TL;DR
This paper presents the first feasibility study of an attack that can fingerprint Software-Defined Networks (SDNs) and then launch resource consumption attacks against them. By exploiting the reactive nature of SDN's control plane, this attack can degrade network performance by overwhelming the controller or filling up data plane flow tables.
What Problem Does This Paper Address?
Software-Defined Networking (SDN) promises a more flexible and programmable network infrastructure by separating the control plane (which decides how traffic flows) from the data plane (which forwards traffic). While this offers many advantages, it also introduces new security vulnerabilities. One key characteristic of SDN is that when a data plane switch encounters a packet it doesn't know how to handle (a "new flow"), it must ask the centralized control plane for instructions (a "flow rule"). This reactive process, while efficient for managing dynamic networks, can be exploited. The paper addresses the lack of understanding and practical demonstration of how this reactive mechanism can be used to attack SDN networks, specifically focusing on fingerprinting SDN deployments and then launching Denial-of-Service (DoS) attacks.
Methods
The authors propose a two-stage attack:
- Fingerprinting SDN Networks: The core idea is to exploit the difference in packet processing time between a "new flow" (requiring a lookup from the control plane) and an "existing flow" (already handled by a pre-existing rule in the data plane).
- Response Time Measurement: When a packet arrives at a switch, if there's no matching flow rule, the switch queries the controller, adding a delay. If a rule exists, the packet is processed faster. The authors formalize this as $T1 = \alpha + \beta$ (new flow, where $\alpha$ is base processing time and $\beta$ is flow setup time) and $T2 = \alpha$ (existing flow).
- SDN Scanner Tool: A novel scanning tool, "SDN scanner," is introduced. It works by sending specially crafted packets to a target network and measuring their response times. By systematically changing header fields of these packets, the scanner can collect multiple $T1$ and $T2$ values.
- Statistical Testing: To determine if the observed response times indicate an SDN network, a statistical t-test is used to check if the collected $T1$ and $T2$ samples are significantly different.
- Launching DoS Attacks: Once an SDN network is identified, the attacker leverages this knowledge to launch resource consumption attacks:
- Control Plane Resource Consumption: The attacker can flood the control plane with a large number of fake flow requests, overwhelming its capacity and potentially causing a DoS.
- Data Plane Resource Consumption: The attacker can generate numerous fake flow rules that are then stored in the data plane's flow tables. This can fill up the tables, preventing legitimate flows from being processed and leading to a DoS.
Datasets
Not discussed in the paper. The authors mention using "real-world networks" for evaluation but do not specify any publicly available datasets. They also set up a controlled test environment for DoS attack simulations.
Key Insights and Ideas
- SDN's Reactive Nature is a Double-Edged Sword: While essential for flexibility, the need for the data plane to query the control plane for new flows creates a predictable timing difference that can be exploited for fingerprinting.
- Remote Fingerprinting is Feasible: The "SDN scanner" demonstrates that an attacker doesn't need privileged access to identify SDN deployments; remote observation of response times is sufficient.
- Simple Attacks Can Be Effective: The proposed DoS attacks, which focus on resource exhaustion (controller capacity or data plane flow table space), are shown to be effective without requiring high-performance or specialized attack tools.
- Practical Implications for SDN Security: This study highlights that SDN, despite its promise, is susceptible to novel attacks that require a shift in security thinking compared to traditional networks.
Key Results
- Fingerprinting Success Rate: The SDN scanner successfully fingerprinted 24 out of 28 target networks, achieving an 85.7% fingerprinting rate.
- DoS Attack Feasibility: The authors set up a test environment with an OpenFlow switch and controller. Figure 2 (though not fully detailed in the abstract) visually suggests the time and bandwidth required for DoS attacks to consume control plane and data plane resources.
Limitations and Future Work
- Limited Real-World Deployment: The authors acknowledge that SDN is not yet widely deployed, making it challenging to collect extensive real-world data. Their evaluation relies on estimating $T1$ values by adding flow setup times from previous work.
- Simplified Test Environment: The DoS attack evaluation was conducted in a controlled test environment, which may not fully replicate the complexity of large-scale, production SDN deployments.
- Future Work: The authors plan to:
- Set up a more realistic SDN network environment for further evaluation.
- Enhance the SDN scanner tool.
- Design new defense solutions against these types of attacks.
Summary updated