Re: [PATCH v2] mm/userfaultfd: fix hugetlb fault mutex hash calculation

["David Hildenbrand (Arm)" <david@kernel.org>]

On 3/7/26 15:35, Jianhui Zhou wrote:
> In mfill_atomic_hugetlb(), linear_page_index() is used to calculate the
> page index for hugetlb_fault_mutex_hash(). However, linear_page_index()
> returns the index in PAGE_SIZE units, while hugetlb_fault_mutex_hash()
> expects the index in huge page units (as calculated by
> vma_hugecache_offset()). This mismatch means that different addresses
> within the same huge page can produce different hash values, leading to
> the use of different mutexes for the same huge page. This can cause
> races between faulting threads, which can corrupt the reservation map
> and trigger the BUG_ON in resv_map_release().
> 
> Fix this by replacing linear_page_index() with vma_hugecache_offset()
> and applying huge_page_mask() to align the address properly. To make
> vma_hugecache_offset() available outside of mm/hugetlb.c, move it to
> include/linux/hugetlb.h as a static inline function.
> 
> Fixes: 60d4d2d2b40e ("userfaultfd: hugetlbfs: add __mcopy_atomic_hugetlb for huge page UFFDIO_COPY")
> Reported-by: syzbot+f525fd79634858f478e7@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=f525fd79634858f478e7
> Cc: stable@vger.kernel.org
> Signed-off-by: Jianhui Zhou <jianhuizzzzz@gmail.com>
> ---
> v2:
> - Remove unnecessary !CONFIG_HUGETLB_PAGE stub for vma_hugecache_offset()
>   (Peter Xu, SeongJae Park)
> 
>  include/linux/hugetlb.h | 11 +++++++++++
>  mm/hugetlb.c            | 11 -----------
>  mm/userfaultfd.c        |  5 ++++-
>  3 files changed, 15 insertions(+), 12 deletions(-)
> 
> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
> index 65910437be1c..f003afe0cc91 100644
> --- a/include/linux/hugetlb.h
> +++ b/include/linux/hugetlb.h
> @@ -796,6 +796,17 @@ static inline unsigned huge_page_shift(struct hstate *h)
>  	return h->order + PAGE_SHIFT;
>  }
>  
> +/*
> + * Convert the address within this vma to the page offset within
> + * the mapping, huge page units here.
> + */
> +static inline pgoff_t vma_hugecache_offset(struct hstate *h,
> +		struct vm_area_struct *vma, unsigned long address)
> +{
> +	return ((address - vma->vm_start) >> huge_page_shift(h)) +
> +		(vma->vm_pgoff >> huge_page_order(h));
> +}

It's hard to put my disgust about the terminology "hugecache" into
words. Not your fault, but we should do better :)

If you're starting to use that from other MM code then hugetlb.c, please
find a better name.

Further, I wonder whether we can avoid passing in "struct hstate *h" and
simply call hstate_vma() internally.

Something like the following to mimic linear_page_index() ?

/**
 * hugetlb_linear_page_index - linear_page_index() but in hugetlb page
 *			       size granularity
 * @vma: ...
 * @address: ...
 *
 * Returns: ...
 */
static inline void hugetlb_linear_page_index(struct vm_area_struct *vma,
		unsigned long address)
{
	struct hstate *h = hstate_vma(vma);

	...
}

-- 
Cheers,

David