Usable Access Control in Cloud Management Systems

Master's thesis of Science and Technology presentation at Aalto University

The thesis will be published at https://github.com/luxas/research

Presentation date: May 22, 2025

Transcript

1. Usable Access Control in Cloud Management Systems Lucas Käldström, Upbound @luxas.dev
2. MSc student Staff Software Engineer Contributor @luxas.dev
3. Outline Project Context: - Find gaps in and improve (usability of) Kubernetes Access Control Evaluated Tools: - Relation-based Access Control with SpiceDB - Analyzable Access Control with Cedar Policy Contributions: - Kubernetes Access Control improvements implemented with Cedar @luxas.dev
4. Production-grade Container Orchestrator Released as open source by Google in 2014, donated to Kubernetes @luxas.dev Published 2013 and 2015
5. Kubernetes Control Plane framework and control plane for workload orchestration @luxas.dev Generic Control Plane Framework Container Orchestration Project Targets Any software system Container Workloads, SDN, etc.
6. Kubernetes Control Plane framework and control plane for workload orchestration @luxas.dev Generic Control Plane Framework Project Targets Any software system
7. Kubernetes Control Plane framework and control plane for workload orchestration @luxas.dev Declarative, uniform, extensible API server Clients Clients Controlled System 1 Read desired state Controlled System 2 Write actual state Act Act Database
8. @luxas.dev Available at https://github.com/luxas/research Published in 2021 More context in my BSc thesis
9. Anatomy of an uniform and extensible API server apiVersion: gateway.networking.k8s.io/v1 kind: Gateway metadata: name: my-route labels: owner: lucas spec: gatewayClassName: test-gateway addresses: - value: 1.2.3.4 status: conditions: - type: Ready Type Metadata always encoded Uniform Object Metadata for all objects User declares the desired state System updates actual state
10. Anatomy of an uniform and extensible API server apiVersion: gateway.networking.k8s.io/v1 kind: Gateway* metadata: name: my-route labels: … Type Metadata always encoded Uniform Object Metadata for all objects list, watch, create, deletecollection: /apis/<group>/<version>/<resource>** get, update, patch, delete: /apis/<group>/<version>/<resource>/<name>** *resource != kind, but almost, simplified here **Plus namespace scoping if namespaced @luxas.dev
11. Kubernetes RBAC Subjects RoleBinding admin RoleBinding deploy-workloads *:* OR Role Bindings @luxas.dev
12. Kubernetes RBAC Subjects RoleBinding admin RoleBinding deploy-workloads Role admin Role deploy-workloads 1:1 *:* OR Role Bindings Roles @luxas.dev
13. Kubernetes RBAC Subjects RoleBinding admin RoleBinding deploy-workloads Role admin Role deploy-workloads 1:1 *:* Rule 1 Rule 2 OR OR (Rule 1 and Role deploy-workloads omitted for brevity) 1:* Role Bindings Roles Role Rules @luxas.dev
14. Kubernetes RBAC Subjects RoleBinding admin RoleBinding deploy-workloads Role admin Role deploy-workloads 1:1 *:* Rule 1 Rule 2 OR OR (Rule 1 and Role deploy-workloads omitted for brevity) 1:* apiGroup resource namespace name AND Role Bindings Roles Role Rules Predicates @luxas.dev
15. Kubernetes RBAC Subjects RoleBinding admin RoleBinding deploy-workloads Role admin Role deploy-workloads 1:1 *:* Rule 1 Rule 2 OR OR (Rule 1 and Role deploy-workloads omitted for brevity) 1:* apiGroup Obj 1 resource namespace name Obj 2 Obj 3 AND Role Bindings Roles Role Rules Predicates Objects @luxas.dev
16. Kubernetes RBAC Subjects RoleBinding admin RoleBinding deploy-workloads Role admin Role deploy-workloads 1:1 *:* Rule 1 Rule 2 OR OR (Rule 1 and Role deploy-workloads omitted for brevity) 1:* apiGroup Obj 1 resource namespace name Obj 2 Obj 3 AND Role Bindings Roles Role Rules Predicates Objects “RBAC” “ABAC” @luxas.dev
17. Kubernetes RBAC mapping into ReBAC In recent years, it has become increasingly popular to turn the access control problem into a graph reachability problem (“Relation-Based Access Control”) @luxas.dev Published 2013 and 2019
18. @luxas @luxas API Server Slide from my talk at Disobey 18 User Authenticator ReBAC Server Flow Control Authorizer Admission Privilege Escalation Prevention Authorization State Update List Filter ReBAC Database Storage Audit Log Policy Engine Check ListResources WriteRelations
19. Kubernetes RBAC mapping into ReBAC However, due to Kubernetes RBAC’s (ironically) heavy ABAC use, a graph-based mapping wasn’t expressive enough. The lack of expressiveness in the graph resulted in the actual access control logic moving to the (Turing-complete) reconciler from user intent to graph representation. In other words, the result was not very analyzable. ⇒ I pivoted from graph- to logic-based methods @luxas.dev
20. Usability issues with Kubernetes RBAC
21. Kubernetes Access Control Request Flow @luxas.dev
22. Kubernetes Access Control Request Flow
23. Kubernetes API Server structure Authenticators RequestInfo UserInfo 401 Webhook OIDC CA Slide from my and Jimmy’s talk at KubeCon Salt Lake City
24. Kubernetes API Server structure Authenticators Authorizers RequestInfo UserInfo 401 403 RequestInfo UserInfo Body Webhook OIDC CA Webhook RBAC Slide from my and Jimmy’s talk at KubeCon Salt Lake City
25. Kubernetes API Server structure Authenticators Authorizers RequestInfo UserInfo Mutating/Validating Admission Controllers* Storage 401 403 40X RequestInfo UserInfo Body Body Webhook OIDC CA Webhook RBAC CEL Webhook 200 * Admission only for CREATE/UPDATE/PATCH/DELETE Slide from my and Jimmy’s talk at KubeCon Salt Lake City
26. Authorization and Admission Control are separate The request payload is not available in the authorization stage. Thus, we need to “over-grant” in authorization, and remove the extra permissions in the admission stage. create, update, delete gateways — object=* oldobject=* Authorization .class != test-gateway .class == test-gateway Admission Control Amount of permissions for lucas Desired permissions
27. Authorization and Admission Control are separate There is a ValidationAdmissionPolicy API in Kubernetes, which allows specifying the deny rule using Common Expression Language (CEL). @luxas.dev create, update, delete gateways — object=* oldobject=* Kubernetes RBAC .class != test-gateway .class == test-gateway CEL Rule Amount of permissions for lucas Desired permissions
28. Authorization and Admission Control are separate RBAC Role Allow in authorization RBAC Role Binding Allow in authorization CEL Policy Deny in admission create, update, delete gateways — object=* oldobject=* .class != test-gateway .class == test-gateway Kubernetes RBAC CEL Rule Amount of permissions for lucas Desired permissions
29. Admission Control is not available for reads ❌ Admission only applies to writes, so reads are still unconstrained to all gateway classes. ✅ There is an experimental feature to include label and field selectors 🟡 Selectors won’t become accessible in Kubernetes RBAC, however. create, update, delete gateways — object=* oldobject=* .class != test-gateway .class == test-gateway Kubernetes RBAC CEL Rule Amount of permissions for lucas Desired permissions
30. Research Question 1: How to unify Kubernetes authorization and admission? @luxas.dev
31. Desired properties of an Authorization Engine @luxas.dev
32. Expressive
33. Fast Expressive
34. Fast Safe Expressive
35. Fast Safe Analyzable Expressive
36. Fast Safe Analyzable Expressive Correct
37. Published 2024 @luxas.dev
38. Open Source Authorization Engine @luxas.dev
39. Open Source Authorization Engine @luxas.dev Aims to be expressive, fast, safe, and analyzable
40. Maintains a decidable encoding into Satisfiability Modulo Theories* Open Source Authorization Engine @luxas.dev Aims to be expressive, fast, safe, and analyzable *Means in practice that the language does not have loops or ordered lists
41. Maintains a decidable encoding into Satisfiability Modulo Theories* Open Source Authorization Engine @luxas.dev Aims to be expressive, fast, safe, and analyzable *Means in practice that the language does not have loops or ordered lists Supports RBAC, ReBAC and ABAC paradigms
42. Maintains a decidable encoding into Satisfiability Modulo Theories* Open Source Authorization Engine @luxas.dev Aims to be expressive, fast, safe, and analyzable *Means in practice that the language does not have loops or ordered lists AWS is donating Cedar to the CNCF Supports RBAC, ReBAC and ABAC paradigms
43. 1. Developer Experience is key @luxas.dev
44. 2. Analyze: Which policy is larger? old @luxas.dev
45. 2. Analyze: Which policy is larger? old new @luxas.dev
46. 3. Formal Verification Image Source: https://aws.amazon.com/blogs/opensource/lean-into-verified-software-development/ @luxas.dev
47. Research Question 2: Can Kubernetes access control be mapped into SMT? @luxas.dev
48. MSc thesis project: Integrate Kubernetes and Cedar 1. Improve policy authoring usability with typed schema and autocompletion 2. Unify policy authoring for both authorization and admission 3. Unify policy authoring targeting label/field selectors for reads and writes 4. Find an expression language that allow policy querying and comparisons 5. Solve hard authorization problems once, re-use for various user experiences @luxas.dev
49. 1. Improve policy authoring usability with typed schema @luxas.dev
50. Kubernetes API Server /openapi/v3/<group> API Discovery Document /apis/<group>/<version> 1. Improve policy authoring usability with typed schema Project Schema IDE Dev loop @luxas.dev
51. 2. Unify policy authoring for both authorization and admission Previous example shown in the project’s proposed syntax. Only one policy object is needed, not three like before. @luxas.dev
52. 2. Unify policy authoring for both authorization and admission Authorization RequestInfo UserInfo Kubernetes Path for write request Project Partial Evaluation Yes, No, Maybe 403 Policies Webhook Authentication @luxas.dev
53. 2. Unify policy authoring for both authorization and admission Authorization RequestInfo UserInfo Admission Control RequestInfo UserInfo Body Body Kubernetes Path for write request Project Partial Evaluation Yes, No, Maybe Full Evaluation Yes, No 403 403 Policies Webhooks Authentication Storage @luxas.dev
54. 3. Unify policy authoring targeting selectors for reads and writes The last example, but for any action, including reads. Predicates targeting resource.stored determine if a concrete object is allowed to be read from storage. @luxas.dev
55. 3. Unify policy authoring targeting selectors for reads and writes Authorization Kubernetes Path for read request Project Full Evaluation Yes, No 403 Authentication Storage Example selectors: “Label owner value in (‘alice’, ‘bob’)” “Field .spec.gatewayClassName != ‘production’” @luxas.dev Policies RequestInfo UserInfo Selectors
56. 3. Unify policy authoring targeting selectors for reads and writes Authorization RequestInfo UserInfo Selectors Kubernetes Path for read request Project 1. Partial Evaluation => Yes, No, Maybe 2. If Maybe, turn Selectors and Residual into SMT => Yield Yes or No 403 Policies Authentication Storage Authorize IFF: ∀o : objectSelected(o) ⇒ isAuthorized(o) ≡ ∃o : objectSelected(o) ∧ ¬isAuthorized(o) = UNSAT @luxas.dev
57. 4. Find an expression language that allow policy analysis Ask “what resources can principal P perform what actions on?” Ask “what principals can perform what actions on resource R?”
58. 4. Find an expression language that allow policy analysis Ask “what resources can principal P perform what actions on?” Ask “what principals can perform what actions on resource R?” Check for logical inconsistencies in a policy set Check for equality Prevent privilege escalation ← No effect ← Allow shadows allow ← Deny shadows allow Allow policy Deny policy
59. 4. Find an expression language that allow policy analysis Queries achievable through Partial Evaluation @luxas.dev
60. 4. Find an expression language that allow policy analysis Queries achievable through Partial Evaluation If the policies are compiled into equivalent SMT formulas, then: Equality if: policy1 ⊨ policy2 ∧ policy2 ⊨ policy1 Policy 2 > Policy 1 if: policy1 ⊨ policy2 ∧ ¬(policy2 ⊨ policy1) Policy change not an escalation if: current_privileges ⊨ new_privileges @luxas.dev
61. 5. Write backend once, use for multiple “frontends” Kubernetes CEL (portion w/o loops) Kubernetes RBAC New Selector-based Authorization paradigm? New Multi-cluster Policies? Project SMT Solvers @luxas.dev Policies Engine
62. Results ✅ There seems to be a path forward for all five goals & two research questions
63. Results ✅ There seems to be a path forward for all five goals & two research questions ✅ I already made two contributions to Cedar
64. Results ✅ There seems to be a path forward for all five goals & two research questions ✅ I already made two contributions to Cedar ✅ Potential interest to contribute some of these features to Kubernetes
65. Results ✅ There seems to be a path forward for all five goals & two research questions ✅ I already made two contributions to Cedar ✅ Potential interest to contribute some of these features to Kubernetes 🟡 More work to be done in Cedar to get production-ready partial evaluation & SMT compilation
66. Results ✅ There seems to be a path forward for all five goals & two research questions ✅ I already made two contributions to Cedar ✅ Potential interest to contribute some of these features to Kubernetes 🟡 More work to be done in Cedar to get production-ready partial evaluation & SMT compilation 🟡 Need to keep getting Kubernetes user feedback
67. Results ✅ There seems to be a path forward for all five goals & two research questions ✅ I already made two contributions to Cedar ✅ Potential interest to contribute some of these features to Kubernetes 🟡 More work to be done in Cedar to get production-ready partial evaluation & SMT compilation 🟡 Need to keep getting Kubernetes user feedback 🟡 Time will tell how useful the analysis properties are in practice
68. Results ✅ There seems to be a path forward for all five goals & two research questions ✅ I already made two contributions to Cedar ✅ Potential interest to contribute some of these features to Kubernetes 🟡 More work to be done in Cedar to get production-ready partial evaluation & SMT compilation 🟡 Need to keep getting Kubernetes user feedback 🟡 Time will tell how useful the analysis properties are in practice ❌ Will always exist cases where analyzable access control isn’t expressive enough
69. Thanks! Please give feedback! Email: lucas@luxaslabs.com Bluesky: @luxas.dev LinkedIn: luxas CNCF/Kubernetes Slack: luxas Credits: Icons by Flaticon