7 Concept and Workflow #

KIWI NG installs packages into your appliance from the repositories defined in the image description. Therefore at least one repository must be defined, as KIWI NG will otherwise not be able to fetch any packages.

A repository is added to the description via the repository element, which is a child of the top-level image element:

<image schemaversion="7.4" name="{exc_image_base_name}">
    <!-- snip -->
    <repository type="rpm-md" alias="kiwi" priority="1">
        <source path="obs://Virtualization:Appliances:Builder/openSUSE_Leap_15.3"/>
    </repository>
    <repository type="rpm-md" alias="OS" imageinclude="true">
        <source path="{exc_repo}"/>
    </repository>
</image>

In the above snippet we defined two repositories:

The repository element accepts one source child element, which contains the URL to the repository in an appropriate format and the following optional attributes:

The package element represents a single package to be installed (or removed), whose name is specified via the mandatory name attribute:

<image schemaversion="7.4" name="{exc_image_base_name}">
    <!-- snip -->
    <packages type="bootstrap">
        <package name="udev"/>
    </packages>
</image>

which adds the package udev to the list of packages to be added to the initial filesystem. Note, that the value that you pass via the name attribute is passed directly to the used package manager. Thus, if the package manager supports other means how packages can be specified, you may pass them in this context too. For example, RPM based package managers (like dnf or zypper) can install packages via their Provides:. This can be used to add a package that provides a certain capability (e.g. Provides: /usr/bin/my-binary) via:

<image schemaversion="7.4" name="{exc_image_base_name}">
    <!-- snip -->
    <packages type="bootstrap">
        <package name="/usr/bin/my-binary"/>
    </packages>
</image>

Whether this works depends on the package manager and on the environment that is being used. In the Open Build Service, certain Provides either are not visible or cannot be properly extracted from the KIWI NG description. Therefore, relying on Provides is not recommended.

Packages can also be included only on specific host architectures via the arch attribute. KIWI NG compares the arch attributes value with the host architecture that builds the image according to the output of uname -m.

<image schemaversion="7.4" name="{exc_image_base_name}">
    <!-- snip -->
    <packages type="image">
        <package name="grub2"/>
        <package name="grub2-x86_64-efi" arch="x86_64"/>
        <package name="shim" arch="x86_64"/>
    </packages>
</image>

which results in grub2-x86_64-efi and shim being only installed if the build host is a 64bit x86 machine, but grub2 will be installed independent of the architecture.

It is sometimes necessary to include additional packages into the image which are not available in the package manager’s native format. KIWI NG supports the inclusion of ordinary tar archives via the archive element, whose name attribute specifies the filename of the archive (KIWI NG looks for the archive in the image description folder).

<packages type="image">
    <archive name="custom-program1.tgz"/>
    <archive name="custom-program2.tar"/>
</packages>

KIWI NG will extract the archive into the root directory of the image using GNU tar, thus only archives supported by it can be included. When multiple archive elements are specified then they will be applied in a top to bottom order. If a file is already present in the image, then the file from the archive will overwrite it (same as with the image overlay).

KIWI NG supports two different methods how packages can be removed from the appliance:

Both types of removals take place after config.sh is run in the Section 7.10.1, “The Prepare Step” (see also Section 7.6, “User Defined Scripts”).

Warning

An uninstall packages request deletes:

• the listed packages,

• the packages dependent on the listed ones, and

• any orphaned dependency of the listed packages.

Use this feature with caution as it can easily cause the removal of sensitive tools leading to failures in later build stages.

Removing packages via type="uninstall" can be used to completely remove a build time tool (e.g. a compiler) without having to specify a all dependencies of that tool (as one would have when using type="delete"). Consider the following example where we wish to compile a custom program in config.sh. We ship its source code via an archive element and add the build tools (ninja, meson and clang) to <packages type="image"> and <packages type="uninstall">:

<image schemaversion="7.4" name="{exc_image_base_name}">
    <!-- snip -->
    <packages type="image">
        <package name="ca-certificates"/>
        <package name="coreutils"/>
        <package name="ninja"/>
        <package name="clang"/>
        <package name="meson"/>
        <archive name="foo_app_sources.tar.gz"/>
    </packages>
    <!-- These packages will be uninstalled after running config.sh -->
    <packages type="uninstall">
        <package name="ninja"/>
        <package name="meson"/>
        <package name="clang"/>
    </packages>
</image>

The tools meson, clang and ninja are then available during the Section 7.10.1, “The Prepare Step” and can thus be used in config.sh (for further details, see Section 7.6, “User Defined Scripts”), for example to build foo_app:

pushd /opt/src/foo_app
mkdir build
export CC=clang
meson build
cd build && ninja && ninja install
popd

The <packages type="uninstall"> element will make sure that the final appliance will no longer contain our tools required to build foo_app, thus making our image smaller.

There are also other use cases for type="uninstall", especially for specialized appliances. For containers one can often remove the package shadow (it is required to setup new user accounts) or any left over partitioning tools (parted or fdisk). All networking tools can be safely uninstalled in images for embedded devices without a network connection.

KIWI NG supports the inclusion of openSUSE products or of namedCollections (patterns in SUSE based distributions or groups for RedHat based distributions). These can be added via the product and namedCollection child elements, which both take the mandatory name attribute and the optional arch attribute.

product and namedCollection can be utilized to shorten the list of packages that need to be added to the image description tremendously. A named pattern, specified with the namedCollection element is a representation of a predefined list of packages. Specifying a pattern will install all packages listed in the named pattern. Support for patterns is distribution specific and available in SLES, openSUSE, CentOS, RHEL and Fedora. The optional patternType attribute on the packages element allows you to control the installation of dependent packages. You may assign one of the following values to the patternType attribute:

Packages can be explicitly marked to be ignored for installation inside a packages collection. This useful to exclude certain packages from being installed when using patterns with patternType="plusRecommended" as shown in the following example:

<image schemaversion="7.4" name="{exc_image_base_name}">
    <packages type="image" patternType="plusRecommended">
        <namedCollection name="network-server"/>
        <package name="grub2"/>
        <package name="kernel"/>
        <ignore name="ejabberd"/>
        <ignore name="puppet-server"/>
    </packages>
</image>

Packages can be marked as ignored during the installation by adding a ignore child element with the mandatory name attribute set to the package’s name. Optionally one can also specify the architecture via the arch similarly to The package element.

Warning

Adding ignore elements as children of a <packages type="delete"> or a <packages type="uninstall"> element has no effect! The packages will still get deleted.

When creating a custom script it usually takes some iterations of try and testing until a final stable state is reached. To support developers with this task KIWI NG calls scripts associated with a screen session. The connection to screen is only done if KIWI NG is called with the --debug option.

In this mode a script can start like the following template:

# The magic bits are still not set

echo "break"
/bin/bash

At call time of the script a screen session executes and you get access to the break in shell. From this environment the needed script code can be implemented. Once the shell is closed the KIWI NG process continues.

Apart from providing a full featured terminal throughout the execution of the script code, there is also the advantage to have control on the session during the process of the image creation. Listing the active sessions for script execution can be done as follows:

$ sudo screen -list

There is a screen on:
     19699.pts-4.asterix     (Attached)
1 Socket in /run/screens/S-root.

Note

As shown above the screen session(s) to execute script code provides extended control which could also be considered a security risk. Because of that KIWI NG only runs scripts through screen when explicitly enabled via the --debug switch. For production processes all scripts should run in their native way and should not require a terminal to operate correctly !

7.6.1.1 Script Template for config.sh / images.sh #

KIWI NG provides a collection of methods and variables that supports users with custom operations. For details see Functions and Variables Provided by KIWI NG. The following template shows how to import this information in your script:

#======================================
# Include functions & variables
#--------------------------------------
test -f /.kconfig && . /.kconfig
test -f /.profile && . /.profile

...

Warning

Modifications of the unpacked root tree

Keep in mind that there is only one unpacked root tree the script operates in. This means that all changes are permanent and will not be automatically restored!

The dracut initrd system uses systemd to implement a predefined workflow of services which are documented in the bootup man page at:

http://man7.org/linux/man-pages/man7/dracut.bootup.7.html

To hook in a custom boot script into this workflow it’s required to provide a dracut module which is picked up by dracut at the time KIWI NG calls it. The module files can be either provided as a package or as part of the overlay directory in your image description

The following example demonstrates how to include a custom hook script right before the system rootfs gets mounted.

Note

Declaring Extra Tools for Hook Scripts

The install() function called by dracut can define extra tools needed by a defined hook script. The “inst_multiple” command and its parameters inform dracut to include these extra tools/items in the initrd.

The tools/items defined here can be any file, but are usually executables and libraries needed by the hook script.

• Each file MUST be included in the Kiwi description either in a package, archive, or in the “root” tree in the image description directory.

• The parameters of the inst_multiple command are space separated.

• Each parameter can be a single executable name if it exists in /bin, /sbin, /usr/bin, or /usr/sbin directories.

• Otherwise, a full pathname to the file is required. This is usually true for libraries and other special files.

That’s it! At the time KIWI NG calls dracut the 90my-module will be taken into account and is installed into the generated initrd. At boot time systemd calls the scripts as part of the dracut-pre-mount.service.

The dracut system offers a lot more possibilities to customize the initrd than shown in the example above. For more information, visit the dracut project page.

A dracut generated initrd in a KIWI NG image build process includes one or more of the KIWI NG provided dracut modules. The following list documents the available kernel boot parameters for this modules:

rd.debug rd.kiwi.debug

less /run/initramfs/log/boot.kiwi

As the first step, KIWI NG creates an unpackaged image tree, also called “root tree”. This directory will be the installation target for software packages to be installed during the image creation process.

For the package installation, KIWI NG relies on the package manager specified in the packagemanager element in config.xml. KIWI NG supports the following package managers: dnf, zypper (default) and apt.

The prepare step consists of the following substeps:

KIWI NG creates the final image during the create step: it converts the unpacked root tree into one or multiple output files appropriate for the respective build type.

It is possible to create multiple images from the same unpacked root tree, for example, a self installing OEM image and a virtual machine image from the same image description. The only prerequisite is that both image types are specified in config.xml.

During the create step the following operations are performed by KIWI NG: