Re: [PATCH 11/18] staging: gasket: always allow root open for write

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Sun, Jul 15, 2018 at 12:11:47PM +0300, Dmitry Torokhov wrote:
> On Sun, Jul 15, 2018 at 12:05 PM Greg Kroah-Hartman
> <gregkh@linuxfoundation.org> wrote:
> >
> > On Fri, Jul 13, 2018 at 10:58:09PM -0700, Todd Poynor wrote:
> > > From: Todd Poynor <toddpoynor@google.com>
> > >
> > > Always allow root to open device for writing.
> > >
> > > Drop special-casing of ioctl permissions for root vs. owner.
> > >
> > > Reported-by: Dmitry Torokhov <dtor@chromium.org>
> > > Signed-off-by: Zhongze Hu <frankhu@chromium.org>
> > > Signed-off-by: Todd Poynor <toddpoynor@google.com>
> > > ---
> > >  drivers/staging/gasket/apex_driver.c  |  9 +++------
> > >  drivers/staging/gasket/gasket_core.c  |  8 +++++---
> > >  drivers/staging/gasket/gasket_ioctl.c | 15 ++++++---------
> > >  3 files changed, 14 insertions(+), 18 deletions(-)
> > >
> > > diff --git a/drivers/staging/gasket/apex_driver.c b/drivers/staging/gasket/apex_driver.c
> > > index b1318482ba65..ffe11d8168ea 100644
> > > --- a/drivers/staging/gasket/apex_driver.c
> > > +++ b/drivers/staging/gasket/apex_driver.c
> > > @@ -644,13 +644,10 @@ static bool is_gcb_in_reset(struct gasket_dev *gasket_dev)
> > >  static uint apex_ioctl_check_permissions(struct file *filp, uint cmd)
> > >  {
> > >       struct gasket_dev *gasket_dev = filp->private_data;
> > > -     int root = capable(CAP_SYS_ADMIN);
> > > -     int is_owner = gasket_dev->dev_info.ownership.is_owned &&
> > > -                    current->tgid == gasket_dev->dev_info.ownership.owner;
> > > +     fmode_t write;
> > >
> > > -     if (root || is_owner)
> > > -             return 1;
> > > -     return 0;
> > > +     write = filp->f_mode & FMODE_WRITE;
> >
> > Ok, this is insane.  You don't change, or check, the permissions on a
> > file handle while it is already open, as you only check the permissions
> > on OPEN, not on WRITE.  See the recent rant from Linus on the linux-api
> > list for yet-another-long-threaad in which he explains this.
> >
> > So this whole ioctl can just be removed, it is totally crazy and wrong
> > and should just be removed.
> 
> No, the code checks whether the requested ioctl command is compatible
> with the mode the file handle was open with. There are some ioctls
> that are allowed on file handle opened for read and others that
> require file handle to be opened for write. That is all. It does not
> change permissions past open.

That's really not obvious here :)

And is odd on a whole other set of "crazy design", but ok, let's let it
live for now.

I can't wait for people to just realize this whole "new" subsystem can
be replaced with UIO, but that's a topic for a different thread...

thanks,

greg k-h