2024-01-12 01:39:49

by Suren Baghdasaryan

Subject: [PATCH 1/1] userfaultfd: avoid huge_zero_page in UFFDIO_MOVE

While testing UFFDIO_MOVE ioctl, syzbot triggered VM_BUG_ON_PAGE caused
by a call to PageAnonExclusive() with a huge_zero_page as a parameter.
UFFDIO_MOVE does not yet handle zeropages and returns EBUSY when one is
encountered. Add an early huge_zero_page check in the PMD move path
to avoid this situation.

Reported-by: [email protected]
Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
Signed-off-by: Suren Baghdasaryan <[email protected]>
---
Applies cleanly over linux-next, mm-stable and mm-unstable branches

mm/userfaultfd.c | 6 ++++++
1 file changed, 6 insertions(+)

diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
index 216ab4c8621f..20e3b0d9cf7e 100644
--- a/mm/userfaultfd.c
+++ b/mm/userfaultfd.c
@@ -1393,6 +1393,12 @@ ssize_t move_pages(struct userfaultfd_ctx *ctx, struct mm_struct *mm,
err = -ENOENT;
break;
}
+ /* Avoid moving zeropages for now */
+ if (is_huge_zero_pmd(*src_pmd)) {
+ spin_unlock(ptl);
+ err = -EBUSY;
+ break;
+ }

/* Check if we can move the pmd without splitting it. */
if (move_splits_huge_pmd(dst_addr, src_addr, src_start + len) ||
--
2.43.0.275.g3460e3d667-goog

2024-01-12 08:57:26

by David Hildenbrand

Subject: Re: [PATCH 1/1] userfaultfd: avoid huge_zero_page in UFFDIO_MOVE

On 12.01.24 02:39, Suren Baghdasaryan wrote:
> While testing UFFDIO_MOVE ioctl, syzbot triggered VM_BUG_ON_PAGE caused
> by a call to PageAnonExclusive() with a huge_zero_page as a parameter.
> UFFDIO_MOVE does not yet handle zeropages and returns EBUSY when one is
> encountered. Add an early huge_zero_page check in the PMD move path
> to avoid this situation.
>
> Reported-by: [email protected]
> Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
> Signed-off-by: Suren Baghdasaryan <[email protected]>
> ---
> Applies cleanly over linux-next, mm-stable and mm-unstable branches
>
> mm/userfaultfd.c | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
> index 216ab4c8621f..20e3b0d9cf7e 100644
> --- a/mm/userfaultfd.c
> +++ b/mm/userfaultfd.c
> @@ -1393,6 +1393,12 @@ ssize_t move_pages(struct userfaultfd_ctx *ctx, struct mm_struct *mm,
> err = -ENOENT;
> break;
> }
> + /* Avoid moving zeropages for now */
> + if (is_huge_zero_pmd(*src_pmd)) {
> + spin_unlock(ptl);
> + err = -EBUSY;
> + break;
> + }

As an alternative (and more future proof?) just reject anything that is
not an anon folio as well?

Anyhow, this should handle the issue at hand.

Acked-by: David Hildenbrand <[email protected]>

--
Cheers,

David / dhildenb

2024-01-12 17:31:55

by Suren Baghdasaryan

Subject: Re: [PATCH 1/1] userfaultfd: avoid huge_zero_page in UFFDIO_MOVE

On Fri, Jan 12, 2024 at 12:57 AM David Hildenbrand <[email protected]> wrote:
>
> On 12.01.24 02:39, Suren Baghdasaryan wrote:
> > While testing UFFDIO_MOVE ioctl, syzbot triggered VM_BUG_ON_PAGE caused
> > by a call to PageAnonExclusive() with a huge_zero_page as a parameter.
> > UFFDIO_MOVE does not yet handle zeropages and returns EBUSY when one is
> > encountered. Add an early huge_zero_page check in the PMD move path
> > to avoid this situation.
> >
> > Reported-by: [email protected]
> > Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
> > Signed-off-by: Suren Baghdasaryan <[email protected]>
> > ---
> > Applies cleanly over linux-next, mm-stable and mm-unstable branches
> >
> > mm/userfaultfd.c | 6 ++++++
> > 1 file changed, 6 insertions(+)
> >
> > diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
> > index 216ab4c8621f..20e3b0d9cf7e 100644
> > --- a/mm/userfaultfd.c
> > +++ b/mm/userfaultfd.c
> > @@ -1393,6 +1393,12 @@ ssize_t move_pages(struct userfaultfd_ctx *ctx, struct mm_struct *mm,
> > err = -ENOENT;
> > break;
> > }
> > + /* Avoid moving zeropages for now */
> > + if (is_huge_zero_pmd(*src_pmd)) {
> > + spin_unlock(ptl);
> > + err = -EBUSY;
> > + break;
> > + }
>
> As an alternative (and more future proof?) just reject anything that is
> not an anon folio as well?

Yes, but I'm going to remove this check shortly, once zeropage
movement patch is ready.

>
> Anyhow, this should handle the issue at hand.
>
> Acked-by: David Hildenbrand <[email protected]>

Thanks!

>
> --
> Cheers,
>
> David / dhildenb
>