[tor-talk] possible to identify tor user via hardware DRM?

[Seth David Schoen]

Nick M. Daly writes:

> I've actually been reading similar concerns (about hardware-based
> tracking) on the FreedomBox list, and I have no basis for validating the
> claims.  Wondering if anybody could shed some light on just how serious
> these claims are and how they'd impact Tor users (or privacy generally)?
> The mail seems full of generalizations, but the author seems genuine:
> 
> http://lists.alioth.debian.org/pipermail/freedombox-discuss/2012-June/004049.html

I find this message misleading in various ways.  The basic thing that
I've been telling people is that there are few situations in which
either PSN or TPM uniqueness makes things qualitatively worse.

There are lots of hardware unique IDs.  On Linux, try "sudo lshw" and
be surprised at all the things that have unique serial numbers.  There
are also things that are unique about your machine that are not
hardware serial numbers, like filesystem serial numbers and observed
combinations of software configurations.

These can be bad for privacy because software can tell which computer
it's running on.  If the software has an adversarial relationship with
you, it can then use that information in a way that you don't like.
We would be better off in some regards if operating systems let us
hide local uniqueness from software so that the software couldn't tell
what machine it was running on, or set fake values for these unique
identifiers.

Some proprietary software including Microsoft Windows already makes
a sophisticated profile of the local machine, including many kinds of
observations, to tie a copy of the software (or an "activation") to a
particular device (!).

The only substantive difference with the TPM uniqueness is that the TPM
uniqueness lets you prove (like a smartcard) to a remote system that
you're running some software on the same machine as before.  Even if
the OS did let you set fake values when software tried to examine the
system it was running on, the remote system could see that the
TPM-related values were fake.  That's useful for some applications,
including but not limited to DRM-like ones.

I've argued that this is bad in some ways, but at least you can still
turn off the TPM.  Then your system can't attempt to offer that kind
of proof.  As far as I know, turning off the TPM is pretty robust:
it really is turned off.

All of these things are anonymity problems in particular when some
software on your computer is actively _trying_ to tell someone else
what machine you're running on, either because it's programmed to do
so or because someone has broken into your computer and installed
spyware and is trying to use it to monitor you.  If you're not in
that situation, there is nothing especially magical about having
unique hardware IDs in your machine, because everyone's machine has
some uniqueness, and (for the most part) that uniqueness isn't part
of standard network protocols like TCP/IP and doesn't automatically
leak out to anyone and everyone you communicate with over the
Internet.  (There is a possible exception about clock skew, which you
can read about in Steven Murdoch's paper from 2006.)

Similarly, having a GPS receiver in your phone does not mean that
everyone you send an SMS to or everyone you call will learn your
exact physical location.  However, it does mean that if there's
spyware on your phone, that spyware is able to use the GPS to learn
your location and leak it.  If you're worried about spyware threats
on your phone, which can be quite a realistic concern, the GPS
itself isn't necessarily the unique core of the threat, because
there are also lots of other things in the phone that can be read to
help physically locate you (like wifi base station MAC addresses,
taking photographs of your surroundings with the phone's camera,
recording the identities and signal strengths of the GSM base
stations your phone sees...).  So a more fundamental question might
be whether your phone operating system is able to either prevent
you from getting malware or prevent the malware from accessing the
sensors on your phone.

In the case of a desktop PC, the hardware uniqueness is _there to
be read by software_, and if it's in a TPM it _may be able to give
the software remotely verifiable cryptographic proof that the
software is really running on the machine containing that particular
TPM_.  In neither case does the hardware uniqueness directly
broadcast itself to other machines, and in neither case does the
hardware uniqueness prevent the operating system from preventing
other software from reading it.

If you do have some kind of software running on your machine that's
trying to track you or trying to help other people track you,
hardware uniqueness is one thing that the software might look at.
But if you're a Tor user, a more basic thing for the software to
try to do is make network connections to leak your real IP address
in order to associate your Tor-based network activity with your
non-Tor-based network activity.  That might be even easier because
the tracking software could just try to make a direct network
connection.

If you're not using Tor, at least not at a particular moment, but
are still concerned about tracking, there's another problem, which
is that all existing browsers _already_ reveal a great deal of
software-based uniqueness to any interested web site, usually enough
to make your browser unique.  See

https://panopticlick.eff.org/browser-uniqueness.pdf

This is important because it doesn't require there to be any
malicious software on your computer, just a traditional web browser.

One of the defenses people have talked about against hardware
fingerprinting is running inside a virtual machine.  Normally,
software inside the virtual machine, even if it's malicious,
doesn't learn much about the physical machine that hosts the VM.
If you always use Tor inside a VM, then even if there's a bug
that lets someone take over your computer (or if they trick you
into installing spyware), the malicious software won't be able
to read much real uniqueness from the host hardware, unless
there's also a bug in the VM software.

Running in a VM isn't exactly a defense against software
fingerprinting (like browser fingerprinting) if you use the VM
for various non-Tor activities that you don't want to be linked
to one another, because the software configuration inside the VM
might be, or become, sufficiently different from others that it
can be recognized.  There's probably more research to be done
about the conditions under which VMs can be uniquely identified
both "from the inside" by malware, and remotely by remote
software fingerprinting, absent VM bugs that give unintended
access to the host.

-- 
Seth Schoen  <schoen at eff.org>
Senior Staff Technologist                       https://www.eff.org/
Electronic Frontier Foundation                  https://www.eff.org/join
454 Shotwell Street, San Francisco, CA  94110   +1 415 436 9333 x107