On Tue, Feb 06, 2024 at 05:37:22PM -0500, Kent Overstreet wrote:
> On Wed, Feb 07, 2024 at 09:01:05AM +1100, Dave Chinner wrote:
> > On Tue, Feb 06, 2024 at 03:18:51PM -0500, Kent Overstreet wrote:
> > > +static int ioctl_getfsuuid(struct file *file, void __user *argp)
> > > +{
> > > + struct super_block *sb = file_inode(file)->i_sb;
> > > +
> > > + if (!sb->s_uuid_len)
> > > + return -ENOIOCTLCMD;
> > > +
> > > + struct fsuuid2 u = { .len = sb->s_uuid_len, };
> > > + memcpy(&u.uuid[0], &sb->s_uuid, sb->s_uuid_len);
> > > +
> > > + return copy_to_user(argp, &u, sizeof(u)) ? -EFAULT : 0;
> > > +}
> >
> > Can we please keep the declarations separate from the code? I always
> > find this sort of implicit scoping of variables both difficult to
> > read (especially in larger functions) and a landmine waiting to be
> > tripped over. This could easily just be:
> >
> > static int ioctl_getfsuuid(struct file *file, void __user *argp)
> > {
> > struct super_block *sb = file_inode(file)->i_sb;
> > struct fsuuid2 u = { .len = sb->s_uuid_len, };
> >
> > ....
> >
> > and then it's consistent with all the rest of the code...
>
> The way I'm doing it here is actually what I'm transitioning my own code
> to - the big reason being that always declaring variables at the tops of
> functions leads to separating declaration and initialization, and worse
> it leads people to declaring a variable once and reusing it for multiple
> things (I've seen that be a source of real bugs too many times).
>
I still think this is of questionable value. I know I've mentioned
similar concerns to Dave's here on the bcachefs list, but still have not
really seen any discussion other than a bit of back and forth on the
handful of generally accepted (in the kernel) uses of this sort of thing
for limiting scope in loops/branches and such.
I was skimming through some more recent bcachefs patches the other day
(the journal write pipelining stuff) where I came across one or two
medium length functions where this had proliferated, and I found it kind
of annoying TBH. It starts to almost look like there are casts all over
the place and it's a bit more tedious to filter out logic from the
additional/gratuitous syntax, IMO.
That's still just my .02, but there was also previous mention of
starting/having discussion on this sort of style change. Is that still
the plan? If so, before or after proliferating it throughout the
bcachefs code? ;) I am curious if there are other folks in kernel land
who think this makes enough sense that they'd plan to adopt it. Hm?
Brian
> But I won't push that in this patch, we can just keep the style
> consistent for now.
>
> > > +/* Returns the external filesystem UUID, the same one blkid returns */
> > > +#define FS_IOC_GETFSUUID _IOR(0x12, 142, struct fsuuid2)
> > > +
> >
> > Can you add a comment somewhere in the file saying that new VFS
> > ioctls should use the "0x12" namespace in the range 142-255, and
> > mention that BLK ioctls should be kept within the 0x12 {0-141}
> > range?
>
> Well, if we're going to try to keep the BLK_ and FS_IOC_ ioctls in
> separate ranges, then FS_IOC_ needs to move to something else becasue
> otherwise BLK_ won't have a way to expand.
>
> So what else -
>
> ioctl-number.rst has a bunch of other ranges listed for fs.h, but 0x12
> appears to be the only one without conflicts - all the other ranges seem
> to have originated with other filesystems.
>
> So perhaps I will take Darrick's nak (0x15) suggestion after all.
>
