[LTP] [PATCH] fcntl.2: F_OFD_XXX needs flock64

[Jeff Layton <jlayton@poochiereds.net>]

On Wed, 2016-08-17 at 15:19 +0200, Cyril Hrubis wrote:
> Hi!
> > 
> > > 
> > > Hmm, I do not think that this is a good idea. The usuall way how to
> > > handle missing constants are fallback definitions such as:
> > > 
> > > #ifndef F_OFD_FOO
> > > # define F_OFD_FOO xyz
> > > #endif
> > > 
> > > This wouldn't do much.
> > > 
> > > Also these should be only disable on 32bit if __USE_FILE_OFFSET64 is
> > > not
> > > defined. But you likely meant that here.
> > > 
> > 
> > That's the usual way, but in this case we wouldn't have a fallback
> > constant. You'd just get an error about F_OFD_* being undefined at
> > build time, which I think is what we'd want here. It's better to fail
> > to compile than to build a binary that passes a bogus struct into the
> > kernel.
> 
> You probably misunderstand what I was trying to say. If you look at the
> > sources out there (for instance at https://codesearch.debian.net/) most
> of it has fallback definitions for F_OFD_* constants included in its own
> header files since these flags are relatively new. Not defining these
> would not accomplish anything.
> 
> One option would be to define them to something invalid such as INT_MAX
> so that it's rejected by kernel on runtime. But I do not think this is
> very good idea either.
> 

Yeah, not much we can do about people that define them on their own. If
you do that, then you're basically saying "I know what I'm doing".

Still, I think it's worthwhile to do this in glibc since we _can_
prevent this problem for folks who aren't doing that.

-- 
Jeff Layton <jlayton@poochiereds.net>