Re: [PATCH 2/2] drivers: core: Remove glue dirs from sysfs earlier

[Benjamin Herrenschmidt]

On Mon, 2018-07-02 at 19:26 -0700, Linus Torvalds wrote:
> On Mon, Jul 2, 2018 at 7:15 PM Linus Torvalds
> <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> > 
> > It's whitespace-damaged on purpose. It's probably broken shit. DO NOT
> > USE UNDER ANY CIRCUMSTANCES.  Think of it more as a "something like
> > this might work, but probably doesn't". Maybe it gives you an idea,
> > although that idea might be "Linus has finally lost it".
> 
> Even if it were to work, it should probably just be done inside kernfs
> and exposed as some kind of "kernfs_remove_if_empty()" function.

That would be harder, we'd have to expose it via a
kobject_del_if_empty() then.

Since we control completely when things are added/removed from the
gluedir and have a big fat mutex for it, I don't think locking is much
of an issue. At least in that specific case.

> We happen to know that we hold the lock that creates all the entries
> in the glue directory (and we don't allow users to move or create
> stuff, afaik, even if we alloc chmod etc), so there should be no
> races, but a generic kernfs helper function would probably have to get
> proper locks and check it the right way.

Yes that or document very clearly under what circumstances that
function can be used.

>                 Linus