Received-SPF: pass (google.com: domain of selinux-refpolicy-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <0397bf1e-415f-b045-5a6d-c12fde48a65a@linux.microsoft.com>
Date:   Fri, 9 Dec 2022 09:29:16 -0700
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:102.0) Gecko/20100101
 Thunderbird/102.5.1
Subject: Re: [refpolicy3 RFC] Split broad file contexts
Content-Language: en-US
To:     Chris PeBenito <chpebeni@linux.microsoft.com>,
        SELinux Reference Policy mailing list 
        <selinux-refpolicy@vger.kernel.org>
References: <d83ef10f-ae8b-08d2-55b7-66f2cf12ed9a@linux.microsoft.com>
From:   Matthew Sheets <masheets@linux.microsoft.com>
In-Reply-To: <d83ef10f-ae8b-08d2-55b7-66f2cf12ed9a@linux.microsoft.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 12/9/2022 9:09 AM, Chris PeBenito wrote:
> In refpolicy2, we have several types, such as bin_t, that have file 
> contexts related to other modules, e.g.:
> 
> /etc/acpi/actions(/.*)?            gen_context(system_u:object_r:bin_t,s0)
> 
> /usr/lib/mailman/bin(/.*)?        gen_context(system_u:object_r:bin_t,s0)
> 
> /var/mailman/bin(/.*)?            gen_context(system_u:object_r:bin_t,s0)
> 
> 
> relate to acpi and mailman.
> 
> Should we continue to put all of the bin_t labeling in files.cas or 
> should we split it back to the individual modules?  This was originally 
> done in refpolicy2 so users could look in a single place for everything 
> about bin_t for encapsulation.  This is nice for users, but not so nice 
> for maintenance and version control.
> 
> Since cascade has the "extend" feature, we can split up the labeling 
> among relevant modules, and tooling can construct a single unified view 
> of the file contexts of bin_t and the like.
> 
> For example, instead of this in file.cas:
> 
> resource bin_t inherits executable {
>    ...many fcs...
>    file_context(/etc/acpi/actions(/.*)?, any);
> }
> 
> we have this in acpi.cas:
> 
> extend bin_t {
>    file_context(/etc/acpi/actions(/.*)?, any);
> }
> 
> 
> Thoughts?
> 
> 

If I turn of a module I want as little residue as possible in other modules.

I think using extend is the best option.  Keep all relative types and 
labeling in the singular module so if it is not used we heavily reduce 
the possibility of unexpected labeling/access.

The users should understand that a single cas file does not define the 
whole policy, so if they see something that may be "off" with labeling 
they may need to look further.  And with the new tooling, looking 
further should be easy.