[gentoo-hardened] Hard Disk Encryption

[Boyd Waters]

One aspect of computer data security that has become important to me
is the obfuscation of data on the computer hard disk. This will slow
down data retrieval by unauthorized people who manage to gain
unrestricted physical access to the computer.

Such a scenario is becoming more likely with the rise in popularity of
laptop computers as primary workstation platforms. (I am writing this
on a laptop that is the only machine I use.) If someone walks off with
your laptop computer, then a robust access-control-list (ACL) policy
is not going to do you any good; one may simply mount the hard disk
via a less-secure system and have unrestricted access to the data.

I decided to encrypt the entire disk on my computer: not just per-user
partitions, but the entire root (and swap, too). The reason for this
is that I could not think of a straightforward way to prevent
important security tokens from landing in non-user areas, such as /etc
or /var/lib.

Critics of whole-disk encryption claim that it's a waste of time to
encrypt common files, such as the ~2GB of files that comprise the
libraries and binaries of a typical Linux installation. Or worse, that
such an approach can lead to known-plaintext attacks. Well: I don't
see a significant performance penalty, and cipher-block-chaining mode
can make known-plaintext attacks more difficult.

I have been using an encrypted root hard disk for a year. I back up my
data to other encrypted volumes. No data loss so far. I use the
kerneli CryptoAPI with 2.4.20.

This thread on Gentoo Forums has seen recent activity, discussing the
use of loop-AES for hard disk encryption with 2.4.x kernels:
http://forums.gentoo.org/viewtopic.php?t=31363

My recent efforts have focused on deployment of encrypted hard disks
in the 2.5.x kernels. The CryptoAPI is now part of the mainline Linux
tree, but loop device encryption is not.

An encrypted loopback block device driver has been implemented by
CryptoAPI developers as a thin layer on top of the new Linux 2.5 block
device system. I believe this approach may be best for the long term,
but there is relatively little discussion of this on the
cryptoapi-devel mailing list. And using this driver requires patches
to util-linux that have not yet been incorporated into the standard
util-linux distribution.

In contrast, the loop-AES implementation has been ported to the 2.5.x
kernels; this is a simple port of the 2.4.x block device to the 2.5
kernel. It may not be able to take advantage of the scatterlist
blockdev optimizations that the cryptoAPI incorporates. But it has a
responsive developer, and it seems to work better with recent versions
of util-linux (2.11z).

I have not yet been able to tweak my init RAMdisks that I used for
2.4.x root encryption so that they will successfully boot a 2.5
system. But I think I am close to doing so: another week or two at the
most.

I believe that block-level encryption is a far better approach for
on-the-disk data protection than a stacked filesystem such as
FiST. Creation of a FiST encryption layer would be upside-down, I
think: a standard filesystem such as e2fs that "hosts" encrypted
data. With FiST, the atomic units exposed to the encryption layer are
file-system units: file names, dirents, etc. With block-level
encryption, you encrypt blocks: opaque chunks of bytes. (FiST might be
very powerful for implementing policy-based access, though...)

I am very pleased by the Gentoo-Hardened effort. I think that the
current focus on learning about SELinux is vital: since security is a
_process_ rather than a _technology_, we need to learn how to
implement meaningful security policies on our Gentoo boxes.

I think that this Disk Encryption is completely orthogonal to this
effort: advances in this arena will not interfere with whatever
security policies are developed by the SELinux initiative. Block-level
encryption is (almost) completely transparent to a running system, so
ACLs and SELinux PSMs can run on top of such with *no* modification.

I think that other enabling technologies like token-based access
control (smart cards, USB dongles) fall somewhere in the middle: the
technology (device drivers) needs to be developed, but security
policies should be likewise developed to take advantage of these.

So:
I am working on block-level encryption. I think it is particluarly
important for laptop users. I would like to co-ordinate my efforts
with the Gentoo-Hardened project.

Cheers!

-- boyd
watersb on Gentoo Forums




--
gentoo-hardened@g.o mailing list