CVE-2021-37661: Crash caused by integer conversion to unsigned in TensorFlow

Impact An attacker can cause a denial of service in boostedtreescreatequantilestreamresource by using negative arguments:

python import tensorflow as tf from tensorflow.python.ops import genboostedtreesops import numpy as np

v= tf.Variable([0.0, 0.0, 0.0, 0.0, 0.0]) genboostedtreesops.boostedtreescreatequantilestreamresource( quantilestreamresourcehandle = v.handle, epsilon = [74.82224], numstreams = [-49], maxelements = np.int32(586))

The implementation does not validate that numstreams only contains non-negative numbers. In turn, this results in using this value to allocate memory:

cc class BoostedTreesQuantileStreamResource : public ResourceBase { public: BoostedTreesQuantileStreamResource(const float epsilon, const int64 maxelements, const int64 numstreams) : arebucketsready(false), epsilon(epsilon), numstreams(numstreams), maxelements(maxelements) { streams.reserve(numstreams); ... } }

However, reserve receives an unsigned integer so there is an implicit conversion from a negative value to a large positive unsigned. This results in a crash from the standard library.

Patches We have patched the issue in GitHub commit 8a84f7a2b5a2b27ecf88d25bad9ac777cd2f7992.

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.

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

Attribution This vulnerability has been reported by members of the Aivul Team from Qihoo 360.