linux-kernel - Re: [PATCH 17/19] sched: Inherit task cookie on fork()

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]

Date:   Mon, 10 May 2021 17:38:18 -0400
From:   Chris Hyser <chris.hyser@...cle.com>
To:     Joel Fernandes <joel@...lfernandes.org>
Cc:     Peter Zijlstra <peterz@...radead.org>,
        Josh Don <joshdon@...gle.com>, Ingo Molnar <mingo@...nel.org>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        Valentin Schneider <valentin.schneider@....com>,
        Mel Gorman <mgorman@...e.de>,
        LKML <linux-kernel@...r.kernel.org>,
        Thomas Glexiner <tglx@...utronix.de>
Subject: Re: [PATCH 17/19] sched: Inherit task cookie on fork()

On 5/10/21 4:47 PM, Joel Fernandes wrote:
> On Mon, May 10, 2021 at 12:23 PM Chris Hyser <chris.hyser@...cle.com> wrote:

>>>> +void sched_core_fork(struct task_struct *p)
>>>> +{
>>>> +       RB_CLEAR_NODE(&p->core_node);
>>>> +       p->core_cookie = sched_core_clone_cookie(current);
>>>
>>> Does this make sense also for !CLONE_THREAD forks?
>>
>> Yes. Given the absence of a cgroup interface, fork inheritance (clone the cookie) is the best way to create shared
>> cookie hierarchies. The security issue you mentioned was handled in my original code by setting a unique cookie on
>> 'exec', but Peter took that out for the reason mentioned above. It was part of the "lets get this in compromise" effort.
> 
> Thanks for sharing the history of it. I guess one can argue that this
> policy is better to be hardcoded in userspace since core-scheduling
> can be used for non-security usecases as well. Maybe one could simply
> call the prctl(2) from userspace if they so desire, before calling
> exec() ?

I think the defining use case is a container's init. If the cookie is set for it by the container creator and without 
any other user code knowing about core_sched, every descendant spawned will have the same cookie and be in the same 
core_sched group much like the cgroup interface had provided. If we create a unique cookie in the kernel either on fork 
or exec, we are secure, but we will now have 1000's of core sched groups.

CLEAR was also removed (temporarily, I hope) because a core_sched knowledgeable program in the example core_sched 
container group should not be able to remove itself from _all_ core sched groups. It can modify it's cookie, but that is 
no different than the normal case.

Both of these beg for a kernel policy, but that discussion was TBD.

-chrish