[pulseaudio-discuss] [PATCH v2] memblock: Detect block ID collisions in pa_memimport_get

[Peter Meerwald]

> > Make sure we don't hand out the wrong block, in case we have the
> > same block ID in different SHM segments.
> 
> typo below, looks good otherwise

maybe it doesn't look that good; the message triggers from time to time in 
testing and then crashes my application

need to investigate, p.

> > ---
> > 
> > I pushed the second patch, which actually fixes the problem.
> > 
> > Looking at the code, I couldn't find a way b->type could be anything
> > else than PA_MEMBLOCK_IMPORTED, so I think the assert is safe.
> > 
> >  src/pulsecore/memblock.c | 7 +++++++
> >  1 file changed, 7 insertions(+)
> > 
> > diff --git a/src/pulsecore/memblock.c b/src/pulsecore/memblock.c
> > index eb85d1f..047a44a 100644
> > --- a/src/pulsecore/memblock.c
> > +++ b/src/pulsecore/memblock.c
> > @@ -1019,6 +1019,13 @@ pa_memblock* pa_memimport_get(pa_memimport *i, uint32_t block_id, uint32_t shm_i
> >      pa_mutex_lock(i->mutex);
> >  
> >      if ((b = pa_hashmap_get(i->blocks, PA_UINT32_TO_PTR(block_id)))) {
> > +        pa_assert(b->type == PA_MEMBLOCK_IMPORTED);
> > +        if (b->per_type.imported.segment->memory.id != shm_id) {
> > +            pa_log("Cannot import memory due to to block ID collision (block ID %u, expected shm ID %u, found shm ID %u)!",
> 
> to to 
> 
> > +                   block_id, shm_id, b->per_type.imported.segment->memory.id);
> > +            b = NULL;
> > +            goto finish;
> > +        }
> >          pa_memblock_ref(b);
> >          goto finish;
> >      }
> > 
> 
> 

-- 

Peter Meerwald
+43-664-2444418 (mobile)