Re: [RFC PATCH] mm/mmap: Fix uprobe anon page be overwritten when expanding vma during mremap

[Lorenzo Stoakes <lorenzo.stoakes@oracle.com>]

On Tue, May 27, 2025 at 12:42:38PM +0100, Lorenzo Stoakes wrote:
> On Mon, May 26, 2025 at 08:46:07PM +0200, David Hildenbrand wrote:
> > On 26.05.25 17:48, Oleg Nesterov wrote:
> > > Hi Lehui,
> > >
> > > As I said, I don't understand mm/, so can't comment, but...
> > >
> > > On 05/26, Pu Lehui wrote:
> > > >
> > > > To make things simpler, perhaps we could try post-processing, that is:
> > > >
> > > > diff --git a/mm/mremap.c b/mm/mremap.c
> > > > index 83e359754961..46a757fd26dc 100644
> > > > --- a/mm/mremap.c
> > > > +++ b/mm/mremap.c
> > > > @@ -240,6 +240,11 @@ static int move_ptes(struct pagetable_move_control
> > > > *pmc,
> > > >                  if (pte_none(ptep_get(old_pte)))
> > > >                          continue;
> > > >
> > > > +               /* skip move pte when expanded range has uprobe */
> > > > +               if (unlikely(pte_present(*new_pte) &&
> > > > +                            vma_has_uprobes(pmc->new, new_addr, new_addr +
> > > > PAGE_SIZE)))
>
> This feels like a horrible hack, note that we also move page tables at higher
> page table levels _anyway_ so this would be broken by that (unless uprobes split
> huge mappings).
>
> If it's uprobe code that's putting the new PTE in place, then this is
> just... yeah. I'm with David's suggestion of just disallowing this scenario, I
> really dislike the idea that we're ok with an invalid condition being ok, only
> to cover off this one specific case.

CORRECTION: Oleg's suggestion :P Sorry Oleg, misread...

>
>
> > > > +                       continue;
> > > > +
> > >
> > > I was thinking about
> > >
> > > 	WARN_ON(!pte_none(*new_pte))
> > >
> > > at the start of the main loop.
> > >
> > > Obviously not to fix the problem, but rather to make it more explicit.
> >
> > Yeah, WARN_ON_ONCE().
> >
> > We really should fix the code to not install uprobes into the area we are
> > moving.
> >
> > Likely, the correct fix will be to pass the range as well to uprobe_mmap(),
> > and passing that range to build_probe_list().
> >
> > Only when growing using mremap(), we want to call it on the extended range
> > only.
>
> We might be able to implement a simpler version of the proposed patch though
> which might avoid us needing to do something like this.
>
> Having a look...
>
> >
> > --
> > Cheers,
> >
> > David / dhildenb
> >
>