Slashdot Mirror
IBM On Trusted Computing, Linux
An anonymous reader writes "A number of IBM's computers have been available with an "embedded security subsystem (ESS)" for some time now. This site lists three research papers regarding the new TCPA (Trusted Computing Platform Initiative) security chip developed by IBM, including the full GPL-ed source code to a Linux driver for this chip. In particular, the 'Why TCPA?' paper claims that IBM's TCPA chip is in fact of extremely limited use for DRM, as it contains no tamper resistance; the chip is designed to fend off software attacks, not physical attacks. An interesting take from a company with very solid products."
~~~
"Trust the computer. The computer is your friend." :)
- bumpersticker seen on car in Microsoft parking lot
These are absolutely terrific articles. Their distribution of an open source TCPA linux module satisfies a lot of concerns and questions many of us had about TCPA in a concrete and specific manner.
One concern still exists: that DRM and Palladium will be used to create a "mainstream" set of M$ applications which give people the illusion of security, while concentrating most of the information and control in the hands of the few.
The most important step people in the open source community can take next are to get a system with a TCPA chip and start developing drivers, firewall systems, proxies and applications that make good solid use of the technology: tsshd, tsquid, tsftp, thttpsd, tbsd, toggd, tnamed, texim, tkonq...
Ignore any threat of the local attack, the remote attack is the important one.
Watch out with that line of thinking... The ideal system has reasonable internal security as well. If a disgruntled employee can get access to these public/private key pairs, you're worse off than before, because you still maintain the illusion of security.
... let's just, for a moment, cast aside paranoid suspicion (and I'm a paranoid & suspicious chap!). IFF these papers are correct TCPA is an encrypted storage location with some extra logic. In this location the user can store ~/.ssh/*.key and make shure the application interacting with the logic isn't sniffing the un-encrypted stream to some remote location. This NEEDS to be embedded in the BIOS to prevent kernel backdooring and simply embeds chain of trust throughout the hardware. I'd like to see this chip bussed to a smartcard to authenticate private root keys to a hashed ssh-agent binary (whether roaming on different PCs or on my own WS...)
I'm also shure that MediaPlayer 10 will be DRMd to the marrow but take note that in the past ridiculously encumbered online music services went titsup in no time while more reasonable services (Apple) seem to strke a balance.
In the past ID tracking such as the PentiumIII ID were dealt with properly so I don't think abuses would be tolerated. People always enjoys the empowering thought of having the option to take a free ride and imposing a "police" computer would vastly outrage the consumer base.
So long as the control on the hadware keys is left to the users I agree with this particular spin from IBM; it's just a secure smartcard system.
It still CAN be extended to require encryption and trust all the way to the DVI interface but that I think would require a heck of a business infrastructure to implement, maintain and persuasion effort.
And given IBM's perspective there's no interest in the user base to proceed in further HW lockdown... all WE would do is to sign on OUR terms a kernel build and that's it; once that's in place, the chip will process OUR keys in OUR best interest... and if some pigopolist wants force something down our throat their business model will fail (as it has repeatedly done).
I'd go for it, just for the sake of my ssh/gpg keyring, and in the future credit card numbers... do you trust an ecommerce site asking to handle it for you?)
Mi domando chi à il mandante di tutte le cazzate che faccio - Altan
There ability to be attach locally is like cable modems firmware chips. Descused on CableModemHack.com (A website of tools to uncap your cable modem), an effort to replace the firmware locally is underway for a lot of models of cable modems. It seems that cable modems are wonderful against software attacks, but very open to hardware attacks.
Hardware attacks, I guess, are not a common senerio that hardware designers really think much about.
No.
Is there a private key that third parties know that it is impossible for the owner of the computer to know?
The paper makes it sound like all key pairs are either randomly generated or that the chip can be fed a public key. However it is a bit vague, and I suspect the answer is that there are also non-random pairs in there, where third parties know the private key but you don't. They skirt around this by saying "Bios startup is quite complex" but I think the real answer is that there unless hashes have matched up to a point these secret public keys are inaccessible.
This system is absolutely useless for security as all exploits actually cause supposedly correct programs to follow the wrong instructions. This is like claiming current systems are secure because you cannot change the microcode and invent new machine instructions. It's purpose is so that it is impossible to get any kind of modified or different operating system in there, and still be able to run DRM programs, which could decode information using the secret key.
The fact that IBM and everybody else has refused to answer this question (I think the answer here was skirted around with some bullshit about the "BIOS startup being quite complex") makes me think they are lying.
The fact that having a high-speed encryption chip is quite useful is being used to hide the real purpose. Do you really think the same people who think Winmodems are a good idea are that interested in adding hardware just to speed up a function that can be done in software?
They also make a point about the random key generation, which is interesting, because it keeps the private key completely in the hardware where no program can see it and thus be fooled to reveal it. However I am curious if this is actually a defense against any real exploits. I have not heard of exploits that involve revealing the private key of a previously-negotiated pair, most involve fooling the system into doing something unwanted through an already opened and legitimate channel, or fooling it into using another public key that the attacker already knows the private one for. Can any experts find any real exploits where a temporary and untransmitted private key was revealed? If not then I would also suspect this is a smoke-screen, attempting to turn the fact that the chip has secret keys into a benefit. I would also think that 99% of the benifit, if any, could be achieved by loading the chip with a random pair and then making sure the program has eradicated all knowledge of the pair. There have been expoits in weak random number generators, and in this case the random number generator is in hardware and no longer easily fixed.
and even the fact that you can generate key pairs
And You Lost
I think there is a communication problem here. The article used "remote" to mean not-in-hardware; i.e., all software. It didn't mean just over the network.
An employee can get to the keys, but only by hacking the hardware. A possibility (as clearly explained in the articles), but not likely. It's also questionable when getting these keys would _do_; they only seem useful for the single machine itself. And I'd presume a good admin would clear/reset any keys if the machine is transfered to another employee.
I will NOT use any system that involves FORCED TCPA, DRM, Fritz or any implementation of M$ code, embeded OS, hidden serial numbers, or any other Big Brother features.
You people can use all that stuff you like, I will not.
And when they finally force people to use it, in that you can not connect to other systems that DO use it unless you too use it, then that's when I become a total luddite and will just go live in the woods and live off the land...
This whole thing stinks and no one can convince me otherwise...
When I have my own chip fab.
Choice of masters is not freedom.
I downloaded this driver when it first came out (even though I don't have the hardware) and it looks like it only has low-level hardware communication code in it. To make it at all useful, you'll need a library that marshals and unmarshals commands to the chip, and I haven't seen such a library anywhere.
Excuse my ignorance, but would it not be possible to write a low-level program to intercept calls to a chip like this and reply to such calls with whatever answers you would like? I'm not a programmer nor software engineer but it seems to me that software calls to a hardware address could be redirected to a routine preloaded into RAM automatically and since the hardware address would be "hard-coded" it would be trivial to spoof the replying address if that were needed. If such is the case, all we would need is a software solution to overcome the "trusted computer" hardware handcuffs. Or no?
In the "No Need for TCPA" paper, the author argues that TCPA isn't practical for DRM because:
How could content providers recognize which reported PCR values were good, given the myriad platforms, operating system versions, and frequent software patches?
This misses the point that the content providers don't need to check all those platforms/OSs/patches. They just need to check that all these components contain a signature from an "approved" authority. Without all required signatures, the user doesn't see/hear anything. The part about tampering with the hardware is relevant, however.
Graham
Graham