Hi: Russ Humphries here.  There’s been a lot of attention this week paid to memory attacks against disk encryption technologies and I wanted to provide some commentary and thoughts. The focus of these conversations is centering on investigating the contents of a computer’s memory – if it’s running or shortly after it has been recently powered down; where ‘recently’ could be seconds to perhaps minutes. The concept that memory retains a ‘ghost image’ of what was last stored on it has been well documented and is an industry-wide issue.

However, the current debate has an interesting angle to it - specifically a method has been detailed in which an application might be able to reconstruct an encryption key, which might have been used for almost any security purpose, from these ghost images.

Since disk encryption is a topic that gains headlines perhaps it was inevitable that the practical demonstration of this key-reconstruction would be to investigate a computer’s memory to ‘break disk encryption products’ and potentially access data stored on the hard drive.

The thing to keep in mind here is the old adage of balancing security, usability and risk.  For example BitLocker provides several options that allow for a user (or more likely Administrator) to increase their security protections but at the cost of somewhat lowering ease-of-use.  BitLocker supports options that will not allow a machine to boot – or resume from hibernate – until the user can:

·         Enter a PIN

·         Insert a USB stick that contains a secret Key

·         … and as of Windows Vista SP1 both enter a PIN and insert the USB stick!

We provide best practice guidance in the Data Encryption Toolkit (http://www.microsoft.com/technet/security/guidance/clientsecurity/dataencryption/analysis/4e6ce820-fcac-495a-9f23-73d65d846638.mspx ) that describes the various manners in which the above choices can be made and also provides advice to help improve security, such as disabling ‘sleep mode’ – forcing a user to hibernate and thus allowing memory to lose the ghost images discussed. These power management settings can all be configured centrally using Group Policy Objects.

Now with the above context in mind, I’d like to take a step back and, from a BitLocker perspective, detail some of the assumptions that have to be made for this attack to be successful:

·         Physical access to the machine

·         The user’s laptop would likely have to be in sleep mode, rather than hibernate mode or powered off

·         The user would have chosen not to implement multi-factor pre-boot authentication

·         The person who finds/steals the laptop must be knowledgeable and interested enough to execute this attack on the laptop they just stole

I would posit that the opportunistic laptop thief is somewhat unlikely to carry a separate laptop on which they will have installed tools that allow them to reconstruct cryptographic keys – or for that matter have a can of compressed air handy. 

Targeted theft is, of course, an entirely different threat model!

Let me also point out that BitLocker allows an administrator to, quite easily, change the protection method for a laptop, even remotely [but assuming some form of connectivity], by having a script execute.  Thanks to BitLocker’s design, which implements key abstraction, a script can be executed that adds pre-boot protection mechanisms without requiring the re-encryption of the hard disk. This script can therefore execute very quickly.

Let me close by clearly stating that quality security research helps our customers and the industry in general raise the security bar, and I applaud it; but let’s also keep in mind that technologies like BitLocker provide a very valuable service to users and helps them protect data on their PCs. BitLocker’s range of deployment options, ranging from single-factor authentication with sleep mode to TPM+PIN+USB with hibernation only, allow customers to find the right balance of security and convenience for their data; the documentation of one attack method, that can be mitigated through these policy choices, does not equate to a class of data protection products being rendered ‘useless’ as has been reported in some circles.

-Russ