GSMem: Data Exfiltration from Air-Gapped Computers over GSM Frequencies

Mordechai Guri, Assaf Kachlon, Ofer Hasson, Gabi Kedma, Yisroel Mirsky, Yuval Elovici

TL;DR

This paper introduces GSMem, a novel method for exfiltrating data from air-gapped computers by modulating electromagnetic signals on GSM frequencies. It leverages standard computer components to transmit data and a compromised mobile phone's baseband firmware to receive it, demonstrating a covert channel that bypasses traditional security measures.

What Problem Does This Paper Address?

Air-gapped networks are designed to be physically and logically isolated from external networks to protect sensitive data. However, recent research has shown that even these highly secured systems are not entirely immune to attacks. A significant challenge in compromising air-gapped systems lies in exfiltrating data once an attacker gains access. Traditional methods like USB drives are often restricted, and wireless exfiltration is typically blocked. This paper addresses the gap in covert data exfiltration from air-gapped systems by exploring a method that utilizes readily available components and existing cellular infrastructure, which might be overlooked by security-conscious organizations.

Methods

The core of GSMem lies in its bifurcated malware: a transmitter on the air-gapped computer and a receiver on a nearby mobile phone.

Transmitter: * Signal Generation: The transmitter exploits the multi-channel memory architecture of modern CPUs. By initiating rapid, continuous data transfers between the CPU and RAM using specific CPU instructions (like SSE's MOVNTDQ), it generates electromagnetic radiation (EMR) at frequencies close to the RAM's I/O bus clock. This process effectively turns the computer's memory bus into a low-power antenna. * Modulation: A variant of Amplitude Shift Keying (ASK) is used. A '1' is transmitted by utilizing multiple memory channels for a set duration (T), increasing the EMR amplitude. A '0' is represented by the normal, lower-amplitude casual emissions. * Bit Framing: To aid synchronization and differentiate between '1's and '0's, data is framed with a preamble sequence ('1010') and 12-bit payloads. This framing helps the receiver dynamically adjust to varying signal strengths and potential movement. * Stealth: The transmitter is designed to be stealthy, with a small memory and CPU footprint, running as a single thread without elevated privileges and avoiding direct API calls that could be monitored.

Receiver: * Baseband Firmware Modification: The receiver component is implemented by modifying the firmware of a mobile phone's baseband processor. This is crucial because the baseband processor handles the radio frequency (RF) communication and has direct access to the necessary hardware for receiving cellular signals. The authors used the open-source OsmocomBB software on a Motorola C123 phone for this purpose. * Signal Reception and Demodulation: The receiver samples the amplitude of the carrier wave, performs noise mitigation (using a moving average filter), detects the preamble sequence for synchronization, and then demodulates the B-ASK signal to extract the binary data. * Frequency Selection: The receiver scans for the best carrier frequency (fc) within the GSM bands that offers the best carrier-to-interference ratio (CIR), assuming lower average amplitude indicates less interference.

Datasets

Not discussed in the paper. The experiments involved transmitting and receiving data between custom-built hardware and software configurations.

Key Insights and Ideas

  • Exploiting Ubiquitous Components: The most significant insight is the realization that standard computer components (CPU, memory bus) and mobile phone basebands, which are present in virtually all devices, can be repurposed for covert communication. This bypasses the need for specialized hardware that might raise suspicion.
  • "Security Through Obscurity" is Insufficient: The paper highlights that relying on the closed nature of baseband firmware, while providing a degree of obscurity, is not a foolproof security measure. Skilled attackers can still find ways to exploit these systems.
  • Memory Bus as an Antenna: The novel approach of using the memory bus as an antenna for transmitting EMR is a key technical contribution. This avoids the need for external antennas or modifications to the computer's physical structure.
  • Baseband as a Covert Receiver: Leveraging the baseband processor's inherent RF reception capabilities and its isolation from the main OS makes it an ideal, stealthy receiver for covert signals.
  • Dynamic Synchronization is Crucial: The need for dynamic synchronization (using preamble detection) is highlighted, especially for scenarios where the mobile receiver might be moving, requiring the receiver to adapt to changing signal strengths.

Key Results

  • Mobile Phone Receiver: With a modified Motorola C123 phone as the receiver, GSMem achieved effective transmission distances of 1-5.5 meters. The bit rate was limited to 1-2 bit/s, suitable for exfiltrating small amounts of data like passwords or encryption keys.
  • Dedicated Hardware Receiver: Using a Software Defined Radio (SDR) with a dedicated antenna, the effective transmission distance significantly increased to over 30 meters, with bit rates reaching 1000 bit/s and a BER of approximately 0.087%.
  • Impact of Memory Channels: The experiments showed that using more memory channels (e.g., quad-channel vs. dual-channel) and higher RAM frequencies increased the signal amplitude and improved the effective range.
  • Positioning Matters: The orientation and position of the receiver relative to the transmitter significantly impacted the Signal-to-Noise Ratio (SNR), with the front of the workstation chassis often providing the best results.
  • Data Transmission Times: The paper provides a table (Table 6) showing the time required to transmit various data types (e.g., MAC Address, MD5 hash, RSA Private Key) using both the mobile phone receiver and the SDR receiver, illustrating the dramatic speed difference.

Limitations and Future Work

  • Mobile Phone Receiver Limitations: The primary limitation acknowledged is the low bit rate and limited effective distance when using a standard mobile phone as a receiver. This is attributed to the limited processing power and access to the DSP capabilities of older, low-end phones (like the Motorola C123) that were used due to their open-source firmware.
  • Modern Smartphones: The authors suggest that implementing GSMem on modern smartphones with more advanced basebands and higher processing capabilities could yield significantly better results in terms of reception quality and transfer rates.
  • Alternative Modulation Schemes: While B-ASK was used for simplicity, the paper notes that exploring other modulation schemes could improve performance.
  • Android Receiver: The paper briefly mentions an optional implementation for receiving GSMem signals on an untampered Android smartphone by exploiting the "neighbor cell jamming" phenomenon. This method is limited to very close proximity (10cm) but demonstrates a potential receiver that doesn't require baseband modification.
  • Countermeasures: The paper discusses potential countermeasures, including "zone" approaches (prohibiting mobile phones), physical shielding, and behavioral analysis, but acknowledges the challenges in detecting baseband-level receiver activity.

Summary updated

BibTeX
@inproceedings{guri2015gsmem,
  author = {Mordechai Guri and Assaf Kachlon and Ofer Hasson and Gabi Kedma and Yisroel Mirsky and Yuval Elovici},
  title = {{GSMem: Data Exfiltration from Air-Gapped Computers over GSM Frequencies}},
  booktitle = {USENIX Security},
  year = {2015},
}
Figure