Re: [PATCH 1/3] usercopy: Handle vm_map_ram() areas

[Kees Cook <keescook@chromium.org>]

On Mon, Jun 13, 2022 at 07:02:10PM +0200, Uladzislau Rezki wrote:
> On Mon, Jun 13, 2022 at 05:44:35PM +0100, Matthew Wilcox wrote:
> > On Mon, Jun 13, 2022 at 09:23:15AM -0700, Kees Cook wrote:
> > > On Sun, Jun 12, 2022 at 10:32:25PM +0100, Matthew Wilcox (Oracle) wrote:
> > > > vmalloc does not allocate a vm_struct for vm_map_ram() areas.  That causes
> > > > us to deny usercopies from those areas.  This affects XFS which uses
> > > > vm_map_ram() for its directories.
> > > > 
> > > > Fix this by calling find_vmap_area() instead of find_vm_area().
> > > 
> > > Thanks for the fixes!
> > > 
> > > > [...]
> > > > +		/* XXX: We should also abort for free vmap_areas */
> > > 
> > > What's needed to detect this?
> > 
> > I'm not entirely sure.  I only just learned of the existence of this
> > struct ;-)
> > 
> >         /*
> >          * The following two variables can be packed, because
> >          * a vmap_area object can be either:
> >          *    1) in "free" tree (root is free_vmap_area_root)
> >          *    2) or "busy" tree (root is vmap_area_root)
> >          */
> >         union {
> >                 unsigned long subtree_max_size; /* in "free" tree */
> >                 struct vm_struct *vm;           /* in "busy" tree */
> >         };
> > 
> > Hmm.  Actually, we only search vmap_area_root, so I suppose it can't
> > be a free area.  So this XXX can be removed, as we'll get NULL back
> > if we've got a pointer to a free area.  Ulad, do I have this right?
> >
> Yep, we find here only allocated areas which bind to the "vmap_area_root"
> tree, so it can not be a freed area. So we will not get a pointer to the
> free area :)

Thanks!

I've tweaked the patch to drop the XXX comment.

-- 
Kees Cook