Re: [PATCH] userfaultfd: change src_folio after ensuring it's unpinned in UFFDIO_MOVE

[Suren Baghdasaryan <surenb@google.com>]

On Thu, Apr 4, 2024 at 1:32 PM Peter Xu <peterx@redhat.com> wrote:
>
> On Thu, Apr 04, 2024 at 06:21:50PM +0100, Matthew Wilcox wrote:
> > On Thu, Apr 04, 2024 at 10:17:26AM -0700, Lokesh Gidra wrote:
> > > -           folio_move_anon_rmap(src_folio, dst_vma);
> > > -           WRITE_ONCE(src_folio->index, linear_page_index(dst_vma, dst_addr));
> > > -
> > >             src_pmdval = pmdp_huge_clear_flush(src_vma, src_addr, src_pmd);
> > >             /* Folio got pinned from under us. Put it back and fail the move. */
> > >             if (folio_maybe_dma_pinned(src_folio)) {
> > > @@ -2270,6 +2267,9 @@ int move_pages_huge_pmd(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd, pm
> > >                     goto unlock_ptls;
> > >             }
> > >
> > > +           folio_move_anon_rmap(src_folio, dst_vma);
> > > +           WRITE_ONCE(src_folio->index, linear_page_index(dst_vma, dst_addr));
> > > +
> >
> > This use of WRITE_ONCE scares me.  We hold the folio locked.  Why do
> > we need to use WRITE_ONCE?  Who's looking at folio->index without
> > holding the folio lock?
>
> Seems true, but maybe suitable for a separate patch to clean it even so?
> We also have the other pte level which has the same WRITE_ONCE(), so if we
> want to drop we may want to drop both.

Yes, I'll do that separately and will remove WRITE_ONCE() in both places.

>
> I just got to start reading some the new move codes (Lokesh, apologies on
> not be able to provide feedbacks previously..), but then I found one thing
> unclear, on special handling of private file mappings only in userfault
> context, and I didn't know why:
>
> lock_vma():
>         if (vma) {
>                 /*
>                  * lock_vma_under_rcu() only checks anon_vma for private
>                  * anonymous mappings. But we need to ensure it is assigned in
>                  * private file-backed vmas as well.
>                  */
>                 if (!(vma->vm_flags & VM_SHARED) && unlikely(!vma->anon_vma))
>                         vma_end_read(vma);
>                 else
>                         return vma;
>         }
>
> AFAIU even for generic users of lock_vma_under_rcu(), anon_vma must be
> stable to be used.  Here it's weird to become an userfault specific
> operation to me.
>
> I was surprised how it worked for private file maps on faults, then I had a
> check and it seems we postponed such check until vmf_anon_prepare(), which
> is the CoW path already, so we do as I expected, but seems unnecessary to
> that point?
>
> Would something like below make it much cleaner for us?  As I just don't
> yet see why userfault is special here.
>
> Thanks,
>
> ===8<===
> diff --git a/mm/memory.c b/mm/memory.c
> index 984b138f85b4..d5cf1d31c671 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -3213,10 +3213,8 @@ vm_fault_t vmf_anon_prepare(struct vm_fault *vmf)
>
>         if (likely(vma->anon_vma))
>                 return 0;
> -       if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
> -               vma_end_read(vma);
> -               return VM_FAULT_RETRY;
> -       }
> +       /* We shouldn't try a per-vma fault at all if anon_vma isn't solid */
> +       WARN_ON_ONCE(vmf->flags & FAULT_FLAG_VMA_LOCK);
>         if (__anon_vma_prepare(vma))
>                 return VM_FAULT_OOM;
>         return 0;
> @@ -5817,9 +5815,9 @@ struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
>          * find_mergeable_anon_vma uses adjacent vmas which are not locked.
>          * This check must happen after vma_start_read(); otherwise, a
>          * concurrent mremap() with MREMAP_DONTUNMAP could dissociate the VMA
> -        * from its anon_vma.
> +        * from its anon_vma.  This applies to both anon or private file maps.
>          */
> -       if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma))
> +       if (unlikely(!(vma->vm_flags & VM_SHARED) && !vma->anon_vma))
>                 goto inval_end_read;
>
>         /* Check since vm_start/vm_end might change before we lock the VMA */
> diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
> index f6267afe65d1..61f21da77dcd 100644
> --- a/mm/userfaultfd.c
> +++ b/mm/userfaultfd.c
> @@ -72,17 +72,8 @@ static struct vm_area_struct *lock_vma(struct mm_struct *mm,
>         struct vm_area_struct *vma;
>
>         vma = lock_vma_under_rcu(mm, address);
> -       if (vma) {
> -               /*
> -                * lock_vma_under_rcu() only checks anon_vma for private
> -                * anonymous mappings. But we need to ensure it is assigned in
> -                * private file-backed vmas as well.
> -                */
> -               if (!(vma->vm_flags & VM_SHARED) && unlikely(!vma->anon_vma))
> -                       vma_end_read(vma);
> -               else
> -                       return vma;
> -       }
> +       if (vma)
> +               return vma;
>
>         mmap_read_lock(mm);
>         vma = find_vma_and_prepare_anon(mm, address);
> --
> 2.44.0
>
>
> --
> Peter Xu
>