[PATCH v6 2/5] userfaultfd: add /dev/userfaultfd for fine grained access control

[Axel Rasmussen]

On Wed, Aug 17, 2022 at 11:26 PM Greg KH <gregkh at linuxfoundation.org> wrote:
>
> On Wed, Aug 17, 2022 at 02:47:25PM -0700, Axel Rasmussen wrote:
> > +static int userfaultfd_dev_open(struct inode *inode, struct file *file)
> > +{
> > +     return 0;
>
> If your open does nothing, no need to list it here at all, right?
>
> > +}
> > +
> > +static long userfaultfd_dev_ioctl(struct file *file, unsigned int cmd, unsigned long flags)
> > +{
> > +     if (cmd != USERFAULTFD_IOC_NEW)
> > +             return -EINVAL;
> > +
> > +     return new_userfaultfd(flags);
> > +}
> > +
> > +static const struct file_operations userfaultfd_dev_fops = {
> > +     .open = userfaultfd_dev_open,
> > +     .unlocked_ioctl = userfaultfd_dev_ioctl,
> > +     .compat_ioctl = userfaultfd_dev_ioctl,
>
> Why do you need to set compat_ioctl?  Shouldn't it just default to the
> existing one?

I took some more time looking at this today, and I think it actually
has to be the way it is.

I didn't find anywhere we noticed compat_ioctl unset, and default to
the "normal" one (e.g. see the compat ioctl syscall definition in
fs/ioctl.c). It looks to me like it really does need some value. It's
common to use compat_ptr_ioctl for this, but since we're interpreting
the arg as a scalar not as a pointer, doing that here would be
incorrect.

It looks like there are other existing examples that do it the same
way, e.g. seccomp_notify_ops in linux/seccomp.c.

>
> And why is this a device node at all?  Shouldn't the syscall handle all
> of this (to be honest, I didn't read anything but the misc code, sorry.)
>
> thanks,
>
> greg k-h