| Package | Affected versions | Patched versions |
|---|---|---|
| tensorflow | ≥ 2.0.0, < 2.0.3≥ 2.1.0, < 2.1.2≥ 2.2.0, < 2.2.1≥ 2.3.0, < 2.3.1< 1.15.4 | 2.0.32.1.22.2.12.3.11.15.4 |
| tensorflow-cpu | ≥ 2.0.0, < 2.0.3≥ 2.1.0, < 2.1.2≥ 2.2.0, < 2.2.1≥ 2.3.0, < 2.3.1< 1.15.4 | 2.0.32.1.22.2.12.3.11.15.4 |
| tensorflow-gpu | ≥ 2.0.0, < 2.0.3≥ 2.1.0, < 2.1.2≥ 2.2.0, < 2.2.1≥ 2.3.0, < 2.3.1< 1.15.4 | 2.0.32.1.22.2.12.3.11.15.4 |
| tensorflow | ≥ 2.0.0, < 2.0.3≥ 2.1.0, < 2.1.2≥ 2.2.0, < 2.2.1≥ 2.3.0, < 2.3.1< 1.15.4 | 2.0.32.1.22.2.12.3.11.15.4 |
By controlling the fill argument of tf.strings.as_string, a malicious attacker is able to trigger a format string vulnerability due to the way the internal format use in a printf call is constructed: https://github.com/tensorflow/tensorflow/blob/0e68f4d3295eb0281a517c3662f6698992b7b2cf/tensorflow/core/kernels/as_string_op.cc#L68-L74
This can result in unexpected output:
In [1]: tf.strings.as_string(input=[1234], width=6, fill='-')
Out[1]: <tf.Tensor: shape=(1,), dtype=string, numpy=array(['1234 '], dtype=object)>
In [2]: tf.strings.as_string(input=[1234], width=6, fill='+')
Out[2]: <tf.Tensor: shape=(1,), dtype=string, numpy=array([' +1234'], dtype=object)>
In [3]: tf.strings.as_string(input=[1234], width=6, fill="h")
Out[3]: <tf.Tensor: shape=(1,), dtype=string, numpy=array(['%6d'], dtype=object)>
In [4]: tf.strings.as_string(input=[1234], width=6, fill="d")
Out[4]: <tf.Tensor: shape=(1,), dtype=string, numpy=array(['12346d'], dtype=object)>
In [5]: tf.strings.as_string(input=[1234], width=6, fill="o")
Out[5]: <tf.Tensor: shape=(1,), dtype=string, numpy=array(['23226d'], dtype=object)>
In [6]: tf.strings.as_string(input=[1234], width=6, fill="x")
Out[6]: <tf.Tensor: shape=(1,), dtype=string, numpy=array(['4d26d'], dtype=object)>
In [7]: tf.strings.as_string(input=[1234], width=6, fill="g")
Out[7]: <tf.Tensor: shape=(1,), dtype=string, numpy=array(['8.67458e-3116d'], dtype=object)>
In [8]: tf.strings.as_string(input=[1234], width=6, fill="a")
Out[8]: <tf.Tensor: shape=(1,), dtype=string, numpy=array(['0x0.00ff7eebb4d4p-10226d'], dtype=object)>
In [9]: tf.strings.as_string(input=[1234], width=6, fill="c")
Out[9]: <tf.Tensor: shape=(1,), dtype=string, numpy=array(['\xd26d'], dtype=object)>
In [10]: tf.strings.as_string(input=[1234], width=6, fill="p")
Out[10]: <tf.Tensor: shape=(1,), dtype=string, numpy=array(['0x4d26d'], dtype=object)>
In [11]: tf.strings.as_string(input=[1234], width=6, fill='m')
Out[11]: <tf.Tensor: shape=(1,), dtype=string, numpy=array(['Success6d'], dtype=object)>
However, passing in n or s results in segmentation fault.
We have patched the issue in 33be22c65d86256e6826666662e40dbdfe70ee83 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.
Running one of these products? Scan your site to see whether it exposes a vulnerable version.
Scan my site free