Thoughts on correct support for Windows Previous versions

[Jeremy Allison]

On Sun, Jan 25, 2015 at 05:49:35PM -0800, Richard Sharpe wrote:
> When we implemented support for Windows Previous Versions at Panzura
> we started with one of the existing Shadow Copy modules and adapted it
> to the SnapShots provided by the file system (ZFS-based), and life
> seemed good.
> 
> Then in came the bug reports from QA and we were forced to check what
> Windows actually did.
> 
> At first I assumed that Windows would return an error like
> STATUS_ACCESS_DENIED if one tried to open a previous version for
> write-like access, eg, opening a previous version in Notepad or
> wordpad, but this turned out to be incorrect. In fact, Windows allows
> you to do that, at least according to our tests, but returns
> STATUS_MEDIA_WRITE_PROTECTED.
> 
> We also found that Windows will ask for FS_INFO on open files,
> presumably to check that the volume is write protected.
> 
> This does lead to some interesting problems, because it means that
> when the client asks for WRITE access to a file and the ACL allows it,
> you then try to open the file for WRITE access, but the file system
> denies it (or at least ours did) and returns EROFS. You have to turn
> around and open the file RO and pretend that you opened it for WRITE
> to get the correct behavior.
> 
> Unfortunately, at Panzura we modified Samba too much, and I am now
> convinced that most of the changes could have been done in the VFS
> module we had.
> 
> However, there are a few small changes I think are needed. One is to
> add a field to the FSP to record whether or not a file is a previous
> version. This would only ever be set by a VFS module providing
> previous versions functionality, but it seems simpler than adorning
> the FSP with an extension and is needed in one place in Samba.
> 
> The other one is to pass an FSP into one function and to return the
> correct value when Windows asks for FS_INFO on an open file.
> 
> Attached is a patch for discussion that makes those changes.
> 
> However, this might not be enough because VFS calls like ntimes does
> not get passed an FSP (something I want to fix at some time as well,
> as soon as 4.2 is finally cut.) (The only way to return
> STATUS_MEDIA_WRITE_PROTECTED from VFS modules is if all functions that
> implement WRITE-like behavior get access to the FSP.)
> 
> Over the next few weeks I will try to get a capture of the behavior of
> Windows with respect to getting FS_INFO on an open file to document
> this.

Yep, that's the next step. We need to know exactly what
behavior we need to emulate to make the Windows Previous Versions
work like a Windows server, then I'll be happy to add tests
and make any needed VFS changes.

Cheers,

	Jeremy.