This page provides an introduction to the network XML format. For background information on the concepts referred to here, consult the network driver architecture page.
The root element required for all virtual networks is
named network and has no configurable attributes
(although since 0.10.0 there is one
optional read-only attribute - when examining the live
configuration of a network, the
attribute connections, if present, specifies the
number of guest interfaces currently connected via this
network). The network XML format is
available since 0.3.0
The first elements provide basic metadata about the virtual network.
<network ipv6='yes'>
<name>default</name>
<uuid>3e3fce45-4f53-4fa7-bb32-11f34168b82b</uuid>
...
namename element provides
a short name for the virtual network. This name should
consist only of alpha-numeric characters and is required
to be unique within the scope of a single host. It is
used to form the filename for storing the persistent
configuration file. Since 0.3.0uuiduuid element provides
a globally unique identifier for the virtual network.
The format must be RFC 4122 compliant, eg 3e3fce45-4f53-4fa7-bb32-11f34168b82b.
If omitted when defining/creating a new network, a random
UUID is generated. Since 0.3.0ipv6='yes'ipv6='yes' enables
a network definition with no IPv6 gateway addresses specified
to have guest-to-guest communications. For further information,
see the example below for the example with no gateway addresses.
Since 1.0.1The next set of elements control how a virtual network is provided connectivity to the physical LAN (if at all).
...
<bridge name="virbr0" stp="on" delay="5"/>
<domain name="example.com"/>
<forward mode="nat" dev="eth0"/>
...
bridgename attribute on the bridge element
defines the name of a bridge device which will be used to construct
the virtual network. The virtual machines will be connected to this
bridge device allowing them to talk to each other. The bridge device
may also be connected to the LAN. It is recommended that bridge
device names started with the prefix vir, but the name
virbr0 is reserved for the "default" virtual
network. This element should always be provided when defining
a new network with a <forward> mode of
"nat" or "route" (or an isolated network with
no <forward> element).
Attribute stp specifies if Spanning Tree Protocol
is 'on' or 'off' (default is
'on'). Attribute delay sets the bridge's forward
delay value in seconds (default is 0).
Since 0.3.0
domainname attribute on the domain
element defines the DNS domain of the DHCP server. This
element is optional, and is only used for those networks with
a <forward> mode of "nat" or "route" (or an
isolated network with no <forward>
element). Since 0.4.5
forwardforward element indicates that
the virtual network is to be connected to the physical
LAN.Since 0.3.0.
The mode attribute determines the method of
forwarding. If there is no forward element, the
network will be isolated from any other network (unless a
guest connected to that network is acting as a router, of
course). The following are valid settings
for mode (if there is a forward
element but mode is not specified, mode='nat' is
assumed):
natdev attribute is set, the
firewall rules will restrict forwarding to the named
device only. Inbound connections from other networks are
all prohibited; all connections between guests on the same
network, and to/from the host to the guests, are
unrestricted and not NATed.Since
0.4.2
Since 1.0.3 it is possible to
specify a public IPv4 address and port range to be used for
the NAT by using the <nat> subelement.
The address range is set with the <address>
subelements and start and stop
attributes:
...
<forward mode='nat'>
<nat>
<address start='1.2.3.4' end='1.2.3.10'/>
</nat>
</forward>
...
An singe IPv4 address can be set by setting
start and end attributes to
the same value.
The port range to be used for the <nat> can
be set via the subelement <port>:
...
<forward mode='nat'>
<nat>
<port start='500' end='1000'/>
</nat>
</forward>
...
routedev
attribute is set, firewall rules will restrict forwarding
to the named device only. This presumes that the local LAN
router has suitable routing table entries to return
traffic to this host. All incoming and outgoing sessions
to guest on these networks are unrestricted. (To restrict
incoming traffic to a guest on a routed network, you can
configure nwfilter rules
on the guest's interfaces.)
Since 0.4.2
bridge<bridge name='xyz'/> element has been
specified, Since 0.9.4), 2) an
existing Open vSwitch bridge that was configured outside of
libvirt (if both a <bridge name='xyz'/>
element and a <virtualport
type='openvswitch'/> have been
specified Since 0.10.0) 3) an
interface or group of interfaces to be used for a "direct"
connection via macvtap using macvtap's "bridge" mode (if
the forward element has one or
more <interface>
subelements, Since 0.9.4)
(see Direct
attachment to physical interface for descriptions of
the various macvtap modes). libvirt doesn't attempt to
manage the bridge interface at all, thus
the <bridge> element's stp
and delay attributes are not allowed; no
iptables rules, IP addresses, or DHCP/DNS services are
added; at the IP level, the guest interface appears to be
directly connected to the physical
interface.Since 0.9.4
private<interface> subelements
of the <forward> element; when using
802.1Qbh mode (as indicated by
the <virtualport> type attribute - note
that this requires an 802.1Qbh-capable hardware switch),
each physical interface can only be in use by a single
guest interface at a time; in modes other than 802.1Qbh,
multiple guest interfaces can share each physical
interface (libvirt will attempt to balance usage between
all available interfaces).Since
0.9.4
vepa<interface> subelements of
the <forward> element; multiple guest
interfaces can share each physical interface (libvirt will
attempt to balance usage between all available
interfaces).Since 0.9.4
passthrough<interface> subelements of
the <forward> element. Each physical
interface can only be in use by a single guest interface
at a time, so libvirt will keep track of which interfaces
are currently in use, and only assign unused interfaces
(if there are no available physical interfaces when a
domain interface is being attached, an error will be
logged, and the operation causing the attach will fail
(usually either a domain start, or a hotplug interface
attach to a domain).Since 0.9.4
hostdev<virtualport>
element. 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. Since 0.10.0
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, vlan tag, 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 0.10.0, you
should use a standard
<hostdev> device definition in the
domain's configuration to assign the device to the guest
instead of defining an <interface
type='network'> pointing to a network
with <forward mode='hostdev'/>.
<forward> element can
have multiple <interface> subelements, each
one giving the name of a physical interface that can be used
for this network Since 0.9.4:
...
<forward mode='passthrough'>
<interface dev='eth10'/>
<interface dev='eth11'/>
<interface dev='eth12'/>
<interface dev='eth13'/>
<interface dev='eth14'/>
</forward>
...
since 0.10.0,
<interface> also has an optional read-only
attribute - when examining the live configuration of a
network, the attribute connections, if present,
specifies the number of guest interfaces currently connected
via this physical interface.
Additionally, since 0.9.10, libvirt
allows a shorthand for specifying all virtual interfaces
associated with a single physical function, by using
the <pf> subelement to call out the
corresponding physical interface associated with multiple
virtual interfaces:
...
<forward mode='passthrough'>
<pf dev='eth0'/>
</forward>
...
When a guest interface is being constructed, libvirt will pick an interface from this list to use for the connection. In modes where physical interfaces can be shared by multiple guest interfaces, libvirt will choose the interface that currently has the least number of connections. For those modes that do not allow sharing of the physical device (in particular, 'passthrough' mode, and 'private' mode when using 802.1Qbh), libvirt will choose an unused physical interface or, if it can't find an unused interface, fail the operation.
since 0.10.0 When using forward
mode 'hostdev', the interface pool is specified with a list
of <address> elements, each of which has
< type> (must always be 'pci',
<domain>, <bus>,
<slot>, and <function>
attributes.
...
<forward mode='hostdev' managed='yes'>
<address type='pci' domain='0' bus='4' slot='0' function='1'/>
<address type='pci' domain='0' bus='4' slot='0' function='2'/>
<address type='pci' domain='0' bus='4' slot='0' function='3'/>
</forward>
...
Alternatively the interface pool can also be defined using a
single physical function <pf> subelement to
call out the corresponding physical interface associated with
multiple virtual interfaces (similar to passthrough mode):
...
<forward mode='hostdev' managed='yes'>
<pf dev='eth0'/>
</forward>
...
...
<forward mode='nat' dev='eth0'/>
<bandwidth>
<inbound average='1000' peak='5000' burst='5120'/>
<outbound average='128' peak='256' burst='256'/>
</bandwidth>
...
This part of network XML provides setting quality of service. Incoming
and outgoing traffic can be shaped independently. The
bandwidth element can have at most one inbound
and at most one outbound child elements. Leaving any of these
children element out result in no QoS applied on that traffic direction.
So, when you want to shape only network's incoming traffic, use
inbound only, and vice versa. Each of these elements have one
mandatory attribute average. It specifies average bit rate on
interface being shaped. Then there are two optional attributes:
peak, which specifies maximum rate at which bridge can send
data, and burst, amount of bytes that can be burst at
peak speed. Accepted values for attributes are integer
numbers, The units for average and peak attributes
are kilobytes per second, and for the burst just kilobytes.
The rate is shared equally within domains connected to the network.
Moreover, bandwidth element can be included in
portgroup element.
Since 0.9.4
...
<devices>
<interface type='bridge'>
<vlan trunk='yes'>
<tag id='42'/>
<tag id='47'/>
</vlan>
<source bridge='ovsbr0'/>
<virtualport type='openvswitch'>
<parameters interfaceid='09b11c53-8b5c-4eeb-8f00-d84eaa0aaa4f'/>
</virtualport>
</interface>
<devices>
...
If (and only if) the network type supports vlan tagging
transparent to the guest, an optional <vlan>
element can specify one or more vlan tags to apply to the
traffic of all guests using this
network Since 0.10.0. (openvswitch
and type='hostdev' SR-IOV networks do support transparent vlan
tagging of guest traffic; everything else, 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 specific vlans.) As expected, the tag
attribute specifies which vlan tag to use. If a network has more
than one <vlan> element defined, it is
assumed that the user wants to do VLAN trunking using all the
specified tags. In the case that vlan trunking with a single tag
is desired, the optional attribute trunk='yes' can
be added to the vlan element.
<vlan> elements can also be specified in
a <portgroup> element, as well as directly in
a domain's <interface> element. In the case
that a vlan tag is specified in multiple locations, the setting
in <interface> takes precedence, followed by
the setting in the <portgroup> selected by
the interface config. The <vlan>
in <network> will be selected only if none is
given in <portgroup>
or <interface>.
...
<forward mode='private'/>
<interface dev="eth20"/>
<interface dev="eth21"/>
<interface dev="eth22"/>
<interface dev="eth23"/>
<interface dev="eth24"/>
</forward>
<portgroup name='engineering' default='yes'>
<virtualport type='802.1Qbh'>
<parameters profileid='test'/>
</virtualport>
<bandwidth>
<inbound average='1000' peak='5000' burst='5120'/>
<outbound average='1000' peak='5000' burst='5120'/>
</bandwidth>
</portgroup>
<portgroup name='sales'>
<virtualport type='802.1Qbh'>
<parameters profileid='salestest'/>
</virtualport>
<bandwidth>
<inbound average='500' peak='2000' burst='2560'/>
<outbound average='128' peak='256' burst='256'/>
</bandwidth>
</portgroup>
...
Since 0.9.4
A portgroup provides a method of easily putting guest
connections to the network into different classes, with each
class potentially having a different level/type of service.
Since 0.9.4 Each
network can have multiple portgroup elements (and one of those
can optionally be designated as the 'default' portgroup for the
network), and each portgroup has a name, as well as various
subelements associated with it. The currently supported
subelements are <bandwidth>
(documented here)
and <virtualport>
(documented here).
If a domain interface definition specifies a portgroup (by
adding a portgroup attribute to
the <source> subelement), that portgroup's
info will be merged into the interface's configuration. If no
portgroup is given in the interface definition, and one of the
network's portgroups has default='yes', that
default portgroup will be used. If no portgroup is given in the
interface definition, and there is no default portgroup, then
none will be used. Any <bandwidth>
specified directly in the domain XML will take precedence over
any setting in the chosen portgroup. if
a <virtualport> is specified in the portgroup
(and/or directly in the network definition), the multiple
virtualports will be merged, and any parameter that is specified
in more than one virtualport, and is not identical, will be
considered an error, and will prevent the interface from
starting.
The final set of elements define the addresses (IPv4 and/or
IPv6, as well as MAC) to be assigned to the bridge device
associated with the virtual network, and optionally enable DHCP
services. These elements are only valid for isolated networks
(no forward element specified), and for those with
a forward mode of 'route' or 'nat'.
...
<mac address='00:16:3E:5D:C7:9E'/>
<domain name="example.com"/>
<dns>
<txt name="example" value="example value" />
<srv service='name' protocol='tcp' domain='test-domain-name' target='.' port='1024' priority='10' weight='10'/>
<host ip='192.168.122.2'>
<hostname>myhost</hostname>
<hostname>myhostalias</hostname>
</host>
</dns>
<ip address="192.168.122.1" netmask="255.255.255.0">
<dhcp>
<range start="192.168.122.100" end="192.168.122.254" />
<host mac="00:16:3e:77:e2:ed" name="foo.example.com" ip="192.168.122.10" />
<host mac="00:16:3e:3e:a9:1a" name="bar.example.com" ip="192.168.122.11" />
</dhcp>
</ip>
<ip family="ipv6" address="2001:db8:ca2:2::1" prefix="64" />
</network>
macaddress attribute defines a MAC
(hardware) address formatted as 6 groups of 2-digit
hexadecimal numbers, the groups separated by colons
(eg, "52:54:00:1C:DA:2F"). This MAC address is
assigned to the bridge device when it is created. Generally
it is best to not specify a MAC address when creating a
network - in this case, if a defined MAC address is needed for
proper operation, libvirt will automatically generate a random
MAC address and save it in the config. Allowing libvirt to
generate the MAC address will assure that it is compatible
with the idiosyncrasies of the platform where libvirt is
running. Since 0.8.8
dnstxtdns element can have 0 or more txt elements.
Each txt element defines a DNS TXT record and has two attributes, both
required: a name that can be queried via dns, and a value that will be
returned when that name is queried. names cannot contain embedded spaces
or commas. value is a single string that can contain multiple values
separated by commas. Since 0.9.3
hosthost element within dns is the
definition of DNS hosts to be passed to the DNS service. The IP
address is identified by the ip attribute and the names
for that IP address are identified in the hostname
sub-elements of the host element.
Since 0.9.3
srvdns element can have also 0 or more srv
record elements. Each srv record element defines a DNS SRV record
and has 2 mandatory and 5 optional attributes. The mandatory attributes
are service name and protocol (tcp, udp) and the optional attributes are
target, port, priority, weight and domain as defined in DNS server SRV
RFC (RFC 2782).
Since 0.9.9
ipaddress attribute defines an IPv4 address in
dotted-decimal format, or an IPv6 address in standard
colon-separated hexadecimal format, that will be configured on
the bridge
device associated with the virtual network. To the guests this IPv4
address will be their IPv4 default route. For IPv6, the default route is
established via Router Advertisement.
For IPv4 addresses, the netmask
attribute defines the significant bits of the network address,
again specified in dotted-decimal format. For IPv6 addresses,
and as an alternate method for IPv4 addresses, you can specify
the significant bits of the network address with the prefix
attribute, which is an integer (for example, netmask='255.255.255.0'
could also be given as prefix='24'. The family
attribute is used to specify the type of address - 'ipv4' or 'ipv6'; if no
family is given, 'ipv4' is assumed. A network can have more than
one of each family of address defined, but only a single IPv4 address can have a
dhcp or tftp element. Since 0.3.0
IPv6, multiple addresses on a single network, family, and
prefix are support Since 0.8.7.
Similar to IPv4, one IPv6 address per network can also have
a dhcp definition. Since 1.0.1
tftpip element there is an optional tftp
element. The presence of this element and of its attribute
root enables TFTP services. The attribute specifies
the path to the root directory served via TFTP. tftp is not
supported for IPv6 addresses, and can only be specified on a single IPv4 address
per network.
Since 0.7.1
dhcpip element there is an
optional dhcp element. The presence of this element
enables DHCP services on the virtual network. It will further
contain one or more range elements. The
dhcp element supported for both
IPv4 Since 0.3.0
and IPv6 Since 1.0.1, but
only for one IP address of each type per network.
rangestart and end attributes on the
range element specify the boundaries of a pool of
addresses to be provided to DHCP clients. These two addresses
must lie within the scope of the network defined on the parent
ip element. There may be zero or more
range elements specified.
Since 0.3.0
range can be specified for one IPv4 address,
one IPv6 address, or both. Since 1.0.1
hostdhcp element there may be zero or more
host elements. These specify hosts which will be given
names and predefined IP addresses by the built-in DHCP server. Any
IPv4 host element must specify the MAC address of the host to be assigned
a given name (via the mac attribute), the IP to be
assigned to that host (via the ip attribute), and the
name to be given that host by the DHCP server (via the
name attribute). Since 0.4.5
An IPv6 host element differs slightly from that for IPv4:
there is no mac attribute since a MAC address has no
defined meaning in IPv6. Instead, the name attribute is
used to identify the host to be assigned the IPv6 address. For DHCPv6,
the name is the plain name of the client host sent by the
client to the server. Note that this method of assigning a
specific IP address can also be used instead of the mac
attribute for IPv4. Since 1.0.1
bootpbootp
element specifies BOOTP options to be provided by the DHCP
server for IPv4 only.
Two attributes are supported: file is mandatory and
gives the file to be used for the boot image; server is
optional and gives the address of the TFTP server from which the boot
image will be fetched. server defaults to the same host
that runs the DHCP server, as is the case when the tftp
element is used. The BOOTP options currently have to be the same
for all address ranges and statically assigned addresses.Since 0.7.1 (server since 0.7.3).
This example is the so called "default" virtual network. It is provided and enabled out-of-the-box for all libvirt installations. This is a configuration that allows guest OS to get outbound connectivity regardless of whether the host uses ethernet, wireless, dialup, or VPN networking without requiring any specific admin configuration. In the absence of host networking, it at least allows guests to talk directly to each other.
<network>
<name>default</name>
<bridge name="virbr0" />
<forward mode="nat"/>
<ip address="192.168.122.1" netmask="255.255.255.0">
<dhcp>
<range start="192.168.122.2" end="192.168.122.254" />
</dhcp>
</ip>
<ip family="ipv6" address="2001:db8:ca2:2::1" prefix="64" />
</network>
Below is a variation of the above example which adds an IPv6 dhcp range definition.
<network>
<name>default6</name>
<bridge name="virbr0" />
<forward mode="nat"/>
<ip address="192.168.122.1" netmask="255.255.255.0">
<dhcp>
<range start="192.168.122.2" end="192.168.122.254" />
</dhcp>
</ip>
<ip family="ipv6" address="2001:db8:ca2:2::1" prefix="64" >
<dhcp>
<range start="2001:db8:ca2:2:1::10" end="2001:db8:ca2:2:1::ff" />
</dhcp>
</ip>
</network>
This is a variant on the default network which routes traffic
from the virtual network to the LAN without applying any NAT.
It requires that the IP address range be pre-configured in the
routing tables of the router on the host network. This example
further specifies that guest traffic may only go out via the
eth1 host network device.
<network>
<name>local</name>
<bridge name="virbr1" />
<forward mode="route" dev="eth1"/>
<ip address="192.168.122.1" netmask="255.255.255.0">
<dhcp>
<range start="192.168.122.2" end="192.168.122.254" />
</dhcp>
</ip>
<ip family="ipv6" address="2001:db8:ca2:2::1" prefix="64" />
</network>
Below is another IPv6 variation. Instead of a dhcp range being specified, this example has a couple of IPv6 host definitions. Note that most of the dhcp host definitions use an "id" (client id or DUID) since this has proven to be a more reliable way of specifying the interface and its association with an IPv6 address. The first is a DUID-LLT, the second a DUID-LL, and the third a DUID-UUID. Since 1.0.3
<network>
<name>local6</name>
<bridge name="virbr1" />
<forward mode="route" dev="eth1"/>
<ip address="192.168.122.1" netmask="255.255.255.0">
<dhcp>
<range start="192.168.122.2" end="192.168.122.254" />
</dhcp>
</ip>
<ip family="ipv6" address="2001:db8:ca2:2::1" prefix="64" >
<dhcp>
<host name="paul" ip="2001:db8:ca2:2:3::1" />
<host id="0:1:0:1:18:aa:62:fe:0:16:3e:44:55:66" ip="2001:db8:ca2:2:3::2" />
<host id="0:3:0:1:0:16:3e:11:22:33" name="ralph" ip="2001:db8:ca2:2:3::3" />
<host id="0:4:7e:7d:f0:7d:a8:bc:c5:d2:13:32:11:ed:16:ea:84:63" name="badbob" ip="2001:db8:ca2:2:3::4" />
</dhcp>
</ip>
</network>
This variant provides a completely isolated private network
for guests. The guests can talk to each other, and the host
OS, but cannot reach any other machines on the LAN, due to
the omission of the forward element in the XML
description.
<network>
<name>private</name>
<bridge name="virbr2" />
<ip address="192.168.152.1" netmask="255.255.255.0">
<dhcp>
<range start="192.168.152.2" end="192.168.152.254" />
</dhcp>
</ip>
<ip family="ipv6" address="2001:db8:ca2:3::1" prefix="64" />
</network>
This variation of an isolated network defines only IPv6. Note that most of the dhcp host definitions use an "id" (client id or DUID) since this has proven to be a more reliable way of specifying the interface and its association with an IPv6 address. The first is a DUID-LLT, the second a DUID-LL, and the third a DUID-UUID. Since 1.0.3
<network>
<name>sixnet</name>
<bridge name="virbr6" />
<ip family="ipv6" address="2001:db8:ca2:6::1" prefix="64" >
<dhcp>
<host name="peter" ip="2001:db8:ca2:6:6::1" />
<host id="0:1:0:1:18:aa:62:fe:0:16:3e:44:55:66" ip="2001:db8:ca2:6:6::2" />
<host id="0:3:0:1:0:16:3e:11:22:33" name="dariusz" ip="2001:db8:ca2:6:6::3" />
<host id="0:4:7e:7d:f0:7d:a8:bc:c5:d2:13:32:11:ed:16:ea:84:63" name="anita" ip="2001:db8:ca2:6:6::4" />
</dhcp>
</ip>
</network>
Since 0.9.4 This shows how to use a pre-existing host bridge "br0". The guests will effectively be directly connected to the physical network (i.e. their IP addresses will all be on the subnet of the physical network, and there will be no restrictions on inbound or outbound connections).
<network>
<name>host-bridge</name>
<forward mode="bridge"/>
<bridge name="br0"/>
</network>
Since 0.9.4, QEMU and KVM only, requires Linux kernel 2.6.34 or newer This shows how to use macvtap to connect to the physical network directly through one of a group of physical devices (without using a host bridge device). As with the host bridge network, the guests will effectively be directly connected to the physical network so their IP addresses will all be on the subnet of the physical network, and there will be no restrictions on inbound or outbound connections. Note that, due to a limitation in the implementation of macvtap, these connections do not allow communication directly between the host and the guests - if you require this you will either need the attached physical switch to be operating in a mirroring mode (so that all traffic coming to the switch is reflected back to the host's interface), or provide alternate means for this communication (e.g. a second interface on each guest that is connected to an isolated network). The other forward modes that use macvtap (private, vepa, and passthrough) would be used in a similar fashion.
<network>
<name>direct-macvtap</name>
<forward mode="bridge">
<interface dev="eth20"/>
<interface dev="eth21"/>
<interface dev="eth22"/>
<interface dev="eth23"/>
<interface dev="eth24"/>
</forward>
</network>
A valid network definition can contain no IPv4 or IPv6 addresses. Such a definition can be used for a "very private" or "very isolated" network since it will not be possible to communicate with the virtualization host via this network. However, this virtual network interface can be used for communication between virtual guest systems. This works for IPv4 and (Since 1.0.1) IPv6. However, the new ipv6='yes' must be added for guest-to-guest IPv6 communication.
<network ipv6='yes'>
<name>nogw</name>
<uuid>7a3b7497-1ec7-8aef-6d5c-38dff9109e93</uuid>
<bridge name="virbr2" stp="on" delay="0" />
<mac address='00:16:3E:5D:C7:9E'/>
</network>