Re: A _good_ and valid use for TPM

[phcoder]

> For some reason he wants to store the data encrypted in multiple
> locations rather than using a simple terminal to retreive the data
> over network which makes things needlessly hard.
He perhaps needs important amount of computing power. And in his case 
"all in centre" may require too much bandwidth
> Now I am not sure how secure this solution is. You can usually remove
> the battery to reset BIOS password, reflash the BIOS, etc.
Many boards save the data in flash memory so removing power won't reset 
password. Second flash chip if it's dedicated can be covered with 
concrete too and resetting pins can be removed. Besides with coreboot 
everything this can be well controlled - you can embed the config to flash.
> Since manufacturers claim (or used to) that you can pry the TPM chip
> off your board and it will still work the board is bootstrapped by the
> main CPU, not the TPM. This makes it possible to short some pins on
> the TPM chip so that is cannot be accessed during boot, boot a virtual
> machine, and have the BIOS initialize the chip inside that.
It would require some modifications to virtual machine to skip some 
initilisation but is entirely possible and needs to be done only once to 
cover 99% of motherboards
> There's also the possibility to remove the RAM from a running computer
> given you find out what kind of RAM it uses and get a different
> compatible computer.
concrete :)
> Generally this shifts the attack from the realm of plain vandalism to
> the realm of planned attack which is certainly a bonus.
>
> Still I would rather rely on a custom solution because I would know
> exactly what it does. The manufacturers of PC mainboards tend to not
> release exact specifications and there are often serious problems.
>
> Still finding the flaw in the particular mainboard would probably take
> some non-trivial effort.
There are only few kinds of tpm chips so it's enough that someone cracks 
 the corresponding ship to make the attack trivial. As a matter of fact 
few year from now it may be easier to get a universal reader for all tpm 
chips then a reader for a specific flash chip
> If the attacker just wants to break something there would likely be
> easier targets. If you are specifically targeted you are doomed.
Yes. Once an attacker has the device he is able to retrieve all the data 
in. Only putting physical obstacles may slow the attacker down. And I 
doubt that a cost of such operation can be over $10000 no matter what 
protection you use.
> Now to the TPM support in GRUB.
>
> This makes the TPM support debate seem quite pointless.
It isn't. Supporing tpm may help it becoming widespread, commonplace and 
acceptable, exactly what we try to avoid

Regards
Vladimir 'phcoder' Serbinenko