Re: Pseudorandom Flow Labels

[Brian E Carpenter <[e-mail]>]

Setting the flow label by calling random() cannot be a stateless
method - it would mean storing the value for use on future packets
of the same flow. We need a stateless method.

    Brian

On 2011-04-07 10:16, Fernando Gont wrote:
> Hi, Shane,
> 
> On 06/04/2011 06:44 p.m., Shane Amante wrote:
> 
>>> * We want Flow Labels that unpredictable by off-path attackers
>>> (history has taught us that this is a good proactive measure) * We
>>> want an algorithm for generating FL that produces FLs that do not 
>>> repeat with a high frequency (i.e., they are distributed normally)
>> I like your (attempt at) a more precise definition that Thomas has
>> been asking for.
>>
>> I would think another desirable property of (host-generated?)
>> flow-labels might be that, by default, they strive to preserve
>> privacy of the transmitter.  IOW, flow-labels cannot be used to track
>> individuals (over time), because they are traceable back to a
>> particular implementation or, worse, a specific device.
> 
> While I agree with this, this seems to be like a requirement for
> psedu-randomness.
> 
> That is, if all hosts in my local network set the FL with random(), and
> only my hosts implements the hash based scheme in e.g.
> draft-gont-flowlabel-security, then you might argue that the FL could be
> exploited to track me (or well, actually identify my host in my local
> network, even if I'm using privacy addresses).
> 
> But this is already the case for virtually all protocol parameters on
> which there's some room for choice. (From TCP's initial window to IPv6
> Hop Limit value, etc.)
> 
> Thanks,