Re: [Linuxptp-users] Multicast group (re)join issue vs IGMP snooping

[Richard C. <ric...@gm...>]

On Thu, Feb 17, 2022 at 11:29:03PM +0000, Keller, Jacob E wrote:

> I think we at least need to re-send these when the link changes,

We do, because the link down/up causes the transport to be closed and
opened again.

> though
> we perhaps were thinking that the kernel does this for us? Maybe thats a
> bug or maybe we have to set an option?
> 
> There is no mention about whether we would need to re-issue this socket
> option in the "man 7 ip" manual page.

>From the application POV, there is only joining the group.  After
that, there is nothing more to do.

> > Maybe it is Linux kernel issue.

Then take it up on the netdev list please.

> However for comparison ptpd2 implements 
> > periodic multicast group (re)join (optionmaster_igmp_refresh_interval):
> > src/dep/net.c:netInitMulticastIPv4() 
> > https://github.com/ptpd/ptpd/blob/master/src/dep/net.c#L605 
> > https://github.com/ptpd/ptpd/blob/master/src/dep/net.c#L2278 
> > https://github.com/ptpd/ptpd/blob/master/src/protocol.c#L607 
> > https://github.com/ptpd/ptpd/blob/master/src/protocol.c#L1120 Other 
> > proprietary PTP devices send IGMP packets periodically every 1-2s and 

Crazy to spam the network like that.

> > work with the switches despite IGMP snooping enabled...
> > 
> 
> This sounds like what we should be doing as well.

No, I don't agree.  This is no different than ARP.  If the kernel
needs to re-send IGMP, then let it.

Cisco's PTP switch has issues.  Have you looked at the residence
times?  I don't feel like implementing workarounds for buggy hardware.

Thanks,
Richard