latency injection attack mitigation

[Zenaan Harkness]

There are two primary network stream origin node "latency injection"
identification attacks, passive and active.

Each attack begins with the attacker establishing one or more network
streams through a target node to stalk, and attempting to manipulate
its io streams.


## Passive latency injection attacks

A passive attack involves stalking two nodes to ascertain whether
they are the same node, or to ascertain whether one node is in the
path of the other or otherwise effects the other.

Our garlick/pipe/iqnets net attempts to mitigate such passive attacks
with chaff file, packet switching with pre-agreed routes and
pre-agreed b/w, and recording and booting nodes which fail to deliver
(whilst we must also consider an innocent end user node who is under
attack, and how to identify and correct for this).



## Active latency injection attacks

An active latency injection attack is performed by having control of
a physical node directly in the path of a stalked node - in practice
the simplest version of this is the first physical hop for an end
user node, which is typically one of their ISP's routers.

Attacker simply suspends a stalked node's entire internet connection
for a suitable number of milliseconds (say 57ms, or whatever is large
enough given typical/ default iqnets config), and if attacker sees
a stream somewhere else under the attacker's watch, display's a
latency spike of around 57 ms, the attacker has with high probability
identified the target they are stalking.

If attacker's probability of ID of stalked node is only say 30% (due
to network jitter or whatever), simply repeat the latency injection a
couple more times, to increase to certainty.

Such low magnitude latency injections (only 57ms, or even less), can
be readily used over dozens or hundreds of targets, to bisect a large
set of stalked nodes, to determine specific nodes of interest.

The ONLY (caution, absolutism in operation here!) obvious mitigation
to such active stalking (and still only for some cases) is:

  - create/use dark links, not just govnet links (you need at least
    one "first hop" link you can trust, which also means hardware you
    can trust)
    - "link" as in physical connection, i.e. "immediate/ first/ next
      hop from me"

  - maintain govenet (ISP) routes for others, not immediately for
    yourself

  - only use govnet for yourself as a fall back, and if you don't
    care about being actively stalked

  - monitor all first-hop/friend nodes, including dark links, for
    signs of latency injection attack/victim in operation



## Other mitigations:

  - stream splitting, possibly to a high degree ("many micro
    streams") and separate routes for each micro stream

  - widely and finely dispersed content addressed content
    distributiona and retrieval, thus:
    - eliminating fat streams
    - eliminating single points of failure
    - thereby reducing possible identification of that nodes which
      fail, and nodes that up/download

  - server-less web "applications"
    - content addressed DHT based forums, web sites etc

    - requires effective (incentivized) distributed cacheing models

    - for rare/old content, "cache of last resort" is an hopefully
      solvable problem - in the case of clear net content producers,
      presumably such content producers have an interest in making
      their content available, except when they don't or when they
      modify older content, mitigated with
      - content addressed content (git style ob stores)
      - distributed cacheing
      - end users who take an interest in archiveing old content and
        holding content producers/ reporters to account for what they
        did in fact say

  - for phone calls, which require real time / low latency, fixed
    rate streams, you're going to need dark links, or you are going
    to be exposed to active latency injection attacks (although, in
    the middle of a phone call, if the overall network params are
    right, latency injections should be audibly identifiable at least)