Files in this item



application/pdfUJCICH-DISSERTATION-2020.pdf (2MB)
(no description provided)PDF


Title:Securing the software-defined networking control plane by using control and data dependency techniques
Author(s):Ujcich, Benjamin E
Director of Research:Sanders, William H
Doctoral Committee Chair(s):Sanders, William H
Doctoral Committee Member(s):Bates, Adam; Gunter, Carl A; Iyer, Ravishankar K
Department / Program:Electrical & Computer Eng
Discipline:Electrical & Computer Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):software-defined networking (SDN)
network operating system
Open Network Operating System (ONOS)
data provenance
program analysis
access control
information flow control (IFC)
vulnerability discovery
exploit generation
causal analysis
information flow
data flow
control flow
computer networking
network applications
Abstract:Software-defined networking (SDN) fundamentally changes how network and security practitioners design, implement, and manage their networks. SDN decouples the decision-making about traffic forwarding (i.e., the control plane) from the traffic being forwarded (i.e., the data plane). SDN also allows for network applications, or apps, to programmatically control network forwarding behavior and policy through a logically centralized control plane orchestrated by a set of SDN controllers. As a result of logical centralization, SDN controllers act as network operating systems in the coordination of shared data plane resources and comprehensive security policy implementation. SDN can support network security through the provision of security services and the assurances of policy enforcement. However, SDN’s programmability means that a network’s security considerations are different from those of traditional networks. For instance, an adversary who manipulates the programmable control plane can leverage significant control over the data plane’s behavior. In this dissertation, we demonstrate that the security posture of SDN can be enhanced using control and data dependency techniques that track information flow and enable understanding of application composability, control and data plane decoupling, and control plane insight. We support that statement through investigation of the various ways in which an attacker can use control flow and data flow dependencies to influence the SDN control plane under different threat models. We systematically explore and evaluate the SDN security posture through a combination of runtime, pre-runtime, and post-runtime contributions in both attack development and defense designs. We begin with the development a conceptual accountability framework for SDN. We analyze the extent to which various entities within SDN are accountable to each other, what they are accountable for, mechanisms for assurance about accountability, standards by which accountability is judged, and the consequences of breaching accountability. We discover significant research gaps in SDN’s accountability that impact SDN’s security posture. In particular, the results of applying the accountability framework showed that more control plane attribution is necessary at different layers of abstraction, and that insight motivated the remaining work in this dissertation. Next, we explore the influence of apps in the SDN control plane’s secure operation. We find that existing access control protections that limit what apps can do, such as role-based access controls, prove to be insufficient for preventing malicious apps from damaging control plane operations. The reason is SDN’s reliance on shared network state. We analyze SDN’s shared state model to discover that benign apps can be tricked into acting as “confused deputies”; malicious apps can poison the state used by benign apps, and that leads the benign apps to make decisions that negatively affect the network. That violates an implicit (but unenforced) integrity policy that governs the network’s security. Because of the strong interdependencies among apps that result from SDN’s shared state model, we show that apps can be easily co-opted as “gadgets,” and that allows an attacker who minimally controls one app to make changes to the network state beyond his or her originally granted permissions. We use a data provenance approach to track the lineage of the network state objects by assigning attribution to the set of processes and agents responsible for each control plane object. We design the ProvSDN tool to track API requests from apps as they access the shared network state’s objects, and to check requests against a predefined integrity policy to ensure that low-integrity apps cannot poison high-integrity apps. ProvSDN acts as both a reference monitor and an information flow control enforcement mechanism. Motivated by the strong inter-app dependencies, we investigate whether implicit data plane dependencies affect the control plane’s secure operation too. We find that data plane hosts typically have an outsized effect on the generation of the network state in reactive-based control plane designs. We also find that SDN’s event-based design, and the apps that subscribe to events, can induce dependencies that originate in the data plane and that eventually change forwarding behaviors. That combination gives attackers that are residing on data plane hosts significant opportunities to influence control plane decisions without having to compromise the SDN controller or apps. We design the EventScope tool to automatically identify where such vulnerabilities occur. EventScope clusters apps’ event usage to decide in which cases unhandled events should be handled, statically analyzes controller and app code to understand how events affect control plane execution, and identifies valid control flow paths in which a data plane attacker can reach vulnerable code to cause unintended data plane changes. We use EventScope to discover 14 new vulnerabilities, and we develop exploits that show how such vulnerabilities could allow an attacker to bypass an intended network (i.e., data plane) access control policy. This research direction is critical for SDN security evaluation because such vulnerabilities could be induced by host-based malware campaigns. Finally, although there are classes of vulnerabilities that can be removed prior to deployment, it is inevitable that other classes of attacks will occur that cannot be accounted for ahead of time. In those cases, a network or security practitioner would need to have the right amount of after-the-fact insight to diagnose the root causes of such attacks without being inundated with too much informa- tion. Challenges remain in 1) the modeling of apps and objects, which can lead to overestimation or underestimation of causal dependencies; and 2) the omission of a data plane model that causally links control and data plane activities. We design the PicoSDN tool to mitigate causal dependency modeling challenges, to account for a data plane model through the use of the data plane topology to link activities in the provenance graph, and to account for network semantics to appropriately query and summarize the control plane’s history. We show how prior work can hinder investigations and analysis in SDN-based attacks and demonstrate how PicoSDN can track SDN control plane attacks.
Issue Date:2020-07-16
Rights Information:Copyright 2020 Benjamin E. Ujcich.
Date Available in IDEALS:2020-10-07
Date Deposited:2020-08

This item appears in the following Collection(s)

Item Statistics