CVE-2021-29581: Segfault in CTCBeamSearchDecoder in TensorFlow

Background

TensorFlow is an open-source machine learning framework widely used for various applications. It provides a comprehensive ecosystem of tools, libraries, and resources for developing and deploying machine learning models. The CTCBeamSearchDecoder is a component in TensorFlow that performs beam search decoding for Connectionist Temporal Classification (CTC) models. It is used for tasks such as speech recognition and handwriting recognition.

Vulnerability Detail

The CTCBeamSearchDecoder in TensorFlow is vulnerable to a segmentation fault caused by a lack of validation in the tf.raw_ops.CTCBeamSearchDecoder function. This vulnerability allows an attacker to trigger a denial of service (DoS) attack by providing an empty input tensor, which leads to the function attempting to read data from a null buffer. This can result in a segmentation fault and crash the application. The impact of this vulnerability is limited to DoS attacks and does not allow for arbitrary code execution or data exfiltration.

pip install --upgrade "tensorflow==2.21.0"

Workaround

As a workaround, users can add input validation to their code before calling the CTCBeamSearchDecoder function. This can be done by checking if the input tensor is empty and handling this case appropriately to avoid triggering the vulnerability. Here is an example of how to add input validation:

import tensorflow as tf

inputs = tf.constant([], shape=[18, 8, 0], dtype=tf.float32)
sequence_length = tf.constant([11, -43, -92, 11, -89, -83, -35, -100], shape=[8], dtype=tf.int32)
beam_width = 10
top_paths = 3
merge_repeated = True

if inputs.shape[2] == 0:
    # Handle empty input tensor
    print('Input tensor is empty')
else:
    tf.raw_ops.CTCBeamSearchDecoder(
        inputs=inputs, sequence_length=sequence_length, beam_width=beam_width,
        top_paths=top_paths, merge_repeated=merge_repeated)

Conclusion

The vulnerability in the CTCBeamSearchDecoder component of TensorFlow allows for a denial of service attack through a segmentation fault. It has been patched in TensorFlow versions 2.5.0, 2.4.2, 2.3.3, 2.2.3, and 2.1.4. Users are advised to upgrade to one of these patched versions to mitigate the risk. If immediate upgrading is not possible, adding input validation to the code can serve as a temporary workaround. For further support on vulnerability remediation, please contact DevNack.com.

See every vulnerability recorded for tensorflow →

FAQ