CVE-2021-29548

TensorFlow QuantizedBatchNormWithGlobalNormalization runtime division by zero vulnerability

• CVSS 5.5
• CWE-369 Divide By Zero
• Design Defects
• Remote

TensorFlow is an end-to-end open source platform for machine learning. An attacker can cause a runtime division by zero error and denial of service in `tf.raw_ops.QuantizedBatchNormWithGlobalNormalization`. This is because the implementation(https://github.com/tensorflow/tensorflow/blob/55a97caa9e99c7f37a0bbbeb414dc55553d3ae7f/tensorflow/core/kernels/quantized_batch_norm_op.cc) does not validate all constraints specified in the op's contract(https://www.tensorflow.org/api_docs/python/tf/raw_ops/QuantizedBatchNormWithGlobalNormalization). The fix will be included in TensorFlow 2.5.0. We will also cherrypick this commit on TensorFlow 2.4.2, TensorFlow 2.3.3, TensorFlow 2.2.3 and TensorFlow 2.1.4, as these are also affected and still in supported range.

CVE

CVE-2021-29548

CWE

CWE-369 Divide By Zero

CVSS base score

5.5

Published

2021-05-14

OWASP

A08 Software and Data Integrity Failures

Orthogonal defect classification

Function

Code defect classification

Incorrect Functionality

Category

Design Defects

Subcategory

Inadequate Error Handling

Accessibility scope

Remote

Impact

Denial of Service (DoS)

Affected component

TensorFlow

Fixed by upgrading

Yes

Solution

Update TensorFlow to version 2.5.0 or higher.

import tensorflow as tf

input_tensor = tf.constant([1, 2, 3], dtype=tf.qint8)
scale = tf.constant([1.0], dtype=tf.float32)
offset = tf.constant([0.0], dtype=tf.float32)
mean = tf.constant([1.0], dtype=tf.float32)
variance = tf.constant([0.0], dtype=tf.float32)

result = tf.raw_ops.QuantizedBatchNormWithGlobalNormalization(
    input=input_tensor,
    scale=scale,
    offset=offset,
    mean=mean,
    variance=variance,
    epsilon=1e-5
)

print(result)

import tensorflow as tf

def quantized_batch_norm_with_global_normalization(input_tensor, scale, offset, mean, variance, epsilon=1e-5):
    if tf.reduce_any(variance <= 0):
        raise ValueError("Variance must be greater than zero to avoid division by zero.")

    return tf.raw_ops.QuantizedBatchNormWithGlobalNormalization(
        input=input_tensor,
        scale=scale,
        offset=offset,
        mean=mean,
        variance=variance,
        epsilon=epsilon
    )

try:
    input_tensor = tf.constant([1, 2, 3], dtype=tf.qint8)
    scale = tf.constant([1.0], dtype=tf.float32)
    offset = tf.constant([0.0], dtype=tf.float32)
    mean = tf.constant([1.0], dtype=tf.float32)
    variance = tf.constant([0.0], dtype=tf.float32)

    result = quantized_batch_norm_with_global_normalization(input_tensor, scale, offset, mean, variance)
except ValueError as e:
    print(e)

@misc{vaitp:cve202129548,
  title        = {{TensorFlow QuantizedBatchNormWithGlobalNormalization runtime division by zero vulnerability}},
  author       = {Bogaerts, Fr\'ed\'eric and Ivaki, Naghmeh and Fonseca, Jos\'e},
  year         = {2021},
  note         = {VAITP Python Vulnerability Dataset, entry CVE-2021-29548},
  howpublished = {\url{https://netpack.pt/vaitp/vulnerability/CVE-2021-29548/}}
}