Scroll to navigation

VMDEBOOTSTRAP(8) System Manager's Manual VMDEBOOTSTRAP(8)
" "vmdebootstrap"

NAME

vmdebootstrap - install basic Debian system into virtual disk image

PURPOSE

vmdebootstrap is a helper to install basic Debian system into virtual disk image. It wraps debootstrap. You need to run vmdebootstrap as root. If the --verbose option is not used, no output will be sent to the command line. If the --log option is not used, no output will be sent to any log files either.

To use the image, you probably want to create a virtual machine using your preferred virtualization technology, such as kvm or qemu. Configure the virtual machine to use the image you've created. Then start the virtual machine and log into it via its console to configure it. The image has an empty root password and will not have networking configured by default. Set the root password before you configure networking.

SYNOPSIS

$ sudo vmdebootstrap --image=FILE --size=SIZE [--mirror=URL] [--distribution=NAME]


Options

write output to FILE, instead of standard output
report what is going on
opposite of --verbose
put created disk image in FILE
create a disk image of size SIZE (1000000000) in bytes. Suffixes k,K,M,G,T are supported, see qemu-img(1) for more detail.
tar up the disk's contents in FILE
use MIRROR as package source (http://httpredir.debian.org/debian/)
architecture to use (amd64) --- if using an architecture which the host system cannot execute, ensure the --foreign option is also used.
release to use (defaults to stable). The release needs to be a valid Debian or Ubuntu release name or codename.
Supply options and arguments to debootstrap, separated by spaces. e.g. --debootstrapopts="variant=buildd no-check-gpg components=main,contrib". See debootstrap (1) for more information. This option replaces the --variant support in previous versions.
set the directory containing debootstrap scripts.
install PACKAGE onto system
install package in DEB file onto system (not from mirror) - all dependencies must be available in the specified distribution.
do not install a linux package
If --no-kernel is not used and the auto-selection of the linux-image-586 or linux-image-armmp or linux-image-$ARCH package is not suitable, the kernel PACKAGE name can be specified explicitly.
enable DHCP on eth0
set root password
lock root account so they cannot login?
run SCRIPT after setting up system. If the script does not exist in the current working directory, /usr/share/vmdebootstrap/examples/ will be checked as a fallback. The script needs to be executable and is passed the root directory of the debootstrap and the image name as the only arguments. Use chroot if you need to execute binaries within the chroot created by debootstrap.
set name to HOSTNAME (debian)
create USER with PASSWORD. The USERSTRING needs to be of the format: USER/PASSSWORD.
change the owner of the final image from root to the specified user.
configure image to use a serial console (Wheezy only)
(Wheezy only.) Set the command to manage the serial console which will be appended to /etc/inittab. Default is /sbin/getty \-L ttyS0 115200 vt100, resulting in a line:

"S0:23:respawn:/sbin/getty \-L ttyS0 115200 vt100"


install sudo, and if user is created, add them to sudo group
If specified, create a /boot partition of the given size within the image. Debootstrapping will fail if this is too small for the selected kernel package and upgrading such a kernel package is likely to need two or three times the space of the installed kernel.
Filesystem to use for the /boot partition. (default ext2)
Flag to set on the first partition. (default none)
Space to leave at start of the image for bootloader
Filesystem to use for the / (root) partition. (default ext4)
Partition type to use for this image. (default msdos)
If specified, create a swap partition of the given size within the image. Debootstrapping will fail if this results in a root partition which is too small for the selected packages. The minimum swap space is 256MB as the default memory allocation of QEMU is 128MB. A default 1GB image is not likely to have enough space for a swap partition as well.
Path to the binfmt_handler to enable foreign support in debootstrap. e.g. /usr/bin/qemu-arm-static Note: foreign debootstraps may take a significant amount of time to complete and debootstrap will retry five times if packages fail to install by default.
Setup image for UEFI boot
opposite of --use-uefi
Size of EFI System Partition - requires use-uefi
install extlinux (deprecated: default will change in a future release to use grub)
Skip installation of extlinux. Needs grub, a customize script or alternative bootloader to make the image bootable. extlinux is deprecated and this will become the default in a future release.
Run install-mbr (default if extlinux used)
opposite of --mbr
Run mksquashfs against the rootfs using xz compression --- requires squashfs-tools to be installed. The squashfs and other files needed to use the squashfs to make a bootable system will be put into the specified directory. The directory will contain a filesystem.squashfs as well as the top level contents of the boot/ directory. (If using UEFI, the boot/efi directory as well.) By default, mksquashfs is allowed to use all processors which may result in high load. squashfs can also have issues with large root filesystems. These errors can result in truncated files. This is a known bug in squashfs. vmdebootstrap will fail if the squashed filesystem is less than 1MB.
Use the specified mirror and distribution to create a suitable apt source inside the VM. Can be useful if debootstrap fails to create it automatically.
Use the specified mirror inside the image instead of the mirror used to build the image. This is useful if you have a local mirror to make building the image quicker but the image needs to run even if that mirror is not available. Requires --configure-apt
Disable extlinux installation and configure grub2 instead. grub2 will be added to the list of packages to install. update-grub will be called once the debootstrap is complete and grub-install will be called in the image.
Disable installation of acpid if not required, otherwise acpid will be installed if --foreign is not used.
Skip optimizing image for compression and keep a sparse image.
opposite of --sparse
Output a list of package names installed inside the image. Useful if you need to track the relevant source packages used inside the image for licence compliance.
Do not build, just test that the options are valid.
Skip the call to update-initramfs for reasons of speed or practicality.
Convert the final raw image to qcow2 format.
Use Predictable Network Interface Names
Do not use Predictable Network Interface Names using systemd-networkd.



CONFIGURATION FILES AND SETTINGS

write out the entire current configuration
clear list of configuration files to read
add FILE to config files



LOGGING

write log entries to FILE (default is to not write log files at all); use "syslog" to log to system log, or "none" to disable logging.
log at LEVEL, one of debug, info, warning, error, critical, fatal (default: debug).
rotate logs larger than SIZE, zero for never (default: 0)
keep last N logs (10)
set permissions of new log files to MODE (octal; default 0600)



PERFORMANCE

make memory profiling dumps using METHOD, which is one of: none, simple, meliae, or heapy (default: simple)
make memory profiling dumps at least SECONDS apart



NETWORKING

Wheezy support

The --enable-networking option uses the /etc/network/interfaces.d/ source directory, with the default settings for lo and eth0 being added to /etc/network/interfaces.d/setup. Other networking configuration can be specified using a customisation script. Localhost settings would be:

auto lo
iface lo inet loopback


If --enable-dhcp is specified, these settings are also included into /etc/network/interfaces.d/setup:

auto eth0
iface eth0 inet dhcp


In addition, wheezy images do not boot if the roottype is specified as the default of ext4, so vmdebootstrap will fail if a --roottype is not specified or is specified as ext4.

Jessie and later

In addition, systemd in jessie or later introduces PredictableNetworkInterfaceNames which are enabled using the systemd-networkd service. If this option is disabled, traditional interface names (like eth0) will be used and the predictable names masked using udev. Implementing the mask requires updating the initramfs, so the --update-initramfs option must not be disabled.

If DHCP is also enabled, the following configuration is used:

/etc/systemd/network/99-dhcp.network


systemd will use the first available match, so this can be overridden by putting another file into place using the customisation scripts, using a lower sorting filename.

Stretch and later

There is no need to use the --enable-dhcp option when using systemd for networking with stretch or sid. systemd-resolved is enabled instead if systemd-networkd is specified. (--enable-dhcp would simply add an unused entry to /etc/network/interfaces for eth0.)

[Match]
Name=en*
[Network]
DHCP=yes


BOOTLOADERS

Unless the --no-extlinux or --grub options are specified, the image will use extlinux as a boot loader. bootsize is not recommended when using extlinux --- use grub instead.

NOTE:

Unlike grub, extlinux support requires the installation of packages outside the image which are used to install the extlinux bootloader inside the image. extlinux support also involves the use of sync which can cause issues on systems with multiple filesystems mounted, particularly over a network or when building multiple images simultaneously. Therefore, extlinux is deprecated in vmdebootstrap. The default will change in a future release and extlinux support may be dropped once Stretch is released.


extlinux support issues with ext4

VMs using ext4 may not boot when using extlinux - unless the build is performed on Jessie. Builds using ext2 and ext3 work normally.

IMPORTANT:

This problem depends on the external distribution, not the distribution you are trying to build. When building on Jessie, extlinux succeeds but when building on Stretch or Sid, extlinux fails to make a bootable system if the filesystem of that system is ext4. ext2 and ext3 work.


Version 1.6 of vmdebootstrap adds a warning but allows the build to proceed (to allow for the bug to be fixed). Sadly, downgrading the version of extlinux to the version in Jessie does not fix the problem when building on stretch or sid. Hence, vmdebootstrap can only output a warning.

SEE ALSO:

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=833057


Versions of grub2 in wheezy

Grub2 in wheezy can fail to install in the VM, at which point vmdebootstrap will fall back to extlinux. It may still be possible to complete the installation of grub2 after booting the VM as the problem may be related to the need to use loopback devices during the grub-install operation. Details of the error will appear in the vmdebootstrap log file, if enabled with the --log option.

NOTE:

grub-legacy is not supported.


vmdebootstrap also supports EFI. See UEFI.

Use --use-uefi to use grub-efi instead of grub-pc. If the default 5MB is not enough space, use the --esp-size option to specify a different size for the EFI partition. Registered firmware is not supported as it would need to be done after boot. If the system you are creating is for more than just a VM or live image, you will likely need a larger ESP, up to 500MB.

UEFI

UEFI support requires Grub and vmdebootstrap contains a configuration table of the UEFI components required for supported architectures.

There are issues with running UEFI with QEMU on some architectures and a customisation script is available for amd64:

# vmdebootstrap --verbose --image jessie-uefi.img --grub  --use-uefi \


  --customize ./examples/qemu-efi-bochs-drm.sh


vmdebootstrap supports UEFI for images and for squashfs but the necessary behaviour is different. With an image, an ESP vfat partition is created. With squashfs, the EFI files will be copied into an efi/ directory in the squashfs output directory instead.

There is EFI firmware available to use with QEMU when testing images built using the UEFI support, but this software is in Debian non-free due to patent concerns. If you choose to install ovmf to test UEFI builds, a secondary change is also needed to symlink the provided OVMF.fd to the file required by QEMU: bios-256k.bin and then tell QEMU about the location of this file with the -L option:

$ qemu-system-x86_64 -L /usr/share/ovmf/ -machine accel=kvm \


 -m 4096 -smp 2 -drive format=raw,file=test.img


To test the image, also consider using the qemu-wrapper.sh:

$ /usr/share/vmdebootstrap/qemu-wrapper.sh jessie-uefi.img amd64 /usr/share/ovmf/


UBoot

UBoot needs manual configuration via the customisation hook scripts, typically support requires adding u-boot using --package and then copying or manipulating the relevant u-boot files in the customisation script. Examples are included for beaglebone-black.

Some u-boot examples recommend the use of the lba flag on the boot partition, so use the --bootflag option where relevant.

INSTALLATION IMAGES AND VIRTUAL MACHINES

:file:vmdebootstrap is aimed principally at creating virtual machines, not installers or prebuilt installation images. It is possible to create prebuilt installation images for some devices but this depends on the specific device. (A 'prebuilt installation image' is a single image file which can be written to physical media in a single operation and which allows the device to boot directly into a fully installed system --- in a similar way to how a virtual machine would behave.)

vmdebootstrap assumes that all operations take place on a local image file or directory, not a physical block device / removable media.

vmdebootstrap is intended to be used with tools like qemu on the command line to launch a new virtual machine. Not all devices have virtualisation support in hardware.

This has implications for u-boot support in some cases. If the device can support reading the bootloader from a known partition, like the beaglebone-black, then vmdebootstrap can provide space for the bootloader and the image will work as a prebuilt installation image. If the device expects that the bootloader exists at a specific offset and therefore requires that the bootloader is written as an image not as a binary which can be copied into an existing partition, vmdebootstrap is unable to include that bootloader image into the virtual machine image.

The beagleboneblack.sh script in the examples/ directory provides a worked example to create a prebuilt installation image. However, the beagleboneblack itself does not support virtualisation in hardware, so is unable to launch a virtual machine. Other devices, like the Cubietruck or Wandboard need u-boot at a predefined offset but can launch a virtual machine using qemu, so the cubietruck and wandboard6q scripts in the examples/ directory relate to building images for virtual machines once the device is already installed and booted into a suitable kernel.

It is possible to wrap vmdebootstrap in such a way as to prepare a physical block device with a bootloader image and then deploy the bootstrap on top. However, this does require physical media to be inserted and removed each time the wrapper is executed. To do this, use the --tarball option instead of the --image option. Then setup the physical media and bootloader image manually, as required for the device, redefine the partitions to make space for the rootfs, create a filesystem on the physical media and unpack the vmdebootstrap tarball onto that filesystem. Once you have working media, an image can be created using dd to read back from the media to an image file, allowing other media to be written with a single image file.

EXAMPLE

To create an image for the stable release of Debian:

sudo vmdebootstrap --image test.img --size 1G \


   --log test.log --log-level debug --verbose \


   --mirror http://mirror.lan/debian/


To run the test image, make sure it is writeable. Use the --owner option to set mode 0644 for the specified user or use chmod manually:

sudo chmod a+w ./test.img


If --log is also used, consider using --log-mode as well so that the logfile is readable by the owner. By default, the log file permissions are 0o600. The logfile itself will be owned by root.

Execute using qemu, e.g. on amd64 using qemu-system-x86_64:

qemu-system-x86_64 -drive format=raw,file=./test.img


(This loads the image in a new window.) Note the use of -drive file=<img>,format=raw which is needed for newer versions of QEMU.

There is a bin/qemu-wrapper.sh <image> <arch> script for simple calls where the --owner option is used, e.g.:

$ /usr/share/vmdebootstrap/qemu-wrapper.sh jessie.img amd64


There is EFI firmware available to use with QEMU when testing images built using the UEFI support, but this software is in Debian non-free due to patent concerns. If you choose to install ovmf to test UEFI builds, a secondary change is also needed to symlink the provided OVMF.fd to the file required by QEMU: bios-256k.bin and then tell QEMU about the location of this file with the -L option:

$ qemu-system-x86_64 -L /usr/share/ovmf/ -machine accel=kvm \


 -m 4096 -smp 2 -drive format=raw,file=test.img


To use the -nographic option, ensure that the --serial-console option is supplied to vmdebootstrap and use -monitor none when booting the image with QEMU.

For further examples, including u-boot support for beaglebone-black, see /usr/share/vmdebootstrap/examples

NOTES

If you get problems with the bootstrap process, run a similar bootstrap call directly and chroot into the directory to investigate the failure. The actual debootstrap call is part of the vmdebootstrap logfile. The debootstrap logfile, if any, will be copied into your current working directory on error.

debootstrap will download all the apt archive files into the apt cache and does not remove them before starting the configuration of the packages. This can mean that debootstrap can fail due to a lack of space on the device if the VM size is small. vmdebootstrap cleans up the apt cache once debootstrap has finished but this doesn't help if the package unpack or configuration steps use up all of the space in the meantime. Avoid this problem by specifying a larger size for the image.

CAUTION:

if you are also using a separate /boot partition in your options to vmdebootstrap it may well be the boot partition which needs to be enlarged rather than the entire image.


It is advisable to change the mirror in the example scripts to a mirror closer to your location, particularly if you need to do repeated builds. Use the --apt-mirror option to specify the apt mirror to be used inside the image, after boot.

There are two types of examples for ARM devices available with vmdebootstrap: prebuilt installation images (like the beaglebone-black) and virtual machine images (cubietruck and wandboard). ARM devices which do not support hypervisor mode and which also rely on the bootloader being at a specific offset instead of using a normal partition will not be supportable by vmdebootstrap. Similarly, devices which support hypervisor will only be supported using virtual machine images, unless the bootloader can be executed from a normal partition.

If the host device has a limited amount of RAM or simply to use a different TMP directory when preparing the filesystems, set the TMPDIR or TEMP or TMP environment variables, in line with the underlying support in the python tempfile module.

DEVELOPING

Testing vmdebootstrap from git

vmdebootstrap uses yarn for the test suite, available in the cmdtest package. YARN is a scenario testing tool. Scenarios are written in mostly human readable language, however, they are not free form text. For more information on YARN see the homepage:

$ sudo apt -y install cmdtest


All commits must pass at least the fast tests. All merges into master need to pass a full test. All additions of new functionality must add fast and build tests --- fast tests for any new options and build tests which exercise the new functionality. Build tests can add checks for particular support on the machine running the test and skip if not found or add new environment settings to selectively run some build tests instead of all.

If no arguments are given, the full test suite will be executed:

$ yarns/run-tests


WARNING:

Do not run the full test suite if your connection to a Debian mirror is limited or metered. Each build requires a minimum of 2GB free space in tmpfs. A full test takes at least 10 minutes.


When limiting the run to specific tests, each --env option needs to be specified separately:

$ sudo yarns/run-tests --env TESTS=build --env MIRROR=http://mirror/debian


To run a single test, use the --run option to specify the name of the scenario (option can be repeated).

pre-commit

All vmdebootstrap developers need to run the fast tests as a pre-commit hook --- any patches which fail this test will be rejected:

$ ln -s ../../pre-commit.sh .git/hooks/pre-commit


The pre-commit hook just runs the fast tests which do not require sudo.

Fast tests

The fast checks validate the handling of incompatible option arguments:

$ yarns/run-tests --env TESTS=fast


Fast tests typically take a few seconds to run.

Build tests

The slow / build tests build multiple images and use sudo --- a local mirror is strongly recommended.

$ sudo yarns/run-tests --env TESTS=build --env MIRROR=http://mirror/debian


If MIRROR is not specified, a default mirror of http://httpredir.debian.org/debian/ will be used.

LAVA tests

There is an example lava-submit.py script which can be edited to automatically submit QEMU tests to a specified LAVA instance. The images themselves will use local file:// URLs and therefore the lava-dispatcher needs to be installed locally. Configuring LAVA for these tests is a separate topic --- please ask on the vmdebootstrap mailing list.

AUTHOR

Neil Williams

COPYRIGHT

2019, Neil Williams

March 31, 2019 1.11