Re: [PATCH] mkfs.xfs.8: clarify DAX-vs-reflink restrictions in the mkfs.xfs man page

["Darrick J. Wong" <djwong@kernel.org>]

On Fri, Aug 27, 2021 at 10:39:18AM -0500, Eric Sandeen wrote:
> Now that we have the tristate dax mount options, it is possible
> to enable DAX mode for non-reflinked files on a reflink-capable
> filesystem.  Clarify this in the mkfs.xfs manpage.
> 
> Reported-by: Bill O'Donnell <bodonnel@redhat.com>
> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
> ---
> 
> diff --git a/man/man8/mkfs.xfs.8 b/man/man8/mkfs.xfs.8
> index a7f70285..84ac50e8 100644
> --- a/man/man8/mkfs.xfs.8
> +++ b/man/man8/mkfs.xfs.8
> @@ -316,12 +316,20 @@ option set. When the option
>  is used, the reference count btree feature is not supported and reflink is
>  disabled.
>  .IP
> -Note: the filesystem DAX mount option (
> +Note: the filesystem-wide DAX mount options (
>  .B \-o dax
> -) is incompatible with
> -reflink-enabled XFS filesystems.  To use filesystem DAX with XFS, specify the
> +and
> +.B \-o dax=always
> +) are incompatible with
> +reflink-enabled XFS filesystems.  To use filesystem-wide DAX with XFS, specify the
>  .B \-m reflink=0
>  option to mkfs.xfs to disable the reflink feature.
> +Alternatey, use the

"Alternately..."

> +.B \-o dax=inode
> +mount option to selectively enable DAX mode on non-reflinked files.

/me wonders if this is dangerously close to wading into the muck that is
"Use dax=inode or even no dax option at all, then make sure nobody
shares your file's data blocks to set the DAX fsxattr flag on the file,
and /then/ you can have DAX."

<shrug> But the only blocker I can see is the misspelling above.

--D

> +See
> +.BR xfs (5)
> +for more information.
>  .RE
>  .PP
>  .PD 0
>