Re: [PATCH 0/2] (Was: BUG_ON in rcu_sync_func triggered)

From: Oleg Nesterov
Date: Thu Oct 13 2016 - 13:10:56 EST


On 09/26, Oleg Nesterov wrote:
>
> Hello,
>
> The patches do not depend on each other.

Yes,

> 1/2 is the trivial fix, imo -stable material. The bug is very old it seems,
> but today this race (leading to unbalanced unlock) manifests itself via
> mysterious BUG_ON's in rcu/sync.c.

Yes. Al, could you take it? Or how else we should route it?

> 2/2 is old, I forgot to send it before. It was already reviewed by Dave and Jan,
> but the generic/068 test from xfstests triggered the warning. This was fixed by
> dbad7c993053 "xfs: stop holding ILOCK over filldir callbacks" so we can hopefully
> kill the early-lockdep-release kludge.

Lets ignore it for now, it does lead to false-positives. Thanks again Jan and Dave
for your help.

Oleg.