CVE-2020-15207
To mimic Python's indexing with negative values, TFLite uses ResolveAxis
to
convert negative values to positive indices. However, the only check that the
converted index is now valid is only present in debug
builds:
// Handle negative index. A positive index 'p_idx' can be represented as a
// negative index 'n_idx' as: n_idx = p_idx-num_dims
// eg: For num_dims=3, [0, 1, 2] is the same as [-3, -2, -1] */
int current = axis[idx] < 0 ? (axis[idx] + num_dims) : axis[idx];
TFLITE_DCHECK(current >= 0 && current < num_dims);
If the DCHECK
does not trigger, then code execution moves ahead with a
negative index. This, in turn, results in accessing data out of bounds which
results in segfaults and/or data corruption.
TensorFlow 1.15.0, 1.15.1, 1.15.2, 1.15.3, 2.0.0, 2.0.1, 2.0.2, 2.1.0, 2.1.1, 2.2.0, 2.3.0.
We have patched the issue in 2d88f470dea2671b430884260f3626b1fe99830a and will release patch releases for all versions between 1.15 and 2.3.
We recommend users to upgrade to TensorFlow 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1.
Please consult our security guide for more information regarding the security model and how to contact us with issues and questions.
This vulnerability has been reported by members of the Aivul Team from Qihoo 360.