Although all storage pool backends share the same public APIs and XML format, they have varying levels of capabilities. Some may allow creation of volumes, others may only allow use of pre-existing volumes. Some may have constraints on volume size, or placement.
The top level tag for a storage pool document is 'pool'. It has
a single attribute type, which is one of dir,
fs, netfs, disk,
iscsi, logical, scsi
(all since 0.4.1),
mpath (since 0.7.1),
rbd (since 0.9.13),
sheepdog (since 0.10.0),
gluster (since 1.2.0),
zfs (since 1.2.8),
vstorage (since 3.1.0),
or iscsi-direct (since 4.7.0).
This corresponds to the
storage backend drivers listed further along in this document.
<pool type="iscsi"> <name>virtimages</name> <uuid>3e3fce45-4f53-4fa7-bb32-11f34168b82b</uuid> <allocation>10000000</allocation> <capacity>50000000</capacity> <available>40000000</available> ...
nameuuidallocationcapacityavailableSome pools support optional features:
... <features> <cow state='no'> </features> ...
Valid features are:
cowbtrfs filesystem. If not set then libvirt
will attempt to disable COW on any btrfs filesystems.
Since 6.6.0.
A single source element is contained within the top level
pool element. This tag is used to describe the source of
the storage pool. The set of child elements that it will contain
depend on the pool type, but come from the following child elements:
...
<source>
<host name="iscsi.example.com"/>
<device path="iqn.2013-06.com.example:iscsi-pool"/>
<auth type='chap' username='myname'>
<secret usage='mycluster_myname'/>
</auth>
<vendor name="Acme"/>
<product name="model"/>
</source>
...
... <source> <device path='/dev/mapper/mpatha' part_separator='no'/> <format type='gpt'/> </source> ...
... <source> <adapter type='scsi_host' name='scsi_host1'/> </source> ...
...
<source>
<adapter type='scsi_host'>
<parentaddr unique_id='1'>
<address domain='0x0000' bus='0x00' slot='0x1f' addr='0x2'/>
</parentaddr>
</adapter>
</source>
...
... <source> <adapter type='fc_host' parent='scsi_host5' wwnn='20000000c9831b4b' wwpn='10000000c9831b4b'/> </source> ...
...
<source>
<host name='localhost'/>
<dir path='/var/lib/libvirt/images'/>
<format type='nfs'/>
<protocol ver='3'/>
</source>
...
devicefs, logical, disk,
iscsi, iscsi-direct, zfs,
vstorage).
May be repeated multiple times depending on backend driver. Contains
a required attribute path which is either the fully
qualified path to the block device node or for iscsi
or iscsi-direct the iSCSI Qualified Name (IQN).
Since 0.4.1
An optional attribute part_separator for each
path may be supplied. Valid values for the attribute
may be either "yes" or "no". This attribute is to be used for a
disk pool type using a path to a
device mapper multipath device. Setting the attribute to "yes"
causes libvirt to attempt to generate and find target volume path's
using a "p" separator. The default algorithm used by device mapper
is to add the "p" separator only when the source device path ends
with a number; however, it's possible to configure the devmapper
device to not use 'user_friendly_names' thus creating partitions
with the "p" separator even when the device source path does not
end with a number.
Since 1.3.1
dirdir, netfs, gluster),
or optionally to select a subdirectory
within a pool that resembles a filesystem (pool
type gluster). May
only occur once. Contains a single attribute path
which is the fully qualified path to the backing directory or
for a netfs pool type using format
type "cifs", the path to the Samba share without the leading slash.
Since 0.4.1adapterscsi). May only occur once.
namevirsh nodedev-list
scsi_host command followed by a combination of
lspci and virsh nodedev-dumpxml
scsi_hostN commands to find the scsi_hostN
to be used. Since 0.6.2
It is further recommended to utilize the
parentaddr element since it's possible to have
the path to which the scsi_hostN uses change between system
reboots. Since 1.2.7
typename attribute is
specified, then it defaults to "scsi_host". To keep backwards
compatibility, this attribute is optional only for the
"scsi_host" adapter, but is mandatory for the "fc_host" adapter.
Since 1.0.5
A "fc_host" capable scsi_hostN can be determined by using
virsh nodedev-list --cap fc_host.
Since 1.2.8
Note: Regardless of whether a "scsi_host" adapter type is defined
using a name or a parentaddr, it
should refer to a real scsi_host adapter as found through a
virsh nodedev-list scsi_host and virsh
nodedev-dumpxml scsi_hostN on one of the scsi_host's
displayed. It should not refer to a "fc_host" capable scsi_hostN
nor should it refer to the vHBA created for some "fc_host"
adapter. For a vHBA the nodedev-dumpxml
output parent setting will be the "fc_host" capable scsi_hostN
value. Additionally, do not refer to an iSCSI scsi_hostN for the
"scsi_host" source. An iSCSI scsi_hostN's
nodedev-dumpxml output parent field is generally
"computer". This is a libvirt created parent value indicating
no parent was defined for the node device.
wwnn and wwpnwwnn) and
"World Wide Port Name" (wwpn) are used by the
"fc_host" adapter to uniquely identify the vHBA device in the
Fibre Channel storage fabric. If the vHBA device already exists
as a Node Device, then libvirt will use it; otherwise, the vHBA
will be created using the provided values. It is considered a
configuration error use the values from the HBA as those would
be for a "scsi_host" type pool instead. The
wwnn and wwpn have very specific
format requirements based on the hypervisor being used, thus
care should be taken if you decide to generate your own to
follow the standards; otherwise, the pool will fail to start
with an opaque error message indicating failure to write to
the vport_create file during vport create/delete due to
"No such file or directory".
Since 1.0.4
parentparent_wwnn and parent_wwpnparent to specify which scsi_host
to use by name, it's possible to provide the wwnn and wwpn of
the parent to be used for the vHBA in order to ensure that
between reboots or after a hardware configuration change that
the scsi_host parent name doesn't change. Both the parent_wwnn
and parent_wwpn must be provided.
Since 3.0.0
parent_fabric_wwnparent to specify which scsi_host
to use by name, it's possible to provide the fabric_wwn on which
the scsi_host exists. This provides flexibility for choosing
a scsi_host that may be available on the fabric rather than
requiring a specific parent by wwnn or wwpn to be available.
Since 3.0.0
managedparentaddrname attribute to more uniquely identify the
SCSI host. Using a combination of the unique_id
attribute and the address element to formulate
a PCI address, a search will be performed of the
/sys/class/scsi_host/hostNN links for a
matching PCI address with a matching unique_id
value in the /sys/class/scsi_host/hostNN/unique_id
file. The value in the "unique_id" file will be unique enough
for the specific PCI address. The hostNN will be
used by libvirt as the basis to define which SCSI host is to
be used for the currently booted system.
Since 1.2.7
addressdomain (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). The PCI address can be determined by
listing the /sys/bus/pci/devices and the
/sys/class/scsi_host directories in order to
find the expected scsi_host device. The address will be
provided in a format such as "0000:00:1f:2" which can be
used to generate the expected PCI address
"domain='0x0000' bus='0x00' slot='0x1f' function='0x0'".
Optionally, using the combination of the commands 'virsh
nodedev-list scsi_host' and 'virsh nodedev-dumpxml' for a
specific list entry and converting the resulting
path element as the basis to formulate the
correctly formatted PCI address.
unique_idparentaddr attribute used to determine
which of the scsi_host adapters for the provided PCI address
should be used. The value is determine by contents of the
unique_id file for the specific scsi_host adapter.
For a PCI address of "0000:00:1f:2", the unique identifier files
can be found using the command
find -H /sys/class/scsi_host/host*/unique_id |
xargs grep '[0-9]'. Optionally, the
virsh nodedev-dumpxml scsi_hostN' of a
specific scsi_hostN list entry will list the
unique_id value.
hostnetfs, iscsi,
iscsi-direct,
rbd, sheepdog, gluster). Will be
used in combination with a directory
or device element. Contains an attribute name
which is the hostname or IP address of the server. May optionally
contain a port attribute for the protocol specific
port number. Duplicate storage pool definition checks may perform
a cursory check that the same host name by string comparison in the
new pool does not match an existing pool's source host name when
combined with the directory or device
element. Name resolution of the provided hostname or IP address
is left to the storage driver backend interactions with the remote
server. See the storage driver page for
any restrictions for specific storage backends.
Since 0.4.1initiatoriscsi-direct pool in order to provide
the iSCSI Qualified Name (IQN) to communicate with the pool's
device target IQN. There is one sub-element
iqn with the name attribute to describe
the IQN for the initiator.
Since 4.7.0authauth element provides the
authentication credentials needed to access the source by the
setting of the type attribute (pool
types iscsi, iscsi-direct, rbd).
The type
must be either "chap" or "ceph". Use "ceph" for
Ceph RBD (Rados Block Device) network sources and use "iscsi" for CHAP
(Challenge-Handshake Authentication Protocol) iSCSI
targets. Additionally a mandatory attribute
username identifies the username to use during
authentication as well as a sub-element secret with
a 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 manages the password.
The secret element requires 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. Since 0.9.7 for "ceph" and
1.1.1 for "chap"
namelogical, rbd,
sheepdog, gluster). Contains a
string identifier.
Since 0.4.5formatfs, netfs, disk,
logical). This
contains a single attribute type whose value is
backend specific. This is typically used to indicate filesystem
type, or network filesystem type, or partition table type, or
LVM metadata type. All drivers are required to have a default
value for this, so it is optional. Since 0.4.1protocolnetfs Storage Pool provide a mechanism to
define which NFS protocol version number will be used to contact
the server's NFS service. The attribute ver accepts
an unsigned integer as the version number to use.
Since 5.1.0vendorname whose value is backend
specific. Since 0.8.4productname whose value
is backend specific. Since 0.8.4
A single target element is contained within the top level
pool element for some types of pools (pool
types dir, fs, netfs,
logical, disk, iscsi,
scsi, mpath, zfs).
This tag is used to describe the mapping of
the storage pool into the host filesystem. It can contain the following
child elements:
...
<target>
<path>/dev/disk/by-path</path>
<permissions>
<owner>107</owner>
<group>107</group>
<mode>0744</mode>
<label>virt_image_t</label>
</permissions>
</target>
</pool>
path/dev/ may seem
like the logical choice, however, devices nodes there are not
guaranteed stable across reboots, since they are allocated on
demand. It is preferable to use a stable location such as one
of the /dev/disk/by-{path|id|uuid|label} locations.
For logical and zfs pool types, a
provided value is ignored and a default path generated.
For a Multipath pool (type mpath), the provided
value is ignored and the default value of "/dev/mapper" is used.
Since 0.4.1
permissionsmode element contains the octal permission set.
The mode defaults to 0711 when not provided.
The owner element contains the numeric user ID.
The group element contains the numeric group ID.
If owner or group aren't specified when
creating a directory, the UID and GID of the libvirtd process are used.
The label element contains the MAC (eg SELinux)
label string.
Since 0.4.1
For running directory or filesystem based pools, these fields
will be filled with the values used by the existing directory.
Since 1.2.16
If a storage pool exposes information about its underlying
placement / allocation scheme, the device element
within the source element may contain information
about its available extents. Some pools have a constraint that
a volume must be allocated entirely within a single constraint
(eg disk partition pools). Thus the extent information allows an
application to determine the maximum possible size for a new
volume
For storage pools supporting extent information, within each
device element there will be zero or more freeExtent
elements. Each of these elements contains two attributes, start
and end which provide the boundaries of the extent on the
device, measured in bytes. Since 0.4.1
The optional refresh element can control how the pool and
associated volumes are refreshed (pool type rbd). The
allocation attribute of the volume child element
controls the method used for computing the allocation of a volume. The
valid attribute values are default to compute the actual
usage or capacity to use the logical capacity for cases where
computing the allocation is too expensive. The following XML snippet
shows the syntax:
<pool type="rbd">
<name>myrbdpool</name>
...
<source/>
...
<refresh>
<volume allocation='capacity'/>
</refresh>
...
</pool>
Since 5.2.0
Usage of Storage Pool Namespaces provides a mechanism to provide pool type specific data in a free form or arbitrary manner via XML syntax targeted solely for the needs of the specific pool type which is not otherwise supported in standard XML. For the "fs" and "netfs" pool types this provides a mechanism to provide additional mount options on the command line. For the "rbd" pool this provides a mechanism to override default settings for RBD configuration options.
Usage of namespaces comes with no support guarantees. It is intended for developers testing out a concept prior to requesting an explicitly supported XML option in libvirt, and thus should never be used in production.
fs:mount_optsfs or netfs type storage
pools. In order to designate that the Storage Pool will be using
the mechanism, the pool element must be modified to
provide the XML namespace attribute syntax as follows:
xmlns:fs='http://libvirt.org/schemas/storagepool/fs/1.0'
The fs:mount_opts defines the mount options by
specifying multiple fs:option subelements with
the attribute name specifying the mount option to
be added. The value of the named option is not checked since
it's possible options don't exist on all distributions. It is
expected that proper and valid options will be supplied for the
target host.
<pool type="netfs" xmlns:fs='http://libvirt.org/schemas/storagepool/fs/1.0'>
<name>nfsimages</name>
...
<source>
...
</source>
...
<target>
...
</target>
<fs:mount_opts>
<fs:option name='sync'/>
<fs:option name='lazytime'/>
</fs:mount_opts>
</pool>
...
Since 5.1.0.rbd:config_optsrbd type
storage pools. In order to designate that the Storage Pool
will be using the mechanism, the pool element
must be modified to provide the XML namespace attribute
syntax as follows:
xmlns:rbd='http://libvirt.org/schemas/storagepool/rbd/1.0'
The rbd:config_opts defines the configuration options
by specifying multiple rbd:option subelements with
the attribute name specifying the configuration option
to be added and value specifying the configuration
option value. The name and value for each option is only checked
to be not empty. The name and value provided are not checked since
it's possible options don't exist on all distributions. It is
expected that proper and valid options will be supplied for the
target host.
<pool type="rbd" xmlns:rbd='http://libvirt.org/schemas/storagepool/rbd/1.0'>
<name>myrbdpool</name>
...
<source>
...
</source>
...
<target>
...
</target>
...
<rbd:config_opts>
<rbd:option name='client_mount_timeout' value='45'/>
<rbd:option name='rados_mon_op_timeout' value='20'/>
<rbd:option name='rados_osd_op_timeout' value='10'/>
</rbd:config_opts>
</pool>
Since 5.1.0.
A storage volume will generally be either a file or a device
node; since 1.2.0, an optional
output-only attribute type lists the actual type
(file, block, dir, network, netdir or ploop), which is also available
from virStorageVolGetInfo(). The storage volume
XML format is available since 0.4.1
<volume type='file'> <name>sparse.img</name> <key>/var/lib/xen/images/sparse.img</key> <allocation>0</allocation> <capacity unit="T">1</capacity> ...
namesource device path
device and next partition number to be created. For example, if
the source device path is /dev/sdb and there are no
partitions on the disk, then the name must be sdb1 with the next
name being sdb2 and so on.
Since 0.4.1keyallocationlogical
pool will not automatically expand volume's allocation when it
gets full; the user is responsible for doing that or configuring
dmeventd to do so automatically.unit can be specified to adjust the passed value.
Values can be: 'B' or 'bytes' for bytes, 'KB' (kilobytes,
103 or 1000 bytes), 'K' or 'KiB' (kibibytes,
210 or 1024 bytes), 'MB' (megabytes, 106
or 1,000,000 bytes), 'M' or 'MiB' (mebibytes, 220
or 1,048,576 bytes), 'GB' (gigabytes, 109 or
1,000,000,000 bytes), 'G' or 'GiB' (gibibytes, 230
or 1,073,741,824 bytes), 'TB' (terabytes, 1012 or
1,000,000,000,000 bytes), 'T' or 'TiB' (tebibytes,
240 or 1,099,511,627,776 bytes), 'PB' (petabytes,
1015 or 1,000,000,000,000,000 bytes), 'P' or 'PiB'
(pebibytes, 250 or 1,125,899,906,842,624 bytes),
'EB' (exabytes, 1018 or 1,000,000,000,000,000,000
bytes), or 'E' or 'EiB' (exbibytes, 260 or
1,152,921,504,606,846,976 bytes). Since
0.4.1, multi-character unit since
0.9.11capacityunit attribute can be
specified with the same semantics as for allocation
This is compulsory when creating a volume.
Since 0.4.1physicalunit
will be in bytes.
Since 3.0.0sourcetarget
A single target element is contained within the top level
volume element. This tag is used to describe the mapping of
the storage volume into the host filesystem. It can contain the following
child elements:
...
<target>
<path>/var/lib/virt/images/sparse.img</path>
<format type='qcow2'/>
<permissions>
<owner>107</owner>
<group>107</group>
<mode>0744</mode>
<label>virt_image_t</label>
</permissions>
<timestamps>
<atime>1341933637.273190990</atime>
<mtime>1341930622.047245868</mtime>
<ctime>1341930622.047245868</ctime>
</timestamps>
<encryption type='...'>
...
</encryption>
<compat>1.1</compat>
<nocow/>
<clusterSize unit='KiB'>64</clusterSize>
<features>
<lazy_refcounts/>
</features>
</target>
pathformattype attribute. Consult the
storage driver page for the list of valid
volume format type values for each specific pool. The
format will be ignored on input for pools without a
volume format type value and the default pool format will be used.
Since 0.4.1permissionsmode element contains the octal permission set.
The mode defaults to 0600 when not provided.
The owner element contains the numeric user ID.
The group element contains the numeric group ID.
If owner or group aren't specified when
creating a supported volume, the UID and GID of the libvirtd process
are used. The label element contains the MAC (eg SELinux)
label string.
For existing directory or filesystem based volumes, these fields
will be filled with the values used by the existing file.
Since 0.4.1
timestampsatime, btime, ctime
and mtime hold the access, birth, change and
modification time of the volume, where known. The used time
format is <seconds>.<nanoseconds> since the
beginning of the epoch (1 Jan 1970). If nanosecond resolution
is 0 or otherwise unsupported by the host OS or filesystem,
then the nanoseconds part is omitted. This is a readonly
attribute and is ignored when creating a volume.
Since 0.10.0
encryptioncompattype='qcow2' volumes. Valid values are 0.10
and 1.1 so far, specifying QEMU version the images should
be compatible with. If the feature element is present,
1.1 is used.
Since 1.1.0 If omitted, 0.10 is used.
Since 1.1.2
nocowclusterSizefeaturesqcow2 now.
Valid sub-elements are:
<lazy_refcounts/> - allow delayed reference
counter updates. Since 1.1.0
A single backingStore element is contained within the top level
volume element. This tag is used to describe the optional copy
on write, backing store for the storage volume. It can contain the following
child elements:
...
<backingStore>
<path>/var/lib/virt/images/master.img</path>
<format type='raw'/>
<permissions>
<owner>107</owner>
<group>107</group>
<mode>0744</mode>
<label>virt_image_t</label>
</permissions>
</backingStore>
</volume>
pathformatpermissionspermissions documentation for explanation
of individual fields.
Since 0.6.0
Here are a couple of examples, for a more complete set demonstrating every type of storage pool, consult the storage driver page
<pool type="dir">
<name>virtimages</name>
<target>
<path>/var/lib/virt/images</path>
</target>
</pool>
<pool type="iscsi">
<name>virtimages</name>
<source>
<host name="iscsi.example.com"/>
<device path="iqn.2013-06.com.example:iscsi-pool"/>
<auth type='chap' username='myuser'>
<secret usage='libvirtiscsi'/>
</auth>
</source>
<target>
<path>/dev/disk/by-path</path>
</target>
</pool>
<volume>
<name>sparse.img</name>
<allocation>0</allocation>
<capacity unit="T">1</capacity>
<target>
<path>/var/lib/virt/images/sparse.img</path>
<permissions>
<owner>107</owner>
<group>107</group>
<mode>0744</mode>
<label>virt_image_t</label>
</permissions>
</target>
</volume>
<volume>
<name>MyLuks.img</name>
<capacity unit="G">5</capacity>
<target>
<path>/var/lib/virt/images/MyLuks.img</path>
<format type='raw'/>
<encryption format='luks'>
<secret type='passphrase' uuid='f52a81b2-424e-490c-823d-6bd4235bc572'/>
</encryption>
</target>
</volume>