Persistent attach of device with varying backend name

I am trying to persistently attach a device which has a varying backend name.

This is the guide that I followed to get the backend name

It returns something like “dm-85”, but it changes with each boot, so I cant do qvm-block attach -p vmname dom0:dm-85 since it won’t exist on next boot.

I can script it to attach it on boot, but then if I restart a VM it won’t be attached anymore.

What could be some solutions?

Could you use cron or systemd to periodically check if the VM is running, and if it is, to then attach the device?

Use libxl hook as I’ve suggested in another topic:

sudo mkdir /etc/libvirt/hooks/
cat << 'EOF' | sudo tee /etc/libvirt/hooks/libxl >/dev/null

if [ "$guest_name" = "yourqubename" ] && [ "$libvirt_operation" = "started" ]; then
        exec 0</dev/null
        exec 1>/dev/null
        exec 2>/dev/null
        dev=$(basename $(readlink /dev/YOUR_LVM_VG/YOUR_LVM_IMAGE))
        if ! qvm-block attach "$guest_name" dom0:$dev; then
            qubesdb-write /qubes-block-devices/$dev/desc "YOUR_LVM_IMAGE"
            qvm-block attach "$guest_name" dom0:$dev
    ) & disown
sudo chmod +x /etc/libvirt/hooks/libxl

Thank you, works like a charm

Some tips for future readers, to debug you can restart libvirtd manually instead of rebooting completely, systemctl restart libvirtd

exit 0 as quickly as possible if it’s not your target operation/VM

the subshell and redirection of output seems to prevent locking/hanging, if it’s not used the VM wouldn’t boot properly, probably for the same reason a terminal might hang until you hit cancel. And if that happens you can barely kill the VM because it’s in a locked state

look for alternative faster qvm- tools like qubesdb-read rather than qubes-block ls, any time wasted in the hook is time lost on boot

Some useful links here

My script is highly personalized with personal device names and VM names, so not useful for the public, but I hope this helps someone else solve the same issue :slight_smile: