Hi Julien, On Thu, Jan 13, 2011 at 12:18:49PM +0100, Julien BLACHE wrote: > > This is fine with me, but shouldn't we instead use a virtual package such as > > (the currently not yet existing package) 'multicast-routing-daemon' that > > both > > smcroute is *NOT* a multicast routing daemon :) > > A multicast routing daemon is a daemon that manages the multicast > routing table dynamically based on IGMPv3 signalling.
Speaking about real multicast routing you're right and I totally agree that smcroute is *not* a multicast routing daemon in this sense. But from the point of view of the kernel smcroute claims that it *is* a multicast routing daemon by calling setsockopt() with MRT_INIT or MRT6_INIT. And exactly this setsockopt() call is the reason why you cannot start another (or a real) multicast routing daemon when smcroute is running: The kernel by design refuses to setsockopt() with MRT_INIT on another socket as long as the previous MRT_INIT-ialized socket is still open. Thus I maintain my suggestion to add Provides:/Conflicts: dependencies to a virtual package (e.g. 'multicast-routing-daemon', but feel free to suggest another name). > The proper thing to do here is to demote smcroute to Priority: extra and > leave pimd at Priority: optional. This would be fine with me as well. > The pimd/xorp situation should be investigated, as I'm not sure both can > be installed at the same time either. They cannot because multicast routing always needs access to the multicast routing table (MRT), which the kernel lets you only access through a socket after calling setsockopt() with MRT_INIT (see above). Regards, Micha -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
