"John W. Linville" wrote :
> On Mon, Jul 09, 2007 at 07:54:39PM +0900, Masakazu Mokuno wrote:
> > This patch removes some duplicated wireless ioctl entries in the array
> > 'struct ioctl_trans ioctl_start[]' of fs/compat_ioctl.c
> >
> > These entries are registered twice like:
> >
> > COMPATIBLE_IOCTL(SIOCGIWPRIV)
> >
> > and
> >
> > HANDLE_IOCTL(SIOCGIWPRIV, do_wireless_ioctl)
> >
> >
> > Signed-off-by: Masakazu Mokuno <[email protected]>
> > ---
> > fs/compat_ioctl.c | 3 ---
> > 1 file changed, 3 deletions(-)
> >
> > --- a/fs/compat_ioctl.c
> > +++ b/fs/compat_ioctl.c
> > @@ -3156,12 +3156,9 @@ COMPATIBLE_IOCTL(SIOCSIWSENS)
> > COMPATIBLE_IOCTL(SIOCGIWSENS)
> > COMPATIBLE_IOCTL(SIOCSIWRANGE)
> > COMPATIBLE_IOCTL(SIOCSIWPRIV)
> > -COMPATIBLE_IOCTL(SIOCGIWPRIV)
> > COMPATIBLE_IOCTL(SIOCSIWSTATS)
> > -COMPATIBLE_IOCTL(SIOCGIWSTATS)
> > COMPATIBLE_IOCTL(SIOCSIWAP)
> > COMPATIBLE_IOCTL(SIOCGIWAP)
> > -COMPATIBLE_IOCTL(SIOCSIWSCAN)
> > COMPATIBLE_IOCTL(SIOCSIWRATE)
> > COMPATIBLE_IOCTL(SIOCGIWRATE)
> > COMPATIBLE_IOCTL(SIOCSIWRTS)
>
> As I read the code in compat_ioctl.c, it looks to me like the
> COMPATIBLE_IOCTL definitions are the ones that are actually being
> used today. Do you agree?
Actually, you are wrong, and Masakazu is right. All those
ioctls contains a pointer and should go through the pointer
conversion.
The reason why Masakazu sent that patch is that he actually
stumbled on the problem and tested it.
> Given the...stability...of the wireless extensions API, if we are going
> to remove one or the other of these not-quite-duplicate definitions,
> shouldn't we remove the HANDLE_IOCTL defintions instead?
I don't understand, you are in favor of breaking the API ?
Regards,
Jean
On Fri, Jul 27, 2007 at 10:13:01AM -0700, Jean Tourrilhes wrote:
> "John W. Linville" wrote :
> > On Mon, Jul 09, 2007 at 07:54:39PM +0900, Masakazu Mokuno wrote:
> > > This patch removes some duplicated wireless ioctl entries in the array
> > > 'struct ioctl_trans ioctl_start[]' of fs/compat_ioctl.c
> > >
> > > These entries are registered twice like:
> > >
> > > COMPATIBLE_IOCTL(SIOCGIWPRIV)
> > >
> > > and
> > >
> > > HANDLE_IOCTL(SIOCGIWPRIV, do_wireless_ioctl)
> > >
> > >
> > > Signed-off-by: Masakazu Mokuno <[email protected]>
> > As I read the code in compat_ioctl.c, it looks to me like the
> > COMPATIBLE_IOCTL definitions are the ones that are actually being
> > used today. Do you agree?
>
> Actually, you are wrong, and Masakazu is right. All those
> ioctls contains a pointer and should go through the pointer
> conversion.
Masakazu replied in agreement that the COMPATIBLE_IOCTL entries are
the effective ones i.e. the code currently uses those entries and
the others are currently just wasting space.
Perhaps the HANDLE_IOCTL entries are indeed the correct and intended ones. You seem to be indicating so.
> The reason why Masakazu sent that patch is that he actually
> stumbled on the problem and tested it.
The only problem stated is the not-quite-duplicate entries. If there
is an actual problem (and not just sloppy code) then that makes the
situation more clear.
What is the manifestation of the problem?
> > Given the...stability...of the wireless extensions API, if we are going
> > to remove one or the other of these not-quite-duplicate definitions,
> > shouldn't we remove the HANDLE_IOCTL defintions instead?
>
> I don't understand, you are in favor of breaking the API ?
I'm not sure how an honest reading could imply that.
If this fixes a bug, then fine. If we are trading the one "duplicate"
entry we have been using for one that hasn't been in use, it doesn't
make much sense.
John
--
John W. Linville
[email protected]