Re: ISCSI: User authentication vs. Machine Authentication for iSC SI

["John Hufferd" <hufferd@us.ibm.com>]

Mark,
You clearly understand part of the problem well.  You need to step a bit
further.
Just recommending the names be the same, does not address what happens
when they are not.  So let me take you a bit further.

A system that has an iSCSI Initiator Node Name "iqn.com.mother.jump"
is given the UserID of "Wonder".
Now assume that "Wonder" is Authenticated. Now the iSCSI client system with
UserID
of "Wonder" can now masquerade as any other Authorized iSCSI Initiator
Node, by putting that name on its Login.

This seems like it is not a real good idea, so the target system needs to
prevent this.  That means that the Target System needs to bind together
in some way the UserID and the approprate iSCSI Initiator Node Name.

So only recommending that the UserID and iSCSI Initiator NodeName
be the same means that everyone MUST implement a relationship
binding Database that can be used to ensure that no one can masquerade
as some other iSCSI Initiator Node.

This also means that it needs to be tied into the UserID authorization
Database/Directory such that when name etc. are change (which
often they do), the iSCSI Target Device relationship/Binding Database
is also updated.  If this is a manual process, it will get out on hand
very quickly.

For those of you that think this is a very simple one to one table, you
might want to read Marks note again, he said that he could have
a different UserID for each network he logs into.  If there is any
cross connect to a specific iSCSI Target this means that the
iSCSI Target Device needs to perhaps carry more then one
UserID for each iSCSI Initiator Node Name.

That may not be a significant problem since the networks
may never interconnect.  Opinions would be useful here.

Is their any other reason that an iSCSI Initiator Node would  have
more then one UserID?

Does anyone have a technique to ensure that only the iSCSI
Initiator Node Name need be managed?



.
.
.
John L. Hufferd
Senior Technical Staff Member (STSM)
IBM/SSG San Jose Ca
Main Office (408) 256-0403, Tie: 276-0403,  eFax: (408) 904-4688
Home Office (408) 997-6136
Internet address: hufferd@us.ibm.com


Mark Bakke <mbakke@cisco.com>@cisco.com on 08/30/2001 01:07:35 PM

Sent by:  mbakke@cisco.com


To:   Stephen Bailey <steph@cs.uchicago.edu>
cc:   ips@ece.cmu.edu, John Hufferd/San Jose/IBM@IBMUS
Subject:  Re: ISCSI: User authentication vs. Machine Authentication for iSC
      SI



Stephen Bailey wrote:
>
> > The new Name (UserID) is exactly what was implied at the meeting in
> > Irvine.
>
> Oh.  How horrible.

Ideally, you're right.  For instance, if iSCSI was using AuthMethod=CHAP,
then it would be great if the CHAP username (N) specified was identical
to the InitiatorName.

In the draft, that's not how it is.  The user names for CHAP and SRP
(I'm not sure about the others) are specified separately from the
InitiatorName, so there's currently nothing in the protocol to prevent
them from being different.

In practise, there might be times when having them be different
makes the "fit in" better with whatever administration scheme is
out there.  For instance, if my desktop was assigned some iSCSI
storage, and my username and password were kept on a RADIUS server
somewhere, for use by file servers, web servers, and iSCSI servers
or devices, my user name is likely to be a non-globally-unique
string, probably "mbakke".  However, that's not likely to be
my InitiatorName.  To make them the same, I would either have to
change my InitiatorName on my desktop to "mbakke", to fit in with
my login on all of our web and file servers, or the admin would
have to create a new login for my iSCSI stuff on the RADIUS server,
and maintain two of them.  Changing my login on everything to
match an iSCSI name is likely not an option.

Another possible place this could happen is if I am obtaining
iSCSI "services" from more than one administratively-separate
environment.  Each of these environments may prefer to assign
me a user name for login that matches their own internal scheme.
I may have some control over this, but in the end, it has to be
unique within the service-provider's scheme.  This is very
similar to setting up a user-name for your ISP at home; you
may request a particular name, but if the ISP already has one,
you have to pick again.

The point is that if one had two of these environments providing
services, they may each need a separate user name.

Of course, the latter example could be solved by having the
admins accept the initiator name that's already on my iSCSI
client as the user name; these are supposed to be unique
anyway.  The only trouble then is the same as before; if I am
getting more than just iSCSI services from this provider, they
would then have to maintain two user names for me.

A third reason this might happen is that some centralized
password authenticated services such as RADIUS may have
length restrictions on their user names that are shorter than
what is allowed in an iSCSI initiator name.

Anyway, implementations have the most flexibility the way
these protocols (SRP, CHAP) are currently documented.  I think
that we could recommend that the iSCSI initiator name be the
same wherever possible, but I don't see how we can require it.

>
> > A number of us, I am sure, think this is a very poor direction for an
> > implementation.
>
> Agreed.  I just wanted to walk the space a little bit.  My point is
> that while traditional SCSI passthru does confer control privs (and is
> protected as such), it doesn't need to.  Specifically, a client using
> SCSI passthru is not actually given the identity of the host OS, per
> se.
>
> > So I believe we must consider such a potential application as
> > probably a rouge application and do nothing to help this, and work
> > to prevent it.
>
> I understand what you're saying, but I'm not sure it confers much
> architectural direction.  There already must be a concept of identity
> which is independent of anything bound to a particular system (e.g. IP
> address, hardware network port etc.), so there's no way to exclude
> that.
>
> > The important issue is, can we make the iSCSI Initiator Node Name be
> > used as the UserID.


Again, I think that the best we can do is recommend that they
be the same, but I don't think we can take the step of requiring
it.


> Agreed.  Are there minutes to the meeting?
>
> Thanks,
>   Steph

--
Mark A. Bakke
Cisco Systems
mbakke@cisco.com
763.398.1054