CVE-2021-29595: TensorFlow - Division by Zero in DepthToSpace TFLite Operator

Background

TensorFlow is an end-to-end open source platform for machine learning. It provides a comprehensive ecosystem for building and deploying machine learning models. The DepthToSpace TFLite operator is a part of TensorFlow and is used for rearranging data in a tensor. This operator is vulnerable to a division by zero error, which can be exploited by an attacker. The vulnerability exists in the implementation of the operator, specifically in the depth_to_space.cc file. The affected versions of TensorFlow include 2.1.4, 2.2.3, 2.3.3, and 2.4.2.

Vulnerability Detail

The vulnerability in the DepthToSpace TFLite operator arises from a division by zero error. An attacker can craft a malicious model in which the params->block_size parameter is set to 0. When this model is executed, the division operation in the operator's implementation will result in a division by zero error. This can lead to various consequences, including crashes, denial of service, or even potential remote code execution. The issue has been identified and a fix will be included in TensorFlow 2.5.0. Additionally, the fix will be backported to TensorFlow 2.4.2, 2.3.3, 2.2.3, and 2.1.4 to address the vulnerability in the supported versions.

Workaround

As a temporary workaround, users can avoid the vulnerability by ensuring that the params->block_size parameter in the DepthToSpace TFLite operator is never set to 0. This can be achieved by validating and sanitizing the input values before executing the model. Here is an example code snippet that demonstrates the workaround:

import tensorflow as tf

# Validate and sanitize the block_size parameter
block_size = get_block_size()
if block_size == 0:
    block_size = 1

# Create and execute the model
model = create_model()
output = model.predict(block_size=block_size)

Conclusion

In conclusion, the DepthToSpace TFLite operator in TensorFlow is vulnerable to a division by zero error. This can be exploited by an attacker by crafting a malicious model. The issue has been identified and a fix will be included in TensorFlow 2.5.0. Users of the affected versions can expect the fix to be backported and included in the respective patch releases. It is recommended to update TensorFlow to the latest version once the fix is available. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow →

FAQ