Re: Opiated file systems

[Adam Back <aba@dcs.ex.ac.uk>]

Rob <WlkngOwl@unix.asb.com> writes:
> > Some more thoughts on encrypted file system design criteria.
> > A wish list:
> > 
> > - Choice of secret key encryption algorithms (IDEA, 3DES, MDC, Blowfish)
> 
> Nice in theory. Awful in practice.  Requires code for managing keys, 
> encrypting and decrypting for ea. algorithm be resident in memory.  

No need to have all the code in memory, I think you ought to be able
to keep the code in an overlay, or something of that nature.

> For some systems (MSDOS), free memory is at a premium... (one of the 
> reasons SecureDrive is popular is because it takes only 2.5k).  It's 
> not worth wasting memory for handling several algorithms when only 
> one is going to be used in most cases.

I take your point about memory consumption under DOS, though
presumably it would be possible these days to load data high.  I think
SFS supports this for instance.

> [...]
>  
> > - High performance (hand optimised assembler for each architecture)
> 
> So much for maintaining code across platforms.

Only the secret key algorithms, 80x86 code would be the highest
priority coz they're the slowest (well the old ones are).  You're
going to need very OS specific code for the low level parts of the
file system anyway.

> > - Compression
> 
> Not worthwhile.  Use a Stacker or JAM driver over the encrypted 
> partition on a PC, for instance.  Keep compression and crypto 
> separate utilities... keeps bugs from one interfering with another 
> and reduces complexity of both drivers; also, if one wants crypto 
> w/out compression or compression w/out crypto, no wasted memory (see 
> above about RAM being at a premium).

One of the requirements some earlier posters gave was that they would
like to be able to accesss the same data with different OSes (the
example was to access the same partition with linux,win95, and winNT).
Double space isn't available for unix.  (There was a read-only version
for linux, but this seems to be currently unmaintained.)

So one reason to include a compression module would be for portability
of data.

I agree with your points about separating concerns and keeping crypto
and compression reasonably separate to keep bugs in compression code
having the potential to affect crypto code.  It would probably be
prudent to use OS segment protection, and page locking to protect
crypto code and data from being accidentally overwritten, and written
to swap respectively.

> > - Ability to chain algorithms (IDEA and then 3DES for example)
> 
> Why?  Doesn't necessarily increase security, esp. considering the 
> performance hit (memory... see above, time, key management).

Hmm, say that you were using MDC with md5 as the hash (before
Dobertin's recent pronouncements, naturally no one would do this now),
if Dobertin comes through with the general case you might wish you had
combined it with 3DES...

For the paranoid only, but a nice option I think,

> [uncontenious stuff]

> > - Facility for duress key, with the real data hidden in the unused
> >   space of the first encrypted drive.  To increase the plausible
> 
> Huh?!?

Encrypted filesystems are for hiding data from other parties.  If your
threat model includes law enforcement such a feature would be most
useful.  You would have data which you would not mind agents
obtaining, and have the "real" data hidden in a second file system.

When you are supeonaed for your key, you reveal the 1st file systems
key.  That key does not reveal anything about the 2nd partition, not
even it's existance.

The second file system would be hidden in the blocks not used with the
1st file system.  To access the 2nd file system you would need to tell
the file system driver the keys for both file systems -- the 1st key
so that it could find which were the unused blocks in the 1st file
system, the 2nd key for access to the hidden file system.

> >   deniability all unused blocks within a file system should be filled
> >   with garbage, so that it is not possible to tell if there is more
> >   data there.
> 
> If the algorithm is good, this shouldn't matter.  The only way a 
> person could tell if a sector is unused is if that person was able to 
> mount the partition already.

Yes, that's the idea -- your 1st key has been supeoned is the threat
model.

> > - File system steganographically hidden in files on another file
> >   system (encrypted or not).  Support for a wide selection of file
> >   formats (Aiff, Wave, Midi, JPEG, GIF, RGB, MPEG).
> 
> Now this is getting nutty!!!  Never mind the size, 

4 gig disks are cheap these days

> compleixty and amazing slowness of such a driver...  you'd have to
> have audio or video files of gigabytes in size to be able to store
> anything of use.

Indeed you would.  So?

> > - Ability to use stegoed file system in files on an unencrypted
> >   file system, and boot from a floppy to access stegoed file system,
> >   with no other traces left on hard disk.
> 
> Why?  The authorities would wonder why you have an 8 Gig JPG on your 
> disk and figure you're using it for stego, or you're crazy, or both, 
> and have you committed.

Nah, you'd buy some video editing equipment, perhaps a photo-CD
recorder, start a business doing photo-retouching (I know someone who
does this btw, they use DAT tapes just to shift the Gbs they get
through), video editing, whatever.  All quite plausible.

Adam
--
only quiche eaters need ritalin...