Re: [nfsv4] I-D Action: draft-ietf-nfsv4-integrity-measurement-01.txt

[Chuck Lever <[e-mail]>]

> On Sep 26, 2018, at 6:12 PM, Bruce Fields <[e-mail]> wrote:
> 
> On Wed, Sep 26, 2018 at 01:27:41PM -0700, Chuck Lever wrote:
>>> On Sep 26, 2018, at 1:21 PM, [e-mail] wrote:
>>> 	"A keyed hash authenticates the identity of the last modifier of
>>> 	a file's content and serves as a strong check of the content's
>>> 	integrity.  For the purposes of this document, we will refer to
>>> 	this as file provenance information.  The Linux Integrity
>>> 	Measurement Architecture (IMA) is an example of a mechanism for
>>> 	assessing file content provenance [IMA-WP]."
>>> 
>>> I think it's also true that this draft is sufficient to allow a client
>>> to support IMA?  That also seems worth mentioning.
>> 
>> I'm confused by your remark. I'm describing a transport
>> protocol between a NFS client and server. Why would it
>> not be sufficient to implement client support?
> 
> Oh, I was pretty sure it was, just thought it was worth making that
> explicit.  "The Linux Integrity Measurement Architecture (IMA) is an
> example of a mechanism for assessing file content provenance, and the
> protocol described in this draft could be used to implement IMA."  Or
> something like that.
> 
>>> 	"To ensure interoperability among accessors of this information
>>> 	when it is stored on NFS version 4.2 servers, this information
>>> 	MUST be self- describing."
>>> 
>>> How do you propose that an implementation would make the information
>>> self-describing?
>> 
>> That is out of scope for this draft.
>> 
>> But typically if an integrity service wanted to do this, it
>> could use XML or JSON to describe the information, for example.
>> 
>> An HMAC has a particular structure (RFC 2104) that should be
>> easy to recognize.
> 
> OK, but you do have to know how all the other clients are formatting the
> information.  I don't quite understand what's meant by ensuring
> interoperability, then.
> 
> I guess I expected to see something like say a 32-bit "type" field with
> implementations able to register different "type" values so you could at
> least tell whether it should bother trying to parse a given file
> provenance field.

I don't think it's helpful to insert the WG and IANA in the
process of deciding the type of information in the FPI payload,
especially when the structure of that payload is out of the WG's
hands.

Either the clients look at a 32-bit enum value, or they look at
the payload and figure out what it is. I can't see how it makes
any operational difference.


> Maybe it's not necessary.  I wonder what would would happen in practice.
> I geuss if people deployed two different clients that both claimed
> support for file provenance, they might be surprised when they started
> randomly throwing errors when one client tries to read files written by
> another.

Typically, an NFS server serves a fleet of clients that have a
common administrator who provisions and configures these clients
to recognize the same FPI payload.

The consequences of not recognizing the file provenance information
are not dire. In those cases, clients simply don't allow access to
the file's content, and the implementations are free to report to
the administrator that the FPI format is not recognized.

Would it help to state that in the draft?

--
Chuck Lever