Whats Wrong with PIC

1
Whats Wrong with PIC

• Bernard Aboba
• Microsoft
• bernarda_at_microsoft.com

2
It Seemed Like Such a Good Idea

• Legacy authentication methods are currently
  popular in network access
• CHAP, EAP-MD5
• One-Time-Password (OTP)
• Generic Token Card
• SecurID
• Certificate-based authentication is supported
  within IKE
• What if
• Legacy authentication methods could be used to
  enable certificate enrollment?
• Replacement of legacy (AAA) infrastructure were
  not required in order to deploy PKI?
• Support for legacy authentication could be
  achieved without changes to IKE?
• Wouldnt life be WONDERFUL??

3
All We Need Is an Enrollment Protocol

• A protocol which can transform legacy weak
  credentials into strong credentials!
• Use legacy authentication mechanism to
  authenticate client to the server
• Server provides client with a certificate
• Client can then use certificate authentication
  within IKE
• Everyone wins
• We dont have to modify IKE
• Customers can deploy certificates, while
  continuing to leverage legacy authentication
  infrastructure.

4
Deployment Scenarios

• Short-term transition
• VPN customer desires quick transition to
  certificate infrastructure
• Enables legacy authentication enrollment
  protocol for a short period
• Users obtain long-term certificates
• Certificates used for authentication within IKE
• Legacy authentication disabled
• Long-term usage
• Customer desires benefits of certificate
  infrastructure without the pain
• Legacy authentication enabled with and without
  enrollment protocol for a considerable period
  (months, years)
• Enrollment protocol used to provide short-term
  certificates (e.g. days)
• Short-term certificates used for authentication
  within IKE

5
So Many Candidates

• XAUTH
• Enabled tunneling of legacy methods within IKE
• Typically run after group pre-shared key
  authentication
• Required modifications to IKE
• GETCERT
• Enabled tunneling of legacy methods within TLS
  w/server auth
• Certificate obtained as a result of
  authentication
• No modifications to IKE required
• PIC
• Enabled tunneling of legacy methods within ISAKMP
  w/server auth
• Certificate obtained as a result of
  authentication
• No modifications to IKE required

6
All With the Same Flaw!

• Man-in-the-middle attack
• Vulnerable protocols PIC, XAUTH, GETCERT, PANA
  over TLS, EAP TTLS, PEAP, HTTP digest over TLS
• Problems
• Server-only authentication within the tunnel
• No binding of tunnels to auth methods inside
  them
• Credential reuse
• Use of legacy methods inside and outside the
  tunnel

7
How It Works

• Attacker connects to enrollment server
• Enrollment server authenticates to attacker
• Attacker does not authenticate to enrollment
  server
• Attacker obtains tunnel keys
• Attacker masquerades as an access server
• Could be PPP, 802.1X, PPTP, even SIP server
• Assumption is that credentials are reused on
  multiple media
• Assumption is that legacy method is used within
  and outside of enrollment protocol
• Client authenticates to attacker
• Attacker tunnels authentication to enrollment
  server
• Result attacker authenticates successfully,
  obtains clients certificate, access to the
  network

8
Can It Be Fixed?

• Sure
• For auth methods deriving keys
• Compound MACs can be exchanged
• Compound keys can be derived
• For auth methods not deriving keys
• Most of what we were attempting to address CHAP,
  EAP-MD5, OTP, GTC, SECURID
• Just change the method to incorporate tunnel
  keys..
• Just change the method to derive keys
• Not really.
• Thats why they call it Legacy!

9
You Cant Get There From Here!

• The flaw is inherent in the long term usage
  scenario
• Enrollment protocol used long term
• Creates lots of time to carry out attack
• Legacy auth methods used inside and outside of
  enrollment protocol
• Creates pool of potential client victims
• Server-only authentication within the tunnel
• Enables an attacker to set up a tunnel without
  authenticating itself
• Reliance on AAA infrastructure serving many uses
• Institutionalizes credential reuse
• Reliance on legacy authentication methods
• Methods without mutual auth or key derivation
• Creates vulnerability to connection hijacking
• Some vulnerable to dictionary attack

10
Lessons Learned

• Legacy authentication methods do not provide a
  good foundation for anything other than a short
  transition
• Proliferation of CHAP as an authentication method
  in IETF protocols is not a good idea
• Authentication tunneling is not a miracle cure