Re: draft-macaulay-6man-packet-stain-00

[Joel jaeggli <[e-mail]>]

On 4/8/12 00:10 , Brian E Carpenter wrote:
> On 2012-04-07 17:17, Joel jaeggli wrote:
>> I hesitate to suggest this because I'll probably turn into a pillar of
>> salt at some point for harping on it. however...
> 
> Just don't look back (towards IPv4).
> 
>>
>> Getting new extension headers generally parsed is a high bar to get
>> over. 
> 
> I have another concern; this draft appears to state that some
> middlebox inserts an extension header into a packet on the fly:
>   "IPv6 packet staining support consists of labeling datagrams with
>    security reputation information through the addition of an IPv6
>    destination option in the packet header by packet manipulation
>    devices (PMDs) in the carrier or enterprise network."
> 
> I'm not aware of any provision in RFC 2460 allowing this, or of any
> other extension header that is inserted by a middlebox. The implications
> for MTU size and fragmentation are clear.

yes, it occured to me that an alternate approach eg encapsulation would
also necessitate the generation fragmented outer packets (which would be
legal)

>> That said within one domain of control you might be able to fit a
>> subset of the information you're looking to carry into the  20 bits
>> available in flow label. 6437 probably provides enough cover/instruction
>> to allow for that.
> 
> Given that the first paragraph of the Introduction to the draft is almost
> identical to the same paragraph in 6437, you may be on to something.
> However, it's only conformant if (a) the resulting values belong to a
> reasonably uniform distribution and are hard to predict and (b) the method
> MUST NOT be used for packets whose flow label is already non-zero.

If one reset the flow label at ingress due to security considerations
(namely in this case that the contents of the flow label cannot be
trusted since they contain information) then all flow labels you
encounter when it comes time to stain them, will be zero.

>     Brian
>