TensorFlow is an end-to-end open source platform for machine learning. In affected versions an attacker can cause undefined behavior via binding a reference to null pointer in tf.raw_ops.SparseFillEmptyRows. The shape inference implementation does not validate that the input arguments are not empty tensors. We have patched the issue in GitHub commit 578e634b4f1c1c684d4b4294f9e5281b2133b3ed. The fix will be included in TensorFlow 2.6.0. We will also cherrypick this commit on TensorFlow 2.5.1, TensorFlow 2.4.3, and TensorFlow 2.3.4, as these are also affected and still in supported range.
The product accesses or uses a pointer that has not been initialized.
| Name | Vendor | Start Version | End Version |
|---|---|---|---|
| Tensorflow | 2.3.0 (including) | 2.3.4 (excluding) | |
| Tensorflow | 2.4.0 (including) | 2.4.3 (excluding) | |
| Tensorflow | 2.5.0 (including) | 2.5.0 (including) | |
| Tensorflow | 2.6.0-rc0 (including) | 2.6.0-rc0 (including) | |
| Tensorflow | 2.6.0-rc1 (including) | 2.6.0-rc1 (including) | |
| Tensorflow | 2.6.0-rc2 (including) | 2.6.0-rc2 (including) |
If the pointer contains an uninitialized value, then the value might not point to a valid memory location. This could cause the product to read from or write to unexpected memory locations, leading to a denial of service. If the uninitialized pointer is used as a function call, then arbitrary functions could be invoked. If an attacker can influence the portion of uninitialized memory that is contained in the pointer, this weakness could be leveraged to execute code or perform other attacks. Depending on memory layout, associated memory management behaviors, and product operation, the attacker might be able to influence the contents of the uninitialized pointer, thus gaining more fine-grained control of the memory location to be accessed.