Security Through Publicity Eric Osterweil Dan Massey Batsukh Tsendjav Beichuan Zhang Lixia Zhang.

Security Through Publicity

Eric Osterweil

Dan Massey

Batsukh Tsendjav

Beichuan Zhang

Lixia Zhang

Motivation Security threats are a driving force in current

protocol design Public key cryptography is common tool

DNSSEC authenticates DNS messages Various BGP Security authenticates routing And many many more…..

Protocols are now established relatively mature

Deployment is essentially non-existent Everything works if only there was a PKI….

Example: DNS Security

Caching DNS Server

End-user

www.darpa.mil

www.darpa.mil = 192.5.18.195Plus (RSA) signature by the darpa.mil private key

Attacker can not forge this answer without the darpa.mil private key.

Our Problem: How Do You Get The Public Key?

Authoritative DNS Servers

Public Key Infrastructure Well known hierarchical PKIs

Ex: Web certificate authorities exist Protocols propose rigid PKIs

DNSSEC follows DNS tree Internet routing follows address registration

But This Assumes that Everyone agrees on the hierarchy Hierarchy members agree to manage keys

DNSSEC Hierarchy

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

DNSSEC Hierarchical PKI DNSSEC PKI follows the DNS tree hierarchy

Root private key signs edu public key Edu private key signs ucla.edu public key Ucla.edu private key signs cs.ucla.edu public key

But this assumes that… Hierarchy members agree to manage keys

Root, com, edu, etc not motivated to sign until lower level zones sign

Lower level zones get little benefit with PKI via root, com, edu, etc.

Everyone agrees on the hierarchy Some signatures naturally deviate from tree Ex: netsec.cs.colostate.edu signs netsec.cs.ucla.edu

Webs and Reputations Web of Trust (PGP)

Small World effect Trust is not transitive, or explicit Only addresses keys (no accountability for actions) No root of trust graph = no stipulated trusted authority Webs tend to be incomplete

Reputation Systems Generally create a high-level trust rating

Looks like a credit score Trust is subjective in large systems No central authority to set reputation rules

If there was such an authority, we would make it a CA!

Our Proposed Solution: PSKI Predicated on the Public Space and that it is

a complete data set of actions Data guaranteed to be complete, not correct!

Protocols that use the PSKI must perform all actions in the public space Forcing all data into public view can create

problems for incorrect data…. Beyond the Web of Trust:

Web of Trust does not represent actions Tracing bad behavior is not possible

What About Privacy? The PSKI is initially designed to work in

systems where privacy is not an issue We feel that the initial protocols that use the

PSKI will operate on public data sets (well known data)

Example: DNS Security No privacy concern in posting zone keys and

signatures used to authenticate zone keys.

Public Space in DNS DNSSEC defines it own semantics for storing

keys and signing records. The public space then mandates that these

actions must be made public. PSKI lists all DNSKEYs every reported to belong

to the zone All on-tree signatures and all off-tree signatures

Some PSKI semantics added for storing this PSKI enforces completeness rule

Resolvers judge trustworthiness

PSKI - Components Entities:

The public key for a zone May be conflicts

(two keys both claim to be ucla.edu) And its associated actions

Trust Graph: Graph RRSIG records that

represent cross-signed DNSKEYs

Actions: Cryptographic audit-trail

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

DNSSEC in the PSKI

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Going Forward Construct rigorous semantics Investigate issues surrounding

privacy Grouping Entities

Similar to Zones in DNSSEC Keys are 1-to-1 with Entities

BUT apps like DNSSEC zones are n-to-1

Going Forward (2) Lack of a PKI has been a

major barrier for sometime Current protocols (DNSSEC,

secure routing, etc.) are being gated

Can we store complete information?

What kind of abstraction crystallizes zones and signatures?

Thank You

Questions?

Goals Developing key infrastructures for the

Internet Goals for this key infrastructure

offer a rigorous framework must scale must impart some semantics that facilitate

trust assessment

Backup

Observations Internet-scale key

infrastructures do not exist PKIs seem too rigid for such a

scale Web of trust does not impart

enough rigor for trust New secure protocols need to

be built, and need a generic infrastructure

PSKI Details - Entities Key ID Key Inception / Expiration

PSKI Details - Trust Graph Entities Entity cross signatures Lapses of Entity registration

An Entity is allowed to expire, then renewed later

Rollover information

PSKI Details - Actions Lookup-key Entity Action Type Inception / Expiration Target of Action

PSKI Details - Entities’ Actions Entities relate to their actions with meta-

data: How often an Entity has signed for data How many active/unexpired Links to actions Current conflicts (with other Entity

signatures) Total number of conflicts for this Entity