This series of patch is the minimal required to get a working proof
of concept implementation of fine grained access control in libvirt.
This demonstrates
- Obtaining a client identity from a socket
- Ensuring RPC calls are executed with the correct identity sset
- A policykit access driver that checks based on access vector alone
- A SELinux access driver that checks based on access vector + object
- A set of hooks in the QEMU driver to protect virDomainObjPtr access
Things that are not done
- APIs for changing the real/effective identity post-connect
- A simple RBAC access driver for doing (Access vector, object)
checks
- SELinux policy for the SELinux driver
- Access control hooks on all other QEMU driver methods
- Access control hooks in LXC, UML, other libvirtd side drivers
- Access control hooks in storage, network, interface, etc drivers
- Document WTF todo to propagate SELinux contexts across TCP
sockets using IPSec. Any hints welcome...
- Lots more I can't think of right now
I should note that the policykit driver is mostly useless because it
is unable to let you do checks on anything other than permission name
and UNIX process ID at this time. In theory it is supposed to be
extendable to allow other types of identity information besides
the process ID, and to include some kind of object identiers in
the permission check, but no one seems to be attacking this.
So I expect the simple RBAC driver to be the most used one in the
common case usage of libvirt