Ocfs2 devel - Jun 2019 - [PATCH 2/5] vfs: create a generic checking function for FS_IOC_FSSETXATTR

On Wed, Jun 26, 2019 at 05:11:33AM +0100, Al Viro wrote:
> On Tue, Jun 25, 2019 at 07:32:18PM -0700, Darrick J. Wong wrote:
> > --- a/fs/btrfs/ioctl.c
> > +++ b/fs/btrfs/ioctl.c
> > @@ -373,10 +373,9 @@ static int check_xflags(unsigned int flags)
> >  static int btrfs_ioctl_fsgetxattr(struct file *file, void __user
*arg)
> >  {
> >  	struct btrfs_inode *binode = BTRFS_I(file_inode(file));
> > -	struct fsxattr fa;
> > -
> > -	memset(&fa, 0, sizeof(fa));
> > -	fa.fsx_xflags = btrfs_inode_flags_to_xflags(binode->flags);
> > +	struct fsxattr fa = {
> > +		.fsx_xflags = btrfs_inode_flags_to_xflags(binode->flags),
> > +	};
> 
> Umm...  Sure, there's no padding, but still - you are going to copy
that thing
> to userland...  How about
> 
> static inline void simple_fill_fsxattr(struct fsxattr *fa, unsigned xflags)
> {
> 	memset(fa, 0, sizeof(*fa));
> 	fa->fsx_xflags = xflags;
> }
> 
> and let the compiler optimize the crap out?
The v2 series used to do that, but Christoph complained that having a
helper for a two-line memset and initialization was silly[1] so now we
have this version.

I don't mind reinstating it as a static inline helper, but I'd like some
input from any of the btrfs developers (or you, Al) about which form is
preferred.

--D

[1] https://lkml.org/lkml/2019/6/25/533

On Wed, Jun 26, 2019 at 08:35:42AM -0700, Darrick J. Wong
wrote:
> > static inline void simple_fill_fsxattr(struct fsxattr *fa, unsigned
xflags)
> > {
> > 	memset(fa, 0, sizeof(*fa));
> > 	fa->fsx_xflags = xflags;
> > }
> > 
> > and let the compiler optimize the crap out?
> 
> The v2 series used to do that, but Christoph complained that having a
> helper for a two-line memset and initialization was silly[1] so now we
> have this version.
> 
> I don't mind reinstating it as a static inline helper, but I'd like
some
> input from any of the btrfs developers (or you, Al) about which form is
> preferred.
I complained having that helper in btrfs.  I think Al wants a generic
one, which at least makes a little more sense.

That being said I wonder if we should lift these attr ioctls to
file op methods and deal with all that crap in VFS code instead of
having all those duplicated ioctl parsers.