Re: [PATCH] filemap: replace pte_offset_map() with pte_offset_map_nolock()

[Andrew Morton]

On Wed, 13 Mar 2024 09:29:13 +0800 Peng Zhang <zhangpeng362@xxxxxxxxxx> wrote:

> From: ZhangPeng <zhangpeng362@xxxxxxxxxx>
> 
> The vmf->ptl in filemap_fault_recheck_pte_none() is still set from
> handle_pte_fault(). But at the same time, we did a pte_unmap(vmf->pte).
> After a pte_unmap(vmf->pte) unmap and rcu_read_unlock(), the page table
> may be racily changed and vmf->ptl maybe fails to protect the actual
> page table.
> Fix this by replacing pte_offset_map() with pte_offset_map_nolock().
> 
> ...
>
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -3207,7 +3207,8 @@ static vm_fault_t filemap_fault_recheck_pte_none(struct vm_fault *vmf)
>  	if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID))
>  		return 0;
>  
> -	ptep = pte_offset_map(vmf->pmd, vmf->address);
> +	ptep = pte_offset_map_nolock(vma->vm_mm, vmf->pmd, vmf->address,
> +				     &vmf->ptl);
>  	if (unlikely(!ptep))
>  		return VM_FAULT_NOPAGE;
>  

whoops, I'm still sitting on this because I didn't know whether we
should backport it.

And...  guess what I say next.  Can we please describe what are the
userspace visible effects of the bug?