Re: Race condition between "read CFQ stats" and "block device shutdown"

[Tejun Heo <tj-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

Hello,

On Wed, Sep 04, 2013 at 08:45:33AM -0700, Anatol Pomozov wrote:
> I am not an expect in block code, so I have a few questions here:
> 
> - are we sure that this operation is atomic? What if blkg->q becomes
> dead right after we checked it, and blkg->q->queue_lock got invalid so
> we have the same crash as before?

request_queue lock switching is something inherently broken in block
layer.  It's unsalvageable.  Maybe we can drop lock switching once
blk-mq is fully merged.

> - do we need to check blk_queue_dying() in every single place where we
> iterate over blkcg->blkg_list?
> 
> - would it be better to remove block cgroup from blkcg->blkg_list
> before queue marked as dead?

I have no idea.  No matter what we do, it's working around something
fundamentally broken in a half-assed way.  Hmmm... there was an effort
to remove the lock switching.  I don't think it actually happened
tho. :(

Thanks.

-- 
tejun