Re: [PATCH] userfaultfd: hugetlbfs: Fix userfaultfd_huge_must_wait pte access

[David Hildenbrand <david@redhat.com>]

On 26.06.2018 15:24, Janosch Frank wrote:
> Use huge_ptep_get to translate huge ptes to normal ptes so we can
> check them with the huge_pte_* functions. Otherwise some architectures
> will check the wrong values and will not wait for userspace to bring
> in the memory.
> 
> Signed-off-by: Janosch Frank <frankja@linux.ibm.com>
> Fixes: 369cd2121be4 ("userfaultfd: hugetlbfs: userfaultfd_huge_must_wait for hugepmd ranges")
> ---
>  fs/userfaultfd.c | 12 +++++++-----
>  1 file changed, 7 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> index 123bf7d516fc..594d192b2331 100644
> --- a/fs/userfaultfd.c
> +++ b/fs/userfaultfd.c
> @@ -222,24 +222,26 @@ static inline bool userfaultfd_huge_must_wait(struct userfaultfd_ctx *ctx,
>  					 unsigned long reason)
>  {
>  	struct mm_struct *mm = ctx->mm;
> -	pte_t *pte;
> +	pte_t *ptep, pte;
>  	bool ret = true;
>  
>  	VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
>  
> -	pte = huge_pte_offset(mm, address, vma_mmu_pagesize(vma));
> -	if (!pte)
> +	ptep = huge_pte_offset(mm, address, vma_mmu_pagesize(vma));
> +
> +	if (!ptep)
>  		goto out;
>  
>  	ret = false;
> +	pte = huge_ptep_get(ptep);
>  
>  	/*
>  	 * Lockless access: we're in a wait_event so it's ok if it
>  	 * changes under us.
>  	 */
> -	if (huge_pte_none(*pte))
> +	if (huge_pte_none(pte))
>  		ret = true;
> -	if (!huge_pte_write(*pte) && (reason & VM_UFFD_WP))
> +	if (!huge_pte_write(pte) && (reason & VM_UFFD_WP))
>  		ret = true;
>  out:
>  	return ret;
> 

Indeed, this only works by luck on x86 and ppc, but not on s390x.

Reviewed-by: David Hildenbrand <david@redhat.com>

-- 

Thanks,

David / dhildenb