Auditing Cloud Administrators Using Information Flow Tracking Afshar David ACM Scalable Trusted Computing Workshop Raleigh, North Carolina October 2012
Cloud Computing Is Not Trusted 2
Admins at Infrastructure-as-a-Service (IaaS) Providers 3 VMM User VM Management Stack
Restricting Admins Is Not the Solution 4 VMM User VM Management Stack I cannot: Install commodity applications I want. Change system configurations. Write my own scripts in Perl or Python. Monitor resource usages. See the logs for troubleshooting.
H-one Provides Logs for Auditing 5 We propose auditing. H-one performs no access control. Auditing has been used in other domains. Auditing deters misbehaving. Helps to assign liability of events. No unnecessary restrictions for admins. Auditing has 2 stages: Generating logs Inspecting the logs
What are the logging challenges in H-one? 6 GOALS Complete Effici ent Privacy Preserving Data: From VMs to Admins From Admins to VMs Minimal Storage Costs Logs related to different customers should be separate. To achieve these goals H-one uses Information Flow Tracking
Example 1: Benign Admin Tasks: VM Backup 7 VMM User VMManagement Stack Disk Kernel User Disk Image H-one Module
Example 2: Benign Admin Tasks: Backup for 2 VMs 8 VMM User VM 2User VM 1Management Stack Disk Kernel Disk 1 Disk 2 H-one Module
Example 3: Adversarial Admin 9 VMM User VMManagement Stack Disk Kernel User Disk Image H-one Module
Using Information Flow Tracking 10 GOALS Complete Effici ent Privacy Preserving H-one tracks any data flow inside management stack. By following information flows, just the required data at appropriate points get logged. Tracking flows lets us know leaked data belong to which user.
We use Xen hypervisor for our prototype. We use a customized LSM module for labeling and tracking information flows protecting the integrity of the H-one logging system We use the concept of the “exporter” processes similar to DStar paper for tracking networking communications. N. Zeldovich, S. Boyd-Wickizer, and D. Mazieres, “Securing Distributed Systems with Information Flow Control,” in Proceedings of the USENIX Symposium on Networked Systems Design and Implementation (NSDI), 2008, pp. 293–308. Implementation 11
Information Flow Tracking reduces the logging cost. Our filtering daemon can further reduce the log size in specific scenarios based on the context. Filtering daemon understands the legitimate flows of information and filters the corresponding logs. Realtime Filtering of Logs 12
13 Questions ?! Discussion ?!
Label Propagation 14
15 Questions ?! Discussion ?!