CVE-2021-29614: TensorFlow GPU - Incorrect Results and Python Interpreter Crash in tf.io.decode_raw

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. TensorFlow GPU is a specific package that enables GPU acceleration for TensorFlow. It allows users to leverage the power of GPUs to accelerate their machine learning computations, resulting in faster training and inference times.

Vulnerability Detail

The vulnerability in TensorFlow GPU resides in the implementation of tf.io.decode_raw function. When combining the fixed_length parameter with wider datatypes, the function produces incorrect results and crashes the Python interpreter. The issue stems from a bug in the padded version of the function, which is responsible for decoding raw data. The bug arises from a confusion about pointer arithmetic rules. The code incorrectly computes the width of each output element and determines the size needed for the output tensor. This leads to an out-of-bounds write, causing the interpreter to crash. While the provided reproducer results in a crash, more severe attacks can be mounted using this vulnerability, as it allows writing to periodically placed locations in memory.

pip install --upgrade "tensorflow-gpu==2.12.0"

Workaround

As a temporary workaround, users can avoid combining the fixed_length parameter with wider datatypes in the tf.io.decode_raw function. By using narrower datatypes or not specifying the fixed_length parameter, the vulnerability can be mitigated. Here is an example of a workaround that uses a narrower datatype:

import tensorflow as tf

# Instead of:
# raw_data = ...
# decoded_data = tf.io.decode_raw(raw_data, tf.float64, fixed_length=10)

# Use:
raw_data = ...
decoded_data = tf.io.decode_raw(raw_data, tf.float32, fixed_length=10)

Conclusion

In conclusion, the vulnerability in TensorFlow GPU's tf.io.decode_raw function can lead to incorrect results and crashes in the Python interpreter. It is important for users to update their TensorFlow GPU package to the latest patched version once it becomes available. In the meantime, a temporary workaround is provided to mitigate the vulnerability. For further support and assistance with vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-gpu →

FAQ