[PATCH] event: Cheking for NULL before dereferencing the pointer.

[Pekka Paalanen]

On Fri, 09 May 2014 15:21:51 +0530
Srivardhan <sri.hebbar at samsung.com> wrote:

> 
> 
> > -----Original Message-----
> > From: Pekka Paalanen [mailto:ppaalanen at gmail.com]
> > Sent: Friday, May 09, 2014 3:09 PM
> > To: Srivardhan
> > Cc: 'Hardening'; wayland-devel at lists.freedesktop.org
> > Subject: Re: [PATCH] event: Cheking for NULL before dereferencing the
> > pointer.
> > 
> > On Fri, 09 May 2014 14:56:14 +0530
> > Srivardhan <sri.hebbar at samsung.com> wrote:
> > 
> > >
> > >
> > > > -----Original Message-----
> > > > From: wayland-devel [mailto:wayland-devel-
> > > > bounces at lists.freedesktop.org] On Behalf Of Hardening
> > > > Sent: Friday, May 09, 2014 1:08 PM
> > > > To: wayland-devel at lists.freedesktop.org
> > > > Subject: Re: [PATCH] event: Cheking for NULL before dereferencing
> > > > the pointer.
> > > >
> > > > Le 09/05/2014 08:43, Srivardhan Hebbar a écrit :
> > > > > Checking for NULL before dereferencing the wl_event_source pointer
> > > > > so as to avoid crash.
> > > > >
> > > > > Signed-off-by: Srivardhan Hebbar <sri.hebbar at samsung.com>
> > > > > ---
> > > > >   src/event-loop.c |    7 ++++++-
> > > > >   1 file changed, 6 insertions(+), 1 deletion(-)
> > > > >
> > > > > diff --git a/src/event-loop.c b/src/event-loop.c index
> > > > > 9790cde..b62d16e 100644
> > > > > --- a/src/event-loop.c
> > > > > +++ b/src/event-loop.c
> > > > > @@ -312,7 +312,12 @@ wl_event_source_check(struct
> > wl_event_source
> > > > *source)
> > > > >   WL_EXPORT int
> > > > >   wl_event_source_remove(struct wl_event_source *source)
> > > > >   {
> > > > > -	struct wl_event_loop *loop = source->loop;
> > > > > +	struct wl_event_loop *loop;
> > > > > +
> > > > > +	if (source == NULL)
> > > > > +		return 0;
> > > > > +
> > > > > +	loop = source->loop;
> > > > >
> > > > >   	/* We need to explicitly remove the fd, since closing the fd
> > > > >   	 * isn't enough in case we've dup'ed the fd. */
> > > > >
> > > >
> > > > Hello Srivardhan,
> > > >
> > > > do you have a case where this check is hit ? I may be wrong but
> > > > having no loop associated with a source event is not supposed to
> > > > happen. So my guess is that a caller of wl_event_source_remove has
> > > > forgotten to nullify the
> > > event
> > > > source after the call.
> > >
> > > Hi,
> > >
> > > I was going through the code of Weston. In the main function in
> > > compositor.c
> > > wl_event_loop_add_signal() is called to allocate the memory for
> > > signals[](Line no: 4196. struct wl_event_source *signals[4]). The
> > > function returns NULL if memory allocation failed. After that there is
> > > no NULL check for 'signals'. In the end while clearing up, this
> > > function is called. So if memory allocation failed then a NULL is
> > > passed to this function. Hence adding code to check for the NULL.
> > 
> > Hi,
> > 
> > I think we should be fixing the caller instead of
> wl_event_source_remove(). I
> > do not believe NULL is a valid argument for it, so the bug is in the
> caller.
> > 
> > If any wl_event_loop_add_signal() in Weston fails, it would be a reason to
> > exit with an error.
> > 
> 
> Oh... k... Will fix that and send the patch...
> 
> In general isn't it fine to check for NULL before dereferencing? 

Checking is one thing, silently hiding bugs is another thing.

If NULL is a legal input, then of course it needs to be checked.

If NULL can happen, but is a runtime error, the program needs to be
vocal about it, e.g. relay the error back to the caller.

If API specification says NULL is not a valid input, putting an
assert() would be fine, since violating that is a programmer error in
the caller.

I think wl_event_source_remove() falls into the last category. All
functions in wayland-util.h belong to this category, too.


Thanks,
pq