Received-SPF: pass (google.com: domain of selinux-refpolicy-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: machinectl shell policy
To:     Russell Coker <russell@coker.com.au>,
        Dominick Grift <dominick.grift@defensec.nl>
Cc:     selinux-refpolicy@vger.kernel.org
References: <8322849.62pqQp6Oog@liv> <1723812.Y751QtlBzf@liv>
 <ypjl8s9m5nbi.fsf@defensec.nl> <5735537.jZnottUgFY@liv>
From:   Chris PeBenito <pebenito@ieee.org>
Message-ID: <f142c9ee-ce7e-7f7d-dec4-9d93ddd49c43@ieee.org>
Date:   Mon, 4 Jan 2021 09:48:42 -0500
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.6.0
MIME-Version: 1.0
In-Reply-To: <5735537.jZnottUgFY@liv>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 12/25/20 4:16 AM, Russell Coker wrote:
> On Friday, 25 December 2020 6:58:41 PM AEDT Dominick Grift wrote:
>> Russell Coker <russell@coker.com.au> writes:
>>> On Thursday, 24 December 2020 7:37:50 PM AEDT Dominick Grift wrote:
>>>>> To enable "machinectl shell" on recent versions of systemd we need
>>>>> something like the above policy (which is not complete or ideal, still
>>>>> doesn't work so no point polishing it) and something for the below.
>>>>> What
>>>>> is the below about?
>>>>
>>>> this should be thoroughly addressed. machined creates a login pty that
>>>> gets relabeled on login as per type_change rules.
>>>
>>> Currently it's not being relabeled on Debian, but that's a separate issue.
>>
>> Maybe the required type_change rules arent present?
> 
> Here is all the policy to make it work.  Maybe we should have a type like
> system_dbusd_devpts_t for this.  This is not policy for inclusion, this is
> policy to discuss before writing that policy.
> 
> term_user_pty(user_systemd_t, user_devpts_t)
> term_login_pty(devpts_t)
> allow user_systemd_t user_devpts_t:chr_file rw_file_perms;
> 
> # for machinectl shell
> allow sysadm_t systemd_machined_t:dbus send_msg;
> systemd_manage_userdb_runtime_dirs(systemd_machined_t)
> systemd_manage_userdb_runtime_sock_files(systemd_machined_t)
> term_use_ptmx(systemd_machined_t)
> dev_getattr_fs(systemd_machined_t)
> term_getattr_pty_fs(systemd_machined_t)
> allow systemd_machined_t sysadm_t:dbus send_msg;
> allow systemd_machined_t devpts_t:chr_file rw_file_perms;
> allow system_dbusd_t systemd_machined_t:fd use;
> allow system_dbusd_t devpts_t:chr_file { read write };
> allow system_dbusd_t ptmx_t:chr_file { read write };
> allow sysadm_t systemd_machined_t:fd use;
> allow user_systemd_t shell_exec_t:file entrypoint;

The pty stuff seems to make sense, but I'm curious why there is a transition 
into user_systemd_t for the shell.

> allow user_systemd_t systemd_machined_t:fd use;
> allow user_systemd_t self:process signal;
> allow user_t systemd_machined_t:fd use;
> allow user_t user_systemd_t:fifo_file { getattr write };
> allow user_t init_t:process signal;


>>> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=892001
>>>
>>> We have work in progress on dbus-broker in Debian.  Would it make sense to
>>> only support dbus-broker in SE Linux policy?  Being forced to use only 1
>>> of
>>> the 2 dbus programs (and the newer and faster 1 of the 2) is a very small
>>> trade-off, smaller than some of the other trade-offs for running SE Linux.

I'd prefer to keep both unless it becomes onerous.


>> should probably be able to support both (conditionally) but could get messy
> 
> Currently we have a heap of ifdef systemd in the policy, as probably the only
> people not wanting dbus-broker will be the ones not wanting systemd we could
> include it in the same ifdef rules.

The "else" of the ifdef can work.

-- 
Chris PeBenito