Re: about R1bis

[Erik Nordmark <erik.nordmark@sun.com>]

marcelo bagnulo braun wrote:
> Two more issues related to R1bis that came to my mind:
>
> 1 - Is it important to defend a peer from a fake R1bis?
> I mean, R1bis may enable some attacks similar to sending a fake TCP RST 
> packet.
> If a peer receives a R1bis packet refering to an established context, it 
> will try to restablish the context, sending an I2 packet. If we assume 
> that the R1bis packet is fake and that the validator is not correct, the 
> receiver will siliently discard the I2 packet. At this point the 
> receiver will try to resend the I2 packet or try to send an I1 packet 
> again. I2 will continue to be discarded, but I1 packet will get a R2 
> reply, which imho will be good enough in this case. I guess that this 
> covers this case pretty good, so i don't think that anything else is 
> needed.

ok

And faking an R1bis requires knowing the context tag.

> 2 - Do we want that the failure detection probes trigger a context 
> recovery mechanisms i.e. R1bis?

Yes.

> In the considered scenario, when the initial path between the ulids is 
> working fine, the transmission of probes to verify that the current path 
> is working may result in triggering the context recovery mechanisms i.e. 
> the transmission of a R1bis packet. Such behaviour may not be the 
> optimum for dealing with the case of the server that wants to discard 

But I don't think we need to have optimality; question is whether it 
works or not.

  Erik

> the clients contexts, in order to let the clients to handle fault 
> tolernace (the scenario suggested by Iljitsch in the ams meeting). A 
> possibility for dealing with this would be that probes that verify that 
> the current path is working do not contain context specific information 
> (as oposed to probes used for exploring the alternative paths, that in 
> order to be used for flooding prevention must carry context specific 
> information). I know that this would imply a change in the failure 
> detection protocol in Jari's draft, but maybe it would be worth to 
> consider this issue.
>
> Regards, marcelo
>
> PS: agree that interaction with context confusion mechanisms needs to be 
> be clearly understood in order to get the full picture about this R1bis 
> thing