RE: Security Gateways

[Robert Snively <rsnively@Brocade.COM>]

Dear David,

I would like to address the rather serious question you have
raised here about both architectural layering and market
requirements.  After considerable thought, based on the
explanations below, I have concluded that the FCIP RFC should
specify that security is optional to implement and optional to use.

1)  Concerning architectural layering:

It is clear that the problems of security are well understood
within IETF. Numerous solutions have been proposed and
many of those have become standard implementations.  The most
successful of those (as an example, IPSec) have used the
principles of architectural layering very well.  IPSec
provides a secure mechanism for transporting any IP-based
protocol if the protocol requires such security.  However,
the overlaying protocols (as an example, TCP) have no requirement
to actually make use of IPSec, and in fact have no particular
requirement to make use of IP.  This is as it should be.
Each layer is implemented independently of the other.  The
services that each layer implements depend on the market
requirements for the environment.  By that logic, it is 
unnecessary for FCIP to specify any security mechanism at all, 
since so many security mechanisms are available.  However,
FCIP has provided the additional guidance that, if security
in a TCP environment is used, it shall be provided at a layer
below the TCP interface.  In particular, it recommends the
use of IPSec for TCP/IP environments (and will recommend the
appropriate IPSec options).  This additional guidance has nothing
to do with the architecture or definition of FCIP and is 
merely a recommendation that should improve interoperability.

2)  Concerning market requirements:

A very high percentage of storage environments today manage
their configurations very carefully.  Such careful management is
necessary to guarantee redundant paths for proper availability,
to provide sufficient paths to provide the required performance, and to
guarantee known paths to improve reparability and consistency
of behavior.  As a side effect, a very high percentage of
the paths of a storage environment are physically secured and have
no requirement for additional security mechanisms.  At present,
the only paths that are using security are those very few paths
that leave one physically secure environment to transport data to
another physically secure environment.  Those few paths almost universally
use security mechanisms at a lower level (as an example, a virtual
private network) to achieve their goals.  As a first guess, such
paths are well under 1% of the storage paths being used today.

I do not believe that this paradigm will change significantly over
the next several years, although of course the percentage of paths
requiring transport security may rise over 1%.  The requirements for
storage performance and the desire to maintain the storage devices
and many of their primary processors in a physically secure
environment will assure similar configurations.

In this environment, the market will demand low cost and high
performance for the great majority of interconnects and will demand
security for a relatively low percentage of interconnects.  

What this implies for technologies like FCIP is that security
be available as a separate option, outside the primary FCIP
definition.  While still allowing the integration of security
inside an FCIP unit, the IETF draft for FCIP  must also allow security
to be provided either not at all or by an outside mechanism.  
The most common outside mechanism would be a gateway box of some sort.
In many cases, the gateway box will be a secure router placed on the
external link and servicing a large number of locally attached
unsecured FCIP boxes.

CONCLUSIONS:

The FCIP draft should continue to specify a security mechanism,
with IPSec being the appropriate candidate, to guarantee interoperability
when security is provided.  The FCIP draft should be modified to
indicate that security is optional to implement and optional to use.
In addition, it should explicitly indicate that external gateway
boxes are allowed implementation mechanisms for security.

As a side question, note that management of storage area networks
and FCIP boxes is a separate question and is outside the scope of
the FCIP draft.  Both Fibre Channel and Ethernet management paths
have security appropriate to the particular management mechanism.
As an example, web-based management mechanisms use web-based
security tools. 

Bob Snively                        e-mail:    rsnively@brocade.com
Brocade Communications Systems     phone:  408 487 8135
1745 Technology Drive
San Jose, CA 95110



-----Original Message-----
From: Black_David@emc.com [mailto:Black_David@emc.com]
Sent: Tuesday, July 24, 2001 7:11 PM
To: ips@ece.cmu.edu
Subject: Security Gateways


The following issue was hidden in my long set of
comments on the -03 version of FCIP:

> > Delete 12 b).  If an FCIP entity is operating with an external
> > security gateway, only the interface on the public side of the
> > gateway is compliant with this specification.  The interface
> > between the FCIP entity and the gateway is not compliant because
> > it is lacking required security features - the FCIP entity
> > *includes* the security gateway in this structure.
> 
> Please post this as a separate issue because several of the
> FCIP Authors believe it is appropriate for FCIP and I cannot
> represent their opinions.

The issue is not whether it's "appropriate".  The issue
is that if an implementation uses an FCIP Entity plus
an external security gateway, the only interface that
conforms to the forthcoming RFC is the public/external
interface on the security gateway.  The interface between
the FCIP Entity and the security gateway is private
and fails to conform to the security that will be
required of all FCIP implementations.

The above paragraph also applies to iSCSI (substitute iSCSI
for FCIP in all instances).  Let me also note that iSCSI's
ability to use a security gateway is not final at this
juncture.  The spectrum of security possibilities includes
things like SRP keying of ESP and IPsec transport mode that
would make external gateways difficult or impossible to use.

Those who care about being able to use security gateways
(or think that there's no need to support their use)
should speak up on the list, in London, and/or in Orange
County (I would expect the decision not to be made prior
to Orange County) and *EXPLAIN WHY* [technical rationale].

Thanks,
--David

---------------------------------------------------
David L. Black, Senior Technologist
EMC Corporation, 42 South St., Hopkinton, MA  01748
+1 (508) 435-1000 x75140     FAX: +1 (508) 497-8500
black_david@emc.com       Mobile: +1 (978) 394-7754
---------------------------------------------------