Open Stack

Hello Robert!

The key splitting in XTS is “minor”, say 256 bit key, top half of key (k1) is used across the whole volume, and the
bottom half (k2) for creating the per sector “tweaks”. The per sector variant, aka initialization vector, does not need
separate storage space because it can be computed from the sector-id and key.

For a single 128-bit data block P in a sector, the encryption can be described as C = T  AES-ENCRYPTk1(T  P), where T = AES-ENCRYPTk2(i)* αj.
In this case, i is the address for the sector, j represents the position of the 128-bit block within the sector, α is a primitive element of the Galois field GF(2128) and the multiplication is defined in the same field as described in [1].
(Excerpted from : http://www.intel.com/content/www/us/en/intelligent-systems/embedded-systems-training/ia-high-performance-storage-encryption-paper.html)

So, using is AES-XTS is a good default, the approach JHU-APL has taken.

The key length should not be a detraction, the keys can be auto-generated. If a user provided key is short, I can be extended with random bits.

Regards
Malini
From: Clark, Robert Graham [mailto:robert.clark at hp.com]
Sent: Monday, February 11, 2013 2:29 AM
To: OpenStack Development Mailing List
Cc: Benjamin, Bruce P.
Subject: Re: [openstack-dev] Volume Encryption

Both CBC-AES and XTS-AES are perfectly well supported in modern OS’s after all we’re all talking about using AES, only the mode of operation is being discussed.

While it’s true that those deploying OpenStack will want to use something that’s OS supported, we should ensure that the most prudent default setting is set. As a secondary point this is a completely appropriate venue for this discussion. It’s important that as a community we have a discussion about the use of crypto primitives, algorithms, modes etc.

I’m  not an expert on crypt but have worked in this space for a few years, I think using CBC with suitable IV would provide a similar assurance levels as XTS, the advantage perhaps with CBC is that one wouldn’t need to use such large key sizes. If memory serves XTS does some peculiar key splitting which forces the user to use either 256 or 512bit keys.

In either case I think the data will probably be sufficiently secure on disk. When appraising/attacking any modern crypto system I will always go after the key server first, where would be a good place to read about the forthcoming ‘enhanced key management server’ ?

From: Bhandaru, Malini K [mailto:malini.k.bhandaru at intel.com]
Sent: 09 February 2013 00:19
To: OpenStack Development Mailing List
Cc: Benjamin, Bruce P.
Subject: Re: [openstack-dev] Volume Encryption

+1 Supporting the  current best set of algorithms.
Malini

From: Caitlin Bestler [mailto:caitlin.bestler at nexenta.com]
Sent: Friday, February 08, 2013 3:13 PM
To: OpenStack Development Mailing List
Cc: Benjamin, Bruce P.
Subject: Re: [openstack-dev] Volume Encryption

On 2/8/2013 1:57 PM, Benjamin, Bruce P. wrote:
Bryan D. Payne wrote:
> If memory serves me right, XTS has some known issues (in particular
> data integrity issues and reply attacks).  I typically still prefer to
> use CBC as it is time tested and works nicely if you handle your IV's
> properly.

We understand that CBC has some watermarking issues for storage encryption use.  XTS is a NIST-approved cryptographic standard for this purpose.  http://csrc.nist.gov/publications/nistpubs/800-38E/nist-sp-800-38E.pdf.  You’re right that this doesn’t provide integrity checks, but the SP800-38E standard states “In the absence of authentication or access control, XTS-AES provides more protection than the other approved confidentiality-only modes against unauthorized manipulation of the encrypted data.”  Also note that cryptsetup for dm-crypt uses XTS as the default mode now.  http://www.spinics.net/lists/dm-crypt/msg04885.html.   The normal usage of XTS would be in an encryption module that would reside directly with the hard drive platter that would be storing the encrypted data.  In our case, though we’re sending the data over iSCSI to a remote drive, we believe this encryption mode can still support a reasonably secure solution, assuming that an enhanced key management server (forthcoming) will be implemented.  If the key is kept from compromise, the encrypted data cannot be easily manipulated or substituted in its encrypted form, and it would basically randomly corrupt data within that block.


I do not see any point in discussing which encryption algorithms will be supported in an OpenStack forum.

If a given encryption algorithm is supported by most operating systems (translation: Linux) then customers
will expect that option to be available.

And if an encryption algorithm is *not* supported by those same algorithms then very few customers would
accept an encryption solution based on software written in python.

So ultimately we are going to accept the determination of the OS vendors and chip developers.
There's no point in debating these issues here.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20130211/9885b5de/attachment.html>