Princeton Deep-Freezes RAM To Sidestep Encryption
Written by Evan SchumanMy favorite aspect of writing about security is that software designers work so hard to craft sophisticated, multi-level strategies to protect data, which is then periodically sidestepped by some kid with $18 worth of equipment, engaging in an easy blunt force attack that no one anticipated.
Case in point: Some clever folk at Princeton came up with this delicious and low-cost way around typical encryption methods. It’s based on RAM chip’s tendency to keep all data–including encryption keys–for seconds or even minutes after power is cut off. The Princeton approach then used a common dust removal aerosol can to deep-freeze the chip before killing power, a cooling tactic that sharply lengthy how long the RAM remembered the desired key data. From there, simple pattern recognition software quickly found the keys. This Princeton video demo is worth watching.
-Marc
February 28th, 2008 at 8:06 am
It’s a widely held tenet in security that if someone has physical access to your machine, they own your machine.
This attack was not unforeseen. Not to take anything away from their work: the Princeton team did the legwork to turn it from theoretical to demonstrable, and produced their striking demo. I think the most surprising aspect of their work was the length of time the room temperature DRAM held a usable image, and the hardest part was identifying and teasing the password out of the image.
Other attacks using a powered-on vector have long been known: an external machine reading a perfect memory image over a firewire port using raw DMA access was demonstrated years ago. “Chip clips” have long been used by reverse engineers to read the contents of RAM in a powered up machine.
Bitlocker, Safeboot, and any “whole disk encryption products” that don’t use hardware-based encryption have always been more about security theater than security. They’re exactly like IPSec and SSL in that regard. They’re certainly better than nothing and do stop certain avenues of attack, but have never provided cryptographic assurance that an attacker at an endpoint cannot get to the data. The closest we have right now are TPM chips and on-chip encrypting secure smart cards; even then the satellite TV pirates learned long ago how to hack keys out of smart cards.
The lesson most non-security people need to take away from this is: if the machine is in the physical hands of an attacker, there is a risk that its contents can be compromised. With encryption it is a much smaller risk, but it is always present and can never be made zero.
February 28th, 2008 at 12:53 pm
Very…Cool!… and it works on Mac and Linux, now I bet those two are having fits over this.
February 29th, 2008 at 12:35 pm
It reminds me of when I hacked into the UNIVAC used on ship for distributing messages and arranging them for print for storage purposes. The funniest thing was the blank look on the techs’ faces; I wasn’t part of the sanctified crowd so I couldn’t possibly do it (which gained me access a couple of other times over the years). The image of a thing is not the thing itself; absolute statements other than regarding the network of representation are potentially quite dangerous. And, for that matter, one really doesn’t equal one.
–Glenn