Re: [PATCH] securityfs: Add missing d_delete() call on removal

[Kees Cook]

On Wed, May 06, 2020 at 05:02:52AM +0100, Al Viro wrote:
> On Tue, May 05, 2020 at 08:28:33PM -0700, Kees Cook wrote:
> > On Wed, May 06, 2020 at 02:14:31AM +0100, Al Viro wrote:
> > > On Tue, May 05, 2020 at 04:40:35PM -0700, Kees Cook wrote:
> > > > After using simple_unlink(), a call to d_delete() is needed in addition
> > > > to dput().
> > > > 
> > > > Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> > > > ---
> > > > Is this correct? I went looking around and there are a lot of variations
> > > > on the simple_unlink() pattern...
> > > > 
> > > > Many using explicit locking and combinations of d_drop(), __d_drop(), etc.
> > > 
> > > Quite a few of those should switch to simple_recursive_removal().  As for
> > > securityfs...  d_drop() is _probably_ a saner variant, but looking at the
> > > callers of that thing... at least IMA ones seem to be garbage.
> > 
> > Hmm, I dunno. I hadn't looked at these yet. I'm not sure what's needed
> > for those cases.
> > 
> > Is my patch to add d_delete() correct, though? I'm trying to construct
> > the right set of calls for pstore's filesystem, and I noticed that most
> > will do simple_unlink(), d_delete(), dput(), but securityfs seemed to be
> > missing it.
> 
> d_drop().  d_delete() is for the situations when you want the sucker
> to become a hashed negative, if at all possible.

I'm not sure what that means. :) Should stuff like apparmorfs be changed
to d_drop()?

> Re pstore: context, please.

Just posted the whole series:
https://lore.kernel.org/lkml/20200506152114.50375-1-keescook@xxxxxxxxxxxx/

But the specific question was driven by this patch:
https://lore.kernel.org/lkml/20200506152114.50375-11-keescook@xxxxxxxxxxxx/

-- 
Kees Cook