Re: iSCSI: Proxies (was Security in iSCSI)

["Jim Hafner" <hafner@almaden.ibm.com>]

David,

I entirely agree with all the points you make.  I have argued that there is
in the current draft no definition of a proxy, no model for such things and
only vague references to them.  I have even suggested that the term "proxy"
and all related text be removed from the document (including the vague
status code of "proxy required").

It turns out that for the purposes of iSCSI, proxies can be implemented
without any change to the protocol (provided you have a fairly intelligent
iSCSI-aware proxy).  All you need to do is have the actual target advertise
(via th normal discovery mechanisms, like SLP, iSNS, admin, etc.) that it's
network addresses are those of the proxy machine and only allow the actual
proxy to make direct connection to the target (the target would not
authenticate and allow login from any other source).  You get proxy
function without any protocol changes.  Security is initiator to proxy (by
the rules configured for that proxy) and proxy to target (by the rules
configured for that proxy-target pair).  The rest is completely transparent
to the protocol and to the host.   This description fits yours where the
proxy participates in the security context directly.

If this model of proxy requires too much function in the proxy boxes, then
we can revisit this issue latter in the iSCSI layer directly.  In the
meantime, I vote (with you) that we put a lid on this issue of proxy (by
removing the notion) and get on with completing the hard issues that are
still on the table for basic operation.

Jim Hafner


Black_David@emc.com@ece.cmu.edu on 08/23/2001 10:14:30 am

Sent by:  owner-ips@ece.cmu.edu


To:   Julian Satran/Haifa/IBM@IBMIL, ips@ece.cmu.edu
cc:
Subject:  iSCSI: Proxies (was Security in iSCSI)



Julian,

> You seem to missing my main point. I am ready to take out the
> table but not the general notion of a cryptographic digest.
> To do the later I have to have the community understand that
> their are left with no protocol provided means to authenticate
> data that pass through iSCSI proxies.

Actually, I was hoping that you weren't going there.  Just when
one thinks one has obtained a good understanding of a network
technology/protocol/solution/etc., adding the notion of proxies
is almost certain to restore the previous state of confusion.
Multicast also has this property, but fortunately, I haven't
seen anyone in the WG take any serious interest in it (and
woe betide anyone who takes this as an invitation ;-) ).

My primary concern is one of schedule - if full support for
proxies is a requirement for the first version of the iSCSI
RFC, the schedule guesstimate I posted yesterday is probably
optimistic by 6+ months.  Issues that come up in the area of
proxies that will take a while to resolve will likely include:

- What's the right scope of proxy support?  How much
     consideration for SCSI level proxies to which other
     transports needs to be reflected in the iSCSI standard?
- SCSI is not an end-to-end transport independent protocol.
     T10 had the opportunity to go in this direction in the
     SCSI GPP work) and chose not to do so. Extensive iSCSI
     work on proxies risks running into SCSI architecture,
     and our experience with ACA suggests that they'll take
     a while to resolve.
- Security interaction with proxies is a very complicated
     subject.  IPsec and NATs is a well-known tarpit that
     is finally getting resolved years later.  I hesitate
     to promise that the iSCSI proxy issues will be significantly
     simpler by comparison.
- There is precedent for proxies adapting to the requirements
     of protocols - TLS requires http proxies to accommodate
     it rather than the other way around.  Whether proxies
     should drive iSCSI security architecture or iSCSI
     security architecture should drive proxy design will
     at the minimum be fodder for serious debate.

If we want to get the iSCSI spec done in a reasonable timeframe,
I suggest closing the lid on this Pandora's box labeled
"proxies" and deferring serious work on iSCSI proxy support
to a future revision.  The separate header and data CRCs are a
useful step towards making life better for those who want to
build proxies, but I hesitate to go much further in that
direction, particularly in the area of security.  As things
currently stand, one way to authenticate data passing through
an iSCSI proxy is to have the iSCSI proxy participate in the
authentication which seems like an acceptable situation (and
one that has a straightforward security explanation).

Comments?

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
---------------------------------------------------