Windows vs POSIX directory rename with open files behaviour

[Jeremy Allison]

On Mon, Jan 26, 2015 at 06:07:00PM +0100, Ralph Böhme wrote:
> On Fri, Jan 23, 2015 at 10:20:44AM -0800, Jeremy Allison wrote:
> > On Fri, Jan 23, 2015 at 05:47:44PM +0100, Ralph Böhme wrote:
> > > > > I'd say it's not a bug, but Apple is trying to squeeze out POSIX
> > > > > semantics.
> > > > 
> > > > Well that's what the POSIX capability bit is for.
> > > 
> > > We don't have that in an SMB2 context. :) I'm piggy packing POSIX
> > > behaviour on top of vfs_fruit, probably just like Apple is piggy
> > > packing it left and right as needed in order to make their POSIX
> > > client system happy.
> > > 
> > > > > The original bug report that shoved me into investigating this was
> > > > > saying that the OS X Finder occasionally wasn't able to rename folder
> > > > > on a Samba 4.1.16 server, for the reason that there were open files
> > > > > (.DS_Store, used by the Finder itself to store directory appearance
> > > > > properties).
> > > > 
> > > > Interesting. I think we need to know *exactly* what semantics
> > > > this requires before we try and emulate it.
> > > 
> > > Not sure to what *this* refers to here, but if I thing for the
> > > *rename* semantics we need exactly this: in an SMB2 context with AAPL
> > > extension allow directory renames without checking for open files in
> > > the filesystem tree rooted in directory.
> > > 
> > > > Sounds like a server bug to me.
> > > 
> > > I don't think so. Sounds like poor-mans SMB2 POSIX extensions to me. :)
> > > 
> > > > After all, those clients would fail against Windows in the same way.
> > > 
> > > Definitely.
> > > 
> > > > Unless it's the addition of the AAPL create context switch that
> > > > changes the behavior.
> > > 
> > > The behaviour is the same irrespective of AAPL create contexts.
> > > 
> > > Thanks for taking time and looking into this!
> > 
> > Well thanks for doing the work to track it down.
> > 
> > It's certainly an Apple server bug though.
> 
> A feature, I swear! :)
> 
> > Now question is, who are we bug-compatible with ? :-) :-).
> 
> I'd like vfs_fruit to be as bug compatible as possible to Apple's
> implementation.

Yeah, but in this case Apple's implementation is just buggy.