Problem using the latest D-Bus enabled GDM on Solaris

[Brian Cameron]

Havoc:

Thanks for the help.  Running with truss helped to highlight the problem,
and it turned out to be issues where the code was passing NULL string
variables into g_debug statements.  I was able to fix that and now the
GDM D-Bus branch is working great on Solaris.  Thanks for the tips
about how to improve the generate_address function, and explaining
things.

Brian


> The key piece of info is that you aren't using the session bus, you're
> instead creating your own server. (Which is probably right for this.)
> 
> One guess is that you need to call
> dbus_connection_set_unix_user_function(), see
> http://dbus.freedesktop.org/doc/api/html/group__DBusConnection.html
> 
> Another guess is that in handle_connection(), you need to ref the
> connection and set it up with the main loop.
> 
> The socket should work fine afaik, though AIUI there are
> platform-specific details of how filesystem permissions apply to
> sockets. strace/truss should easily reveal if there's a permission
> problem.
> 
> I'm not sure there's a point creating a random /tmp name for your
> socket, btw. It creates a couple of problems (which you may have
> solved, but maybe with code that could be avoided):
>  1) when/how does the /tmp name get deleted?
>  2) how does a client find out the name?
> 
> The reason the session bus creates a random socket name is that it
> needs a new socket for every user session.
> 
> But the system bus just uses /var/lib/dbus/system_bus_socket or
> something like that; by avoiding /tmp there's no security issue with a
> well-known fixed name. Just use a directory that users can't write to.
> 
> With gdm you might be able to have something like
> /var/lib/gdm/socket_:0,  socket_:1, perhaps.
> 
> For the session bus, the cleanup problem is solved in two ways. On
> Linux, by using abstract sockets; and on other platforms the best I
> came up with was the dbus-cleanup-sockets binary that comes with dbus,
> which is meant to run in a cron job or something. It would not work
> (without modification) for gdm's sockets, but the same idea would. (I
> think it's just broken that sockets persist on the filesystem after
> their server is gone - a longstanding unix design screwup.)
> 
> For the session bus, the discovery problem is solved with the
> DBUS_SESSION_BUS_ADDRESS environment variable.
> 
> One last note, if you do use a random socket name, you can avoid the
> #ifdef linux thing in your code; just always use the tmpdir=/tmp
> version. If you specify tmpdir then abstract sockets are automatically
> used when available.
> 
> If you use a fixed socket name, you can avoid the ifdef by just never
> using abstract sockets, since their primary benefit is avoiding the
> cleanup problem, and with fixed socket names cleanup doesn't really
> matter.
> 
> So either way you should not need any ifdef here.
> 
> Havoc