[PATCH] flock() files even with a permissive share mode.

[Christof Schmitt]

On Sat, Feb 02, 2019 at 06:26:17PM +0100, Volker Lendecke via samba-technical wrote:
> On Fri, Feb 01, 2019 at 08:38:04PM -0500, Matthew Stickney via samba-technical wrote:
> > I've only just begun poking around the samba code, so apologies if I've
> > missed something obvious (or if this is the wrong list for this sort of
> > question).
> 
> This is exactly right. The problem with this API is -- nobody cares.
> You are the first one in MANY years who has taken a look.
> 
> Whenever I have the chance to meet NFS people, I literally beg for
> their precious time to discuss interoperable locking. We need to at
> least talk about the intended semantics: What does a share mode in SMB
> mean for NFS share reservations, and how do we properly communicate
> that? What about mandatory vs advisory? What about leases/oplocks vs
> delegations? What is a good break mechanism? We need to sort this out
> in user space before we can even start approaching the kernel.
> 
> The NFS people I met do not see this as a priority, nobody ever could
> allocate time. Locking seems to be a topic nobody outside the SMB
> world ever looks at. We need to keep the VFS call in, because I know
> of at least one vendor who has this implemented in his proprietary
> file system. For the rest -- I don't think it matters if we keep it
> in.

Do you mean what we have implemented for gpfs? Or is there another file
system that can enforce share modes?

I see the problems with the flock call. The main advantage of keeping it
would be having an easy way to test the vfs call. We could add a test
for this to autobuild to ensure that the codepath is excersized, but if
gpfs is the only actual user, that probably would fall on me.

Christof