Re: Securing data in memory (was "Locking physical memory (fwd)

Bill Stewart wrote:

> Back in the mid-80s, Princeton University got some SDI money for the
> Massive Memory Machine Project, researching what you could do if
> you had enough memory for anything you wanted.  Even with the
> CS approaches of the time, you'd do a lot of things differently.
> Their non-massive toy machine was a 1.5MIPS VAX with 128MB RAM -
> which took 10 extension racks to hold all the RAM.
> These days 128MB is cheap, as are 500 MIPS P2s, but 50GB of RAM
> is still bigger than most machines' busses will hold,
> and even 2GB of DRAM is a lot bigger and more power-hungry than
> most laptops can really support, much less 2GB of fast SRAM.

There used to be a rule of thumb that you'd want roughly a megabyte
per megahertz.  Today, with our 200+ MHz processors, we tend to have
considerably less memory than this.  My Gateway PC from a couple of
years ago came with a 200 MHz Pentium Pro but only 32 MB of memory.

Most PCs today are not well balanced architecturally.  They should
really have a couple hundred megabytes of memory.  Memory is cheap
enough today that this can be added, but the motherboard configuration
may limit the amount.  If you had this much memory, swapping to disk
would be a smaller problem.

As for the idea of running without a disk drive, see the handheld PDAs
like the PalmPilot and the Microsoft WinCE machines.  The problem with
these from the security standpoint is that the memory is non-volatile.
With a PC, we don't worry too much about junk in memory because we turn
it off occasionally.  We are concerned about the disk because that is
where the persistent storage is.

But with a memory-only machine, you end up using your memory as a de-facto
disk drive, so that the contents of memory become just as sensitive as
the contents of disk would be on a PC.  Someone who gets hold of your
PDA can dump out memory and find sensitive data which is stored there.
Theoretically you can use a password to protect data, but then there are
tradeoffs between security and ease of use, and you may end up leaving
data exposed for extended periods.

But speaking of using passwords to protect data, how about an encrypted
swapfile for a PC?  There are fast enough ciphers today that the virtual
memory system could encrypt data as it swaps to the disk, and decrypt
as it loads back into memory.  You'd type in the passphrase at boot time.

This would solve the problem of sensitive data leaking onto the disk via
the swapfile.  There could still be traces in memory, but at least this
most obvious leak would be plugged.  You don't have to wait for apps to
be rewritten to use secure memory allocation drivers, you get the benefit
immediately.

Can existing encrypting-filesystem drivers be used for swapfiles?

On Mon, Nov 30, 1998 at 06:00:02PM -0000, lcs Mixmaster Remailer wrote:
> 
> But speaking of using passwords to protect data, how about an encrypted
> swapfile for a PC?  There are fast enough ciphers today that the virtual
> memory system could encrypt data as it swaps to the disk, and decrypt
> as it loads back into memory.  You'd type in the passphrase at boot time.

	Why the hell would you need a passphrase or any persistant security
for something transient like the swap ?   Might just as well choose a
completely random key (from /dev/urandom perhaps) and make every effort
to erase and forget it on system shutdown or crash.  In fact an algorithm
that initialized a crypto engine and then forgot the key used would
be ideal, provided only that it remains possible to recover blocks of
swap out of order from the order they were written in (they are quite
likely to get swapped back in a very different order than they were
written out, so simple stream ciphers are hard to use).

	There is nothing in a swapfile of value beyond an instantation
of the OS, except of course for snooping and debugging crashes.

	The only real hastle with doing this in real OS's is that the
swap may get initialized before a lot of randomness gets collected on
startup.   One might have to start encrypting swap after it was enabled
(but probably before much actual swapping).



-- 
	Dave Emery N1PRE,  die@die.com  DIE Consulting, Weston, Mass. 
PGP fingerprint = 2047/4D7B08D1 DE 6E E1 CC 1F 1D 96 E2  5D 27 BD B0 24 88 C3 18

At 10:00 AM -0800 11/30/98, lcs Mixmaster Remailer wrote:

>There used to be a rule of thumb that you'd want roughly a megabyte
>per megahertz.  Today, with our 200+ MHz processors, we tend to have
>considerably less memory than this.  My Gateway PC from a couple of
>years ago came with a 200 MHz Pentium Pro but only 32 MB of memory.
>
>Most PCs today are not well balanced architecturally.  They should
>really have a couple hundred megabytes of memory.  Memory is cheap
>enough today that this can be added, but the motherboard configuration
>may limit the amount.  If you had this much memory, swapping to disk
>would be a smaller problem.

Some of us are more balanced...

My G3 Powerbook has 160 MB of RAM for a 240 MHz processor.

My recollection is that "Amdahl's Law" was only a rule of thumb within an
order of magnitude or so. I'm not a computer architect, so I don't what a
good ratio today would be.

--Tim May

"I swear to tell the truth, the whole truth, just the way the President did."
---------:---------:---------:---------:---------:---------:---------:----
Timothy C. May              | Crypto Anarchy: encryption, digital money,
ComSec 3DES:   831-728-0152 | anonymous networks, digital pseudonyms, zero
W.A.S.T.E.: Corralitos, CA  | knowledge, reputations, information markets,
Licensed Ontologist         | black markets, collapse of governments.

lcs Mixmaster Remailer wrote:
> Most PCs today are not well balanced architecturally.  They should
> really have a couple hundred megabytes of memory.  Memory is cheap
> enough today that this can be added, but the motherboard configuration
> may limit the amount.  If you had this much memory, swapping to disk
> would be a smaller problem.

Unless you're running NT. I have a box with NT server and 256MB RAM, and
I'm pretty sure about 100MB of that has never seen any data. The disk
still rattles away, even though the machine is typically left switched
on. For example, if you run Word, log out, then run Word again, it still
seems to go for the disk. 

Cheers,
Frank O'Dwyer.