1. © Ravi Sandhu www.list.gmu.edu The Secure Information Sharing Problem and Solution Approaches Ravi Sandhu Professor of Information Security and Assurance George Mason University www.list.gmu.edu

2. © Ravi Sandhu www.list.gmu.edu 2 Three Themes Secure Information Sharing (IS) Share but Protect Mother of all Security Problems Trusted Computing (TC) Policy-Enforcement- Implementation Layers (PEI) & Usage Control Models (UCON) Work in Progress

3. © Ravi Sandhu www.list.gmu.edu 3 Basic premise Software alone cannot provide an adequate foundation for trust Old style Trusted Computing (1970 – 1990s) Multics system Capability-based computers –Intel 432 vis a vis Intel 8086 Trust with security kernel based on military-style security labels –Orange Book: eliminate trust from applications Whats new (2000s) Hardware and cryptography-based root of trust –Trust within a platform –Trust across platforms Rely on trust in applications –No Trojan Horses or –Mitigate Trojan Horses and bugs by legal and reputational recourse What is Trusted Computing (TC)? Massive paradigm shift Prevent information leakage by binding information to Trusted Viewers on the client How best to leverage this technology?

4. © Ravi Sandhu www.list.gmu.edu 4 What is Information Sharing? The mother of all security problems Share but protect Requires controls on the client Server-side controls do not scale to high assurance Bigger than (but includes) Retail DRM (Digital Rights Management) Enterprise DRM

5. © Ravi Sandhu www.list.gmu.edu 5 What is Information Sharing? Strength of Enforcement Content type and valueWeakMediumStrong Sensitive and proprietaryPassword-protected documentsSoftware-based client controls for documents Hardware based trusted viewers, displays and inputs Revenue drivenIEEE, ACM digital libraries protected by server access controls DRM-enabled media players such as for digital music and eBooks Dongle-based copy protection, hardware based trusted viewers, displays and inputs Sensitive and revenueAnalyst and business reports protected by server access controls Software-based client controls for documents Hardware based trusted viewers, displays and inputs Roshan Thomas and Ravi Sandhu, Towards a Multi-Dimensional Characterization of Dissemination Control. POLICY04.

6. FunctionalityStrength of enforcement SimpleComplexWeak/MediumStrong Legally enforceable versus system enforced rights. Reliance on legal enforcement; Limited system enforced controls. Strong system- enforceable rights, revocable rights. Dissemination chains and flexibility. Limited to one-step disseminations. Flexible, multi-step, and multi-point.Mostly legal enforcement;System enforceable controls. Object types supported.Simple, read-only and single- version objects. Support for complex, multi-version objects. Support for object sensitivity/confidentiality. Reliance on legally enforceable rights. System supported and enforceable rights and sanitization on multiple versions. Persistence and modifiability of rights and licenses. Immutable, persistent and viral on all disseminated copies. Not viral and modifiable by recipient.Reliance on legally enforceable rights. System enforceable. Online versus offline access and persistent client-side copies No offline access and no client- side copies. Allows offline access to client-side copies. Few unprotected copies are tolerated. No unprotected copies are tolerated. Usage controlsControl of basic dissemination.Flexible, rule-based usage controls on instances. Some usage abuse allowed.No potential for usage abuse. Preservation of attribution. Recipient has legal obligation to give attribution to disseminator. System-enabled preservation and trace- back of the attribution chain back to original disseminator. Attribution can only be legally enforced. Attribution is system enforced. RevocationSimple explicit revocations.Complex policy-based revocation.No timeliness guarantees.Guaranteed to take immediate effect. Support for derived and value-added objects. Not supported.Supported.Reliance on legally enforceable rights. System enforceable rights for derived and valued-added objects. Integrity protection for disseminated objects. Out of band or non-crypto based validation. Cryptographic schemes for integrity validation. Off-line validation.High-assurance cryptographic validation. AuditAudit support for basic dissemination operations. Additional support for the audit of instance usage. Offline audit analysis.Real-time audit analysis and alerts. PaymentSimple payment schemes (if any). Multiple pricing models and payment schemes including resale. Tolerance of some revenue loss. No revenue loss; Objective is to maximize revenue. With current state of knowledge the information sharing space is too complex to characterize in a comprehensive manner Look for sweet spots that are of practical interest and where progress (and killer products) can be made Roshan Thomas and Ravi Sandhu, Towards a Multi-Dimensional Characterization of Dissemination Control. POLICY04.

7. © Ravi Sandhu www.list.gmu.edu 7 Classic Approaches to Information Sharing Discretionary Access Control (DAC), Lampson 1971 Fundamentally broken Controls access to the original but not to copies (or extracts) Mandatory Access Control (MAC), Bell-LaPadula 1971 Solves the problem for coarse-grained sharing –Thorny issues of covert channels, inference, aggregation remain but can be confronted Does not scale to fine-grained sharing –Super-exponential explosion of security labels is impractical –Fallback to DAC for fine-grained control (as per the Orange Book) is pointless Originator Control (ORCON), Graubart 1989 Propagated access control lists: let copying happen but propagate ACLs to copies (or extracts) Not very successful

8. © Ravi Sandhu www.list.gmu.edu 8 Modern Approach to Information Sharing Prevent leakage by binding information to Trusted Viewers on the client Use a mix of cryptographic and access control techniques Cryptography and Trusted Computing primitives enable encapsulation of content in a Trusted Viewer Trusted Viewer cannot see plaintext unless it has the correct keys Access control enables fine-grained control and flexible policy enforcement by the Trusted Viewer Trusted Viewer will not display plaintext (even though it can) unless policy requirements are met Enables policy flexibility and policy-mechanism separation Without use of hardware-rooted Trusted Computing assurance of client-side controls is very weak

9. © Ravi Sandhu www.list.gmu.edu 9 PEI Models Framework

10. © Ravi Sandhu www.list.gmu.edu 10 Scoping Information Sharing Problem: Objective Layer Scoping the Problem Read-only (versus read-write) Document-level controls (versus query-level control) Superdistribution (encrypt once, access wherever authorized) Support for off-line access without advance set-up (with usage limits) Scoping the Solution Two-phase enforcement –Enroll TPM (Trusted Platform Module) and LaGrande equipped client computer into a Group –Within the Group impose additional policy-based controls Required to support super-distribution Supports fine-grained controls and policy flexibility Limits instant and pre- emptive revocation

11. © Ravi Sandhu www.list.gmu.edu 11 PEI Models Framework

12. © Ravi Sandhu www.list.gmu.edu 12 Various states of a member in a group Initial state: Never been a member State I Currently a member State II Past member State III enrolldis- enroll enroll

13. © Ravi Sandhu www.list.gmu.edu 13 Access policies for State I Initial state: Never been a member State I Currently a member State II Past member State III enrolldis- enroll 1. Straight-forward. User has no access to any group documents. enroll

14. © Ravi Sandhu www.list.gmu.edu 14 Access policies for State II Initial state: Never been a member State I Currently a member State II Past member State III enrolldis- enroll 1. Access to current documents only (or) 2. Access to current documents and past documents 3. Access can be further restricted with rate and/or usage limits 4. Access can be further restricted on basis of individual user credentials enroll

15. © Ravi Sandhu www.list.gmu.edu 15 Access policies for State III Initial state: Never been a member State I Currently a member State II Past member State III enrolldis- enroll 1. Past member loses access to all documents (or) 2. can access any document created during his membership (or) 3. can access documents he accessed during membership (or) 4. can access all documents created before he left the group (this includes the ones created before his join time) 5. all subject to possible additional rate, usage and user credential restrictions enroll

16. © Ravi Sandhu www.list.gmu.edu 16 Access policies for State II.re-enroll Initial state: Never been a member State I Currently a member State II Past member State III enrolldis- enroll 1. No rejoin of past members is allowed, rejoin with new ID (or) 2. Past members rejoin the group just like any other user who has never been a member 3. The same access policies defined during his prior membership should again be enforced (or) 4. access policies could vary between membership cycles enroll

17. © Ravi Sandhu www.list.gmu.edu 17 Use-case 1 Access policies: II.1, III.1, II.re-enroll.1 Intels ViiV: A typical scenario in libraries, book-stores, cafes, etc. a.A master system could be ViiV enabled which subscribes to various kinds of channels from its content providers b.Many devices can in-turn subscribe to this master device and receive content and thus form a group c.When a device leaves the group, it loses access to all the downloaded content. Leaving the group could be determined by various mechanisms depending on context and available technology (location, network connectivity, etc.). Many universities and corporations allow access to their content as long as one is within their network. Once the user leaves the network, the user loses access to the content.

18. © Ravi Sandhu www.list.gmu.edu 18 Use-case 2 Access policies: II.1, III.2, II.re-enroll.1: Project out-sourcing: –A financial organization could recruit a software-consulting firm to provide software solutions. This forms a temporary group. –The incoming members (from the software firm) cannot access any past documents. These could be design documents that were created in collaboration with a different external organization. –When they finish the project and leave the group, they can continue to have access to the documents exchanged during their membership for future reference and to add to their profile. This is dependant on financial institutions policy.

19. © Ravi Sandhu www.list.gmu.edu 19 Use-case 3 Access policies: II.2, III.1, II.re-enroll.1: An employee in a company can access all the current documents. When he employee quits, he should lose access to all documents. DoD projects / contracts have a multi tiered structure. A member of a contracting company may be authorized to access certain set of documents only for the duration of the project – once the project is over, the contractors right to use the document is automatically voided. In a supply chain situation, there are lots of partners and suppliers who will send quotes for a given proposal. They need to have access to the proposal and related content. But once the quote/response is submitted, their membership context for that particular or group of proposals ceases and they shouldnt have access to any of the older content that they had access to.

20. © Ravi Sandhu www.list.gmu.edu 20 Use-case 4 Access policies: II.2, III.2, II.re-enroll.1: Collaborative product development: –In the case of several automobile models, there are product twins – models from the same company that resemble each other, except for the division's brand name and price tag. –In such instances, there could be either a loose collaboration (e.g. shared design team, parts ordering/manufacturing but different factories) or a tight collaboration (e.g. joint manufacturing of two different models). –In either case, the members from different parties join hands and share documents actively. –They will need access to both old documents and current documents. Even after the collaboration period, they will need access to the old documents for further refinement and production.

21. © Ravi Sandhu www.list.gmu.edu 21 Various states of an object (document) in a group Initial state: Never been a group doc State I Currently a group doc State II Past group doc State III addremove add

22. © Ravi Sandhu www.list.gmu.edu 22 Access policies for State I Initial state: Never been a group doc State I Currently a group doc State II Past group doc State III addremove 1. Straight-forward. No access to group members. add

23. © Ravi Sandhu www.list.gmu.edu 23 Access policies for State II Initial state: Never been a group doc State I Currently a group doc State II Past group doc State III addremove 1. Access allowed only to current group members 2. Access allowed to current and past group members add

24. © Ravi Sandhu www.list.gmu.edu 24 Access policies for State III Initial state: Never been a group doc State I Currently a group doc State II Past group doc State III addremove 1. No one can access 2. Any one can access 3. Past members can access add

25. © Ravi Sandhu www.list.gmu.edu 25 Access policies for State II.re-add Initial state: Never been a group doc State I Currently a group doc State II Past group doc State III addremove 1. Cannot be re-added. 2.When a document is re-added, it will be treated as a new document that is added into the group. 3.Only current members can access. 4.Past members and current members can access add

26. © Ravi Sandhu www.list.gmu.edu 26 Policy Models Idealized policy: Instant revocation Pre-emptive revocation Enforcement models will specify degree of approximation (among other details)

27. © Ravi Sandhu www.list.gmu.edu 27 Policy Models Group membership control Group-admins enroll and dis-enroll members Group-admins add/remove documents. For concreteness we assume specific group-level policies Members cannot access documents created prior to joining Past-members can access documents created during (most recent) membership Past documents cannot be accessed by anybody Documents cannot be re-added The PEI models have specific points where such policies are enforced and remain robust to changes in policy details

28. © Ravi Sandhu www.list.gmu.edu 28 Usage Control: The UCON Model unified model integrating authorization obligation conditions and incorporating continuity of decisions mutability of attributes

29. © Ravi Sandhu www.list.gmu.edu 29 UCON Policy Model Operations that we need to model: Document read by a member. Adding/removing a member to/from the group Adding/removing a document to/from the group Member attributes Member: boolean TS-join: join time TS-leave: leave time Document attributes D-Member: boolean D-TS-join: join time D-TS-leave: leave time

30. © Ravi Sandhu www.list.gmu.edu 30 Policy model: member enroll/dis-enroll Initial state: Never been a member State I Currently a member State II Past member State III enrolldis- enroll member TS-join TS-leave null True time of join null enroll False time of join time of leave dis-enroll enroll enroll, dis-enroll: authorized to Group-Admins UCON elements: Pre-Authorization, attribute predicates, attribute mutability enroll

31. © Ravi Sandhu www.list.gmu.edu 31 Policy model: document add/remove Initial state: Never been a group doc State I Currently a group doc State II Past group doc State III add remove D-member D-TS-join D-TS-leave null True time of join null add False time of join time of leave remove add, remove : authorized to Group-Admins add UCON elements: Pre-Authorization, attribute predicates, attribute mutability add

32. © Ravi Sandhu www.list.gmu.edu 32 Policy model: document read (S,O,read) Pre-authorization check member(S) null AND D-member(O) null AND TS-join(S) null AND D-TS-join(O) null AND –TS-leave(S) = null AND TS-join(S) D-TS-join(O) OR –TS-leave(S) null AND TS-join(S) D-TS-join(O) TS-leave(O) Ongoing-authorization check: terminate if D-TS-leave(O) null Details depend on details of group-level policy UCON elements: Pre-Authorization, attribute predicates, attribute mutability Ongoing-authorization

33. © Ravi Sandhu www.list.gmu.edu 33 PEI Models Framework

34. © Ravi Sandhu www.list.gmu.edu 34 Enforcement Models Design Principle Do not inject new policy Focus on trade-offs for instant and pre-emptive revocation versus off-line access Faithful Enforcement w/o Off-line Access (Faithful Model): We need continuous online touch (at start of every access and during access) Continuous on-line touch can only be approximated Usage-limited Off-line Access (Approximate Model): We need online touch periodically after some duration (at start of every access and during access) –Duration between online touches can be based on time, but time is not practical for TPM-based TC –Duration between online touches can be based on usage count, which is practical for TPM-based TC

35. © Ravi Sandhu www.list.gmu.edu 35 Enforcement Architecture 3 1 2 4 5 Group-AdminMember Joining Member Control Center (CC) 7 Faithful Model: steps 3 and 4 are coupled Approximate Model: steps 3 and 4 are de-coupled D-Member 6 Member enroll and dis-enroll (steps 1-2, 5) Document add and remove (step 6, 7) Read policy enforcement (step 3) Attribute update (step 4) Two sets of attributes Authoritative: as known to the CC Local: as known on a members computer

36. © Ravi Sandhu www.list.gmu.edu 36 UCON Enforcement Models Member attributes Member-a, Member-l: boolean TS-join-a, TS-join-l: join time TS-leave-a, TS-leave-l: leave time Document attributes D-Member-a, D-Member-l: boolean D-TS-join-a, D-TS-join-l: join time D-TS-leave-a, D-TS-leave-l: leave time Additional Member attributes for Approximate model: refresh_count: decremented on every access refresh_count_reset: refresh_count is reset to this value when it hits 0

37. © Ravi Sandhu www.list.gmu.edu 37 Faithful model highlights enroll a member: two steps Step 1: Group-admin issues enrollment token to Joining Member Step 2: Joining Member presents token to CC and receives group membership credential –Group key (symmetric key) –Local attribute values dis-enroll a member Updates authoritative attributes at CC Takes effect on local attributes at next update add a document Updates authoritative attributes at CC remove a document Updates authoritative attributes at CC Propagated to clients as DRLs (Document Revocation List)

38. © Ravi Sandhu www.list.gmu.edu 38 Faithful model highlights: (S,O,read) Pre-Obligation Local attributes of S and O are updated based on authoritative values from CC Local DRL updated from authoritative DRL at CC Pre-Condition Requires connectivity to enable updates Pre-Authorization Based on just updated local attributes of S and O and DRL Ongoing-Obligation Local attributes of S and O continuously updated based on authoritative values from CC Local DRL continuously updated from authoritative DRL at CC Ongoing-Condition Requires connectivity to enable updates Ongoing-Authorization Based on continuously updated local attributes of S and O and DRL UCON elements: Requires full power of UCON

39. © Ravi Sandhu www.list.gmu.edu 39 Approximate model highlights enroll a member: two steps Step 1: Group-admin issues enrollment token to Joining Member Step 2: Joining Member presents token to CC and receives group membership credential –Group key (symmetric key) –Local attribute values dis-enroll a member Updates authoritative attributes at CC Takes effect on local attributes at next update add a document Updates authoritative attributes at CC remove a document Updates authoritative attributes at CC Propagated to clients as DRLs (Document Revocation List) Different from Faithful model

40. © Ravi Sandhu www.list.gmu.edu 40 Approximate model highlights: (S,O,read) Pre-Obligation Local attributes of S and O are periodically updated based on authoritative values from CC Pre-Condition Requires connectivity to enable updates when required Pre-Authorization Based on just updated local attributes of S and O Ongoing-Obligation Local attributes of S and O are continuously periodically updated based on authoritative values from CC Ongoing-Condition Requires connectivity to enable updates when required Ongoing-Authorization Based on continuously periodically updated local attributes of S and O UCON elements: Requires full power of UCON

41. © Ravi Sandhu www.list.gmu.edu 41 PEI Models Framework Out of scope for this talk

42. © Ravi Sandhu www.list.gmu.edu 42 Conclusion Information sharing is an important security problem and a potential growth area Trusted computing is a good fit for solving some sweet spots in this space The PEI models framework is useful in closing the policy-implementation gap UCON is a useful framework for stating policy and enforcement models

43. © Ravi Sandhu www.list.gmu.edu 43 Q&A Secure Information Sharing (IS) Share but Protect Mother of all Security Problems Trusted Computing (TC) Policy-Enforcement- Implementation Layers (PEI) & Usage Control Models (UCON) Work in Progress