Re: [PATCH] nfsd: do not allow exporting of special kernel filesystems

[Christoph Hellwig <hch@lst.de>]

On Wed, Jan 21, 2026 at 11:35:11AM +0100, Amir Goldstein wrote:
> On Wed, Jan 21, 2026 at 11:12 AM Christoph Hellwig <hch@lst.de> wrote:
> >
> > On Wed, Jan 21, 2026 at 09:50:27AM +0100, Amir Goldstein wrote:
> > > pidfs and nsfs recently gained support for encode/decode of file handles
> > > via name_to_handle_at(2)/opan_by_handle_at(2).
> > >
> > > These special kernel filesystems have custom ->open() and ->permission()
> > > export methods, which nfsd does not respect and it was never meant to be
> > > used for exporting those filesystems by nfsd.
> > >
> > > Therefore, do not allow nfsd to export filesystems with custom ->open()
> > > or ->permission() methods.
> >
> > Yeah, this was added in and not used in the existing export_ops users.
> >
> 
> Not used in existing users (nfsd) on purpose to my understanding
> That's the point of this patch - to fix this misunderstanding

Well, I've dug through the documentation and commits that added it,
and there's absolutely nothing explaining the intent unfortunately.

> >
> > Please spell out here why ->open and ->permission are not allowed.
> > Listing what the code does is generally not that useful, while why
> > it does that provides value.
> 
> This is what I had in the RFC patch:
> /*
> + * Do not allow exporting to NFS filesystems with custom ->open() and
> + * ->permission() ops, which nfsd does not respect (e.g. pidfs, nsfs).
> + */
> 
> I took Chuck's suggestion to rewrite the requirements, but TBH,
> I'd rather not touch the existing comment myself at all.
> I prefer that Check and Jeff apply a separate patch to rewrite the
> documentation if they feel that this is needed or to propose the
> phrasing that they prefer.

I don't really care who writes the documentation, but if we reject
based on the presence of the methods we really need to document
the why.  

> > While looking this I have to say the API documentation for these
> > methods in exportfs.h is unfortunately completely useless as well.
> > It doesn't mention the limitation that it's only used by the
> > non-exportfs code, and also doesn't mention why a file system
> > would implement or have to implement them :(  The commit messages
> > adding them are just as bad as well.
> 
> I will leave that to Christian for a followup patch or to suggest
> the phrasing.

Yes, I think we really need a braindump from Christian here.  If
you don't want to add that to the documentation I'll find some
time to include it into a revision, because documenting these
kinds of things is really essential.  We're already confused only
2 years after this was added, and it's only going to get worse.