CVE-2022-41894: TensorFlow Lite: buffer overflow in CONV_3D_TRANSPOSE op

TensorFlow is an open source platform for machine learning. The reference kernel of the `CONV_3D_TRANSPOSE` TensorFlow Lite operator wrongly increments the data_ptr when adding the bias to the result. Instead of `data_ptr += num_channels;` it should be `data_ptr += output_num_channels;` as if the number of input channels is different than the number of output channels, the wrong result will be returned and a buffer overflow will occur if num_channels > output_num_channels. An attacker can craft a model with a specific number of input channels. It is then possible to write specific values through the bias of the layer outside the bounds of the buffer. This attack only works if the reference kernel resolver is used in the interpreter. We have patched the issue in GitHub commit 72c0bdcb25305b0b36842d746cc61d72658d2941. The fix will be included in TensorFlow 2.11. We will also cherrypick this commit on TensorFlow 2.10.1, 2.9.3, and TensorFlow 2.8.4, as these are also affected and still in supported range.

An adversary targeting an organization's edge AI inference pipeline (e.g., smart camera fleet, federated learning node) crafts a TFLite model where a CONV_3D_TRANSPOSE layer has num_input_channels significantly greater than num_output_channels. The model is delivered via a compromised model update server or malicious model upload to a training/serving platform. When the TFLite runtime loads and executes the model with the reference kernel resolver active, the bias loop overwrites heap memory beyond the output buffer. Since bias values are attacker-controlled, this enables targeted corruption of adjacent heap structures — function pointers, vtables — potentially yielding code execution on edge devices where patching cadence is slow.