Re: [oe-core][PATCH] sysvinit: create tmpfile dir for x11 domain socket

[Randolph Sapp <rs@ti.com>]

On Fri Oct 17, 2025 at 7:01 PM CDT, Randolph Sapp via lists.openembedded.org wrote:
> On Wed Oct 15, 2025 at 12:22 PM CDT, Randolph Sapp via lists.openembedded.org wrote:
>> On Wed Oct 15, 2025 at 9:37 AM CDT, Ross Burton wrote:
>>> On 8 Oct 2025, at 00:48, rs@ti.com wrote:
>>>> 
>>>> From: Randolph Sapp <rs@ti.com>
>>>> 
>>>> Register a volatile directory entry for the x11 domain socket when not
>>>> using systemd. This will make sure the directory is always created with
>>>> the correct permissions. Systemd already provides their own tmpfile.d
>>>> entry for the same behavior.
>>>
>>> We discussed this in the patch review call on Monday and concluded that it
>>> definitely felt like it should be in the xserver recipe.
>>
>> Sure.
>>
>
> I am a little bummed that we don't want to have symmetry between init system
> packages though. Gonna need a gross "VIRTUAL-RUNTIME_init_manager" check
> to make sure we don't clash with tmpfiles.d entry.
>
>>>> Currently some x11 related applications will create this directory if it
>>>> doesn't already exist, but this is not true for everything. In addition,
>>>> if the application in question isn't started as root, it's possible this
>>>> directory can be owned by a non-root user. This isn't an issue by
>>>> itself, but it can potentially lead to problems in a multi-user
>>>> environment.
>>>
>>> Can you clarify what applications will create this directory and which will
>>> not? Surely it’s the responsibility for the xserver to create the sockets?
>>>
>>> Ross
>>
>> Forgive me, as it's been quite some time since I initially tested this. I
>> believe during testing with the current default xsession managers will create it
>> but xwayland sessions will not. Given that systemd has determined they need to
>> ship this as a global tmpfiles.d entry I assume there is more nuance to this
>> than what I have outlined.
>>
>> - Randolph
>
> Ah, now I remember, all rootless X sessions also have the capability of
> generating this socket directory with the wrong owner, given that the server may
> try to make the directory after it's already dropped to user permissions. It's
> normally fine because they throw 777 at it. Not great, but whatever.
>
> - Randolph

Are we still dead set on splitting this into both xwayland and xorg-xserver and
conditionally masking it based on whether or not systemd is selected? Feels odd.