linux-kernel - Re: [PATCH] rt_mutex: correctly initialize lockdep in rt_mutex_init_proxy

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <20170612123443.ddpij6zws6jydfdj@hirez.programming.kicks-ass.net>
Date:   Mon, 12 Jun 2017 14:34:43 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     "Levin, Alexander (Sasha Levin)" <alexander.levin@...izon.com>
Cc:     Linus Torvalds <torvalds@...ux-foundation.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...nel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] rt_mutex: correctly initialize lockdep in
 rt_mutex_init_proxy_locked

On Sun, Jun 11, 2017 at 02:51:09PM +0000, Levin, Alexander (Sasha Levin) wrote:
> On Sat, Jun 10, 2017 at 04:02:12PM +0200, Peter Zijlstra wrote:
> > On Sat, Jun 10, 2017 at 02:48:04AM +0000, Levin, Alexander (Sasha Levin) wrote:
> > > lockdep can't deal with NULL name or key, and doesn't do anything
> > > with the lock when that happens.
> > 
> > Not doing anything is 'right', the proxy stuff won't be lockdep tracked
> > anyway. But yeah, the first thing is a wee bit of a problem, for it will
> > trigger DEBUG_LOCKS_WARN_ON() and fully kill lockdep.
> 
> But don't we want pi_state->pi_mutex tracked by lockdep?

Nope, we can't. That pi_mutex is owned by userspace and not all
operations upon it are visible to the kernel. That is, a userspace
thread can (conceptually) acquire the lock without the kernel ever
knowing.

We typically only create the pi_state when there's contention, at which
point we create the pi_mutex as owned by someone else (hence proxy).

Also, we 'obviously' hold the thing over the return to userspace, which
is something lockdep very much doesn't like.

> > Yeah, no need to do that; all we really need here is something like:
> > 
> > ---
> >  kernel/locking/rtmutex-debug.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/kernel/locking/rtmutex-debug.c b/kernel/locking/rtmutex-debug.c
> > index ac35e648b0e5..8dc647dc4b4b 100644
> > --- a/kernel/locking/rtmutex-debug.c
> > +++ b/kernel/locking/rtmutex-debug.c
> > @@ -175,7 +175,8 @@ void debug_rt_mutex_init(struct rt_mutex *lock, const char *name, struct lock_cl
> >  	lock->name = name;
> >  
> >  #ifdef CONFIG_DEBUG_LOCK_ALLOC
> > -	lockdep_init_map(&lock->dep_map, name, key, 0);
> > +	if (name && key)
> > +		lockdep_init_map(&lock->dep_map, name, key, 0);
> >  #endif
> >  }
> 
> I didn't want to do that because in later calls on that mutex we
> will end up going into lockdep code, and I didn't think that doing
> that without calling lockdep_init_map() initially was safe.

So futex has its own private rt_mutex implementation and interface, none
of which include lockdep hooks. So not initializing the field for that
case should not be a problem.