Out of bounds write in grappler in Tensorflow

ImpactThe function MakeGrapplerFunctionItem takes arguments that determine the sizes of inputs and outputs. If the inputs given are greater than or equal to the sizes of the outputs, an out-of-bounds memory read or a crash is triggered. PatchesWe have patched the issue in GitHub commit...

Segfault in `CompositeTensorVariantToComponents`

ImpactAn input encoded that is not a valid CompositeTensorVariant tensor will trigger a segfault in tf.rawops.CompositeTensorVariantToComponents.pythonimport tensorflow as tfencode = tf.rawops.EmptyTensorListelementdtype=tf.int32, elementshape=10, 15, maxnumelements=2meta=...

`CHECK` fail via inputs in `PyFunc`

ImpactAn input token that is not a UTF-8 bytestring will trigger a CHECK fail in tf.rawops.PyFunc.pythonimport tensorflow as tfvalue = tf.constantvalue=1,2token = b'\xb0'dataType = tf.int32tf.rawops.PyFuncinput=value,token=token,Tout=dataType PatchesWe have patched the issue in GitHub commit...

Segfault in `CompositeTensorVariantToComponents`

ImpactAn input encoded that is not a valid CompositeTensorVariant tensor will trigger a segfault in tf.rawops.CompositeTensorVariantToComponents.pythonimport tensorflow as tfencode = tf.rawops.EmptyTensorListelementdtype=tf.int32, elementshape=10, 15, maxnumelements=2meta=...

Overflow in `ResizeNearestNeighborGrad`

ImpactWhen tf.rawops.ResizeNearestNeighborGrad is given a large size input, it overflows.import tensorflow as tfaligncorners = Truehalfpixelcenters = Falsegrads = tf.constant1, shape=1,8,16,3, dtype=tf.float16size = tf.constant1879048192,1879048192, shape=2,...

`CHECK` fail via inputs in `PyFunc`

ImpactAn input token that is not a UTF-8 bytestring will trigger a CHECK fail in tf.rawops.PyFunc.pythonimport tensorflow as tfvalue = tf.constantvalue=1,2token = b'\xb0'dataType = tf.int32tf.rawops.PyFuncinput=value,token=token,Tout=dataType PatchesWe have patched the issue in GitHub commit...

FractionalMaxPool and FractionalAVGPool heap out-of-bounds acess

ImpactAn input poolingratio that is smaller than 1 will trigger a heap OOB in tf.rawops.FractionalMaxPool and tf.rawops.FractionalAvgPool. PatchesWe have patched the issue in GitHub commit 216525144ee7c910296f5b05d214ca1327c9ce48.The fix will be included in TensorFlow 2.11.0. We will also cherry...

`CHECK` fail via inputs in `SdcaOptimizer`

ImpactInputs densefeatures or examplestatedata not of rank 2 will trigger a CHECK fail in SdcaOptimizer.pythonimport tensorflow as tftf.rawops.SdcaOptimizer sparseexampleindices=4 tf.random.uniform5,5,5,3, dtype=tf.dtypes.int64, maxval=100, sparsefeatureindices=4 tf.random.uniform5,5,5,3,...

`CHECK` fail via inputs in `SparseFillEmptyRowsGrad`

ImpactIf SparseFillEmptyRowsGrad is given empty inputs, TensorFlow will crash.pythonimport tensorflow as tftf.rawops.SparseFillEmptyRowsGrad reverseindexmap=, gradvalues=, name=None PatchesWe have patched the issue in GitHub commit af4a6a3c8b95022c351edae94560acc61253a1b8.The fix will be included...

`CHECK` fail via inputs in `SdcaOptimizer`

ImpactInputs densefeatures or examplestatedata not of rank 2 will trigger a CHECK fail in SdcaOptimizer.pythonimport tensorflow as tftf.rawops.SdcaOptimizer sparseexampleindices=4 tf.random.uniform5,5,5,3, dtype=tf.dtypes.int64, maxval=100, sparsefeatureindices=4 tf.random.uniform5,5,5,3,...

`CHECK` fail via inputs in `SparseFillEmptyRowsGrad`

ImpactIf SparseFillEmptyRowsGrad is given empty inputs, TensorFlow will crash.pythonimport tensorflow as tftf.rawops.SparseFillEmptyRowsGrad reverseindexmap=, gradvalues=, name=None PatchesWe have patched the issue in GitHub commit af4a6a3c8b95022c351edae94560acc61253a1b8.The fix will be included...

`CHECK_EQ` fail via input in `SparseMatrixNNZ`

ImpactAn input sparsematrix that is not a matrix with a shape with rank 0 will trigger a CHECK fail in tf.rawops.SparseMatrixNNZ.pythonimport tensorflow as tftf.rawops.SparseMatrixNNZsparsematrix= PatchesWe have patched the issue in GitHub commit f856d02e5322821aad155dad9b3acab1e9f5d693.The fix...

FractionalMaxPool and FractionalAVGPool heap out-of-bounds acess

ImpactAn input poolingratio that is smaller than 1 will trigger a heap OOB in tf.rawops.FractionalMaxPool and tf.rawops.FractionalAvgPool. PatchesWe have patched the issue in GitHub commit 216525144ee7c910296f5b05d214ca1327c9ce48.The fix will be included in TensorFlow 2.11.0. We will also cherry...

`FractionalMaxPoolGrad` Heap out of bounds read

ImpactIf FractionMaxPoolGrad is given outsize inputs rowpoolingsequence and colpoolingsequence, TensorFlow will crash.pythonimport tensorflow as tftf.rawops.FractionMaxPoolGrad originput = 1, 1, 1, 1, 1, origoutput = 1, 1, 1, outbackprop = 3, 3, 6, rowpoolingsequence = -0x4000000, 1, 1,...

`FractionalMaxPoolGrad` Heap out of bounds read

ImpactIf FractionMaxPoolGrad is given outsize inputs rowpoolingsequence and colpoolingsequence, TensorFlow will crash.pythonimport tensorflow as tftf.rawops.FractionMaxPoolGrad originput = 1, 1, 1, 1, 1, origoutput = 1, 1, 1, outbackprop = 3, 3, 6, rowpoolingsequence = -0x4000000, 1, 1,...

`CHECK_EQ` fail in `tf.raw_ops.TensorListResize`

ImpactIf tf.rawops.TensorListResize is given a nonscalar value for input size, it results CHECK fail which can be used to trigger a denial of service attack.pythonimport numpy as npimport tensorflow as tfa = datastructures.tftensorlistnewelements = tf.constantvalue=3, 4, 5b = np.zeros0, 2, 3,...

`tf.raw_ops.Mfcc` crashes

ImpactIf ThreadUnsafeUnigramCandidateSampler is given input filterbankchannelcount greater than the allowed max size, TensorFlow will crash.pythonimport tensorflow as tftf.rawops.Mfcc spectrogram = 1.38, 6.32, 5.75, 9.51, samplerate = 2, upperfrequencylimit = 5.0, lowerfrequencylimit = 1.0,...

`CHECK_EQ` fail in `tf.raw_ops.TensorListResize`

ImpactIf tf.rawops.TensorListResize is given a nonscalar value for input size, it results CHECK fail which can be used to trigger a denial of service attack.pythonimport numpy as npimport tensorflow as tfa = datastructures.tftensorlistnewelements = tf.constantvalue=3, 4, 5b = np.zeros0, 2, 3,...

`MirrorPadGrad` heap out of bounds read

ImpactIf MirrorPadGrad is given outsize input paddings, TensorFlow will give a heap OOB error.pythonimport tensorflow as tftf.rawops.MirrorPadGradinput=1, paddings=0x77f00000,0xa000000, mode = 'REFLECT' PatchesWe have patched the issue in GitHub commit 717ca98d8c3bba348ff62281fdf38dcb5ea1ec92.The...

`MirrorPadGrad` heap out of bounds read

ImpactIf MirrorPadGrad is given outsize input paddings, TensorFlow will give a heap OOB error.pythonimport tensorflow as tftf.rawops.MirrorPadGradinput=1, paddings=0x77f00000,0xa000000, mode = 'REFLECT' PatchesWe have patched the issue in GitHub commit 717ca98d8c3bba348ff62281fdf38dcb5ea1ec92.The...

Segfault in `tf.raw_ops.TensorListConcat`

ImpactIf tf.rawops.TensorListConcat is given elementshape=, it results segmentation fault which can be used to trigger a denial of service attack.pythonimport tensorflow as tftf.rawops.TensorListConcat inputhandle=tf.data.experimental.tovarianttf.data.Dataset.fromtensorslices1, 2, 3,...

`tf.raw_ops.Mfcc` crashes

ImpactIf ThreadUnsafeUnigramCandidateSampler is given input filterbankchannelcount greater than the allowed max size, TensorFlow will crash.pythonimport tensorflow as tftf.rawops.Mfcc spectrogram = 1.38, 6.32, 5.75, 9.51, samplerate = 2, upperfrequencylimit = 5.0, lowerfrequencylimit = 1.0,...

Segfault in `tf.raw_ops.TensorListConcat`

ImpactIf tf.rawops.TensorListConcat is given elementshape=, it results segmentation fault which can be used to trigger a denial of service attack.pythonimport tensorflow as tftf.rawops.TensorListConcat inputhandle=tf.data.experimental.tovarianttf.data.Dataset.fromtensorslices1, 2, 3,...

Overflow in `tf.keras.losses.poisson`

Impacttf.keras.losses.poisson receives a ypred and ytrue that are passed through functor::mul in BinaryOp. If the resulting dimensions overflow an int32, TensorFlow will crash due to a size mismatch during broadcast assignment.pythonimport numpy as npimport tensorflow as tftruevalue =...

`CHECK` fail in `BCast` overflow

ImpactIf BCast::ToShape is given input larger than an int32, it will crash, despite being supposed to handle up to an int64. An example can be seen in tf.experimental.numpy.outer by passing in large input to the input b.pythonimport tensorflow as tfvalue = tf.constantshape=2, 1024, 1024, 1024,...

Segfault via invalid attributes in `pywrap_tfe_src.cc`

ImpactIf a list of quantized tensors is assigned to an attribute, the pywrap code fails to parse the tensor and returns a nullptr, which is not caught. An example can be seen in tf.compat.v1.extractvolumepatches by passing in quantized tensors as input ksizes.pythonimport numpy as npimport...

FPE in `tf.image.generate_bounding_box_proposals`

ImpactWhen running on GPU, tf.image.generateboundingboxproposals receives a scores input that must be of rank 4 but is not checked.pythonimport tensorflow as tfa = tf.constantvalue=1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0b =...

Segfault via invalid attributes in `pywrap_tfe_src.cc`

ImpactIf a list of quantized tensors is assigned to an attribute, the pywrap code fails to parse the tensor and returns a nullptr, which is not caught. An example can be seen in tf.compat.v1.extractvolumepatches by passing in quantized tensors as input ksizes.pythonimport numpy as npimport...

FPE in `tf.image.generate_bounding_box_proposals`

ImpactWhen running on GPU, tf.image.generateboundingboxproposals receives a scores input that must be of rank 4 but is not checked.pythonimport tensorflow as tfa = tf.constantvalue=1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0b =...

Overflow in `ImageProjectiveTransformV2`

ImpactWhen tf.rawops.ImageProjectiveTransformV2 is given a large output shape, it overflows.pythonimport tensorflow as tfinterpolation = "BILINEAR"fillmode = "REFLECT"images = tf.constant0.184634328, shape=2,5,8,3, dtype=tf.float32transforms = tf.constant0.378575385, shape=2,8,...

Overflow in `tf.keras.losses.poisson`

Impacttf.keras.losses.poisson receives a ypred and ytrue that are passed through functor::mul in BinaryOp. If the resulting dimensions overflow an int32, TensorFlow will crash due to a size mismatch during broadcast assignment.pythonimport numpy as npimport tensorflow as tftruevalue =...

`CHECK` fail in `BCast` overflow

ImpactIf BCast::ToShape is given input larger than an int32, it will crash, despite being supposed to handle up to an int64. An example can be seen in tf.experimental.numpy.outer by passing in large input to the input b.pythonimport tensorflow as tfvalue = tf.constantshape=2, 1024, 1024, 1024,...

Out of bounds segmentation fault due to unequal op inputs in Tensorflow

Impact tf.rawops.DynamicStitch specifies input sizes when it is registered. cppREGISTEROP"DynamicStitch" .Input"indices: N int32" .Input"data: N T" .Output"merged: T" .Attr"N : int = 1" .Attr"T : type" .SetShapeFnDynamicStitchShapeFunction;When it receives a differing number of inputs, such as wh...

Overflow in `ImageProjectiveTransformV2`

ImpactWhen tf.rawops.ImageProjectiveTransformV2 is given a large output shape, it overflows.pythonimport tensorflow as tfinterpolation = "BILINEAR"fillmode = "REFLECT"images = tf.constant0.184634328, shape=2,5,8,3, dtype=tf.float32transforms = tf.constant0.378575385, shape=2,8,...

Overflow in `FusedResizeAndPadConv2D`

ImpactWhen tf.rawops.FusedResizeAndPadConv2D is given a large tensor shape, it overflows.pythonimport tensorflow as tfmode = "REFLECT"strides = 1, 1, 1, 1padding = "SAME"resizealigncorners = Falseinput = tf.constant147, shape=3,3,1,1, dtype=tf.float16size = tf.constant1879048192,1879048192,...

Seg fault in `ndarray_tensor_bridge` due to zero and large inputs

ImpactIf a numpy array is created with a shape such that one element is zero and the others sum to a large number, an error will be raised. E.g. the following raises an error:pythonnp.ones0, 231, 231An example of a proof of concept:pythonimport numpy as npimport tensorflow as tfinputval =...

Out of bounds segmentation fault due to unequal op inputs in Tensorflow

Impact tf.rawops.DynamicStitch specifies input sizes when it is registered. cppREGISTEROP"DynamicStitch" .Input"indices: N int32" .Input"data: N T" .Output"merged: T" .Attr"N : int = 1" .Attr"T : type" .SetShapeFnDynamicStitchShapeFunction;When it receives a differing number of inputs, such as wh...

Overflow in `FusedResizeAndPadConv2D`

ImpactWhen tf.rawops.FusedResizeAndPadConv2D is given a large tensor shape, it overflows.pythonimport tensorflow as tfmode = "REFLECT"strides = 1, 1, 1, 1padding = "SAME"resizealigncorners = Falseinput = tf.constant147, shape=3,3,1,1, dtype=tf.float16size = tf.constant1879048192,1879048192,...

TensorFlow vulnerable to `CHECK` fail in `Save` and `SaveSlices`

ImpactIf Save or SaveSlices is run over tensors of an unsupported dtype, it results in a CHECK fail that can be used to trigger a denial of service attack.pythonimport tensorflow as tffilename = tf.constant""tensornames = tf.constant"" Savedata = tf.casttf.random.uniformshape=1, minval=-10000,...

OpenStack Sushy-Tools and VirtualBMC Improper Preservation of Permissions

An issue was discovered in OpenStack Sushy-Tools through 0.21.0 and VirtualBMC through 2.2.2. Changing the boot device configuration with these packages removes password protection from the managed libvirt XML domain. NOTE: this only affects an "unsupported, production-like configuration."...

Seg fault in `ndarray_tensor_bridge` due to zero and large inputs

ImpactIf a numpy array is created with a shape such that one element is zero and the others sum to a large number, an error will be raised. E.g. the following raises an error:pythonnp.ones0, 231, 231An example of a proof of concept:pythonimport numpy as npimport tensorflow as tfinputval =...

TensorFlow vulnerable to `CHECK` fail in `Save` and `SaveSlices`

ImpactIf Save or SaveSlices is run over tensors of an unsupported dtype, it results in a CHECK fail that can be used to trigger a denial of service attack.pythonimport tensorflow as tffilename = tf.constant""tensornames = tf.constant"" Savedata = tf.casttf.random.uniformshape=1, minval=-10000,...

TensorFlow vulnerable to `CHECK` fail in `ParameterizedTruncatedNormal`

ImpactParameterizedTruncatedNormal assumes shape is of type int32. A valid shape of type int64 results in a mismatched type CHECK fail that can be used to trigger a denial of service attack.pythonimport tensorflow as tfseed = 1618seed2 = 0shape = tf.random.uniformshape=3, minval=-10000,...

TensorFlow vulnerable to `CHECK` fail in `ParameterizedTruncatedNormal`

ImpactParameterizedTruncatedNormal assumes shape is of type int32. A valid shape of type int64 results in a mismatched type CHECK fail that can be used to trigger a denial of service attack.pythonimport tensorflow as tfseed = 1618seed2 = 0shape = tf.random.uniformshape=3, minval=-10000,...

TensorFlow vulnerable to `CHECK` fail in `LRNGrad`

ImpactIf LRNGrad is given an outputimage input tensor that is not 4-D, it results in a CHECK fail that can be used to trigger a denial of service attack.pythonimport tensorflow as tfdepthradius = 1bias = 1.59018219alpha = 0.117728651beta = 0.404427052inputgrads = tf.random.uniformshape=4, 4, 4, 4...

TensorFlow vulnerable to `CHECK` fail in `tf.linalg.matrix_rank`

ImpactWhen tf.linalg.matrixrank receives an empty input a, the GPU kernel gives a CHECK fail that can be used to trigger a denial of service attack.pythonimport tensorflow as tfa = tf.constant, shape=0, 1, 1, dtype=tf.float32tf.linalg.matrixranka=a PatchesWe have patched the issue in GitHub commi...

TensorFlow vulnerable to `CHECK` fail in `MaxPool`

ImpactWhen MaxPool receives a window size input array ksize with dimensions greater than its input tensor input, the GPU kernel gives a CHECK fail that can be used to trigger a denial of service attack.pythonimport tensorflow as tfimport numpy as npinput = np.ones1, 1, 1, 1ksize = 1, 1, 2, 2strid...

TensorFlow vulnerable to `CHECK` fail in `MaxPool`

ImpactWhen MaxPool receives a window size input array ksize with dimensions greater than its input tensor input, the GPU kernel gives a CHECK fail that can be used to trigger a denial of service attack.pythonimport tensorflow as tfimport numpy as npinput = np.ones1, 1, 1, 1ksize = 1, 1, 2, 2strid...

TensorFlow vulnerable to `CHECK` fail in `LRNGrad`

ImpactIf LRNGrad is given an outputimage input tensor that is not 4-D, it results in a CHECK fail that can be used to trigger a denial of service attack.pythonimport tensorflow as tfdepthradius = 1bias = 1.59018219alpha = 0.117728651beta = 0.404427052inputgrads = tf.random.uniformshape=4, 4, 4, 4...

TensorFlow vulnerable to segfault in `SparseBincount`

ImpactIf SparseBincount is given inputs for indices, values, and denseshape that do not make a valid sparse tensor, it results in a segfault that can be used to trigger a denial of service attack.pythonimport tensorflow as tfbinaryoutput = Trueindices = tf.random.uniformshape=, minval=-10000,...