Re: Task destruction

[Roland McGrath]

In the L4 context I think it is reasonable to have two options for how the
sub-hurd case can be.  First, one should be able to have one's own task
server with a whole universe of tasks that does not exist as far as the
outer system is concerned.  This is a microkernel analog closer to having a
virtual machine.  Second, one should be able to use the task server
managing one's own task to make new unregistered tasks that are accessible
by others (mediated by proc servers or whatever access controls the task
server has) analogous to Mach native tasks.  The latter is what we have on
Mach, and is convenient in that you can run gdb on the outer system on an
individual task in the inner system and ps and so forth.  Both modes should
be usable by an unprivileged user, with the proc server understanding
what's going on enough to grant task control access to an appropriate set
of requesters.