[Xorp-users] Multicast problem

[Bruce M Simpson]

Ben Greear wrote:
> ...
> Is there a way to check to see if it's actually doing good things
> in xorpsh?

 > show route table ipv4 multicast fib2mrib

...should tell you.

>
> On a more basic level, can you let me know if this is close to how
> this scenario *should* work:
>
> * kernel mcast sends the pkts to the pimreg device, which shows up on
>  the mcast socket in mfea_mrouter.
>
> * Some piece of xorp (haven't found what yet) then makes a kernel
>  call to set up the MFC for this particular route.

The interaction with the multicast FIB happens in the MFEA (which is now 
mostly merged with the FEA). libmrt might also be worth a look.

>
>
>  I'm guessing fib2mrib or similar logic must query xorp's understanding
>  of the routing tables to set up the MFC correctly?

fib2mrib takes the unicast FIB contents and puts it into the multicast 
RIB, so that RPF checks can happen on each interface when a channel is 
brought up in PIM.

Otherwise we'd have a situation where traffic could be forwarded back 
across the same interface twice, resulting in a packet explosion.

Normally the RPF checks happen in kernel using interface indexes as the key.

>
> * After this, the kernel stops sending pkts to the pimreg device and 
> instead
>   sends them directly out the proper kernel interface.

That's correct, once the REGISTER goes upstream to the RP, the MFC 
should be set up.