Re: ISMS working group and charter problems

[Eliot Lear <[e-mail]>]

Margaret,

My apologies for my delayed response.  I've been away.  The timing of
this charter review was unfortunate.

> (1) I don't think that call home fits within the scope of the ISMS
> group.  I am not necessarily saying that we shouldn't do this somewhere
> in the IETF, just not in this WG.

> 
> This is a security area WG.  It is not charged with producing a
> next-generation of SNMP, just with the (important) task of integrating
> the current generation of SNMP with more widely-deployed security
> mechanisms.  I personally think that tightly-scoped WGs are good.  I
> also think that it would be inappropriate to add new architectural
> features to SNMP in a security area WG.

How many times shall we change SNMP?  ISMS will represent a substantial
change, no matter how it is done.  There will be a new packet format,
and there will be a new authentication architecture for SNMP.  Whether
or not somebody's idea of an architecture for SNMP is changed at this
point is irrelevant.  All tools that use this solution will change.

Doing yet another round of changes to accomplish CH will cause a
continuously unstable network management standard for some period to
come.  It is at best an overly bureaucratic reasoning that says that X
work has to be done in X area when Y group has already opened the
protocol in Y's area.

As I've previously written, all the interest parties from the O&M area
have been in the room.  As others have written, CH requires security
review.  What, then, is the problem?

> 
> (2) I do think that call home represents a significant architectural
> change to SNMP, for the same reasons that Randy Presuhn has offered. I'd
> also like to emphasize one of his points -- we have already defined how
> SNMP runs over TCP (in RFC 3430), so ISMS is not the first group to
> consider this.  Running SNMP over TCP did not require the types of
> operational changes that you are discussing.

Beyond what I wrote above, as others have written, a substantial
architectural change is moving from packet based to session based
security.  Furthermore, had I been paying attention to the SNMP over TCP
discussions at the time I would have raised this concern at that time.

> IMO, the idea of an SNMP command responder (agent) initiating the
> communication is not as simple as you have portrayed it.  There are many
> open questions, like:  How would the command responder know what
> manager(s) to contact? When should the command responder try to initiate
> the communication? How often should it re-try if the initial contact
> fails?  What should it do if the connection drops?

These are valid questions, and worthy of discussion.  They are not
insurmountable questions, however.  I could right now give you my own
proposed answers for each of these, but I propose that they are best
answered in a working group!

> 
> SNMP command responders are stateless entities that simply respond to
> the queries they receive in the order that they receive them. Changing
> them so that they can initiate and effectively manage a communication
> connection to the manager is a pretty big change, IMO.

No it is not.  All that is required is an understanding that responses
to a manager will travel via a pre-established SSH connection.  For
example (and only example), we have an SNMP-TARGET-MIB that already
identifies in the case of a notification the method to communicate.
Conveniently the aforementioned TCP communication has great examples of
precisely how to populate this MIB in its appendix.

> 
> (3) I am not certain that there are a large number of people who want to
> initiative SNMP management communication through a NAT/Firewall who do
> not have the ability to allow/tunnel an SSH connection through that same
> NAT/Firewall.

I believe you are conflating two issues here.  The first is that one can
use SSH tunneling over which to route SNMP queries or responses.  The
second is the ability to cause the remote command generator to use that
tunnel.

> 
> In my experience, NATs and Firewalls are usually placed at the
> boundaries of an administrative area.  In many cases, they are
> intentionally used to prevent access to internal systems from the
> outside.  I have not run into many cases where a network administrator
> has wanted to allow management access to systems from outside of the
> local network, but perhaps you have?

I have, as has Eric, as have the many others who have written to you to
say this at this point.  Any overlay network or service provider that
wishes to offer a managed service to his or her customer will have run
into this problem.  Any enterprise with mobile devices will run into
this problem, particularly with phones.  Now you can say that
enterprises should simply bring up VPNs to solve the problem, and while
that's likely to sell more Cisco hardware, it may not be the right
solution for everyone, particularly if the phones are managed by a vendor.

> 
> If you do believe that this is a common need, I think that you will need
> to better define/explain your use cases.

Margaret, I'm surprised that the one use case I mentioned in my original
note did not resonate with you.  If you run Windows or MacOS or most
commercial variants of Linux your own computer already calls home to
have certain functions managed.  Why is it so difficult to believe that
a device within a network would want to accomplish the same objective
with the existing SNMP framework?

Eliot

_______________________________________________
Ietf mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/ietf