Skip to content

CHECK-fail in LSTM with zero-length input

Low
mihaimaruseac published GHSA-m648-33qf-v3gp Dec 9, 2020

Package

tensorflow, tensorflow-cpu, tensorflow-gpu (tensorflow)

Affected versions

< 2.4.0

Patched versions

1.15.5, 2.0.4, 2.1.3, 2.2.2, 2.3.2, 2.4.0

Description

Impact

Running an LSTM/GRU model where the LSTM/GRU layer receives an input with zero-length results in a CHECK failure when using the CUDA backend.

This can result in a query-of-death vulnerability, via denial of service, if users can control the input to the layer.

Patches

We have patched the issue in GitHub commit 14755416e364f17fb1870882fa778c7fec7f16e3 and will release TensorFlow 2.4.0 containing the patch. TensorFlow nightly packages after this commit will also have the issue resolved.

Since this issue also impacts TF versions before 2.4, we will patch all releases between 1.15 and 2.3 inclusive.

For more information

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

Severity

Low

CVE ID

CVE-2020-26270

Weaknesses

No CWEs