Alyssa Ross <hi@alyssa.is> writes:

> Demi Marie Obenour <demiobenour@gmail.com> writes:
>
>> On 12/10/25 07:47, Alyssa Ross wrote:
>>> diff --git a/host/rootfs/image/usr/bin/vm-start b/host/rootfs/image/usr/bin/vm-start
>>> index 67480e52..c8031eec 100755
>>> --- a/host/rootfs/image/usr/bin/vm-start
>>> +++ b/host/rootfs/image/usr/bin/vm-start
>>> @@ -20,4 +20,21 @@ foreground {
>>>    redirfd -w 2 /dev/null
>>>    s6-svwait -U /run/service/vmm/instance/${1}
>>>  }
>>> -ch-remote --api-socket /run/vm/by-id/${1}/vmm boot
>>> +foreground { ch-remote --api-socket /run/vm/by-id/${1}/vmm boot }
>>> +importas -Siu ?
>>> +if {
>>> +  if -t { test $? -eq 0 }
>>> +
>>> +  # This is technically racy: if somehow we don't get here before the VM boots
>>> +  # and connects to xdg-desktop-portal-spectrum-host, it won't be able to
>>> +  # connect.  The VM rebooting will also break this, because the socket will be
>>> +  # re-created with the wrong mode, but VM reboots are broken anyway at the time
>>> +  # of writing:
>>> +  #
>>> +  # https://github.com/cloud-hypervisor/cloud-hypervisor/issues/7547
>>> +  #
>>> +  # Ideally we'd be able to give a listening socket FD to Cloud Hypervisor for
>>> +  # its VSOCK socket.
>>> +  chown xdp-spectrum-${1} /run/vsock/${1}/vsock
>>
>> It's possible to avoid the race using extended ACLs.
>
> Nice idea!

Actually I don't think it is, sadly.  Even with acls like the following,
when Cloud Hypervisor creates its socket, the mask ends up getting set
to ---, so xdp-spectrum-host still can't connect.  See also[1].

# file: run/vsock/GeOkfl
# owner: root
# group: root
user::rwx
group::r-x
other::r-x
default:user::rwx
default:user:xdp-spectrum-GeOkfl:rwx
default:group::r-x
default:mask::rwx
default:other::r-x

Even making the directory setgid wouldn't help, because the effective
mask applies to /all/ groups.  I don't think there's a way to do this at
the moment without either setting a less restrictive umask on Cloud
Hypervisor, or the approach I sent here.

[1]: https://serverfault.com/questions/833349/why-is-my-unix-socket-created-with-a-different-acl-mask-to-other-files