Re: [patch, validator] fix proc_subdir_lock related deadlock

[Andrew Morton]

Ingo Molnar <mingo@xxxxxxx> wrote:
>
> 
> * Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> 
> > > proc_subdir_lock can also be used from softirq (tasklet) context, which 
> > > may lead to deadlocks.
> > > 
> > > This bug was found via the lock validator:
> > > 
> > 
> > Thanks Ingo,
> > 
> > I stressed in sending the patch that there was a big assumption that 
> > the calls would not be done in (soft)irq context.  I just didn't want 
> > to add overhead if it wasn't needed.  But I guess that this is needed 
> > until we can remove all the instances that use it in softirq context. 
> > But that's for a later patch.
> 
> the validator just found another problem with this lock, pointing out 
> that files_lock nests inside of proc_subdir_lock, and that files_lock is 
> a softirq-unsafe lock, creating another (unlikely but possible) deadlock 
> scenario:

files_lock can be taken on the free_irq() path: proc_kill_inodes().

> ...
> to solve this we must either change files_lock to be softirq-safe too 
> (bleh!), or we must forbid remove_proc_entry() use from softirq 
> contexts. Neither is a happy solution - remove_proc_entry() is used 
> within free_irq(), and who knows how many drivers do free_irq() in 
> softirq/tasklet context ...

free_irq()'s /proc fiddling has always been a pain - we just shouldn't be
doing filesystem things in irq/bh context.

> Andrew, this needs to be resolved before v2.6.16, correct? Steve's patch 
> solves a real bug in the upstream kernel.

It's not a very big bug - I think only Steve hit it, and that with a
stress-test which was somewhat tuned to hit it.

So we can afford to sit on the problem for a while, as long as someone is
working on a broader /proc-sanity fix.  But nobody will do that.

I wonder if we can just punt the unregister_handler_proc/kfree up to a
keventd callback.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/