CVE-2021-37680: Division by zero in TFLite

Sponsored Habilume Habit tracker with a companion that grows with you Complete habits, earn Lume Points, and watch your Lume grow as you stay consistent. App Store Google Play

Background

The TensorFlow Lite (TFLite) package is an open-source deep learning framework developed by TensorFlow. It is designed to run machine learning models on resource-constrained devices such as mobile phones, embedded systems, and microcontrollers. TFLite provides a lightweight and efficient solution for deploying neural networks in edge devices. However, a vulnerability has been discovered in TFLite that can lead to a division by zero error.

Vulnerability Detail

The vulnerability in TFLite arises from a division operation in the implementation of fully connected layers. Specifically, the vulnerable code can be found in the fully_connected.cc file. The division operation is performed to calculate the batch size as input_size / filter->dims->data[1]. An attacker can exploit this vulnerability by crafting a malicious model that sets filter->dims->data[1] to zero. As a result, a division by zero error occurs during the execution of the model. This vulnerability can potentially lead to a crash or undefined behavior of the TFLite runtime.

pip install --upgrade "tensorflow==2.21.0"

Workaround

As a temporary workaround, users can manually check the value of filter->dims->data[1] before executing any TFLite models. If the value is zero, the model should not be executed to avoid the division by zero error. Here is an example code snippet that demonstrates the workaround:

import tensorflow as tf

# Load and preprocess the TFLite model
model = tf.lite.Interpreter('model.tflite')

# Check the value of filter->dims->data[1]
if model.get_tensor(0).shape[1] == 0:
    print('Error: Division by zero detected. Model execution aborted.')
else:
    # Execute the TFLite model
    model.invoke()

Please note that this workaround should only be used until the TensorFlow installation is updated to a patched version.

Conclusion

In conclusion, the vulnerability in TFLite that allows a division by zero error has been addressed by the TensorFlow team. Users are advised to update their TensorFlow installation to one of the patched versions to mitigate the risk. Until the update is applied, a temporary workaround is available to manually check the value of filter->dims->data[1] before executing TFLite models. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow →

FAQ