Multicast group memberships lost if eth0 brought down and up

[Harald Hoyer]

Deron Meranda wrote:
> On Sat, 02 Oct 2004 07:49:01 -0400, Ted Kaczmarek <tedkaz at optonline.net> wrote:
> 
>>up/down cycling of what?
>>Most switches will and should flush all tables on a port state change.
>>I am assuming you are in a switch that is doing igmp snooping as well.
> 
> 
> Thanks for the response, but my problem is for a normal application
> running in user-space, not in the kernel.  IGMP or snooping is not
> involved (at least that the application directly knows about).  The
> application is only using standard RFC 3542 socket API calls.
> 
> What I have is an application program which opens an IPv6 socket,
> binds it to a multicast group address and UDP port, and then adds it
> to the multicast membership (an IPV6_ADD_MEMBERSHIP socket option). 
> Binding allows one to send packets to a multicast group, but only by
> setting the socket option can one also receive packets from the group.
> 
> The problem occurs when the interface is brought down and then back
> up.  The kernel looses track of all of the multicast memberships that
> used to be associated with that interface.  But the application is in
> no way informed that it has lost its membership; the socket descriptor
> is still perfectly valid and will receive no errors.  Packets can
> still be sent successfully to the group.  However the application is
> now in a state where it can send packets but will never receive them.
> 
> I would have expected the kernel to save group memberships when it
> goes down and to re-establish memberships when it's brought back up. 
> It may not even need to save memberships, since all that information
> should be recorded among all the open descriptor structures.  When an
> interface comes back up the kernel should do all the IGMP and
> device-level registering automatically.  Yes, the network peers will
> (and should) see the host leave and rejoin the group, but it should be
> transparent to the application; and the application should contrinue
> to function correctly once the interface is brought back up.
> 
> Or, as a less desirable workaround, provide a way for an application
> to be notified when an interface goes down or back up (yes, there are
> kernel-level notifications, but nothing available at user-space. that
> I know about).
> 

I think this should be discussed on the linux kernel mailing list...
http://www.tux.org/lkml/
http://lkml.org/