Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <15c607d9-c1fa-ca11-d675-8f2b3a6fd15b@redhat.com>
Date:   Thu, 29 Jun 2023 08:11:33 -0400
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.7.1
Subject: Re: [PATCH 1/3] cpuset: Allow setscheduler regardless of manipulated
 task
Content-Language: en-US
To:     =?UTF-8?Q?Michal_Koutn=c3=bd?= <mkoutny@suse.com>,
        linux-kernel@vger.kernel.org, cgroups@vger.kernel.org,
        linux-kselftest@vger.kernel.org
Cc:     Zefan Li <lizefan.x@bytedance.com>, Tejun Heo <tj@kernel.org>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Shuah Khan <shuah@kernel.org>
References: <20230629091146.28801-1-mkoutny@suse.com>
 <20230629091146.28801-2-mkoutny@suse.com>
From:   Waiman Long <longman@redhat.com>
In-Reply-To: <20230629091146.28801-2-mkoutny@suse.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 6/29/23 05:11, Michal Koutný wrote:
> When we migrate a task between two cgroups, one of the checks is a
> verification whether we can modify task's scheduler settings
> (cap_task_setscheduler()).
>
> An implicit migration occurs also when enabling a controller on the
> unified hierarchy (think of parent to child migration). The
> aforementioned check may be problematic if the caller of the migration
> (enabling a controller) has no permissions over migrated tasks.
> For instance, a user's cgroup that ends up running a process of a
> different user. Although cgroup permissions are configured favorably,
> the enablement fails due to the foreign process [1].
>
> Change the behavior by relaxing the permissions check on the unified
> hierarchy (or in v2 mode). This is in accordance with unified hierarchy
> attachment behavior when permissions of the source to target cgroups are
> decisive whereas the migrated task is opaque (as opposed to more
> restrictive check in __cgroup1_procs_write()).

The is_in_v2_mode() check is for supporting the v2 mode in cgroup v1. 
However, there is no controller enabling in v1. So I think you should 
just use cgroup_subsys_on_dfl(cpuset_cgrp_subsys) as the v2 check if 
your focus is just to prevent problem when enabling cpuset controller.


>
> [1] https://github.com/systemd/systemd/issues/18293#issuecomment-831205649
>
> Signed-off-by: Michal Koutný <mkoutny@suse.com>
> ---
>   kernel/cgroup/cpuset.c | 7 +++++++
>   1 file changed, 7 insertions(+)
>
> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> index e4ca2dd2b764..3b5f87a9a150 100644
> --- a/kernel/cgroup/cpuset.c
> +++ b/kernel/cgroup/cpuset.c
> @@ -2495,6 +2495,13 @@ static int cpuset_can_attach(struct cgroup_taskset *tset)
>   		ret = task_can_attach(task, cs->effective_cpus);
>   		if (ret)
>   			goto out_unlock;
> +
> +		/*
> +		 * Skip rights over task check in v2, migration permission derives
> +		 * from hierarchy ownership in cgroup_procs_write_permission()).
> +		 */
> +		if (is_in_v2_mode())
> +			continue;
>   		ret = security_task_setscheduler(task);
>   		if (ret)
>   			goto out_unlock;

This change will likely conflict with the latest cpuset change on 
tracking # of dl tasks in a cpuset. You will have to, at least, move the 
dl task check before the security_task_setscheduler() check.

Another fact about cpuset controller enabling is that both cpus_allowed 
and mems_allowed are empty at that point. You may also add these checks 
as a preconditions for disabling the security_task_setscheduler check.

Cheers,
Longman