Date: Mon, 13 Dec 2010 15:16:05 +0100
From: Levente Peres <sheridan@...sz.org>
To: full-disclosure@...ts.grok.org.uk
Subject: Possible issues with encrypted Linux filesystems?

Dear All,

Yesterday I had a very interesting conversation with Anthony G. Basile, 
Ph. D. of D'Youville College about filesystem security. We thought that 
we should continue this discussion here, so we could all contemplate on 
the possibility of such a thing being possible.

After reading Anthony's article, which you may find here...

http://opensource.dyc.edu/random-vs-encrypted


...I've became worried about something very alarming, which I'd like to hear your opinion about.

You see, it's one thing that you encrypt data, and then make backups, encrypt those backups, and the attacker could get valuable information by comparing the patterns of the two... But when encrypting an entire operating system space, you actually encrypt much more than the data you wish to protect: you encrypt your system files, your packages, all of it. Now this may sound like an ideal thing to do, but I'm not so sure about that anymore.

Now, as we know, most Linux distributions have at least some files, directories, whatever that are bound to be the same on all systems. For example, binaries of gcc, some base directory names like /var, /usr, /home, layouts, and things like that. Even more, if you are using a "standard" distro like CentOS, you are assured to have literally gigabytes of data in forms of binary RPM packages on a default "base" installation, which not only are sure to be the same on all systems, but even their distribution across filesystems are prone to be predictable. For simplicity's sake, let's just put these into one bucket and call them "known artefacts".

I'm now worried that if an attacker knows, or "guesses" that you are using, say, CentOS Linux 5.5, (or at least some mutation of Red Hat), he might use this knowledge of "known artefacts" to his advantage, by starting out from the data he knows "must be there", and looking for it's "patterns". I don't know... This may be a longshot, wishful thinking or both, but somehow it feels to me like it's a lot easier to break a code when you already know exactly what the decrypted data is, and what it looks like. It should be like reverse-engineering ancient-egyptian text by seeing the same damn text in two or three other different languages you can actually understand... Essentially you could at the very least improve your chances at success if you have several certain, fixed points of reference for the decryption procedure (these "artefacts" we mentioned).

I'll dare to go even further... Even if you are not encrypting your entire system, just the data... you could be leaving behind arefacts like file format headers, etc etc... or in case of LVM, logical flesystems within the LVM could leave behind headers, identifiers to mark the type, end or beginning, etc. of FS, whatever. I agree it's not much, and probably no concern, but if you want to be extremely paranoid, it's something.

Now I'm not pretending to be an encryption expert... But I've go to tell it to you, If there's any possibility to this - then it creeps me out. Worst case scenario, we could be looking at the possibility of breaking virtually any
"standard" distro as long as one could "guess" (or "brute-force-guess") the version and type of the distro, AND the system is encrypted along with the data to be protected...

I'd like you guys to put me back to ease by either proving me fatally wrong, or if there's anything to this... well, then we should discuss anyway.

Best Regards,

Levente Peres

_______________________________________________
Full-Disclosure - We believe in it.
Charter: http://lists.grok.org.uk/full-disclosure-charter.html
Hosted and sponsored by Secunia - http://secunia.com/

Powered by blists - more mailing lists