Received-SPF: pass (google.com: best guess record for domain of selinux-refpolicy-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: Access to raw memory: remove or make boolean?
To:     russell@coker.com.au
Cc:     selinux-refpolicy@vger.kernel.org
References: <11011d01-844e-c526-a85f-92a7fc985d16@gmail.com>
 <2111138.8pYWFqxgyc@xev> <710a52cc-5b72-e833-9ac6-4289b1bc0b61@gmail.com>
 <2012273.jey3ENlaR0@xev>
From:   Topi Miettinen <toiwoton@gmail.com>
Message-ID: <769ff926-5cce-a14b-0579-6b81a436c5a9@gmail.com>
Date:   Tue, 25 Feb 2020 10:54:33 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.5.0
MIME-Version: 1.0
In-Reply-To: <2012273.jey3ENlaR0@xev>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: selinux-refpolicy-owner@vger.kernel.org
Precedence: bulk

On 25.2.2020 2.27, Russell Coker wrote:
> On Tuesday, 25 February 2020 2:56:01 AM AEDT Topi Miettinen wrote:
>> The PR would make all these conditional to new boolean,
>> allow_raw_memory_access.
> 
> So if someone needs one of those many accesses (klogd_t or hald_t seems
> likely) then they also get access for things that aren't needed on most
> systems nowadays (EG xserver_t) and things that never made any sense (such as
> colord_t).
> 
> I think it would be best to remove most of those /dev/mem access rules and add
> them back only after testing with recent software and comments about why they
> are needed.
> 
>>> A quick grep of the latest policy turned up the above access to /dev/mem.
>>> Do ddcprobe_t, vbetool_t, and the X server still do that?  mcelog_t, and
>>> klogd_t might have good uses, as might sosreport_t (don't even know what
>>> it does but guessing it's like klogd_t).  rpm_t should maybe transition
>>> to a different domain for whatever it was doing and the same for kudzo_t.
>>>   Vmware  is a bit ugly, so vmware_t might actually do that.  iscsi_t,
>>> mdadm_t, colord_t, and initrc_t should never have needed that.  hald_t,
>>> hald_mac_t and
>>> devicekit_disk_t might have needed it, but hopefully that was fixed a long
>>> time ago.
>>>
>>> Interestingly bootloader_t doesn't have such access even though a quick
>>> inspection of the LILO source code shows that it still probes the boot
>>> order by directly reading the BIOS memory.  I guess no-one uses LILO with
>>> SE Linux.
>> I also don't know most of these programs. Direct memory access was
>> probably needed for X server during SVGA times, at least NVIDIA driver
>> on my system does not seem to need it.
> 
> I think it was needed before KMS.  Is it even possible to run without KMS
> nowadays?

No idea. I'll update the PR so that only ddcprobe_t, vbetool_t, 
mcelog_t, klogd_t, sosreport_t and vmware_t keep the access (subject to 
boolean) and for others it's just removed unconditionally.

-Topi