CVE-2022-23585: Memory Leak in TensorFlow when Decoding PNG Images

Background

TensorFlow is an open-source machine learning framework widely used for developing and training machine learning models. It provides a comprehensive set of tools and libraries for building and deploying machine learning applications. TensorFlow supports various image formats, including PNG, for image processing tasks. However, a vulnerability has been identified in TensorFlow that can lead to a memory leak when decoding PNG images.

Vulnerability Detail

The vulnerability arises during the decoding process of PNG images in TensorFlow. When an invalid image is encountered, TensorFlow can produce a memory leak due to improper handling of allocated buffers. Specifically, after calling the png::CommonInitDecode(..., &decode) function, the decode value contains allocated buffers that can only be freed by calling png::CommonFreeDecode(&decode). However, certain error cases in the function implementation invoke the OP_REQUIRES macro, which abruptly terminates the execution of the function without allowing for the memory free operation. This results in a memory leak. The vulnerability can have significant ramifications, including potential denial-of-service (DoS) attacks and resource exhaustion.

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

Workaround

As a temporary workaround, users can manually free the allocated buffers by calling png::CommonFreeDecode(&decode) in the error handling code paths. This ensures that the memory is properly released even in cases where the function execution is terminated prematurely. Here is an example of how the workaround can be implemented in Python:

import tensorflow as tf

# Decoding PNG image
image_data = ...  # PNG image data

decode = tf.png.decode_png(image_data)

try:
    # Perform image processing
    ...
except Exception as e:
    # Handle error and free allocated buffers
    tf.png.CommonFreeDecode(decode)
    raise e

Conclusion

In conclusion, TensorFlow is susceptible to a memory leak vulnerability when decoding PNG images. This vulnerability can lead to resource exhaustion and potential denial-of-service attacks. The TensorFlow development team has addressed the issue in the upcoming release of TensorFlow 2.8.0 and has backported the fix to earlier versions. Users are strongly advised to upgrade their TensorFlow installations to the latest patched version to mitigate the vulnerability. For further support and assistance with vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow-cpu →

FAQ