XCP storage driver domains

This page describes how to configure a storage driver domain on XCP 1.6. This is experimental software: please send feedback to xen-api@lists.xen.org.

Create a suitable VM

Install a VM in the usual way. A storage driver domain can be PV or HVM. A storage driver domain must have:

• the blkback kernel module
• the XCP block udev scripts (available in RPM and .deb form)
• access to the storage medium (via the network or via a PCI passthrough device)

The blkback kernel module

Check you have the "xen-blkback" module (Linux) and "modprobe" it if its missing.

On Ubuntu 12.10, try

   echo xen-blkback >> /etc/modules

The XCP block udev scripts

These are needed to signal to the toolstack when a disk has been "attached". In Ubuntu the scripts are normally installed via

   apt-get install xen-utils-common

PCI passthrough

Use "lspci" in dom0 to identify the storage controller you want to pass through. Take a note of the "BBBB:DD.F" identification string. Note that unless a multi-function device supports "Function Level Reset" on individual Functions you must pass all functions together as a group.

Use the "xe" CLI to associate this PCI device with your driver VM:

   xe vm-param-set uuid=<VM uuid> other-config:pci=0/BBBB:DD.F,1/BBBB:DD.F

If you shutdown and then restart the VM (not reboot), you should see the device appear in the driver VM with "lspci" while in dom0 the device should be associated with "pciback".

Add a VIF to the "internal management network"

Shut down the driver VM and execute in dom0:

   NET=$(xe network-list other-config:is_host_internal_management_network=true params=uuid --minimal)
   xe vif-create vm-uuid=<my VM uuid> network-uuid=$NET device=0

This will add a special VIF connecting the driver VM to the "host internal management" network.

Install the storage control plane software

Inside the driver VM run:

   git clone https://github.com/djs55/dbus-test
   cd dbus-test
   sudo make install

This will add a python program (/usr/bin/xcp-sm-fs)

It's a good idea to arrange for this program to start automatically when the VM starts, for example by running it from /etc/rc.local There's a config file (/etc/xcp-sm-fs.conf) which has some tweakable settings. Check the VM is binding to port 80 on eth0 using "netstat" (or adjust the configuration if you're using a different interface)

You also need to install some generic storage tools:

   apt-get install blktap-utils nfs-common

Create a directory to contain the disk images

Inside the driver domain:

   sudo mkfs.ext3 /dev/myblockdevice
   sudo mkdir /interestingstorage

Add the filesystem to /etc/fstab so that it is mounted after every boot.

Attach an SR via the driver domain

In dom0, "introduce" an SR:

   SR=$(uuidgen)
   xe sr-introduce uuid=$SR name-label=driverdomain type=ext
   HOST=$(xe host-list params=uuid --minimal)
   PBD=$(xe pbd-create sr-uuid=$SR host-uuid=$HOST device-config:path=/interestingstorage)

Next, associate this host's connection ("PBD") with the driver domain:

   xe pbd-param-set uuid=$PBD other-config:storage_driver_domain=<driver domain uuid>

Finally, try attaching the storage:

   xe pbd-plug uuid=$PBD

This should cause the VM to be started and the storage control plane software to be queried.

Create a VM using the new storage

In dom0 create a blank VM from a template e.g:

   WIN=$(xe vm-install template="Windows XP SP3 (32-bit)" new-name-label=winxp sr-uuid=$SR)

Tweak the default configuration to allow qemu to find the disk device:

   xe vm-param-set uuid=$WIN platform:qemu_disk_cmdline=true

The rest of the VM install should proceed as normal: try inserting a CD via a GUI like XenCenter.