MISFIRE

Browse / AI Framework and Platform Flaws / CVE-2022-36014

TensorFlow vulnerable to null-dereference in `mlir::tfg::TFOp::nameAttr`

CVE-2022-36014 ยท 2022-09-16 moderate

Class
AI Framework and Platform Flaws (filed by a model that read the source)
Technique
Denial of service
Also involves
Unsafe Model Loading and Supply Chain
Affected
tensorflow-cpu, tensorflow (PyPI)
CVSS
CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H
EPSS
0.7% chance of exploitation in 30 days

How it works, in short

Importing a graph definition with null type-list attributes causes a null pointer dereference that crashes the TensorFlow MLIR importer. (Our summary, written by a model. The source text is below.)

Details

Impact

When mlir::tfg::TFOp::nameAttr receives null type list attributes, it crashes.


StatusOr<unsigned> GraphDefImporter::ArgNumType(const NamedAttrList &attrs,
                                                const OpDef::ArgDef &arg_def,
                                                SmallVectorImpl<Type> &types) {
  // Check whether a type list attribute is specified.
  if (!arg_def.type_list_attr().empty()) {
    if (auto v = attrs.get(arg_def.type_list_attr()).dyn_cast<ArrayAttr>()) {
      for (Attribute attr : v) {
        if (auto dtype = attr.dyn_cast<TypeAttr>()) {
          types.push_back(UnrankedTensorType::get(dtype.getValue()));
        } else {
          return InvalidArgument("Expected '", arg_def.type_list_attr(),
                                 "' to be a list of types");
        }
      }
      return v.size();
    }
    return NotFound("Type attr not found: ", arg_def.type_list_attr());
  }

  unsigned num = 1;
  // Check whether a number attribute is specified.
  if (!arg_def.number_attr().empty()) {
    if (auto v = attrs.get(arg_def.number_attr()).dyn_cast<IntegerAttr>()) {
      num = v.getValue().getZExtValue();
    } else {
      return NotFound("Type attr not found: ", arg_def.number_attr());
    }
  }

  // Check for a type or type attribute.
  Type dtype;
  if (arg_def.type() != DataType::DT_INVALID) {
    TF_RETURN_IF_ERROR(ConvertDataType(arg_def.type(), b_, &dtype));
  } else if (arg_def.type_attr().empty()) {
    return InvalidArgument("Arg '", arg_def.name(),
                           "' has invalid type and no type attribute");
  } else {
    if (auto v = attrs.get(arg_def.type_attr()).dyn_cast<TypeAttr>()) {
      dtype = v.getValue();
    } else {
      return NotFound("Type attr not found: ", arg_def.type_attr());
    }
  }
  types.append(num, UnrankedTensorType::get(dtype));
  return num;
}

Patches

We have patched the issue in GitHub commits 3a754740d5414e362512ee981eefba41561a63a6 and a0f0b9a21c9270930457095092f558fbad4c03e5.

The fix will be included in TensorFlow 2.10.0. We will also cherrypick this commit on TensorFlow 2.9.1, TensorFlow 2.8.1, and TensorFlow 2.7.2, as these are also affected and still in supported range.

For more information

Please consult our security guide for more information regarding the security model and how to contact us with issues and questions.

References

Source: GitHub Advisory Database, used under CC BY 4.0.