Hi Benjamin and Jeff,
Following our discussion in Montreal, would the following, or something along
these lines, be OK with you in the Security Considerations section.
When BFD clients are used to modify BFD configuration (as described
in Section 2.1), any authentication and authorization for the BFD
configuration changes have to take place in the BFD clients. For
example, if the BFD client is an IGP then the IGP SHOULD be
authenticated. Also, consideration should be given to the access control of
the BFD clients.
Regards,
Reshad.
On 2018-07-15, 9:05 AM, "Benjamin Kaduk" <ka...@mit.edu> wrote:
On Wed, Jul 11, 2018 at 11:32:18AM -0400, Jeffrey Haas wrote:
> Benjamin,
>
> On Wed, Jul 11, 2018 at 10:02:41AM -0500, Benjamin Kaduk wrote:
> > "may be overkill in some circumstances" sounds exactly like an RFC 2119
> > SHOULD, does it not?
>
> Putting it slightly a different way, I am always wary of trying to embed
too
> much operational and security wisdom in documents for the following
reasons:
> - What's wise in one set of circumstances may not be in another. By being
> proscriptive, you may lead to implementations that lack necessary
> flexibility.
In my opinion, including guidance with the supporting motivation suffices
to leave flexibility for unanticipated future cases with differing needs.
> - You're imposing a level of fate binding between mechanisms that may
> contravene desired behaviors from some operators that have split
> operational roles.
If the stated motivation does not apply to operators with such split roles,
do we not think they are smart enough to see that the prerequisite is not
met and ignore the advice?
> [...]
> > To frame the same idea in a different fashion, we have this nice
security
> > considerations boilerplate for YANG modules, talking about how the usual
> > access methods are NETCONF/RESTCONF, with MTI secure transport of
ssh/TLS.
> > The scheme being described here is effectively providing a new access
> > mechanism (IGP) for a subset of the YANG module,
>
> This is perhaps my personal disconnect.
>
> Much of the point of providing a common configuration grouping for BFD for
> client protocols was to encapsulate, "I'm a client of BFD, here's my
> parameters". An implementation is free to use the "please use bfd with
> these parameters for my protocol" or perhaps ignore them. But in
> circumstances that an operator may wish to limit access to protocol BFD
> behavior, it has the existing ability in NACM to enforce its policy on
those
> BFD nodes within the protocol tree.
>
> What I feel you're saying is we need to call special attention to these
> instantiations that may be imported by some module.
>
> What I'm confused about is why such an import is any more special than any
> other import from another module.
I've been trying to wrap my head around your explanation for the past few
days, and I'm not sure I'm succeeding at it. The only reason I'm raising
this point with this document is because there is text in the document like
"[f]or example, when a new IGP peer is discovered, the IGP would create a
BFD session to the newly discovered peer and similarly when an IGP peer
goes away, the IGP would remove the BFD session to that peer." Imagine if
I was writing a document about a device that controls a physical door, and
the usual way to operate the device is to manually enter a PIN while
physically in front of the door. If I also said "some people expose this
door-unlocking device to the internet and accept unlock requests over
HTTP", that would be incredibly unresponsible of me unless I added some
extra qualifier. Perhaps it would be "and these people are crazy", or
perhaps "but HTTP itself is insecure, so in such situations TLS ought to be
used with mutual authentication, authorization checks for the party
requesting unlocking, and the best practices from RFC 7525".
So, in my head, I see this document as using a not-quite-throwaway example
to make a point about the limitations of the main focus of the document,
and should properly qualify the different security properties of that
example. Your description above still reads to me as if it's focusing on
YANG modules not specified in this document but that also use the same BFD
grouping. In such a case, when the NACM applies, isn't that still going
through normal management paths for that respective YANG module? I'm still
trapped in a rut thinking about "received IGP packet from the network that
instantiates a new IGP session; as a side effect instantiate a BFD session
as well", where the incoming IGP packet is just the IGP implementation and
not a "management function" per se.
-Benjamin
