xref: /aosp_15_r20/external/tensorflow/tensorflow/security/advisory/tfsa-2020-033.md (revision b6fb3261f9314811a0f4371741dbb8839866f948)
1## TFSA-2020-033: CHECK-fail in LSTM with zero-length input
2
3### CVE Number
4CVE-2020-26270
5
6### Impact
7Running an LSTM/GRU model where the LSTM/GRU layer receives an input with
8zero-length results in a `CHECK` failure when using the CUDA backend.
9
10This can result in a query-of-death vulnerability, via denial of service, if
11users can control the input to the layer.
12
13### Patches
14
15We have patched the issue in GitHub commit
16[14755416e364f17fb1870882fa778c7fec7f16e3](https://github.com/tensorflow/tensorflow/commit/14755416e364f17fb1870882fa778c7fec7f16e3)
17and will release TensorFlow 2.4.0 containing the patch. TensorFlow nightly
18packages after this commit will also have the issue resolved.
19
20Since this issue also impacts TF versions before 2.4, we will patch all releases
21between 1.15 and 2.3 inclusive.
22
23### For more information
24Please consult [our security
25guide](https://github.com/tensorflow/tensorflow/blob/master/SECURITY.md) for
26more information regarding the security model and how to contact us with issues
27and questions.
28