linux-kernel - Re: [PATCH v3 11/14] sched: Reject CPU affinity changes based on arch_cpu_allowed

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

Message-ID: <20201119110723.GE3946@willie-the-truck>
Date:   Thu, 19 Nov 2020 11:07:24 +0000
From:   Will Deacon <will@...nel.org>
To:     Quentin Perret <qperret@...gle.com>
Cc:     linux-arm-kernel@...ts.infradead.org, linux-arch@...r.kernel.org,
        linux-kernel@...r.kernel.org,
        Catalin Marinas <catalin.marinas@....com>,
        Marc Zyngier <maz@...nel.org>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Peter Zijlstra <peterz@...radead.org>,
        Morten Rasmussen <morten.rasmussen@....com>,
        Qais Yousef <qais.yousef@....com>,
        Suren Baghdasaryan <surenb@...gle.com>,
        Tejun Heo <tj@...nel.org>, Li Zefan <lizefan@...wei.com>,
        Johannes Weiner <hannes@...xchg.org>,
        Ingo Molnar <mingo@...hat.com>,
        Juri Lelli <juri.lelli@...hat.com>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        kernel-team@...roid.com
Subject: Re: [PATCH v3 11/14] sched: Reject CPU affinity changes based on
 arch_cpu_allowed_mask()

On Thu, Nov 19, 2020 at 09:47:44AM +0000, Quentin Perret wrote:
> On Friday 13 Nov 2020 at 09:37:16 (+0000), Will Deacon wrote:
> > Reject explicit requests to change the affinity mask of a task via
> > set_cpus_allowed_ptr() if the requested mask is not a subset of the
> > mask returned by arch_cpu_allowed_mask(). This ensures that the
> > 'cpus_mask' for a given task cannot contain CPUs which are incapable of
> > executing it, except in cases where the affinity is forced.
> > 
> > Signed-off-by: Will Deacon <will@...nel.org>
> > ---
> >  kernel/sched/core.c | 4 ++++
> >  1 file changed, 4 insertions(+)
> > 
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index 8df38ebfe769..13bdb2ae4d3f 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -1877,6 +1877,7 @@ static int __set_cpus_allowed_ptr_locked(struct task_struct *p,
> >  					 struct rq_flags *rf)
> >  {
> >  	const struct cpumask *cpu_valid_mask = cpu_active_mask;
> > +	const struct cpumask *cpu_allowed_mask = arch_cpu_allowed_mask(p);
> >  	unsigned int dest_cpu;
> >  	int ret = 0;
> >  
> > @@ -1887,6 +1888,9 @@ static int __set_cpus_allowed_ptr_locked(struct task_struct *p,
> >  		 * Kernel threads are allowed on online && !active CPUs
> >  		 */
> >  		cpu_valid_mask = cpu_online_mask;
> > +	} else if (!cpumask_subset(new_mask, cpu_allowed_mask)) {
> > +		ret = -EINVAL;
> > +		goto out;
> 
> So, IIUC, this should make the sched_setaffinity() syscall fail and
> return -EINVAL to userspace if it tries to put 64bits CPUs in the
> affinity mask of a 32 bits task, which I think makes sense.
> 
> But what about affinity change via cpusets? e.g., if a 32 bit task is
> migrated to a cpuset with 64 bit CPUs, then the migration will be
> 'successful' and the task will appear to be in the destination cgroup,
> but the actual affinity of the task will be something completely
> different?

Yeah, the cpuset code ignores the return value of set_cpus_allowed_ptr() in
update_tasks_cpumask() so the failure won't be propagated, but then again I
think that might be the right thing to do. Nothing prevents 32-bit and
64-bit tasks from co-existing in the same cpuseti afaict, so forcing the
64-bit tasks onto the 32-bit-capable cores feels much worse than the
approach taken here imo. Nothing says we _have_ to schedule on all of the
cores in the mask.

The interesting case is what happens if the cpuset for a 32-bit task is
changed to contain only the 64-bit-only cores. I think that's a userspace
bug, but the fallback rq selection should avert disaster.

Will