Re: [kernel-hardening] Re: initcall dependency problem (ns vs. threads)

[Vasiliy Kulikov <segoon@openwall.com>]

On Mon, Aug 01, 2011 at 12:07 -0700, Andrew Morton wrote:
> On Mon, 1 Aug 2011 23:03:41 +0400
> Vasiliy Kulikov <segoon@openwall.com> wrote:
> 
> > > Are we talking about init_ipc_ns.ids[] here?  If so, did you try
> > > initializing the three rwsems at compile-time?
> > > 
> > > That's rather a nasty hack though.  It'd be better to run the mystery
> > > init function before starting the threads.
> > 
> > Looks like it solves the race.
> 
> What patch are you talking about here?

Sorry for short sentences :)  I tried the patch you've suggested -
initialize rw_mutex in the init_ipc_ns declaration.  Surely, it solves a
specific race.  As no kernel threads actually use shm, other fields are
not needed to be initialized before do_initcall().

However, it is a bit ugly as it divides namespace initialization code
into init_ipc_ns initialization and other namespaces.  It's better to
use the same code for all namespaces (as it currently is).


> >  However, I think it should be solved on
> > another level.
> 
> What level?

I mean it is a bug of _implicit_ assume that kthreads don't use ns
related information.  So, AFAICS, it can be fixed 2 ways:

1) Move creations of kernel threads somewhere after namespaces
initializations in the init chain.

2) Deferring threads creation until all ns initialization is done.


> >  Other bugs might be hidden with this race.
> 
> What bugs?

I don't speak about specific bugs (I know the only one, which is this shm
related bug), but I suppose some threads might use some ns related
information as well.  At least I don't see whether it is somehow
explicitly denied currently.

Thanks,

-- 
Vasiliy Kulikov
http://www.openwall.com - bringing security into open computing environments