Re: [PATCH] filemap: avoid truncating 64-bit offset to 32 bits

[Andrew Morton]

On Thu,  2 Jan 2025 11:04:11 -0800 Marco Nelissen <marco.nelissen@xxxxxxxxx> wrote:

> on 32-bit kernels, folio_seek_hole_data() was inadvertently truncating a
> 64-bit value to 32 bits, leading to a possible infinite loop when writing
> to an xfs filesystem.
> 
> ...
>
> +++ b/mm/filemap.c
> @@ -3005,7 +3005,7 @@ static inline loff_t folio_seek_hole_data(struct xa_state *xas,
>  		if (ops->is_partially_uptodate(folio, offset, bsz) ==
>  							seek_data)
>  			break;
> -		start = (start + bsz) & ~(bsz - 1);
> +		start = (start + bsz) & ~((u64)bsz - 1);
>  		offset += bsz;
>  	} while (offset < folio_size(folio));
>  unlock:

Thanks.  I'll add

Fixes: 54fa39ac2e00b ("iomap: use mapping_seek_hole_data")
Cc: <stable@xxxxxxxxxxxxxxx>

The

	offset = offset_in_folio(folio, start) & ~(bsz - 1);

a few lines earlier is worrisome.  I wonder if we should simply make
`bsz' and `offset' have type u64 and sleep well at night.