CVE-2021-29604: TensorFlow Hashtable Lookup Division by Zero Vulnerability

Background

TensorFlow is an end-to-end open source platform for machine learning. It provides a comprehensive ecosystem of tools, libraries, and resources for developing and deploying machine learning models. The TensorFlow Lite (TFLite) implementation of hashtable lookup is vulnerable to a division by zero error. This vulnerability can be exploited by an attacker who crafts a model in such a way that the values tensor's first dimension is set to zero. This vulnerability affects multiple versions of TensorFlow, including 2.1.4, 2.2.3, 2.3.3, and 2.4.2.

Vulnerability Detail

The vulnerability arises due to a division by zero error in the hashtable lookup implementation of TensorFlow Lite. When the values tensor's first dimension is set to zero, a division operation is performed, leading to a runtime error. An attacker can exploit this vulnerability by crafting a malicious model that triggers the division by zero error. The impact of this vulnerability can range from denial of service to potential information disclosure or arbitrary code execution.

pip install --upgrade "tensorflow-cpu==2.21.0"

Workaround

As a temporary workaround, users can ensure that the values tensor's first dimension is always set to a non-zero value in their models. By avoiding a zero value for the first dimension, the division by zero error can be prevented. Here is an example code snippet that demonstrates the workaround:

import tensorflow as tf

# Ensure the first dimension of `values` is non-zero
values = tf.constant([[1, 2, 3], [4, 5, 6]])

# Perform hashtable lookup
result = tf.lookup.experimental.DenseHashTable().lookup(values)

By setting values to a non-zero tensor, the vulnerability can be mitigated.

Conclusion

The division by zero vulnerability in TensorFlow's hashtable lookup implementation can be exploited by crafting a malicious model. To mitigate this vulnerability, users should update their TensorFlow installations to the patched versions once they are available. In the meantime, a temporary workaround is to ensure that the values tensor's first dimension is always set to a non-zero value. For further support on vulnerability remediation, contact DevNack.com.

See every vulnerability recorded for tensorflow-cpu →

FAQ