Re: [patch 6/7] um: Convert mmapper to unlocked_ioctl

[Arnd Bergmann <arndbergmann@googlemail.com>]

On Thursday 15 October 2009, Alan Cox wrote:
> On Thu, 15 Oct 2009 15:00:34 +0200
> Arnd Bergmann <arndbergmann@googlemail.com> wrote:
>
> > However, returning -ENOIOCTLCMD from an *unlocked_ioctl* function
> > automatically gets turned into -EINVAL. It does this to allow
> > the same functions to be used for unlocked_ioctl and compat_ioctl.
> > In effect, this patch is functionally identical to removing the
> > ioctl function, which I think is what should be done here.
> 
> That is wrong.
> 
> SuS requires an unknown ioctl code returns -ENOTTY. If the code is
> currently remapping it to EINVAL then it wants fixing.

Right, I forgot about the EINVAL/ENOTTY difference. The code currently
returns -ENOIOCTLCMD, which is worse. Thomas' patch makes it return
-EINVAL, which as you said is still wrong. Removing the ioctl function
will do the right thing and return -ENOTTY, so that should be done
here in um/mmapper, with an appropriate changelog.

For the common code in fs/ioctl.c, I think the current behaviour is
correct. It returns -EINVAL if the driver returns -ENOIOCTLCMD, iow
"the request [...] argument is not valid for this device", as specified
by http://www.opengroup.org/onlinepubs/009695399/functions/ioctl.html.

Drivers returning ENOIOCTLCMD for every request are broken and should
be changed to have no ioctl function.
	
	Arnd <><