All, 

This question is interesting an esoteric way, as it illustrates a point
about CICM support for high assurance (HA*).  Encrypted objects possess
security properties that are not compatible with unencrypted (or
decrypted) objects.  There are numerous other examples of objects in a
generic SDR that have security properties that are not compatible with
other objects.  

To ensure secure operation (in general) I know of at least two
alternatives.  1. Trust the application software to do right (whatever
that is), and this is repeatedly debunked, (but we still pursue it
anyway) and 2. Define "secure behavior" in a Security Policy and enforce
it by a HA OE from beneath the application SW (the only way that has
really ever really worked for HA in practice).  

Alternative 2 above implies objects with compatible security properties
need to be grouped in security domains and isolated from other
incompatible security domains by a HA OE, and the HA OE must prevent
some of the flow in some directions between some of these domains,
(allowable flow being defined by a Security Policy) including precluding
some of the "echoing" suggested by the question.  I claim there exist a
structure for security domains in generic secure SDRs that does not
require harmful flow (obviating the need to "bend rules" with CDS
guards) among security domains and this structure is required for secure
behavior.  We have a science that produces this structure.  Furthermore,
I claim guards that "sanitize" objects are not logically sound in these
cases.  (These admittedly controversial claims are supported in an IEEE
paper "On The Architecture of Secure SDRs"
<http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=04753440>. ) 

It seems to me our APIs support flow that we do not need in a securely
structured SDR and should not really be allowable in secure SDR OE,
whether or not it is HA.  (*HA is just a term that implies a
successfully enforced security policy.)  My concern is that when we
assume the HW view of a crypto, that is red side / black side, we lack
enough "sides" to get the SDR job done without bending the rules.  It
worked in HW cryptos that connected physically isolated functions with
physical wires but when we emulate that in SW along with additional
radio functions with APIs, we really do need APIs that bend rules.  This
gives rise to questions like this.  I am still hoping to propose a
"guidance" section for CICM on how to structure an SDR to avoid these
issues.  

So that is my opinion. 
Thanks, 
John  

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of
Nanjundiah, Girish
Sent: Friday, May 27, 2011 3:35 PM
To: CICM Discussion List
Subject: Re: [cicm] Moving Packets and Storing Identifiers (was
RE:CICMQuestions)

Hi Lev,

Thanks for the answers. In regards to the answer to the first question,
while we can define the actual mechanism to reflect back the packets
outside of the driver, the driver still needs to call the
function/mechanism that we define within its decrypt() function before
it can expect the packet to appear. Technically, this is still not
completely separated from the driver functionality. Is this still an
appropriate place to echo back black packets?

Thanks,
-Girish 

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of
Novikov, Lev
Sent: Friday, May 27, 2011 12:47 PM
To: CICM Discussion List
Subject: [cicm] Moving Packets and Storing Identifiers (was RE:
CICMQuestions)

Hi Girish,

On 2011-05-27 at 14:50, Girish Nanjundiah wrote:
> 1. When the function encrypt() sends a packet to the Crypto, it
emerges on 
> the black side encrypted and decrypt() is expected to decrypt this
encrypted
> packet. In order for the packet to be decrypted, there needs to be
some 
> function that echoes the encrypted packet back through the black side.
Since
> this echoing back seems to be more of an application level function
than a 
> driver function, what part of the code should be responsible for
echoing the
> encrypted packet back through the Crypto?

Another way of asking this question is: What functions move data from
the 
remote (unprotected) side of a crypto into the local (protected) side?

See also:
"Does CICM only provide (specify) the "red" side API in a multilevel
system?"
http://www.ietf.org/mail-archive/web/cicm/current/msg00014.html
http://www.ietf.org/mail-archive/web/cicm/current/msg00028.html

You correctly point out that this is not a driver-level function. In
fact,
CICM does not define functions for simply moving data into the crypto. 
Therefore, you are free to use whatever transport mechanism works for
you 
(e.g., POSIX socket).

> 2. When create_en/decrypt_conduit is finished executing, it needs to
store 
> an identifier (just a number really) to identify the conduit it has
created 
> with the Crypto. Since both of these functions only return a status
and a 
> CICM::En/Decrypt::Conduit pointer, the only way to store the
identifier for 
> the conduit to use is to add it as a member variable to the Conduit
class. 
> If the variable is to be private, we would also need a simple public
member 
> function to access it. Is there a way to update the CICM API so that
we can 
> store the conduit's identifier in one of the ways I listed?

This is an interesting suggestion. We define KeyId for key identifiers,
but we do not define a ChannelId for channel identifiers. This is
because isn't currently a way to lookup a channel by its identifier
(like there is for keys, modules, and tokens).

This is going to require some discussion. Normally, vendor-specific
attributes are defined by extending the CICM base object and adding
those properties. However, it seems like a common operation to store a
vendor-specific identifier in a CICM object to make it easy to reference
the underlying object later on.

** What do people think about extending the API to allow vendors to
store a 
   single numeric value to uniquely identify the objects to the
underlying 
   system?
   
Lev
_______________________________________________
cicm mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/cicm
_______________________________________________
cicm mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/cicm
_______________________________________________
cicm mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/cicm

Reply via email to