libvirt/docs/formatdomain.rst
Boris Fiuczynski 3bf02acdc5 qemu: allow use of async teardown in domain
Asynchronous teardown can be specified if the QEMU binary supports it by
adding in the domain XML

  <features>
    ...
    <async-teardown enabled='yes|no'/>
    ...
  </features>

By default this new feature is disabled.

Signed-off-by: Boris Fiuczynski <fiuczy@linux.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
2023-07-10 13:28:17 +02:00

375 KiB

Domain XML format

This section describes the XML format used to represent domains, there are variations on the format based on the kind of domains run and the options used to launch them. For hypervisor specific details consult the driver docs

Element and attribute overview

The root element required for all virtual machines is named domain. It has two attributes, the type specifies the hypervisor used for running the domain. The allowed values are driver specific, but include "xen", "kvm", "hvf" (since 8.1.0 and QEMU 2.12), "qemu" and "lxc". The second attribute is id which is a unique integer identifier for the running guest machine. Inactive machines have no id value.

General metadata

<domain type='kvm' id='1'>
  <name>MyGuest</name>
  <uuid>4dea22b3-1d52-d8f3-2516-782e98ab3fa0</uuid>
  <genid>43dc0cf8-809b-4adb-9bea-a9abb5f3d90e</genid>
  <title>A short description - title - of the domain</title>
  <description>Some human readable description</description>
  <metadata>
    <app1:foo xmlns:app1="http://app1.org/app1/">..</app1:foo>
    <app2:bar xmlns:app2="http://app1.org/app2/">..</app2:bar>
  </metadata>
  ...
name

The content of the name element provides a short name for the virtual machine. This name should consist only of alphanumeric characters and is required to be unique within the scope of a single host. It is often used to form the filename for storing the persistent configuration file. Since 0.0.1

uuid

The content of the uuid element provides a globally unique identifier for the virtual machine. The format must be RFC 4122 compliant, eg 3e3fce45-4f53-4fa7-bb32-11f34168b82b. If omitted when defining/creating a new machine, a random UUID is generated. It is also possible to provide the UUID via a SMBIOS System Information specification. Since 0.0.1, sysinfo since 0.8.7

genid

Since 4.4.0 , the genid element can be used to add a Virtual Machine Generation ID which exposes a 128-bit, cryptographically random, integer value identifier, referred to as a Globally Unique Identifier (GUID) using the same format as the uuid. The value is used to help notify the guest operating system when the virtual machine is re-executing something that has already executed before, such as:

  • VM starts executing a snapshot
  • VM is recovered from backup
  • VM is failover in a disaster recovery environment
  • VM is imported, copied, or cloned

The guest operating system notices the change and is then able to react as appropriate by marking its copies of distributed databases as dirty, re-initializing its random number generator, etc.

The libvirt XML parser will accept both a provided GUID value or just <genid/> in which case a GUID will be generated and saved in the XML. For the transitions such as above, libvirt will change the GUID before re-executing.

title

The optional element title provides space for a short description of the domain. The title should not contain any newlines. Since 0.9.10 .

description

The content of the description element provides a human readable description of the virtual machine. This data is not used by libvirt in any way, it can contain any information the user wants. Since 0.7.2

metadata

The metadata node can be used by applications to store custom metadata in the form of XML nodes/trees. Applications must use custom namespaces on their XML nodes/trees, with only one top-level element per namespace (if the application needs structure, they should have sub-elements to their namespace element). Since 0.9.10

Operating system booting

There are a number of different ways to boot virtual machines each with their own pros and cons.

BIOS bootloader

Booting via the BIOS is available for hypervisors supporting full virtualization. In this case the BIOS has a boot order priority (floppy, harddisk, cdrom, network) determining where to obtain/find the boot image.

<!-- Xen with fullvirt loader -->
...
<os>
  <type>hvm</type>
  <loader>/usr/lib/xen/boot/hvmloader</loader>
  <boot dev='hd'/>
</os>
...

<!-- QEMU with default firmware, serial console and SMBIOS -->
...
<os>
  <type>hvm</type>
  <boot dev='cdrom'/>
  <bootmenu enable='yes' timeout='3000'/>
  <smbios mode='sysinfo'/>
  <bios useserial='yes' rebootTimeout='0'/>
</os>
...

<!-- QEMU with UEFI manual firmware and secure boot -->
...
<os>
  <type>hvm</type>
  <loader readonly='yes' secure='yes' type='pflash'>/usr/share/OVMF/OVMF_CODE.fd</loader>
  <nvram template='/usr/share/OVMF/OVMF_VARS.fd'>/var/lib/libvirt/nvram/guest_VARS.fd</nvram>
  <boot dev='hd'/>
</os>
...

<!-- QEMU with UEFI manual firmware, secure boot and with NVRAM type 'file'-->
...
<os>
  <type>hvm</type>
  <loader readonly='yes' secure='yes' type='pflash'>/usr/share/OVMF/OVMF_CODE.fd</loader>
  <nvram type='file' template='/usr/share/OVMF/OVMF_VARS.fd'>
    <source file='/var/lib/libvirt/nvram/guest_VARS.fd'/>
  </nvram>
  <boot dev='hd'/>
</os>
...

<!-- QEMU with UEFI manual firmware, secure boot and with network backed NVRAM'-->
...
<os>
  <type>hvm</type>
  <loader readonly='yes' secure='yes' type='pflash'>/usr/share/OVMF/OVMF_CODE.fd</loader>
  <nvram type='network'>
    <source protocol='iscsi' name='iqn.2013-07.com.example:iscsi-nopool/0'>
      <host name='example.com' port='6000'/>
      <auth username='myname'>
        <secret type='iscsi' usage='mycluster_myname'/>
      </auth>
    </source>
  </nvram>
  <boot dev='hd'/>
</os>
...

<!-- QEMU with automatic UEFI firmware and secure boot -->
...
<os firmware='efi'>
  <type>hvm</type>
  <loader secure='yes'/>
  <boot dev='hd'/>
</os>
...

<!-- QEMU with automatic UEFI stateless firmware for AMD SEV -->
...
<os firmware='efi'>
  <type>hvm</type>
  <loader stateless='yes'/>
  <boot dev='hd'/>
</os>
...
firmware

The firmware attribute allows management applications to automatically fill <loader/> and <nvram/> elements and possibly enable some features required by selected firmware. Accepted values are bios and efi. The selection process scans for files describing installed firmware images in specified location and uses the most specific one which fulfills domain requirements. The locations in order of preference (from generic to most specific one) are:

  • /usr/share/qemu/firmware
  • /etc/qemu/firmware
  • $XDG_CONFIG_HOME/qemu/firmware

For more information refer to firmware metadata specification as described in docs/interop/firmware.json in QEMU repository. Regular users do not need to bother. Since 5.2.0 (QEMU and KVM only) For VMware guests, this is set to efi when the guest uses UEFI, and it is not set when using BIOS. Since 5.3.0 (VMware ESX and Workstation/Player)

type

The content of the type element specifies the type of operating system to be booted in the virtual machine. hvm indicates that the OS is one designed to run on bare metal, so requires full virtualization. linux (badly named!) refers to an OS that supports the Xen 3 hypervisor guest ABI. There are also two optional attributes, arch specifying the CPU architecture to virtualization, and machine referring to the machine type. The Capabilities XML provides details on allowed values for these. If arch is omitted then for most hypervisor drivers, the host native arch will be chosen. For the test, ESX and VMWare hypervisor drivers, however, the i686 arch will always be chosen even on an x86_64 host. Since 0.0.1

firmware

Since 7.2.0 QEMU/KVM only

When using firmware auto-selection there are different features enabled in the firmwares. The list of features can be used to limit what firmware should be automatically selected for the VM. The list of features can be specified using zero or more feature elements. Libvirt will take into consideration only the listed features and ignore the rest when selecting the firmware.

feature

The list of mandatory attributes:

  • enabled (accepted values are yes and no) is used to tell libvirt if the feature must be enabled or not in the automatically selected firmware
  • name the name of the feature, the list of the features:
    • enrolled-keys whether the selected nvram template has default certificate enrolled. Firmware with Secure Boot feature but without enrolled keys will successfully boot non-signed binaries as well. Valid only for firmwares with Secure Boot feature.
    • secure-boot whether the firmware implements UEFI Secure boot feature.
loader

The optional loader tag refers to a firmware blob, which is specified by absolute path, used to assist the domain creation process. It is used by Xen fully virtualized domains as well as setting the QEMU BIOS file path for QEMU/KVM domains. Xen since 0.1.0, QEMU/KVM since 0.9.12 Then, since 1.2.8 it's possible for the element to have two optional attributes: readonly (accepted values are yes and no) to reflect the fact that the image should be writable or read-only. The second attribute type accepts values rom and pflash. It tells the hypervisor where in the guest memory the file should be mapped. For instance, if the loader path points to an UEFI image, type should be pflash. Moreover, some firmwares may implement the Secure boot feature. Attribute secure can be used to tell the hypervisor that the firmware is capable of Secure Boot feature. It cannot be used to enable or disable the feature itself in the firmware. Since 2.1.0. If the loader is marked as read-only, then with UEFI it is assumed that there will be a writable NVRAM available. In some cases, however, it may be desirable for the loader to run without any NVRAM, discarding any config changes on shutdown. The stateless flag (Since 8.6.0) can be used to control this behaviour, when set to yes NVRAM will never be created.

When firmware autoselection is enabled, the format attribute can be used to tell libvirt to only consider firmware builds that are in a specific format. Supported values are raw and qcow2. Since 9.2.0 (QEMU only)

nvram

Some UEFI firmwares may want to use a non-volatile memory to store some variables. In the host, this is represented as a file and the absolute path to the file is stored in this element. Moreover, when the domain is started up libvirt copies so called master NVRAM store file defined in qemu.conf. If needed, the template attribute can be used to per domain override map of master NVRAM stores from the config file. Note, that for transient domains if the NVRAM file has been created by libvirt it is left behind and it is management application's responsibility to save and remove file (if needed to be persistent). Since 1.2.8

Since 8.5.0, it's possible for the element to have type attribute (accepts values file, block and network) in that case the NVRAM storage is described by a <source> sub-element with the same syntax as disk's source. See Hard drives, floppy disks, CDROMs.

Note: network backed NVRAM the variables are not instantiated from the template and it's user's responsibility to provide a valid NVRAM image.

This element supports a format attribute, which has the same semantics as the attribute of the same name for the <loader> element. Since 9.2.0 (QEMU only)

It is not valid to provide this element if the loader is marked as stateless.

boot

The dev attribute takes one of the values "fd", "hd", "cdrom" or "network" and is used to specify the next boot device to consider. The boot element can be repeated multiple times to setup a priority list of boot devices to try in turn. Multiple devices of the same type are sorted according to their targets while preserving the order of buses. After defining the domain, its XML configuration returned by libvirt (through virDomainGetXMLDesc) lists devices in the sorted order. Once sorted, the first device is marked as bootable. Thus, e.g., a domain configured to boot from "hd" with vdb, hda, vda, and hdc disks assigned to it will boot from vda (the sorted list is vda, vdb, hda, hdc). Similar domain with hdc, vda, vdb, and hda disks will boot from hda (sorted disks are: hda, hdc, vda, vdb). It can be tricky to configure in the desired way, which is why per-device boot elements (see Hard drives, floppy disks, CDROMs, Network interfaces, and Host device assignment sections below) were introduced and they are the preferred way providing full control over booting order. The boot element and per-device boot elements are mutually exclusive. Since 0.1.3, per-device boot since 0.8.8

smbios

How to populate SMBIOS information visible in the guest. The mode attribute must be specified, and is either "emulate" (let the hypervisor generate all values), "host" (copy all of Block 0 and Block 1, except for the UUID, from the host's SMBIOS values; the virConnectGetSysinfo call can be used to see what values are copied), or "sysinfo" (use the values in the SMBIOS System Information element). If not specified, the hypervisor default is used. Since 0.8.7

Up till here the BIOS/UEFI configuration knobs are generic enough to be implemented by majority (if not all) firmwares out there. However, from now on not every single setting makes sense to all firmwares. For instance, rebootTimeout doesn't make sense for UEFI, useserial might not be usable with a BIOS firmware that doesn't produce any output onto serial line, etc. Moreover, firmwares don't usually export their capabilities for libvirt (or users) to check. And the set of their capabilities can change with every new release. Hence users are advised to try the settings they use before relying on them in production.

bootmenu

Whether or not to enable an interactive boot menu prompt on guest startup. The enable attribute can be either "yes" or "no". If not specified, the hypervisor default is used. Since 0.8.3 Additional attribute timeout takes the number of milliseconds the boot menu should wait until it times out. Allowed values are numbers in range [0, 65535] inclusive and it is ignored unless enable is set to "yes". Since 1.2.8

bios

This element has attribute useserial with possible values yes or no. It enables or disables Serial Graphics Adapter which allows users to see BIOS messages on a serial port. Therefore, one needs to have Serial port defined. Since 0.9.4 . Since 0.10.2 (QEMU only) there is another attribute, rebootTimeout that controls whether and after how long the guest should start booting again in case the boot fails (according to BIOS). The value is in milliseconds with maximum of 65535 and special value -1 disables the reboot.

Host bootloader

Hypervisors employing paravirtualization do not usually emulate a BIOS, and instead the host is responsible to kicking off the operating system boot. This may use a pseudo-bootloader in the host to provide an interface to choose a kernel for the guest. An example is pygrub with Xen. The Bhyve hypervisor also uses a host bootloader, either bhyveload or grub-bhyve.

...
<bootloader>/usr/bin/pygrub</bootloader>
<bootloader_args>--append single</bootloader_args>
...
bootloader

The content of the bootloader element provides a fully qualified path to the bootloader executable in the host OS. This bootloader will be run to choose which kernel to boot. The required output of the bootloader is dependent on the hypervisor in use. Since 0.1.0

bootloader_args

The optional bootloader_args element allows command line arguments to be passed to the bootloader. Since 0.2.3

Direct kernel boot

When installing a new guest OS it is often useful to boot directly from a kernel and initrd stored in the host OS, allowing command line arguments to be passed directly to the installer. This capability is usually available for both para and full virtualized guests.

...
<os>
  <type>hvm</type>
  <loader>/usr/lib/xen/boot/hvmloader</loader>
  <kernel>/root/f8-i386-vmlinuz</kernel>
  <initrd>/root/f8-i386-initrd</initrd>
  <cmdline>console=ttyS0 ks=http://example.com/f8-i386/os/</cmdline>
  <dtb>/root/ppc.dtb</dtb>
  <acpi>
    <table type='slic'>/path/to/slic.dat</table>
  </acpi>
</os>
...
type

This element has the same semantics as described earlier in the BIOS bootloader section.

loader

This element has the same semantics as described earlier in the BIOS bootloader section.

kernel

The contents of this element specify the fully-qualified path to the kernel image in the host OS.

initrd

The contents of this element specify the fully-qualified path to the (optional) ramdisk image in the host OS.

cmdline

The contents of this element specify arguments to be passed to the kernel (or installer) at boot time. This is often used to specify an alternate primary console (eg serial port), or the installation media source / kickstart file

dtb

The contents of this element specify the fully-qualified path to the (optional) device tree binary (dtb) image in the host OS. Since 1.0.4

acpi

The table element contains a fully-qualified path to the ACPI table. The type attribute contains the ACPI table type (currently only slic is supported) Since 1.3.5 (QEMU) Since 5.9.0 (Xen)

Container boot

When booting a domain using container based virtualization, instead of a kernel / boot image, a path to the init binary is required, using the init element. By default this will be launched with no arguments. To specify the initial argv, use the initarg element, repeated as many time as is required. The cmdline element, if set will be used to provide an equivalent to /proc/cmdline but will not affect init argv.

To set environment variables, use the initenv element, one for each variable.

To set a custom work directory for the init, use the initdir element.

To run the init command as a given user or group, use the inituser or initgroup elements respectively. Both elements can be provided either a user (resp. group) id or a name. Prefixing the user or group id with a + will force it to be considered like a numeric value. Without this, it will be first tried as a user or group name.

<os>
  <type arch='x86_64'>exe</type>
  <init>/bin/systemd</init>
  <initarg>--unit</initarg>
  <initarg>emergency.service</initarg>
  <initenv name='MYENV'>some value</initenv>
  <initdir>/my/custom/cwd</initdir>
  <inituser>tester</inituser>
  <initgroup>1000</initgroup>
</os>

If you want to enable user namespace, set the idmap element. The uid and gid elements have three attributes:

start

First user ID in container. It must be '0'.

target

The first user ID in container will be mapped to this target user ID in host.

count

How many users in container are allowed to map to host's user.

<idmap>
  <uid start='0' target='1000' count='10'/>
  <gid start='0' target='1000' count='10'/>
</idmap>

SMBIOS System Information

Some hypervisors allow control over what system information is presented to the guest (for example, SMBIOS fields can be populated by a hypervisor and inspected via the dmidecode command in the guest). The optional sysinfo element covers all such categories of information. Since 0.8.7

...
<os>
  <smbios mode='sysinfo'/>
  ...
</os>
<sysinfo type='smbios'>
  <bios>
    <entry name='vendor'>LENOVO</entry>
  </bios>
  <system>
    <entry name='manufacturer'>Fedora</entry>
    <entry name='product'>Virt-Manager</entry>
    <entry name='version'>0.9.4</entry>
  </system>
  <baseBoard>
    <entry name='manufacturer'>LENOVO</entry>
    <entry name='product'>20BE0061MC</entry>
    <entry name='version'>0B98401 Pro</entry>
    <entry name='serial'>W1KS427111E</entry>
  </baseBoard>
  <chassis>
    <entry name='manufacturer'>Dell Inc.</entry>
    <entry name='version'>2.12</entry>
    <entry name='serial'>65X0XF2</entry>
    <entry name='asset'>40000101</entry>
    <entry name='sku'>Type3Sku1</entry>
  </chassis>
  <oemStrings>
    <entry>myappname:some arbitrary data</entry>
    <entry>otherappname:more arbitrary data</entry>
  </oemStrings>
</sysinfo>
<sysinfo type='fwcfg'>
  <entry name='opt/com.example/name'>example value</entry>
  <entry name='opt/com.coreos/config' file='/tmp/provision.ign'/>
</sysinfo>
...

The sysinfo element has a mandatory attribute type that determine the layout of sub-elements, with supported values of:

smbios

Sub-elements call out specific SMBIOS values, which will affect the guest if used in conjunction with the smbios sub-element of the os element (see Operating system booting). Each sub-element of sysinfo names a SMBIOS block, and within those elements can be a list of entry elements that describe a field within the block. The following blocks and entries are recognized:

bios

This is block 0 of SMBIOS, with entry names drawn from:

vendor

BIOS Vendor's Name

version

BIOS Version

date

BIOS release date. If supplied, is in either mm/dd/yy or mm/dd/yyyy format. If the year portion of the string is two digits, the year is assumed to be 19yy.

release

System BIOS Major and Minor release number values concatenated together as one string separated by a period, for example, 10.22.

system

This is block 1 of SMBIOS, with entry names drawn from:

manufacturer

Manufacturer of BIOS

product

Product Name

version

Version of the product

serial

Serial number

uuid

Universal Unique ID number. If this entry is provided alongside a top-level uuid element (see General metadata), then the two values must match.

sku

SKU number to identify a particular configuration.

family

Identify the family a particular computer belongs to.

baseBoard

This is block 2 of SMBIOS. This element can be repeated multiple times to describe all the base boards; however, not all hypervisors necessarily support the repetition. The element can have the following children:

manufacturer

Manufacturer of BIOS

product

Product Name

version

Version of the product

serial

Serial number

asset

Asset tag

location

Location in chassis

NB: Incorrectly supplied entries for the bios, system or baseBoard blocks will be ignored without error. Other than uuid validation and date format checking, all values are passed as strings to the hypervisor driver.

chassis

Since 4.1.0, this is block 3 of SMBIOS, with entry names drawn from:

manufacturer

Manufacturer of Chassis

version

Version of the Chassis

serial

Serial number

asset

Asset tag

sku

SKU number

oemStrings

This is block 11 of SMBIOS. This element should appear once and can have multiple entry child elements, each providing arbitrary string data. There are no restrictions on what data can be provided in the entries, however, if the data is intended to be consumed by an application in the guest, it is recommended to use the application name as a prefix in the string. ( Since 4.1.0 )

fwcfg

Some hypervisors provide unified way to tweak how firmware configures itself, or may contain tables to be installed for the guest OS, for instance boot order, ACPI, SMBIOS, etc.

It even allows users to define their own config blobs. In case of QEMU, these then appear under domain's sysfs (if the guest kernel has FW_CFG_SYSFS config option enabled), under /sys/firmware/qemu_fw_cfg. Note, that these values apply regardless the <smbios/> mode under <os/>. Since 6.5.0

Please note that because of limited number of data slots use of fwcfg is strongly discouraged and <oemStrings/> should be used instead.

<sysinfo type='fwcfg'>
  <entry name='opt/com.example/name'>example value</entry>
  <entry name='opt/com.example/config' file='/tmp/provision.ign'/>
</sysinfo>

The sysinfo element can have multiple entry child elements. Each element then has mandatory name attribute, which defines the name of the blob and must begin with opt/ and to avoid clashing with other names is advised to be in form opt/$RFQDN/$name where $RFQDN is a reverse fully qualified domain name you control. Then, the element can either contain the value (to set the blob value directly), or file attribute (to set the blob value from the file).

CPU Allocation

<domain>
  ...
  <vcpu placement='static' cpuset="1-4,^3,6" current="1">2</vcpu>
  <vcpus>
    <vcpu id='0' enabled='yes' hotpluggable='no' order='1'/>
    <vcpu id='1' enabled='no' hotpluggable='yes'/>
  </vcpus>
  ...
</domain>
vcpu

The content of this element defines the maximum number of virtual CPUs allocated for the guest OS, which must be between 1 and the maximum supported by the hypervisor.

cpuset

The optional attribute cpuset is a comma-separated list of physical CPU numbers that domain process and virtual CPUs can be pinned to by default. (NB: The pinning policy of domain process and virtual CPUs can be specified separately by cputune. If the attribute emulatorpin of cputune is specified, the cpuset specified by vcpu here will be ignored. Similarly, for virtual CPUs which have the vcpupin specified, the cpuset specified by cpuset here will be ignored. For virtual CPUs which don't have vcpupin specified, each will be pinned to the physical CPUs specified by cpuset here). Each element in that list is either a single CPU number, a range of CPU numbers, or a caret followed by a CPU number to be excluded from a previous range. Since 0.4.4

current

The optional attribute current can be used to specify whether fewer than the maximum number of virtual CPUs should be enabled. Since 0.8.5

placement

The optional attribute placement can be used to indicate the CPU placement mode for domain process. The value can be either "static" or "auto", but defaults to placement of numatune or "static" if cpuset is specified. Using "auto" indicates the domain process will be pinned to the advisory nodeset from querying numad and the value of attribute cpuset will be ignored if it's specified. If both cpuset and placement are not specified or if placement is "static", but no cpuset is specified, the domain process will be pinned to all the available physical CPUs. Since 0.9.11 (QEMU and KVM only)

vcpus

The vcpus element allows to control state of individual vCPUs. The id attribute specifies the vCPU id as used by libvirt in other places such as vCPU pinning, scheduler information and NUMA assignment. Note that the vCPU ID as seen in the guest may differ from libvirt ID in certain cases. Valid IDs are from 0 to the maximum vCPU count as set by the vcpu element minus 1. The enabled attribute allows to control the state of the vCPU. Valid values are yes and no. hotpluggable controls whether given vCPU can be hotplugged and hotunplugged in cases when the CPU is enabled at boot. Note that all disabled vCPUs must be hotpluggable. Valid values are yes and no. order allows to specify the order to add the online vCPUs. For hypervisors/platforms that require to insert multiple vCPUs at once the order may be duplicated across all vCPUs that need to be enabled at once. Specifying order is not necessary, vCPUs are then added in an arbitrary order. If order info is used, it must be used for all online vCPUs. Hypervisors may clear or update ordering information during certain operations to assure valid configuration. Note that hypervisors may create hotpluggable vCPUs differently from boot vCPUs thus special initialization may be necessary. Hypervisors may require that vCPUs enabled on boot which are not hotpluggable are clustered at the beginning starting with ID 0. It may be also required that vCPU 0 is always present and non-hotpluggable. Note that providing state for individual CPUs may be necessary to enable support of addressable vCPU hotplug and this feature may not be supported by all hypervisors. For QEMU the following conditions are required. vCPU 0 needs to be enabled and non-hotpluggable. On PPC64 along with it vCPUs that are in the same core need to be enabled as well. All non-hotpluggable CPUs present at boot need to be grouped after vCPU 0. Since 2.2.0 (QEMU only)

IOThreads Allocation

IOThreads are dedicated event loop threads for supported disk devices to perform block I/O requests in order to improve scalability especially on an SMP host/guest with many LUNs. Since 1.2.8 (QEMU only)

<domain>
  ...
  <iothreads>4</iothreads>
  ...
</domain>
<domain>
  ...
  <iothreadids>
    <iothread id="2"/>
    <iothread id="4"/>
    <iothread id="6"/>
    <iothread id="8" thread_pool_min="2" thread_pool_max="32">
      <poll max='123' grow='456' shrink='789'/>
    </iothread>
  </iothreadids>
  <defaultiothread thread_pool_min="8" thread_pool_max="16"/>
  ...
</domain>
iothreads

The content of this optional element defines the number of IOThreads to be assigned to the domain for use by supported target storage devices. There should be only 1 or 2 IOThreads per host CPU. There may be more than one supported device assigned to each IOThread. Since 1.2.8

iothreadids

The optional iothreadids element provides the capability to specifically define the IOThread ID's for the domain. By default, IOThread ID's are sequentially numbered starting from 1 through the number of iothreads defined for the domain. The id attribute is used to define the IOThread ID. The id attribute must be a positive integer greater than 0. If there are less iothreadids defined than iothreads defined for the domain, then libvirt will sequentially fill iothreadids starting at 1 avoiding any predefined id. If there are more iothreadids defined than iothreads defined for the domain, then the iothreads value will be adjusted accordingly. Since 1.2.15 The element has two optional attributes thread_pool_min and thread_pool_max which allow setting lower and upper boundary for number of worker threads for given IOThread. While the former can be value of zero, the latter can't. Since 8.5.0 Since 9.4.0 an optional sub-element poll with can be used to override the hypervisor-default interval of polling for the iothread before it switches back to events. The optional attribute max sets the maximum time polling should be used in nanoseconds. Setting max to 0 disables polling. Attributes grow and shrink override (or disable when set to 0 the default steps for increasing/decreasing the polling interval if the set interval is deemed insufficient or extensive.

defaultiothread

This element represents the default event loop within hypervisor, where I/O requests from devices not assigned to a specific IOThread are processed. The element then can have thread_pool_min and/or thread_pool_max attributes, which control the lower and upper boundary for number of worker threads of the default event loop. Emulator might be multithreaded and spawn so called worker threads on demand. In general neither of these attributes should be set (leaving the emulator use its own default values), unless the emulator runs in a real time workload and thus can't afford unpredictability of time it takes to spawn new worker threads. Since 8.5.0

CPU Tuning

<domain>
  ...
  <cputune>
    <vcpupin vcpu="0" cpuset="1-4,^2"/>
    <vcpupin vcpu="1" cpuset="0,1"/>
    <vcpupin vcpu="2" cpuset="2,3"/>
    <vcpupin vcpu="3" cpuset="0,4"/>
    <emulatorpin cpuset="1-3"/>
    <iothreadpin iothread="1" cpuset="5,6"/>
    <iothreadpin iothread="2" cpuset="7,8"/>
    <shares>2048</shares>
    <period>1000000</period>
    <quota>-1</quota>
    <global_period>1000000</global_period>
    <global_quota>-1</global_quota>
    <emulator_period>1000000</emulator_period>
    <emulator_quota>-1</emulator_quota>
    <iothread_period>1000000</iothread_period>
    <iothread_quota>-1</iothread_quota>
    <vcpusched vcpus='0-4,^3' scheduler='fifo' priority='1'/>
    <iothreadsched iothreads='2' scheduler='batch'/>
    <cachetune vcpus='0-3'>
      <cache id='0' level='3' type='both' size='3' unit='MiB'/>
      <cache id='1' level='3' type='both' size='3' unit='MiB'/>
      <monitor level='3' vcpus='1'/>
      <monitor level='3' vcpus='0-3'/>
    </cachetune>
    <cachetune vcpus='4-5'>
      <monitor level='3' vcpus='4'/>
      <monitor level='3' vcpus='5'/>
    </cachetune>
    <memorytune vcpus='0-3'>
      <node id='0' bandwidth='60'/>
    </memorytune>

  </cputune>
  ...
</domain>
cputune

The optional cputune element provides details regarding the CPU tunable parameters for the domain. Note: for the qemu driver, the optional vcpupin and emulatorpin pinning settings are honored after the emulator is launched and NUMA constraints considered. This means that it is expected that other physical CPUs of the host will be used during this time by the domain, which will be reflected by the output of virsh cpu-stats. Since 0.9.0

vcpupin

The optional vcpupin element specifies which of host's physical CPUs the domain vCPU will be pinned to. If this is omitted, and attribute cpuset of element vcpu is not specified, the vCPU is pinned to all the physical CPUs by default. It contains two required attributes, the attribute vcpu specifies vCPU id, and the attribute cpuset is same as attribute cpuset of element vcpu. QEMU driver support since 0.9.0, Xen driver support since 0.9.1

emulatorpin

The optional emulatorpin element specifies which of host physical CPUs the "emulator", a subset of a domain not including vCPU or iothreads will be pinned to. If this is omitted, and attribute cpuset of element vcpu is not specified, "emulator" is pinned to all the physical CPUs by default. It contains one required attribute cpuset specifying which physical CPUs to pin to.

iothreadpin

The optional iothreadpin element specifies which of host physical CPUs the IOThreads will be pinned to. If this is omitted and attribute cpuset of element vcpu is not specified, the IOThreads are pinned to all the physical CPUs by default. There are two required attributes, the attribute iothread specifies the IOThread ID and the attribute cpuset specifying which physical CPUs to pin to. See the IOThreads Allocation section documenting valid values of iothread. Since 1.2.9

shares

The optional shares element specifies the proportional weighted share for the domain. If this is omitted, it defaults to the OS provided defaults. NB, There is no unit for the value, it's a relative measure based on the setting of other VM, e.g. A VM configured with value 2048 will get twice as much CPU time as a VM configured with value 1024. The value should be in range [2, 262144] using cgroups v1, [1, 10000] using cgroups v2. Since 0.9.0

period

The optional period element specifies the enforcement interval (unit: microseconds). Within period, each vCPU of the domain will not be allowed to consume more than quota worth of runtime. The value should be in range [1000, 1000000]. A period with value 0 means no value. Only QEMU driver support since 0.9.4, LXC since 0.9.10

quota

The optional quota element specifies the maximum allowed bandwidth (unit: microseconds). A domain with quota as any negative value indicates that the domain has infinite bandwidth for vCPU threads, which means that it is not bandwidth controlled. The value should be in range [1000, 17592186044415] or less than 0. A quota with value 0 means no value. You can use this feature to ensure that all vCPUs run at the same speed. Only QEMU driver support since 0.9.4, LXC since 0.9.10

global_period

The optional global_period element specifies the enforcement CFS scheduler interval (unit: microseconds) for the whole domain in contrast with period which enforces the interval per vCPU. The value should be in range 1000, 1000000]. A global_period with value 0 means no value. Only QEMU driver support since 1.3.3

global_quota

The optional global_quota element specifies the maximum allowed bandwidth (unit: microseconds) within a period for the whole domain. A domain with global_quota as any negative value indicates that the domain has infinite bandwidth, which means that it is not bandwidth controlled. The value should be in range [1000, 17592186044415] or less than 0. A global_quota with value 0 means no value. Only QEMU driver support since 1.3.3

emulator_period

The optional emulator_period element specifies the enforcement interval (unit: microseconds). Within emulator_period, emulator threads (those excluding vCPUs) of the domain will not be allowed to consume more than emulator_quota worth of runtime. The value should be in range [1000, 1000000]. A period with value 0 means no value. Only QEMU driver support since 0.10.0

emulator_quota

The optional emulator_quota element specifies the maximum allowed bandwidth (unit: microseconds) for domain's emulator threads (those excluding vCPUs). A domain with emulator_quota as any negative value indicates that the domain has infinite bandwidth for emulator threads (those excluding vCPUs), which means that it is not bandwidth controlled. The value should be in range [1000, 17592186044415] or less than 0. A quota with value 0 means no value. Only QEMU driver support since 0.10.0

iothread_period

The optional iothread_period element specifies the enforcement interval (unit: microseconds) for IOThreads. Within iothread_period, each IOThread of the domain will not be allowed to consume more than iothread_quota worth of runtime. The value should be in range [1000, 1000000]. An iothread_period with value 0 means no value. Only QEMU driver support since 2.1.0

iothread_quota

The optional iothread_quota element specifies the maximum allowed bandwidth (unit: microseconds) for IOThreads. A domain with iothread_quota as any negative value indicates that the domain IOThreads have infinite bandwidth, which means that it is not bandwidth controlled. The value should be in range [1000, 17592186044415] or less than 0. An iothread_quota with value 0 means no value. You can use this feature to ensure that all IOThreads run at the same speed. Only QEMU driver support since 2.1.0

vcpusched, iothreadsched and emulatorsched

The optional vcpusched, iothreadsched and emulatorsched elements specify the scheduler type (values batch, idle, fifo, rr) for particular vCPU, IOThread and emulator threads respectively. For vcpusched and iothreadsched the attributes vcpus and iothreads select which vCPUs/IOThreads this setting applies to, leaving them out sets the default. The element emulatorsched does not have that attribute. Valid vcpus values start at 0 through one less than the number of vCPU's defined for the domain. Valid iothreads values are described in the IOThreads Allocation section. If no iothreadids are defined, then libvirt numbers IOThreads from 1 to the number of iothreads available for the domain. For real-time schedulers (fifo, rr), priority must be specified as well (and is ignored for non-real-time ones). The value range for the priority depends on the host kernel (usually 1-99). Since 1.2.13 emulatorsched since 5.3.0

cachetune Since 4.1.0

Optional cachetune element can control allocations for CPU caches using the resctrl on the host. Whether or not is this supported can be gathered from capabilities where some limitations like minimum size and required granularity are reported as well. The required attribute vcpus specifies to which vCPUs this allocation applies. A vCPU can only be member of one cachetune element allocation. The vCPUs specified by cachetune can be identical with those in memorytune, however they are not allowed to overlap. The optional, output only id attribute identifies cache uniquely. Supported subelements are:

cache

This optional element controls the allocation of CPU cache and has the following attributes:

level

Host cache level from which to allocate.

id

Host cache id from which to allocate.

type

Type of allocation. Can be code for code (instructions), data for data or both for both code and data (unified). Currently the allocation can be done only with the same type as the host supports, meaning you cannot request both for host with CDP (code/data prioritization) enabled.

size

The size of the region to allocate. The value by default is in bytes, but the unit attribute can be used to scale the value.

unit (optional)

If specified it is the unit such as KiB, MiB, GiB, or TiB (described in the memory element for Memory Allocation) in which size is specified, defaults to bytes.

monitor Since 4.10.0

The optional element monitor creates the cache monitor(s) for current cache allocation and has the following required attributes:

level

Host cache level the monitor belongs to.

vcpus

vCPU list the monitor applies to. A monitor's vCPU list can only be the member(s) of the vCPU list of the associated allocation. The default monitor has the same vCPU list as the associated allocation. For non-default monitors, overlapping vCPUs are not permitted.

memorytune Since 4.7.0

Optional memorytune element can control allocations for memory bandwidth using the resctrl on the host. Whether or not is this supported can be gathered from capabilities where some limitations like minimum bandwidth and required granularity are reported as well. The required attribute vcpus specifies to which vCPUs this allocation applies. A vCPU can only be member of one memorytune element allocation. The vcpus specified by memorytune can be identical to those specified by cachetune. However they are not allowed to overlap each other. Supported subelements are:

node

This element controls the allocation of CPU memory bandwidth and has the following attributes:

id

Host node id from which to allocate memory bandwidth.

bandwidth

The memory bandwidth to allocate from this node. The value by default is in percentage.

Memory Allocation

<domain>
  ...
  <maxMemory slots='16' unit='KiB'>1524288</maxMemory>
  <memory unit='KiB'>524288</memory>
  <currentMemory unit='KiB'>524288</currentMemory>
  ...
</domain>
memory

The maximum allocation of memory for the guest at boot time. The memory allocation includes possible additional memory devices specified at start or hotplugged later. The units for this value are determined by the optional attribute unit, which defaults to "KiB" (kibibytes, 210 or blocks of 1024 bytes). Valid units are "b" or "bytes" for bytes, "KB" for kilobytes (103 or 1,000 bytes), "k" or "KiB" for kibibytes (1024 bytes), "MB" for megabytes (106 or 1,000,000 bytes), "M" or "MiB" for mebibytes (220 or 1,048,576 bytes), "GB" for gigabytes (109 or 1,000,000,000 bytes), "G" or "GiB" for gibibytes (230 or 1,073,741,824 bytes), "TB" for terabytes (1012 or 1,000,000,000,000 bytes), or "T" or "TiB" for tebibytes (240 or 1,099,511,627,776 bytes). However, the value will be rounded up to the nearest kibibyte by libvirt, and may be further rounded to the granularity supported by the hypervisor. Some hypervisors also enforce a minimum, such as 4000KiB. In case NUMA is configured for the guest (See CPU model and topology) the memory element can be omitted. In the case of crash, optional attribute dumpCore can be used to control whether the guest memory should be included in the generated coredump or not (values "on", "off"). unit since 0.9.11 , dumpCore since 0.10.2 (QEMU only)

maxMemory

The run time maximum memory allocation of the guest. The initial memory specified by either the <memory> element or the NUMA cell size configuration can be increased by hot-plugging of memory to the limit specified by this element. The unit attribute behaves the same as for <memory>. The slots attribute specifies the number of slots available for adding memory to the guest. The bounds are hypervisor specific. Note that due to alignment of the memory chunks added via memory hotplug the full size allocation specified by this element may be impossible to achieve. Since 1.2.14 supported by the QEMU driver.

currentMemory

The actual allocation of memory for the guest. This value can be less than the maximum allocation, to allow for ballooning up the guests memory on the fly. If this is omitted, it defaults to the same value as the memory element. The unit attribute behaves the same as for memory.

Memory Backing

<domain>
  ...
  <memoryBacking>
    <hugepages>
      <page size="1" unit="G" nodeset="0-3,5"/>
      <page size="2" unit="M" nodeset="4"/>
    </hugepages>
    <nosharepages/>
    <locked/>
    <source type="file|anonymous|memfd"/>
    <access mode="shared|private"/>
    <allocation mode="immediate|ondemand" threads='8'/>
    <discard/>
  </memoryBacking>
  ...
</domain>

The optional memoryBacking element may contain several elements that influence how virtual memory pages are backed by host pages.

hugepages

This tells the hypervisor that the guest should have its memory allocated using hugepages instead of the normal native page size. Since 1.2.5 it's possible to set hugepages more specifically per numa node. The page element is introduced. It has one compulsory attribute size which specifies which hugepages should be used (especially useful on systems supporting hugepages of different sizes). The default unit for the size attribute is kilobytes (multiplier of 1024). If you want to use different unit, use optional unit attribute. For systems with NUMA, the optional nodeset attribute may come handy as it ties given guest's NUMA nodes to certain hugepage sizes. From the example snippet, one gigabyte hugepages are used for every NUMA node except node number four. For the correct syntax see NUMA Node Tuning.

nosharepages

Instructs hypervisor to disable shared pages (memory merge, KSM) for this domain. Since 1.0.6

locked

When set and supported by the hypervisor, memory pages belonging to the domain will be locked in host's memory and the host will not be allowed to swap them out, which might be required for some workloads such as real-time. For QEMU/KVM guests, the memory used by the QEMU process itself will be locked too: unlike guest memory, this is an amount libvirt has no way of figuring out in advance, so it has to remove the limit on locked memory altogether. Thus, enabling this option opens up to a potential security risk: the host will be unable to reclaim the locked memory back from the guest when it's running out of memory, which means a malicious guest allocating large amounts of locked memory could cause a denial-of-service attack on the host. Because of this, using this option is discouraged unless your workload demands it; even then, it's highly recommended to set a hard_limit (see Memory Tuning) on memory allocation suitable for the specific environment at the same time to mitigate the risks described above. Since 1.0.6

source

Using the type attribute, it's possible to provide "file" to utilize file memorybacking or keep the default "anonymous". Since 4.10.0 , you may choose "memfd" backing. (QEMU/KVM only)

access

Using the mode attribute, specify if the memory is to be "shared" or "private". This can be overridden per numa node by memAccess.

allocation

Using the optional mode attribute, specify when to allocate the memory by supplying either "immediate" or "ondemand". Since 8.2.0 it is possible to set the number of threads that hypervisor uses to allocate memory via threads attribute. To speed allocation process up, when pinning emulator thread it's recommended to include CPUs from desired NUMA nodes so that allocation threads can have their affinity set.

discard

When set and supported by hypervisor the memory content is discarded just before guest shuts down (or when DIMM module is unplugged). Please note that this is just an optimization and is not guaranteed to work in all cases (e.g. when hypervisor crashes). Since 4.4.0 (QEMU/KVM only)

Memory Tuning

<domain>
  ...
  <memtune>
    <hard_limit unit='G'>1</hard_limit>
    <soft_limit unit='M'>128</soft_limit>
    <swap_hard_limit unit='G'>2</swap_hard_limit>
    <min_guarantee unit='bytes'>67108864</min_guarantee>
  </memtune>
  ...
</domain>
memtune

The optional memtune element provides details regarding the memory tunable parameters for the domain. If this is omitted, it defaults to the OS provided defaults. For QEMU/KVM, the parameters are applied to the QEMU process as a whole. Thus, when counting them, one needs to add up guest RAM, guest video RAM, and some memory overhead of QEMU itself. The last piece is hard to determine so one needs guess and try. For each tunable, it is possible to designate which unit the number is in on input, using the same values as for <memory>. For backwards compatibility, output is always in KiB. unit since 0.9.11 Possible values for all *_limit parameters are in range from 0 to VIR_DOMAIN_MEMORY_PARAM_UNLIMITED.

hard_limit

The optional hard_limit element is the maximum memory the guest can use. The units for this value are kibibytes (i.e. blocks of 1024 bytes). Users of QEMU and KVM are strongly advised not to set this limit as domain may get killed by the kernel if the guess is too low, and determining the memory needed for a process to run is an undecidable problem; that said, if you already set locked in Memory Backing because your workload demands it, you'll have to take into account the specifics of your deployment and figure out a value for hard_limit that is large enough to support the memory requirements of your guest, but small enough to protect your host against a malicious guest locking all memory.

soft_limit

The optional soft_limit element is the memory limit to enforce during memory contention. The units for this value are kibibytes (i.e. blocks of 1024 bytes)

swap_hard_limit

The optional swap_hard_limit element is the maximum memory plus swap the guest can use. The units for this value are kibibytes (i.e. blocks of 1024 bytes). This has to be more than hard_limit value provided

min_guarantee

The optional min_guarantee element is the guaranteed minimum memory allocation for the guest. The units for this value are kibibytes (i.e. blocks of 1024 bytes). This element is only supported by VMware ESX and OpenVZ drivers.

NUMA Node Tuning

<domain>
  ...
  <numatune>
    <memory mode="strict" nodeset="1-4,^3"/>
    <memnode cellid="0" mode="strict" nodeset="1"/>
    <memnode cellid="2" mode="preferred" nodeset="2"/>
  </numatune>
  ...
</domain>
numatune

The optional numatune element provides details of how to tune the performance of a NUMA host via controlling NUMA policy for domain process. NB, only supported by QEMU driver. Since 0.9.3

memory

The optional memory element specifies how to allocate memory for the domain process on a NUMA host. It contains several optional attributes. Attribute mode is either 'interleave', 'strict', 'preferred', or 'restrictive', defaults to 'strict'. The value 'restrictive' specifies using system default policy and only cgroups is used to restrict the memory nodes, and it requires setting mode to 'restrictive' in memnode elements (see quirk below). This exists solely for the purpose of being able to request movement of such memory for a running domain using virsh numatune or virDomainSetNumaParameters and is not guaranteed to happen. Attribute nodeset specifies the NUMA nodes, using the same syntax as attribute cpuset of element vcpu. Attribute placement ( since 0.9.12 ) can be used to indicate the memory placement mode for domain process, its value can be either "static" or "auto", defaults to placement of vcpu, or "static" if nodeset is specified. "auto" indicates the domain process will only allocate memory from the advisory nodeset returned from querying numad, and the value of attribute nodeset will be ignored if it's specified. If placement of vcpu is 'auto', and numatune is not specified, a default numatune with placement 'auto' and mode 'strict' will be added implicitly. Since 0.9.3 See virDomainSetNumaParameters for more information on update of this element.

memnode

Optional memnode elements can specify memory allocation policies per each guest NUMA node. For those nodes having no corresponding memnode element, the default from element memory will be used. Attribute cellid addresses guest NUMA node for which the settings are applied. Attributes mode and nodeset have the same meaning and syntax as in memory element. This setting is not compatible with automatic placement. Note that for memnode this will only guide the memory access for the vCPU threads or similar mechanism and is very hypervisor-specific. This does not guarantee the placement of the node's memory allocation. For proper restriction other means should be used (e.g. different mode, preallocated hugepages).

QEMU Since 1.2.7

Block I/O Tuning

<domain>
  ...
  <blkiotune>
    <weight>800</weight>
    <device>
      <path>/dev/sda</path>
      <weight>1000</weight>
    </device>
    <device>
      <path>/dev/sdb</path>
      <weight>500</weight>
      <read_bytes_sec>10000</read_bytes_sec>
      <write_bytes_sec>10000</write_bytes_sec>
      <read_iops_sec>20000</read_iops_sec>
      <write_iops_sec>20000</write_iops_sec>
    </device>
  </blkiotune>
  ...
</domain>
blkiotune

The optional blkiotune element provides the ability to tune Blkio cgroup tunable parameters for the domain. If this is omitted, it defaults to the OS provided defaults. Since 0.8.8

weight

The optional weight element is the overall I/O weight of the guest. The value should be in the range [100, 1000]. After kernel 2.6.39, the value could be in the range [10, 1000].

device

The domain may have multiple device elements that further tune the weights for each host block device in use by the domain. Note that multiple disks (See Hard drives, floppy disks, CDROMs) can share a single host block device, if they are backed by files within the same host file system, which is why this tuning parameter is at the global domain level rather than associated with each guest disk device (contrast this to the <iotune> element of a disk definition (See Hard drives, floppy disks, CDROMs) which can applies to an individual disk). Each device element has two mandatory sub-elements, path describing the absolute path of the device, and weight giving the relative weight of that device, in the range [100, 1000]. After kernel 2.6.39, the value could be in the range [10, 1000]. Since 0.9.8 Additionally, the following optional sub-elements can be used:

read_bytes_sec

Read throughput limit in bytes per second. Since 1.2.2

write_bytes_sec

Write throughput limit in bytes per second. Since 1.2.2

read_iops_sec

Read I/O operations per second limit. Since 1.2.2

write_iops_sec

Write I/O operations per second limit. Since 1.2.2

Resource partitioning

Hypervisors may allow for virtual machines to be placed into resource partitions, potentially with nesting of said partitions. The resource element groups together configuration related to resource partitioning. It currently supports a child element partition whose content defines the absolute path of the resource partition in which to place the domain. If no partition is listed, then the domain will be placed in a default partition. It is the responsibility of the app/admin to ensure that the partition exists prior to starting the guest. Only the (hypervisor specific) default partition can be assumed to exist by default.

...
<resource>
  <partition>/virtualmachines/production</partition>
</resource>
...

Resource partitions are currently supported by the QEMU and LXC drivers, which map partition paths to cgroups directories, in all mounted controllers. Since 1.0.5

Fibre Channel VMID

The FC SAN can provide various QoS levels and access control depending on the VMID. It can also collect telemetry data at per-VM level which can be used to enhance the IO performance of the VM. This can be configured by using the appid attribute of fibrechannel element. The attribute contains single string (max 128 bytes) and it is used by kernel to create VMID.

...
<resource>
  <fibrechannel appid='userProvidedID'/>
</resource>
...

Using this feature requires Fibre Channel capable HW, kernel compiled with option CONFIG_BLK_CGROUP_FC_APPID and nvme_fc kernel module loaded. Since 7.7.0

CPU model and topology

Requirements for CPU model, its features and topology can be specified using the following collection of elements. Since 0.7.5

...
<cpu match='exact'>
  <model fallback='allow'>core2duo</model>
  <vendor>Intel</vendor>
  <topology sockets='1' dies='1' cores='2' threads='1'/>
  <cache level='3' mode='emulate'/>
  <maxphysaddr mode='emulate' bits='42'/>
  <feature policy='disable' name='lahf_lm'/>
</cpu>
...
<cpu mode='host-model'>
  <model fallback='forbid'/>
  <topology sockets='1' dies='1' cores='2' threads='1'/>
</cpu>
...
<cpu mode='host-passthrough' migratable='off'>
  <cache mode='passthrough'/>
  <maxphysaddr mode='passthrough' limit='39'/>
  <feature policy='disable' name='lahf_lm'/>
...
<cpu mode='maximum' migratable='off'>
  <cache mode='passthrough'/>
  <feature policy='disable' name='lahf_lm'/>
...

In case no restrictions need to be put on CPU model and its features, a simpler cpu element can be used. Since 0.7.6

...
<cpu>
  <topology sockets='1' dies='1' cores='2' threads='1'/>
</cpu>
...
cpu

The cpu element is the main container for describing guest CPU requirements. Its match attribute specifies how strictly the virtual CPU provided to the guest matches these requirements. Since 0.7.6 the match attribute can be omitted if topology is the only element within cpu. Possible values for the match attribute are:

minimum

The specified CPU model and features describes the minimum requested CPU. A better CPU will be provided to the guest if it is possible with the requested hypervisor on the current host. This is a constrained host-model mode; the domain will not be created if the provided virtual CPU does not meet the requirements.

exact

The virtual CPU provided to the guest should exactly match the specification. If such CPU is not supported, libvirt will refuse to start the domain.

strict

The domain will not be created unless the host CPU exactly matches the specification. This is not very useful in practice and should only be used if there is a real reason.

Since 0.8.5 the match attribute can be omitted and will default to exact. Sometimes the hypervisor is not able to create a virtual CPU exactly matching the specification passed by libvirt. Since 3.2.0 , an optional check attribute can be used to request a specific way of checking whether the virtual CPU matches the specification. It is usually safe to omit this attribute when starting a domain and stick with the default value. Once the domain starts, libvirt will automatically change the check attribute to the best supported value to ensure the virtual CPU does not change when the domain is migrated to another host. The following values can be used:

none

Libvirt does no checking and it is up to the hypervisor to refuse to start the domain if it cannot provide the requested CPU. With QEMU this means no checking is done at all since the default behavior of QEMU is to emit warnings, but start the domain anyway.

partial

Libvirt will check the guest CPU specification before starting a domain, but the rest is left on the hypervisor. It can still provide a different virtual CPU.

full

The virtual CPU created by the hypervisor will be checked against the CPU specification and the domain will not be started unless the two CPUs match.

Since 0.9.10 , an optional mode attribute may be used to make it easier to configure a guest CPU to be as close to host CPU as possible. Possible values for the mode attribute are:

custom

In this mode, the cpu element describes the CPU that should be presented to the guest. This is the default when no mode attribute is specified. This mode makes it so that a persistent guest will see the same hardware no matter what host the guest is booted on.

host-model

The host-model mode is essentially a shortcut to copying host CPU definition from capabilities XML into domain XML. Since the CPU definition is copied just before starting a domain, exactly the same XML can be used on different hosts while still providing the best guest CPU each host supports. The match attribute can't be used in this mode. Specifying CPU model is not supported either, but model's fallback attribute may still be used. Using the feature element, specific flags may be enabled or disabled specifically in addition to the host model. This may be used to fine tune features that can be emulated. (Since 1.1.1) . Libvirt does not model every aspect of each CPU so the guest CPU will not match the host CPU exactly. On the other hand, the ABI provided to the guest is reproducible. During migration, complete CPU model definition is transferred to the destination host so the migrated guest will see exactly the same CPU model for the running instance of the guest, even if the destination host contains more capable CPUs or newer kernel; but shutting down and restarting the guest may present different hardware to the guest according to the capabilities of the new host. Prior to libvirt 3.2.0 and QEMU 2.9.0 detection of the host CPU model via QEMU is not supported. Thus the CPU configuration created using host-model may not work as expected. Since 3.2.0 and QEMU 2.9.0 this mode works the way it was designed and it is indicated by the fallback attribute set to forbid in the host-model CPU definition advertised in domain capabilities XML. When fallback attribute is set to allow in the domain capabilities XML, it is recommended to use custom mode with just the CPU model from the host capabilities XML. Since 1.2.11 PowerISA allows processors to run VMs in binary compatibility mode supporting an older version of ISA. Libvirt on PowerPC architecture uses the host-model to signify a guest mode CPU running in binary compatibility mode. Example: When a user needs a power7 VM to run in compatibility mode on a Power8 host, this can be described in XML as follows :

<cpu mode='host-model'>
  <model>power7</model>
</cpu>
...
host-passthrough

With this mode, the CPU visible to the guest should be exactly the same as the host CPU even in the aspects that libvirt does not understand. Though the downside of this mode is that the guest environment cannot be reproduced on different hardware. Thus, if you hit any bugs, you are on your own. Further details of that CPU can be changed using feature elements. Migration of a guest using host-passthrough is dangerous if the source and destination hosts are not identical in both hardware, QEMU version, microcode version and configuration. If such a migration is attempted then the guest may hang or crash upon resuming execution on the destination host. Depending on hypervisor version the virtual CPU may or may not contain features which may block migration even to an identical host. Since 6.5.0 optional migratable attribute may be used to explicitly request such features to be removed from (on) or kept in (off) the virtual CPU. This attribute does not make migration to another host safer: even with migratable='on' migration will be dangerous unless both hosts are identical as described above.

maximum

When running a guest with hardware virtualization this CPU model is functionally identical to host-passthrough, so refer to the docs above.

When running a guest with CPU emulation, this CPU model will enable the maximum set of features that the emulation engine is able to support. Note that even with migratable='on' migration will be dangerous unless both hosts are running identical versions of the emulation code.

Since 7.1.0 with the QEMU driver.

Both host-model and host-passthrough modes make sense when a domain can run directly on the host CPUs (for example, domains with type kvm or hvf). The actual host CPU is irrelevant for domains with emulated virtual CPUs (such as domains with type qemu). However, for backward compatibility host-model may be implemented even for domains running on emulated CPUs in which case the best CPU the hypervisor is able to emulate may be used rather then trying to mimic the host CPU model.

If an application does not care about a specific CPU, just wants the best feature set without a need for migration compatibility, the maximum model is a good choice on hypervisors where it is available.

model

The content of the model element specifies CPU model requested by the guest. The list of available CPU models and their definition can be found in directory cpu_map, installed in libvirt's data directory. If a hypervisor is not able to use the exact CPU model, libvirt automatically falls back to a closest model supported by the hypervisor while maintaining the list of CPU features. Since 0.9.10 , an optional fallback attribute can be used to forbid this behavior, in which case an attempt to start a domain requesting an unsupported CPU model will fail. Supported values for fallback attribute are: allow (this is the default), and forbid. The optional vendor_id attribute ( Since 0.10.0 ) can be used to set the vendor id seen by the guest. It must be exactly 12 characters long. If not set the vendor id of the host is used. Typical possible values are "AuthenticAMD" and "GenuineIntel".

vendor

Since 0.8.3 the content of the vendor element specifies CPU vendor requested by the guest. If this element is missing, the guest can be run on a CPU matching given features regardless on its vendor. The list of supported vendors can be found in cpu_map/*_vendors.xml.

topology

The topology element specifies requested topology of virtual CPU provided to the guest. Four attributes, sockets, dies, cores, and threads, accept non-zero positive integer values. They refer to the total number of CPU sockets, number of dies per socket, number of cores per die, and number of threads per core, respectively. The dies attribute is optional and will default to 1 if omitted, while the other attributes are all mandatory. Hypervisors may require that the maximum number of vCPUs specified by the cpus element equals to the number of vcpus resulting from the topology.

feature

The cpu element can contain zero or more feature elements used to fine-tune features provided by the selected CPU model. The list of known feature names can be found in the same file as CPU models. The meaning of each feature element depends on its policy attribute, which has to be set to one of the following values:

force

The virtual CPU will claim the feature is supported regardless of it being supported by host CPU.

require

Guest creation will fail unless the feature is supported by the host CPU or the hypervisor is able to emulate it.

optional

The feature will be supported by virtual CPU if and only if it is supported by host CPU.

disable

The feature will not be supported by virtual CPU.

forbid

Guest creation will fail if the feature is supported by host CPU.

Since 0.8.5 the policy attribute can be omitted and will default to require.

Individual CPU feature names are specified as part of the name attribute. For example, to explicitly specify the 'pcid' feature with Intel IvyBridge CPU model:

...
<cpu match='exact'>
  <model fallback='forbid'>IvyBridge</model>
  <vendor>Intel</vendor>
  <feature policy='require' name='pcid'/>
</cpu>
...
cache

Since 3.3.0 the cache element describes the virtual CPU cache. If the element is missing, the hypervisor will use a sensible default.

level

This optional attribute specifies which cache level is described by the element. Missing attribute means the element describes all CPU cache levels at once. Mixing cache elements with the level attribute set and those without the attribute is forbidden.

mode

The following values are supported:

emulate

The hypervisor will provide a fake CPU cache data.

passthrough

The real CPU cache data reported by the host CPU will be passed through to the virtual CPU.

disable

The virtual CPU will report no CPU cache of the specified level (or no cache at all if the level attribute is missing).

maxphysaddr

Since 8.7.0 the maxphysaddr element describes the virtual CPU address size in bits. The hypervisor default is used if the element is missing.

mode

This mandatory attribute specifies how the address size is presented. The follow modes are supported:

passthrough

The number of physical address bits reported by the host CPU will be passed through to the virtual CPUs

emulate

The hypervisor will define a specific value for the number of bits of physical addresses via the bits attribute, (optional since 9.2.0)

The number of bits cannot exceed the number of physical address bits supported by the hypervisor.

bits

The bits attribute is mandatory if the mode attribute is set to emulate and specifies the virtual CPU address size in bits.

limit

The limit attribute can be used to restrict the maximum value of address bits for passthrough mode, i.e. in case the host CPU reports more bits than that, limit is used. Since 9.3.0

Guest NUMA topology can be specified using the numa element. Since 0.9.8

...
<cpu>
  ...
  <numa>
    <cell id='0' cpus='0-3' memory='512000' unit='KiB' discard='yes'/>
    <cell id='1' cpus='4-7' memory='512000' unit='KiB' memAccess='shared'/>
  </numa>
  ...
</cpu>
...

Each cell element specifies a NUMA cell or a NUMA node. cpus specifies the CPU or range of CPUs that are part of the node. Since 6.5.0 For the qemu driver, if the emulator binary supports disjointed cpus ranges in each cell, the sum of all CPUs declared in each cell will be matched with the maximum number of virtual CPUs declared in the vcpu element. This is done by filling any remaining CPUs into the first NUMA cell. Users are encouraged to supply a complete NUMA topology, where the sum of the NUMA CPUs matches the maximum virtual CPUs number declared in vcpus, to make the domain consistent across qemu and libvirt versions. memory specifies the node memory in kibibytes (i.e. blocks of 1024 bytes). Since 6.6.0 the cpus attribute is optional and if omitted a CPU-less NUMA node is created. Since 1.2.11 one can use an additional unit attribute (See Memory Allocation) to define units in which memory is specified. Since 1.2.7 all cells should have id attribute in case referring to some cell is necessary in the code, otherwise the cells are assigned ids in the increasing order starting from 0. Mixing cells with and without the id attribute is not recommended as it may result in unwanted behaviour. Since 1.2.9 the optional attribute memAccess can control whether the memory is to be mapped as "shared" or "private". This is valid only for hugepages-backed memory and nvdimm modules. Each cell element can have an optional discard attribute which fine tunes the discard feature for given numa node as described under Memory Backing. Accepted values are yes and no. Since 4.4.0

This guest NUMA specification is currently available only for QEMU/KVM and Xen.

A NUMA hardware architecture supports the notion of distances between NUMA cells. Since 3.10.0 it is possible to define the distance between NUMA cells using the distances element within a NUMA cell description. The sibling sub-element is used to specify the distance value between sibling NUMA cells. For more details, see the chapter explaining the system's SLIT (System Locality Information Table) within the ACPI (Advanced Configuration and Power Interface) specification.

...
<cpu>
  ...
  <numa>
    <cell id='0' cpus='0,4-7' memory='512000' unit='KiB'>
      <distances>
        <sibling id='0' value='10'/>
        <sibling id='1' value='21'/>
        <sibling id='2' value='31'/>
        <sibling id='3' value='41'/>
      </distances>
    </cell>
    <cell id='1' cpus='1,8-10,12-15' memory='512000' unit='KiB' memAccess='shared'>
      <distances>
        <sibling id='0' value='21'/>
        <sibling id='1' value='10'/>
        <sibling id='2' value='21'/>
        <sibling id='3' value='31'/>
      </distances>
    </cell>
    <cell id='2' cpus='2,11' memory='512000' unit='KiB' memAccess='shared'>
      <distances>
        <sibling id='0' value='31'/>
        <sibling id='1' value='21'/>
        <sibling id='2' value='10'/>
        <sibling id='3' value='21'/>
      </distances>
    </cell>
    <cell id='3' cpus='3' memory='512000' unit='KiB'>
      <distances>
        <sibling id='0' value='41'/>
        <sibling id='1' value='31'/>
        <sibling id='2' value='21'/>
        <sibling id='3' value='10'/>
      </distances>
    </cell>
  </numa>
  ...
</cpu>
...

Describing distances between NUMA cells is currently only supported by Xen and QEMU. If no distances are given to describe the SLIT data between different cells, it will default to a scheme using 10 for local and 20 for remote distances.

ACPI Heterogeneous Memory Attribute Table

...
<cpu>
  ...
  <numa>
    <cell id='0' cpus='0-3' memory='2097152' unit='KiB' discard='yes'>
      <cache level='1' associativity='direct' policy='writeback'>
        <size value='10' unit='KiB'/>
        <line value='8' unit='B'/>
      </cache>
    </cell>
    <cell id='1' cpus='4-7' memory='512000' unit='KiB' memAccess='shared'/>
    <interconnects>
      <latency initiator='0' target='0' type='access' value='5'/>
      <latency initiator='0' target='0' cache='1' type='access' value='10'/>
      <bandwidth initiator='0' target='0' type='access' value='204800' unit='KiB'/>
    </interconnects>
  </numa>
  ...
</cpu>
...

Since 6.6.0 the cell element can have a cache child element which describes memory side cache for memory proximity domains. The cache element has a level attribute describing the cache level and thus the element can be repeated multiple times to describe different levels of the cache.

The cache element then has following mandatory attributes:

level

Level of the cache this description refers to.

associativity

Describes cache associativity (accepted values are none, direct and full).

policy

Describes cache write associativity (accepted values are none, writeback and writethrough).

The cache element has two mandatory child elements then: size and line which describe cache size and cache line size. Both elements accept two attributes: value and unit which set the value of corresponding cache attribute.

The NUMA description has an optional interconnects element that describes the normalized memory read/write latency, read/write bandwidth between Initiator Proximity Domains (Processor or I/O) and Target Proximity Domains (Memory).

The interconnects element can have zero or more latency child elements to describe latency between two memory nodes and zero or more bandwidth child elements to describe bandwidth between two memory nodes. Both these have the following mandatory attributes:

initiator

Refers to the source NUMA node

target

Refers to the target NUMA node

type

The type of the access. Accepted values: access, read, write

value

The actual value. For latency this is delay in nanoseconds, for bandwidth this value is in kibibytes per second. Use additional unit attribute to change the units.

To describe latency from one NUMA node to a cache of another NUMA node the latency element has optional cache attribute which in combination with target attribute creates full reference to distant NUMA node's cache level. For instance, target='0' cache='1' refers to the first level cache of NUMA node 0.

Events configuration

It is sometimes necessary to override the default actions taken on various events. Not all hypervisors support all events and actions. The actions may be taken as a result of calls to libvirt APIs virDomainReboot , virDomainShutdown , or virDomainShutdownFlags . Using virsh reboot or virsh shutdown would also trigger the event.

...
<on_poweroff>destroy</on_poweroff>
<on_reboot>restart</on_reboot>
<on_crash>restart</on_crash>
<on_lockfailure>poweroff</on_lockfailure>
...

The following collections of elements allow the actions to be specified when a guest OS triggers a lifecycle operation. A common use case is to force a reboot to be treated as a poweroff when doing the initial OS installation. This allows the VM to be re-configured for the first post-install bootup.

on_poweroff

The content of this element specifies the action to take when the guest requests a poweroff.

on_reboot

The content of this element specifies the action to take when the guest requests a reboot.

on_crash

The content of this element specifies the action to take when the guest crashes.

Each of these states allow for the same four possible actions.

destroy

The domain will be terminated completely and all resources released.

restart

The domain will be terminated and then restarted with the same configuration.

preserve

The domain will be terminated and its resource preserved to allow analysis.

rename-restart

The domain will be terminated and then restarted with a new name. (Only supported by the libxl hypervisor driver.)

QEMU/KVM/HVF supports the on_poweroff and on_reboot events handling the destroy and restart actions, but the combination of on_poweroff set to restart and on_reboot set to destroy is forbidden.

The on_crash event supports these additional actions since 0.8.4 .

coredump-destroy

The crashed domain's core will be dumped, and then the domain will be terminated completely and all resources released

coredump-restart

The crashed domain's core will be dumped, and then the domain will be restarted with the same configuration

Since 3.9.0 , the lifecycle events can be configured via the virDomainSetLifecycleAction API.

The on_lockfailure element ( since 1.0.0 ) may be used to configure what action should be taken when a lock manager loses resource locks. The following actions are recognized by libvirt, although not all of them need to be supported by individual lock managers. When no action is specified, each lock manager will take its default action.

poweroff

The domain will be forcefully powered off.

restart

The domain will be powered off and started up again to reacquire its locks.

pause

The domain will be paused so that it can be manually resumed when lock issues are solved.

ignore

Keep the domain running as if nothing happened.

Power Management

Since 0.10.2 it is possible to forcibly enable or disable BIOS advertisements to the guest OS. (NB: Only qemu driver support)

...
<pm>
  <suspend-to-disk enabled='no'/>
  <suspend-to-mem enabled='yes'/>
</pm>
...
pm

These elements enable ('yes') or disable ('no') BIOS support for S3 (suspend-to-mem) and S4 (suspend-to-disk) ACPI sleep states. If nothing is specified, then the hypervisor will be left with its default value. Note: This setting cannot prevent the guest OS from performing a suspend as the guest OS itself can choose to circumvent the unavailability of the sleep states (e.g. S4 by turning off completely).

Hypervisor features

Hypervisors may allow certain CPU / machine features to be toggled on/off.

...
<features>
  <pae/>
  <acpi/>
  <apic/>
  <hap/>
  <privnet/>
  <hyperv mode='custom'>
    <relaxed state='on'/>
    <vapic state='on'/>
    <spinlocks state='on' retries='4096'/>
    <vpindex state='on'/>
    <runtime state='on'/>
    <synic state='on'/>
    <stimer state='on'>
      <direct state='on'/>
    </stimer>
    <reset state='on'/>
    <vendor_id state='on' value='KVM Hv'/>
    <frequencies state='on'/>
    <reenlightenment state='on'/>
    <tlbflush state='on'/>
    <ipi state='on'/>
    <evmcs state='on'/>
  </hyperv>
  <kvm>
    <hidden state='on'/>
    <hint-dedicated state='on'/>
    <poll-control state='on'/>
    <pv-ipi state='off'/>
    <dirty-ring state='on' size='4096'/>
  </kvm>
  <xen>
    <e820_host state='on'/>
    <passthrough state='on' mode='share_pt'/>
  </xen>
  <pvspinlock state='on'/>
  <gic version='2'/>
  <ioapic driver='qemu'/>
  <hpt resizing='required'>
    <maxpagesize unit='MiB'>16</maxpagesize>
  </hpt>
  <vmcoreinfo state='on'/>
  <smm state='on'>
    <tseg unit='MiB'>48</tseg>
  </smm>
  <htm state='on'/>
  <ccf-assist state='on'/>
  <msrs unknown='ignore'/>
  <cfpc value='workaround'/>
  <sbbc value='workaround'/>
  <ibs value='fixed-na'/>
  <tcg>
    <tb-cache unit='MiB'>128</tb-cache>
  </tcg>
  <async-teardown enabled='yes'/>
</features>
...

All features are listed within the features element, omitting a togglable feature tag turns it off. The available features can be found by asking for the capabilities XML and domain capabilities XML, but a common set for fully virtualized domains are:

pae

Physical address extension mode allows 32-bit guests to address more than 4 GB of memory.

acpi

ACPI is useful for power management, for example, with KVM or HVF guests it is required for graceful shutdown to work.

apic

APIC allows the use of programmable IRQ management. Since 0.10.2 (QEMU only) there is an optional attribute eoi with values on and off which toggles the availability of EOI (End of Interrupt) for the guest.

hap

Depending on the state attribute (values on, off) enable or disable use of Hardware Assisted Paging. The default is on if the hypervisor detects availability of Hardware Assisted Paging.

viridian

Enable Viridian hypervisor extensions for paravirtualizing guest operating systems

privnet

Always create a private network namespace. This is automatically set if any interface devices are defined. This feature is only relevant for container based virtualization drivers, such as LXC.

hyperv

Enable various features improving behavior of guests running Microsoft Windows.

Feature Description Value Since
relaxed Relax constraints on timers on, off 1.0.0 (QEMU 2.0)
vapic Enable virtual APIC on, off 1.1.0 (QEMU 2.0)
spinlocks Enable spinlock support on, off; retries - at least 4095 1.1.0 (QEMU 2.0)
vpindex Virtual processor index on, off 1.3.3 (QEMU 2.5)
runtime Processor time spent on running guest code and on behalf of guest code on, off 1.3.3 (QEMU 2.5)
synic Enable Synthetic Interrupt Controller (SynIC) on, off 1.3.3 (QEMU 2.6)
stimer Enable SynIC timers, optionally with Direct Mode support on, off; direct - on,off 1.3.3 (QEMU 2.6), direct mode 5.7.0 (QEMU 4.1)
reset Enable hypervisor reset on, off 1.3.3 (QEMU 2.5)
vendor_id Set hypervisor vendor id on, off; value - string, up to 12 characters 1.3.3 (QEMU 2.5)
frequencies Expose frequency MSRs on, off 4.7.0 (QEMU 2.12)
reenlightenment Enable re-enlightenment notification on migration on, off 4.7.0 (QEMU 3.0)
tlbflush Enable PV TLB flush support on, off 4.7.0 (QEMU 3.0)
ipi Enable PV IPI support on, off 4.10.0 (QEMU 3.1)
evmcs Enable Enlightened VMCS on, off 4.10.0 (QEMU 3.1)
avic Enable use Hyper-V SynIC with hardware APICv/AVIC on, off 8.10.0 (QEMU 6.2)

Since 8.0.0 , the hypervisor can be configured further by setting the mode attribute to one of the following values:

custom

Set exactly the specified features.

passthrough

Enable all features currently supported by the hypervisor, even those that libvirt does not understand. Migration of a guest using passthrough is dangerous if the source and destination hosts are not identical in both hardware, QEMU version, microcode version and configuration. If such a migration is attempted then the guest may hang or crash upon resuming execution on the destination host. Depending on hypervisor version the virtual CPU may or may not contain features which may block migration even to an identical host.

The mode attribute can be omitted and will default to custom.

pvspinlock

Notify the guest that the host supports paravirtual spinlocks for example by exposing the pvticketlocks mechanism. This feature can be explicitly disabled by using state='off' attribute.

kvm

Various features to change the behavior of the KVM hypervisor.

Feature Description Value Since
hidden Hide the KVM hypervisor from standard MSR based discovery on, off 1.2.8 (QEMU 2.1.0)
hint-dedicated Allows a guest to enable optimizations when running on dedicated vCPUs on, off 5.7.0 (QEMU 2.12.0)
poll-control Decrease IO completion latency by introducing a grace period of busy waiting on, off 6.10.0 (QEMU 4.2)
pv-ipi Paravirtualized send IPIs on, off 7.10.0 (QEMU 3.1)
dirty-ring Enable dirty ring feature on, off; size - must be power of 2, range [1024,65536] 8.0.0 (QEMU 6.1)
xen

Various features to change the behavior of the Xen hypervisor.

Feature Description Value Since
e820_host Expose the host e820 to the guest (PV only) on, off 6.3.0
passthrough Enable IOMMU mappings allowing PCI passthrough on, off; mode - optional string sync_pt or share_pt 6.3.0
pmu

Depending on the state attribute (values on, off, default on) enable or disable the performance monitoring unit for the guest. Since 1.2.12

vmport

Depending on the state attribute (values on, off, default on) enable or disable the emulation of VMware IO port, for vmmouse etc. Since 1.2.16

gic

Enable for architectures using a General Interrupt Controller instead of APIC in order to handle interrupts. For example, the 'aarch64' architecture uses gic instead of apic. The optional attribute version specifies the GIC version; however, it may not be supported by all hypervisors. Accepted values are 2, 3 and host. Since 1.2.16

smm

Depending on the state attribute (values on, off, default on) enable or disable System Management Mode. Since 2.1.0

Optional sub-element tseg can be used to specify the amount of memory dedicated to SMM's extended TSEG. That offers a fourth option size apart from the existing ones (1 MiB, 2 MiB and 8 MiB) that the guest OS (or rather loader) can choose from. The size can be specified as a value of that element, optional attribute unit can be used to specify the unit of the aforementioned value (defaults to 'MiB'). If set to 0 the extended size is not advertised and only the default ones (see above) are available.

If the VM is booting you should leave this option alone, unless you are very certain you know what you are doing.

This value is configurable due to the fact that the calculation cannot be done right with the guarantee that it will work correctly. In QEMU, the user-configurable extended TSEG feature was unavailable up to and including pc-q35-2.9. Starting with pc-q35-2.10 the feature is available, with default size 16 MiB. That should suffice for up to roughly 272 vCPUs, 5 GiB guest RAM in total, no hotplug memory range, and 32 GiB of 64-bit PCI MMIO aperture. Or for 48 vCPUs, with 1TB of guest RAM, no hotplug DIMM range, and 32GB of 64-bit PCI MMIO aperture. The values may also vary based on the loader the VM is using.

Additional size might be needed for significantly higher vCPU counts or increased address space (that can be memory, maxMemory, 64-bit PCI MMIO aperture size; roughly 8 MiB of TSEG per 1 TiB of address space) which can also be rounded up.

Due to the nature of this setting being similar to "how much RAM should the guest have" users are advised to either consult the documentation of the guest OS or loader (if there is any), or test this by trial-and-error changing the value until the VM boots successfully. Yet another guiding value for users might be the fact that 48 MiB should be enough for pretty large guests (240 vCPUs and 4TB guest RAM), but it is on purpose not set as default as 48 MiB of unavailable RAM might be too much for small guests (e.g. with 512 MiB of RAM).

See Memory Allocation for more details about the unit attribute. Since 4.5.0 (QEMU only)

ioapic

Tune the I/O APIC. Possible values for the driver attribute are: kvm (default for KVM domains) and qemu which puts I/O APIC in userspace which is also known as a split I/O APIC mode. Since 3.4.0 (QEMU/KVM only)

hpt

Configure the HPT (Hash Page Table) of a pSeries guest. Possible values for the resizing attribute are enabled, which causes HPT resizing to be enabled if both the guest and the host support it; disabled, which causes HPT resizing to be disabled regardless of guest and host support; and required, which prevents the guest from starting unless both the guest and the host support HPT resizing. If the attribute is not defined, the hypervisor default will be used. Since 3.10.0 (QEMU/KVM only).

The optional maxpagesize subelement can be used to limit the usable page size for HPT guests. Common values are 64 KiB, 16 MiB and 16 GiB; when not specified, the hypervisor default will be used. Since 4.5.0 (QEMU/KVM only).

vmcoreinfo

Enable QEMU vmcoreinfo device to let the guest kernel save debug details. Since 4.4.0 (QEMU only)

htm

Configure HTM (Hardware Transational Memory) availability for pSeries guests. Possible values for the state attribute are on and off. If the attribute is not defined, the hypervisor default will be used. Since 4.6.0 (QEMU/KVM only)

nested-hv

Configure nested HV availability for pSeries guests. This needs to be enabled from the host (L0) in order to be effective; having HV support in the (L1) guest is very desirable if it's planned to run nested (L2) guests inside it, because it will result in those nested guests having much better performance than they would when using KVM PR or TCG. Possible values for the state attribute are on and off. If the attribute is not defined, the hypervisor default will be used. Since 4.10.0 (QEMU/KVM only)

msrs

Some guests might require ignoring unknown Model Specific Registers (MSRs) reads and writes. It's possible to switch this by setting unknown attribute of msrs to ignore. If the attribute is not defined, or set to fault, unknown reads and writes will not be ignored. Since 5.1.0 (bhyve only)

ccf-assist

Configure ccf-assist (Count Cache Flush Assist) availability for pSeries guests. Possible values for the state attribute are on and off. If the attribute is not defined, the hypervisor default will be used. Since 5.9.0 (QEMU/KVM only)

cfpc

Configure cfpc (Cache Flush on Privilege Change) availability for pSeries guests. Possible values for the value attribute are broken (no protection), workaround (software workaround available) and fixed (fixed in hardware). If the attribute is not defined, the hypervisor default will be used. Since 6.3.0 (QEMU/KVM only)

sbbc

Configure sbbc (Speculation Barrier Bounds Checking) availability for pSeries guests. Possible values for the value attribute are broken (no protection), workaround (software workaround available) and fixed (fixed in hardware). If the attribute is not defined, the hypervisor default will be used. Since 6.3.0 (QEMU/KVM only)

ibs

Configure ibs (Indirect Branch Speculation) availability for pSeries guests. Possible values for the value attribute are broken (no protection), workaround (count cache flush), fixed-ibs (fixed by serializing indirect branches), fixed-ccd (fixed by disabling the cache count) and fixed-na (fixed in hardware - no longer applicable). If the attribute is not defined, the hypervisor default will be used. Since 6.3.0 (QEMU/KVM only)

tcg

Various features to change the behavior of the TCG accelerator.

Feature Description Value Since
tb-cache The size of translation block cache size an integer (a multiple of MiB) 8.0.0
async-teardown

Depending on the enabled attribute (values yes, no) enable or disable QEMU asynchronous teardown to improve memory reclaiming on a guest. Since 9.6.0 (QEMU only)

Time keeping

The guest clock is typically initialized from the host clock. Most operating systems expect the hardware clock to be kept in UTC, and this is the default. Windows, however, expects it to be in so called 'localtime'.

...
<clock offset='localtime'>
  <timer name='rtc' tickpolicy='catchup' track='guest'>
    <catchup threshold='123' slew='120' limit='10000'/>
  </timer>
  <timer name='pit' tickpolicy='delay'/>
</clock>
...
clock

The offset attribute takes four possible values, allowing fine grained control over how the guest clock is synchronized to the host. NB, not all hypervisors support all modes.

utc

The guest clock will always be synchronized to UTC when booted. Since 0.9.11 'utc' mode can be converted to 'variable' mode, which can be controlled by using the adjustment attribute. If the value is 'reset', the conversion is never done (not all hypervisors can synchronize to UTC on each boot; use of 'reset' will cause an error on those hypervisors). A numeric value forces the conversion to 'variable' mode using the value as the initial adjustment. The default adjustment is hypervisor specific.

localtime

The guest clock will be synchronized to the host's configured timezone when booted, if any. Since 0.9.11, the adjustment attribute behaves the same as in 'utc' mode.

timezone

The guest clock will be synchronized to the requested timezone using the timezone attribute. Since 0.7.7

variable

The guest clock will have an arbitrary offset applied relative to UTC or localtime, depending on the basis attribute. The delta relative to UTC (or localtime) is specified in seconds, using the adjustment attribute. The guest is free to adjust the RTC over time and expect that it will be honored at next reboot. This is in contrast to 'utc' and 'localtime' mode (with the optional attribute adjustment='reset'), where the RTC adjustments are lost at each reboot. Since 0.7.7 Since 0.9.11 the basis attribute can be either 'utc' (default) or 'localtime'.

absolute

The guest clock will be always set to the value of the start attribute at startup of the domain. The start attribute takes an epoch timestamp. Since 8.4.0.

A clock may have zero or more timer sub-elements. Since 0.8.0

timer

Each timer element requires a name attribute, and has other optional attributes that depend on the name specified. Various hypervisors support different combinations of attributes.

name

The name attribute selects which timer is being modified, and can be one of "platform" (currently unsupported), "hpet" (xen, qemu, lxc), "kvmclock" (qemu), "pit" (qemu), "rtc" (qemu, lxc), "tsc" (xen, qemu -since 3.2.0 ), "hypervclock" (qemu - since 1.2.2 ) or "armvtimer" (qemu - since 6.1.0 ). The hypervclock timer adds support for the reference time counter and the reference page for iTSC feature for guests running the Microsoft Windows operating system.

track

The track attribute specifies what the timer tracks, and can be "boot", "guest", or "wall", or "realtime". Only valid for name="rtc" or name="platform".

tickpolicy

The tickpolicy attribute determines what happens when QEMU misses a deadline for injecting a tick to the guest. This can happen, for example, because the guest was paused.

delay

Continue to deliver ticks at the normal rate. The guest OS will not notice anything is amiss, as from its point of view time will have continued to flow normally. The time in the guest should now be behind the time in the host by exactly the amount of time during which ticks have been missed.

catchup

Deliver ticks at a higher rate to catch up with the missed ticks. The guest OS will not notice anything is amiss, as from its point of view time will have continued to flow normally. Once the timer has managed to catch up with all the missing ticks, the time in the guest and in the host should match.

merge

Merge the missed tick(s) into one tick and inject. The guest time may be delayed, depending on how the OS reacts to the merging of ticks

discard

Throw away the missed ticks and continue with future injection normally. The guest OS will see the timer jump ahead by a potentially quite significant amount all at once, as if the intervening chunk of time had simply not existed; needless to say, such a sudden jump can easily confuse a guest OS which is not specifically prepared to deal with it. Assuming the guest OS can deal correctly with the time jump, the time in the guest and in the host should now match.

If the policy is "catchup", there can be further details in the catchup sub-element.

catchup

The catchup element has three optional attributes, each a positive integer. The attributes are threshold, slew, and limit.

Note that hypervisors are not required to support all policies across all time sources

frequency

The frequency attribute is an unsigned integer specifying the frequency at which name="tsc" runs.

mode

The mode attribute controls how the name="tsc" timer is managed, and can be "auto", "native", "emulate", "paravirt", or "smpsafe". Other timers are always emulated.

present

The present attribute can be "yes" or "no" to specify whether a particular timer is available to the guest.

Performance monitoring events

Some platforms allow monitoring of performance of the virtual machine and the code executed inside. To enable the performance monitoring events you can either specify them in the perf element or enable them via virDomainSetPerfEvents API. The performance values are then retrieved using the virConnectGetAllDomainStats API. Since 2.0.0

...
<perf>
  <event name='cmt' enabled='yes'/>
  <event name='mbmt' enabled='no'/>
  <event name='mbml' enabled='yes'/>
  <event name='cpu_cycles' enabled='no'/>
  <event name='instructions' enabled='yes'/>
  <event name='cache_references' enabled='no'/>
  <event name='cache_misses' enabled='no'/>
  <event name='branch_instructions' enabled='no'/>
  <event name='branch_misses' enabled='no'/>
  <event name='bus_cycles' enabled='no'/>
  <event name='stalled_cycles_frontend' enabled='no'/>
  <event name='stalled_cycles_backend' enabled='no'/>
  <event name='ref_cpu_cycles' enabled='no'/>
  <event name='cpu_clock' enabled='no'/>
  <event name='task_clock' enabled='no'/>
  <event name='page_faults' enabled='no'/>
  <event name='context_switches' enabled='no'/>
  <event name='cpu_migrations' enabled='no'/>
  <event name='page_faults_min' enabled='no'/>
  <event name='page_faults_maj' enabled='no'/>
  <event name='alignment_faults' enabled='no'/>
  <event name='emulation_faults' enabled='no'/>
</perf>
...
event name Description stats parameter name
cmt usage of l3 cache in bytes by applications running on the platform perf.cmt
mbmt total system bandwidth from one level of cache perf.mbmt
mbml bandwidth of memory traffic for a memory controller perf.mbml
cpu_cycles the count of CPU cycles (total/elapsed) perf.cpu_cycles
instructions the count of instructions by applications running on the platform perf.instructions
cache_references the count of cache hits by applications running on the platform perf.cache_references
cache_misses the count of cache misses by applications running on the platform perf.cache_misses
branch_instructions the count of branch instructions by applications running on the platform perf.branch_instructions
branch_misses the count of branch misses by applications running on the platform perf.branch_misses
bus_cycles the count of bus cycles by applications running on the platform perf.bus_cycles
stalled_cycles_frontend the count of stalled CPU cycles in the frontend of the instruction processor pipeline by applications running on the platform perf.stalled_cycles_frontend
stalled_cycles_backend the count of stalled CPU cycles in the backend of the instruction processor pipeline by applications running on the platform perf.stalled_cycles_backend
ref_cpu_cycles the count of total CPU cycles not affected by CPU frequency scaling by applications running on the platform perf.ref_cpu_cycles
cpu_clock the count of CPU clock time, as measured by a monotonic high-resolution per-CPU timer, by applications running on the platform perf.cpu_clock
task_clock the count of task clock time, as measured by a monotonic high-resolution CPU timer, specific to the task that is run by applications running on the platform perf.task_clock
page_faults the count of page faults by applications running on the platform. This includes minor, major, invalid and other types of page faults perf.page_faults
context_switches the count of context switches by applications running on the platform perf.context_switches
cpu_migrations the count of CPU migrations, that is, where the process moved from one logical processor to another, by applications running on the platform perf.cpu_migrations
page_faults_min the count of minor page faults, that is, where the page was present in the page cache, and therefore the fault avoided loading it from storage, by applications running on the platform perf.page_faults_min
page_faults_maj the count of major page faults, that is, where the page was not present in the page cache, and therefore had to be fetched from storage, by applications running on the platform perf.page_faults_maj
alignment_faults the count of alignment faults, that is when the load or store is not aligned properly, by applications running on the platform perf.alignment_faults
emulation_faults the count of emulation faults, that is when the kernel traps on unimplemented instructions and emulates them for user space, by applications running on the platform

perf.emulation_faults

Devices

The final set of XML elements are all used to describe devices provided to the guest domain. All devices occur as children of the main devices element. Since 0.1.3

...
<devices>
  <emulator>/usr/lib/xen/bin/qemu-dm</emulator>
</devices>
...
emulator

The contents of the emulator element specify the fully qualified path to the device model emulator binary. The capabilities XML specifies the recommended default emulator to use for each particular domain type / architecture combination.

To help users identifying devices they care about, every device can have direct child alias element which then has name attribute where users can store identifier for the device. The identifier has to have "ua-" prefix and must be unique within the domain. Additionally, the identifier must consist only of the following characters: [a-zA-Z0-9_-]. Since 3.9.0

<devices>
  <disk type='file'>
    <alias name='ua-myDisk'/>
  </disk>
  <interface type='network' trustGuestRxFilters='yes'>
    <alias name='ua-myNIC'/>
  </interface>
  ...
</devices>

Hard drives, floppy disks, CDROMs

Any device that looks like a disk, be it a floppy, harddisk, cdrom, or paravirtualized driver is specified via the disk element.

...
<devices>
  <disk type='file' snapshot='external'>
    <driver name="tap" type="aio" cache="default"/>
    <source file='/var/lib/xen/images/fv0' startupPolicy='optional'>
      <seclabel relabel='no'/>
    </source>
    <target dev='hda' bus='ide'/>
    <iotune>
      <total_bytes_sec>10000000</total_bytes_sec>
      <read_iops_sec>400000</read_iops_sec>
      <write_iops_sec>100000</write_iops_sec>
    </iotune>
    <boot order='2'/>
    <encryption type='...'>
      ...
    </encryption>
    <shareable/>
    <serial>
      ...
    </serial>
  </disk>
    ...
  <disk type='network'>
    <driver name="qemu" type="raw" io="threads" ioeventfd="on" event_idx="off"/>
    <source protocol="sheepdog" name="image_name">
      <host name="hostname" port="7000"/>
    </source>
    <target dev="hdb" bus="ide"/>
    <boot order='1'/>
    <transient/>
    <address type='drive' controller='0' bus='1' unit='0'/>
  </disk>
  <disk type='network'>
    <driver name="qemu" type="raw"/>
    <source protocol="rbd" name="image_name2">
      <host name="hostname" port="7000"/>
      <snapshot name="snapname"/>
      <config file="/path/to/file"/>
      <auth username='myuser'>
        <secret type='ceph' usage='mypassid'/>
      </auth>
    </source>
    <target dev="hdc" bus="ide"/>
  </disk>
  <disk type='block' device='cdrom'>
    <driver name='qemu' type='raw'/>
    <target dev='hdd' bus='ide' tray='open'/>
    <readonly/>
  </disk>
  <disk type='network' device='cdrom'>
    <driver name='qemu' type='raw'/>
    <source protocol="http" name="url_path" query="foo=bar&amp;baz=flurb>
      <host name="hostname" port="80"/>
      <cookies>
        <cookie name="test">somevalue</cookie>
      </cookies>
      <readahead size='65536'/>
      <timeout seconds='6'/>
    </source>
    <target dev='hde' bus='ide' tray='open'/>
    <readonly/>
  </disk>
  <disk type='network' device='cdrom'>
    <driver name='qemu' type='raw'/>
    <source protocol="https" name="url_path">
      <host name="hostname" port="443"/>
      <ssl verify="no"/>
    </source>
    <target dev='hdf' bus='ide' tray='open'/>
    <readonly/>
  </disk>
  <disk type='network' device='cdrom'>
    <driver name='qemu' type='raw'/>
    <source protocol="ftp" name="url_path">
      <host name="hostname" port="21"/>
    </source>
    <target dev='hdg' bus='ide' tray='open'/>
    <readonly/>
  </disk>
  <disk type='network' device='cdrom'>
    <driver name='qemu' type='raw'/>
    <source protocol="ftps" name="url_path">
      <host name="hostname" port="990"/>
    </source>
    <target dev='hdh' bus='ide' tray='open'/>
    <readonly/>
  </disk>
  <disk type='network' device='cdrom'>
    <driver name='qemu' type='raw'/>
    <source protocol="tftp" name="url_path">
      <host name="hostname" port="69"/>
    </source>
    <target dev='hdi' bus='ide' tray='open' rotation_rate='7200'/>
    <readonly/>
  </disk>
  <disk type='block' device='lun'>
    <driver name='qemu' type='raw'/>
    <source dev='/dev/sda'>
      <slices>
        <slice type='storage' offset='12345' size='123'/>
      </slices>
      <reservations managed='no'>
        <source type='unix' path='/path/to/qemu-pr-helper' mode='client'/>
      </reservations>
    </source>
    <target dev='sda' bus='scsi' rotation_rate='1'/>
    <address type='drive' controller='0' bus='0' target='3' unit='0'/>
  </disk>
  <disk type='block' device='disk'>
    <driver name='qemu' type='raw'/>
    <source dev='/dev/sda'/>
    <geometry cyls='16383' heads='16' secs='63' trans='lba'/>
    <blockio logical_block_size='512' physical_block_size='4096'/>
    <target dev='hdj' bus='ide'/>
  </disk>
  <disk type='volume' device='disk'>
    <driver name='qemu' type='raw'/>
    <source pool='blk-pool0' volume='blk-pool0-vol0'/>
    <target dev='hdk' bus='ide'/>
  </disk>
  <disk type='network' device='disk'>
    <driver name='qemu' type='raw'/>
    <source protocol='iscsi' name='iqn.2013-07.com.example:iscsi-nopool/2'>
      <host name='example.com' port='3260'/>
      <auth username='myuser'>
        <secret type='iscsi' usage='libvirtiscsi'/>
      </auth>
    </source>
    <target dev='vda' bus='virtio'/>
  </disk>
  <disk type='network' device='lun'>
    <driver name='qemu' type='raw'/>
    <source protocol='iscsi' name='iqn.2013-07.com.example:iscsi-nopool/1'>
      <host name='example.com' port='3260'/>
      <auth username='myuser'>
        <secret type='iscsi' usage='libvirtiscsi'/>
      </auth>
    </source>
    <target dev='sdb' bus='scsi'/>
  </disk>
  <disk type='network' device='disk'>
    <driver name='qemu' type='raw'/>
    <source protocol='nfs' name='PATH'>
      <host name='example.com'/>
      <identity user='USER' group='GROUP'/>
    </source>
    <target dev='vda' bus='virtio'/>
  </disk>
  <disk type='network' device='lun'>
    <driver name='qemu' type='raw'/>
    <source protocol='iscsi' name='iqn.2013-07.com.example:iscsi-nopool/0'>
      <host name='example.com' port='3260'/>
      <initiator>
        <iqn name='iqn.2013-07.com.example:client'/>
      </initiator>
    </source>
    <target dev='sdb' bus='scsi'/>
  </disk>
  <disk type='volume' device='disk'>
    <driver name='qemu' type='raw'/>
    <source pool='iscsi-pool' volume='unit:0:0:1' mode='host'/>
    <target dev='vdb' bus='virtio'/>
  </disk>
  <disk type='volume' device='disk'>
    <driver name='qemu' type='raw'/>
    <source pool='iscsi-pool' volume='unit:0:0:2' mode='direct'/>
    <target dev='vdc' bus='virtio'/>
  </disk>
  <disk type='file' device='disk'>
    <driver name='qemu' type='qcow2' queues='4' queue_size='256' />
    <source file='/var/lib/libvirt/images/domain.qcow'/>
    <backingStore type='file'>
      <format type='qcow2'/>
      <source file='/var/lib/libvirt/images/snapshot.qcow'/>
      <backingStore type='block'>
        <format type='raw'/>
        <source dev='/dev/mapper/base'/>
        <backingStore/>
      </backingStore>
    </backingStore>
    <target dev='vdd' bus='virtio'/>
  </disk>
  <disk type='nvme' device='disk'>
    <driver name='qemu' type='raw'/>
    <source type='pci' managed='yes' namespace='1'>
      <address domain='0x0000' bus='0x01' slot='0x00' function='0x0'/>
    </source>
    <target dev='vde' bus='virtio'/>
  </disk>
  <disk type='vhostuser' device='disk'>
    <driver name='qemu' type='raw'/>
    <source type='unix' path='/tmp/vhost-blk.sock'>
      <reconnect enabled='yes' timeout='10'/>
    </source>
    <target dev='vdf' bus='virtio'/>
  </disk>
</devices>
...
disk

The disk element is the main container for describing disks and supports the following attributes:

type

Valid values are "file", "block", "dir" ( since 0.7.5 ), "network" ( since 0.8.7 ), or "volume" ( since 1.0.5 ), or "nvme" ( since 6.0.0 ), or "vhostuser" ( since 7.1.0 ) and refer to the underlying source for the disk. Since 0.0.3

device

Indicates how the disk is to be exposed to the guest OS. Possible values for this attribute are "floppy", "disk", "cdrom", and "lun", defaulting to "disk".

Using "lun" ( since 0.9.10 ) is only valid when the type is "block" or "network" for protocol='iscsi' or when the type is "volume" when using an iSCSI source pool for mode "host" or as an NPIV virtual Host Bus Adapter (vHBA) using a Fibre Channel storage pool. Configured in this manner, the LUN behaves identically to "disk", except that generic SCSI commands from the guest are accepted and passed through to the physical device. Also note that device='lun' will only be recognized for actual raw devices, but never for individual partitions or LVM partitions (in those cases, the kernel will reject the generic SCSI commands, making it identical to device='disk'). Since 0.1.4

model

Indicates the emulated device model of the disk. Typically this is indicated solely by the bus property but for bus "virtio" the model can be specified further with "virtio-transitional", "virtio-non-transitional", or "virtio". See Virtio transitional devices for more details. Since 5.2.0

rawio

Indicates whether the disk needs rawio capability. Valid settings are "yes" or "no" (default is "no"). If any one disk in a domain has rawio='yes', rawio capability will be enabled for all disks in the domain (because, in the case of QEMU, this capability can only be set on a per-process basis). This attribute is only valid when device is "lun". NB, rawio intends to confine the capability per-device, however, current QEMU implementation gives the domain process broader capability than that (per-process basis, affects all the domain disks). To confine the capability as much as possible for QEMU driver as this stage, sgio is recommended, it's more secure than rawio. Since 0.9.10

sgio

If supported by the hypervisor and OS, indicates whether unprivileged SG_IO commands are filtered for the disk. Valid settings are "filtered" or "unfiltered" where the default is "filtered". Only available when the device is 'lun'. Since 1.0.2

snapshot

Indicates the default behavior of the disk during disk snapshots: internal requires a file format such as qcow2 that can store both the snapshot and the data changes since the snapshot; external will separate the snapshot from the live data; no means the disk will not participate in snapshots; and manual allows snapshotting done via an unmanaged storage provider. Read-only disks default to no, while the default for other disks depends on the hypervisor's capabilities. Some hypervisors allow a per-snapshot choice as well, during domain snapshot creation. Not all snapshot modes are supported; for example, enabling snapshots with a transient disk generally does not make sense. Since 0.9.5

source

Representation of the disk source depends on the disk type attribute value as follows:

file

The file attribute specifies the fully-qualified path to the file holding the disk. Since 0.0.3

Since 9.0.0 a new optional attribute fdgroup can be added instructing to access the disk via file descriptors associated to the domain object via the virDomainFDAssociate() API rather than opening the files. The files do not necessarily have to be accessible by libvirt via the filesystem. The filename passed via file can still be used to generate paths to write into image metadata when doing block operations but libvirt will not access these natively.

block

The dev attribute specifies the fully-qualified path to the host device to serve as the disk. Since 0.0.3

dir

The dir attribute specifies the fully-qualified path to the directory to use as the disk. Since 0.7.5

network

The protocol attribute specifies the protocol to access to the requested image. Possible values are "nbd", "iscsi", "rbd", "sheepdog", "gluster", "vxhs", "nfs", "http", "https", "ftp", ftps", or "tftp".

For any protocol other than nbd an additional attribute name is mandatory to specify which volume/image will be used.

For "nbd", the name attribute is optional. TLS transport for NBD can be enabled by setting the tls attribute to yes. For the QEMU hypervisor, usage of a TLS environment can also be globally controlled on the host by the nbd_tls and nbd_tls_x509_cert_dir in /etc/libvirt/qemu.conf. ('tls' Since 4.5.0 ) Since 8.2.0 the optional attribute tlsHostname can be used to override the expected host name of the NBD server used for TLS certificate verification.

For protocols http and https an optional attribute query specifies the query string. ( Since 6.2.0 )

For "iscsi" ( since 1.0.4 ), the name attribute may include a logical unit number, separated from the target's name by a slash (e.g., iqn.2013-07.com.example:iscsi-pool/1). If not specified, the default LUN is zero.

For "vxhs" ( since 3.8.0 ), the name is the UUID of the volume, assigned by the HyperScale server. Additionally, an optional attribute tls (QEMU only) can be used to control whether a VxHS block device would utilize a hypervisor configured TLS X.509 certificate environment in order to encrypt the data channel. For the QEMU hypervisor, usage of a TLS environment can also be globally controlled on the host by the vxhs_tls and vxhs_tls_x509_cert_dir or default_tls_x509_cert_dir settings in the file /etc/libvirt/qemu.conf. If vxhs_tls is enabled, then unless the domain tls attribute is set to "no", libvirt will use the host configured TLS environment. If the tls attribute is set to "yes", then regardless of the qemu.conf setting, TLS authentication will be attempted.

Since 0.8.7

volume

The underlying disk source is represented by attributes pool and volume. Attribute pool specifies the name of the storage pool (managed by libvirt) where the disk source resides. Attribute volume specifies the name of storage volume (managed by libvirt) used as the disk source. The value for the volume attribute will be the output from the "Name" column of a virsh vol-list [pool-name] command.

Use the attribute mode ( since 1.1.1 ) to indicate how to represent the LUN as the disk source. Valid values are "direct" and "host". If mode is not specified, the default is to use "host". Using "direct" as the mode value indicates to use the storage pool's source element host attribute as the disk source to generate the libiscsi URI (e.g. 'file=iscsi://example.com:3260/iqn.2013-07.com.example:iscsi-pool/1'). Using "host" as the mode value indicates to use the LUN's path as it shows up on host (e.g. 'file=/dev/disk/by-path/ip-example.com:3260-iscsi-iqn.2013-07.com.example:iscsi-pool-lun-1'). Using a LUN from an iSCSI source pool provides the same features as a disk configured using type 'block' or 'network' and device of 'lun' with respect to how the LUN is presented to and may be used by the guest. Since 1.0.5

nvme

To specify disk source for NVMe disk the source element has the following attributes:

type

The type of address specified in address sub-element. Currently, only pci value is accepted.

managed

This attribute instructs libvirt to detach NVMe controller automatically on domain startup (yes) or expect the controller to be detached by system administrator (no).

namespace

The namespace ID which should be assigned to the domain. According to NVMe standard, namespace numbers start from 1, including.

The difference between <disk type='nvme'> and <hostdev/> is that the latter is plain host device assignment with all its limitations (e.g. no live migration), while the former makes hypervisor to run the NVMe disk through hypervisor's block layer thus enabling all features provided by the layer (e.g. snapshots, domain migration, etc.). Moreover, since the NVMe disk is unbinded from its PCI driver, the host kernel storage stack is not involved (compared to passing say /dev/nvme0n1 via <disk type='block'> and therefore lower latencies can be achieved.

vhostuser

Enables the hypervisor to connect to another process using vhost-user protocol. Requires shared memory configured for the VM, for more details see access mode for memoryBacking element (See Memory Backing).

The source element has following mandatory attributes:

type

The type of char device. Currently only unix type is supported.

path

Path to the unix socket to be used as disk source.

Note that the vhost server replaces both the disk frontend and backend thus almost all of the disk properties can't be configured via the <disk> XML for this disk type. Additionally features such as blockjobs, incremental backups and snapshots are not supported for this disk type.

With "file", "block", and "volume", one or more optional sub-elements seclabel (See Security label) can be used to override the domain security labeling policy for just that source file. (NB, for "volume" type disk, seclabel is only valid when the specified storage volume is of 'file' or 'block' type). since 0.9.9

The source element may also have the index attribute with same semantics the index attribute of backingStore.

The source element may contain the following sub elements:

host

When the disk type is "network", the source may have zero or more host sub-elements used to specify the hosts to connect. The host element supports 4 attributes, viz. "name", "port", "transport" and "socket", which specify the hostname, the port number, transport type and path to socket, respectively. The meaning of this element and the number of the elements depend on the protocol attribute.

Protocol Meaning Number of hosts Default port
nbd a server running nbd-server only one 10809
iscsi an iSCSI server only one 3260
rbd monitor servers of RBD one or more librados default
sheepdog one of the sheepdog servers (default is localhost:7000) zero or one 7000
gluster a server running glusterd daemon one or more ( Since 2.1.0 ), just one prior to that 24007
vxhs a server running Veritas HyperScale daemon only one 9999
nfs a server running Network File System only one ( Since 7.0.0 ) must be omitted

gluster supports "tcp", "rdma", "unix" as valid values for the transport attribute. nbd supports "tcp" and "unix". Others only support "tcp". If nothing is specified, "tcp" is assumed. If the transport is "unix", the socket attribute specifies the path to an AF_UNIX socket. nfs only supports the use of a "tcp" transport, and does not support using a port at all so it must be omitted.

snapshot

The name attribute of snapshot element can optionally specify an internal snapshot name to be used as the source for storage protocols. Supported for 'rbd' since 1.2.11 (QEMU only).

config

The file attribute for the config element provides a fully qualified path to a configuration file to be provided as a parameter to the client of a networked storage protocol. Supported for 'rbd' since 1.2.11 (QEMU only).

auth

Since libvirt 3.9.0 , the auth element is supported for a disk type "network" that is using a source element with the protocol attributes "rbd" or "iscsi". If present, the auth element provides the authentication credentials needed to access the source. It includes a mandatory attribute username, which identifies the username to use during authentication, as well as a sub-element secret with mandatory attribute type, to tie back to a libvirt secret object that holds the actual password or other credentials (the domain XML intentionally does not expose the password, only the reference to the object that does manage the password). Known secret types are "ceph" for Ceph RBD network sources and "iscsi" for CHAP authentication of iSCSI targets. Both will require either a uuid attribute with the UUID of the secret object or a usage attribute matching the key that was specified in the secret object.

encryption

Since libvirt 3.9.0 , the encryption can be a sub-element of the source element for encrypted storage sources. If present, specifies how the storage source is encrypted See the Storage Encryption page for more information. Note that the 'qcow' format of encryption is broken and thus is no longer supported for use with disk images. ( Since libvirt 4.5.0 )

reservations

Since libvirt 4.4.0 , the reservations can be a sub-element of the source element for storage sources (QEMU driver only). If present it enables persistent reservations for SCSI based disks. The element has one mandatory attribute managed with accepted values yes and no. If managed is enabled libvirt prepares and manages any resources needed. When the persistent reservations are unmanaged, then the hypervisor acts as a client and the path to the server socket must be provided in the child element source, which currently accepts only the following attributes: type with one value unix, path path to the socket, and finally mode which accepts one value client specifying the role of hypervisor. It's recommended to allow libvirt manage the persistent reservations.

initiator

Since libvirt 4.7.0 , the initiator element is supported for a disk type "network" that is using a source element with the protocol attribute "iscsi". If present, the initiator element provides the initiator IQN needed to access the source via mandatory attribute name.

address

For disk of type nvme this element specifies the PCI address of the host NVMe controller. Since 6.0.0

slices

The slices element using its slice sub-elements allows configuring offset and size of either the location of the image format (slice type='storage') inside the storage source or the guest data inside the image format container (future expansion). The offset and size values are in bytes. Since 6.1.0

ssl

For https and ftps accessed storage it's possible to tweak the SSL transport parameters with this element. The verify attribute allows to turn on or off SSL certificate validation. Supported values are yes and no. Since 6.2.0

cookies

For http and https accessed storage it's possible to pass one or more cookies. The cookie name and value must conform to the HTTP specification. Since 6.2.0

readahead

The readahead element has a size attribute which specifies the size of the readahead buffer in bytes for protocols which support it. Note that '0' is considered as if the value is not provided. Since 6.2.0

timeout

The timeout element has a seconds attribute which specifies the connection timeout in seconds for protocols which support it. Note that '0' is considered as if the value is not provided. Since 6.2.0

identity

When using an nfs protocol, this is used to provide information on the configuration of the user and group. The element has two attributes, user and group. The user can provide these elements as user or group strings, or as user and group ID numbers directly if the string is formatted using a "+" at the beginning of the ID number. If either of these attributes is omitted, then that field is assumed to be the default value for the current system. If both user and group are intended to be default, then the entire element may be omitted.

reconnect

For disk type vhostuser configures reconnect timeout if the connection is lost. This is set with the two mandatory attributes enabled and timeout. For disk type network and protocol nbd the QEMU NBD reconnect delay can be set via attribute delay:

enabled

If the reconnect feature is enabled, accepts yes and no

timeout

The amount of seconds after which hypervisor tries to reconnect.

delay

Only for NBD hosts. The amount of seconds during which all requests are paused and will be rerun after a successful reconnect. After that time, any delayed requests and all future requests before a successful reconnect will immediately fail. If not set the default QEMU value is 0.

For a "file" or "volume" disk type which represents a cdrom or floppy (the device attribute), it is possible to define policy what to do with the disk if the source file is not accessible. (NB, startupPolicy is not valid for "volume" disk unless the specified storage volume is of "file" type). This is done by the startupPolicy attribute ( since 0.9.7 ), accepting these values:

mandatory fail if missing for any reason (the default)
requisite fail if missing on boot up, drop if missing on migrate/restore/revert
optional drop if missing at any start attempt

Since 1.1.2 the startupPolicy is extended to support hard disks besides cdrom and floppy. On guest cold bootup, if a certain disk is not accessible or its disk chain is broken, with startupPolicy 'optional' the guest will drop this disk. This feature doesn't support migration currently.

backingStore

This element describes the backing store used by the disk specified by sibling source element. Since 1.2.4. If the hypervisor driver does not support the backingStoreInput ( Since 5.10.0 ) domain feature the backingStore is ignored on input and only used for output to describe the detected backing chains of running domains. If backingStoreInput is supported the backingStore is used as the backing image of source or other backingStore overriding any backing image information recorded in the image metadata. An empty backingStore element means the sibling source is self-contained and is not based on any backing store. For the detected backing chain information to be accurate, the backing format must be correctly specified in the metadata of each file of the chain (files created by libvirt satisfy this property, but using existing external files for snapshot or block copy operations requires the end user to pre-create the file correctly). The following attributes are supported in backingStore:

type

The type attribute represents the type of disk used by the backing store, see disk type attribute above for more details and possible values.

index

This attribute is only valid in output (and ignored on input) and it can be used to refer to a specific part of the disk chain when doing block operations (such as via the virDomainBlockRebase API). For example, vda[2] refers to the backing store with index='2' of the disk with vda target.

Moreover, backingStore supports the following sub-elements:

format

The format element contains type attribute which specifies the internal format of the backing store, such as raw or qcow2.

The format element can contain metadata_cache subelement, which has identical semantics to the identically named subelement of driver of a disk.

source

This element has the same structure as the source element in disk. It specifies which file, device, or network location contains the data of the described backing store.

backingStore

If the backing store is not self-contained, the next element in the chain is described by nested backingStore element.

mirror

This element is present if the hypervisor has started a long-running block job operation, where the mirror location in the source sub-element will eventually have the same contents as the source, and with the file format in the sub-element format (which might differ from the format of the source). The details of the source sub-element are determined by the type attribute of the mirror, similar to what is done for the overall disk device element. The job attribute mentions which API started the operation ("copy" for the virDomainBlockRebase API, or "active-commit" for the virDomainBlockCommit API), since 1.2.7 . The attribute ready, if present, tracks progress of the job: yes if the disk is known to be ready to pivot, or, since 1.2.7 , abort or pivot if the job is in the process of completing. If ready is not present, the disk is probably still copying. For now, this element only valid in output; it is ignored on input. The source sub-element exists for all two-phase jobs since 1.2.6 . Older libvirt supported only block copy to a file, since 0.9.12 ; for compatibility with older clients, such jobs include redundant information in the attributes file and format in the mirror element.

target

The target element controls the bus / device under which the disk is exposed to the guest OS. The dev attribute indicates the "logical" device name. The actual device name specified is not guaranteed to map to the device name in the guest OS. Treat it as a device ordering hint. The optional bus attribute specifies the type of disk device to emulate; possible values are driver specific, with typical values being "ide", "scsi", "virtio", "xen", "usb", "sata", or "sd" "sd" since 1.1.2 . If omitted, the bus type is inferred from the style of the device name (e.g. a device named 'sda' will typically be exported using a SCSI bus). The optional attribute tray indicates the tray status of the removable disks (i.e. CDROM or Floppy disk), the value can be either "open" or "closed", defaults to "closed". NB, the value of tray could be updated while the domain is running. The optional attribute removable sets the removable flag for USB or SCSI disks, and its value can be either "on" or "off", defaulting to "off". The optional attribute rotation_rate sets the rotation rate of the storage for disks on a SCSI, IDE, or SATA bus. Values in the range 1025 to 65534 are used to indicate rotational media speed in revolutions per minute. A value of one is used to indicate solid state, or otherwise non-rotational, storage. These values are not required to match the values of the underlying host storage. Since 0.0.3; bus attribute since 0.4.3; tray attribute since 0.9.11; "usb" attribute value since after 0.4.4; "sata" attribute value since 0.9.7; "removable" attribute value since 1.1.3; "rotation_rate" attribute value since 7.3.0

iotune

The optional iotune element provides the ability to provide additional per-device I/O tuning, with values that can vary for each device (contrast this to the blkiotune element (See Block I/O Tuning), which applies globally to the domain). Currently, the only tuning available is Block I/O throttling for qemu. This element has optional sub-elements; any sub-element not specified or given with a value of 0 implies no limit. Since 0.9.8

total_bytes_sec

The optional total_bytes_sec element is the total throughput limit in bytes per second. This cannot appear with read_bytes_sec or write_bytes_sec.

read_bytes_sec

The optional read_bytes_sec element is the read throughput limit in bytes per second.

write_bytes_sec

The optional write_bytes_sec element is the write throughput limit in bytes per second.

total_iops_sec

The optional total_iops_sec element is the total I/O operations per second. This cannot appear with read_iops_sec or write_iops_sec.

read_iops_sec

The optional read_iops_sec element is the read I/O operations per second.

write_iops_sec

The optional write_iops_sec element is the write I/O operations per second.

total_bytes_sec_max

The optional total_bytes_sec_max element is the maximum total throughput limit in bytes per second. This cannot appear with read_bytes_sec_max or write_bytes_sec_max.

read_bytes_sec_max

The optional read_bytes_sec_max element is the maximum read throughput limit in bytes per second.

write_bytes_sec_max

The optional write_bytes_sec_max element is the maximum write throughput limit in bytes per second.

total_iops_sec_max

The optional total_iops_sec_max element is the maximum total I/O operations per second. This cannot appear with read_iops_sec_max or write_iops_sec_max.

read_iops_sec_max

The optional read_iops_sec_max element is the maximum read I/O operations per second.

write_iops_sec_max

The optional write_iops_sec_max element is the maximum write I/O operations per second.

size_iops_sec

The optional size_iops_sec element is the size of I/O operations per second.

Throughput limits since 1.2.11 and QEMU 1.7

group_name

The optional group_name provides the ability to share I/O throttling quota between multiple drives. This prevents end-users from circumventing a hosting provider's throttling policy by splitting 1 large drive in N small drives and getting N times the normal throttling quota. Any name may be used.

group_name since 3.0.0 and QEMU 2.4

total_bytes_sec_max_length

The optional total_bytes_sec_max_length element is the maximum duration in seconds for the total_bytes_sec_max burst period. Only valid when the total_bytes_sec_max is set.

read_bytes_sec_max_length

The optional read_bytes_sec_max_length element is the maximum duration in seconds for the read_bytes_sec_max burst period. Only valid when the read_bytes_sec_max is set.

write_bytes_sec_max

The optional write_bytes_sec_max_length element is the maximum duration in seconds for the write_bytes_sec_max burst period. Only valid when the write_bytes_sec_max is set.

total_iops_sec_max_length

The optional total_iops_sec_max_length element is the maximum duration in seconds for the total_iops_sec_max burst period. Only valid when the total_iops_sec_max is set.

read_iops_sec_max_length

The optional read_iops_sec_max_length element is the maximum duration in seconds for the read_iops_sec_max burst period. Only valid when the read_iops_sec_max is set.

write_iops_sec_max

The optional write_iops_sec_max_length element is the maximum duration in seconds for the write_iops_sec_max burst period. Only valid when the write_iops_sec_max is set.

Throughput length since 2.4.0 and QEMU 2.6

driver

The optional driver element allows specifying further details related to the hypervisor driver used to provide the disk. Since 0.1.8

  • If the hypervisor supports multiple backend drivers, then the name attribute selects the primary backend driver name, while the optional type attribute provides the sub-type. For example, xen supports a name of "tap", "tap2", "phy", or "file", with a type of "aio", while qemu only supports a name of "qemu", but multiple types including "raw", "bochs", "qcow2", and "qed".

  • The optional cache attribute controls the cache mechanism, possible values are "default", "none", "writethrough", "writeback", "directsync" (like "writethrough", but it bypasses the host page cache) and "unsafe" (host may cache all disk io, and sync requests from guest are ignored). Since 0.6.0, "directsync" since 0.9.5, "unsafe" since 0.9.7

  • The optional error_policy attribute controls how the hypervisor will behave on a disk read or write error, possible values are "stop", "report", "ignore", and "enospace". Since 0.8.0, "report" since 0.9.7 The default is left to the discretion of the hypervisor. There is also an optional rerror_policy that controls behavior for read errors only. Since 0.9.7 . If no rerror_policy is given, error_policy is used for both read and write errors. If rerror_policy is given, it overrides the error_policy for read errors. Also note that "enospace" is not a valid policy for read errors, so if error_policy is set to "enospace" and no rerror_policy is given, the read error policy will be left at its default.

  • The optional io attribute controls specific policies on I/O; qemu guests support "threads" and "native" Since 0.8.8 , io_uring Since 6.3.0 (QEMU 5.0) .

  • The optional ioeventfd attribute allows users to set domain I/O asynchronous handling for disk device. The default is left to the discretion of the hypervisor. Accepted values are "on" and "off". Enabling this allows qemu to execute VM while a separate thread handles I/O. Typically guests experiencing high system CPU utilization during I/O will benefit from this. On the other hand, on overloaded host it could increase guest I/O latency. Since 0.9.3 (QEMU and KVM only) In general you should leave this option alone, unless you are very certain you know what you are doing.

  • The optional event_idx attribute controls some aspects of device event processing. The value can be either 'on' or 'off' - if it is on, it will reduce the number of interrupts and exits for the guest. The default is determined by QEMU; usually if the feature is supported, default is on. In case there is a situation where this behavior is suboptimal, this attribute provides a way to force the feature off. Since 0.9.5 (QEMU and KVM only) In general you should leave this option alone, unless you are very certain you know what you are doing.

  • The optional copy_on_read attribute controls whether to copy read backing file into the image file. The value can be either "on" or "off". Copy-on-read avoids accessing the same backing file sectors repeatedly and is useful when the backing file is over a slow network. By default copy-on-read is off. Since 0.9.10 (QEMU and KVM only)

  • The optional discard attribute controls whether discard requests (also known as "trim" or "unmap") are ignored or passed to the filesystem. The value can be either "unmap" (allow the discard request to be passed) or "ignore" (ignore the discard request). Since 1.0.6 (QEMU and KVM only)

  • The optional detect_zeroes attribute controls whether to detect zero write requests. The value can be "off", "on" or "unmap". First two values turn the detection off and on, respectively. The third value ("unmap") turns the detection on and additionally tries to discard such areas from the image based on the value of discard above (it will act as "on" if discard is set to "ignore"). NB enabling the detection is a compute intensive operation, but can save file space and/or time on slow media. Since 2.0.0

  • The optional iothread attribute assigns the disk to an IOThread as defined by the range for the domain iothreads value. (See IOThreads Allocation). Multiple disks may be assigned to the same IOThread and are numbered from 1 to the domain iothreads value. Available for a disk device target configured to use "virtio" bus and "pci" or "ccw" address types. Since 1.2.8 (QEMU 2.1)

  • The optional queues attribute specifies the number of virt queues for virtio-blk. ( Since 3.9.0 )

  • The optional queue_size attribute specifies the size of each virt queue for virtio-blk. ( Since 7.8.0 )

  • For virtio disks, Virtio-related options can also be set. ( Since 3.5.0 )

  • The optional metadata_cache subelement controls aspects related to the format specific caching of storage image metadata. Note that this setting applies only on the top level image; the identically named subelement of backingStore's format element can be used to specify cache settings for the backing image.

    Since 7.0.0 the maximum size of the metadata cache of qcow2 format driver of the qemu hypervisor can be controlled via the max_size subelement (see example below).

    The optional discard_no_unref attribute can be set to control the way the qemu hypervisor handles guest discard commands inside the qcow2 image. When enabled, a discard request from within the guest will mark the qcow2 cluster as zero, but will keep the reference/offset of that cluster. But it will still pass the discard further to the lower layer. This will resolve fragmentation within the qcow2 image. Since 9.5.0

    In the majority of cases the default configuration used by the hypervisor is sufficient so modifying this setting should not be necessary. For specifics on how the metadata cache of qcow2 in qemu behaves refer to the qemu qcow2 cache docs

    Example:

    <disk type='file' device='disk'>
      <driver name='qemu' type='qcow2'>
        <metadata_cache>
          <max_size unit='bytes'>1234</max_size>
        </metadata_cache>
      </driver>
      <source file='/var/lib/libvirt/images/domain.qcow'/>
      <backingStore type='file'>
        <format type='qcow2'>
          <metadata_cache>
            <max_size unit='bytes'>1234</max_size>
          </metadata_cache>
        </format>
        <source file='/var/lib/libvirt/images/snapshot.qcow'/>
        <backingStore/>
      </backingStore>
      <target dev='vdd' bus='virtio'/>
    </disk>
backenddomain

The optional backenddomain element allows specifying a backend domain (aka driver domain) hosting the disk. Use the name attribute to specify the backend domain name. Since 1.2.13 (Xen only)

boot

Specifies that the disk is bootable. The order attribute determines the order in which devices will be tried during boot sequence. On the S390 architecture only the first boot device is used. The optional loadparm attribute is an 8 character string which can be queried by guests on S390 via sclp or diag 308. Linux guests on S390 can use loadparm to select a boot entry. Since 3.5.0 The per-device boot elements cannot be used together with general boot elements in BIOS bootloader section. Since 0.8.8

encryption

Starting with libvirt 3.9.0 the encryption element is preferred to be a sub-element of the source element. If present, specifies how the volume is encrypted using "qcow". See the Storage Encryption page for more information.

readonly

If present, this indicates the device cannot be modified by the guest. For now, this is the default for disks with attribute device='cdrom'.

shareable

If present, this indicates the device is expected to be shared between domains (assuming the hypervisor and OS support this), which means that caching should be deactivated for that device.

transient

If present, this indicates that changes to the device contents should be reverted automatically when the guest exits. With some hypervisors, marking a disk transient prevents the domain from participating in migration, snapshots, or blockjobs. Only supported in vmx hypervisor (Since 0.9.5) and qemu hypervisor (Since 6.9.0).

In cases where the source image of the <transient/> disk is supposed to be shared between multiple concurrently running VMs the optional shareBacking attribute should be set to yes. Note that hypervisor drivers may need to hotplug such disk and thus it works only with configurations supporting hotplug. Since 7.4.0

serial

If present, this specify serial number of virtual hard drive. For example, it may look like <serial>WD-WMAP9A966149</serial>. Not supported for scsi-block devices, that is those using disk type 'block' using device 'lun' on bus 'scsi'. Since 0.7.1

Note that depending on hypervisor and device type the serial number may be truncated silently. IDE/SATA devices are commonly limited to 20 characters. SCSI devices depending on hypervisor version are limited to 20, 36 or 247 characters.

Hypervisors may also start rejecting overly long serials instead of truncating them in the future so it's advised to avoid the implicit truncation by testing the desired serial length range with the desired device and hypervisor combination.

wwn

If present, this element specifies the WWN (World Wide Name) of a virtual hard disk or CD-ROM drive. It must be composed of 16 hexadecimal digits. Since 0.10.1

vendor

If present, this element specifies the vendor of a virtual hard disk or CD-ROM device. It must not be longer than 8 printable characters. Since 1.0.1

product

If present, this element specifies the product of a virtual hard disk or CD-ROM device. It must not be longer than 16 printable characters. Since 1.0.1

address

If present, the address element ties the disk to a given slot of a controller (the actual <controller> device can often be inferred by libvirt, although it can be be explicitly specified. See Controllers). The type attribute is mandatory, and is typically "pci" or "drive". For a "pci" controller, additional attributes for bus, slot, and function must be present, as well as optional domain and multifunction. Multifunction defaults to 'off'; any other value requires QEMU 0.1.3 and libvirt 0.9.7 . For a "drive" controller, additional attributes controller, bus, target ( libvirt 0.9.11 ), and unit are available, each defaulting to 0.

auth

Starting with libvirt 3.9.0 the auth element is preferred to be a sub-element of the source element. The element is still read and managed as a disk sub-element. It is invalid to use auth as both a sub-element of disk and source. The auth element was introduced as a disk sub-element in libvirt 0.9.7.

geometry

The optional geometry element provides the ability to override geometry settings. This mostly useful for S390 DASD-disks or older DOS-disks. 0.10.0

cyls

The cyls attribute is the number of cylinders.

heads

The heads attribute is the number of heads.

secs

The secs attribute is the number of sectors per track.

trans

The optional trans attribute is the BIOS-Translation-Modus (none, lba or auto)

blockio

If present, the blockio element allows to override any of the block device properties listed below. Since 0.10.2 (QEMU and KVM)

logical_block_size

The logical block size the disk will report to the guest OS. For Linux this would be the value returned by the BLKSSZGET ioctl and describes the smallest units for disk I/O.

physical_block_size

The physical block size the disk will report to the guest OS. For Linux this would be the value returned by the BLKPBSZGET ioctl and describes the disk's hardware sector size which can be relevant for the alignment of disk data.

Filesystems

A directory on the host that can be accessed directly from the guest. since 0.3.3, since 0.8.5 for QEMU/KVM

...
<devices>
  <filesystem type='template'>
    <source name='my-vm-template'/>
    <target dir='/'/>
  </filesystem>
  <filesystem type='mount' accessmode='passthrough' multidevs='remap'>
    <driver type='path' wrpolicy='immediate'/>
    <source dir='/export/to/guest'/>
    <target dir='/import/from/host'/>
    <readonly/>
  </filesystem>
  <filesystem type='mount' accessmode='mapped' fmode='644' dmode='755'>
    <driver type='path'/>
    <source dir='/export/to/guest'/>
    <target dir='/import/from/host'/>
    <readonly/>
  </filesystem>
  <filesystem type='file' accessmode='passthrough'>
    <driver type='loop' format='raw'/>
    <source file='/export/to/guest.img'/>
    <target dir='/import/from/host'/>
    <readonly/>
  </filesystem>
  <filesystem type='mount' accessmode='passthrough'>
      <driver type='virtiofs' queue='1024'/>
      <binary path='/usr/libexec/virtiofsd' xattr='on'>
         <cache mode='always'/>
         <sandbox mode='namespace'/>
         <lock posix='on' flock='on'/>
         <thread_pool size='16'/>
      </binary>
      <source dir='/path'/>
      <target dir='mount_tag'/>
  </filesystem>
  <filesystem type='mount'>
      <driver type='virtiofs' queue='1024'/>
      <source socket='/tmp/sock'/>
      <target dir='tag'/>
  </filesystem>
  ...
</devices>
...
filesystem

The filesystem attribute type specifies the type of the source. The possible values are:

mount

A host directory to mount in the guest. Used by LXC, OpenVZ (since 0.6.2) and QEMU/KVM (since 0.8.5) . This is the default type if one is not specified. This mode also has an optional sub-element driver, with an attribute type='path' or type='handle' (since 0.9.7) . The driver block has an optional attribute wrpolicy that further controls interaction with the host page cache; omitting the attribute gives default behavior, while the value immediate means that a host writeback is immediately triggered for all pages touched during a guest file write operation (since 0.9.10) . Since 6.2.0 , type='virtiofs' is also supported. Using virtiofs requires setting up shared memory, see the guide: Virtiofs

template

OpenVZ filesystem template. Only used by OpenVZ driver.

file

A host file will be treated as an image and mounted in the guest. The filesystem format will be autodetected. Only used by LXC driver.

block

A host block device to mount in the guest. The filesystem format will be autodetected. Only used by LXC driver (since 0.9.5) .

ram

An in-memory filesystem, using memory from the host OS. The source element has a single attribute usage which gives the memory usage limit in KiB, unless units are specified by the units attribute. Only used by LXC driver. (since 0.9.13)

bind

A directory inside the guest will be bound to another directory inside the guest. Only used by LXC driver (since 0.9.13)

The filesystem element has an optional attribute accessmode which specifies the security mode for accessing the source (since 0.8.5) . Currently this only works with type='mount' for the QEMU/KVM driver. For driver type virtiofs, only passthrough is supported. For other driver types, the possible values are:

passthrough

The source is accessed with the permissions of the user inside the guest. This is the default accessmode if one is not specified. More info

mapped

The source is accessed with the permissions of the hypervisor (QEMU process). More info

squash

Similar to 'passthrough', the exception is that failure of privileged operations like 'chown' are ignored. This makes a passthrough-like mode usable for people who run the hypervisor as non-root. More info

Since 5.2.0 , the filesystem element has an optional attribute model with supported values "virtio-transitional", "virtio-non-transitional", or "virtio". See Virtio transitional devices for more details.

The filesystem element has optional attributes fmode and dmode. These two attributes control the creation mode for files and directories when used with the mapped value for accessmode (since 6.10.0, requires QEMU 2.10 ). If not specified, QEMU creates files with mode 600 and directories with mode 700. The setuid, setgid, and sticky bit are unsupported.

The filesystem element has an optional attribute multidevs which specifies how to deal with a filesystem export containing more than one device, in order to avoid file ID collisions on guest when using 9pfs ( since 6.3.0, requires QEMU 4.2 ). This attribute is not available for virtiofs. The possible values are:

default

Use QEMU's default setting (which currently is warn).

remap

This setting allows guest to access multiple devices per export without encountering misbehaviours. Inode numbers from host are automatically remapped on guest to actively prevent file ID collisions if guest accesses one export containing multiple devices.

forbid

Only allow to access one device per export by guest. Attempts to access additional devices on the same export will cause the individual filesystem access by guest to fail with an error and being logged (once) as error on host side.

warn

This setting resembles the behaviour of 9pfs prior to QEMU 4.2, that is no action is performed to prevent any potential file ID collisions if an export contains multiple devices, with the only exception: a warning is logged (once) on host side now. This setting may lead to misbehaviours on guest side if more than one device is exported per export, due to the potential file ID collisions this may cause on guest side in that case.

driver

The optional driver element allows specifying further details related to the hypervisor driver used to provide the filesystem. Since 1.0.6

  • If the hypervisor supports multiple backend drivers, then the type attribute selects the primary backend driver name, while the format attribute provides the format type. For example, LXC supports a type of "loop", with a format of "raw" or "nbd" with any format. QEMU supports a type of "path" or "handle", but no formats. Virtuozzo driver supports a type of "ploop" with a format of "ploop".
  • For virtio-backed devices, Virtio-related options can also be set. ( Since 3.5.0 )
  • For virtiofs, the queue attribute can be used to specify the queue size (i.e. how many requests can the queue fit). ( Since 6.2.0 )
binary

The optional binary element can tune the options for virtiofsd. All of the following attributes and elements are optional. The attribute path can be used to override the path to the daemon. Attribute xattr enables the use of filesystem extended attributes. Caching can be tuned via the cache element, possible mode values being none and always. Locking can be controlled via the lock element - attributes posix and flock both accepting values on or off. ( Since 6.2.0 ) The sandboxing method used by virtiofsd can be configured with the sandbox element, possible mode values being namespace and chroot, see the virtiofsd documentation for more details. ( Since 7.2.0 ) Element thread_pool accepts one attribute size which defines the maximum thread pool size. A value of "0" disables the pool. The thread pool helps increase the number of requests in flight when used with storage that has a higher latency. However, it has an overhead, and so for fast, low latency filesystems, it may be best to turn it off. ( Since 8.5.0 )

source

The resource on the host that is being accessed in the guest. The name attribute must be used with type='template', and the dir attribute must be used with type='mount'. For virtiofs, the socket attribute can be used to connect to a virtiofsd daemon launched outside of libvirt. In that case, the target element does not apply and neither do most virtiofs-related options, since they are controlled by virtiofsd, not libvirtd. The usage attribute is used with type='ram' to set the memory limit in KiB, unless units are specified by the units attribute.

target

Where the source can be accessed in the guest. For most drivers this is an automatic mount point, but for QEMU/KVM this is merely an arbitrary string tag that is exported to the guest as a hint for where to mount.

readonly

Enables exporting filesystem as a readonly mount for guest, by default read-write access is given (currently only works for QEMU/KVM driver).

space_hard_limit

Maximum space available to this guest's filesystem. Since 0.9.13

space_soft_limit

Maximum space available to this guest's filesystem. The container is permitted to exceed its soft limits for a grace period of time. Afterwards the hard limit is enforced. Since 0.9.13

Device Addresses

Many devices have an optional <address> sub-element to describe where the device is placed on the virtual bus presented to the guest. If an address (or any optional attribute within an address) is omitted on input, libvirt will generate an appropriate address; but an explicit address is required if more control over layout is required. See below for device examples including an address element.

Every address has a mandatory attribute type that describes which bus the device is on. The choice of which address to use for a given device is constrained in part by the device and the architecture of the guest. For example, a <disk> device uses type='drive', while a <console> device would use type='pci' on i686 or x86_64 guests, or type='spapr-vio' on PowerPC64 pseries guests. Each address type has further optional attributes that control where on the bus the device will be placed:

pci

PCI addresses have the following additional attributes: domain (a 2-byte hex integer, not currently used by qemu), bus (a hex value between 0 and 0xff, inclusive), slot (a hex value between 0x0 and 0x1f, inclusive), and function (a value between 0 and 7, inclusive). Also available is the multifunction attribute, which controls turning on the multifunction bit for a particular slot/function in the PCI control register ( since 0.9.7, requires QEMU 0.13 ). multifunction defaults to 'off', but should be set to 'on' for function 0 of a slot that will have multiple functions used. ( Since 4.10.0 ), PCI address extensions depending on the architecture are supported. For example, PCI addresses for S390 guests will have a zpci child element, with two attributes: uid (a hex value between 0x0001 and 0xffff, inclusive), and fid (a hex value between 0x00000000 and 0xffffffff, inclusive) used by PCI devices on S390 for User-defined Identifiers and Function Identifiers. Since 1.3.5 , some hypervisor drivers may accept an <address type='pci'/> element with no other attributes as an explicit request to assign a PCI address for the device rather than some other type of address that may also be appropriate for that same device (e.g. virtio-mmio). The relationship between the PCI addresses configured in the domain XML and those seen by the guest OS can sometime seem confusing: a separate document describes how PCI addresses work in more detail.

drive

Drive addresses have the following additional attributes: controller (a 2-digit controller number), bus (a 2-digit bus number), target (a 2-digit target number), and unit (a 2-digit unit number on the bus).

virtio-serial

Each virtio-serial address has the following additional attributes: controller (a 2-digit controller number), bus (a 2-digit bus number), and slot (a 2-digit slot within the bus).

ccid

A CCID address, for smart-cards, has the following additional attributes: bus (a 2-digit bus number), and slot attribute (a 2-digit slot within the bus). Since 0.8.8.

usb

USB addresses have the following additional attributes: bus (a hex value between 0 and 0xfff, inclusive), and port (a dotted notation of up to four octets, such as 1.2 or 2.1.3.1).

spapr-vio

On PowerPC pseries guests, devices can be assigned to the SPAPR-VIO bus. It has a flat 32-bit address space; by convention, devices are generally assigned at a non-zero multiple of 0x00001000, but other addresses are valid and permitted by libvirt. Each address has the following additional attribute: reg (the hex value address of the starting register). Since 0.9.9.

ccw

S390 guests with a machine value of s390-ccw-virtio use the native CCW bus for I/O devices. CCW bus addresses have the following additional attributes: cssid (a hex value between 0 and 0xfe, inclusive), ssid (a value between 0 and 3, inclusive) and devno (a hex value between 0 and 0xffff, inclusive). Partially specified bus addresses are not allowed. If omitted, libvirt will assign a free bus address with cssid=0xfe and ssid=0. Virtio-ccw devices must have their cssid set to 0xfe. Since 1.0.4

virtio-mmio

This places the device on the virtio-mmio transport, which is currently only available for some armv7l and aarch64 virtual machines. virtio-mmio addresses do not have any additional attributes. Since 1.1.3 If the guest architecture is aarch64 and the machine type is virt, libvirt will automatically assign PCI addresses to devices; however, the presence of a single device with virtio-mmio address in the guest configuration will cause libvirt to assign virtio-mmio addresses to all further devices. Since 3.0.0

isa

ISA addresses have the following additional attributes: iobase and irq. Since 1.2.1

unassigned

For PCI hostdevs, <address type='unassigned'/> allows the admin to include a PCI hostdev in the domain XML definition, without making it available for the guest. This allows for configurations in which Libvirt manages the device as a regular PCI hostdev, regardless of whether the guest will have access to it. <address type='unassigned'/> is an invalid address type for all other device types. Since 6.0.0

QEMU's virtio devices have some attributes related to the virtio transport under the driver element: The iommu attribute enables the use of emulated IOMMU by the device. The attribute ats controls the Address Translation Service support for PCIe devices. This is needed to make use of IOTLB support (see IOMMU devices). Possible values are on or off. Since 3.5.0

The attribute packed controls if QEMU should try to use packed virtqueues. Compared to regular split queues, packed queues consist of only a single descriptor ring replacing available and used ring, index and descriptor buffer. This can result in better cache utilization and performance. If packed virtqueues are actually used depends on the feature negotiation between QEMU, vhost backends and guest drivers. Possible values are on or off. Since 6.3.0 (QEMU and KVM only)

This optional attribute page_per_vq controls the layout of the notification capabilities exposed to the guest. When enabled, each virtio queue will have a dedicated page on the device BAR exposed to the guest. It is recommended to be used when vDPA is enabled on the hypervisor, as it enables mapping the notification area to the physical device, which is only supported in page granularity. The default is determined by QEMU. Since 7.9.0 (QEMU 2.8) Note: In general you should leave this option alone, unless you are very certain you know what you are doing.

Virtio transitional devices

Since 5.2.0 , some of QEMU's virtio devices, when used with PCI/PCIe machine types, accept the following model values:

virtio-transitional

This device can work both with virtio 0.9 and virtio 1.0 guest drivers, so it's the best choice when compatibility with older guest operating systems is desired. libvirt will plug the device into a conventional PCI slot.

virtio-non-transitional

This device can only work with virtio 1.0 guest drivers, and it's the recommended option unless compatibility with older guest operating systems is necessary. libvirt will plug the device into either a PCI Express slot or a conventional PCI slot based on the machine type, resulting in a more optimized PCI topology.

virtio

This device will work like a virtio-non-transitional device when plugged into a PCI Express slot, and like a virtio-transitional device otherwise; libvirt will pick one or the other based on the machine type. This is the best choice when compatibility with libvirt versions older than 5.2.0 is necessary, but it's otherwise not recommended to use it.

While the information outlined above applies to most virtio devices, there are a few exceptions:

  • for SCSI controllers, there is no virtio model available due to historical reasons: use virtio-scsi instead, which behaves the same as virtio does for other devices. Both virtio-transitional and virtio-non-transitional work with SCSI controllers;
  • some devices, such as GPUs and input devices (keyboard, tablet and mouse), are only defined in the virtio 1.0 spec and as such don't have a transitional variant: the only accepted model is virtio, which will result in a non-transitional device.

For more details see the qemu patch posting and the virtio-1.0 spec.

Controllers

Depending on the guest architecture, some device buses can appear more than once, with a group of virtual devices tied to a virtual controller. Normally, libvirt can automatically infer such controllers without requiring explicit XML markup, but sometimes it is necessary to provide an explicit controller element, notably when planning the PCI topology for guests where device hotplug is expected.

...
<devices>
  <controller type='ide' index='0'/>
  <controller type='virtio-serial' index='0' ports='16' vectors='4'/>
  <controller type='virtio-serial' index='1'>
    <address type='pci' domain='0x0000' bus='0x00' slot='0x0a' function='0x0'/>
  </controller>
  <controller type='scsi' index='0' model='virtio-scsi'>
    <driver iothread='4'/>
    <address type='pci' domain='0x0000' bus='0x00' slot='0x0b' function='0x0'/>
  </controller>
  <controller type='xenbus' maxGrantFrames='64' maxEventChannels='2047'/>
  ...
</devices>
...

Each controller has a mandatory attribute type, which must be one of 'ide', 'fdc', 'scsi', 'sata', 'usb', 'ccid', 'virtio-serial' or 'pci', and a mandatory attribute index which is the decimal integer describing in which order the bus controller is encountered (for use in controller attributes of <address> elements). Since 1.3.5 the index is optional; if not specified, it will be auto-assigned to be the lowest unused index for the given controller type. Some controller types have additional attributes that control specific features, such as:

virtio-serial

The virtio-serial controller has two additional optional attributes ports and vectors, which control how many devices can be connected through the controller. Since 5.2.0 , it supports an optional attribute model which can be 'virtio', 'virtio-transitional', or 'virtio-non-transitional'. See Virtio transitional devices for more details.

scsi

A scsi controller has an optional attribute model, which is one of 'auto', 'buslogic', 'ibmvscsi', 'lsilogic', 'lsisas1068', 'lsisas1078', 'virtio-scsi', 'vmpvscsi', 'virtio-transitional', 'virtio-non-transitional', 'ncr53c90' (as builtin implicit controller only), 'am53c974', 'dc390'. See Virtio transitional devices for more details.

usb

A usb controller has an optional attribute model, which is one of "piix3-uhci", "piix4-uhci", "ehci", "ich9-ehci1", "ich9-uhci1", "ich9-uhci2", "ich9-uhci3", "vt82c686b-uhci", "pci-ohci", "nec-xhci", "qusb1" (xen pvusb with qemu backend, version 1.1), "qusb2" (xen pvusb with qemu backend, version 2.0) or "qemu-xhci". Additionally, since 0.10.0 , if the USB bus needs to be explicitly disabled for the guest, model='none' may be used. Since 1.0.5 , no default USB controller will be built on s390. Since 1.3.5 , USB controllers accept a ports attribute to configure how many devices can be connected to the controller.

ide

Since 3.10.0 for the vbox driver, the ide controller has an optional attribute model, which is one of "piix3", "piix4" or "ich6".

xenbus

Since 5.2.0 , the xenbus controller has an optional attribute maxGrantFrames, which specifies the maximum number of grant frames the controller makes available for connected devices. Since 6.3.0 , the xenbus controller supports the optional maxEventChannels attribute, which specifies maximum number of event channels (PV interrupts) that can be used by the guest.

Note: The PowerPC64 "spapr-vio" addresses do not have an associated controller.

For controllers that are themselves devices on a PCI or USB bus, an optional sub-element <address> can specify the exact relationship of the controller to its master bus, with semantics described in the Device Addresses section.

An optional sub-element driver can specify the driver specific options:

queues

The optional queues attribute specifies the number of queues for the controller. For best performance, it's recommended to specify a value matching the number of vCPUs. Since 1.0.5 (QEMU and KVM only)

cmd_per_lun

The optional cmd_per_lun attribute specifies the maximum number of commands that can be queued on devices controlled by the host. Since 1.2.7 (QEMU and KVM only)

max_sectors

The optional max_sectors attribute specifies the maximum amount of data in bytes that will be transferred to or from the device in a single command. The transfer length is measured in sectors, where a sector is 512 bytes. Since 1.2.7 (QEMU and KVM only)

ioeventfd

The optional ioeventfd attribute specifies whether the controller should use I/O asynchronous handling or not. Accepted values are "on" and "off". Since 1.2.18

iothread

Supported for controller type scsi using model virtio-scsi for address types pci and ccw since 1.3.5 (QEMU 2.4) . The optional iothread attribute assigns the controller to an IOThread as defined by the range for the domain iothreads (See IOThreads Allocation). Each SCSI disk assigned to use the specified controller will utilize the same IOThread. If a specific IOThread is desired for a specific SCSI disk, then multiple controllers must be defined each having a specific iothread value. The iothread value must be within the range 1 to the domain iothreads value.

virtio options

For virtio controllers, Virtio-related options can also be set. ( Since 3.5.0 )

USB companion controllers have an optional sub-element <master> to specify the exact relationship of the companion to its master controller. A companion controller is on the same bus as its master, so the companion index value should be equal. Not all controller models can be used as companion controllers and libvirt might provide some sensible defaults (settings of master startport and function of an address) for some particular models. Preferred companion controllers are ich-uhci[123].

...
<devices>
  <controller type='usb' index='0' model='ich9-ehci1'>
    <address type='pci' domain='0' bus='0' slot='4' function='7'/>
  </controller>
  <controller type='usb' index='0' model='ich9-uhci1'>
    <master startport='0'/>
    <address type='pci' domain='0' bus='0' slot='4' function='0' multifunction='on'/>
  </controller>
  ...
</devices>
...

PCI controllers have an optional model attribute; possible values for this attribute are

  • pci-root, pci-bridge ( since 1.0.5 )
  • pcie-root, dmi-to-pci-bridge ( since 1.1.2 )
  • pcie-root-port, pcie-switch-upstream-port, pcie-switch-downstream-port ( since 1.2.19 )
  • pci-expander-bus, pcie-expander-bus ( since 1.3.4 )
  • pcie-to-pci-bridge ( since 4.3.0 )

The root controllers (pci-root and pcie-root) have an optional pcihole64 element specifying how big (in kilobytes, or in the unit specified by pcihole64's unit attribute) the 64-bit PCI hole should be. Some guests (like Windows XP or Windows Server 2003) might crash when QEMU and Seabios are recent enough to support 64-bit PCI holes, unless this is disabled (set to 0). Since 1.1.2 (QEMU only)

PCI controllers also have an optional subelement <model> with an attribute name. The name attribute holds the name of the specific device that qemu is emulating (e.g. "i82801b11-bridge") rather than simply the class of device ("pcie-to-pci-bridge", "pci-bridge"), which is set in the controller element's model attribute. In almost all cases, you should not manually add a <model> subelement to a controller, nor should you modify one that is automatically generated by libvirt. Since 1.2.19 (QEMU only).

PCI controllers also have an optional subelement <target> with the attributes and subelements listed below. These are configurable items that 1) are visible to the guest OS so must be preserved for guest ABI compatibility, and 2) are usually left to default values or derived automatically by libvirt. In almost all cases, you should not manually add a <target> subelement to a controller, nor should you modify the values in the those that are automatically generated by libvirt. Since 1.2.19 (QEMU only).

chassisNr

PCI controllers that have attribute model="pci-bridge", can also have a chassisNr attribute in the <target> subelement, which is used to control QEMU's "chassis_nr" option for the pci-bridge device (normally libvirt automatically sets this to the same value as the index attribute of the pci controller). If set, chassisNr must be between 1 and 255.

chassis

pcie-root-port and pcie-switch-downstream-port controllers can also have a chassis attribute in the <target> subelement, which is used to set the controller's "chassis" configuration value, which is visible to the virtual machine. If set, chassis must be between 0 and 255.

port

pcie-root-port and pcie-switch-downstream-port controllers can also have a port attribute in the <target> subelement, which is used to set the controller's "port" configuration value, which is visible to the virtual machine. If set, port must be between 0 and 255.

hotplug

pci-root (Since 7.9.0), pcie-root-port (Since 6.3.0) and pcie-switch-downstream-port controllers (Since 6.3.0) can also have a hotplug attribute in the <target> subelement, which is used to disable hotplug/unplug of devices on a particular controller. For the pci-root controller, the setting affects the ACPI based hotplug. For the rest, the setting affects both ACPI based hotplug as well as PCIE native hotplug. The default setting of hotplug is on; it should be set to off to disable hotplug/unplug of devices on a particular controller.

busNr

pci-expander-bus and pcie-expander-bus controllers can have an optional busNr attribute (1-254). This will be the bus number of the new bus; All bus numbers between that specified and 255 will be available only for assignment to PCI/PCIe controllers plugged into the hierarchy starting with this expander bus, and bus numbers less than the specified value will be available to the next lower expander-bus (or the root-bus if there are no lower expander buses). If you do not specify a busNumber, libvirt will find the lowest existing busNumber in all other expander buses (or use 256 if there are no others) and auto-assign the busNr of that found bus - 2, which provides one bus number for the pci-expander-bus and one for the pci-bridge that is automatically attached to it (if you plan on adding more pci-bridges to the hierarchy of the bus, you should manually set busNr to a lower value).

A similar algorithm is used for automatically determining the busNr attribute for pcie-expander-bus, but since the pcie-expander-bus doesn't have any built-in pci-bridge, the 2nd bus-number is just being reserved for the pcie-root-port that must necessarily be connected to the bus in order to actually plug in an endpoint device. If you intend to plug multiple devices into a pcie-expander-bus, you must connect a pcie-switch-upstream-port to the pcie-root-port that is plugged into the pcie-expander-bus, and multiple pcie-switch-downstream-ports to the pcie-switch-upstream-port, and of course for this to work properly, you will need to decrease the pcie-expander-bus' busNr accordingly so that there are enough unused bus numbers above it to accommodate giving out one bus number for the upstream-port and one for each downstream-port (in addition to the pcie-root-port and the pcie-expander-bus itself).

node

Some PCI controllers (pci-expander-bus for the pc machine type, pcie-expander-bus for the q35 machine type and, since 3.6.0 , pci-root for the pseries machine type) can have an optional <node> subelement within the <target> subelement, which is used to set the NUMA node reported to the guest OS for that bus - the guest OS will then know that all devices on that bus are a part of the specified NUMA node (it is up to the user of the libvirt API to attach host devices to the correct pci-expander-bus when assigning them to the domain).

index

pci-root controllers for pSeries guests use this attribute to record the order they will show up in the guest. Since 3.6.0

For machine types which provide an implicit PCI bus, the pci-root controller with index=0 is auto-added and required to use PCI devices. pci-root has no address. PCI bridges are auto-added if there are too many devices to fit on the one bus provided by pci-root, or a PCI bus number greater than zero was specified. PCI bridges can also be specified manually, but their addresses should only refer to PCI buses provided by already specified PCI controllers. Leaving gaps in the PCI controller indexes might lead to an invalid configuration.

...
<devices>
  <controller type='pci' index='0' model='pci-root'/>
  <controller type='pci' index='1' model='pci-bridge'>
    <address type='pci' domain='0' bus='0' slot='5' function='0' multifunction='off'/>
  </controller>
</devices>
...

For machine types which provide an implicit PCI Express (PCIe) bus (for example, the machine types based on the Q35 chipset), the pcie-root controller with index=0 is auto-added to the domain's configuration. pcie-root has also no address, provides 31 slots (numbered 1-31) that can be used to attach PCIe or PCI devices (although libvirt will never auto-assign a PCI device to a PCIe slot, it will allow manual specification of such an assignment). Devices connected to pcie-root cannot be hotplugged. If traditional PCI devices are present in the guest configuration, a pcie-to-pci-bridge controller will automatically be added: this controller, which plugs into a pcie-root-port, provides 31 usable PCI slots (1-31) with hotplug support ( since 4.3.0 ). If the QEMU binary doesn't support the corresponding device, then a dmi-to-pci-bridge controller will be added instead, usually at the defacto standard location of slot=0x1e. A dmi-to-pci-bridge controller plugs into a PCIe slot (as provided by pcie-root), and itself provides 31 standard PCI slots (which also do not support device hotplug). In order to have hot-pluggable PCI slots in the guest system, a pci-bridge controller will also be automatically created and connected to one of the slots of the auto-created dmi-to-pci-bridge controller; all guest PCI devices with addresses that are auto-determined by libvirt will be placed on this pci-bridge device. ( since 1.1.2 ).

Domains with an implicit pcie-root can also add controllers with model='pcie-root-port', model='pcie-switch-upstream-port', and model='pcie-switch-downstream-port'. pcie-root-port is a simple type of bridge device that can connect only to one of the 31 slots on the pcie-root bus on its upstream side, and makes a single (PCIe, hotpluggable) port available on the downstream side (at slot='0'). pcie-root-port can be used to provide a single slot to later hotplug a PCIe device (but is not itself hotpluggable - it must be in the configuration when the domain is started). ( since 1.2.19 )

pcie-switch-upstream-port is a more flexible (but also more complex) device that can only plug into a pcie-root-port or pcie-switch-downstream-port on the upstream side (and only before the domain is started - it is not hot-pluggable), and provides 32 ports on the downstream side (slot='0' - slot='31') that accept only pcie-switch-downstream-port devices; each pcie-switch-downstream-port device can only plug into a pcie-switch-upstream-port on its upstream side (again, not hot-pluggable), and on its downstream side provides a single hotpluggable pcie port that can accept any standard pci or pcie device (or another pcie-switch-upstream-port), i.e. identical in function to a pcie-root-port. ( since 1.2.19 )

...
<devices>
  <controller type='pci' index='0' model='pcie-root'/>
  <controller type='pci' index='1' model='pcie-root-port'>
    <address type='pci' domain='0x0000' bus='0x00' slot='0x01' function='0x0'/>
  </controller>
  <controller type='pci' index='2' model='pcie-to-pci-bridge'>
    <address type='pci' domain='0x0000' bus='0x01' slot='0x00' function='0x0'/>
  </controller>
</devices>
...

Device leases

When using a lock manager, it may be desirable to record device leases against a VM. The lock manager will ensure the VM won't start unless the leases can be acquired.

...
<devices>
  ...
  <lease>
    <lockspace>somearea</lockspace>
    <key>somekey</key>
    <target path='/some/lease/path' offset='1024'/>
  </lease>
  ...
</devices>
...
lockspace

This is an arbitrary string, identifying the lockspace within which the key is held. Lock managers may impose extra restrictions on the format, or length of the lockspace name.

key

This is an arbitrary string, uniquely identifying the lease to be acquired. Lock managers may impose extra restrictions on the format, or length of the key.

target

This is the fully qualified path of the file associated with the lockspace. The offset specifies where the lease is stored within the file. If the lock manager does not require an offset, just pass 0.

Host device assignment

USB / PCI / SCSI devices

USB, PCI and SCSI devices attached to the host can be passed through to the guest using the hostdev element. since after 0.4.4 for USB, 0.6.0 for PCI (KVM only) and 1.0.6 for SCSI (KVM only) :

...
<devices>
  <hostdev mode='subsystem' type='usb'>
    <source startupPolicy='optional' guestReset='off'>
      <vendor id='0x1234'/>
      <product id='0xbeef'/>
    </source>
    <boot order='2'/>
  </hostdev>
</devices>
...

or:

...
<devices>
  <hostdev mode='subsystem' type='pci' managed='yes'>
    <source writeFiltering='no'>
      <address domain='0x0000' bus='0x06' slot='0x02' function='0x0'/>
    </source>
    <boot order='1'/>
    <rom bar='on' file='/etc/fake/boot.bin'/>
  </hostdev>
</devices>
...

or:

...
<devices>
  <hostdev mode='subsystem' type='scsi' sgio='filtered' rawio='yes'>
    <source>
      <adapter name='scsi_host0'/>
      <address bus='0' target='0' unit='0'/>
    </source>
    <readonly/>
    <address type='drive' controller='0' bus='0' target='0' unit='0'/>
  </hostdev>
</devices>
...

or:

...
<devices>
  <hostdev mode='subsystem' type='scsi'>
    <source protocol='iscsi' name='iqn.2014-08.com.example:iscsi-nopool/1'>
      <host name='example.com' port='3260'/>
      <auth username='myuser'>
        <secret type='iscsi' usage='libvirtiscsi'/>
      </auth>
      <initiator>
        <iqn name='iqn.2020-07.com.example:test'/>
      </initiator>
    </source>
    <address type='drive' controller='0' bus='0' target='0' unit='0'/>
  </hostdev>
</devices>
...

or:

...
<devices>
  <hostdev mode='subsystem' type='scsi_host'>
    <source protocol='vhost' wwpn='naa.50014057667280d8'/>
  </hostdev>
</devices>
...

or:

...
<devices>
  <hostdev mode='subsystem' type='mdev' model='vfio-pci'>
  <source>
    <address uuid='c2177883-f1bb-47f0-914d-32a22e3a8804'/>
  </source>
  </hostdev>
  <hostdev mode='subsystem' type='mdev' model='vfio-ccw'>
  <source>
    <address uuid='9063cba3-ecef-47b6-abcf-3fef4fdcad85'/>
  </source>
  <address type='ccw' cssid='0xfe' ssid='0x0' devno='0x0001'/>
  </hostdev>
</devices>
...
hostdev

The hostdev element is the main container for describing host devices. For each device, the mode is always "subsystem" and the type is one of the following values with additional attributes noted.

usb

USB devices are detached from the host on guest startup and reattached after the guest exits or the device is hot-unplugged.

pci

For PCI devices, when managed is "yes" it is detached from the host before being passed on to the guest and reattached to the host after the guest exits. If managed is omitted or "no", the user is responsible to call virNodeDeviceDetachFlags (or virsh nodedev-detach before starting the guest or hot-plugging the device and virNodeDeviceReAttach (or virsh nodedev-reattach) after hot-unplug or stopping the guest.

scsi

For SCSI devices, user is responsible to make sure the device is not used by host. If supported by the hypervisor and OS, the optional sgio ( since 1.0.6 ) attribute indicates whether unprivileged SG_IO commands are filtered for the disk. Valid settings are "filtered" or "unfiltered", where the default is "filtered". The optional rawio ( since 1.2.9 ) attribute indicates whether the lun needs the rawio capability. Valid settings are "yes" or "no". See the rawio description within the Hard drives, floppy disks, CDROMs section. If a disk lun in the domain already has the rawio capability, then this setting not required.

scsi_host

since 2.5.0 For SCSI devices, user is responsible to make sure the device is not used by host. This type passes all LUNs presented by a single HBA to the guest. Since 5.2.0, the model attribute can be specified further with "virtio-transitional", "virtio-non-transitional", or "virtio". Virtio transitional devices for more details.

mdev

For mediated devices ( Since 3.2.0 ) the model attribute specifies the device API which determines how the host's vfio driver will expose the device to the guest. Currently, model='vfio-pci', model='vfio-ccw' ( Since 4.4.0 ) and model='vfio-ap' ( Since 4.9.0 ) is supported. MDEV section provides more information about mediated devices as well as how to create mediated devices on the host. Since 4.6.0 (QEMU 2.12) an optional display attribute may be used to enable or disable support for an accelerated remote desktop backed by a mediated device (such as NVIDIA vGPU or Intel GVT-g) as an alternative to emulated Video devices. This attribute is limited to model='vfio-pci' only. Supported values are either on or off (default is 'off'). It is required to use a graphical framebuffer (See Graphical framebuffers) in order to use this attribute, currently only supported with VNC, Spice and egl-headless graphics devices. Since version 5.10.0 , there is an optional ramfb attribute for devices with model='vfio-pci'. Supported values are either on or off (default is 'off'). When enabled, this attribute provides a memory framebuffer device to the guest. This framebuffer will be used as a boot display when a vgpu device is the primary display.

Note: There are also some implications on the usage of guest's address type depending on the model attribute, see the address element below.

Note: The managed attribute is only used with type='pci' and is ignored by all the other device types, thus setting managed explicitly with other than a PCI device has the same effect as omitting it. Similarly, model attribute is only supported by mediated devices and ignored by all other device types.

source

The source element describes the device as seen from the host using the following mechanism to describe:

usb

The USB device can either be addressed by vendor / product id using the vendor and product elements or by the device's address on the host using the address element.

Since 1.0.0 , the source element of USB devices may contain startupPolicy attribute which can be used to define policy what to do if the specified host USB device is not found. The attribute accepts the following values:

mandatory fail if missing for any reason (the default)
requisite fail if missing on boot up, drop if missing on migrate/restore/revert
optional drop if missing at any start attempt

Since 8.6.0, the source element can contain guestReset attribute with the following value:

off all guest initiated device reset requests are ignored

uninitialized

device request is ignored if device is initialized, otherwise reset is performed

on device is reset on every guest initiated request

This attribute can be helpful when assigning an USB device with a firmware that crashes on reset.

pci

PCI devices can only be described by their address. Since 6.8.0 (Xen only) , the source element of a PCI device may contain the writeFiltering attribute to control write access to the PCI configuration space. By default Xen only allows writes of known safe values to the configuration space. Setting writeFiltering='no' will allow all writes to the device's PCI configuration space.

scsi

SCSI devices are described by both the adapter and address elements. The address element includes a bus attribute (a 2-digit bus number), a target attribute (a 10-digit target number), and a unit attribute (a 20-digit unit number on the bus). Not all hypervisors support larger target and unit values. It is up to each hypervisor to determine the maximum value supported for the adapter.

Since 1.2.8 , the source element of a SCSI device may contain the protocol attribute. When the attribute is set to "iscsi", the host device XML follows the network disk device (See Hard drives, floppy disks, CDROMs) using the same name attribute and optionally using the auth element to provide the authentication credentials to the iSCSI server.

Since 6.7.0, the optional initiator sub-element controls the IQN of the initiator ran by the hypervisor via it's <iqn name='iqn...' subelement.

scsi_host

Since 2.5.0 , multiple LUNs behind a single SCSI HBA are described by a protocol attribute set to "vhost" and a wwpn attribute that is the vhost_scsi wwpn (16 hexadecimal digits with a prefix of "naa.") established in the host configfs.

mdev

Mediated devices ( Since 3.2.0 ) are described by the address element. The address element contains a single mandatory attribute uuid.

vendor, product

The vendor and product elements each have an id attribute that specifies the USB vendor and product id. The ids can be given in decimal, hexadecimal (starting with 0x) or octal (starting with 0) form.

boot

Specifies that the device is bootable. The order attribute determines the order in which devices will be tried during boot sequence. The per-device boot elements cannot be used together with general boot elements in BIOS bootloader section. Since 0.8.8 for PCI devices, Since 1.0.1 for USB devices.

rom

The rom element is used to change how a PCI device's ROM is presented to the guest. The optional bar attribute can be set to "on" or "off", and determines whether or not the device's ROM will be visible in the guest's memory map. (In PCI documentation, the "rombar" setting controls the presence of the Base Address Register for the ROM). If no rom bar is specified, the qemu default will be used (older versions of qemu used a default of "off", while newer qemus have a default of "on"). Since 0.9.7 (QEMU and KVM only) . The optional file attribute contains an absolute path to a binary file to be presented to the guest as the device's ROM BIOS. This can be useful, for example, to provide a PXE boot ROM for a virtual function of an sr-iov capable ethernet device (which has no boot ROMs for the VFs). Since 0.9.10 (QEMU and KVM only) . The optional enabled attribute can be set to no to disable PCI ROM loading completely for the device; if PCI ROM loading is disabled through this attribute, attempts to tweak the loading process further using the bar or file attributes will be rejected. Since 4.3.0 (QEMU and KVM only) .

address

The address element for USB devices has a bus and device attribute to specify the USB bus and device number the device appears at on the host. The values of these attributes can be given in decimal, hexadecimal (starting with 0x) or octal (starting with 0) form. For PCI devices the element carries 4 attributes allowing to designate the device as can be found with the lspci or with virsh nodedev-list. For SCSI devices a 'drive' address type must be used. For mediated devices, which are software-only devices defining an allocation of resources on the physical parent device, the address type used must conform to the model attribute of element hostdev, e.g. any address type other than PCI for vfio-pci device API or any address type other than CCW for vfio-ccw device API will result in an error. See the Device Addresses section for more details on the address element.

driver

PCI devices can have an optional driver subelement that specifies which backend driver to use for PCI device assignment. Use the name attribute to select either "vfio" (for the new VFIO device assignment backend, which is compatible with UEFI SecureBoot) or "kvm" (the legacy device assignment handled directly by the KVM kernel module) Since 1.0.5 (QEMU and KVM only, requires kernel 3.6 or newer) . When specified, device assignment will fail if the requested method of device assignment isn't available on the host. When not specified, the default is "vfio" on systems where the VFIO driver is available and loaded, and "kvm" on older systems, or those where the VFIO driver hasn't been loaded Since 1.1.3 (prior to that the default was always "kvm").

readonly

Indicates that the device is readonly, only supported by SCSI host device now. Since 1.0.6 (QEMU and KVM only)

shareable

If present, this indicates the device is expected to be shared between domains (assuming the hypervisor and OS support this). Only supported by SCSI host device. Since 1.0.6

Note: Although shareable was introduced in 1.0.6 , it did not work as as expected until 1.2.2 .

Block / character devices

Block / character devices from the host can be passed through to the guest using the hostdev element. This is only possible with container based virtualization. Devices are specified by a fully qualified path. since after 1.0.1 for LXC :

...
<hostdev mode='capabilities' type='storage'>
  <source>
    <block>/dev/sdf1</block>
  </source>
</hostdev>
...
...
<hostdev mode='capabilities' type='misc'>
  <source>
    <char>/dev/input/event3</char>
  </source>
</hostdev>
...
...
<hostdev mode='capabilities' type='net'>
  <source>
    <interface>eth0</interface>
  </source>
</hostdev>
...
hostdev

The hostdev element is the main container for describing host devices. For block/character device passthrough mode is always "capabilities" and type is "storage" for a block device, "misc" for a character device and "net" for a host network interface.

source

The source element describes the device as seen from the host. For block devices, the path to the block device in the host OS is provided in the nested "block" element, while for character devices the "char" element is used. For network interfaces, the name of the interface is provided in the "interface" element.

Redirected devices

USB device redirection through a character device is supported since after 0.9.5 (KVM only) :

...
<devices>
  <redirdev bus='usb' type='spicevmc'/>
  <redirdev bus='usb' type='tcp'>
    <source mode='connect' host='localhost' service='4000'/>
    <boot order='1'/>
  </redirdev>
  <redirfilter>
    <usbdev class='0x08' vendor='0x1234' product='0xbeef' version='2.56' allow='yes'/>
    <usbdev allow='no'/>
  </redirfilter>
</devices>
...
redirdev

The redirdev element is the main container for describing redirected devices. bus must be "usb" for a USB device. An additional attribute type is required, matching one of the supported serial device types (See Consoles, serial, parallel & channel devices), to describe the host side of the tunnel; type='tcp' or type='spicevmc' (which uses the usbredir channel of a SPICE graphics device (See Graphical framebuffers)) are typical.

The redirdev element has an optional sub-element <address> which can tie the device to a particular controller. Further sub-elements, such as <source>, may be required according to the given type, although a <target> sub-element is not required (since the consumer of the character device is the hypervisor itself, rather than a device visible in the guest).

boot

Specifies that the device is bootable. The order attribute determines the order in which devices will be tried during boot sequence. The per-device boot elements cannot be used together with general boot elements in BIOS bootloader section. ( Since 1.0.1 )

redirfilter

Theredirfilterelement is used for creating the filter rule to filter out certain devices from redirection. It uses sub-element <usbdev> to define each filter rule. class attribute is the USB Class code, for example, 0x08 represents mass storage devices. The USB device can be addressed by vendor / product id using the vendor and product attributes. version is the device revision from the bcdDevice field (not the version of the USB protocol). These four attributes are optional and -1 can be used to allow any value for them. allow attribute is mandatory, 'yes' means allow, 'no' for deny.

Smartcard devices

A virtual smartcard device can be supplied to the guest via the smartcard element. A USB smartcard reader device on the host cannot be used on a guest with simple device passthrough, since it will then not be available on the host, possibly locking the host computer when it is "removed". Therefore, some hypervisors provide a specialized virtual device that can present a smartcard interface to the guest, with several modes for describing how credentials are obtained from the host or even a from a channel created to a third-party smartcard provider. Since 0.8.8

...
<devices>
  <smartcard mode='host'/>
  <smartcard mode='host-certificates'>
    <certificate>cert1</certificate>
    <certificate>cert2</certificate>
    <certificate>cert3</certificate>
    <database>/etc/pki/nssdb/</database>
  </smartcard>
  <smartcard mode='passthrough' type='tcp'>
    <source mode='bind' host='127.0.0.1' service='2001'/>
    <protocol type='raw'/>
    <address type='ccid' controller='0' slot='0'/>
  </smartcard>
  <smartcard mode='passthrough' type='spicevmc'/>
</devices>
...

The <smartcard> element has a mandatory attribute mode. The following modes are supported; in each mode, the guest sees a device on its USB bus that behaves like a physical USB CCID (Chip/Smart Card Interface Device) card.

host

The simplest operation, where the hypervisor relays all requests from the guest into direct access to the host's smartcard via NSS. No other attributes or sub-elements are required. See below about the use of an optional <address> sub-element.

host-certificates

Rather than requiring a smartcard to be plugged into the host, it is possible to provide three NSS certificate names residing in a database on the host. These certificates can be generated via the command certutil -d /etc/pki/nssdb -x -t CT,CT,CT -S -s CN=cert1 -n cert1, and the resulting three certificate names must be supplied as the content of each of three <certificate> sub-elements. An additional sub-element <database> can specify the absolute path to an alternate directory (matching the -d option of the certutil command when creating the certificates); if not present, it defaults to /etc/pki/nssdb.

passthrough

Rather than having the hypervisor directly communicate with the host, it is possible to tunnel all requests through a secondary character device to a third-party provider (which may in turn be talking to a smartcard or using three certificate files). In this mode of operation, an additional attribute type is required, matching one of the supported serial device types (See Consoles, serial, parallel & channel devices), to describe the host side of the tunnel; type='tcp' or type='spicevmc' (which uses the smartcard channel of a SPICE graphics device (See Graphical framebuffers)) are typical. Further sub-elements, such as <source>, may be required according to the given type, although a <target> sub-element is not required (since the consumer of the character device is the hypervisor itself, rather than a device visible in the guest).

Each mode supports an optional sub-element <address> (See Device Addresses), which fine-tunes the correlation between the smartcard and a ccid bus controller. For now, qemu only supports at most one smartcard, with an address of bus=0 slot=0.

Network interfaces

...
<devices>
  <interface type='direct' trustGuestRxFilters='yes'>
    <source dev='eth0'/>
    <mac address='52:54:00:5d:c7:9e'/>
    <boot order='1'/>
    <rom bar='off'/>
    <acpi index='4'/>
  </interface>
</devices>
...

There are several possibilities for specifying a network interface visible to the guest. Each subsection below provides more details about common setup options.

Since 1.2.10 ), the interface element property trustGuestRxFilters provides the capability for the host to detect and trust reports from the guest regarding changes to the interface mac address and receive filters by setting the attribute to yes. The default setting for the attribute is no for security reasons and support depends on the guest network device model as well as the type of connection on the host - currently it is only supported for the virtio device model and for macvtap connections on the host.

Each <interface> element has an optional <address> sub-element that can tie the interface to a particular pci slot, with attribute type='pci' as documented in the Device Addresses section.

Since 6.6.0 , one can force libvirt to keep the provided MAC address when it's in the reserved VMware range by adding a type="static" attribute to the <mac/> element. Note that this attribute is useless if the provided MAC address is outside of the reserved VMWare ranges.

Since 7.3.0, one can set the ACPI index against network interfaces. With some operating systems (eg Linux with systemd), the ACPI index is used to provide network interface device naming, that is stable across changes in PCI addresses assigned to the device. This value is required to be unique across all devices and be between 1 and (16*1024-1).

Virtual network

This is the recommended config for general guest connectivity on hosts with dynamic / wireless networking configs. (or multi-host environments where the host hardware details are described separately in a <network> definition Since 0.9.4 ).

Provides a connection whose details are described by the named network definition. Depending on the virtual network's "forward mode" configuration, the network may be totally isolated (no <forward> element given), NAT'ing to an explicit network device or to the default route (<forward mode='nat'>), routed with no NAT (<forward mode='route'/>), or connected directly to one of the host's network interfaces (via macvtap) or bridge devices ((<forward mode='bridge|private|vepa|passthrough'/> Since 0.9.4 )

For networks with a forward mode of bridge, private, vepa, and passthrough, it is assumed that the host has any necessary DNS and DHCP services already setup outside the scope of libvirt. In the case of isolated, nat, and routed networks, DHCP and DNS are provided on the virtual network by libvirt, and the IP range can be determined by examining the virtual network config with 'virsh net-dumpxml [networkname]'. There is one virtual network called 'default' setup out of the box which does NAT'ing to the default route and has an IP range of 192.168.122.0/255.255.255.0. Each guest will have an associated tun device created with a name of vnetN, which can also be overridden with the <target> element (see Overriding the target element).

When the source of an interface is a network, a portgroup can be specified along with the name of the network; one network may have multiple portgroups defined, with each portgroup containing slightly different configuration information for different classes of network connections. Since 0.9.4 .

When a guest is running an interface of type network may include a portid attribute. This provides the UUID of an associated virNetworkPortPtr object that records the association between the domain interface and the network. This attribute is read-only since port objects are create and deleted automatically during startup and shutdown. Since 5.1.0

Also, similar to direct network connections (described below), a connection of type network may specify a virtualport element, with configuration data to be forwarded to a vepa (802.1Qbg) or 802.1Qbh compliant switch ( Since 0.8.2 ), or to an Open vSwitch virtual switch ( Since 0.9.11 ).

Since the actual type of switch may vary depending on the configuration in the <network> on the host, it is acceptable to omit the virtualport type attribute, and specify attributes from multiple different virtualport types (and also to leave out certain attributes); at domain startup time, a complete <virtualport> element will be constructed by merging together the type and attributes defined in the network and the portgroup referenced by the interface. The newly-constructed virtualport is a combination of them. The attributes from lower virtualport can't make change on the ones defined in higher virtualport. Interface takes the highest priority, portgroup is lowest priority. ( Since 0.10.0 ). For example, in order to work properly with both an 802.1Qbh switch and an Open vSwitch switch, you may choose to specify no type, but both a profileid (in case the switch is 802.1Qbh) and an interfaceid (in case the switch is Open vSwitch) (you may also omit the other attributes, such as managerid, typeid, or profileid, to be filled in from the network's <virtualport>). If you want to limit a guest to connecting only to certain types of switches, you can specify the virtualport type, but still omit some/all of the parameters - in this case if the host's network has a different type of virtualport, connection of the interface will fail.

...
<devices>
  <interface type='network'>
    <source network='default'/>
  </interface>
  ...
  <interface type='network'>
    <source network='default' portgroup='engineering'/>
    <target dev='vnet7'/>
    <mac address="00:11:22:33:44:55"/>
    <virtualport>
      <parameters instanceid='09b11c53-8b5c-4eeb-8f00-d84eaa0aaa4f'/>
    </virtualport>
  </interface>
</devices>
...
Bridge to LAN

This is the recommended config for general guest connectivity on hosts with static wired networking configs.

Provides a bridge from the VM directly to the LAN. This assumes there is a bridge device on the host which has one or more of the hosts physical NICs attached. The guest VM will have an associated tun device created with a name of vnetN, which can also be overridden with the <target> element (see Overriding the target element). The tun device will be attached to the bridge. The IP range / network configuration is whatever is used on the LAN. This provides the guest VM full incoming & outgoing net access just like a physical machine.

On Linux systems, the bridge device is normally a standard Linux host bridge. On hosts that support Open vSwitch, it is also possible to connect to an Open vSwitch bridge device by adding a <virtualport type='openvswitch'/> to the interface definition. ( Since 0.9.11 ). The Open vSwitch type virtualport accepts two parameters in its <parameters> element - an interfaceid which is a standard uuid used to uniquely identify this particular interface to Open vSwitch (if you do not specify one, a random interfaceid will be generated for you when you first define the interface), and an optional profileid which is sent to Open vSwitch as the interfaces "port-profile".

...
<devices>
  ...
  <interface type='bridge'>
    <source bridge='br0'/>
  </interface>
  <interface type='bridge'>
    <source bridge='br1'/>
    <target dev='vnet7'/>
    <mac address="00:11:22:33:44:55"/>
  </interface>
  <interface type='bridge'>
    <source bridge='ovsbr'/>
    <virtualport type='openvswitch'>
      <parameters profileid='menial' interfaceid='09b11c53-8b5c-4eeb-8f00-d84eaa0aaa4f'/>
    </virtualport>
  </interface>
  ...
</devices>
...

On hosts that support Open vSwitch on the kernel side and have the Midonet Host Agent configured, it is also possible to connect to the 'midonet' bridge device by adding a <virtualport type='midonet'/> to the interface definition. ( Since 1.2.13 ). The Midonet virtualport type requires an interfaceid attribute in its <parameters> element. This interface id is the UUID that specifies which port in the virtual network topology will be bound to the interface.

...
<devices>
  ...
  <interface type='bridge'>
    <source bridge='br0'/>
  </interface>
  <interface type='bridge'>
    <source bridge='br1'/>
    <target dev='vnet7'/>
    <mac address="00:11:22:33:44:55"/>
  </interface>
  <interface type='bridge'>
    <source bridge='midonet'/>
    <virtualport type='midonet'>
      <parameters interfaceid='0b2d64da-3d0e-431e-afdd-804415d6ebbb'/>
    </virtualport>
  </interface>
  ...
</devices>
...
Userspace (SLIRP or passt) connection

The user type connects the guest interface to the outside via a transparent userspace proxy that doesn't require any special system privileges, making it usable in cases when libvirt itself is running with no privileges (e.g. libvirt's "session mode" daemon, or when libvirt is run inside an unprivileged container).

By default, this user proxy is done with QEMU's internal SLIRP driver which has DHCP & DNS services that give the guest IP addresses starting from 10.0.2.15, a default route of 10.0.2.2 and DNS server of 10.0.2.3. Since 3.8.0 it is possible to override the default network address by including an ip element specifying an IPv4 address in its one mandatory attribute, address. Optionally, a second ip element with a family attribute set to "ipv6" can be specified to add an IPv6 address to the interface. address. Optionally, address prefix can be specified.

...
<devices>
  <interface type='user'/>
  ...
  <interface type='user'>
    <mac address="00:11:22:33:44:55"/>
    <ip family='ipv4' address='172.17.2.0' prefix='24'/>
    <ip family='ipv6' address='2001:db8:ac10:fd01::' prefix='64'/>
  </interface>
</devices>
...

Since 9.0.0 an alternate backend implementation of the user interface type can be selected by setting the interface's <backend> subelement type attribute to passt. In this case, the passt transport (https://passt.top) is used. Similar to SLIRP, passt has an internal DHCP server that provides a requesting guest with one ipv4 and one ipv6 address; it then uses userspace proxies and a separate network namespace to provide outgoing UDP/TCP/ICMP sessions, and optionally redirect incoming traffic destined for the host toward the guest instead.

When the passt backend is used, the <backend> attribute logFile can be used to tell the passt process for this interface where to write its message log, and the <source> attribute dev can tell it to use a particular host interface to derive the routes given to the guest for forwarding traffic upstream.

Additionally, when passt is used, multiple <portForward> elements can be added to forward incoming network traffic for the host to this guest interface. Each <portForward> must have a proto attribute (set to tcp or udp) and optional original address (if not specified, then all incoming sessions to any host IP for the given proto/port(s) will be forwarded to the guest).

The decision of which ports to forward is described with zero or more <range> subelements of <portForward> (if there is no <range> then all ports for the given proto/address will be forwarded). Each <range> has a start and optional end attribute. If end is omitted then a single port will be forwarded, otherwise all ports between start and end (inclusive) will be forwarded. If the port number(s) should remain unmodified as the session is forwarded, no further options are needed, but if the guest is expecting the sessions on a different port, then this should be specified with the to attribute of <range> - the port number of each forwarded session in the range will be offeset by "to -start". A <range> element can also be used to specify a range of ports that should not be forwarded. This is done by setting the range's exclude attribute to yes. This may not seem very useful, but can be when it is desirable to forward a long range of ports with the exception of some subset.

...
<devices>
  ...
  <interface type='user'>
    <backend type='passt' logFile='/tmp/passt.log'/>
    <mac address="00:11:22:33:44:55"/>
    <source dev='eth0'/>
    <ip family='ipv4' address='172.17.2.4' prefix='24'/>
    <ip family='ipv6' address='2001:db8:ac10:fd01::20'/>
    <portForward proto='tcp'>
      <range start='2022' to='22'/>
    </portForward>
    <portForward proto='udp' address='1.2.3.4'>
      <range start='5000' end='5020' to='6000'/>
      <range start='5010' end='5015' exclude='yes'/>
    </portForward>
    <portForward proto='tcp' address='2001:db8:ac10:fd01::1:10'>
      <range start='80'/>
      <range start='443' to='344'/>
    </portForward>
  </interface>
</devices>
...
Generic ethernet connection

Provides a means to use a new or existing tap device (or veth device pair, depending on the needs of the hypervisor driver) that is partially or wholly setup external to libvirt (either prior to the guest starting, or while the guest is being started via an optional script specified in the config).

The name of the tap device can optionally be specified with the dev attribute of the <target> element. If no target dev is specified, libvirt will create a new standard tap device with a name of the pattern "vnetN", where "N" is replaced with a number. If a target dev is specified and that device doesn't exist, then a new standard tap device will be created with the exact dev name given. If the specified target dev does exist, then that existing device will be used. Usually some basic setup of the device is done by libvirt, including setting a MAC address, and the IFF_UP flag, but if the dev is a pre-existing device, and the managed attribute of the target element is also set to "no" (the default value is "yes"), even this basic setup will not be performed - libvirt will simply pass the device on to the hypervisor with no setup at all. Since 5.7.0 Using managed='no' with a pre-created tap device is useful because it permits a virtual machine managed by an unprivileged libvirtd to have emulated network devices based on tap devices.

After creating/opening the tap device, an optional shell script (given in the path attribute of the <script> element) will be run. Since 0.2.1 Also, after detaching/closing the tap device, an optional shell script (given in the path attribute of the <downscript> element) will be run. Since 6.4.0 These can be used to do whatever extra host network integration is required.

...
<devices>
  <interface type='ethernet'>
    <script path='/etc/qemu-ifup-mynet'/>
    <downscript path='/etc/qemu-ifdown-mynet'/>
  </interface>
  ...
  <interface type='ethernet'>
    <target dev='mytap1' managed='no'/>
    <model type='virtio'/>
  </interface>
</devices>
...
Direct attachment to physical interface
Provides direct attachment of the virtual machine's NIC to the given physical interface of the host. Since 0.7.7 (QEMU and KVM only)
This setup requires the Linux macvtap driver to be available. (Since Linux 2.6.34.) One of the modes 'vepa' ( 'Virtual Ethernet Port Aggregator'), 'bridge' or 'private' can be chosen for the operation mode of the macvtap device, 'vepa' being the default mode. The individual modes cause the delivery of packets to behave as follows:

If the model type is set to virtio and interface's trustGuestRxFilters attribute is set to yes, changes made to the interface mac address, unicast/multicast receive filters, and vlan settings in the guest will be monitored and propagated to the associated macvtap device on the host ( Since 1.2.10 ). If trustGuestRxFilters is not set, or is not supported for the device model in use, an attempted change to the mac address originating from the guest side will result in a non-working network connection.

vepa

All VMs' packets are sent to the external bridge. Packets whose destination is a VM on the same host as where the packet originates from are sent back to the host by the VEPA capable bridge (today's bridges are typically not VEPA capable).

bridge

Packets whose destination is on the same host as where they originate from are directly delivered to the target macvtap device. Both origin and destination devices need to be in bridge mode for direct delivery. If either one of them is in vepa mode, a VEPA capable bridge is required.

private

All packets are sent to the external bridge and will only be delivered to a target VM on the same host if they are sent through an external router or gateway and that device sends them back to the host. This procedure is followed if either the source or destination device is in private mode.

passthrough

This feature attaches a virtual function of a SRIOV capable NIC directly to a VM without losing the migration capability. All packets are sent to the VF/IF of the configured network device. Depending on the capabilities of the device additional prerequisites or limitations may apply; for example, on Linux this requires kernel 2.6.38 or newer. Since 0.9.2

...
<devices>
  ...
  <interface type='direct' trustGuestRxFilters='no'>
    <source dev='eth0' mode='vepa'/>
  </interface>
</devices>
...

The network access of direct attached virtual machines can be managed by the hardware switch to which the physical interface of the host machine is connected to.

The interface can have additional parameters as shown below, if the switch is conforming to the IEEE 802.1Qbg standard. The parameters of the virtualport element are documented in more detail in the IEEE 802.1Qbg standard. The values are network specific and should be provided by the network administrator. In 802.1Qbg terms, the Virtual Station Interface (VSI) represents the virtual interface of a virtual machine. Since 0.8.2

Please note that IEEE 802.1Qbg requires a non-zero value for the VLAN ID.

managerid

The VSI Manager ID identifies the database containing the VSI type and instance definitions. This is an integer value and the value 0 is reserved.

typeid

The VSI Type ID identifies a VSI type characterizing the network access. VSI types are typically managed by network administrator. This is an integer value.

typeidversion

The VSI Type Version allows multiple versions of a VSI Type. This is an integer value.

instanceid

The VSI Instance ID Identifier is generated when a VSI instance (i.e. a virtual interface of a virtual machine) is created. This is a globally unique identifier.

...
<devices>
  ...
  <interface type='direct'>
    <source dev='eth0.2' mode='vepa'/>
    <virtualport type="802.1Qbg">
      <parameters managerid="11" typeid="1193047" typeidversion="2" instanceid="09b11c53-8b5c-4eeb-8f00-d84eaa0aaa4f"/>
    </virtualport>
  </interface>
</devices>
...

The interface can have additional parameters as shown below if the switch is conforming to the IEEE 802.1Qbh standard. The values are network specific and should be provided by the network administrator. Since 0.8.2

profileid

The profile ID contains the name of the port profile that is to be applied to this interface. This name is resolved by the port profile database into the network parameters from the port profile, and those network parameters will be applied to this interface.

...
<devices>
  ...
  <interface type='direct'>
    <source dev='eth0' mode='private'/>
    <virtualport type='802.1Qbh'>
      <parameters profileid='finance'/>
    </virtualport>
  </interface>
</devices>
...
PCI Passthrough

A PCI network device (specified by the <source> element) is directly assigned to the guest using generic device passthrough, after first optionally setting the device's MAC address to the configured value, and associating the device with an 802.1Qbh capable switch using an optionally specified <virtualport> element (see the examples of virtualport given above for type='direct' network devices). Note that - due to limitations in standard single-port PCI ethernet card driver design - only SR-IOV (Single Root I/O Virtualization) virtual function (VF) devices can be assigned in this manner; to assign a standard single-port PCI or PCIe ethernet card to a guest, use the traditional <hostdev> device definition and Since 0.9.11

To use VFIO device assignment rather than traditional/legacy KVM device assignment (VFIO is a new method of device assignment that is compatible with UEFI Secure Boot), a type='hostdev' interface can have an optional driver sub-element with a name attribute set to "vfio". To use legacy KVM device assignment you can set name to "kvm" (the default is "vfio" on systems where the VFIO driver is available, and "kvm" on older systems. Since 1.1.3 (prior to that the default was always "kvm").

Note that this "intelligent passthrough" of network devices is very similar to the functionality of a standard <hostdev> device, the difference being that this method allows specifying a MAC address and <virtualport> for the passed-through device. If these capabilities are not required, if you have a standard single-port PCI, PCIe, or USB network card that doesn't support SR-IOV (and hence would anyway lose the configured MAC address during reset after being assigned to the guest domain), or if you are using a version of libvirt older than 0.9.11, you should use standard <hostdev> to assign the device to the guest instead of <interface type='hostdev'/>.

Similar to the functionality of a standard <hostdev> device, when managed is "yes", it is detached from the host before being passed on to the guest, and reattached to the host after the guest exits. If managed is omitted or "no", the user is responsible to call virNodeDeviceDettach (or virsh nodedev-detach) before starting the guest or hot-plugging the device, and virNodeDeviceReAttach (or virsh nodedev-reattach) after hot-unplug or stopping the guest.

...
<devices>
  <interface type='hostdev' managed='yes'>
    <driver name='vfio'/>
    <source>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x07' function='0x0'/>
    </source>
    <mac address='52:54:00:6d:90:02'/>
    <virtualport type='802.1Qbh'>
      <parameters profileid='finance'/>
    </virtualport>
  </interface>
</devices>
...
vDPA devices

A vDPA network device can be used to provide wire speed network performance within a domain. A vDPA device is a specialized type of network device that uses a datapath that complies with the virtio specification but has a vendor-specific control path. To use such a device with libvirt, the host device must already be bound to the appropriate device-specific vDPA driver. This creates a vDPA char device (e.g. /dev/vhost-vdpa-0) that can be used to assign the device to a libvirt domain. Since 6.9.0 (QEMU only, requires QEMU 5.1.0 or newer)

...
<devices>
  <interface type='vdpa'>
    <source dev='/dev/vhost-vdpa-0'/>
  </interface>
</devices>
...
Teaming a virtio/hostdev NIC pair

Since 6.1.0 (QEMU and KVM only, requires QEMU 4.2.0 or newer and a guest virtio-net driver supporting the "failover" feature, such as the one included in Linux kernel 4.18 and newer) The <teaming> element of two interfaces can be used to connect them as a team/bond device in the guest (assuming proper support in the hypervisor and the guest network driver).

...
<devices>
  <interface type='network'>
    <source network='mybridge'/>
    <mac address='00:11:22:33:44:55'/>
    <model type='virtio'/>
    <teaming type='persistent'/>
    <alias name='ua-backup0'/>
  </interface>
  <interface type='network'>
    <source network='hostdev-pool'/>
    <mac address='00:11:22:33:44:55'/>
    <model type='virtio'/>
    <teaming type='transient' persistent='ua-backup0'/>
  </interface>
</devices>
...

The second interface in this example is referencing a network that is a pool of SRIOV VFs (i.e. a "hostdev network"). You could instead directly reference an SRIOV VF device:

...
  <interface type='hostdev'>
    <source>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x07' function='0x0'/>
    </source>
    <mac address='00:11:22:33:44:55'/>
    <teaming type='transient' persistent='ua-backup0'/>
  </interface>
...

The <teaming> element required attribute type will be set to either persistent to indicate a device that should always be present in the domain, or transient to indicate a device that may periodically be removed, then later re-added to the domain. When type="transient", there should be a second attribute to <teaming> called persistent - this attribute should be set to the alias name of the other device in the pair (the one that has <teaming type="persistent'/>).

In the particular case of QEMU, libvirt's <teaming> element is used to setup a virtio-net "failover" device pair. For this setup, the persistent device must be an interface with <model type="virtio"/>, and the transient device must be <interface type='hostdev'/> (or <interface type='network'/> where the referenced network defines a pool of SRIOV VFs). The guest will then have a simple network team/bond device made of the virtio NIC + hostdev NIC pair. In this configuration, the higher-performing hostdev NIC will normally be preferred for all network traffic, but when the domain is migrated, QEMU will automatically unplug the VF from the guest, and then hotplug a similar device once migration is completed; while migration is taking place, network traffic will use the virtio NIC. (Of course the emulated virtio NIC and the hostdev NIC must be connected to the same subnet for bonding to work properly).

Since 7.1.0 The <teaming> element can also be added to a plain <hostdev> device.

...
  <hostdev mode='subsystem' type='pci' managed='no'>
    <source>
      <address domain='0x0000' bus='0x00' slot='0x07' function='0x0'/>
    </source>
    <teaming type='transient' persistent='ua-backup0'/>
  </hostdev>
...

This device must be a network device, but not necessarily an SRIOV VF. Using plain <hostdev> rather than <interface type='hostdev'> or <interface type='network'> is useful if the device that will be assigned with VFIO is a standard NIC (not a VF) or if libvirt doesn't have the necessary resources and privileges to set the VF's MAC address (e.g. if libvirt is running unprivileged, or in a container). This of course means that the user (or another application) is responsible for setting the MAC address of the device in a way such that it will survive guest driver initialization. For standard NICs (i.e. not an SRIOV VF) this probably means that the NIC's factory-programmed MAC address will need to be used for the teaming pair (since any driver init in the guest will reset the MAC back to factory). If it is an SRIOV VF, then its MAC address will need to be set via the VF's PF, e.g. if you are going to use VF 2 of the PF enp2s0f1, you would use something like this command:

ip link set enp2s0f1 vf 2 mac 52:54:00:11:22:33

NB1: Since you must know the alias name of the virtio NIC when configuring the hostdev NIC, it will need to be manually set in the virtio NIC's configuration (as with all other manually set alias names, this means it must start with "ua-").

NB2: Currently the only implementation of the guest OS virtio-net driver supporting virtio-net failover requires that the MAC addresses of the virtio and hostdev NIC must match. Since that may not always be a requirement in the future, libvirt doesn't enforce this limitation - it is up to the person/management application that is creating the configuration to assure the MAC addresses of the two devices match.

NB3: Since the PCI addresses of the SRIOV VFs on the hosts that are the source and destination of the migration will almost certainly be different, either higher level management software will need to modify the <source> of the hostdev NIC (<interface type='hostdev'>) at the start of migration, or (a simpler solution) the configuration will need to use a libvirt "hostdev" virtual network that maintains a pool of such devices, as is implied in the example's use of the libvirt network named "hostdev-pool" - as long as the hostdev network pools on both hosts have the same name, libvirt itself will take care of allocating an appropriate device on both ends of the migration. Similarly the XML for the virtio interface must also either work correctly unmodified on both the source and destination of the migration (e.g. by connecting to the same bridge device on both hosts, or by using the same virtual network), or the management software must properly modify the interface XML during migration so that the virtio device remains connected to the same network segment before and after migration.

Multicast tunnel

A multicast group is setup to represent a virtual network. Any VMs whose network devices are in the same multicast group can talk to each other even across hosts. This mode is also available to unprivileged users. There is no default DNS or DHCP support and no outgoing network access. To provide outgoing network access, one of the VMs should have a 2nd NIC which is connected to one of the first 4 network types and do the appropriate routing. The multicast protocol is compatible with that used by user mode linux guests too. The source address used must be from the multicast address block.

...
<devices>
  <interface type='mcast'>
    <mac address='52:54:00:6d:90:01'/>
    <source address='230.0.0.1' port='5558'/>
  </interface>
</devices>
...
TCP tunnel

A TCP client/server architecture provides a virtual network. One VM provides the server end of the network, all other VMS are configured as clients. All network traffic is routed between the VMs via the server. This mode is also available to unprivileged users. There is no default DNS or DHCP support and no outgoing network access. To provide outgoing network access, one of the VMs should have a 2nd NIC which is connected to one of the first 4 network types and do the appropriate routing.

...
<devices>
  <interface type='server'>
    <mac address='52:54:00:22:c9:42'/>
    <source address='192.168.0.1' port='5558'/>
  </interface>
  ...
  <interface type='client'>
    <mac address='52:54:00:8b:c9:51'/>
    <source address='192.168.0.1' port='5558'/>
  </interface>
</devices>
...
UDP unicast tunnel

A UDP unicast architecture provides a virtual network which enables connections between QEMU instances using QEMU's UDP infrastructure. The xml "source" address is the endpoint address to which the UDP socket packets will be sent from the host running QEMU. The xml "local" address is the address of the interface from which the UDP socket packets will originate from the QEMU host. Since 1.2.20

...
<devices>
  <interface type='udp'>
    <mac address='52:54:00:22:c9:42'/>
    <source address='127.0.0.1' port='11115'>
      <local address='127.0.0.1' port='11116'/>
    </source>
  </interface>
</devices>
...
Null network interface

An unconnected network interface sounds pretty pointless, but can show up for example with VMWare without any specified network to be connected to. Since 8.7.0

...
<devices>
  <interface type='null'>
    <mac address='52:54:00:22:c9:42'/>
  </interface>
</devices>
...
VMWare Distributed Switch

Interface can be connected to VMWare Distributed Switch, but since libvirt cannot provide information about that architecture, the information presented here is only what can be gathered from the VM configuration. VMs with this interface type can be created, so that editing of the XML works properly, however libvirt cannot guarantee that any changes in these parameters will be valid in the hypervisor. Since 8.7.0

...
<devices>
  <interface type='vds'>
    <mac address='52:54:00:22:c9:42'/>
    <source switchid='12345678-1234-1234-1234-123456789abc' portid='6' portgroupid='pg-4321' connectionid='12345'/>
  </interface>
</devices>
...
Setting the NIC model
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet1'/>
    <model type='ne2k_pci'/>
  </interface>
</devices>
...

For hypervisors which support this, you can set the model of emulated network interface card.

The values for type aren't defined specifically by libvirt, but by what the underlying hypervisor supports (if any). For QEMU and KVM you can get a list of supported models with these commands:

qemu -net nic,model=? /dev/null
qemu-kvm -net nic,model=? /dev/null

Typical values for QEMU and KVM include: ne2k_isa i82551 i82557b i82559er ne2k_pci pcnet rtl8139 e1000 virtio. Since 5.2.0 , virtio-transitional and virtio-non-transitional values are supported. See Virtio transitional devices for more details. Since 9.3.0 igb is also supported.

Setting NIC driver-specific options
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet1'/>
    <model type='virtio'/>
    <driver name='vhost' txmode='iothread' ioeventfd='on' event_idx='off' queues='5' rx_queue_size='256' tx_queue_size='256'>
      <host csum='off' gso='off' tso4='off' tso6='off' ecn='off' ufo='off' mrg_rxbuf='off'/>
      <guest csum='off' tso4='off' tso6='off' ecn='off' ufo='off'/>
    </driver>
    </interface>
</devices>
...

Some NICs may have tunable driver-specific options. These are set as attributes of the driver sub-element of the interface definition. Currently the following attributes are available for the virtio NIC driver:

name

The optional name attribute forces which type of backend driver to use. The value can be either 'qemu' (a user-space backend) or 'vhost' (a kernel backend, which requires the vhost module to be provided by the kernel); an attempt to require the vhost driver without kernel support will be rejected. If this attribute is not present, then the domain defaults to 'vhost' if present, but silently falls back to 'qemu' without error. Since 0.8.8 (QEMU and KVM only) For interfaces of type='hostdev' (PCI passthrough devices) the name attribute can optionally be set to "vfio" or "kvm". "vfio" tells libvirt to use VFIO device assignment rather than traditional KVM device assignment (VFIO is a new method of device assignment that is compatible with UEFI Secure Boot), and "kvm" tells libvirt to use the legacy device assignment performed directly by the kvm kernel module (the default is currently "kvm", but is subject to change). Since 1.0.5 (QEMU and KVM only, requires kernel 3.6 or newer) For interfaces of type='vhostuser', the name attribute is ignored. The backend driver used is always vhost-user.

txmode

The txmode attribute specifies how to handle transmission of packets when the transmit buffer is full. The value can be either 'iothread' or 'timer'. Since 0.8.8 (QEMU and KVM only) If set to 'iothread', packet tx is all done in an iothread in the bottom half of the driver (this option translates into adding "tx=bh" to the qemu commandline -device virtio-net-pci option). If set to 'timer', tx work is done in qemu, and if there is more tx data than can be sent at the present time, a timer is set before qemu moves on to do other things; when the timer fires, another attempt is made to send more data. The resulting difference, according to the qemu developer who added the option is: "bh makes tx more asynchronous and reduces latency, but potentially causes more processor bandwidth contention since the CPU doing the tx isn't necessarily the CPU where the guest generated the packets." In general you should leave this option alone, unless you are very certain you know what you are doing.

ioeventfd

This optional attribute allows users to set domain I/O asynchronous handling for interface device. The default is left to the discretion of the hypervisor. Accepted values are "on" and "off". Enabling this allows qemu to execute VM while a separate thread handles I/O. Typically guests experiencing high system CPU utilization during I/O will benefit from this. On the other hand, on overloaded host it could increase guest I/O latency. Since 0.9.3 (QEMU and KVM only) In general you should leave this option alone, unless you are very certain you know what you are doing.

event_idx

The event_idx attribute controls some aspects of device event processing. The value can be either 'on' or 'off' - if it is on, it will reduce the number of interrupts and exits for the guest. The default is determined by QEMU; usually if the feature is supported, default is on. In case there is a situation where this behavior is suboptimal, this attribute provides a way to force the feature off. Since 0.9.5 (QEMU and KVM only) In general you should leave this option alone, unless you are very certain you know what you are doing.

queues

The optional queues attribute controls the number of queues to be used for either Multiqueue virtio-net or vhost-user (See vhost-user interface) network interfaces. Use of multiple packet processing queues requires the interface having the <model type='virtio'/> element. Each queue will potentially be handled by a different processor, resulting in much higher throughput. virtio-net since 1.0.6 (QEMU and KVM only) vhost-user since 1.2.17 (QEMU and KVM only)

rx_queue_size

The optional rx_queue_size attribute controls the size of virtio ring for each queue as described above. The default value is hypervisor dependent and may change across its releases. Moreover, some hypervisors may pose some restrictions on actual value. For instance, latest QEMU (as of 2016-09-01) requires value to be a power of two from [256, 1024] range. Since 2.3.0 (QEMU and KVM only) In general you should leave this option alone, unless you are very certain you know what you are doing.

tx_queue_size

The optional tx_queue_size attribute controls the size of virtio ring for each queue as described above. The default value is hypervisor dependent and may change across its releases. Moreover, some hypervisors may pose some restrictions on actual value. For instance, QEMU v2.9 requires value to be a power of two from [256, 1024] range. In addition to that, this may work only for a subset of interface types, e.g. aforementioned QEMU enables this option only for vhostuser type. Since 3.7.0 (QEMU and KVM only) In general you should leave this option alone, unless you are very certain you know what you are doing.

rss

The rss option enables in-qemu/ebpf RSS for virtio NIC. RSS works with virtio and tap backends only. Virtio NIC will be launched with "rss" property. For now "in-qemu" RSS is supported by libvirt. QEMU may load eBPF RSS if it has CAP_SYS_ADMIN permissions, which is not supported by default in libvirt. Since 8.3.0 and QEMU 5.1 In general you should leave this option alone, unless you are very certain you know what you are doing. Proper RSS configuration depends from vcpu, tap, and vhost settings.

rss_hash_report

The rss_hash_report option enables in-qemu RSS hash report for virtio NIC. Virtio NIC will be launched with a "hash" property. Network packets provided to VM will contain a hash of the packet in the virt header. Usually enabled alongside with rss. Without rss option, the hash report doesn't affect steering itself but provides vnet header with a calculated hash. Since 8.3.0 and QEMU 5.1 In general you should leave this option alone, unless you are very certain you know what you are doing. Proper RSS configuration depends from vcpu, tap, and vhost settings.

virtio options

For virtio interfaces, Virtio-related options can also be set. ( Since 3.5.0 )

Offloading options for the host and guest can be configured using the following sub-elements:

host

The csum, gso, tso4, tso6, ecn and ufo attributes with possible values on and off can be used to turn off host offloading options. By default, the supported offloads are enabled by QEMU. Since 1.2.9 (QEMU only) The mrg_rxbuf attribute can be used to control mergeable rx buffers on the host side. Possible values are on (default) and off. Since 1.2.13 (QEMU only)

guest

The csum, tso4, tso6, ecn and ufo attributes with possible values on and off can be used to turn off guest offloading options. By default, the supported offloads are enabled by QEMU. Since 1.2.9 (QEMU only)

Setting network backend-specific options
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet1'/>
    <model type='virtio'/>
    <backend tap='/dev/net/tun' vhost='/dev/vhost-net'/>
    <driver name='vhost' txmode='iothread' ioeventfd='on' event_idx='off' queues='5'/>
    <tune>
      <sndbuf>1600</sndbuf>
    </tune>
  </interface>
</devices>
...

For tuning the backend of the network, the backend element can be used. The vhost attribute can override the default vhost device path (/dev/vhost-net) for devices with virtio model. The tap attribute overrides the tun/tap device path (default: /dev/net/tun) for network and bridge interfaces. This does not work in session mode. Since 1.2.9

For tap devices there is also sndbuf element which can adjust the size of send buffer in the host. Since 0.8.8

Overriding the target element
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet1'/>
  </interface>
</devices>
...

If no target is specified, certain hypervisors will automatically generate a name for the created tun device. This name can be manually specified, however the name should not start with either 'vnet', 'vif', 'macvtap', or 'macvlan', which are prefixes reserved by libvirt and certain hypervisors. Manually specified targets using these prefixes may be ignored.

Note that for LXC containers, this defines the name of the interface on the host side. Since 1.2.7 , to define the name of the device on the guest side, the guest element should be used, as in the following snippet:

...
<devices>
  <interface type='network'>
    <source network='default'/>
    <guest dev='myeth'/>
  </interface>
</devices>
...
Specifying boot order
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet1'/>
    <boot order='1'/>
  </interface>
</devices>
...

For hypervisors which support this, you can set a specific NIC to be used for network boot. The order attribute determines the order in which devices will be tried during boot sequence. The per-device boot elements cannot be used together with general boot elements in BIOS bootloader section. Since 0.8.8

Interface ROM BIOS configuration
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet1'/>
    <rom bar='on' file='/etc/fake/boot.bin'/>
  </interface>
</devices>
...

For hypervisors which support this, you can change how a PCI Network device's ROM is presented to the guest. The bar attribute can be set to "on" or "off", and determines whether or not the device's ROM will be visible in the guest's memory map. (In PCI documentation, the "rombar" setting controls the presence of the Base Address Register for the ROM). If no rom bar is specified, the qemu default will be used (older versions of qemu used a default of "off", while newer qemus have a default of "on"). The optional file attribute is used to point to a binary file to be presented to the guest as the device's ROM BIOS. This can be useful to provide an alternative boot ROM for a network device. Since 0.9.10 (QEMU and KVM only) .

Setting up a network backend in a driver domain
...
<devices>
  ...
  <interface type='bridge'>
    <source bridge='br0'/>
    <backenddomain name='netvm'/>
  </interface>
  ...
</devices>
...

The optional backenddomain element allows specifying a backend domain (aka driver domain) for the interface. Use the name attribute to specify the backend domain name. You can use it to create a direct network link between domains (so data will not go through host system). Use with type 'ethernet' to create plain network link, or with type 'bridge' to connect to a bridge inside the backend domain. Since 1.2.13 (Xen only)

Quality of service
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet0'/>
    <bandwidth>
      <inbound average='1000' peak='5000' floor='200' burst='1024'/>
      <outbound average='128' peak='256' burst='256'/>
    </bandwidth>
  </interface>
</devices>
...

This part of interface XML provides setting quality of service. Incoming and outgoing traffic can be shaped independently. The bandwidth element and its child elements are described in the QoS section of the Network XML.

Setting VLAN tag (on supported network types only)
...
<devices>
  <interface type='bridge'>
    <vlan>
      <tag id='42'/>
    </vlan>
    <source bridge='ovsbr0'/>
    <virtualport type='openvswitch'>
      <parameters interfaceid='09b11c53-8b5c-4eeb-8f00-d84eaa0aaa4f'/>
    </virtualport>
  </interface>
  <interface type='bridge'>
    <vlan trunk='yes'>
      <tag id='42'/>
      <tag id='123' nativeMode='untagged'/>
    </vlan>
    ...
  </interface>
</devices>
...

If (and only if) the network connection used by the guest supports VLAN tagging transparent to the guest, an optional <vlan> element can specify one or more VLAN tags to apply to the guest's network traffic Since 0.10.0 . Network connections that support guest-transparent VLAN tagging include 1) type='bridge' interfaces connected to an Open vSwitch bridge Since 0.10.0 , 2) SRIOV Virtual Functions (VF) used via type='hostdev' (direct device assignment) Since 0.10.0 , and 3) SRIOV VFs used via type='direct' with mode='passthrough' (macvtap "passthru" mode) Since 1.3.5 . All other connection types, including standard linux bridges and libvirt's own virtual networks, do not support it. 802.1Qbh (vn-link) and 802.1Qbg (VEPA) switches provide their own way (outside of libvirt) to tag guest traffic onto a specific VLAN. Each tag is given in a separate <tag> subelement of <vlan> (for example: <tag id='42'/>). For VLAN trunking of multiple tags (which is supported only on Open vSwitch connections), multiple <tag> subelements can be specified, which implies that the user wants to do VLAN trunking on the interface for all the specified tags. In the case that VLAN trunking of a single tag is desired, the optional attribute trunk='yes' can be added to the toplevel <vlan> element to differentiate trunking of a single tag from normal tagging.

For network connections using Open vSwitch it is also possible to configure 'native-tagged' and 'native-untagged' VLAN modes Since 1.1.0. This is done with the optional nativeMode attribute on the <tag> subelement: nativeMode may be set to 'tagged' or 'untagged'. The id attribute of the <tag> subelement containing nativeMode sets which VLAN is considered to be the "native" VLAN for this interface, and the nativeMode attribute determines whether or not traffic for that VLAN will be tagged.

Isolating guests' network traffic from each other
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <port isolated='yes'/>
  </interface>
</devices>
...

Since 6.1.0. The port element property isolated, when set to yes (default setting is no) is used to isolate this interface's network traffic from that of other guest interfaces connected to the same network that also have <port isolated='yes'/>. This setting is only supported for emulated interface devices that use a standard tap device to connect to the network via a Linux host bridge. This property can be inherited from a libvirt network, so if all guests that will be connected to the network should be isolated, it is better to put the setting in the network configuration. (NB: this only prevents guests that have isolated='yes' from communicating with each other; if there is a guest on the same bridge that doesn't have isolated='yes', even the isolated guests will be able to communicate with it.)

...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet0'/>
    <link state='down'/>
  </interface>
</devices>
...

This element provides means of setting state of the virtual network link. Possible values for attribute state are up and down. If down is specified as the value, the interface behaves as if it had the network cable disconnected. Default behavior if this element is unspecified is to have the link state up. Since 0.9.5

MTU configuration
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet0'/>
    <mtu size='1500'/>
  </interface>
</devices>
...

This element provides means of setting MTU of the virtual network link. Currently there is just one attribute size which accepts a non-negative integer which specifies the MTU size for the interface. Since 3.1.0

Coalesce settings
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet0'/>
    <coalesce>
      <rx>
        <frames max='7'/>
      </rx>
    </coalesce>
  </interface>
</devices>
...

This element provides means of setting coalesce settings for some interface devices (currently only type network and bridge. Currently there is just one attribute, max, to tweak, in element frames for the rx group, which accepts a non-negative integer that specifies the maximum number of packets that will be received before an interrupt. Since 3.3.0

IP configuration
...
<devices>
  <interface type='network'>
    <source network='default'/>
    <target dev='vnet0'/>
    <ip address='192.168.122.5' prefix='24'/>
    <ip address='192.168.122.5' prefix='24' peer='10.0.0.10'/>
    <route family='ipv4' address='192.168.122.0' prefix='24' gateway='192.168.122.1'/>
    <route family='ipv4' address='192.168.122.8' gateway='192.168.122.1'/>
  </interface>
  ...
  <hostdev mode='capabilities' type='net'>
    <source>
      <interface>eth0</interface>
    </source>
    <ip address='192.168.122.6' prefix='24'/>
    <route family='ipv4' address='192.168.122.0' prefix='24' gateway='192.168.122.1'/>
    <route family='ipv4' address='192.168.122.8' gateway='192.168.122.1'/>
  </hostdev>
  ...
</devices>
...

Since 1.2.12 network devices and hostdev devices with network capabilities can optionally be provided one or more IP addresses to set on the network device in the guest. Note that some hypervisors or network device types will simply ignore them or only use the first one. The family attribute can be set to either ipv4 or ipv6, and the address attribute contains the IP address. The optional prefix is the number of 1 bits in the netmask, and will be automatically set if not specified - for IPv4 the default prefix is determined according to the network "class" (A, B, or C - see RFC870), and for IPv6 the default prefix is 64. The optional peer attribute holds the IP address of the other end of a point-to-point network device (since 2.1.0) .

Since 1.2.12 route elements can also be added to define IP routes to add in the guest. The attributes of this element are described in the documentation for the route element in network definitions. This is used by the LXC driver.

...
<devices>
  <interface type='ethernet'>
    <source/>
      <ip address='192.168.123.1' prefix='24'/>
      <ip address='10.0.0.10' prefix='24' peer='192.168.122.5'/>
      <route family='ipv4' address='192.168.42.0' prefix='24' gateway='192.168.123.4'/>
    <source/>
    ...
  </interface>
  ...
</devices>
...

Since 2.1.0 network devices of type "ethernet" can optionally be provided one or more IP addresses and one or more routes to set on the host side of the network device. These are configured as subelements of the <source> element of the interface, and have the same attributes as the similarly named elements used to configure the guest side of the interface (described above).

vhost-user interface

Since 1.2.7 the vhost-user enables the communication between a QEMU virtual machine and other userspace process using the Virtio transport protocol. A char dev (e.g. Unix socket) is used for the control plane, while the data plane is based on shared memory.

...
<devices>
  <interface type='vhostuser'>
    <mac address='52:54:00:3b:83:1a'/>
    <source type='unix' path='/tmp/vhost1.sock' mode='server'/>
    <model type='virtio'/>
  </interface>
  <interface type='vhostuser'>
    <mac address='52:54:00:3b:83:1b'/>
    <source type='unix' path='/tmp/vhost2.sock' mode='client'>
      <reconnect enabled='yes' timeout='10'/>
    </source>
    <model type='virtio'/>
    <driver queues='5'/>
  </interface>
</devices>
...

The <source> element has to be specified along with the type of char device. Currently, only type='unix' is supported, where the path (the directory path of the socket) and mode attributes are required. Both mode='server' and mode='client' are supported. vhost-user requires the virtio model type, thus the <model> element is mandatory. Since 4.1.0 the element has an optional child element reconnect which configures reconnect timeout if the connection is lost. It has two attributes enabled (which accepts yes and no) and timeout which specifies the amount of seconds after which hypervisor tries to reconnect.

Traffic filtering with NWFilter

Since 0.8.0 an nwfilter profile can be assigned to a domain interface, which allows configuring traffic filter rules for the virtual machine. See the nwfilter documentation for more complete details.

...
<devices>
  <interface ...>
    ...
    <filterref filter='clean-traffic'/>
  </interface>
  <interface ...>
    ...
    <filterref filter='myfilter'>
      <parameter name='IP' value='104.207.129.11'/>
      <parameter name='IP6_ADDR' value='2001:19f0:300:2102::'/>
      <parameter name='IP6_MASK' value='64'/>
      ...
    </filterref>
  </interface>
</devices>
...

The filter attribute specifies the name of the nwfilter to use. Optional <parameter> elements may be specified for passing additional info to the nwfilter via the name and value attributes. See the nwfilter docs for info on parameters.

Input devices

Input devices allow interaction with the graphical framebuffer in the guest virtual machine. When enabling the framebuffer, an input device is automatically provided. It may be possible to add additional devices explicitly, for example, to provide a graphics tablet for absolute cursor movement.

...
<devices>
  <input type='mouse' bus='usb'/>
  <input type='keyboard' bus='usb'/>
  <input type='mouse' bus='virtio'/>
  <input type='keyboard' bus='virtio'/>
  <input type='tablet' bus='virtio'/>
  <input type='passthrough' bus='virtio'>
    <source evdev='/dev/input/event1'/>
  </input>
  <input type='evdev'>
    <source dev='/dev/input/event1234' grab='all' repeat='on' grabToggle='ctrl-ctrl'/>
  </input>
</devices>
...
input

The input element has one mandatory attribute, the type whose value can be 'mouse', 'tablet', ( since 1.2.2 ) 'keyboard', ( since 1.3.0 ) 'passthrough' or ( since 7.4.0 ) 'evdev'. The tablet provides absolute cursor movement, while the mouse uses relative movement. The optional bus attribute can be used to refine the exact device type. It takes values "xen" (paravirtualized), "ps2" and "usb" or ( since 1.3.0 ) "virtio".

The input element has an optional sub-element <address> which can tie the device to a particular PCI slot, documented in the Device Addresses section. On S390, address can be used to provide a CCW address for an input device ( since 4.2.0 ). For types passthrough and evdev, the mandatory sub-element source must have an evdev (for passthrough) or dev (for evdev) attribute containing the absolute path to the event device passed through to guests. For type evdev, source has three optional attributes grab with value 'all' which when enabled grabs all input devices instead of just one, repeat with value 'on'/'off' to enable/disable auto-repeat events and grabToggle (since 7.6.0) with values ctrl-ctrl, alt-alt, shift-shift, meta-meta, scrolllock or ctrl-scrolllock to change the grab key combination. input type evdev is currently supported only on linux devices. (KVM only) Since 5.2.0 , the input element accepts a model attribute which has the values 'virtio', 'virtio-transitional' and 'virtio-non-transitional'. See Virtio transitional devices for more details.

The subelement driver can be used to tune the virtio options of the device: Virtio-related options can also be set. ( Since 3.5.0 )

Hub devices

A hub is a device that expands a single port into several so that there are more ports available to connect devices to a host system.

...
<devices>
  <hub type='usb'/>
</devices>
...
hub

The hub element has one mandatory attribute, the type whose value can only be 'usb'.

The hub element has an optional sub-element <address> (See Device Addresses) type='usb' which can tie the device to a particular controller.

Graphical framebuffers

A graphics device allows for graphical interaction with the guest OS. A guest will typically have either a framebuffer or a text console configured to allow interaction with the admin.

...
<devices>
  <graphics type='sdl' display=':0.0'/>
  <graphics type='vnc' port='5904' sharePolicy='allow-exclusive'>
    <listen type='address' address='1.2.3.4'/>
  </graphics>
  <graphics type='rdp' autoport='yes' multiUser='yes' />
  <graphics type='desktop' fullscreen='yes'/>
  <graphics type='spice'>
    <listen type='network' network='rednet'/>
  </graphics>
</devices>
...
graphics

The graphics element has a mandatory type attribute which takes the value sdl, vnc, spice, rdp, desktop or egl-headless:

sdl

This displays a window on the host desktop, it can take 3 optional arguments: a display attribute for the display to use, an xauth attribute for the authentication identifier, and an optional fullscreen attribute accepting values yes or no.

You can use a gl with the enable="yes" property to enable OpenGL support in SDL. Likewise you can explicitly disable OpenGL support with enable="no".

vnc

Starts a VNC server. The port attribute specifies the TCP port number (with -1 as legacy syntax indicating that it should be auto-allocated). The autoport attribute is the new preferred syntax for indicating auto-allocation of the TCP port to use. The passwd attribute provides a VNC password in clear text. If the passwd attribute is set to an empty string, then VNC access is disabled. The keymap attribute specifies the keymap to use. It is possible to set a limit on the validity of the password by giving a timestamp passwdValidTo='2010-04-09T15:51:00' assumed to be in UTC. The connected attribute allows control of connected client during password changes. VNC accepts keep value only since 0.9.3 . NB, this may not be supported by all hypervisors.

The optional sharePolicy attribute specifies vnc server display sharing policy. allow-exclusive allows clients to ask for exclusive access by dropping other connections. Connecting multiple clients in parallel requires all clients asking for a shared session (vncviewer: -Shared switch). This is the default value. force-shared disables exclusive client access, every connection has to specify -Shared switch for vncviewer. ignore welcomes every connection unconditionally since 1.0.6 .

Rather than using listen/port, QEMU supports a socket attribute for listening on a unix domain socket path Since 0.8.8 .

For VNC WebSocket functionality, websocket attribute may be used to specify port to listen on (with -1 meaning auto-allocation and autoport having no effect due to security reasons) Since 1.0.6 .

For VNC, the powerControl attribute can be used to enable VM shutdown, reboot and reset power control features for the VNC client. This is appropriate if the authenticated VNC client user already has administrator privileges in the guest Since 7.1.0.

Although VNC doesn't support OpenGL natively, it can be paired with graphics type egl-headless (see below) which will instruct QEMU to open and use drm nodes for OpenGL rendering.

A VNC server could be optionally mapped to the specific host audio backend using the <audio> sub-element:

<graphics type='vnc' ...>
  <audio id='1'>
</graphics>

Where 1 is an id of the audio device (SeeAudio backends). If no ID is specified, then the default audio backend will be used. Since 7.2.0, qemu.

spice Since 0.8.6

Starts a SPICE server. The port attribute specifies the TCP port number (with -1 as legacy syntax indicating that it should be auto-allocated), while tlsPort gives an alternative secure port number. The autoport attribute is the new preferred syntax for indicating auto-allocation of needed port numbers. The passwd attribute provides a SPICE password in clear text. If the passwd attribute is set to an empty string, then SPICE access is disabled. The keymap attribute specifies the keymap to use. It is possible to set a limit on the validity of the password by giving a timestamp passwdValidTo='2010-04-09T15:51:00' assumed to be in UTC.

The connected attribute allows control of connected client during password changes. SPICE accepts keep to keep client connected, disconnect to disconnect client and fail to fail changing password . NB, this may not be supported by all hypervisors. Since 0.9.3

The defaultMode attribute sets the default channel security policy, valid values are secure, insecure and the default any (which is secure if possible, but falls back to insecure rather than erroring out if no secure path is available). Since 0.9.12

When SPICE has both a normal and TLS secured TCP port configured, it can be desirable to restrict what channels can be run on each port. This is achieved by adding one or more <channel> elements inside the main <graphics> element and setting the mode attribute to either secure or insecure. Setting the mode attribute overrides the default value as set by the defaultMode attribute. (Note that specifying any as mode discards the entry as the channel would inherit the default mode anyways.) Valid channel names include main, display, inputs, cursor, playback, record (all since 0.8.6 ); smartcard ( since 0.8.8 ); and usbredir ( since 0.9.12 ).

<graphics type='spice' port='-1' tlsPort='-1' autoport='yes'>
  <channel name='main' mode='secure'/>
  <channel name='record' mode='insecure'/>
  <image compression='auto_glz'/>
  <streaming mode='filter'/>
  <clipboard copypaste='no'/>
  <mouse mode='client'/>
  <filetransfer enable='no'/>
  <gl enable='yes' rendernode='/dev/dri/by-path/pci-0000:00:02.0-render'/>
</graphics>

Spice supports variable compression settings for audio, images and streaming. These settings are accessible via the compression attribute in all following elements: image to set image compression (accepts auto_glz, auto_lz, quic, glz, lz, off), jpeg for JPEG compression for images over wan (accepts auto, never, always), zlib for configuring wan image compression (accepts auto, never, always) and playback for enabling audio stream compression (accepts on or off). Since 0.9.1

Streaming mode is set by the streaming element, settings its mode attribute to one of filter, all or off. Since 0.9.2

Copy & Paste functionality (via Spice agent) is set by the clipboard element. It is enabled by default, and can be disabled by setting the copypaste property to no. Since 0.9.3

Mouse mode is set by the mouse element, setting its mode attribute to one of server or client. If no mode is specified, the qemu default will be used (client mode). Since 0.9.11

File transfer functionality (via Spice agent) is set using the filetransfer element. It is enabled by default, and can be disabled by setting the enable property to no. Since 1.2.2

Spice may provide accelerated server-side rendering with OpenGL. You can enable or disable OpenGL support explicitly with the gl element, by setting the enable property. (QEMU only, since 1.3.3 ). Note that this only works locally, since this requires usage of UNIX sockets, i.e. using listen types 'socket' or 'none'. For accelerated OpenGL with remote support, consider pairing this element with type egl-headless (see below). However, this will deliver weaker performance compared to native Spice OpenGL support.

By default, QEMU will pick the first available GPU DRM render node. You may specify a DRM render node path to use instead. (QEMU only, since 3.1.0 ).

rdp

Starts a RDP server. The port attribute specifies the TCP port number (with -1 as legacy syntax indicating that it should be auto-allocated). The autoport attribute is the new preferred syntax for indicating auto-allocation of the TCP port to use. In the VirtualBox driver, the autoport will make the hypervisor pick available port from 3389-3689 range when the VM is started. The chosen port will be reflected in the port attribute. The multiUser attribute is a boolean deciding whether multiple simultaneous connections to the VM are permitted. The replaceUser attribute is a boolean deciding whether the existing connection must be dropped and a new connection must be established by the VRDP server, when a new client connects in single connection mode.

desktop

This value is reserved for VirtualBox domains for the moment. It displays a window on the host desktop, similarly to "sdl", but using the VirtualBox viewer. Just like "sdl", it accepts the optional attributes display and fullscreen.

egl-headless Since 4.6.0

This display type provides support for an OpenGL accelerated display accessible both locally and remotely (for comparison, Spice's native OpenGL support only works locally using UNIX sockets at the moment, but has better performance). Since this display type doesn't provide any window or graphical console like the other types, for practical reasons it should be paired with either vnc or spice graphics types. This display type is only supported by QEMU domains (needs QEMU 2.10 or newer). 5.0.0 this element accepts a <gl/> sub-element with an optional attribute rendernode which can be used to specify an absolute path to a host's DRI device to be used for OpenGL rendering.

<graphics type='spice' autoport='yes'/>
<graphics type='egl-headless'>
  <gl rendernode='/dev/dri/renderD128'/>
</graphics>
dbusSince 8.4.0

Export the display over D-Bus. By default, it will use a private bus, except when p2p or address are specified.

<graphics type='dbus'/>

p2p (accepts on or off) enables peer-to-peer connections, established through virDomainOpenGraphics() APIs.

address (accepts a D-Bus address), will connect to the specified bus address.

This element accepts a <gl/> sub-element with an optional attribute rendernode which can be used to specify an absolute path to a host's DRI device to be used for OpenGL rendering.

Copy & Paste functionality is provided thanks to the QEMU clipboard manager and the SPICE vdagent protocol. See qemu-vdagent for more details.

D-Bus can export an audio backend using the <audio> sub-element:

<graphics type='dbus' ...>
  <audio id='1'>
</graphics>

Where 1 is an id of the audio device (See Audio backends).

Graphics device uses a <listen> to set up where the device should listen for clients. It has a mandatory attribute type which specifies the listen type. Only vnc, spice and rdp supports <listen> element. Since 0.9.4 . Available types are:

address

Tells a graphics device to use an address specified in the address attribute, which will contain either an IP address or hostname (which will be resolved to an IP address via a DNS query) to listen on.

It is possible to omit the address attribute in order to use an address from config files Since 1.3.5 .

The address attribute is duplicated as listen attribute in graphics element for backward compatibility. If both are provided they must be equal.

network

This is used to specify an existing network in the network attribute from libvirt's list of configured networks. The named network configuration will be examined to determine an appropriate listen address and the address will be stored in live XML in address attribute. For example, if the network has an IPv4 address in its configuration (e.g. if it has a forward type of route, nat, or no forward type (isolated)), the first IPv4 address listed in the network's configuration will be used. If the network is describing a host bridge, the first IPv4 address associated with that bridge device will be used, and if the network is describing one of the 'direct' (macvtap) modes, the first IPv4 address of the first forward dev will be used.

socket since 2.0.0 (QEMU only)

This listen type tells a graphics server to listen on unix socket. Attribute socket contains a path to unix socket. If this attribute is omitted libvirt will generate this path for you. Supported by graphics type vnc and spice.

For vnc graphics be backward compatible the socket attribute of first listen element is duplicated as socket attribute in graphics element. If graphics element contains a socket attribute all listen elements are ignored.

none since 2.0.0 (QEMU only)

This listen type doesn't have any other attribute. Libvirt supports passing a file descriptor through our APIs virDomainOpenGraphics() and virDomainOpenGraphicsFD(). No other listen types are allowed if this one is used and the graphics device doesn't listen anywhere. You need to use one of the two APIs to pass a FD to QEMU in order to connect to this graphics device. Supported by graphics type vnc and spice.

Video devices

A video device.

...
<devices>
  <video>
    <model type='vga' vram='16384' heads='1'>
      <acceleration accel3d='yes' accel2d='yes'/>
    </model>
    <driver name='qemu'/>
  </video>
</devices>
...
video

The video element is the container for describing video devices. For backwards compatibility, if no video is set but there is a graphics in domain xml, then libvirt will add a default video according to the guest type.

For a guest of type "kvm", the default video is: type with value "cirrus", vram with value "16384" and heads with value "1". By default, the first video device in domain xml is the primary one, but the optional attribute primary ( since 1.0.2 ) with value 'yes' can be used to mark the primary in cases of multiple video device. The non-primary must be type of "qxl" or ( since 2.4.0 ) "virtio".

model

The model element has a mandatory type attribute which takes the value "vga", "cirrus", "vmvga", "xen", "vbox", "qxl" ( since 0.8.6 ), "virtio" ( since 1.3.0 ), "gop" ( since 3.2.0 ), "bochs" ( since 5.6.0 ), "ramfb" ( since 5.9.0 ), or "none" ( since 4.6.0 ), depending on the hypervisor features available.

Note: The purpose of the type none is to instruct libvirt not to add a default video device in the guest (see the video element description above), since such behaviour is inconvenient in cases where GPU mediated devices are meant to be the only rendering device within a guest. If this is your use case specify a none type video device in the XML to stop the default behaviour. Refer to Host device assignment to see how to add a mediated device into a guest.

You can provide the amount of video memory in kibibytes (blocks of 1024 bytes) using vram. This is supported only for guest type of "vz", "qemu", "kvm", "hvf", "vbox", "vmx" and "xen". If no value is provided the default is used. If the size is not a power of two it will be rounded to closest one.

The number of screen can be set using heads. This is supported only for guests type of "vz", "kvm", "hvf", "vbox" and "vmx".

For guest type of "kvm", "hvf" or "qemu" and model type "qxl" there are optional attributes. Attribute ram ( since 1.0.2 ) specifies the size of the primary bar, while the attribute vram specifies the secondary bar size. If ram or vram are not supplied a default value is used. The ram should also be rounded to power of two as vram. There is also optional attribute vgamem ( since 1.2.11 ) to set the size of VGA framebuffer for fallback mode of QXL device. Attribute vram64 ( since 1.3.3 ) extends secondary bar and makes it addressable as 64bit memory.

Since 9.2.0 (QEMU driver only), devices with type "virtio" have an optional blob attribute that can be set to "on" or "off". Setting blob to "on" will enable the use of blob resources in the device. This can accelerate the display path by reducing or eliminating copying of pixel data between the guest and host. Note that blob resource support requires QEMU version 6.1 or newer.

Since 5.9.0 , the model element may also have an optional resolution sub-element. The resolution element has attributes x and y to set the minimum resolution for the video device. This sub-element is valid for model types "vga", "qxl", "bochs", "gop", and "virtio".

acceleration

Configure if video acceleration should be enabled.

accel2d

Enable 2D acceleration (for vbox driver only, since 0.7.1 )

accel3d

Enable 3D acceleration (for vbox driver since 0.7.1 , qemu driver since 1.3.0 )

rendernode

Absolute path to a host's DRI device to be used for rendering (for 'vhostuser' driver only, since 5.8.0 ). If none is specified, libvirt will pick one available.

address

The optional address sub-element can be used to tie the video device to a particular PCI slot. On S390, address can be used to provide the CCW address for the video device ( since 4.2.0 ).

driver

The subelement driver can be used to tune the device:

name

Specify the backend driver to use, either "qemu" or "vhostuser" depending on the hypervisor features available ( since 5.8.0 ). "qemu" is the default QEMU backend. "vhostuser" will use a separate vhost-user process backend (for virtio device).

virtio options

Virtio-related options can also be set (Since 3.5.0)

VGA configuration

Control how the video devices exposed to the guest using the vgaconf attribute which takes the value "io", "on" or "off". At present, it's only applicable to the bhyve's "gop" video model type ( Since 3.5.0 )

Consoles, serial, parallel & channel devices

A character device provides a way to interact with the virtual machine. Paravirtualized consoles, serial ports, parallel ports and channels are all classed as character devices and so represented using the same syntax.

...
<devices>
  <parallel type='pty'>
    <source path='/dev/pts/2'/>
    <target port='0'/>
  </parallel>
  <serial type='pty'>
    <source path='/dev/pts/3'/>
    <target port='0'/>
  </serial>
  <serial type='file'>
    <source path='/tmp/file' append='on'>
      <seclabel model='dac' relabel='no'/>
    </source>
    <target port='0'/>
  </serial>
  <console type='pty'>
    <source path='/dev/pts/4'/>
    <target port='0'/>
  </console>
  <channel type='unix'>
    <source mode='bind' path='/tmp/guestfwd'/>
    <target type='guestfwd' address='10.0.2.1' port='4600'/>
  </channel>
</devices>
...

In each of these directives, the top-level element name (parallel, serial, console, channel) describes how the device is presented to the guest. The guest interface is configured by the target element.

The interface presented to the host is given in the type attribute of the top-level element. The host interface is configured by the source element.

The source element may contain an optional seclabel to override the way that labelling is done on the socket path. If this element is not present, the Security label is inherited from the per-domain setting.

If the interface type presented to the host is "file", then the source element may contain an optional attribute append that specifies whether or not the information in the file should be preserved on domain restart. Allowed values are "on" and "off" (default). Since 1.3.1 .

Regardless of the type, character devices can have an optional log file associated with them. This is expressed via a log sub-element, with a file attribute. There can also be an append attribute which takes the same values described above. Since 1.3.3 .

...
<log file="/var/log/libvirt/qemu/guestname-serial0.log" append="off"/>
...

Each character device element has an optional sub-element <address> which can tie the device to a particular controller (See Controllers) or PCI slot.

For character device with type unix or tcp the source has an optional element reconnect which configures reconnect timeout if the connection is lost. There are two attributes, enabled where possible values are "yes" and "no" and timeout which is in seconds. The reconnect attribute is valid only for connect mode. Since 3.7.0 (QEMU driver only) .

Guest interface

A character device presents itself to the guest as one of the following types.

Parallel port
...
<devices>
  <parallel type='pty'>
    <source path='/dev/pts/2'/>
    <target port='0'/>
  </parallel>
</devices>
...

target can have a port attribute, which specifies the port number. Ports are numbered starting from 0. There are usually 0, 1 or 2 parallel ports.

Serial port
...
<devices>
  <!-- Serial port -->
  <serial type='pty'>
    <source path='/dev/pts/3'/>
    <target port='0'/>
  </serial>
  <!-- Debug port for SeaBIOS / EDK II -->
  <serial type='file'>
    <target type='isa-debug'/>
    <address type='isa' iobase='0x402'/>
    <source path='/tmp/DOMAIN-ovmf.log'/>
  </serial>

</devices>
...
...
<devices>
  <!-- USB serial port -->
  <serial type='pty'>
    <target type='usb-serial' port='0'>
      <model name='usb-serial'/>
    </target>
    <address type='usb' bus='0' port='1'/>
  </serial>
</devices>
...

The target element can have an optional port attribute, which specifies the port number (starting from 0), and an optional type attribute: valid values are, since 1.0.2 , isa-serial (usable with x86 guests), usb-serial (usable whenever USB support is available) and pci-serial (usable whenever PCI support is available); since 3.10.0 , spapr-vio-serial (usable with ppc64/pseries guests), system-serial (usable with aarch64/virt and, since 4.7.0 , riscv/virt guests), sclp-serial (usable with s390 and s390x guests) are available as well and since 8.1.0 isa-debug (usable with x86 guests).

Since 3.10.0 , the target element can have an optional model subelement; valid values for its name attribute are: isa-serial (usable with the isa-serial target type); usb-serial (usable with the usb-serial target type); pci-serial (usable with the pci-serial target type); spapr-vty (usable with the spapr-vio-serial target type); pl011 and, since 4.7.0 , 16550a (usable with the system-serial target type); sclpconsole and sclplmconsole (usable with the sclp-serial target type). Since: 8.1.0, isa-debugcon (usable with the isa-debug target type); provides a virtual console for receiving debug messages from the firmware on x86 platforms. Providing a target model is usually unnecessary: libvirt will automatically pick one that's suitable for the chosen target type, and overriding that value is generally not recommended.

If any of the attributes is not specified by the user, libvirt will choose a value suitable for most users.

Most target types support configuring the guest-visible device address as documented in the Device Addresses section; more specifically, acceptable address types are isa (for isa-serial), usb (for usb-serial), pci (for pci-serial) and spapr-vio (for spapr-vio-serial). The system-serial and sclp-serial target types don't support specifying an address.

For the relationship between serial ports and consoles, refer to the Relationship between serial ports and consoles section.

Console
...
<devices>
  <!-- Serial console -->
  <console type='pty'>
    <source path='/dev/pts/2'/>
   <target type='serial' port='0'/>
  </console>
</devices>
...
...
<devices>
  <!-- KVM virtio console -->
  <console type='pty'>
    <source path='/dev/pts/5'/>
    <target type='virtio' port='0'/>
  </console>
</devices>
...

The console element is used to represent interactive serial consoles. Depending on the type of guest in use and the specifics of the configuration, the console element might represent the same device as an existing serial element or a separate device.

A target subelement is supported and works the same way as with the serial element (See Serial port for details). Valid values for the type attribute are: serial (described below); virtio (usable whenever VirtIO support is available); xen, lxc and openvz (available when the corresponding hypervisor is in use). sclp and sclplm (usable for s390 and s390x QEMU guests) are supported for compatibility reasons but should not be used for new guests: use the sclpconsole and sclplmconsole target models, respectively, with the serial element instead.

Of the target types listed above, serial is special in that it doesn't represents a separate device, but rather the same device as the first serial element. Due to this, there can only be a single console element with target type serial per guest.

Virtio consoles are usually accessible as /dev/hvc[0-7] from inside the guest; for more information, see https://fedoraproject.org/wiki/Features/VirtioSerial. Since 0.8.3

For the relationship between serial ports and consoles, refer to the Relationship between serial ports and consoles section.

Relationship between serial ports and consoles

Due to historical reasons, the serial and console elements have partially overlapping scopes.

In general, both elements are used to configure one or more serial consoles to be used for interacting with the guest. The main difference between the two is that serial is used for emulated, usually native, serial consoles, whereas console is used for paravirtualized ones.

Both emulated and paravirtualized serial consoles have advantages and disadvantages:

  • emulated serial consoles are usually initialized much earlier than paravirtualized ones, so they can be used to control the bootloader and display both firmware and early boot messages;
  • on several platforms, there can only be a single emulated serial console per guest but paravirtualized consoles don't suffer from the same limitation.

A configuration such as:

...
<devices>
  <console type='pty'>
    <target type='serial'/>
  </console>
  <console type='pty'>
    <target type='virtio'/>
  </console>
</devices>
...

will work on any platform and will result in one emulated serial console for early boot logging / interactive / recovery use, and one paravirtualized serial console to be used eg. as a side channel. Most people will be fine with having just the first console element in their configuration, but if a specific configuration is desired then both elements should be specified.

Note that, due to the compatibility concerns mentioned earlier, all the following configurations:

...
<devices>
  <serial type='pty'/>
</devices>
...
...
<devices>
  <console type='pty'/>
</devices>
...
...
<devices>
  <serial type='pty'/>
  <console type='pty'/>
</devices>
...

will be treated the same and will result in a single emulated serial console being available to the guest.

Channel

This represents a private communication channel between the host and the guest.

...
<devices>
  <channel type='unix'>
    <source mode='bind' path='/tmp/guestfwd'/>
    <target type='guestfwd' address='10.0.2.1' port='4600'/>
  </channel>

  <!-- KVM virtio channel -->
  <channel type='pty'>
    <target type='virtio' name='arbitrary.virtio.serial.port.name'/>
  </channel>
  <channel type='unix'>
    <source mode='bind' path='/var/lib/libvirt/qemu/f16x86_64.agent'/>
    <target type='virtio' name='org.qemu.guest_agent.0' state='connected'/>
  </channel>
  <channel type='spicevmc'>
    <target type='virtio' name='com.redhat.spice.0'/>
  </channel>
</devices>
...

This can be implemented in a variety of ways. The specific type of channel is given in the type attribute of the target element. Different channel types have different target attributes.

guestfwd

TCP traffic sent by the guest to a given IP address and port is forwarded to the channel device on the host. The target element must have address and port attributes. Since 0.7.3

virtio

Paravirtualized virtio channel. Channel is exposed in the guest under /dev/vport*, and if the optional element name is specified, /dev/virtio-ports/$name (for more info, please see https://fedoraproject.org/wiki/Features/VirtioSerial). The optional element address can tie the channel to a particular type='virtio-serial' controller, as documented in the Device Addresses section. With qemu, if name is "org.qemu.guest_agent.0", then libvirt can interact with a guest agent installed in the guest, for actions such as guest shutdown or file system quiescing. Since 0.7.7, guest agent interaction since 0.9.10 Moreover, since 1.0.6 it is possible to have source path auto generated for virtio unix channels. This is very useful in case of a qemu guest agent, where users don't usually care about the source path since it's libvirt who talks to the guest agent. In case users want to utilize this feature, they should leave <source> element out. Since 1.2.11 the active XML for a virtio channel may contain an optional state attribute that reflects whether a process in the guest is active on the channel. This is an output-only attribute. Possible values for the state attribute are connected and disconnected.

xen

Paravirtualized Xen channel. Channel is exposed in the guest as a Xen console but identified with a name. Setup and consumption of a Xen channel depends on software and configuration in the guest. See the xen-pv-channel(7) man page for more information. Channel source path semantics are the same as the virtio target type. The state attribute is not supported since Xen channels lack the necessary probing mechanism. Since 2.3.0

spicevmc

Paravirtualized SPICE channel. The domain must also have a SPICE server as a graphics device (See Graphical framebuffers), at which point the host piggy-backs messages across the main channel. The target element must be present, with attribute type='virtio'; an optional attribute name controls how the guest will have access to the channel, and defaults to name='com.redhat.spice.0'. The optional address element can tie the channel to a particular type='virtio-serial' controller. Since 0.8.8

qemu-vdagent

Paravirtualized qemu vdagent channel. This channel implements the SPICE vdagent protocol, but is handled internally by qemu and therefore does not require a SPICE graphics device. Like the spicevmc channel, the target element must be present, with attribute type='virtio'; an optional attribute name controls how the guest will have access to the channel, and defaults to name='com.redhat.spice.0'. The optional address element can tie the channel to a particular type='virtio-serial' controller. Certain vdagent protocol features can by enabled or disabled using the source element.

Copy & Paste functionality is set by the clipboard element. It is disabled by default, and can be enabled by setting the copypaste property to yes. This allows the guest's clipboard to be synchronized with the qemu clipboard manager. This can enable copy and paste between a guest and a client when using a VNC graphics device (See Graphical framebuffers) (when using a VNC client that supports the copy/paste feature) or other graphics types that support the qemu clipboard manager.

Mouse mode is set by the mouse element, setting its mode attribute to one of server or client. If no mode is specified, the qemu default will be used (client mode). Since 8.4.0

Host interface

A character device presents itself to the host as one of the following types.

Domain logfile

This disables all input on the character device, and sends output into the virtual machine's logfile

...
<devices>
  <console type='stdio'>
    <target port='1'/>
  </console>
</devices>
...
Device logfile

A file is opened and all data sent to the character device is written to the file.

...
<devices>
  <serial type="file">
    <source path="/var/log/vm/vm-serial.log"/>
    <target port="1"/>
  </serial>
</devices>
...
Virtual console

Connects the character device to the graphical framebuffer in a virtual console. This is typically accessed via a special hotkey sequence such as "ctrl+alt+3"

...
<devices>
  <serial type='vc'>
    <target port="1"/>
  </serial>
</devices>
...
Null device

Connects the character device to the void. No data is ever provided to the input. All data written is discarded.

...
<devices>
  <serial type='null'>
    <target port="1"/>
  </serial>
</devices>
...
Pseudo TTY

A Pseudo TTY is allocated using /dev/ptmx. A suitable client such as 'virsh console' can connect to interact with the serial port locally.

...
<devices>
  <serial type="pty">
    <source path="/dev/pts/3"/>
    <target port="1"/>
  </serial>
</devices>
...

NB special case if <console type='pty'>, then the TTY path is also duplicated as an attribute tty='/dev/pts/3' on the top level <console> tag. This provides compat with existing syntax for <console> tags.

Host device proxy

The character device is passed through to the underlying physical character device. The device types must match, eg the emulated serial port should only be connected to a host serial port - don't connect a serial port to a parallel port.

...
<devices>
  <serial type="dev">
    <source path="/dev/ttyS0"/>
    <target port="1"/>
  </serial>
</devices>
...
Named pipe

The character device writes output to a named pipe. See pipe(7) for more info.

...
<devices>
  <serial type="pipe">
    <source path="/tmp/mypipe"/>
    <target port="1"/>
  </serial>
</devices>
...
TCP client/server

The character device acts as a TCP client connecting to a remote server.

...
<devices>
  <serial type="tcp">
    <source mode="connect" host="0.0.0.0" service="2445"/>
    <protocol type="raw"/>
    <target port="1"/>
  </serial>
</devices>
 ...

Or as a TCP server waiting for a client connection.

...
<devices>
  <serial type="tcp">
    <source mode="bind" host="127.0.0.1" service="2445"/>
    <protocol type="raw"/>
    <target port="1"/>
  </serial>
</devices>
...

Alternatively you can use telnet instead of raw TCP in order to utilize the telnet protocol for the connection.

Since 0.8.5, some hypervisors support use of either telnets (secure telnet) or tls (via secure sockets layer) as the transport protocol for connections.

...
<devices>
  <serial type="tcp">
    <source mode="connect" host="0.0.0.0" service="2445"/>
    <protocol type="telnet"/>
    <target port="1"/>
  </serial>
  ...
  <serial type="tcp">
    <source mode="bind" host="127.0.0.1" service="2445"/>
    <protocol type="telnet"/>
    <target port="1"/>
  </serial>
</devices>
...

Since 2.4.0, the optional attribute tls can be used to control whether a chardev TCP communication channel would utilize a hypervisor configured TLS X.509 certificate environment in order to encrypt the data channel. For the QEMU hypervisor, usage of a TLS environment can be controlled on the host by the chardev_tls and chardev_tls_x509_cert_dir or default_tls_x509_cert_dir settings in the file /etc/libvirt/qemu.conf. If chardev_tls is enabled, then unless the tls attribute is set to "no", libvirt will use the host configured TLS environment. If chardev_tls is disabled, but the tls attribute is set to "yes", then libvirt will attempt to use the host TLS environment if either the chardev_tls_x509_cert_dir or default_tls_x509_cert_dir TLS directory structure exists.

...
<devices>
  <serial type="tcp">
    <source mode='connect' host="127.0.0.1" service="5555" tls="yes"/>
    <protocol type="raw"/>
    <target port="0"/>
  </serial>
</devices>
...
UDP network console

The character device acts as a UDP netconsole service, sending and receiving packets. This is a lossy service.

...
<devices>
  <serial type="udp">
    <source mode="bind" host="0.0.0.0" service="2445"/>
    <source mode="connect" host="0.0.0.0" service="2445"/>
    <target port="1"/>
  </serial>
</devices>
...
UNIX domain socket client/server

The character device acts as a UNIX domain socket server, accepting connections from local clients.

...
<devices>
  <serial type="unix">
    <source mode="bind" path="/tmp/foo"/>
    <target port="1"/>
  </serial>
</devices>
...
Spice channel

The character device is accessible through spice connection under a channel name specified in the channel attribute. Since 1.2.2

Note: depending on the hypervisor, spiceports might (or might not) be enabled on domains with or without spice graphics (See Graphical framebuffers).

...
<devices>
  <serial type="spiceport">
    <source channel="org.qemu.console.serial.0"/>
    <target port="1"/>
  </serial>
</devices>
...
Nmdm device

The nmdm device driver, available on FreeBSD, provides two tty devices connected together by a virtual null modem cable. Since 1.2.4

...
<devices>
  <serial type="nmdm">
    <source master="/dev/nmdm0A" slave="/dev/nmdm0B"/>
  </serial>
</devices>
...

The source element has these attributes:

master

Master device of the pair, that is passed to the hypervisor. Device is specified by a fully qualified path.

slave

Slave device of the pair, that is passed to the clients for connection to the guest console. Device is specified by a fully qualified path.

Sound devices

A virtual sound card can be attached to the host via the sound element. Since 0.4.3

...
<devices>
  <sound model='es1370'/>
</devices>
...
sound

The sound element has one mandatory attribute, model, which specifies what real sound device is emulated. Valid values are specific to the underlying hypervisor, though typical choices are sb16, es1370, pcspk, ac97 (Since 0.6.0), ich6 (Since 0.8.8), ich9 (Since 1.1.3), usb (Since 1.2.8) and ich7 (Since 6.7.0, bhyve only).

Since 0.9.13 , a sound element with ich6 or ich9 models can have optional sub-elements <codec> to attach various audio codecs to the audio device. If not specified, a default codec will be attached to allow playback and recording.

Valid values are:

  • duplex - advertise a line-in and a line-out
  • micro - advertise a speaker and a microphone
  • output - advertise a line-out Since 4.4.0
...
<devices>
  <sound model='ich6'>
    <codec type='micro'/>
  </sound>
</devices>
...

Since 9.4.0 the usb sound device can be optionally switched into multi-channel mode by using the multichannel attribute:

<sound model='usb' multichannel='yes'/>

Each sound element has an optional sub-element <address> which can tie the device to a particular PCI slot. See Device Addresses.

A sound device could be optionally mapped to the specific host audio backend using the <audio> sub-element:

...
<devices>
  <sound model='ich7'>
     <audio id='1'>
  </sound>
</devices>
...

Where 1 is an id of the audio device (See Audio backends). If no ID is specified, then the default audio backend will be used. Since 6.7.0, bhyve; Since 7.2.0, qemu.

Audio backends

A virtual audio device corresponds to a host audio backend that is mapped to the guest sound device.

type

The required type attribute specifies audio backend type. Currently, the supported values are none, alsa, coreaudio, dbus, jack, oss, pulseaudio, sdl, spice, file.

id

Integer id of the audio device. Must be greater than 0.

timerPeriod

Timer period in microseconds. Must be greater than 0. If omitted, the lowest possible period is used.

All the backends support child element for configuring input and output properties

...
<devices>
  <audio id='1' type='pulseaudio' timerPeriod='40'>
    <input mixingEngine='yes' fixedSettings='yes' voices='1' bufferLength='100'>
      <settings frequency='44100' channels='2' format='s16'/>
    </input>
    <output mixingEngine='yes' fixedSettings='yes' voices='2' bufferLength='100'>
      <settings frequency='22050' channels='4' format='f32'/>
    </output>
  </audio>
</devices>
...

The input and output elements support the same set of attributes and elements

  • mixingEngine

    Control whether the host mixing engine is used to convert between different audio formats and sampling rates. When the mixing engine is disabled it is possible to make use of improved audio formats such as 5.1/7.1. If not specified, a hypervisor default applies.

  • fixedSettings

    Control whether the mixing engine can dynamically choose settings to minimize format conversion. This is only valid when the mixing engine is explicitly enabled.

  • voices

    The number of voices voices to use, usually defaults to 1

  • bufferLength

    The length of the audio buffer in microseconds. Default is backend specific.

The <input> and <output> elements may also permit backend specific options.

When fixed settings are enabled, the <settings> child element is permitted with the following attributes.

  • frequency

    The frequency in HZ, usually defaulting to 44100

  • channels

    The number of channels, usually defaulting to 2. The permitted max number of channels is hypervisor specific.

  • format

    The audio format, one of s8, u8, s16, u16, s32, u32, f32. The default is hypervisor specific.

Note: If no <audio/> element is defined, and the graphics element is set to either vnc or sdl, the libvirtd or virtqemud process will honor the following environment variables:

  • SDL_AUDIODRIVER

    Valid values are pulseaudio, esd, alsa or arts.

  • QEMU_AUDIO_DRV

    Valid values are pa, none, alsa, coreaudio, jack, oss, sdl, spice or wav.

None audio backend

The none audio backend is a dummy backend that does not connect to any host audio framework. It still allows a remote desktop server like VNC to send and receive audio though. This is the default backend when VNC graphics are enabled in QEMU.

Since 7.2.0, qemu

ALSA audio backend

The alsa audio type uses the ALSA host audio device framework.

The following additional attributes are permitted on the <input> and <output> elements

  • dev

    Path to the host device node to connect the backend to. A hypervisor specific default applies if not specified.

<audio id="1" type="alsa">
  <input dev="/dev/dsp0"/>
  <output dev="/dev/dsp1"/>
</audio>

Since 7.2.0, qemu

Coreaudio audio backend

The coreaudio audio backend delegates to a CoreAudio host audio framework for input and output on macOS.

The following additional attributes are permitted on the <input> and <output> elements

  • bufferCount

    The number of buffers. It is recommended to set the bufferLength attribute at the same time.

<audio id="1" type="coreaudio">
  <input bufferCount="50"/>
  <output bufferCount="42"/>
</audio>

Since 7.2.0, qemu

D-Bus audio backend

The dbus audio backend does not connect to any host audio framework. It exports a D-Bus interface when associated with a D-Bus display.

Since 8.4.0, qemu

Jack audio backend

The jack audio backend delegates to a Jack daemon for audio input and output.

The following additional attributes are permitted on the <input> and <output> elements

  • serverName

    Select the Jack server instance to connect to.

  • clientName

    The client name to identify as. The server may modify this to ensure uniqueness unless exactName is enabled

  • connectPorts

    A regular expression of Jack client port names to monitor and connect to.

  • exactName

    Use the exact clientName requested

<audio id="1" type="jack">
  <input serverName="fish" clientName="food" connectPorts="system:capture_[13]" exactName="yes"/>
  <output serverName="fish" clientName="food" connectPorts="system:playback_[13]" exactName="yes"/>
</audio>

Since 7.2.0, qemu

OSS audio backend

The oss audio type uses the OSS host audio device framework.

The following additional attributes are permitted on the <audio> element

  • tryMMap

    Attempt to use mmap for data transfer

  • exclusive

    Enforce exclusive access to the host device

  • dspPolicy

    Set the timing policy of the device, values between -1 and 10. Smaller numbers result in lower latency but higher CPU usage. A negative value requests use of fragment mode.

The following additional attributes are permitted on the <input> and <output> elements

  • dev

    Path to the host device node to connect the backend to. A hypervisor specific default applies if not specified.

  • bufferCount

    The number of buffers. It is recommended to set the bufferLength attribute at the same time.

  • tryPoll

    Attempt to use polling mode

<audio type='oss' id='1' tryMMap='yes' exclusive='yes' dspPolicy='4'>
  <input dev='/dev/dsp0' bufferCount='40' tryPoll='yes'/>
  <output dev='/dev/dsp0' bufferCount='40' tryPoll='yes'/>
</audio>

Since 6.7.0, bhyve; Since 7.2.0, qemu

PulseAudio audio backend

The pulseaudio audio backend delegates to a PulseAudio daemon audio input and output.

The following additional attributes are permitted on the <audio> element

  • serverName

    Hostname of the PulseAudio server

The following additional attributes are permitted on the <input> and <output> elements

  • name

    The sink/source name to use

  • streamName

    The name to identify the stream associated with the VM

  • latency

    Desired latency for the server to target in microseconds

<audio id="1" type="pulseaudio" serverName="acme.example.org">
  <input name="fish" streamName="food" latency="100"/>
  <output name="fish" streamName="food" latency="200"/>
</audio>

Since 7.2.0, qemu

SDL audio backend

The sdl audio backend delegates to the SDL library for audio input and output.

The following additional attributes are permitted on the <audio> element

  • driver

    SDL audio driver. The name attribute specifies SDL driver name, one of esd, alsa, arts, pulseaudio.

The following additional attributes are permitted on the <input> and <output> elements

  • bufferCount

    The number of buffers. It is recommended to set the bufferLength attribute at the same time.

<audio type='sdl' id='1' driver='pulseaudio'>
  <input bufferCount='40'/>
  <output bufferCount='40'/>
</audio>

Since 7.2.0, qemu

Spice audio backend

The spice audio backend is similar to the none backend in that it does not connect to any host audio framework. It exclusively allows a SPICE server to send and receive audio. This is the default backend when SPICE graphics are enabled in QEMU.

<audio type='spice' id='1'/>

Since 7.2.0, qemu

File audio backend

The file audio backend is an output only driver which records audio to a file. The file format is implementation defined, and defaults to WAV with QEMU.

<audio id="1" type="file" path="audio.wav"/>

Since 7.2.0, qemu

Watchdog devices

A virtual hardware watchdog device can be added to the guest via the watchdog element. Since 0.7.3, QEMU and KVM only

The watchdog device requires an additional driver and management daemon in the guest. Just enabling the watchdog in the libvirt configuration does not do anything useful on its own.

Currently libvirt does not support notification when the watchdog fires. This feature is planned for a future version of libvirt.

Having multiple watchdogs is usually not something very common, but be aware that this might happen, for example, when an implicit watchdog device is added as part of another device. For example the iTCO watchdog being part of the ich9 southbridge, which is used with the q35 machine type. Since 9.1.0

...
<devices>
  <watchdog model='i6300esb'/>
</devices>
...
...
<devices>
  <watchdog model='i6300esb' action='poweroff'/>
</devices>
</domain>
model

The required model attribute specifies what real watchdog device is emulated. Valid values are specific to the underlying hypervisor.

QEMU and KVM support:

  • 'itco' - included by default with q35 machine type Since 9.1.0
  • 'i6300esb' - the recommended device, emulating a PCI Intel 6300ESB
  • 'ib700' - emulating an ISA iBase IB700
  • 'diag288' - emulating an S390 DIAG288 device Since 1.2.17
action

The optional action attribute describes what action to take when the watchdog expires. Valid values are specific to the underlying hypervisor.

QEMU and KVM support:

  • 'reset' - default, forcefully reset the guest
  • 'shutdown' - gracefully shutdown the guest (not recommended)
  • 'poweroff' - forcefully power off the guest
  • 'pause' - pause the guest
  • 'none' - do nothing
  • 'dump' - automatically dump the guest Since 0.8.7
  • 'inject-nmi' - inject a non-maskable interrupt into the guest Since 1.2.17

Note 1: the 'shutdown' action requires that the guest is responsive to ACPI signals. In the sort of situations where the watchdog has expired, guests are usually unable to respond to ACPI signals. Therefore using 'shutdown' is not recommended.

Note 2: the directory to save dump files can be configured by auto_dump_path in file /etc/libvirt/qemu.conf.

Memory balloon device

A virtual memory balloon device is added to all Xen and KVM/QEMU guests. It will be seen as memballoon element. It will be automatically added when appropriate, so there is no need to explicitly add this element in the guest XML unless a specific PCI slot needs to be assigned. Since 0.8.3, Xen, QEMU and KVM only Additionally, since 0.8.4 , if the memballoon device needs to be explicitly disabled, model='none' may be used.

Example: automatically added device with KVM

...
<devices>
  <memballoon model='virtio'/>
</devices>
...

Example: manually added device with static PCI slot 2 requested

...
<devices>
  <memballoon model='virtio'>
    <address type='pci' domain='0x0000' bus='0x00' slot='0x02' function='0x0'/>
    <stats period='10'/>
    <driver iommu='on' ats='on'/>
  </memballoon>
</devices>
</domain>
model

The required model attribute specifies what type of balloon device is provided. Valid values are specific to the virtualization platform

  • 'virtio' - default with QEMU/KVM
  • 'virtio-transitional' Since 5.2.0
  • 'virtio-non-transitional' Since 5.2.0
  • 'xen' - default with Xen

See Virtio transitional devices for more details.

autodeflate

The optional autodeflate attribute allows to enable/disable (values "on"/"off", respectively) the ability of the QEMU virtio memory balloon to release some memory at the last moment before a guest's process get killed by Out of Memory killer. Since 1.3.1, QEMU and KVM only

freePageReporting

The optional freePageReporting attribute allows to enable/disable ("on"/"off", respectively) the ability of the QEMU virtio memory balloon to return unused pages back to the hypervisor to be used by other guests or processes. Please note that despite its name it has no effect on free memory as reported by virDomainMemoryStats() and/or virsh dommemstat. Since 6.9.0, QEMU and KVM only

period

The optional period allows the QEMU virtio memory balloon driver to provide statistics through the virsh dommemstat [domain] command. By default, collection is not enabled. In order to enable, use the virsh dommemstat [domain] --period [number] command or virsh edit command to add the option to the XML definition. The virsh dommemstat will accept the options --live, --current, or --config. If an option is not provided, the change for a running domain will only be made to the active guest. If the QEMU driver is not at the right revision, the attempt to set the period will fail. Large values (e.g. many years) might be ignored. Since 1.1.1, requires QEMU 1.5

driver

For model virtio memballoon, Virtio-related options can also be set. ( Since 3.5.0 )

Random number generator device

The virtual random number generator device allows the host to pass through entropy to guest operating systems. Since 1.0.3

Example: usage of the RNG device:

...
<devices>
  <rng model='virtio'>
    <rate period="2000" bytes="1234"/>
    <backend model='random'>/dev/random</backend>
    <!-- OR -->
    <backend model='egd' type='udp'>
      <source mode='bind' service='1234'/>
      <source mode='connect' host='1.2.3.4' service='1234'/>
    </backend>
    <!-- OR -->
    <backend model='builtin'/>
  </rng>
</devices>
...
model

The required model attribute specifies what type of RNG device is provided. Valid values are specific to the virtualization platform:

  • 'virtio' - supported by qemu and virtio-rng kernel module
  • 'virtio-transitional' Since 5.2.0
  • 'virtio-non-transitional' Since 5.2.0

See Virtio transitional devices for more details.

rate

The optional rate element allows limiting the rate at which entropy can be consumed from the source. The mandatory attribute bytes specifies how many bytes are permitted to be consumed per period. An optional period attribute specifies the duration of a period in milliseconds; if omitted, the period is taken as 1000 milliseconds (1 second). Since 1.0.4

backend

The backend element specifies the source of entropy to be used for the domain. The source model is configured using the model attribute. Supported source models are:

random

This backend type expects a non-blocking character device as input. The file name is specified as contents of the backend element. Since 1.3.4 any path is accepted. Before that /dev/random and /dev/hwrng were the only accepted paths. When no file name is specified, the hypervisor default is used. For QEMU, the default is /dev/random. However, the recommended source of entropy is /dev/urandom (as it doesn't have the limitations of /dev/random).

egd

This backend connects to a source using the EGD protocol. The source is specified as a character device. Refer to the Host interface for more information.

builtin

This backend uses qemu builtin random generator, which uses getrandom() syscall as the source of entropy. ( Since 6.1.0 and QEMU 4.2 )

driver

The subelement driver can be used to tune the device:

virtio options

Virtio-related options can also be set. ( Since 3.5.0 )

TPM device

The TPM device enables a QEMU guest to have access to TPM functionality. The TPM device may either be a TPM 1.2 or a TPM 2.0.

The TPM passthrough device type provides access to the host's TPM for one QEMU guest. No other software may be using the TPM device, typically /dev/tpm0, at the time the QEMU guest is started. 'passthrough' since 1.0.5

Example: usage of the TPM passthrough device

...
<devices>
  <tpm model='tpm-tis'>
    <backend type='passthrough'>
      <device path='/dev/tpm0'/>
    </backend>
  </tpm>
</devices>
...

The emulator device type gives access to a TPM emulator providing TPM functionality for each VM. QEMU talks to it over a Unix socket. With the emulator device type each guest gets its own private TPM. 'emulator' since 4.5.0 The state of the TPM emulator can be encrypted by providing an encryption element. 'encryption' since 5.6.0

Example: usage of the TPM Emulator

...
<devices>
  <tpm model='tpm-tis'>
    <backend type='emulator' version='2.0'>
      <encryption secret='6dd3e4a5-1d76-44ce-961f-f119f5aad935'/>
      <active_pcr_banks>
          <sha256/>
      </active_pcr_banks>
    </backend>
  </tpm>
</devices>
...
model

The model attribute specifies what device model QEMU provides to the guest. If no model name is provided, tpm-tis will automatically be chosen for non-PPC64 architectures. Since 4.4.0 , another available choice is the tpm-crb, which should only be used when the backend device is a TPM 2.0. Since 6.1.0 , pSeries guests on PPC64 are supported and the default is tpm-spapr. Since 6.5.0 , a new model called spapr-tpm-proxy was added for pSeries guests. This model only works with the passthrough backend. It creates a TPM Proxy device that communicates with an existing TPM Resource Manager in the host, for example /dev/tpmrm0, enabling the guest to run in secure virtual machine mode with the help of an Ultravisor. Adding a TPM Proxy to a pSeries guest brings no security benefits unless the guest is running on a PPC64 host that has an Ultravisor and a TPM Resource Manager. Only one TPM Proxy device is allowed per guest, but a TPM Proxy device can be added together with other TPM devices.

backend

The backend element specifies the type of TPM device. The following types are supported:

passthrough

Use the host's TPM or TPM Resource Manager device.

This backend type requires exclusive access to a TPM device on the host. An example for such a device is /dev/tpm0. The fully qualified file name is specified by path attribute of the source element. If no file name is specified then /dev/tpm0 is automatically used. Since 6.5.0 , when choosing the spapr-tpm-proxy model, the file name specified is expected to be a TPM Resource Manager device, e.g. /dev/tpmrm0.

emulator

For this backend type the 'swtpm' TPM Emulator must be installed on the host. Libvirt will automatically start an independent TPM emulator for each QEMU guest requesting access to it.

version

The version attribute indicates the version of the TPM. This attribute only works with the emulator backend. The following versions are supported:

  • '1.2' : creates a TPM 1.2
  • '2.0' : creates a TPM 2.0

The default version used depends on the combination of hypervisor, guest architecture, TPM model and backend.

persistent_state

The persistent_state attribute indicates whether 'swtpm' TPM state is kept or not when a transient domain is powered off or undefined. This option can be used for preserving TPM state. By default the value is no. This attribute only works with the emulator backend. The accepted values are yes and no. Since 7.0.0

active_pcr_banks

The active_pcr_banks node is used to define which of the PCR banks of a TPM 2.0 to activate. Valid names are for example sha1, sha256, sha384, and sha512. If this node is provided, the set of PCR banks are activated before every start of a VM and this step is logged in the swtpm's log. If this node is removed or omitted then libvirt will not modify the active PCR banks upon VM start but leave them at their last configuration. This attribute requires that swtpm_setup v0.7 or later is installed and may not have any effect otherwise. The selection of PCR banks only works with the emulator backend. Since 7.10.0

encryption

The encryption element allows the state of a TPM emulator to be encrypted. The secret must reference a secret object that holds the passphrase from which the encryption key will be derived.

NVRAM device

nvram device is always added to pSeries guest on PPC64, and its address is allowed to be changed. Element nvram (only valid for pSeries guest, since 1.0.5 ) is provided to enable the address setting.

Example: usage of NVRAM configuration

...
<devices>
  <nvram>
    <address type='spapr-vio' reg='0x00003000'/>
  </nvram>
</devices>
...
spapr-vio

VIO device address type, only valid for PPC64.

reg

Device address

panic device

panic device enables libvirt to receive panic notification from a QEMU guest. Since 1.2.1, QEMU and KVM only

This feature is always enabled for:

  • pSeries guests, since it's implemented by the guest firmware
  • S390 guests, since it's an integral part of the S390 architecture

For the guest types listed above, libvirt automatically adds a panic element to the domain XML.

Example: usage of panic configuration

...
<devices>
  <panic model='hyperv'/>
  <panic model='isa'>
    <address type='isa' iobase='0x505'/>
  </panic>
</devices>
...
model

The optional model attribute specifies what type of panic device is provided. The panic model used when this attribute is missing depends on the hypervisor and guest arch.

  • 'isa' - for ISA pvpanic device
  • 'pseries' - default and valid only for pSeries guests.
  • 'hyperv' - for Hyper-V crash CPU feature. Since 1.3.0, QEMU and KVM only
  • 's390' - default for S390 guests. Since 1.3.5
  • 'pvpanic' - for PCI pvpanic device Since 9.1.0, QEMU only
address

address of panic. The default ioport is 0x505. Most users don't need to specify an address, and doing so is forbidden altogether for s390, pseries and hyperv models.

Shared memory device

A shared memory device allows to share a memory region between different virtual machines and the host. Since 1.2.10, QEMU and KVM only

...
<devices>
  <shmem name='my_shmem0' role='peer'>
    <model type='ivshmem-plain'/>
    <size unit='M'>4</size>
  </shmem>
  <shmem name='shmem_server'>
    <model type='ivshmem-doorbell'/>
    <size unit='M'>2</size>
    <server path='/tmp/socket-shmem'/>
    <msi vectors='32' ioeventfd='on'/>
  </shmem>
</devices>
...
shmem

The shmem element has one mandatory attribute, name to identify the shared memory. This attribute cannot be directory specific to . or .. as well as it cannot involve path separator /.

The optional role (since 6.6.0) attribute specifies the shared memory is migratable or not. The value can be either "master" or "peer", the former will mean that upon migration, the data in the shared memory is migrated with the domain. There should be only one "master" per shared memory object. Migration with "peer" role is disabled. If migration of such domain is required, the shmem device needs to be unplugged before migration and plugged in at the destination upon successful migration. If the role not specified, the hypervisor default is used. This attribute is currently available only for model type ivshmem-plain and ivshmem-doorbell.

model

Attribute type of the optional element model specifies the model of the underlying device providing the shmem device. The models currently supported are ivshmem (supports both server and server-less shmem, but is deprecated by newer QEMU in favour of the -plain and -doorbell variants), ivshmem-plain (only for server-less shmem) and ivshmem-doorbell (only for shmem with the server).

size

The optional size element specifies the size of the shared memory. This must be power of 2 and greater than or equal to 1 MiB.

server

The optional server element can be used to configure a server socket the device is supposed to connect to. The optional path attribute specifies the absolute path to the unix socket and defaults to /var/lib/libvirt/shmem/$shmem-$name-sock.

msi

The optional msi element enables/disables (values "on"/"off", respectively) MSI interrupts. This option can currently be used only together with the server element. The vectors attribute can be used to specify the number of interrupt vectors. The ioeventd attribute enables/disables (values "on"/"off", respectively) ioeventfd.

Memory devices

In addition to the initial memory assigned to the guest, memory devices allow additional memory to be assigned to the guest in the form of memory modules. A memory device can be hot-plugged or hot-unplugged depending on the guests' memory resource needs. Some hypervisors may require NUMA configured for the guest.

Example: usage of the memory devices

...
<devices>
  <memory model='dimm' access='private' discard='yes'>
    <target>
      <size unit='KiB'>524287</size>
      <node>0</node>
    </target>
  </memory>
  <memory model='dimm'>
    <source>
      <pagesize unit='KiB'>2048</pagesize>
      <nodemask>1-3</nodemask>
    </source>
    <target>
      <size unit='KiB'>524287</size>
      <node>1</node>
    </target>
  </memory>
  <memory model='nvdimm'>
    <uuid>9066901e-c90a-46ad-8b55-c18868cf92ae</uuid>
    <source>
      <path>/tmp/nvdimm</path>
    </source>
    <target>
      <size unit='KiB'>524288</size>
      <node>1</node>
      <label>
        <size unit='KiB'>128</size>
      </label>
      <readonly/>
    </target>
  </memory>
  <memory model='nvdimm' access='shared'>
    <uuid>e39080c8-7f99-4b12-9c43-d80014e977b8</uuid>
    <source>
      <path>/dev/dax0.0</path>
      <alignsize unit='KiB'>2048</alignsize>
      <pmem/>
    </source>
    <target>
      <size unit='KiB'>524288</size>
      <node>1</node>
      <label>
        <size unit='KiB'>128</size>
      </label>
    </target>
  </memory>
  <memory model='virtio-pmem' access='shared'>
    <source>
      <path>/tmp/virtio_pmem</path>
    </source>
    <target>
      <size unit='KiB'>524288</size>
      <address base='0x140000000'/>
    </target>
  </memory>
  <memory model='virtio-mem'>
    <source>
      <nodemask>1-3</nodemask>
      <pagesize unit='KiB'>2048</pagesize>
    </source>
    <target>
      <size unit='KiB'>2097152</size>
      <node>0</node>
      <block unit='KiB'>2048</block>
      <requested unit='KiB'>1048576</requested>
      <current unit='KiB'>524288</current>
      <address base='0x150000000'/>
    </target>
  </memory>
  <memory model='sgx-epc'>
    <source>
      <nodemask>0-1</nodemask>
    </source>
    <target>
      <size unit='KiB'>16384</size>
      <node>0</node>
    </target>
  </memory>
  <memory model='sgx-epc'>
    <target>
      <size unit='KiB'>16384</size>
    </target>
  </memory>
</devices>
...
model

Provide dimm to add a virtual DIMM module to the guest. Since 1.2.14 Provide nvdimm model that adds a Non-Volatile DIMM module. Since 3.2.0 Provide virtio-pmem model to add a paravirtualized persistent memory device. Since 7.1.0 Provide virtio-mem model to add paravirtualized memory device. Since 7.9.0 Provide sgx-epc model to add a SGX enclave page cache (EPC) memory to the guest. Since 8.10.0 and QEMU 7.0.0

access

An optional attribute access ( since 3.2.0 ) that provides capability to fine tune mapping of the memory on per module basis. Values are the same as Memory Backing: shared and private. For nvdimm model, if using real NVDIMM DAX device as backend, shared is required. For virtio-pmem model shared is required.

discard

An optional attribute discard ( since 4.4.0 ) that provides capability to fine tune discard of data on per module basis. Accepted values are yes and no. The feature is described here: Memory Backing. This attribute is allowed only for model='dimm'.

uuid

For pSeries guests, an uuid can be set to identify the nvdimm module. If absent, libvirt will generate an uuid automatically. This attribute is allowed only for model='nvdimm' for pSeries guests. Since 6.2.0

source

For model dimm and model virtio-mem this element is optional and allows to fine tune the source of the memory used for the given memory device. If the element is not provided defaults configured via numatune are used. If the element is provided, then the following optional elements can be provided:

pagesize

This element can be used to override the default host page size used for backing the memory device. The configured value must correspond to a page size supported by the host.

nodemask

This element can be used to override the default set of NUMA nodes where the memory would be allocated.

For model nvdimm the source element is mandatory. The mandatory child element path represents a path in the host that backs the nvdimm module in the guest. The following optional elements may be used:

alignsize

The alignsize element defines the page size alignment used to mmap the address range for the backend path. If not supplied the host page size is used. For example, to mmap a real NVDIMM device a 2M-aligned page may be required, and host page size is 4KB, then we need to set this element to 2MB. Since 5.0.0

pmem

If persistent memory is supported and enabled by the hypervisor in order to guarantee the persistence of writes to the vNVDIMM backend, then use the pmem element in order to utilize the feature. Since 5.0.0

For model virtio-pmem the source element is mandatory. The following optional elements may be used:

path

Represents a path in the host that backs the virtio memory module in the guest. It is mandatory.

For model sgx-epc this element is optional. The following optional elements may be used:

nodemask

This element can be used to override the default set of NUMA nodes where the memory would be allocated. Since 8.10.0 and QEMU 7.0.0

target

The mandatory target element configures the placement and sizing of the added memory from the perspective of the guest.

The mandatory size subelement configures the size of the added memory as a scaled integer. For virtio-mem this represents the maximum possible size exposed to the guest.

The node subelement configures the guest NUMA node to attach the memory to. The element shall be used only if the guest has NUMA nodes configured.

The following optional elements may be used:

label

For NVDIMM type devices one can use label and its subelement size to configure the size of namespaces label storage within the NVDIMM module. The size element has usual meaning described in the Memory Allocation section. label is mandatory for pSeries guests and optional for all other architectures. For QEMU domains the following restrictions apply:

  1. the minimum label size is 128KiB,
  2. the remaining size (total-size - label-size), also called guest area, will be aligned to 4KiB as default. For pSeries guests, the guest area will be aligned down to 256MiB, and the minimum size of the guest area must be at least 256MiB.
readonly

The readonly element is used to mark the vNVDIMM as read-only. Only the real NVDIMM device backend can guarantee the guest write persistence, so other backend types should use the readonly element. Since 5.0.0

block

For virtio-mem only. The size of an individual block, granularity of division of memory block. Must be power of two and at least equal to size of a transparent hugepage (2MiB on x84_64). The default is hypervisor dependent.

requested

For virtio-mem only. The total size exposed to the guest. Must respect block granularity and be smaller than or equal to size.

current

Active XML for virtio-mem model may contain current element that reflects the current size of the corresponding virtio memory device. The element is formatted into live XML and never parsed, i.e. it is output-only element.

address

For virtio-mem and virtio-pmem only. The physical address in memory, where device is mapped. Since 9.4.0

IOMMU devices

The iommu element can be used to add an IOMMU device. Since 2.1.0

Example:

...
<devices>
  <iommu model='intel'>
    <driver intremap='on'/>
  </iommu>
</devices>
...
model

Supported values are intel (for Q35 guests) smmuv3 (since 5.5.0, for ARM virt guests), and virtio (since 8.3.0, for Q35 and ARM virt guests).

driver

The driver subelement can be used to configure additional options, some of which might only be available for certain IOMMU models:

intremap

The intremap attribute with possible values on and off can be used to turn on interrupt remapping, a part of the VT-d functionality. Currently this requires split I/O APIC (<ioapic driver='qemu'/>). Since 3.4.0 (QEMU/KVM only)

caching_mode

The caching_mode attribute with possible values on and off can be used to turn on the VT-d caching mode (useful for assigned devices). Since 3.4.0 (QEMU/KVM only)

eim

The eim attribute (with possible values on and off) can be used to configure Extended Interrupt Mode. A q35 domain with split I/O APIC (as described in Hypervisor features), and both interrupt remapping and EIM turned on for the IOMMU, will be able to use more than 255 vCPUs. Since 3.4.0 (QEMU/KVM only)

iotlb

The iotlb attribute with possible values on and off can be used to turn on the IOTLB used to cache address translation requests from devices. Since 3.5.0 (QEMU/KVM only)

aw_bits

The aw_bits attribute can be used to set the address width to allow mapping larger iova addresses in the guest. Since 6.5.0 (QEMU/KVM only)

The virtio IOMMU devices can further have address element as described in Device addresses (address has to by type of pci).

Vsock

A vsock host/guest interface. The model attribute defaults to virtio. Since 5.2.0 model can also be 'virtio-transitional' and 'virtio-non-transitional', see Virtio transitional devices for more details. The optional attribute address of the cid element specifies the CID assigned to the guest. If the attribute auto is set to yes, libvirt will assign a free CID automatically on domain startup. Since 4.4.0 The optional driver element allows to specify virtio options, see Virtio-related options for more details. Since 7.1.0

...
<devices>
  <vsock model='virtio'>
    <cid auto='no' address='3'/>
  </vsock>
</devices>
...

Crypto

A crypto device. The model attribute defaults to virtio. Since v9.0.0 model supports virtio only. The type attribute defaults to qemu. Since v9.0.0 type supports qemu only. The optional attribute backend is required if the type is qemu, the model attribute can be builtint and lkcf, the optional attribute queues specifies the number of virt queues for virtio crypto.

...
<devices>
  <crypto model='virtio' type='qemu'>
    <backend model='builtin' queues='1'/>
  </crypto>
</devices>
...

Security label

The seclabel element allows control over the operation of the security drivers. There are three basic modes of operation, 'dynamic' where libvirt automatically generates a unique security label, 'static' where the application/administrator chooses the labels, or 'none' where confinement is disabled. With dynamic label generation, libvirt will always automatically relabel any resources associated with the virtual machine. With static label assignment, by default, the administrator or application must ensure labels are set correctly on any resources, however, automatic relabeling can be enabled if desired. 'dynamic' since 0.6.1, 'static' since 0.6.2, and 'none' since 0.9.10.

If more than one security driver is used by libvirt, multiple seclabel tags can be used, one for each driver and the security driver referenced by each tag can be defined using the attribute model

Valid input XML configurations for the top-level security label are:

<seclabel type='dynamic' model='selinux'/>

<seclabel type='dynamic' model='selinux'>
  <baselabel>system_u:system_r:my_svirt_t:s0</baselabel>
</seclabel>

<seclabel type='static' model='selinux' relabel='no'>
  <label>system_u:system_r:svirt_t:s0:c392,c662</label>
</seclabel>

<seclabel type='static' model='selinux' relabel='yes'>
  <label>system_u:system_r:svirt_t:s0:c392,c662</label>
</seclabel>

<seclabel type='none'/>

If no 'type' attribute is provided in the input XML, then the security driver default setting will be used, which may be either 'none' or 'dynamic'. If a 'baselabel' is set but no 'type' is set, then the type is presumed to be 'dynamic'

When viewing the XML for a running guest with automatic resource relabeling active, an additional XML element, imagelabel, will be included. This is an output-only element, so will be ignored in user supplied XML documents

type

Either static, dynamic or none to determine whether libvirt automatically generates a unique security label or not.

model

A valid security model name, matching the currently activated security model. Model dac is not available when guest is run by unprivileged user.

relabel

Either yes or no. This must always be yes if dynamic label assignment is used. With static label assignment it will default to no.

label

If static labelling is used, this must specify the full security label to assign to the virtual domain. The format of the content depends on the security driver in use:

  • SELinux: a SELinux context.
  • AppArmor: an AppArmor profile.
  • DAC: owner and group separated by colon. They can be defined both as user/group names or uid/gid. The driver will first try to parse these values as names, but a leading plus sign can used to force the driver to parse them as uid or gid.
baselabel

If dynamic labelling is used, this can optionally be used to specify the base security label that will be used to generate the actual label. The format of the content depends on the security driver in use. The SELinux driver uses only the type field of the baselabel in the generated label. Other fields are inherited from the parent process when using SELinux baselabels. (The example above demonstrates the use of my_svirt_t as the value for the type field.)

imagelabel

This is an output only element, which shows the security label used on resources associated with the virtual domain. The format of the content depends on the security driver in use

When relabeling is in effect, it is also possible to fine-tune the labeling done for specific source file names, by either disabling the labeling (useful if the file lives on NFS or other file system that lacks security labeling) or requesting an alternate label (useful when a management application creates a special label to allow sharing of some, but not all, resources between domains), since 0.9.9 . When a seclabel element is attached to a specific path rather than the top-level domain assignment, only the attribute relabel or the sub-element label are supported. Additionally, since 1.1.2 , an output-only element labelskip will be present for active domains on disks where labeling was skipped due to the image being on a file system that lacks security labeling.

Key Wrap

The content of the optional keywrap element specifies whether the guest will be allowed to perform the S390 cryptographic key management operations. A clear key can be protected by encrypting it under a unique wrapping key that is generated for each guest VM running on the host. Two variations of wrapping keys are generated: one version for encrypting protected keys using the DEA/TDEA algorithm, and another version for keys encrypted using the AES algorithm. If a keywrap element is not included, the guest will be granted access to both AES and DEA/TDEA key wrapping by default.

<domain>
  ...
  <keywrap>
    <cipher name='aes' state='off'/>
  </keywrap>
  ...
</domain>

At least one cipher element must be nested within the keywrap element.

cipher

The name attribute identifies the algorithm for encrypting a protected key. The values supported for this attribute are aes for encryption under the AES wrapping key, or dea for encryption under the DEA/TDEA wrapping key. The state attribute indicates whether the cryptographic key management operations should be turned on for the specified encryption algorithm. The value can be set to on or off.

Note: DEA/TDEA is synonymous with DES/TDES.

Launch Security

Specifying <launchSecurity type='s390-pv'/> in a s390 domain prepares the guest to run in protected virtualization secure mode, also known as IBM Secure Execution. For more required host and guest preparation steps, see Protected Virtualization on s390 Since 7.6.0

The contents of the <launchSecurity type='sev'> element is used to provide the guest owners input used for creating an encrypted VM using the AMD SEV feature (Secure Encrypted Virtualization). SEV is an extension to the AMD-V architecture which supports running encrypted virtual machine (VMs) under the control of KVM. Encrypted VMs have their pages (code and data) secured such that only the guest itself has access to the unencrypted version. Each encrypted VM is associated with a unique encryption key; if its data is accessed to a different entity using a different key the encrypted guests data will be incorrectly decrypted, leading to unintelligible data. For more information see various input parameters and its format see the SEV API spec Since 4.4.0

<domain>
  ...
  <launchSecurity type='sev' kernelHashes='yes'>
    <policy>0x0001</policy>
    <cbitpos>47</cbitpos>
    <reducedPhysBits>1</reducedPhysBits>
    <dhCert>RBBBSDDD=FDDCCCDDDG</dhCert>
    <session>AAACCCDD=FFFCCCDSDS</session>
  </launchSecurity>
  ...
</domain>
kernelHashes

The optional kernelHashes attribute indicates whether the hashes of the kernel, ramdisk and command line should be included in the measurement done by the firmware. This is only valid if using direct kernel boot. Since 8.0.0

cbitpos

The required cbitpos element provides the C-bit (aka encryption bit) location in guest page table entry. The value of cbitpos is hypervisor dependent and can be obtained through the sev element from the domain capabilities.

reducedPhysBits

The required reducedPhysBits element provides the physical address bit reduction. Similar to cbitpos the value of reduced-phys-bit is hypervisor dependent and can be obtained through the sev element from the domain capabilities.

policy

The required policy element provides the guest policy which must be maintained by the SEV firmware. This policy is enforced by the firmware and restricts what configuration and operational commands can be performed on this guest by the hypervisor. The guest policy provided during guest launch is bound to the guest and cannot be changed throughout the lifetime of the guest. The policy is also transmitted during snapshot and migration flows and enforced on the destination platform. The guest policy is a 4 unsigned byte with the fields shown in Table:

Bit(s) Description
0 Debugging of the guest is disallowed when set
1 Sharing keys with other guests is disallowed when set
2 SEV-ES is required when set
3 Sending the guest to another platform is disallowed when set
4 The guest must not be transmitted to another platform that is not in the domain when set.
5 The guest must not be transmitted to another platform that is not SEV capable when set.
6:15 reserved
16:32 The guest must not be transmitted to another platform with a lower firmware version.
dhCert

The optional dhCert element provides the guest owners base64 encoded Diffie-Hellman (DH) key. The key is used to negotiate a master secret key between the SEV firmware and guest owner. This master secret key is then used to establish a trusted channel between SEV firmware and guest owner.

session

The optional session element provides the guest owners base64 encoded session blob defined in the SEV API spec. See SEV spec LAUNCH_START section for the session blob format.

Example configs

Example configurations for each driver are provide on the driver specific pages listed below