Building An Image¶
Now that you have diskimage-builder properly installed you can get started by building your first disk image.
Our first image is going to be a bootable vm image using one of the standard supported distribution elements (Ubuntu or Fedora).
The following command will start our image build (distro must be either ‘ubuntu’ or ‘fedora’):
disk-image-create <distro> vm
This will create a qcow2 file ‘image.qcow2’ which can then be booted.
It is important to note that we are passing in a list of elements to disk-image-create in our above command. Elements are how we decide what goes into our image and what modifications will be performed.
Some elements provide a root filesystem, such as the ubuntu or fedora element in our example above, which other elements modify to create our image. At least one of these ‘distro elements’ must be specified when performing an image build. It’s worth pointing out that there are many distro elements (you can even create your own), and even multiples for some of the distros. This is because there are often multiple ways to install a distro which are very different. For example: One distro element might use a cloud image while another uses a package installation tool to build a root filesystem for the same distro.
Other elements modify our image in some way. The ‘vm’ element in our example above ensures that our image has a bootloader properly installed. This is only needed for certain use cases and certain output formats and therefore it is not performed by default.
By default a qcow2 image is created by the disk-image-create command. Other output formats may be specified using the -t <format> argument. Multiple output formats can also be specified by comma separation. The supported output formats are:
When building a tgz image, note that the DIB_GZIP_BIN environment variable can be used to set the path of the gzip executable.
Disk Image Layout¶
The disk image layout (like number of images, partitions, LVM, disk encryption) is something which should be set up during the initial image build: it is mostly not possible to change these things later on.
There are currently two defaults:
When using the
vmelement, an element that provides
block-deviceshould be included. Available
block-device-*elements cover the common case of a single partition that fills up the whole disk and used as root device. Currently there are MBR, GPT and EFI versions. For example, to use a GPT disk you could build with
disk-image-create -o output.qcow vm block-device-gpt ubuntu-minimal
When not using the
vmelement a plain filesystem image, without any partitioning, is created.
If you wish to customise the top-level
file from one of the
block-device-* elements, set the environment
variable DIB_BLOCK_DEVICE_CONFIG. This variable must hold YAML
structured configuration data or be a
file:// URL reference to a
on-disk configuration file.
There are a lot of different options for the different levels. The following sections describe each level in detail.
In general each module that depends on another module has a base element that points to the depending base. Also each module has a name that can be used to reference the module.
Tree-Like vs. Complete Digraph Configuration¶
The configuration is specified as a digraph. Each module is a node; a edge is the relation of the current element to its base.
Example: The tree like notation
mkfs: name: root_fs base: root_part mount: mount_point: /
is exactly the same as writing
mkfs: name: root_fs base: root_part mount: name: mount_root_fs base: root_fs mount_point: /
Non existing name and base entries in the tree notation are automatically generated: the name is the name of the base module prepended by the type-name of the module itself; the base element is automatically set to the parent node in the tree.
In mostly all cases the much simpler tree notation can be used. Nevertheless there are some use cases when the more general digraph notation is needed. Example: when there is the need to combine two or more modules into one new, like combining a couple of physical volumes into one volume group.
Tree and digraph notations can be mixed as needed in a configuration.
To provide an interface towards the existing elements, there are currently three fixed keys used - which are not configurable:
root-label: this is the label of the block device that is mounted at /.
image-block-partition: if there is a block device with the name root this is used else the block device with the name image0 is used.
image-path: the path of the image that contains the root file system is taken from the image0.
Module: Local Loop¶
This module generates a local image file and uses the loop device to create a block device from it. The symbolic name for this module is local_loop.
(mandatory) The name of the image. This is used as the name for the image in the file system and also as a symbolic name to be able to reference this image (e.g. to create a partition table on this disk).
(optional) The size of the disk. The size can be expressed using unit names like TiB (1024^4 bytes) or GB (1000^3 bytes). Examples: 2.5GiB, 12KB. If the size is not specified here, the size as given to disk-image-create (–image-size) or the automatically computed size is used.
(optional) The directory where the image is created.
local_loop: name: image0 local_loop: name: data_image size: 7.5GiB directory: /var/tmp
This creates two image files and uses the loop device to use them as block devices. One image file called image0 is created with default size in the default temp directory. The second image has the size of 7.5GiB and is created in the /var/tmp folder.
This module generates partitions on existing block devices. This means that it is possible to take any kind of block device (e.g. LVM, encrypted, …) and create partition information in it.
The symbolic name for this module is partitioning.
It is possible to create primary or logical partitions or a mix of them. The numbering of the primary partitions will start at 1, e.g. /dev/vda1; logical partitions will typically start with 5, e.g. /dev/vda5 for the first partition, /dev/vda6 for the second and so on.
The number of logical partitions created by this module is theoretical unlimited and it was tested with more than 1000 partitions inside one block device. Nevertheless the Linux kernel and different tools (like parted, sfdisk, fdisk) have some default maximum number of partitions that they can handle. Please consult the documentation of the appropriate software you plan to use and adapt the number of partitions.
Partitions are created in the order they are configured. Primary partitions - if needed - must be first in the list.
GPT partitioning requires the
sgdisk tool to be available.
There are the following key / value pairs to define one partition table:
(mandatory) The base device to create the partitions in.
(mandatory) Possible values: ‘mbr’, ‘gpt’ Configure use of either the Master Boot Record (MBR) or GUID Partition Table (GPT) formats
(optional - default value ‘1MiB’; MBR only) Set the alignment of the partition. This must be a multiple of the block size (i.e. 512 bytes). The default of 1MiB (~ 2048 * 512 bytes blocks) is the default for modern systems and known to perform well on a wide range of targets. For each partition there might be some space that is not used - which is align - 512 bytes. For the default of 1MiB exactly 1048064 bytes (= 1 MiB - 512 byte) are not used in the partition itself. Please note that if a boot loader should be written to the disk or partition, there is a need for some space. E.g. grub needs 63 * 512 byte blocks between the MBR and the start of the partition data; this means when grub will be installed, the align must be set at least to 64 * 512 byte = 32 KiB.
(mandatory) A list of dictionaries. Each dictionary describes one partition.
The following key / value pairs can be given for each partition:
(mandatory) The name of the partition. With the help of this name, the partition can later be referenced, e.g. when creating a file system.
(optional) List of flags for the partition. Default: empty. Possible values:
- boot (MBR only)
Sets the boot flag for the partition
- primary (MBR only)
Partition should be a primary partition. If not set a logical partition will be created.
(mandatory) The size of the partition. The size can either be an absolute number using units like 10GiB or 1.75TB or relative (percentage) numbers: in the later case the size is calculated based on the remaining free space.
- type (optional)
The partition type stored in the MBR or GPT partition table entry.
For MBR the default value is ‘0x83’ (Linux Default partition). Any valid one byte hexadecimal value may be specified here.
For GPT the default value is ‘8300’ (Linux Default partition). Any valid two byte hexadecimal value may be specified here. Due to
sgdiskleading ‘0x’ should not be used.
- partitioning: base: image0 label: mbr partitions: - name: part-01 flags: [ boot ] size: 1GiB - name: part-02 size: 100% - partitioning: base: data_image label: mbr partitions: - name: data0 size: 33% - name: data1 size: 50% - name: data2 size: 100% - partitioning: base: gpt_image label: gpt partitions: - name: ESP type: EF00 size: 16MiB - name: data1 size: 1GiB - name: lvmdata type: 8E00 size: 100%
On the image0 two partitions are created. The size of the first is 1GiB, the second uses the remaining free space. On the data_image three partitions are created: all are about 1/3 of the disk size. On the gpt_image three partitions are created: 16MiB one for EFI bootloader, 1GiB Linux filesystem one and rest of disk will be used for LVM partition.
This module generates volumes on existing block devices. This means that it is possible to take any previous created partition, and create volumes information in it.
The symbolic name for this module is lvm.
There are the following key / value pairs to define one set of volumes:
(mandatory) A list of dictionaries. Each dictionary describes one physical volume.
(mandatory) A list of dictionaries. Each dictionary describes one volume group.
(mandatory) A list of dictionaries. Each dictionary describes one logical volume.
The following key / value pairs can be given for each pvs:
(mandatory) The name of the physical volume. With the help of this name, the physical volume can later be referenced, e.g. when creating a volume group.
(mandatory) The name of the partition where the physical volume needs to be created.
(optional) List of options for the physical volume. It can contain any option supported by the pvcreate command.
The following key / value pairs can be given for each vgs:
(mandatory) The name of the volume group. With the help of this name, the volume group can later be referenced, e.g. when creating a logical volume.
(mandatory) The name(s) of the physical volumes where the volume groups needs to be created. As a volume group can be created on one or more physical volumes, this needs to be a list.
(optional) List of options for the volume group. It can contain any option supported by the vgcreate command.
The following key / value pairs can be given for each lvs:
(mandatory) The name of the logical volume. With the help of this name, the logical volume can later be referenced, e.g. when creating a filesystem.
(mandatory) The name of the volume group where the logical volume needs to be created.
(optional) The exact size of the volume to be created. It accepts the same syntax as the -L flag of the lvcreate command.
(optional) The relative size in extents of the volume to be created. It accepts the same syntax as the -l flag of the lvcreate command. Either size or extents need to be passed on the volume creation.
(optional) List of options for the logical volume. It can contain any option supported by the lvcreate command.
- lvm: name: lvm pvs: - name: pv options: ["--force"] base: root vgs: - name: vg base: ["pv"] options: ["--force"] lvs: - name: lv_root base: vg size: 1800M - name: lv_tmp base: vg size: 100M - name: lv_var base: vg size: 500M - name: lv_log base: vg size: 100M - name: lv_audit base: vg size: 100M - name: lv_home base: vg size: 200M
On the root partition a physical volume is created. On that physical volume, a volume group is created. On top of this volume group, six logical volumes are created.
Please note that in order to build images that are bootable using volumes, your ramdisk image will need to have that support. If the image you are using does not have it, you can add the needed modules and regenerate it, by including the dracut-regenerate element when building it.
This module creates file systems on the block device given as base. The following key / value pairs can be given:
(mandatory) The name of the block device where the filesystem will be created on.
(mandatory) The name of the partition. This can be used to reference (e.g. mounting) the filesystem.
(mandatory) The type of the filesystem, like ext4 or xfs.
(optional - defaults to the name) The label of the filesystem. This can be used e.g. by grub or in the fstab.
(optional - defaults to empty list) Options that will passed to the mkfs command.
(optional - no default / not used if not givem) The UUID of the filesystem. Not all file systems might support this. Currently there is support for ext2, ext3, ext4 and xfs.
- mkfs: name: mkfs_root base: root type: ext4 label: cloudimage-root uuid: b733f302-0336-49c0-85f2-38ca109e8bdb opts: "-i 16384"
This module mounts a filesystem. The options are:
(mandatory) The name of the filesystem that will be mounted.
(mandatory) The name of the mount point. This can be used for reference the mount (e.g. creating the fstab).
(mandatory) The mount point of the filesystem.
There is no need to list the mount points in the correct order: an algorithm will automatically detect the mount order.
- mount: name: root_mnt base: mkfs_root mount_point: /
This module creates fstab entries. The following options exists. For details please consult the fstab man page.
(mandatory) The name of the mount point that will be written to fstab.
(mandatory) The name of the fstab entry. This can be used later on as reference - and is currently unused.
(optional, defaults to default) Special mount options can be given. This is used as the fourth field in the fstab entry.
(optional, defaults to 0 - don’t dump) This is passed to dump to determine which filesystem should be dumped. This is used as the fifth field in the fstab entry.
(optional, defaults to 2) Determines the order to run fsck. Please note that this should be set to 1 for the root file system. This is used as the sixth field in the fstab entry.
- fstab: name: var_log_fstab base: var_log_mnt options: nodev,nosuid dump-freq: 2
Legacy global filesystem configuration¶
disk-image-create tool has a number of historic global
disk-related command-line options which are maintained for backwards
compatibility. These options are merged as necessary by the
block-device layer into the active configuration. If you are using
more complicated block-device layouts with multiple partitions, you
may need to take into account the special behaviour described below.
local_loop module will take it’s default size from the
The size of loopback device which the image will be generated in, in gigabytes. If this is left unset, the size will be calculated from the on-disk size of the image and then scaled up by a fixed 60% factor. Can also set
Extra space to add when automatically calculating image size, in megabytes. This overrides the default 60% scale up as described above for
--image-size. Can also set
The special node named
mkfs_root is affected by the following;
this reflects that the standard layout has only a single root
partition so the options are, in effect, global for the default
configuration. Note that if you are using multiple partitions,
settings such as
--mkfs-options will not apply to other
The file-system type for the
mkfs_root node is set by the
FS_TYPE environment variable, and defaults to
should also work. There is no command-line argument for this.
The following options also affect the
Options passed to mkfs when making the root partition. For
ext4partitions, this by default sets a 4k byte-to-inode ratio (see below) and a default journal size of 64MiB. Note
--mkfs-optionsare options passed to the mfks driver (i.e.
mkfs.ext4) rather than
mkfsitself (i.e. arguments come after the initial
mkfs -t <fstype>argument). You also need to be careful with quoting. Can also set
disk-image-createuses a 4k byte-to-inode ratio when creating the filesystem in the image. This allows large ‘whole-system’ images to utilize several TB disks without exhausting inodes. In contrast, when creating images intended for tenant instances, this ratio consumes more disk space than an end-user would expect (e.g. a 50GB root disk has 47GB available). If the image is intended to run within a tens to hundrededs of gigabyte disk, setting the byte-to-inode ratio to the ext4 default of 16k will allow for more usable space on the instance. The default can be overridden by passing
'-i 16384'as a
Only valid for
FS_TYPE==ext4. This value set the filesystem journal size in MB; overriding the default of 64MiB. Note the image size will be grown to fit the journal, unless
DIB_IMAGE_SIZEis explicitly set. Can also set
Only valid for
FS_TYPE==ext4; this value sets the maximum filesystem blocks when resizing. Can also set
The file-system label specified when creating the root file system. Defaults to
xfs. Can also set
If you have 4GB of available physical RAM (as reported by /proc/meminfo MemTotal), or more, diskimage-builder will create a tmpfs mount to build the image in. This will improve image build time by building it in RAM. By default, the tmpfs file system uses 50% of the available RAM. Therefore, the RAM should be at least the double of the minimum tmpfs size required.
For larger images, when no sufficient amount of RAM is available, tmpfs can be disabled completely by passing –no-tmpfs to disk-image-create. ramdisk-image-create builds a regular image and then within that image creates ramdisk.
If tmpfs is not used, you will need enough room in /tmp to store two uncompressed cloud images. If tmpfs is used, you would still need /tmp space for one uncompressed cloud image and about 20% of that image for working files.
To ensure elements can access the network,
/etc/resolv.conf within the chroot with a copy of the
host’s file early in the image creation process.
/etc/resolv.conf can be controlled in a number of ways.
If, during the build, the
/etc/resolv.conf file within the chroot
is replaced with a symlink, this will be retained in the final image
1. If the file is marked immutable, it will also not be touched.
This somewhat odd case was added for installation of the
resolvconfpackage, which replaces
/etc/resolv.confwith a symlink to it’s version. Depending on its contents, and what comes after the installation in the build, this mostly works.
If you would like specific contents within the final
/etc/resolv.conf you can place them into
/etc/resolv.conf.ORIG during the build. As one of the final
steps, this file will be
Chosing an Architecture¶
If needed you can specify an override the architecture selection by passing a
-a argument like:
disk-image-create -a <arch> ...
Notes about PowerPC Architectures¶
PowerPC can operate in either Big or Little Endian mode.
always refers to Big Endian operation. When running in little endian
mode it can be referred to as
ppc64el refers to a
.deb based distribution
ppc64le refers to a
.rpm based distribution.
Regardless of the distribution the kernel architecture is always
Notes about s390x (z Systems) Architecture¶
Images for s390x can only be build on s390x hosts. Trying to build it with the architecture override on other architecture will cause the build to fail.