From mboxrd@z Thu Jan 1 00:00:00 1970 From: Mike Galbraith Date: Fri, 24 Dec 2010 17:24:51 +0000 Subject: Re: [PATCH] sched, cgroup: Use exit hook to avoid use-after-free Message-Id: <1293211491.15747.3.camel@marge.simson.net> List-Id: References: <1277808215.1868.5.camel@laptop> <20101219020313.GJ31750@genesis.frugalware.org> <20101222002248.GP10557@genesis.frugalware.org> <1293006589.2170.41.camel@laptop> <1293007311.11370.172.camel@marge.simson.net> <1293008842.2170.70.camel@laptop> <20101222133154.GS10557@genesis.frugalware.org> <1293026422.2170.136.camel@laptop> <1293027112.2170.140.camel@laptop> <20101222151434.GW10557@genesis.frugalware.org> <1293037718.2170.155.camel@laptop> <1293050173.2170.389.camel@laptop> <1293106330.2170.618.camel@laptop> <1293107624.2170.642.camel@laptop> <1293128670.2170.748.camel@laptop> <1293132304.6798.6.camel@marge.simson.net> <1293132862.25981.22.camel@laptop> <1293187425.7138.2.camel@marge.simson.net> <1293188091.25981.200.camel@laptop> <1293192999.18035.4.camel@marge.simson.net> <1293206353.29444.205.camel@laptop> <1293209297.14433.4.camel@marge.simson.net> <1293210457.29444.206.camel@laptop> In-Reply-To: <1293210457.29444.206.camel@laptop> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Peter Zijlstra Cc: Miklos Vajna , shenghui , kernel-janitors@vger.kernel.org, linux-kernel@vger.kernel.org, mingo@elte.hu, Greg KH , Paul Turner , Yong Zhang , Li Zefan , Paul Menage , Balbir Singh , Srivatsa Vaddagiri On Fri, 2010-12-24 at 18:07 +0100, Peter Zijlstra wrote: > On Fri, 2010-12-24 at 17:48 +0100, Mike Galbraith wrote: > > On Fri, 2010-12-24 at 16:59 +0100, Peter Zijlstra wrote: > > > > > So cgroup moves a task without calling cgroup_subsys::attach() which is > > > odd, but it does have an ::exit method, sadly it calls that _before_ > > > re-assigning the task, which means we have to jump through some hoops. > > > > Could you do the move in cgroup_exit() in the CONFIG_PREEMPT case? > > I'm not really comfortable relying on that.. voluntary might just grow a > cond_resched() somewhere in the exit path and lead us down the same > path, also I think that !preempt smp might have the same race. Yeah, good point. -Mike