1## TFSA-2021-154: Division by zero in TFLite 2 3### CVE Number 4CVE-2021-37680 5 6### Impact 7The implementation of fully connected layers in TFLite is [vulnerable to a 8division by zero 9error](https://github.com/tensorflow/tensorflow/blob/460e000de3a83278fb00b61a16d161b1964f15f4/tensorflow/lite/kernels/fully_connected.cc#L226): 10 11```cc 12const int batch_size = input_size / filter->dims->data[1]; 13``` 14 15An attacker can craft a model such that `filter->dims->data[1]` is 0. 16 17### Patches 18We have patched the issue in GitHub commit 19[718721986aa137691ee23f03638867151f74935f](https://github.com/tensorflow/tensorflow/commit/718721986aa137691ee23f03638867151f74935f). 20 21The fix will be included in TensorFlow 2.6.0. We will also cherrypick this 22commit on TensorFlow 2.5.1, TensorFlow 2.4.3, and TensorFlow 2.3.4, as these are 23also affected and still in supported range. 24 25### For more information 26Please consult [our security 27guide](https://github.com/tensorflow/tensorflow/blob/master/SECURITY.md) for 28more information regarding the security model and how to contact us with issues 29and questions. 30 31### Attribution 32This vulnerability has been reported by members of the Aivul Team from Qihoo 33360. Concurrently, it has also been reported by Yakun Zhang of Baidu Security. 34