Re: [PATCH] mm/hugetlb: Move vmf_anon_prepare upfront in hugetlb_wp

[David Hildenbrand <david@redhat.com>]

On 21.05.24 09:34, Oscar Salvador wrote:
> hugetlb_wp calls vmf_anon_prepare() after having allocated a page, which
> means that we might need to call restore_reserve_on_error() upon error.
> vmf_anon_prepare() releases the vma lock before returning, but
> restore_reserve_on_error() expects the vma lock to be held by the caller.
> 
> Fix it by calling vmf_anon_prepare() before allocating the page.
> 
> Signed-off-by: Oscar Salvador <osalvador@suse.de>
> Fixes: 9acad7ba3e25 ("hugetlb: use vmf_anon_prepare() instead of anon_vma_prepare()")
> ---
> I did not hit this bug, I just spotted this because I was looking at hugetlb_wp
> for some other reason. And I did not want to get creative to see if I could
> trigger this so I could get a backtrace.
> My assumption is that we could trigger this if 1) this was a shared mapping,
> so no anon_vma and 2) we call in GUP code with FOLL_WRITE, which would cause
> the FLAG_UNSHARE to be passed, so we will end up in hugetlb_wp().

FOLL_WRITE should never result in FLAG_UNSHARE.

> 
>   mm/hugetlb.c | 17 +++++++++--------
>   1 file changed, 9 insertions(+), 8 deletions(-)
> 
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 6be78e7d4f6e..eb0d8a45505e 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -6005,6 +6005,15 @@ static vm_fault_t hugetlb_wp(struct folio *pagecache_folio,
>   	 * be acquired again before returning to the caller, as expected.
>   	 */
>   	spin_unlock(vmf->ptl);
> +
> +	/*
> +	 * When the original hugepage is shared one, it does not have
> +	 * anon_vma prepared.
> +	 */
> +	ret = vmf_anon_prepare(vmf);
> +	if (unlikely(ret))
> +		goto out_release_old;
> +
>   	new_folio = alloc_hugetlb_folio(vma, vmf->address, outside_reserve);
>   
>   	if (IS_ERR(new_folio)) {
> @@ -6058,14 +6067,6 @@ static vm_fault_t hugetlb_wp(struct folio *pagecache_folio,
>   		goto out_release_old;
>   	}
>   
> -	/*
> -	 * When the original hugepage is shared one, it does not have
> -	 * anon_vma prepared.
> -	 */
> -	ret = vmf_anon_prepare(vmf);
> -	if (unlikely(ret))
> -		goto out_release_all;
> -
>   	if (copy_user_large_folio(new_folio, old_folio, vmf->real_address, vma)) {
>   		ret = VM_FAULT_HWPOISON_LARGE | VM_FAULT_SET_HINDEX(hstate_index(h));
>   		goto out_release_all;

The joy of hugetlb reservation code.

LGTM

-- 
Cheers,

David / dhildenb