CVE-2021-29544: Denial of Service in TensorFlow

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 is widely used in various domains, including research, industry, and academia. It allows users to build and train machine learning models using high-level APIs and provides support for distributed computing and deployment on different platforms. However, a vulnerability has been discovered in TensorFlow that can lead to a denial of service attack.

Vulnerability Detail

The vulnerability, identified as CVE-2021-29544, exists in the tf.raw_ops.QuantizeAndDequantizeV4Grad function of TensorFlow. An attacker can exploit this vulnerability by triggering a CHECK-fail in the function, which occurs due to the lack of validation for the rank of certain input tensors. As a result, these tensors are passed as they are to another function, QuantizeAndDequantizePerChannelGradientImpl, which requires the rank to be 1. This mismatch in rank triggers a CHECK failure, leading to a denial of service condition. The vulnerability affects versions of TensorFlow up to 2.4.3.

pip install --upgrade "tensorflow==2.21.0"

Workaround

A possible workaround for this vulnerability is to add explicit rank validation for the input tensors in the QuantizeAndDequantizeV4Grad function. This can be done by modifying the implementation code to check the rank of the input_* tensors before passing them to QuantizeAndDequantizePerChannelGradientImpl. Here is an example of how the code can be modified:

# Original code
if vec<T>(input_shape.dim_size(0)) != 1:
  CHECK(false) << "Rank of input_* tensors must be 1.";

# Modified code
if input_shape.dims() != 1:
  CHECK(false) << "Rank of input_* tensors must be 1.";

Conclusion

In conclusion, a denial of service vulnerability has been identified in TensorFlow versions up to 2.4.3. This vulnerability can be exploited by an attacker to disrupt the service or render machine learning models unavailable. Users are advised to update their TensorFlow installations to version 2.5.0 once it is released. In the meantime, a workaround is available to mitigate the risk. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow →

FAQ