Re: [RFC 0/28] Patches to pass vfsmount to LSM inode security hooks

[Christoph Hellwig <hch@infradead.org>]

On Wed, Feb 07, 2007 at 01:58:15AM -0800, Andreas Gruenbacher wrote:
> > It's not actually a pathname we care about, but a vfsmount + dentry
> > combo.  That one means as much in nfsd as elsewhere.  We want nfsd
> > to obey r/o or noatime mount flags if /export/foo is exported with them
> > but /foo not.
> 
> This assumes that you can discover from the file handle which export it 
> belongs to. And right, this can be done using the fsid export option. Okay, 
> I'm now convinced that passing the vfsmounts from the svc_export to the vfs_ 
> helpers is good at least for *something*. I'll update the patches.

As I mentioned before I have a big patchset reworking this area, so please
wait a week or two for me two finish it instead of creating lots of rejects.

> > This doesn't matter.  hardlinks are per definition on the same vfsmount.
> 
> Hardlinks are per definition on the same filesystem. This doesn't mean that 
> you can't have two filehandles for the same file but with different 
> vfsmounts -- you can have the same file visible in multiple locations, each 
> with their own vfsmount.

Sure, but that doesn't matter here.  Exports are per-vfsmount, and
each export has the vfsmount stored in it.  So we know which of the
vfsmount we want before we even resolve the dentry.  And yes, if
I have /export/foo mounted with option but /foo now I want those
options to apply to the nfs export.  Note that we actually have another
little bug in this area currently:  if you export two different
patch that have the same underlying filesystems they actually do get
the same file handle unless using fsid=.  We probably need to put
a check into svc_export_parse so that we don't export the same
device twice without using fsid=.