Full Disk Encryption - Xen, Windows and Linux?

Bofh To asks: "I'm in an industry that, more or less, requires full disk encryption, and to accomplish this, we use Pointsec on Windows. For the past 8 years, I've been running Linux on my work laptop, and this is the first time I'm running in a Windows only environment. I am interested in changing that, because I want to use Linux as my main platform, and only drop in to Windows when necessary (and use crossover if at all possible). I'm also interested in Xen, and would like to see if I can use that to virtualize Windows under Linux. My thought is that, as long as Pointsec is in dom0 and I use virtual disks for the Windows VM, I should be covered. The problem is that I'd also like a machine that is usable, as opposed to waiting endlessly as the virtual memory, virtual machine, pointsec, and xen all thrash around while I'm working on the machine. Has anyone used Pointsec for Linux, with Xen? "

Look at dm-crypt

[Cheeziologist]

I know you asked about people using pointsec with Linux, but have you considered using the device mapper to do hard disk encryption for you? On my laptop, I have the entire hd encrypted using aes and sha256, using the kernel's dm-crypt abilities and the cryptsetup [endorphin.org] program. To do this, you need to have a small partition to boot from that contains the kernel (and an initramfs if you don't build it into the kernel). From there you unencrypt the drive, pivot root, and continue booting. Additionally, if your intent is to run the virtual windows encrypted, you can use cryptsetup to manage the the device or files to keep the windows files on. There are many good tutorials on using dm-crypt, and can definitely tell you more than I can easily explain.

Re:

[phrasebook]

How's the performance of dm-crypt for you?

I use it on my swap and /home partitions on my laptop, but when doing heavy writing to the disk, the whole machine locks up for 1 or 2 seconds at a time - no mouse movement, no sound, no cursor - then it resumes. These freezes occur every 10 seconds or so as data gets flushed out to the disk.

Normal reading/writing load is ok, but doing something like an rsync backup kills responsiveness.

It seems to get a bit better if I renice kcryptd and kjournald. Any experience w

Re:

[Jah-Wren Ryel]

I use it on my swap and /home partitions on my laptop, but when doing heavy writing to the disk, the whole machine locks up for 1 or 2 seconds at a time - no mouse movement, no sound, no cursor - then it resumes. These freezes occur every 10 seconds or so as data gets flushed out to the disk.

From the dm-crypt faq: [saout.de]
Q: My system hangs for some time in regular intervals when writing to encrypted disks.
A: You are probably using Linux 2.6.4. Du to the introduction of kthread pdflush is running at nice level -10,

Re:

[phrasebook]

I only started using dm-crypt with 2.6.18 in Debian etch. But kcryptd and kjournald are -5 by default, so maybe renicing those is all there is to it. Weird thing is I've not seen many people mentioning the problem. I can only assume most don't experience it.

Re:Look at dm-crypt

[rjforster]

Not knowing the exact details of the requirement, but Pointsec is FULL disk encryption. This matters.

To the original poster:
I think this is one of those 'suck it and see' situations. Processors are getting faster all the time. Disks are getting faster too, especially solid state drives. So the trade offs between different performance areas are changing all the time. Eg today you might notice the crypto delays, tomorrow you might not because you essentially have a dedicated core doing disk crypto.

Last year I ran tests with Pointsec for a different situation and it was pretty good with a flash drive. Not _quite_ as good as a FDE competitor but not far off. This wasn't on a fancy new laptop with decent dual core processor either. For these tests I got a free eval copy of Pointsec. They were nice, helpful guys when I spoke with them, perhaps you could get an eval copy too.

Another alternative is a hardware solution such as Flagstone from Stonewood. Full hard drive speed and full OS compatibility.

Re:Look at dm-crypt

[swillden]

Not knowing the exact details of the requirement, but Pointsec is FULL disk encryption. This matters.

As is the proposed dm-crypt configuration. In both cases you have a small unencrypted boot section containing no sensitive data and everything else is encrypted.

The only difference from a security perspective is that you can't audit Pointsec.

Re:

[rjforster]

With Pointsec only the MBR plus a couple of other sectors are unencrypted. There is no small partition in plain text which is what I understand dm-crypt to be. Please correct me if I'm wrong.

Re:

[mangu]

This "plain text" partition isn't text at all, it's just a set of routines to load enough decryption software into the system to use the rest of the disk. There's no sensitive data there because it's all public software anyhow.

Re:

[Matthew Bafford]

In cryptographic terms it is plain text.

Yes, but isn't it irrelevant plain text? What does it have to do with the encrypted data? It's just a compiled public algorithm, isn't it?

Maybe I'm missing something.

At some point something has to be plain text. You can't have everything encrypted without something being unencrypted and runnable (even if it's in a chip or in the boot sector).

Re:

[swillden]

With Pointsec only the MBR plus a couple of other sectors are unencrypted. There is no small partition in plain text which is what I understand dm-crypt to be. Please correct me if I'm wrong.

You're correct, but the difference is irrelevant: it doesn't matter if it's a few KB or a few MB that is unencrypted, the key is that all of the functional system and its data is encrypted, including all swap.

Actually, dm-crypt and Linux can do one thing that Pointsec, AFAIK, does not do, which is to take advantage of a TPM-enabled machine. Given a TPM, TPM-enabled BIOS, TPM-enabled GRUB and Linux kernel, you can bind a portion of the master decryption key to the boot state, ensuring that any attempt

Re:

[rjforster]

I would expect a few kB to be easier than a few MB to audit, otherwise your point is valid.

As for the auditing, I would take closed source but CAPS (or similar) approved[1] over open source non-CAPS. Because it _has_ been audited as part of the approval process. Of course, at this level the rubber hose hack is the best way of getting to the data.

I don't know about the TPM side of things.

Cheers

[1] By CAPS approved I'll take the 'commerical' version of a product certified for classified data use. In other wor

looking forward to replies on this one

[Cybersonic]

I have not tried out Pointsec yet, but its a solution my company sells so I should learn it :) I certified myself in PGP, which unfortunately does not support full disk encryption on Linux, just Windows and soon OSX... It also does not support dual boot on Windows. (its a shim into ntloader - but after the actual boot loader the 'pgp' os which asks for the decryption key during boot is linux, so I KNOW they have linux expertise...)

I kind of like the roll your own approach to the Linux full disk encryption scenario, but most large organizations balk at anything thats not a commercial solution

Re:

[Paracelcus]

I can tell you that Compusec's free security software does the job very well, you cannot use it with an MBR boot manager like GRUB or Lilo as it's loader resides in the MBR. http://www.ce-infosys.com/english/downloads/free_c ompusec/index.html [ce-infosys.com]

As to virtualization, I use VMware to run Windows under Linux.

Debian's new installer is spiffy

[deftcoder]

The latest version of Debian Stable, codenamed 'Etch', has the ability to set up a fully-encrypted system (except for /boot of course) right from the installer.

It's amazingly simple to use, and great for laptops. (I'm running it on my dual-core laptop)

Check it out: http://www.us.debian.org/CD/ [debian.org]

Re:

[kestasjk]

I thought I covered this when I said

if it gets it from somewhere else but it is read by the kernel in /boot that's also no good because the kernel could be replaced.

If you're protecting against theft having an unencrypted kernel read the password is fine. But if you're protecting against theft why both with full disk encryption; why not just encrypt specific files or use a virtual encrypted drive like TrueCrypt?

The main reason for full disk encryption instead of alternatives is that it makes it impossible to modify any part of the operating system while the machine is offline; so you can have a system running in an insecure envir

Re:

[Bishop]

Full disk encryption and more targeted encryption such as file encryption solve two different pieces of the same problem. Full disk encryption (and all-but-boot encryption) protect against accidental exposures and residual data. Accidental exposures are from copying a file from an encrypted part of the drive to non encrypted part. And residual data are just the little bits of the plain text left from when the encrypted data was decrypted.

Full disk encryption is convenient security. It is seamless and the pe

Re:

[Bishop]

If an adversary can access the hardware then all bets are off. It does not matter if the kernel could be trojaned as the hardware could be just as easily trojaned with a physical key logger or similar. All (consumer) computer systems are susceptible to this attack. Even TPM systems.

Carrying /boot on a removable device is easy. A tiny USB drive would be sufficient.

These questions always make me smile...

[Anonymous Coward]

I always find these types of "Ask Slashdot" amusing. People ask about what security product to use in their enterprise, how it will work with Linux etc etc. All perfectly valid questions, but utterly pointless in a corporate context because guess what? It's the Information Security Policy (& CISO) which will dictate who can and can't authorise new encryption products, changes to production environments, installation of non-standard baseline software (and the list goes on & on). If the OP really does

Re:

[zCyl]

What's wrong with a little public brainstorming? I don't know what business you work in where the IT guys come pre-programmed with all the answers.

Re:

[hey!]

Probably a large one.

If you're talking a thousand or so employees or less, you have about a dozen or so IT guys, so you head over to where they take lunch and you shoot the shit with them, and they can probably agree it would be cool to look at solution X on Linux.

If you're talking an outfit with a thousand or so IT guys, then the answers are likely to be preprogrammed unless you can get to somebody high enough. Even then they're going to be more interested in keeping their headaches minimized than making

Re:

[jsellens]

One "obvious" approach is to run the corporate windows, and then run linux under vmware running on windows.

Re:

[Gothmolly]

While the ISO may _set_ policy, he's usually the last person you want to actually ask about a technical solution to anything. "What does Gartner say?" is not a technology strategy.

Compusec by CE-Infosys

[mlts]

I have seen a product for FDE for Linux, although its not open sourced at all. CE-Infosys's Compusec. The nice thing, its usable at no charge, so it may be worth a look on a non-production box. However, I don't know much about it, and have not tested it.

gordon freeman approves...

[User 956]

I'm also interested in Xen, and would like to see if I can use that to virtualize Windows under Linux.

I'm not sure about that, but I'm sure Xen would be a great place to store backups to keep them from prying eyes. [wikipedia.org] Who needs encryption when you have a low-gravity parallel dimension as a safe-deposit box?

Have you considered Pointsec on Linux?

[swillden]

They have a Linux version [pointsec.com]. Then your virtualized Windows image will also be encrypted. BTW, for virtualizing Windows, I'd recommend you get a copy of VMWare, rather than using Xen. The open source virtualization tools are coming along, but at this point in time VMWare will perform much better.

Re:

[Mr2cents]

It's always confusing when they say something like Pointsec for Linux 2.0. Is it (Pointsec for Linux) 2.0 or Pointsec for (Linux 2.0)? It always takes me a few seconds before my mind settles on the first ;).

Re:

[bratgitarre]

I know. The non-associativity of the English language is frequently a problem to me, a non-native speaker. It would help if people would use hyphenation to break disambiguities, e.g. instead of writing "infamous WMD evidence", write "infamous WMD-evidence" (I made this up on the spot, there are certainly better examples).

Re:

[bill_mcgonigle]

The Ask Slashdot probably changed after you posted your comment. As of now the question ends with, "Has anyone used Pointsec for Linux, with Xen?", so he's apparently considering it.

to all the people advocating dm-crypt

[Alex]

It kind of sounds like the poster is stuck with pointsec, and bearing in mind his "industry" requires full disk encryption. I'd imagine he probably doesn't get much choice how its done.

Alex

Do you really need full hard disk encryption?

[Futurepower(R)]

If you can accept just having some partitions encrypted, TrueCrypt is wonderful [truecrypt.org].

Known plaintext attack on encryption

[Anonymous Coward]

Is full disk encryption a good idea? With the operating system within the encrypted partition, it gives a LARGE amount known plaintext to mount an
attack.

Re:

[DanLake]

There is not really any known plaintext on the disk. If The disk is 40GB, and the O/S is 1GB, how exactly will you know which bits of ciphertext correspond to which bits of operating system? You wouldn't. Any block of the O/S could be in any block of the disk, possibly fragmented, possibly random data in slack space at the ends of files, different package and kernel versions, etc.

Known plaintext is when someone tells you here are is 64 bits of ciphertext (i.e. from DES), and then also gives you the 64 bits

Re:

[tigersha]

If you take two machines, install both of them right after the other both disks will have the same sectors written, almost guaranteed. File system fragmentation only happens when you write to files and this never happens to the OS (only when you reinstall or upgrade). Having two machines installed right after the other happens a lot and the chances that the OS is on a predictable place if you only stole one laptop is very high. The parent poster is right.

Is there any way of deliberately fragment a disk duri

I wish FDE were more common with UNIXs

[mlts]

I sort of wish FDE were more common with UNIXs, where the only real point of attack would be the small amount of code in the MBR. With a TPM chip, even that is protected, so an attacker would have to physically disassemble the TPM chip and pull the key out of its physical RAM cells (which is pretty hard on even an unprotected chip, unless you are a large corp or government with an up to date chip fab.) /boot can be compromised fairly easily with a clever keylogger and some way to store the decryption key c

BestCrypt

[Sedennial]

We just finished evaluation of a number of products as we also require full disk encryption. We are purchasing BestCrypt from Jetico [jetico.com]. It also handles encryption of pagefile, swap files, swap partitions, and hibernation files.

Xen + Encryption + LVM + RAID on Debian

[Burz]

Here is a general overview [lxer.com] of the steps needed to set this up on Debian. Also take note of the responses from Sander.

It is probably more than you are looking for, since it doesn't sound like you want RAID. But that part is easily skipped. The LVM part I would keep, as logical volumes will make managing the virtual machines that much easier.

Actually, a lot of this (the LVM and encryption parts) should be doable from the Debian 4.0 installer.