Impact The code for boosted trees in TensorFlow is still missing validation. As a result, attackers can trigger denial of service (via dereferencing nullptrs or via CHECK-failures) as well as abuse undefined behavior (binding references to nullptrs). An attacker can also read and write from heap buffers, depending on the API that gets used and the arguments that are passed to the call.

Note: Given that the boosted trees implementation in TensorFlow is unmaintained, it is recommend to no longer use these APIs. Instead, please use the downstream TensorFlow Decision Forests project which is newer and supports more features. We will deprecate TensorFlow's boosted trees APIs in subsequent releases.

Patches We have patched the issue in GitHub commit 5c8c9a8bfe750f9743d0c859bae112060b216f5c.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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.

Impact The shape inference function for Transpose is vulnerable to a heap buffer overflow:

python import tensorflow as tf @tf.function def test(): y = tf.rawops.Transpose(x=[1,2,3,4],perm=[-10]) return y

test()

This occurs whenever perm contains negative elements. The shape inference function does not validate that the indices in perm are all valid: cc for (int32t i = 0; i < rank; ++i) { int64t inidx = data[i]; if (inidx >= rank) { return errors::InvalidArgument("perm dim ", inidx, " is out of range of input rank ", rank); } dims[i] = c->Dim(input, inidx); }

where Dim(tensor, index) accepts either a positive index less than the rank of the tensor or the special value -1 for unknown dimensions.

Patches We have patched the issue in GitHub commit c79ba87153ee343401dbe9d1954d7f79e521eb14.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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.

Impact Several TensorFlow operations are missing validation for the shapes of the tensor arguments involved in the call. Depending on the API, this can result in undefined behavior and segfault or CHECK-fail related crashes but in some scenarios writes and reads from heap populated arrays are also possible.

We have discovered these issues internally via tooling while working on improving/testing GPU op determinism. As such, we don't have reproducers and there will be multiple fixes for these issues.

Patches We have patched the issue in GitHub commits 68422b215e618df5ad375bcdc6d2052e9fd3080a, 4d74d8a00b07441cba090a02e0dd9ed385145bf4, 579261dcd446385831fe4f7457d802a59685121d, da4aad5946be30e5f049920fa076e1f7ef021261, 4dddb2fd0b01cdd196101afbba6518658a2c9e07, and e7f497570abb6b4ae5af4970620cd880e4c0c904.

These fixes will be included in TensorFlow 2.7.0. We will also cherrypick these commits on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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.

Impact The shape inference code for the Cudnn operations in TensorFlow can be tricked into accessing invalid memory, via a heap buffer overflow:

python import tensorflow as tf

@tf.function def func(): return tf.rawops.CudnnRNNV3( input=[0.1, 0.1], inputh=[0.5], inputc=[0.1, 0.1, 0.1], params=[0.5, 0.5], sequencelengths=[-1, 0, 1]) func() This occurs because the ranks of the input, inputh and inputc parameters are not validated, but code assumes they have certain values:

cc auto inputshape = c->input(0); auto inputhshape = c->input(1); auto seqlength = c->Dim(inputshape, 0); auto batchsize = c->Dim(inputshape, 1); // assumes rank >= 2 auto numunits = c->Dim(inputhshape, 2); // assumes rank >= 3

Patches We have patched the issue in GitHub commit af5fcebb37c8b5d71c237f4e59c6477015c78ce6.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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.

Impact TensorFlow's savedmodelcli tool is vulnerable to a code injection as it calls eval on user supplied strings python def preprocessinputexprsargstring(inputexprsstr): ... for inputraw in filter(bool, inputexprsstr.split(';')): ... inputkey, expr = inputraw.split('=', 1) inputdict[inputkey] = eval(expr) ... This can be used by attackers to run arbitrary code on the plaform where the CLI tool runs. However, given that the tool is always run manually, the impact of this is not severe. We have patched this by adding a safe flag which defaults to True and an explicit warning for users.

Patches We have patched the issue in GitHub commit 8b202f08d52e8206af2bdb2112a62fafbc546ec7.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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 Omer Kaspi from Vdoo.

Impact TensorFlow's Grappler optimizer has a use of unitialized variable:

cc const NodeDef dequeuenode; for (const auto& trainnode : trainnodes) { if (IsDequeueOp(trainnode)) { dequeuenode = trainnode; break; } }

if (dequeuenode) { ... }

If the trainnodes vector (obtained from the saved model that gets optimized) does not contain a Dequeue node, then dequeuenode is left unitialized.

Patches We have patched the issue in GitHub commit 68867bf01239d9e1048f98cbad185bf4761bedd3.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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 Qian Feng from Baidu Security Team.

Impact The async implementation of CollectiveReduceV2 suffers from a memory leak and a use after free:

python import tensorflow as tf tf.rawops.CollectiveReduceV2( input=[], groupsize=[-10, -10, -10], groupkey=[-10, -10], instancekey=[-10], orderingtoken=[], mergeop='Mul', finalop='Div')

This occurs due to the asynchronous computation and the fact that objects that have been std::move()d from are still accessed:

cc auto donewithcleanup = colparams, done = std::move(done) { done(); colparams->Unref(); }; OPREQUIRESOKASYNC(c, FillCollectiveParams(colparams, REDUCTIONCOLLECTIVE, /groupsize/ c->input(1), /groupkey/ c->input(2), /instancekey/ c->input(3)), done);

Here, done is already moved from by the time OPREQUIRESOKASYNC macro needs to invoke it in case of errors. In this case, we get an undefined behavior, which can manifest via crashes, std::badalloc throws or just memory leaks.

Patches We have patched the issue in GitHub commit ca38dab9d3ee66c5de06f11af9a4b1200da5ef75.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, as this version is the only one that is also affected.

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.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, TensorFlow's savedmodelcli tool is vulnerable to a code injection. This can be used to open a reverse shell. This code path was maintained for compatibility reasons as the maintainers had several test cases where numpy expressions were used as arguments. However, given that the tool is always run manually, the impact of this is still not severe. The maintainers have now removed the safe=False argument, so all parsing is done without calling eval. The patch is available in versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, the implementation of tf.rawops.EditDistance has incomplete validation. Users can pass negative values to cause a segmentation fault based denial of service. In multiple places throughout the code, one may compute an index for a write operation. However, the existing validation only checks against the upper bound of the array. Hence, it is possible to write before the array by massaging the input to generate negative values for loc. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

Impact The ImmutableConst operation in TensorFlow can be tricked into reading arbitrary memory contents:

python import tensorflow as tf with open('/tmp/test','wb') as f: f.write(b'\xe2'128) data = tf.rawops.ImmutableConst(dtype=tf.string,shape=3,memoryregionname='/tmp/test') print(data) This is because the tstring TensorFlow string class has a special case for memory mapped strings but the operation itself does not offer any support for this datatype.

Patches We have patched the issue in GitHub commit 3712a2d3455e6ccb924daa5724a3652a86f6b585 and GitHub commit 1cb6bb6c2a6019417c9adaf9e6843ba75ee2580b. The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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.

Impact The shape inference code for AllToAll can be made to execute a division by 0:

python import tensorflow as tf @tf.function def func(): return tf.rawops.AllToAll( input=[0.0, 0.1652, 0.6543], groupassignment=[1, -1], concatdimension=0, splitdimension=0, splitcount=0)

func()

This occurs whenever the splitcount argument is 0: cc TFRETURNIFERROR(c->GetAttr("splitcount", &splitcount)); ... for (int32t i = 0; i < rank; ++i) { ... dims[i] = c->MakeDim(c->Value(dims[i]) / splitcount); ... }

Patches We have patched the issue in GitHub commit a8ad3e5e79c75f36edb81e0ba3f3c0c5442aeddc.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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.

Impact The implementation of ParallelConcat misses some input validation and can produce a division by 0:

python import tensorflow as tf

@tf.function def test(): y = tf.rawops.ParallelConcat(values=[['tf']],shape=0) return y

test()

Patches We have patched the issue in GitHub commit f2c3931113eaafe9ef558faaddd48e00a6606235.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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.

Impact While calculating the size of the output within the tf.range kernel, there is a conditional statement of type int64 = condition ? int64 : double. Due to C++ implicit conversion rules, both branches of the condition will be cast to double and the result would be truncated before the assignment. This result in overflows:

python import tensorflow as tf

tf.sparse.eye(numrows=9223372036854775807, numcolumns=None) Similarly, tf.range would result in crashes due to overflows if the start or end point are too large.

python import tensorflow as tf

tf.range(start=-1e+38, limit=1)

Patches We have patched the issue in GitHub commits 6d94002a09711d297dbba90390d5482b76113899 (merging #51359) and 1b0e0ec27e7895b9985076eab32445026ae5ca94 (merging #51711).

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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 externally via GitHub issue, GitHub issue and GitHub issue.

Impact The implementations for convolution operators trigger a division by 0 if passed empty filter tensor arguments.

Patches We have patched the issue in GitHub commit f2c3931113eaafe9ef558faaddd48e00a6606235. The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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.

Impact The implementation of SplitV can trigger a segfault is an attacker supplies negative arguments:

python import tensorflow as tf

tf.rawops.SplitV( value=tf.constant([]), sizesplits=[-1, -2] ,axis=0, numsplit=2) This occurs whenever sizesplits contains more than one value and at least one value is negative. Patches We have patched the issue in GitHub commit 25d622ffc432acc736b14ca3904177579e733cc6.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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.

Impact The code behind tf.function API can be made to deadlock when two tf.function decorated Python functions are mutually recursive:

python import tensorflow as tf

@tf.function() def fun1(num): if num == 1: return print(num) fun2(num-1)

@tf.function() def fun2(num): if num == 0: return print(num) fun1(num-1)

fun1(9)

This occurs due to using a non-reentrant Lock Python object.

Loading any model which contains mutually recursive functions is vulnerable. An attacker can cause denial of service by causing users to load such models and calling a recursive tf.function, although this is not a frequent scenario.

Patches We have patched the issue in GitHub commit afac8158d43691661ad083f6dd9e56f327c1dcb7.

The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, 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.

Impact The implementation of tf.rawops.LSTMBlockCell does not fully validate the input arguments. This results in a CHECK-failure which can be used to trigger a denial of service attack:

python import tensorflow as tf

tf.rawops.LSTMBlockCell( x=tf.constant(0.837607, shape=[28,29], dtype=tf.float32), csprev=tf.constant(0, shape=[28,17], dtype=tf.float32), hprev=tf.constant(0.592631638, shape=[28,17], dtype=tf.float32), w=tf.constant(0.887386262, shape=[46,68], dtype=tf.float32), wci=tf.constant(0, shape=[], dtype=tf.float32), wcf=tf.constant(0, shape=[17], dtype=tf.float32), wco=tf.constant(0.592631638, shape=[28,17], dtype=tf.float32), b=tf.constant(0.75259006, shape=[68], dtype=tf.float32), forgetbias=1, cellclip=0, usepeephole=False) The code does not validate the ranks of any of the arguments to this API call. This results in CHECK-failures when the elements of the tensor are accessed. Patches We have patched the issue in GitHub commit 803404044ae7a1efac48ba82d74111fce1ddb09a. The fix will be included in TensorFlow 2.9.0. We will also cherrypick this commit on TensorFlow 2.8.1, TensorFlow 2.7.2, and TensorFlow 2.6.4, 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 Neophytos Christou from Secure Systems Lab at Brown University.

Impact The implementation of tf.ragged.constant does not fully validate the input arguments. This results in a denial of service by consuming all available memory:

python import tensorflow as tf tf.ragged.constant(pylist=[],raggedrank=8968073515812833920) Patches We have patched the issue in GitHub commit bd4d5583ff9c8df26d47a23e508208844297310e.

The fix will be included in TensorFlow 2.9.0. We will also cherrypick this commit on TensorFlow 2.8.1, TensorFlow 2.7.2, and TensorFlow 2.6.4, 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 externally via a GitHub issue.

Impact The tf.compat.v1.signal.rfft2d and tf.compat.v1.signal.rfft3d lack input validation and under certain condition can result in crashes (due to CHECK-failures).

Patches We have patched the issue in GitHub commit 0a8a781e597b18ead006d19b7d23d0a369e9ad73 (merging GitHub PR #55274).

The fix will be included in TensorFlow 2.9.0. We will also cherrypick this commit on TensorFlow 2.8.1, TensorFlow 2.7.2, and TensorFlow 2.6.4, 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 externally via a GitHub issue.

Impact The implementation of tf.rawops.Conv3DBackpropFilterV2 does not fully validate the input arguments. This results in a CHECK-failure which can be used to trigger a denial of service attack:

python import tensorflow as tf

tf.rawops.Conv3DBackpropFilterV2( input=tf.constant(.5053710941, shape=[2,2,2,2,1], dtype=tf.float16), filtersizes=tf.constant(0, shape=[], dtype=tf.int32), outbackprop=tf.constant(.5053710941, shape=[2,2,2,2,1], dtype=tf.float16), strides=[1, 1, 1, 1, 1], padding="VALID", dataformat="NDHWC", dilations=[1, 1, 1, 1, 1]) The code does not validate that the filtersizes argument is a vector. Patches We have patched the issue in GitHub commit 174c5096f303d5be7ed2ca2662b08371bff4ab88.

The fix will be included in TensorFlow 2.9.0. We will also cherrypick this commit on TensorFlow 2.8.1, TensorFlow 2.7.2, and TensorFlow 2.6.4, 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 Neophytos Christou from Secure Systems Lab at Brown University.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, there is a potential for segfault / denial of service in TensorFlow by calling tf.compat.v1. ops which don't yet have support for quantized types, which was added after migration to TensorFlow 2.x. In these scenarios, since the kernel is missing, a nullptr value is passed to ParseDimensionValue for the pyvalue argument. Then, this is dereferenced, resulting in segfault. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, certain TFLite models that were created using TFLite model converter would crash when loaded in the TFLite interpreter. The culprit is that during quantization the scale of values could be greater than 1 but code was always assuming sub-unit scaling. Thus, since code was calling QuantizeMultiplierSmallerThanOneExp, the TFLITECHECKLT assertion would trigger and abort the process. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, the implementation of tf.rawops.SpaceToBatchND (in all backends such as XLA and handwritten kernels) is vulnerable to an integer overflow: The result of this integer overflow is used to allocate the output tensor, hence we get a denial of service via a CHECK-failure (assertion failure), as in TFSA-2021-198. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, the implementation of tf.rawops.SparseTensorToCSRSparseMatrix does not fully validate the input arguments. This results in a CHECK-failure which can be used to trigger a denial of service attack. The code assumes denseshape is a vector and indices is a matrix (as part of requirements for sparse tensors) but there is no validation for this. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, the implementation of tf.rawops.SparseTensorDenseAdd does not fully validate the input arguments. In this case, a reference gets bound to a nullptr during kernel execution. This is undefined behavior. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, the implementation of tf.rawops.LoadAndRemapMatrix does not fully validate the input arguments. This results in a CHECK-failure which can be used to trigger a denial of service attack. The code assumes initializingvalues is a vector but there is no validation for this before accessing its value. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, the implementation of tf.rawops.TensorSummaryV2 does not fully validate the input arguments. This results in a CHECK-failure which can be used to trigger a denial of service attack. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, the implementation of tf.histogramfixedwidth is vulnerable to a crash when the values array contain Not a Number (NaN) elements. The implementation assumes that all floating point operations are defined and then converts a floating point result to an integer index. If values contains NaN then the result of the division is still NaN and the cast to int32 would result in a crash. This only occurs on the CPU implementation. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, the macros that TensorFlow uses for writing assertions (e.g., CHECKLT, CHECKGT, etc.) have an incorrect logic when comparing sizet and int values. Due to type conversion rules, several of the macros would trigger incorrectly. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, multiple TensorFlow operations misbehave in eager mode when the resource handle provided to them is invalid. In graph mode, it would have been impossible to perform these API calls, but migration to TF 2.x eager mode opened up this vulnerability. If the resource handle is empty, then a reference is bound to a null pointer inside TensorFlow codebase (various codepaths). This is undefined behavior. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.