mm: userfaultfd: fix unexpected change to src_folio when UFFDIO_MOVE fails

Message ID 20240222080815.46291-1-zhengqi.arch@bytedance.com
State New
Headers
Series mm: userfaultfd: fix unexpected change to src_folio when UFFDIO_MOVE fails |

Commit Message

Qi Zheng Feb. 22, 2024, 8:08 a.m. UTC
  After ptep_clear_flush(), if we find that src_folio is pinned we will fail
UFFDIO_MOVE and put src_folio back to src_pte entry, but the change to
src_folio->{mapping,index} is not restored in this process. This is not
what we expected, so fix it.

Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
---
 mm/userfaultfd.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
  

Comments

David Hildenbrand Feb. 22, 2024, 8:43 a.m. UTC | #1
On 22.02.24 09:08, Qi Zheng wrote:
> After ptep_clear_flush(), if we find that src_folio is pinned we will fail
> UFFDIO_MOVE and put src_folio back to src_pte entry, but the change to
> src_folio->{mapping,index} is not restored in this process. This is not
> what we expected, so fix it.
> 
> Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
> ---
>   mm/userfaultfd.c | 6 +++---
>   1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
> index 4744d6a96f96..503ea77c81aa 100644
> --- a/mm/userfaultfd.c
> +++ b/mm/userfaultfd.c
> @@ -1008,9 +1008,6 @@ static int move_present_pte(struct mm_struct *mm,
>   		goto out;
>   	}
>   
> -	folio_move_anon_rmap(src_folio, dst_vma);
> -	WRITE_ONCE(src_folio->index, linear_page_index(dst_vma, dst_addr));
> -
>   	orig_src_pte = ptep_clear_flush(src_vma, src_addr, src_pte);
>   	/* Folio got pinned from under us. Put it back and fail the move. */
>   	if (folio_maybe_dma_pinned(src_folio)) {
> @@ -1019,6 +1016,9 @@ static int move_present_pte(struct mm_struct *mm,
>   		goto out;
>   	}
>   
> +	folio_move_anon_rmap(src_folio, dst_vma);
> +	WRITE_ONCE(src_folio->index, linear_page_index(dst_vma, dst_addr));
> +
>   	orig_dst_pte = mk_pte(&src_folio->page, dst_vma->vm_page_prot);
>   	/* Follow mremap() behavior and treat the entry dirty after the move */
>   	orig_dst_pte = pte_mkwrite(pte_mkdirty(orig_dst_pte), dst_vma);

Indeed, LGTM.

Reviewed-by: David Hildenbrand <david@redhat.com>
  
Andrew Morton Feb. 22, 2024, 9 p.m. UTC | #2
On Thu, 22 Feb 2024 16:08:15 +0800 Qi Zheng <zhengqi.arch@bytedance.com> wrote:

> After ptep_clear_flush(), if we find that src_folio is pinned we will fail
> UFFDIO_MOVE and put src_folio back to src_pte entry, but the change to
> src_folio->{mapping,index} is not restored in this process. This is not
> what we expected, so fix it.
> 
> Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")

What are the expected worst-case userspace-visible runtime effects of
this flaw?
  
Suren Baghdasaryan Feb. 22, 2024, 9:41 p.m. UTC | #3
On Thu, Feb 22, 2024 at 12:43 AM David Hildenbrand <david@redhat.com> wrote:
>
> On 22.02.24 09:08, Qi Zheng wrote:
> > After ptep_clear_flush(), if we find that src_folio is pinned we will fail
> > UFFDIO_MOVE and put src_folio back to src_pte entry, but the change to
> > src_folio->{mapping,index} is not restored in this process. This is not
> > what we expected, so fix it.
> >
> > Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
> > Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
> > ---
> >   mm/userfaultfd.c | 6 +++---
> >   1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
> > index 4744d6a96f96..503ea77c81aa 100644
> > --- a/mm/userfaultfd.c
> > +++ b/mm/userfaultfd.c
> > @@ -1008,9 +1008,6 @@ static int move_present_pte(struct mm_struct *mm,
> >               goto out;
> >       }
> >
> > -     folio_move_anon_rmap(src_folio, dst_vma);
> > -     WRITE_ONCE(src_folio->index, linear_page_index(dst_vma, dst_addr));
> > -
> >       orig_src_pte = ptep_clear_flush(src_vma, src_addr, src_pte);
> >       /* Folio got pinned from under us. Put it back and fail the move. */
> >       if (folio_maybe_dma_pinned(src_folio)) {
> > @@ -1019,6 +1016,9 @@ static int move_present_pte(struct mm_struct *mm,
> >               goto out;
> >       }
> >
> > +     folio_move_anon_rmap(src_folio, dst_vma);
> > +     WRITE_ONCE(src_folio->index, linear_page_index(dst_vma, dst_addr));
> > +
> >       orig_dst_pte = mk_pte(&src_folio->page, dst_vma->vm_page_prot);
> >       /* Follow mremap() behavior and treat the entry dirty after the move */
> >       orig_dst_pte = pte_mkwrite(pte_mkdirty(orig_dst_pte), dst_vma);
>
> Indeed, LGTM.
>
> Reviewed-by: David Hildenbrand <david@redhat.com>

Thanks for catching this! Makes total sense to check before modification.

Reviewed-by: Suren Baghdasaryan <surenb@google.com>

>
> --
> Cheers,
>
> David / dhildenb
>
  
Suren Baghdasaryan Feb. 22, 2024, 9:56 p.m. UTC | #4
On Thu, Feb 22, 2024 at 1:00 PM Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Thu, 22 Feb 2024 16:08:15 +0800 Qi Zheng <zhengqi.arch@bytedance.com> wrote:
>
> > After ptep_clear_flush(), if we find that src_folio is pinned we will fail
> > UFFDIO_MOVE and put src_folio back to src_pte entry, but the change to
> > src_folio->{mapping,index} is not restored in this process. This is not
> > what we expected, so fix it.
> >
> > Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
>
> What are the expected worst-case userspace-visible runtime effects of
> this flaw?

It can cause rmap for that page to be invalid. I guess memory
corruption might be the visible effect?
  
David Hildenbrand Feb. 28, 2024, 10:46 a.m. UTC | #5
On 22.02.24 22:56, Suren Baghdasaryan wrote:
> On Thu, Feb 22, 2024 at 1:00 PM Andrew Morton <akpm@linux-foundation.org> wrote:
>>
>> On Thu, 22 Feb 2024 16:08:15 +0800 Qi Zheng <zhengqi.arch@bytedance.com> wrote:
>>
>>> After ptep_clear_flush(), if we find that src_folio is pinned we will fail
>>> UFFDIO_MOVE and put src_folio back to src_pte entry, but the change to
>>> src_folio->{mapping,index} is not restored in this process. This is not
>>> what we expected, so fix it.
>>>
>>> Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
>>
>> What are the expected worst-case userspace-visible runtime effects of
>> this flaw?
> 
> It can cause rmap for that page to be invalid. I guess memory
> corruption might be the visible effect?

At least swapout+migration would no longer work, because we might fail 
to locate the mappings of that folio.

Memory corruption, not sure.
  

Patch

diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
index 4744d6a96f96..503ea77c81aa 100644
--- a/mm/userfaultfd.c
+++ b/mm/userfaultfd.c
@@ -1008,9 +1008,6 @@  static int move_present_pte(struct mm_struct *mm,
 		goto out;
 	}
 
-	folio_move_anon_rmap(src_folio, dst_vma);
-	WRITE_ONCE(src_folio->index, linear_page_index(dst_vma, dst_addr));
-
 	orig_src_pte = ptep_clear_flush(src_vma, src_addr, src_pte);
 	/* Folio got pinned from under us. Put it back and fail the move. */
 	if (folio_maybe_dma_pinned(src_folio)) {
@@ -1019,6 +1016,9 @@  static int move_present_pte(struct mm_struct *mm,
 		goto out;
 	}
 
+	folio_move_anon_rmap(src_folio, dst_vma);
+	WRITE_ONCE(src_folio->index, linear_page_index(dst_vma, dst_addr));
+
 	orig_dst_pte = mk_pte(&src_folio->page, dst_vma->vm_page_prot);
 	/* Follow mremap() behavior and treat the entry dirty after the move */
 	orig_dst_pte = pte_mkwrite(pte_mkdirty(orig_dst_pte), dst_vma);