Re: [PATCH v2 1/5] fat: allocate persistent inode numbers

["J. Bruce Fields" <bfields@fieldses.org>]

On Thu, Sep 13, 2012 at 02:38:11AM +0900, OGAWA Hirofumi wrote:
> "J. Bruce Fields" <bfields@fieldses.org> writes:
> 
> >> On other view (as server side solution), we are thinking there is
> >> possible to make the stable filehandle on FAT if we disabled some
> >> operations (e.g. rename(), unlink()) which change inode location in FAT.
> >> 
> >> Yes, it would be stable, but supporting limited operations.
> >> 
> >> This is server side solution, and we comparing it with client solution.
> >
> > Is that useful to anyone?
> 
> Good question. I'm not sure though, Namjae is asking. And I was asked
> about stable read-only export in past.
> 
> >> >> LOOKUP return NFS FH->[inode number changed at NFS Server] ->
> >> >> But we still use old NFS FH returned from LOOKUP for any file
> >> >> operation(write,read,etc..)
> >> >> -> ESTALE will be returned.
> >> 
> >> Yes.  And I'm expecting as client side solution,
> >> 
> >> -> ESTALE will be returned -> discard old FH -> restart from LOOKUP ->
> >> make cached inode -> use returned new FH.
> >> 
> >> Yeah, I know this is unstable (there is no perfect solution for now),
> >
> > You may end up with a totally different file, of course:
> >
> > 	client:			server:
> >
> > 	open "/foo/bar"
> > 				rename "/foo/baz"->"/foo/bar"
> > 	write to file
> >
> > And now we're writing to the file that was originally named /foo/baz
> > when we should have gotten ESTALE.
> 
> I see. So, client can't solve the ESTALE if inode cache was evicted,
> right? (without application changes)

I don't see how.


As another server-side workaround: maybe they could try tuning the inode
caching to make eviction less likely?

Grepping around...  Documentation/sysctl/vm.txt mentions a
vfs_cache_pressure parameter.

--b.