VYPR
← All advisories
Moderate severityNVD Advisory· Published Dec 10, 2020· Updated Aug 4, 2024

CHECK-fail in LSTM with zero-length input in TensorFlow

CVE-2020-26270

Description

In affected versions of TensorFlow 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. This is fixed in versions 1.15.5, 2.0.4, 2.1.3, 2.2.2, 2.3.2, and 2.4.0.

Affected packages

Versions sourced from the GitHub Security Advisory.

PackageAffected versionsPatched versions
tensorflowPyPI
< 1.15.51.15.5
tensorflowPyPI
>= 2.0.0, < 2.0.42.0.4
tensorflowPyPI
>= 2.1.0, < 2.1.32.1.3
tensorflowPyPI
>= 2.2.0, < 2.2.22.2.2
tensorflowPyPI
>= 2.3.0, < 2.3.22.3.2
tensorflow-cpuPyPI
< 1.15.51.15.5
tensorflow-cpuPyPI
>= 2.0.0, < 2.0.42.0.4
tensorflow-cpuPyPI
>= 2.1.0, < 2.1.32.1.3
tensorflow-cpuPyPI
>= 2.2.0, < 2.2.22.2.2
tensorflow-cpuPyPI
>= 2.3.0, < 2.3.22.3.2
tensorflow-gpuPyPI
< 1.15.51.15.5
tensorflow-gpuPyPI
>= 2.0.0, < 2.0.42.0.4
tensorflow-gpuPyPI
>= 2.1.0, < 2.1.32.1.3
tensorflow-gpuPyPI
>= 2.2.0, < 2.2.22.2.2
tensorflow-gpuPyPI
>= 2.3.0, < 2.3.22.3.2

Affected products

1

Patches

1
14755416e364

Prevent CHECK-fail in LSTM/GRU with zero-length input.

https://github.com/tensorflow/tensorflowMihai MaruseacDec 8, 2020via ghsa
commit
1 file changed · +6 −2
  • tensorflow/stream_executor/cuda/cuda_dnn.cc+6 −2 modified
    @@ -1468,7 +1468,9 @@ class CudnnRnnSequenceTensorDescriptor
   static port::StatusOr<CudnnRnnSequenceTensorDescriptor> Create(
       GpuExecutor* parent, int max_seq_length, int batch_size, int data_size,
       cudnnDataType_t data_type) {
-    CHECK_GT(max_seq_length, 0);
+    if (max_seq_length <= 0) {
+      return port::Status(port::error::INVALID_ARGUMENT, "max_seq_length <= 0");
+    }
     int dims[] = {batch_size, data_size, 1};
     int strides[] = {dims[1] * dims[2], dims[2], 1};
     TensorDescriptor tensor_desc = CreateTensorDescriptor();
@@ -1486,7 +1488,9 @@ class CudnnRnnSequenceTensorDescriptor
       GpuExecutor* parent, int max_seq_length, int batch_size, int data_size,
       const absl::Span<const int>& seq_lengths, bool time_major,
       cudnnDataType_t data_type) {
-    CHECK_GT(max_seq_length, 0);
+    if (max_seq_length <= 0) {
+      return port::Status(port::error::INVALID_ARGUMENT, "max_seq_length <= 0");
+    }
     int dims[] = {batch_size, data_size, 1};
     int strides[] = {dims[1] * dims[2], dims[2], 1};
     TensorDescriptor tensor_desc = CreateTensorDescriptor();

Vulnerability mechanics

Generated by null/stub on May 9, 2026. Inputs: CWE entries + fix-commit diffs from this CVE's patches. Citations validated against bundle.

References

7

News mentions

0

No linked articles in our index yet.