[support] Bug MCoA + correction

[Romain KUNTZ]

Hi Kien,

Indeed, in the case you describe the current design is not good.  
Always removing the BULE upon disconnection seems like to easiest way  
to handle that, though it may not be the most optimized. I'll test  
your 2nd method and apply it when I'll update the MCoA code.

Thank you,
romain

On 2008/10/03, at 11:15, NGUYEN Duc Kien wrote:

> Hi,
>
> Here a bug in MCoA (20080108) + the cause + our correction.
>
> (-) Circumstance:
>
> We have MR who has 2 egress interface iface1 & iface2. First of all  
> we disconnect iface1, then iface2. After that iface1is reconnected:  
> in this case MR has only one interface iface1 connected.
>
> (-) Error:
>
> We always see 2 bules on MR/2 bces on HA, and the applications which  
> use MCoAv6 like Interpol continue to take iface2’s BID to mark the  
> packages -> corresponding flows are stopped because iface2 remains  
> still off-line.
>
> (-) Cause:
>
> MCoA keeps the last bule even the corresponding interface is  
> disconnected, and when the other interfaces reconnect, it is always  
> not deleted.
>
> (-) Correction:
>
> - 1st method: Delete the bule corresponding to interface off-line  
> when another interface connects.
>
> - 2rd method: Personally, I prefer to delete the last bule to  
> simplify some xfrm problems in IPv4.
>
> I think in the actual MCoA code, you want to keep it because the  
> time to (delete + then create) > the time to modify tunnels. It is  
> rather useful in the case we have only one interface like UMIP to  
> make handover; but with MCoA that is not very important. I thus  
> modify in src/mn.c->mcoa_mn_chk_ho_verdict (): remove “active_coa >  
> 1 &&“.
>
> Regards,
>
> Kien.
>
> _______________________________________________
> Support mailing list
> Support at ml.nautilus6.org
> http://ml.nautilus6.org/mailman/listinfo/support