Re: [PATCH] mm: fix apply_to_existing_page_range()

[David Hildenbrand]

On 09.04.25 12:23, Kirill A. Shutemov wrote:
> On Wed, Apr 09, 2025 at 11:52:43AM +0200, David Hildenbrand wrote:
> > On 09.04.25 11:40, Kirill A. Shutemov wrote:
> > > In the case of apply_to_existing_page_range(), apply_to_pte_range() is
> > > reached with 'create' set to false. When !create, the loop over the PTE
> > > page table is broken.
> > >
> > > apply_to_pte_range() will only move to the next PTE entry if 'create' is
> > > true or if the current entry is not pte_none().
> > >
> > > This means that the user of apply_to_existing_page_range() will not have
> > > 'fn' called for any entries after the first pte_none() in the PTE page
> > > table.
> > >
> > > Fix the loop logic in apply_to_pte_range().
> > >
> > > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> > > Fixes: be1db4753ee6 ("mm/memory.c: add apply_to_existing_page_range() helper")
> > > Cc: Daniel Axtens <dja@xxxxxxxxxx>
> > > ---
> > >    mm/memory.c | 4 ++--
> > >    1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/mm/memory.c b/mm/memory.c
> > > index fb7b8dc75167..2094564f4dfb 100644
> > > --- a/mm/memory.c
> > > +++ b/mm/memory.c
> > > @@ -2907,11 +2907,11 @@ static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
> > >    	if (fn) {
> > >    		do {
> > >    			if (create || !pte_none(ptep_get(pte))) {
> > > -				err = fn(pte++, addr, data);
> > > +				err = fn(pte, addr, data);
> > >    				if (err)
> > >    					break;
> > >    			}
> > > -		} while (addr += PAGE_SIZE, addr != end);
> > > +		} while (pte++, addr += PAGE_SIZE, addr != end);
> > >    	}
> > >    	*mask |= PGTBL_PTE_MODIFIED;
> >
> > LGTM. just curious, did you run into any actual issues that are worth
> > describing?
>
> I stepped on it in my non-upstream code debugging. I am not sure how it
> affects existing users.
>
> > It should affect apply_to_existing_page_range() users where create==false.
> > There are not many, and likely most PTEs in the range they are passing are
> > all non-none.
>
> Or we just silently leak memory :P

That's exactly what I am trying to figure out: is there something 
upstream that could actually run into this such that we should CC stable?

--
Cheers,

David / dhildenb