Latest leases patchset - getting there !

[Jeremy Allison]

On Fri, Nov 14, 2014 at 08:29:00AM +0100, Stefan (metze) Metzmacher wrote:
> Am 14.11.2014 um 08:12 schrieb Jeremy Allison:
> > On Thu, Nov 13, 2014 at 08:09:33PM -0800, Jeremy Allison wrote:
> >> On Fri, Nov 14, 2014 at 03:02:19AM +0100, Stefan (metze) Metzmacher wrote:
> >>>
> >>> This is fixed with todo19.diff.txt :-)
> >>>
> >>> There we just need to check what to do in the timeout handler.
> >>
> >> W00t! Great Metze !!!!!
> >>
> >> I think that's the last outstanding issue, right ?
> > 
> > FYI. I'm making progress with the breaking2
> > test.
> > 
> > I understand what is going on, but not quite
> > sure yet how to get to what is needed...
> > 
> > The 'will_overwrite' is the key. I think we
> > can separate out the lease break inside
> > delay_for_oplock() when will_overwrite
> > is true from a direct break to NONE,
> > to a break to READ - and then have
> > the break to NONE happen when the
> > ftruncate happens when the open
> > actually takes place).
> 
> Yep, I think the truncate needs to be wrapped into
> contend_level2_oplocks_begin() contend_level2_oplocks_end()

It already is. Check the inside of the vfs_set_filelen()
call.

> And we only need to downgrade to none in delay_for_oplocks()
> for real oplocks.

Yep, we should go to READ for leases, and let
the truncate break to none.