3

I have configured a disk encryption in linux using LUKS with TPM with secure boot.

The encryption key is loaded from TPM from NV memory using command TPM_NV_ReadValueAuth.

The problem is that it is read as a clear text and it can be captured on LPC bus with a cheap logical analyser.

My assumption is that the attacker can have physical access to HW and therefore would be able to get the key and decrypt disk.

How to avoid sending clear text key from TPM and avoid eavesdropping or sniffing the key traveling from TPM to RAM?

1

I don't think this problem can be solved, as someone capable of soldering wires to the LPC bus can solder them to the memory modules and get the keys from memory.

The best alternative I know about is to simply keep the keys separate from the machine, like on a flash drive and only connecting it when necessary after making sure the machine hasn't been tampered with. That way, someone stealing the machine has no way of decrypting its storage, no matter how hard he tries (excluding quantum computers).

  • I think that sniffing communication on memory is much more harder as you are not able to store data because of speed of memory and I am not aware of any cost affordable sniffer for memory or something similar. Whereas you can get the key leaving from TPM using 500$ analyzer and hour of analysis of data. Would be any difference using for example Intel Platform Trust Technology or ARM TrustZone? Would be the key still easily eavesdropped? I don't have experience with that. – user1563721 Nov 26 '15 at 21:24
  • @user1563721 any platform where the memory modules are easily accessible is vulnerable. You could get additional protection by using an SoC where both the key storage (NVRAM), RAM and CPU are on the same package, making eavesdropping much harder (bonus points if the SoC is designed to be tamper-proof). Unfortunately I don't know what TurstZone nor Platform Trust Technology is but as a general rule I wouldn't trust proprietary crypto/security technologies. – André Borie Nov 26 '15 at 21:40
  • @user1563721 also, I was talking about a government-level attacker with almost unlimited resources, but you are right that sniffing memory would require a bit more than a 500$ logic analyzer. – André Borie Nov 26 '15 at 21:41
  • Even a government adversary does not necessarily have access to that technology, at least not at large. They still buy incredibly overpriced (scam level) PCIe-based DMA attack hardware with simple push-button "extract TrueCrypt key" functionality because they don't know how to use a $10 PCIe device to get the same functionality. – guest Nov 19 '17 at 4:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.