1## TFSA-2020-007: Null pointer dereference in TFLite 2 3### CVE Number 4CVE-2020-15209 5 6### Impact 7A crafted TFLite model can force a node to have as input a tensor backed by a 8`nullptr` buffer. This can be achieved by changing a buffer index in the 9flatbuffer serialization to convert a read-only tensor to a read-write one. The 10runtime assumes that these buffers are written to before a possible read, hence 11they are [initialized with 12`nullptr`](https://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/lite/core/subgraph.cc#L1224-L1227): 13```cc 14 TfLiteTensorReset(type, name, ConvertArrayToTfLiteIntArray(rank, dims), 15 GetLegacyQuantization(quantization), 16 /*buffer=*/nullptr, required_bytes, allocation_type, 17 nullptr, is_variable, &tensor); 18``` 19 20However, by changing the buffer index for a tensor and implicitly converting 21that tensor to be a read-write one, as there is nothing in the model that writes 22to it, we get a null pointer dereference. 23 24### Vulnerable Versions 25TensorFlow 1.15.0, 1.15.1, 1.15.2, 1.15.3, 2.0.0, 2.0.1, 2.0.2, 2.1.0, 2.1.1, 262.2.0, 2.3.0. 27 28### Patches 29We have patched the issue in 30[0b5662bc](https://github.com/tensorflow/tensorflow/commit/0b5662bc) and will 31release patch releases for all versions between 1.15 and 2.3. 32 33We recommend users to upgrade to TensorFlow 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 342.3.1. 35 36### Workarounds 37A potential workaround would be to add a custom `Verifier` to the model loading 38code to ensure that no operator reuses tensors as both inputs and outputs. Care 39should be taken to check all types of inputs (i.e., constant or variable tensors 40as well as optional tensors). 41 42### For more information 43Please consult [our security 44guide](https://github.com/tensorflow/tensorflow/blob/master/SECURITY.md) for 45more information regarding the security model and how to contact us with issues 46and questions. 47 48### Attribution 49This vulnerability has been discovered by members of the Aivul Team and is also 50discoverable through a variant analysis of [another 51vulnerability reported by members of the Aivul Team from Qihoo 52360](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/security/advisory/tfsa-2020-005.md). 53