git bluetoothd getting disconnected from dbus

I compiled a new version of bluez to try help test an
issue
(https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/issues/880), but now bluetoothd is being disconnected from dbus.

This is on a fedora 32, so the dbus is actually
https://github.com/bus1/dbus-broker.

The bluz revision is 3146b7a0785758be2d4e894d44e99d29c0db983e.

The behavior is that I can start bluetoothd with

$ sudo ./src/bluetoothd -n --debug='*' -f ./src/main.conf

and pulse audio with

$ ./src/pulseaudio -p $(pwd)/src/ --daemonize=no -v -v -v

Everything looks fine until I turn on a bluetooth device that was already paired and trusted. As far as I can tell, it connects correctly, but then the bluetoothd log shows

May 10 10:18:41 raspberrypi.local bluetoothd[3127]: profiles/audio/transport.c:transport_update_playing() /org/bluez/hci0/dev_EC_81_93_4A_C7_7E/fd0 State=TRANSPORT_STATE_ACTIVE Playing=1
May 10 10:18:41 raspberrypi.local bluetoothd[3127]: Disconnected from D-Bus. Exiting.

I can't quite figure out what is going on, but dbus-monitor around the time shows:

signal time=1589131121.073746 sender=org.freedesktop.DBus -> destination=:1.65 serial=4294967295 path=/org/freedesktop/DBus; interface=org.freedesktop.DBus; member=NameLost
string "org.bluez"
signal time=1589131121.073832 sender=org.freedesktop.DBus -> destination=(null destination) serial=4294967295 path=/org/freedesktop/DBus; interface=org.freedesktop.DBus; member=NameOwnerChanged
string "org.bluez"
string ":1.65"
string ""

Which is suspicious. I have attached the full logs starting just before the bluetooth speaker is turned on.

What I have checked so far

* I removed the system bluez and pulse audio
* I checked that ./src/bluetooth.conf has been copied to /etc/dbus-1/system.d/bluetooth.conf

Any suggestions as to where I should look next?

Thanks,
Rafael

Hi Rafael,

I suspect this is due to fd hand over, I usually have to SELINUX=disabled to disable SELinux policies when running bluetoothd from source because it seems there is some policy the prevents fd passing over D-Bus, we should probably add this to HACKING if we confirm that this is in fact SELinux policy.

Luiz Augusto von Dentz