Re: [Roll] Roman Danyliw's No Objection on draft-ietf-roll-turnon-rfc8138-14: (with COMMENT)

["Pascal Thubert (pthubert)" <[e-mail]>]

Hello Roman

Many thanks for your time on this review, and for your comments below : )

Let's see:
 
> ----------------------------------------------------------------------
> COMMENT:
> ----------------------------------------------------------------------
> 
> Section 5.  Typo. s/an homogenous/a homogeneous/

Fixed

> 
> Section 7.  Editorial. To be clearer on where the attacker is in the topology and
> on who is incurring the cost:
> 
> OLD
> An attacker in the middle of the network may reset the "T" flag to cause extra
> energy spending in its subDAG.
> 
> NEW
> An on-path attacker may reset the “T” flag to force additional energy
> consumption by the nodes in the subDAG.
> 

Hum, I believe we're not there yet, because worded as this it looks like the attacker is along a unicast path. This is not the case.

The "T" flag is in a message (the DIO) that is propagated in a fashion that is akin to multicast though it's not. The RPL node receives DIO messages from neighbors that are willing to be parents. If this node decides to also be a parent, it will regenerate a DIO that contains some fields unchanged, including the DODAG configuration option. 

In that process, the bad guy may change some fields it shouldn't, including the "T" flag, but it's just one of many fields in that same situation. His children and their descendant will repeat that wrong setting. Note that this is not a tree but a DODAG. So you get DIOs from not one (candidate) parent but multiple ones. A descendant may discover an inconsistency between its parents (if some descend from the attacker and others do not) and could raise an alert, but the case can be normal, e.g., during a transition. RPL does not really specify that, it is left to implementations, e.g., constrained nodes will not spend code looking for anomalies.

Bottom line, there is no path to be on. There's a wave coming, the attacker relays that wave, and modifies it. If it is in the middle of the network this has an effect - on its descendants. If the attacker is at the leaf edge, the attack as no effect since this node has no descendant. 

What do you think?

Take care,

Pascal