Transcript slides
Architectural issues for
network-layer identifiers
Stefan Savage
Dept of Computer Science & Engineering
UC San Diego
Historical context
I
n the beginning... it was amazing the net worked
at all.
Everyone was a good actor.
Existing Internet design
Focused on universal connectivity
IP address Identifiers purely for the purpose of
connectivity
Actively trying to introduce homogeneous substrate
Dst address for routing, Src to identify destination for replies
Strictly voluntary
Unbound usage model
Security not a significant consideration in the network
layer; trust everyone equally
Cryptography expensive relative to transport
Cryptographic abstractions limited
True when IPSec designed also
What has changed?
Many users/providers don’t want homogeneity
Most src addresses today are NATed
We want to limit who can talk to whom
Huge growth in criminal activity
10s of millions of compromised machines
Sophisticated abuse of network layer
Problems
Network architecture provides “how”
Security questions are mainly about “who” and
“what”
Ad hoc, brittle mappings between two
Firewalls (address, port)
Ingress/egress filtering
DDoS filtering (ttl hack, blackholing, etc)
Key issue
Can’t count on src address being correct or global
Even if it is correct only represents existence of endpoint
Worth rethinking…
How might we design packet identifiers to
provide useful attribution?
Attribution – working definition:
The act of linking identity with action
Uses
Authentication: who wants to do that?
Situational awareness: who is doing that now?
Access control
Operational response (e.g. filtering DDoS, BotNet C&C)
Forensics: who did that in the past?
Investigatory, evidentiary
Design options
Meaning of identifier
Network attribute
Physical attribute
Location: place packet sent from (used today in payment sys)
Originator: machine packet sent from
User attribute
IP address: topological endpoint
Path: topological route (StackPI)
Capability: right to access something
Principal: evidence of individual
Scope of identifier (local, global, in-between)
Who can interpret (anyone, trusted party, hybrid)
New opportunity
Crypto has advanced significantly
Many operations are comparatively cheap now
10’s of microseconds
Line-rate hardware implementations feasible
Completely new kinds of cryptography
Groups, aggregates, append-only, IBE, Attributebased crypto, homomorphic crypto, broadcast
systems, etc
Its not just encrypt, hash and sign anymore…
New tools provide new design opportunities
Remaining agenda
Revisiting the Cryptographic toolbox (Boneh)
Local identifiers for access control (Casado)
Global identifiers for forensics (Savage)
Attribution
To whom