Re: [PATCH] Enable DBus activation on non-systemd systems.

[Mark Hindley <mark at hindley.org.uk>]

[-- Attachment #1: Type: text/plain, Size: 3762 bytes --]

Marcel,

Thanks

On Fri, Nov 12, 2021 at 03:57:59PM +0100, Marcel Holtmann wrote:
> > --- a/client/dbus-proxy.c
> > +++ b/client/dbus-proxy.c
> > @@ -855,13 +855,12 @@ bool dbus_proxy_init(void)
> > 	l_dbus_set_disconnect_handler(dbus, dbus_disconnect_callback, NULL,
> > 									NULL);
> > 
> > +	get_managed_objects();
> > 	if (command_is_interactive_mode())
> > 		l_dbus_add_service_watch(dbus, IWD_SERVICE,
> > 						service_appeared_callback,
> > 						service_disappeared_callback,
> > 						NULL, NULL);
> > -	else
> > -		get_managed_objects();
> 
> I don’t understand this change. The whole point is to actually wait for the
> service to appear.

Yes, but it is a chicken and egg situation: plain DBus activation relies on the
first call on the bus to trigger the activation[1].  The call at the end of
service_appeared_callback() is never going to be reached as interactive iwct is
just waiting having called service_disappeared_callback().

> 
> > 
> > 	return true;
> > }
> > diff --git a/src/main.c b/src/main.c
> > index 989665e4..8e3a5ca7 100644
> > --- a/src/main.c
> > +++ b/src/main.c
> > @@ -191,9 +191,6 @@ static void request_name_callback(struct l_dbus *dbus, bool success,
> > 		goto fail_exit;
> > 	}
> > 
> > -	if (!l_dbus_object_manager_enable(dbus, "/"))
> > -		l_warn("Unable to register the ObjectManager");
> > -
> > 	if (!l_dbus_object_add_interface(dbus, IWD_BASE_PATH,
> > 						IWD_DAEMON_INTERFACE,
> > 						NULL) ||
> > @@ -240,6 +237,9 @@ static void dbus_ready(void *user_data)
> > {
> > 	struct l_dbus *dbus = user_data;
> > 
> > +	if (!l_dbus_object_manager_enable(dbus, "/"))
> > +		l_warn("Unable to register the ObjectManager");
> > +
> 
> This change is unclear to me. Why would I register the ObjectManager if I
> can’t acquire the name. I don’t get how this is suppose to work. 

Without this change, once the name is acquired, dbus still fails to find the
ObjectManager path as it hasn't (yet) been enabled. So the first call to
get_managed_objects() fails if it was also the one that caused the service
activation.

> This is the right thing to do.

I wondered about the order here too. I know that they are different bindings and
that it is deprecated, so it may well not be the most convincing argument, but
GDBus docs specify this requirement[2] and use it in their examples[3] where
object paths are registered on bus acquisition rather than name acquisition.

> 
> > 	l_dbus_name_acquire(dbus, "net.connman.iwd", false, false, false,
> > 				request_name_callback, NULL);
> > 
> > diff --git a/src/net.connman.iwd.service b/src/net.connman.iwd.service.in
> > similarity index 77%
> > rename from src/net.connman.iwd.service
> > rename to src/net.connman.iwd.service.in
> > index d8ece4c3..a7cb7edd 100644
> > --- a/src/net.connman.iwd.service
> > +++ b/src/net.connman.iwd.service.in
> > @@ -1,5 +1,5 @@
> > [D-BUS Service]
> > Name=net.connman.iwd
> > -Exec=/bin/false
> > +Exec=@libexecdir@/iwd
> > User=root
> > SystemdService=iwd.service
> 
> This change is not for upstream git tree. Use --disable-systemd-service and
> then provide your own files in the distro packaging.

But this isn't about distro specific packaging, I think it is a setup that works
both with systemd or without it. Debian has both systemd and several other inits
available so building with --disabled-systemd-service is not an option.

Thanks

Mark

[1]  https://dbus.freedesktop.org/doc/dbus-specification.html#message-bus-starting-services

[2]  https://www.freedesktop.org/software/gstreamer-sdk/data/docs/latest/gio/ch30s03.html

[3]  https://gitlab.gnome.org/GNOME/glib/-/blob/HEAD/gio/tests/gdbus-example-server.c