svn commit: r268074 - head/sys/kern

[Mateusz Guzik]

On Tue, Jul 01, 2014 at 05:05:24PM +0200, Gary Jennejohn wrote:
> On Tue, 1 Jul 2014 16:19:20 +0200
> Mateusz Guzik <mjguzik at gmail.com> wrote:
> 
> > On Tue, Jul 01, 2014 at 09:40:57AM -0400, John Baldwin wrote:
> > > On Tuesday, July 01, 2014 2:29:15 am Mateusz Guzik wrote:
> > > > Modified: head/sys/kern/kern_sig.c
> > > > 
> > > ==============================================================================
> > > > --- head/sys/kern/kern_sig.c	Tue Jul  1 06:23:48 2014	(r268073)
> > > > +++ head/sys/kern/kern_sig.c	Tue Jul  1 06:29:15 2014	(r268074)
> > > > @@ -3453,10 +3453,6 @@ sigacts_copy(struct sigacts *dest, struc
> > > >  int
> > > >  sigacts_shared(struct sigacts *ps)
> > > >  {
> > > > -	int shared;
> > > >  
> > > > -	mtx_lock(&ps->ps_mtx);
> > > > -	shared = ps->ps_refcnt > 1;
> > > > -	mtx_unlock(&ps->ps_mtx);
> > > > -	return (shared);
> > > > +	return (ps->ps_refcnt > 1);
> > > >  }
> > > 
> > > You should KASSERT() in sigacts_shared that P_HADTHREADS is not set so that 
> > > new code does not call this function unsafely in the future.
> > > 
> > 
> > Looks more like !P_HADTHREADS || p->p_singlethread != NULL on the first
> > sight, but yeah, I'll add an assertion.
> > 
> 
> Also, please fix the build with gcc, which dies with an error that
> newsigacts may be use unitialzed.
> 
> What I did was set both oldsigacts and newsigacts to NULL before this
> line at around 622:
> 
> if (sigacts_share(p->p_sigacts)) {
> 
> which allowed me to remove the else branch for setting oldsigacts to
> NULL.
> 

Sorry for breakage.

Does the following fix the problem for you?

diff --git a/sys/kern/kern_exec.c b/sys/kern/kern_exec.c
index 91d161d..dc4eb9b 100644
--- a/sys/kern/kern_exec.c
+++ b/sys/kern/kern_exec.c
@@ -625,8 +625,10 @@ interpret:
 		oldsigacts = p->p_sigacts;
 		newsigacts = sigacts_alloc();
 		sigacts_copy(newsigacts, oldsigacts);
-	} else
+	} else {
 		oldsigacts = NULL;
+		newsigacts = NULL; /* satisfy gcc */
+	}
 
 	PROC_LOCK(p);
 	if (oldsigacts)

-- 
Mateusz Guzik <mjguzik gmail.com>