Headline
CVE-2022-35981: Fix security vulnerability with FractionalMaxPoolGrad · tensorflow/tensorflow@8741e57
TensorFlow is an open source platform for machine learning. FractionalMaxPoolGrad validates its inputs with CHECK failures instead of with returning errors. If it gets incorrectly sized inputs, the CHECK failure can be used to trigger a denial of service attack. We have patched the issue in GitHub commit 8741e57d163a079db05a7107a7609af70931def4. 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. There are no known workarounds for this issue.
@@ -124,7 +124,7 @@ def _ValidateFractionalMaxPoolResult(self, input_tensor, pooling_ratio, Returns: None “"” with self.cached_session() as sess: with self.cached_session(): p, r, c = nn_ops.fractional_max_pool_v2( input_tensor, pooling_ratio, @@ -155,7 +155,7 @@ def _testVisually(self): overlapping)) rand_mat = self._PRNG.randint(10, size=tensor_shape) pooling_ratio = [1, math.sqrt(2), math.sqrt(2), 1] with self.cached_session() as sess: with self.cached_session(): p, r, c = nn_ops.fractional_max_pool_v2( rand_mat, pooling_ratio, @@ -630,6 +630,29 @@ def testWhenRepeatedMaxValueInPoolingRegion(self): self.assertAllClose(expected_input_backprop_overlapping, input_backprop_overlapping)
def testInvalidSeqRaiseErrorForFractionalMaxPoolGrad(self): with self.assertRaises(errors.InvalidArgumentError): with self.cached_session() as _: overlapping = True orig_input = constant_op.constant( .453409232, shape=[1, 7, 13, 1], dtype=dtypes.float32) orig_output = constant_op.constant( .453409232, shape=[1, 7, 13, 1], dtype=dtypes.float32) out_backprop = constant_op.constant( .453409232, shape=[1, 7, 13, 1], dtype=dtypes.float32) row_pooling_sequence = constant_op.constant( 0, shape=[5], dtype=dtypes.int64) col_pooling_sequence = constant_op.constant( 0, shape=[5], dtype=dtypes.int64) t = gen_nn_ops.FractionalMaxPoolGrad( orig_input=orig_input, orig_output=orig_output, out_backprop=out_backprop, row_pooling_sequence=row_pooling_sequence, col_pooling_sequence=col_pooling_sequence, overlapping=overlapping) self.evaluate(t)
if __name__ == "__main__": test.main()
Related news
### Impact `FractionalMaxPoolGrad` validates its inputs with `CHECK` failures instead of with returning errors. If it gets incorrectly sized inputs, the `CHECK` failure can be used to trigger a denial of service attack: ```python import tensorflow as tf overlapping = True orig_input = tf.constant(.453409232, shape=[1,7,13,1], dtype=tf.float32) orig_output = tf.constant(.453409232, shape=[1,7,13,1], dtype=tf.float32) out_backprop = tf.constant(.453409232, shape=[1,7,13,1], dtype=tf.float32) row_pooling_sequence = tf.constant(0, shape=[5], dtype=tf.int64) col_pooling_sequence = tf.constant(0, shape=[5], dtype=tf.int64) tf.raw_ops.FractionalMaxPoolGrad(orig_input=orig_input, orig_output=orig_output, out_backprop=out_backprop, row_pooling_sequence=row_pooling_sequence, col_pooling_sequence=col_pooling_sequence, overlapping=overlapping) ``` ### Patches We have patched the issue in GitHub commit [8741e57d163a079db05a7107a7609af70931def4](https://github.com/tensorflow/tensorflow/commit/8741...