Re: [nfsv4] No WG meeting for IETF 91

["J. Bruce Fields" <[e-mail]>]

On Sat, Oct 04, 2014 at 12:24:30AM -0400, Thomas Haynes wrote:
> On Oct 3, 2014, at 3:59 PM, Marc Eshel <[e-mail]> wrote:
> > I am not sure why we are wasting time arguing about the obvious, xattr are 
> > out there and used
> 
> Are they? How much?
> 
> The argument being presented is that we should support xattrs
> because they are there. 

That's overstating it; section 2

	http://tools.ietf.org/html/draft-naik-nfsv4-xattrs-01#section-2

*does* have examples, though though not with a lot of detail about any
particular example.

Personally my main worry was the confusion between xattr support and acl
and selinux support.  Because that's almost universally what people have
given me as use cases in casual interactions.

I'm mostly reassured that there are in fact people using them for other
things, though absent more evidence I'm still sort of neutral on whether
it's worth doing.

I guess what would really make it easy for me would be *one* example
with evidence and details showing how somebody's life would be improved
with xattr support.  Ideally a real "somebody" who would be sufficiently
motivated to actually test our implementation, so we'd have a pretty
good idea whether we'd met requirements or not.

So the reason I was suggesting e.g. scanning Linux filesystems or
looking at data dependencies would be that we might be able to scare up
some developer with a good use case.  And, I don't know, maybe that
could be easier than getting some customer to describe their internal
use.

--b.

> 
> What drives us to take the work? Where do we lose "customers"
> if we do not do xattrs?
> 
> Is it a feature that will be used once it is delivered?
> 
> 
> Here is a link which talks about them being not used as much:
> 
> http://www.lesbonscomptes.com/pages/extattrs.html
> 
> And while it helps the draft because it blames the lack of NFS support,
> the tone on various different online articles I've read today is that
> xattrs while present are not used that much.
> 
> Labeled NFS was in somewhat the same boat until the seVIrt use
> case was presented to the group.
> 
> 
> > and for NFS to stay relevant we need to make it support 
> > all file system features and xattr is supported by most file systems.
> 
> 
> Yes, it is a filesystem. It is not just a transport protocol. 
> 
> There were two things that enabled Labeled NFS to finally proceed:
> 
> 1) The spec detailed how different MAC implementations could co-exist.
> 
> 2) The requirements document broke down the responsibilities of the client
> and the server.
> 
> As it currently stands, the xattr draft either implicitly assumes that the
> only clients will be Linux based or ignores the interaction between different
> client implementations.
> 
> Also note that it helped that each of the MAC implementations that were
> considered for LNFS had well defined semantics (and in some cases
> actual specifications).
> 
> 
> > We 
> > should focuses and use the limited time that the WG member can spend here 
> > (like was said before, it is no one full time job) on how to implement the 
> > right support for xattr.
> 
> 
> If we only visit it every 4 months, nothing will be done.
> 
> 
> > 
> > Again, at some point we need to reach consensus and move on, I though that 
> > we passed that point but if some one object to the need of xattr to be 
> > part on the NFS protocol please speak up and if not lets just talk about 
> > how to do it. 
> > 
> > 
> 
> I've been objecting since it was first proposed.
> 
> What is the use case that makes us want to do it?
> 
> How do different client OSes interact?
> a) In namespaces
> b) Securely
> c) With the limits
> 
> 
> 
> 
> 
> _______________________________________________
> nfsv4 mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/nfsv4