Thread: [tboot-devel] Intel's P-MAPS research project

SourceForge Headquarters 1320 Columbia Street Suite 310 San Diego, CA 92101 +1 (858) 422-6466

I recently learned about Intel's P-MAPS research project which
provides an alternative way of using TPM+TXT to provide attestations
and sealing in the context of a standard OS. Here is a link to the
Intel Research blog post:

http://blogs.intel.com/research/2009/04/p-maps_an_on-demand_hardware-r.php

and here is an article in Dr Dobbs Journal which goes into more detail:

http://www.ddj.com/mobile/218401423

The goal is to allow applications running in a standard OS like Linux
or Windows to be able to gain hardware protection from corruption of
other processes or of the OS. This is a hard problem to solve due to
the complexity of modern OS's. P-MAPS bypasses the OS by loading a
Measured Virtual Machine Monitor (MVMM) which runs the OS as a VM.
Then a P-MAPS aware application can make special VM calls directly
into P-MAPS, going around the OS, to request protection. P-MAPS
monitors and virtualizes the OS's page tables and is able to protect
all of the application's pages from rogue access, either from the OS
or other processes.

Because P-MAPS mostly confines its attention to memory management, it
can be relatively small for a VMM. It doesn't have to worry about
virtualizing devices or networks or I/O or having to load lots of
different drivers. It mostly just manages page tables. This means that
the OS is removed from the Trusted Computing Base (TCB) which greatly
reduces the amount of code which has to be correct in order to achieve
security.

P-MAPS is also able to perform attestation ("Quote") and sealing on
behalf of protected applications, allowing apps to protect secrets
from other applications and from the OS, and to attest to outside
parties that their data is safe.

Among other nice features, P-MAPS uses smart loading, such that when
no applications are currently requesting P-MAPS services, it unloads
itself completely and switches the OS from being in a VM back to being
in a normal, non-virtualized mode. Then when a process requests P-MAPS
protection, it re-virtualizes the OS, including doing a TXT launch of
the P-MAPS MVMM.

All in all this sounds like an amazing range of functionality, a real
tour de force to get all of these technologies (TPM, TXT, VM) working
together successfully. But the net result is a tremendously useful
package that neatly bypasses the dilemma of security vs complexity.
Most solutions today either provide potentially high security with
relatively limited functionality, like Jon McCune's Flicker, or
provide a much wider set of functions, like TBOOT+XEN, at the expense
of a large TCB which inherently undercuts security goals. P-MAPS
appears to be the first solution I've seen that could provide high
security via a small TCB, while retaining the functionality provided
by a standard OS.

Unfortunately, as a research project it does not sound like something
which is likely to be made available to experimenters any time soon. I
hope Intel will find a way to make the code available as it has done
with TBOOT. P-MAPS is IMO even better suited as a framework for
providing meaningful TXT based protections to today's application
developers.

Hal Finey