.

<screen>
$ wpa_passphrase ESSID PSK > /etc/wpa_supplicant.conf</screen>

After you have edited the <literal>wpa_supplicant.conf</literal>,
you need to restart the wpa_supplicant service.

<screen>
$ systemctl restart wpa_supplicant.service</screen>
</para>


</section>


<section><title>Ad-hoc configuration</title>

<para>You can use <option>networking.localCommands</option> to specify
shell commands to be run at the end of
<literal>network-setup.service</literal>.  This is useful for doing
network configuration not covered by the existing NixOS modules.  For
instance, to statically configure an IPv6 address:

<programlisting>
networking.localCommands =
  ''
    ip -6 addr add 2001:610:685:1::1/64 dev eth0
  '';
</programlisting>

</para>

</section>


<!-- TODO: OpenVPN, NAT -->


</section>


<!--===============================================================-->

<section><title>Linux kernel</title>

<para>You can override the Linux kernel and associated packages using
the option <option>boot.kernelPackages</option>.  For instance, this
selects the Linux 3.10 kernel:
<programlisting>
boot.kernelPackages = pkgs.linuxPackages_3_10;
</programlisting>
Note that this not only replaces the kernel, but also packages that
are specific to the kernel version, such as the NVIDIA video drivers.
This ensures that driver packages are consistent with the
kernel.</para>

<para>The default Linux kernel configuration should be fine for most
users.  You can see the configuration of your current kernel in
<filename>/run/booted-system/kernel-modules/config</filename>.  If you
want to change the kernel configuration, you can use the
<option>packageOverrides</option> feature (see <xref
linkend="sec-customising-packages" />).  For instance, to enable
support for the kernel debugger KGDB:

<programlisting>
nixpkgs.config.packageOverrides = pkgs:
  { linux_3_4 = pkgs.linux_3_4.override {
      extraConfig =
        ''
          KGDB y
        '';
    };
  };
</programlisting>

<varname>extraConfig</varname> takes a list of Linux kernel
configuration options, one per line.  The name of the option should
not include the prefix <literal>CONFIG_</literal>.  The option value
is typically <literal>y</literal>, <literal>n</literal> or
<literal>m</literal> (to build something as a kernel module).</para>

<para>Kernel modules for hardware devices are generally loaded
automatically by <command>udev</command>.  You can force a module to
be loaded via <option>boot.kernelModules</option>, e.g.
<programlisting>
boot.kernelModules = [ "fuse" "kvm-intel" "coretemp" ];
</programlisting>
If the module is required early during the boot (e.g. to mount the
root file system), you can use
<option>boot.initrd.extraKernelModules</option>:
<programlisting>
boot.initrd.extraKernelModules = [ "cifs" ];
</programlisting>
This causes the specified modules and their dependencies to be added
to the initial ramdark.</para>

<para>Kernel runtime parameters can be set through
<option>boot.kernel.sysctl</option>, e.g.
<programlisting>
boot.kernel.sysctl."net.ipv4.tcp_keepalive_time" = 120;
</programlisting>
sets the kernel’s TCP keepalive time to 120 seconds.  To see the
available parameters, run <command>sysctl -a</command>.</para>

</section>


<!-- Apache; libvirtd virtualisation -->


</chapter>