This can cause some rules to fail and eg remove dm-* devices. Replace it with
what is really needed to hide mounted (and other ignored) devices from
qubes-block-devices.
When device is ejected by some VM (state=6, effectively inactive), it should be
removed from xenstore to free slot for some another device. This should be done
by libxl toolstack, but not implemented in xen 4.1 - AFAIR done in xen 4.2.
Create lockfile to set it proper permissions. Without it the first use
(qvm-start netvm) will create it with root:root and 600.
Without xend, no one sets dom0 name...
This is core part of migration. Things not migrated yet:
- DispVM (qubes_restore needs to be almost rewritten)
- VM xen config files should be fixed (use "script:" prefix in block device description, perhaps generate this files on VM start)
Huge, slow xend not needed any more, now it conflicts with libxl
We want to set up a default dvm if needed at boot time; for this,
the default netvm must have been already started. Therefore, we
need a qubes_dvm script, that executes after qubes_netvm.
Make meminfo-writer compute used memory, and report to qmemman only if
it has changed significantly enough. As it is written in C, its code is
much faster that qmemman-server; also in the idle case, it saves on xenstore
communication overhead. Allows to send updates up to 10 times per second,
with CPU load on the VM below 0.1%.
Instead of hardcoded savefile name, use a symlink in
/var/run/qubes. Tools should set this symlink to a correct
savefile. Also, test whether the savefile is older than the
template root.img.
