CVE-2021-37667: Reference binding to nullptr in unicode encoding in TensorFlow

Background

TensorFlow is an open-source machine learning framework developed by Google. It provides a comprehensive ecosystem of tools, libraries, and resources for building and deploying machine learning models. TensorFlow supports various programming languages, including Python, C++, and Java. This vulnerability affects the TensorFlow package and its related distributions, such as tensorflow-cpu and tensorflow-gpu.

Vulnerability Detail

The vulnerability, identified as CVE-2021-37667, allows an attacker to cause undefined behavior by binding a reference to a null pointer in the unicode encoding functionality of TensorFlow. Specifically, the issue lies in the tf.raw_ops.UnicodeEncode function. By exploiting this vulnerability, an attacker can trigger undefined behavior, potentially leading to a denial of service (DoS) or arbitrary code execution.

pip install --upgrade "tensorflow==2.21.0"

Workaround

As a workaround, users can avoid the vulnerable code path by not using the tf.raw_ops.UnicodeEncode function with null pointer references. Instead, they can validate the input tensors and ensure that the input_splits tensor is not empty before calling the function. Here's an example of how to perform the validation:

import tensorflow as tf
from tensorflow.python.ops import gen_string_ops

def unicode_encode_safe(input_values, input_splits, output_encoding='UTF-8', errors='ignore', replacement_char='a'):
  if input_splits.shape[0] == 0:
    raise ValueError('input_splits tensor cannot be empty')
  return gen_string_ops.unicode_encode(
    input_values=input_values,
    input_splits=input_splits,
    output_encoding=output_encoding,
    errors=errors,
    replacement_char=replacement_char
  )

unicode_encode_safe([], [], 'UTF-8', 'ignore', 'a')

By adding the validation check, users can ensure that the input_splits tensor is not empty before proceeding with the encoding operation.

Conclusion

In conclusion, the vulnerability CVE-2021-37667 in TensorFlow allows an attacker to cause undefined behavior by binding a reference to a null pointer in the unicode encoding functionality. The TensorFlow team has released patched versions to address this vulnerability, and users are advised to update their installations. By following the recommended steps, users can mitigate the risk of exploitation. For further support on vulnerability remediation, you can contact DevNack.com.

See every vulnerability recorded for tensorflow →

FAQ