[dm-crypt] dm-crypt Digest, Vol 27, Issue 2
arno at wagner.name
Thu Sep 1 19:24:40 CEST 2011
all this is really just a side issue to disk encryption
as many of the base asumptions do not apply in a cloud
setting. For example, disk encryption on a laptop does not
really protect you from an attacker with repeated undetected
access, but from the situation when you lose the machine,
know it and it stays lost. An attcker with repeated access
can do a lot of things that completely negate all benefits
of disk encryption. In the cloud, repeated access is the
the primary model.
Some more comments below.
On Thu, Sep 01, 2011 at 04:30:23PM +0300, Yaron Sheffer wrote:
> Hi Arno,
> Encryption of data-at-rest in the public cloud is not "pointless",
> it is yet another layer of security. Just as people encrypt their
> laptops even though they are password protected.
If we are talking secure cloud storage, then the encryption did
not happen in the cloud and the situation is different. As network
bandwiths are low in comparison with encryption speeds encrypting
the data before putting it in the cloud is easy. I do know that
there are cloud storage providers that encrypt in the cloud,
but these do not qualify as "secure" and the encryption is pure
parketing that does not add any quantifiable amount of security.
Als, this type of storage will typically be file-based.
> The cloud provider does not have access to "everything", certainly
> not when we're talking about data at rest, where the keys may have
> come and gone months ago but the data is still there. Moreover, the
What prevents the cloud provider from regularly pulling
keys from VMs and storing them? It may take a little effort,
but encrypted data is relatively easy to recognize, and pulling
encryption keys from, say, a Linux kernel, is easy if you have raw
memory access. For encrypted block devices you just need to find
the key once.
I see no problem at all with scripting this with reasonable
efffectiveness in, say, 4-8 developer weeks. And given that
Amazon (and others) cooperate with TLAs at least in the US, I
am sure they have been asked to put such "key grabbers" into
place and have them ready. Now, I am not saying they run these
routinely and on all instances, but I think that it would be
entirely feasible and cause quite low overhead to do so.
Side note: If it is encrypted block storage "at rest", the
cloud provider can just wait until you access it again and
then grab your keys.
> cloud provider is not the only or the most important threat. By the
> way, I am not claiming that the permission system is broken.
If the permission system is unbroken, then encryption does
not add anything, see above.
> Attacks on encrypted data are no harder or more expensive in the
> cloud than on physical disks. If you parallelize things, your
> throughput is limited by the disks physical access, just as for
> "real" disks.
This staement does not make sense. Maybe you mean it the other
way round? Still, if you can actually brute-force (or somthing
close to it) with reasonable cost, I would regard the crypto
as broken. That would not be the fault of the cloud.
But you forget that access (physical, logical) in the cloud is not
under your control and in fact you may not even be able to tell
what access security measures are in place, and how good they
are. Certification, SLAs and vendor promises have turned out
to be meaningless time and again. As this applies to your system
image and even your system memory as well, encryption does not
solve the problem. It does however increase complexity.
> This is a solution for a very real problem. But I don't want to go
> commercial again...
The problem is very real, I agree on that. However, block-device
encryption is not the solution. The real danger is that people
could think it was a solution, i.e. think their data was
secure in the cloud because "it is encrypted". But the cold fact
is that you cannot trust a cloud not under your control with
confidential data. That is why confidential data does not
belong there. For a "private" cloud, things are different.
> On 09/01/2011 02:38 PM, dm-crypt-request at saout.de wrote:
> Message: 2
> Date: Thu, 1 Sep 2011 13:27:24 +0200
> From: Arno Wagner<arno at wagner.name>
> To: dm-crypt at saout.de
> Subject: Re: [dm-crypt] Blog post on FDE and integrity protection
> Message-ID:<20110901112724.GB4617 at tansi.org>
> Content-Type: text/plain; charset=us-ascii
> Disk encryption in a non-private cloud is pretty pointless.
> The cloud provider can access everything. An attacker should
> reliably be kept from accessing your storage, otherwise you are
> screwed anyways. I know, people are doing this, but they are
> kidding themselves.
> For your EBS scenario, true, block-level encryption
> can be done, but it is irrelevant. Encryption is not the
> right way to fix a broken cloud permission system. Critical
> encrypted data should never be decrypted in the cloud. It
> is just not secure. On the other hand, attacks that
> manipulate encrypted images are not relevant for lower
> security requirements, as they are very hard (expensive)
> to do.
> This makes integtity protection of encrypted data in the cloud
> a complete non-issue. This is a solution without a problem.
> On Thu, Sep 01, 2011 at 01:51:38PM +0300, Yaron Sheffer wrote:
> >>Hi Arno,
> >>Thank you for reviewing my post. Please see my comments below.
> >> Yaron
> >>>Message: 3
> >>>Date: Wed, 31 Aug 2011 23:29:40 +0200
> >>>From: Arno Wagner<arno at wagner.name>
> >>>To: dm-crypt at saout.de
> >>>Subject: Re: [dm-crypt] Blog post on FDE and integrity protection
> >>>Message-ID:<20110831212940.GB25013 at tansi.org>
> >>>Content-Type: text/plain; charset=us-ascii
> >>>Commercial, for sure. It combines fragments from well-known
> >>>facts and marketing speech. And it has not understood the
> >>>problem, advertizing for SAN/cloud services, where storage is
> >>>not block-based but file-based.
> >>The most commonly used public cloud is Amazon WS. This cloud offers
> >>two storage possibilities, S3 which is object ("file") storage, and
> >>EBS which is block storage, and is exposed to the application as a
> >>disk volume. The post is about EBS, sorry if that wasn't clear.
> >>>I should also note to anyone contemplating "solution" 3
> >>>that RAID1 does not read both devices on read access,
> >>>and inconsistencies will only show up if you or your
> >>>distro does RAID consistency checks.
> >>This is correct, thanks.
> >>>And of course the whole article does not apply to the
> >>>SAN/cloud setting in the first place, as the attack
> >>>scenario is for an unmapped encrypted filesystem and
> >>>an attacker getting write access to that, i.e. the
> >>>encrypted raw (block) view needs to be exported to
> >>>the attacker. I do not see how that would be done in the
> >>>SAN/Cloud setting. These do their own filesystem
> >>>and block encryption must be done below the FS layer,
> >>>there is no way around that.
> >>The attack scenario is for someone who has access (possibly limited
> >>access) to your cloud account to detach your EBS volume from its
> >>current virtual server, attach it to a different server, and then
> >>modify the (encrypted) storage. This is all completely doable and
> >>actually standard procedure on AWS.
> >>>On Wed, Aug 31, 2011 at 04:25:51PM +0200, Heinz Diehl wrote:
> >>>>On 31.08.2011, Yaron Sheffer wrote:
> dm-crypt mailing list
> dm-crypt at saout.de
Arno Wagner, Dr. sc. techn., Dipl. Inform., CISSP -- Email: arno at wagner.name
GnuPG: ID: 1E25338F FP: 0C30 5782 9D93 F785 E79C 0296 797F 6B50 1E25 338F
Cuddly UI's are the manifestation of wishful thinking. -- Dylan Evans
If it's in the news, don't worry about it. The very definition of
"news" is "something that hardly ever happens." -- Bruce Schneier
More information about the dm-crypt