Re: [PATCH v2] fs/stat: set attributes_mask for STATX_ATTR_DAX

["Darrick J. Wong" <darrick.wong@oracle.com>]

On Wed, Nov 25, 2020 at 10:08:40AM +0800, XiaoLi Feng wrote:
> From: Xiaoli Feng <fengxiaoli0714@gmail.com>
> 
> keep attributes and attributes_mask are consistent for
> STATX_ATTR_DAX.
> ---
> Hi,
> Please help to review this patch. I send this patch because xfstests generic/532
> is failed for dax test.
> 
>  fs/stat.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/stat.c b/fs/stat.c
> index dacecdda2e79..4619b3fc9694 100644
> --- a/fs/stat.c
> +++ b/fs/stat.c
> @@ -80,8 +80,10 @@ int vfs_getattr_nosec(const struct path *path, struct kstat *stat,
>  	if (IS_AUTOMOUNT(inode))
>  		stat->attributes |= STATX_ATTR_AUTOMOUNT;
>  
> -	if (IS_DAX(inode))
> +	if (IS_DAX(inode)) {
>  		stat->attributes |= STATX_ATTR_DAX;
> +		stat->attributes_mask |= STATX_ATTR_DAX;

From the discussion of the V1 patch:

Doesn't it make more sense for /filesystems/ driver to set the attr_mask
bit when the filesystem is capable of DAX?  Surely we should be able to
tell applications that DAX is a possibility for the fs even if it's not
enabled on this specific file.

IOWs the place to make this change is in the ext2/ext4/fuse/xfs code,
not in the generic vfs.

--D

> +	}
>  
>  	if (inode->i_op->getattr)
>  		return inode->i_op->getattr(path, stat, request_mask,
> -- 
> 2.18.1
>