livemedia-creator
=================

:Authors:
    Brian C. Lane <bcl@redhat.com>

livemedia-creator uses `Anaconda <https://github.com/rhinstaller/anaconda>`_,
`kickstart <https://github.com/rhinstaller/pykickstart>`_ and `Lorax
<https://github.com/rhinstaller/lorax>`_ to create bootable media that use the
same install path as a normal system installation. It can be used to make live
isos, bootable (partitioned) disk images, tarfiles, and filesystem images for
use with virtualization and container solutions like libvirt, docker, and
OpenStack.

The general idea is to use qemu with kickstart and an Anaconda boot.iso to
install into a disk image and then use the disk image to create the bootable
media.

livemedia-creator --help will describe all of the options available. At the
minimum you need:

``--make-iso`` to create a final bootable .iso or one of the other ``--make-*`` options.

``--iso`` to specify the Anaconda install media to use with qemu.

``--ks`` to select the kickstart file describing what to install.

To use livemedia-creator with virtualization you will need to have qemu-kvm installed.

If you are going to be using Anaconda directly, with ``--no-virt`` mode, make sure
you have the anaconda-tui package installed.

Conventions used in this document:

``lmc`` is an abbreviation for livemedia-creator.

``builder`` is the system where livemedia-creator is being run

``image`` is the disk image being created by running livemedia-creator


livemedia-creator cmdline arguments
-----------------------------------

.. argparse::
   :ref: pylorax.cmdline.lmc_parser
   :prog: livemedia-creator


Quickstart
----------

Run this to create a bootable live iso::

    sudo livemedia-creator --make-iso \
    --iso=/extra/iso/boot.iso --ks=./docs/fedora-livemedia.ks

You can run it directly from the lorax git repo like this::

    sudo PATH=./src/sbin/:$PATH PYTHONPATH=./src/ ./src/sbin/livemedia-creator \
    --make-iso --iso=/extra/iso/boot.iso \
    --ks=./docs/fedora-livemedia.ks --lorax-templates=./share/

You can observe the installation using vnc. The logs will show what port was
chosen, or you can use a specific port by passing it. eg. ``--vnc vnc:127.0.0.1:5``

This is usually a good idea when testing changes to the kickstart. lmc tries
to monitor the logs for fatal errors, but may not catch everything.


How ISO creation works
----------------------

There are 2 stages, the install stage which produces a disk or filesystem image
as its output, and the boot media creation which uses the image as its input.
Normally you would run both stages, but it is possible to stop after the
install stage, by using ``--image-only``, or to skip the install stage and use
a previously created disk image by passing ``--disk-image`` or ``--fs-image``

When creating an iso qemu boots using the passed Anaconda installer iso
and installs the system based on the kickstart. The ``%post`` section of the
kickstart is used to customize the installed system in the same way that
current spin-kickstarts do.

livemedia-creator monitors the install process for problems by watching the
install logs. They are written to the current directory or to the base
directory specified by the --logfile command. You can also monitor the install
by using a vnc client. This is recommended when first modifying a kickstart,
since there are still places where Anaconda may get stuck without the log
monitor catching it.

The output from this process is a partitioned disk image. kpartx can be used
to mount and examine it when there is a problem with the install. It can also
be booted using kvm.

When creating an iso the disk image's / partition is copied into a formatted
filesystem image which is then used as the input to lorax for creation of the
final media.

The final image is created by lorax, using the templates in /usr/share/lorax/live/
or the live directory below the directory specified by ``--lorax-templates``. The
templates are written using the Mako template system with some extra commands
added by lorax.

.. note::
    The output from --make-iso includes the artifacts used to create the boot.iso;
    the kernel, initrd, the squashfs filesystem, etc. If you only want the
    boot.iso you can pass ``--iso-only`` and the other files will be removed. You
    can also name the iso by using ``--iso-name my-live.iso``.


Kickstarts
----------

The docs/ directory includes several example kickstarts, one to create a live
desktop iso using GNOME, and another to create a minimal disk image. When
creating your own kickstarts you should start with the minimal example, it
includes several needed packages that are not always included by dependencies.

Or you can use existing spin kickstarts to create live media with a few
changes. Here are the steps I used to convert the Fedora XFCE spin.

1. Flatten the xfce kickstart using ksflatten
2. Add zerombr so you don't get the disk init dialog
3. Add clearpart --all
4. Add swap partition
5. bootloader target
6. Add shutdown to the kickstart
7. Add network --bootproto=dhcp --activate to activate the network
   This works for F16 builds but for F15 and before you need to pass
   something on the cmdline that activate the network, like sshd:

    ``livemedia-creator --kernel-args="sshd"``

8. Add a root password::

    rootpw rootme
    network --bootproto=dhcp --activate
    zerombr
    clearpart --all
    bootloader --location=mbr
    part swap --size=512
    shutdown

9. In the livesys script section of the %post remove the root password. This
   really depends on how the spin wants to work. You could add the live user
   that you create to the %wheel group so that sudo works if you wanted to.

    ``passwd -d root > /dev/null``

10. Remove /etc/fstab in %post, dracut handles mounting the rootfs

    ``cat /dev/null > /dev/fstab``

    Do this only for live iso's, the filesystem will be mounted read only if
    there is no /etc/fstab

11. Don't delete initramfs files from /boot in %post
12. When creating live iso's you need to have, at least, these packages in the %package section::
    dracut-config-generic
    dracut-live
    -dracut-config-rescue
    grub-efi
    memtest86+
    syslinux

One drawback to using qemu is that it pulls the packages from the repo each
time you run it. To speed things up you either need a local mirror of the
packages, or you can use a caching proxy. When using a proxy you pass it to
livemedia-creator like this:

    ``--proxy=http://proxy.yourdomain.com:3128``

You also need to use a specific mirror instead of mirrormanager so that the
packages will get cached, so your kickstart url would look like:

    ``url --url="http://dl.fedoraproject.org/pub/fedora/linux/development/rawhide/x86_64/os/"``

You can also add an update repo, but don't name it updates. Add --proxy to it
as well.


Anaconda image install (no-virt)
--------------------------------

You can create images without using qemu by passing ``--no-virt`` on the
cmdline. This will use Anaconda's directory install feature to handle the
install.  There are a couple of things to keep in mind when doing this:

1. It will be most reliable when building images for the same release that the
   host is running. Because Anaconda has expectations about the system it is
   running under you may encounter strange bugs if you try to build newer or
   older releases.

2. Make sure selinux is set to permissive or disabled. It won't install
   correctly with selinux set to enforcing yet.

3. It may totally trash your host. So far I haven't had this happen, but the
   possibility exists that a bug in Anaconda could result in it operating on
   real devices. I recommend running it in a virt or on a system that you can
   afford to lose all data from.

The logs from anaconda will be placed in an ./anaconda/ directory in either
the current directory or in the directory used for --logfile

Example cmdline:

``sudo livemedia-creator --make-iso --no-virt --ks=./fedora-livemedia.ks``

.. note::
    Using no-virt to create a partitioned disk image (eg. --make-disk or
    --make-vagrant) will only create disks usable on the host platform (BIOS
    or UEFI). You can create BIOS partitioned disk images on UEFI by using
    virt.


AMI Images
----------

Amazon EC2 images can be created by using the --make-ami switch and an appropriate
kickstart file. All of the work to customize the image is handled by the kickstart.
The example currently included was modified from the cloud-kickstarts version so
that it would work with livemedia-creator.

Example cmdline:

``sudo livemedia-creator --make-ami --iso=/path/to/boot.iso --ks=./docs/fedora-livemedia-ec2.ks``

This will produce an ami-root.img file in the working directory.

At this time I have not tested the image with EC2. Feedback would be welcome.


Appliance Creation
------------------

livemedia-creator can now replace appliance-tools by using the --make-appliance
switch. This will create the partitioned disk image and an XML file that can be
used with virt-image to setup a virtual system.

The XML is generated using the Mako template from
/usr/share/lorax/appliance/libvirt.xml You can use a different template by
passing ``--app-template <template path>``

Documentation on the Mako template system can be found at the `Mako site
<http://docs.makotemplates.org/en/latest/index.html>`_

The name of the final output XML is appliance.xml, this can be changed with
``--app-file <file path>``

The following variables are passed to the template:

    ``disks``
       A list of disk_info about each disk.
       Each entry has the following attributes:

        ``name``
        base name of the disk image file

        ``format``
        "raw"

        ``checksum_type``
        "sha256"

        ``checksum``
        sha256 checksum of the disk image

    ``name``
    Name of appliance, from --app-name argument

    ``arch``
    Architecture

    ``memory``
    Memory in KB (from ``--ram``)

    ``vcpus``
    from ``--vcpus``

    ``networks``
    list of networks from the kickstart or []

    ``title``
    from ``--title``

    ``project``
    from ``--project``

    ``releasever``
    from ``--releasever``

The created image can be imported into libvirt using:

    ``virt-image appliance.xml``

You can also create qcow2 appliance images using ``--image-type=qcow2``, for example::

    sudo livemedia-creator --make-appliance --iso=/path/to/boot.iso --ks=./docs/fedora-minimal.ks \
    --image-type=qcow2 --app-file=minimal-test.xml --image-name=minimal-test.img


Filesystem Image Creation
-------------------------

livemedia-creator can be used to create un-partitined filesystem images using
the ``--make-fsimage`` option. As of version 21.8 this works with both qemu and
no-virt modes of operation. Previously it was only available with no-virt.

Kickstarts should have a single / partition with no extra mountpoints.

    ``livemedia-creator --make-fsimage --iso=/path/to/boot.iso --ks=./docs/fedora-minimal.ks``

You can name the output image with ``--image-name`` and set a label on the filesystem with ``--fs-label``


TAR File Creation
-----------------

The ``--make-tar`` command can be used to create a tar of the root filesystem. By
default it is compressed using xz, but this can be changed using the
``--compression`` and ``--compress-arg`` options. This option works with both virt and
no-virt install methods.

As with ``--make-fsimage`` the kickstart should be limited to a single / partition.

For example::

    livemedia-creator --make-tar --iso=/path/to/boot.iso --ks=./docs/fedora-minimal.ks \
    --image-name=fedora-root.tar.xz


Live Image for PXE Boot
-----------------------

The ``--make-pxe-live`` command will produce squashfs image containing live root
filesystem that can be used for pxe boot. Directory with results will contain
the live image, kernel image, initrd image and template of pxe configuration
for the images.


Atomic Live Image for PXE Boot
------------------------------

The ``--make-ostree-live`` command will produce the same result as ``--make-pxe-live``
for installations of Atomic Host.  Example kickstart for such an installation
using Atomic installer iso with local repo included in the image can be found
in docs/rhel-atomic-pxe-live.ks.

The PXE images can also be created with ``--no-virt`` by using the example
kickstart in docs/fedora-atomic-pxe-live-novirt.ks. This also works inside the
mock environment.


Using Mock and --no-virt to Create Images
-----------------------------------------

As of lorax version 22.2 you can use livemedia-creator and anaconda version
22.15 inside of a mock chroot with --make-iso and --make-fsimage.

.. note::
    As of mock 1.3.4 you need to use ``--old-chroot`` with mock. Mock now defaults to using systemd-nspawn
    which cannot create the needed loop device nodes. Passing ``--old-chroot`` will use the old system
    where ``/dev/loop*`` is setup for you.

On the host system:

1. yum install -y mock

2. Add a user to the mock group to use for running mock. eg. builder

3. Create a new /etc/mock/ config file based on the rawhide one, or modify the
   existing one so that the following options are setup::

       config_opts['chroot_setup_cmd'] = 'install @buildsys-build anaconda-tui lorax'

       # build results go into /home/builder/results/
       config_opts['plugin_conf']['bind_mount_opts']['dirs'].append(('/home/builder/results','/results/'))

   If you are creating images for a branched release of Fedora you should also enable
   the updates-testing repository so that you get the latest builds in your mock chroot.

The following steps are run as the builder user who is a member of the mock
group.

4. Make a directory for results matching the bind mount above
   ``mkdir ~/results/``

5. Copy the example kickstarts
   ``cp /usr/share/docs/lorax/*ks .``

6. Make sure tar and dracut-network are in the %packages section and that the
   ``url points to the correct repo``

7. Init the mock
   ``mock -r fedora-rawhide-x86_64 --old-chroot --init``

8. Copy the kickstart inside the mock
   ``mock -r fedora-rawhide-x86_64 --old-chroot --copyin ./fedora-minimal.ks /root/``

9. Make a minimal iso::

        mock -r fedora-rawhide-x86_64 --old-chroot --chroot -- livemedia-creator --no-virt \
        --resultdir=/results/try-1 --logfile=/results/logs/try-1/try-1.log \
        --make-iso --ks /root/fedora-minimal.ks

Results will be in ./results/try-1 and logs under /results/logs/try-1/
including anaconda logs and livemedia-creator logs. The new iso will be
located at ~/results/try-1/images/boot.iso, and the ~/results/try-1/
directory tree will also contain the vmlinuz, initrd, etc.


Using Mock and qemu to Create Images
------------------------------------

Version 25.0 of livemedia-creator switches to using qemu for virtualization.
This allows creation of all image types, and use of the KVM on the host if
/dev/kvm is present in the mock environment.

On the host system:

1. yum install -y mock

2. Add a user to the mock group to use for running mock. eg. builder

3. Create a new /etc/mock/ config file based on the rawhide one, or modify the
   existing one so that the following options are setup::

       config_opts['chroot_setup_cmd'] = 'install @buildsys-build lorax qemu'

       # build results go into /home/builder/results/
       config_opts['plugin_conf']['bind_mount_opts']['dirs'].append(('/home/builder/results','/results/'))

   If you are creating images for a branched release of Fedora you should also enable
   the updates-testing repository so that you get the latest builds in your mock chroot.

The following steps are run as the builder user who is a member of the mock
group.

4. Make a directory for results matching the bind mount above
   ``mkdir ~/results/``

5. Copy the example kickstarts
   ``cp /usr/share/docs/lorax/*ks .``

6. Make sure tar and dracut-network are in the %packages section and that the
   ``url points to the correct repo``

7. Init the mock
   ``mock -r fedora-rawhide-x86_64 --old-chroot --init``

8. Copy the kickstart inside the mock
   ``mock -r fedora-rawhide-x86_64 --old-chroot --copyin ./fedora-minimal.ks /root/``

9. Copy the Anaconda boot.iso inside the mock
   ``mock -r fedora-rawhide-x86_64 --old-chroot --copyin ./boot.iso /root/``

10. Make a minimal iso::

        mock -r fedora-rawhide-x86_64 --old-chroot --chroot -- livemedia-creator \
        --resultdir=/results/try-1 --logfile=/results/logs/try-1/try-1.log \
        --make-iso --ks /root/fedora-minimal.ks --iso /root/boot.iso

Results will be in ./results/try-1 and logs under /results/logs/try-1/
including anaconda logs and livemedia-creator logs. The new iso will be
located at ~/results/try-1/images/boot.iso, and the ~/results/try-1/
directory tree will also contain the vmlinuz, initrd, etc.

This will run qemu without kvm support, which is going to be very slow. You can
add ``mknod /dev/kvm c 10 232;`` to create the device node before running lmc.


OpenStack Image Creation
------------------------

OpenStack supports partitioned disk images so ``--make-disk`` can be used to
create images for importing into glance, OpenStack's image storage component.
You need to have access to an OpenStack provider that allows image uploads, or
setup your own using the instructions from the `RDO Project
<https://www.rdoproject.org/Quickstart>`_.

The example kickstart, fedora-openstack.ks, is only slightly different than the
fedora-minimal.ks one.  It adds the cloud-init and cloud-utils-growpart
packages. OpenStack supports setting up the image using cloud-init, and
cloud-utils-growpart will grow the image to fit the instance's disk size.

Create a qcow2 image using the kickstart like this:

    ``sudo livemedia-creator --make-disk --iso=/path/to/boot.iso --ks=/path/to/fedora-openstack.ks --image-type=qcow2``

.. note::
    On the RHEL7 version of lmc ``--image-type`` isn't supported. You can only create a bare partitioned disk image.

Import the resulting disk image into the OpenStack system, either via the web UI, or glance on the cmdline::

    glance image-create --name "fedora-openstack" --is-public true --disk-format qcow2 \
    --container-format bare --file ./fedora-openstack.qcow2

If qcow2 wasn't used then ``--disk-format`` should be set to raw.


Docker Image Creation
---------------------

Use lmc to create a tarfile as described in the `TAR File Creation`_ section, but substitute the
fedora-docker.ks example kickstart which removes the requirement for core files and the kernel.

You can then import the tarfile into docker like this (as root):

    ``cat /var/tmp/fedora-root.tar.xz | docker import - fedora-root``

And then run bash inside of it:

    ``sudo docker run -i -t fedora-root /bin/bash``


Open Container Initiative Image Creation
----------------------------------------

The OCI is a new specification that is still being worked on. You can read more about it at
`the Open Container Initiative website <https://www.opencontainers.org/>`_. You can create
OCI images using the following command::

    sudo livemedia-creator --make-oci --oci-config /path/to/config.json --oci-runtime /path/to/runtime.json \
    --iso=/path/to/boot.iso --ks=/path/to/fedora-minimal.ks

You must provide the config.json and runtime.json files to be included in the bundle,
their specifications can be found `on the OCI github project <https://github.com/opencontainers/specs>`_
output will be in the results directory with a default name of bundle.tar.xz

This will work with ``--no-virt`` and inside a mock since it doesn't use any
partitioned disk images.


Vagrant Image Creation
----------------------

`Vagrant <https://www.vagrantup.com/>`_ images can be created using the following command::

    sudo livemedia-creator --make-vagrant --vagrant-metadata /path/to/metadata.json \
    --iso=/path/to/boot.iso --ks=/path/to/fedora-vagrant.ks

The image created is a `vagrant-libvirt
<https://github.com/pradels/vagrant-libvirt>`_ provider image and needs to have
vagrant setup with libvirt before you can use it.

The ``--vagrant-metadata`` file is optional, it will create a minimal one by
default, and if one is passed it will make sure the disk size  is setup
correctly. If you pass a ``--vagrant-vagrantfile`` it will be included in the
image verbatim. By default no vagrantfile is created.

There is an example Vagrant kickstart file in the docs directory that sets up
the vagrant user with the default insecure SSH pubkey and a few useful
utilities.

This also works with ``--no-virt``, but will not work inside a mock due to its
use of partitioned disk images and qcow2.


Creating UEFI disk images with virt
-----------------------------------

Partitioned disk images can only be created for the same platform as the host system (BIOS or
UEFI). You can use virt to create BIOS images on UEFI systems, and it is also possible
to create UEFI images on BIOS systems using OVMF. You first need to setup your system with
the OVMF firmware. The details can be `found here linux-kvm OVMF page <http://www.linux-kvm.org/page/OVMF>`_
but it amounts to:

1. Download the firmware.repo from `Gerd Hoffmann <https://www.kraxel.org/repos/>`_ and install it
   in /etc/yum.repos.d/

2. Install the edk2.git-ovmf-x64 package

3. Copy /usr/share/edk2.git/ovmf-x64/OVMF_CODE-pure-efi.fd to /usr/share/OVMF/OVMF_CODE.fd

4. Copy /usr/share/edk2.git/ovmf-x64/OVMF_VARS-pure-efi.fd to /usr/share/OVMF/OVMF_VARS.fd

Now you can run livemedia-creator with ``--virt-uefi`` to boot and install using UEFI::

    sudo livemedia-creator --make-disk --virt-uefi --iso=/path/to/boot.iso \
    --ks=/path/to/fedora-minimal.ks

Make sure that the kickstart you are using creates a /boot/efi partition by including this::

    part /boot/efi --fstype="efi" --size=500

.. note::
    When using the resulting image with the current version of OVMF (edk2.git-ovmf-x64-0-20151103.b1295.ge5cffca)
    it will not boot automatically because there is a problem with the fallback path.
    You can boot it by entering the UEFI shell and running EFI/fedora/shim.efi and
    then using efibootmgr to setup the correct boot entry.


Debugging problems
------------------

Sometimes an installation will get stuck. When using qemu the logs will
be written to ./virt-install.log and most of the time any problems that happen
will be near the end of the file. lmc tries to detect common errors and will
cancel the installation when they happen. But not everything can be caught.
When creating a new kickstart it is helpful to use vnc so that you can monitor
the installation as it happens, and if it gets stuck without lmc detecting the
problem you can switch to tty1 and examine the system directly.

If it does get stuck the best way to cancel is to use kill -9 on the qemu pid,
lmc will detect that the process died and cleanup.

If lmc didn't handle the cleanup for some reason you can do this:
1. ``sudo umount /tmp/lmc-XXXX`` to unmount the iso from its mountpoint.
2. ``sudo rm -rf /tmp/lmc-XXXX``
3. ``sudo rm /var/tmp/lmc-disk-XXXXX`` to remove the disk image.

Note that lmc uses the lmc- prefix for all of its temporary files and
directories to make it easier to find and clean up leftovers.

The logs from the qemu run are stored in virt-install.log, logs from
livemedia-creator are in livemedia.log and program.log

You can add ``--image-only`` to skip the .iso creation and examine the resulting
disk image. Or you can pass ``--keep-image`` to keep it around after the iso has
been created.

Cleaning up aborted ``--no-virt`` installs can sometimes be accomplished by
running the ``anaconda-cleanup`` script. As of Fedora 18 anaconda is
multi-threaded and it can sometimes become stuck and refuse to exit. When this
happens you can usually clean up by first killing the anaconda process then
running ``anaconda-cleanup``.


Hacking
-------

Development on this will take place as part of the lorax project, and on the
anaconda-devel-list mailing list, and `on github <https://github.com/rhinstaller/lorax>`_

Feedback, enhancements and bugs are welcome.  You can use `bugzilla
<https://bugzilla.redhat.com/enter_bug.cgi?product=Fedora&component=lorax>`_ to
report bugs against the lorax component.