README - pub/scm/linux/kernel/git/maz/vminstall - Git at Google


            
            TiNLV: This is Not LibVirt


            <boring>

            
            Everything here is licensed under the GPLv2. Please see the LICENSE

            
            file at the root of the directory.

            
            </boring>


            This repository contains a bag of scripts I cobbled together to easily

            
            install and run KVM/arm64 UEFI based VMs, acting as a middle ground

            
            between the full complexity of the QEMU command line and the awkward

            
            abstraction provided by LibVirt. The VMs can either be ephemeral (they

            
            are destroyed on exit) or made persistent (though no effort is made to

            
            keep them running).


            There are essentially three major scripts here:


            - runefivm.sh runs a guest, taking a bunch of parameters such as the

            
              disk image, the EFI binary, the number of CPUs...


            - runinstallvm.sh is a very simple wrapper for runefivm.sh, creating

            
              the storage for the VM, and either destroying it once it ends, or

            
              moving it into a safe location for later reuse.


            - build-cd-image is only slightly related to the above. It takes a

            
              Debian or Ubuntu CD/DVD image, and messes with it enough that it

            
              becomes an unattended installer. Now also deals with Fedora DVD

            
              images.


            All the scripts are expected to be run from the vminstall

            
            directory. Yes, this is a braindead expectation, and maybe I'll fix it

            
            at some point. Maybe.


            * runefivm.sh [QEMU args]


              runefivm.sh is entirely driven by environment variables, and

            
              provides more or less sensible defaults when the variable is not

            
              defined:


              - EFI: The EFI (AAVMF) binary. Defaults to

            
                /usr/share/AAVMF/AAVMF_CODE.fd


              - EFIVAR: The per-VM EFI variable container. Defauts to no variables.


              - DISK: The raw file containing the VM mass storage. Defaults to

            
                none.


              - DISKBACKEND: The type of mass storage device exposed to the

            
                guest. Can be either "virtio" (the default) or "nvme".


              - CONSOLE: The console type that is exposed to both the guest and

            
                the user. By default (and for anything other that "virtio"), this

            
                is the usual PL011, connected to stdio. When set to "virtio", an

            
                additional virtio console is created, both the guest's own traffic

            
                and the QEMU monitor is directed to it, while the standard PL011

            
                (which is still present) is redirected to /dev/null. This requires

            
                guest support, whih is why this isn't the default.


              - BUS: The bus on which the virtio devices are created. Can be

            
                "device" or "pci", defaults to "pci".


              - NETIF: The host network interface for this VM. Can be a tap or

            
                macvtap interface (which enables vhost), or "none" (which

            
                indicates that no networking is to be provided). When absent,

            
                userspace networking is provided to the guest. Multiple taps

            
                can be passed as a colon separated list, each of them resulting

            
                in a guest network interface.


              - VIRTTYPE: The QEMU machine type, and comma-separated options

            
                ("virt,virtualization=on" for an EL2 VM, for example). Defaults

            
                to "virt".


              - GIC: The GIC version. Use "2" to force GICv2, "3" for GICv3+ITS,

            
                "3,noits" for GICv3 without an ITS. Defaults to "host", which is

            
                whichever is physically present.

            
                On a system without a GIC, or if you are masochistic, "none" will

            
                use QEMU's GICv2 emulation. It may work, it may not.


              - BIT: 32 or 64, defaults to 64. If using 32, you will have to

            
                provide the path to the EFI path using the EFI variable.


              - CPUS: Number of CPUs, defaults to 4.


              - RAM: Memory size, defaults to 1G.


              - CDROM: A cdrom image provided to the guest, defaults to no cdrom

            
                at all. Can be a file, but also remote resources as as supported by

            
                QEMU (http://mylocalserver:8000/cdimage.iso for example).


              - QEMU: The QEMU binary, defaults to /usr/bin/qemu-system-aarch64.


              Any additional parameter is directly passed to QEMU.


            * runinstallvm.sh [QEMU args]


              All the environment variables used by runefivm.sh are equaly valid

            
              for runinstallvm.sh, with the exception of DISK and EFIVAR (which it

            
              overrides). In addition, it provides the following:


              - FORMAT: Format of the virtio storage, as understood by qemu-img,

            
                such as qcow2. Default is "raw".


              - SIZE: Size of the virtio disk storage. Defaults to 16GB.


              - KEEPME: When provided, create a directory with that name, and

            
                store the VM disk and EFI variables in that directory. Also

            
                generate a "runvm.sh" script in the same directory, which calls

            
                runefivm.sh with the right environment variables.


              - TEMPLATE: A disk image that is used to pre-populates the virtio

            
                disk (useful for things like mini.iso that will pull everything

            
                from the network).


              Any additional parameter is directly passed to QEMU and potentially

            
              preserved in the "runvm.sh" script.


            * build-cd-image source.iso target.iso


              - Debian/Ubuntu: Takes source.iso, adds preseed.cfg to it, add a new

            
                grub entry for unattended installation, repack the whole thing as

            
                target.iso. The only VM user is root, password is root. Don't try

            
                this on a publicly visible box... Depends on the xorriso package

            
                being installed. If the PRESEED_PATH environment variable is

            
                defined, feed this to the installer as part of the kernel command

            
                line. This is useful to point the installer to a remote preseed

            
                file (PRESEED_PATH="url=http://my.remote.box/preseed.cfg").


            Example:


              $ ./build-cd-image ../debian-9.0.0-arm64-netinst.iso ../debian-9.0.0-arm64-netinst-preseed.iso

            
              $ CDROM=../debian-9.0.0-arm64-netinst-preseed.iso RAM=2G KEEPME=stretch ./runinstallvm.sh

            
              [halt the VM]

            
              $ ./stretch/runvm.sh


              - Fedora: Take source.iso, adds a kickstart file and a new grub

            
                entry, and do the same magic to generate an unattended

            
                installer. Note that this will only work with a full DVD image,

            
                and it will probably stop working if packages are missing from the

            
                image. Same user/passwd as above. Only tested with Fedora 26

            
                Server DVD image.
	TiNLV: This is Not LibVirt

	<boring>
	Everything here is licensed under the GPLv2. Please see the LICENSE
	file at the root of the directory.
	</boring>

	This repository contains a bag of scripts I cobbled together to easily
	install and run KVM/arm64 UEFI based VMs, acting as a middle ground
	between the full complexity of the QEMU command line and the awkward
	abstraction provided by LibVirt. The VMs can either be ephemeral (they
	are destroyed on exit) or made persistent (though no effort is made to
	keep them running).

	There are essentially three major scripts here:

	- runefivm.sh runs a guest, taking a bunch of parameters such as the
	disk image, the EFI binary, the number of CPUs...

	- runinstallvm.sh is a very simple wrapper for runefivm.sh, creating
	the storage for the VM, and either destroying it once it ends, or
	moving it into a safe location for later reuse.

	- build-cd-image is only slightly related to the above. It takes a
	Debian or Ubuntu CD/DVD image, and messes with it enough that it
	becomes an unattended installer. Now also deals with Fedora DVD
	images.

	All the scripts are expected to be run from the vminstall
	directory. Yes, this is a braindead expectation, and maybe I'll fix it
	at some point. Maybe.

	* runefivm.sh [QEMU args]

	runefivm.sh is entirely driven by environment variables, and
	provides more or less sensible defaults when the variable is not
	defined:

	- EFI: The EFI (AAVMF) binary. Defaults to
	/usr/share/AAVMF/AAVMF_CODE.fd

	- EFIVAR: The per-VM EFI variable container. Defauts to no variables.

	- DISK: The raw file containing the VM mass storage. Defaults to
	none.

	- DISKBACKEND: The type of mass storage device exposed to the
	guest. Can be either "virtio" (the default) or "nvme".

	- CONSOLE: The console type that is exposed to both the guest and
	the user. By default (and for anything other that "virtio"), this
	is the usual PL011, connected to stdio. When set to "virtio", an
	additional virtio console is created, both the guest's own traffic
	and the QEMU monitor is directed to it, while the standard PL011
	(which is still present) is redirected to /dev/null. This requires
	guest support, whih is why this isn't the default.

	- BUS: The bus on which the virtio devices are created. Can be
	"device" or "pci", defaults to "pci".

	- NETIF: The host network interface for this VM. Can be a tap or
	macvtap interface (which enables vhost), or "none" (which
	indicates that no networking is to be provided). When absent,
	userspace networking is provided to the guest. Multiple taps
	can be passed as a colon separated list, each of them resulting
	in a guest network interface.

	- VIRTTYPE: The QEMU machine type, and comma-separated options
	("virt,virtualization=on" for an EL2 VM, for example). Defaults
	to "virt".

	- GIC: The GIC version. Use "2" to force GICv2, "3" for GICv3+ITS,
	"3,noits" for GICv3 without an ITS. Defaults to "host", which is
	whichever is physically present.
	On a system without a GIC, or if you are masochistic, "none" will
	use QEMU's GICv2 emulation. It may work, it may not.

	- BIT: 32 or 64, defaults to 64. If using 32, you will have to
	provide the path to the EFI path using the EFI variable.

	- CPUS: Number of CPUs, defaults to 4.

	- RAM: Memory size, defaults to 1G.

	- CDROM: A cdrom image provided to the guest, defaults to no cdrom
	at all. Can be a file, but also remote resources as as supported by
	QEMU (http://mylocalserver:8000/cdimage.iso for example).

	- QEMU: The QEMU binary, defaults to /usr/bin/qemu-system-aarch64.

	Any additional parameter is directly passed to QEMU.

	* runinstallvm.sh [QEMU args]

	All the environment variables used by runefivm.sh are equaly valid
	for runinstallvm.sh, with the exception of DISK and EFIVAR (which it
	overrides). In addition, it provides the following:

	- FORMAT: Format of the virtio storage, as understood by qemu-img,
	such as qcow2. Default is "raw".

	- SIZE: Size of the virtio disk storage. Defaults to 16GB.

	- KEEPME: When provided, create a directory with that name, and
	store the VM disk and EFI variables in that directory. Also
	generate a "runvm.sh" script in the same directory, which calls
	runefivm.sh with the right environment variables.

	- TEMPLATE: A disk image that is used to pre-populates the virtio
	disk (useful for things like mini.iso that will pull everything
	from the network).

	Any additional parameter is directly passed to QEMU and potentially
	preserved in the "runvm.sh" script.

	* build-cd-image source.iso target.iso

	- Debian/Ubuntu: Takes source.iso, adds preseed.cfg to it, add a new
	grub entry for unattended installation, repack the whole thing as
	target.iso. The only VM user is root, password is root. Don't try
	this on a publicly visible box... Depends on the xorriso package
	being installed. If the PRESEED_PATH environment variable is
	defined, feed this to the installer as part of the kernel command
	line. This is useful to point the installer to a remote preseed
	file (PRESEED_PATH="url=http://my.remote.box/preseed.cfg").

	Example:

	$ ./build-cd-image ../debian-9.0.0-arm64-netinst.iso ../debian-9.0.0-arm64-netinst-preseed.iso
	$ CDROM=../debian-9.0.0-arm64-netinst-preseed.iso RAM=2G KEEPME=stretch ./runinstallvm.sh
	[halt the VM]
	$ ./stretch/runvm.sh

	- Fedora: Take source.iso, adds a kickstart file and a new grub
	entry, and do the same magic to generate an unattended
	installer. Note that this will only work with a full DVD image,
	and it will probably stop working if packages are missing from the
	image. Same user/passwd as above. Only tested with Fedora 26
	Server DVD image.