Impact Remote Server-Side Request Forgery (SSRF) Issue: TorchServe default configuration lacks proper input validation, enabling third parties to invoke remote HTTP download requests and write files to the disk. This issue could be taken advantage of to compromise the integrity of the system and sensitive data. This issue is present in versions 0.1.0 to 0.8.1. Mitigation: The user is able to load the model of their choice from any URL that they would like to use. The user of TorchServe is responsible for configuring both the allowedurls and specifying the model URL to be used. A pull request to warn the user when the default value for allowedurls is used has been merged - https://github.com/pytorch/serve/pull/2534. TorchServe release 0.8.2 includes this change.

Patches

TorchServe release 0.8.2 includes fixes to address the previously listed issue:

https://github.com/pytorch/serve/releases/tag/v0.8.2

Tags for upgraded DLC release User can use the following new image tags to pull DLCs that ship with patched TorchServe version 0.8.2: x86 GPU

v1.9-pt-ec2-2.0.1-inf-gpu-py310 v1.8-pt-sagemaker-2.0.1-inf-gpu-py310

x86 CPU

v1.8-pt-ec2-2.0.1-inf-cpu-py310 v1.7-pt-sagemaker-2.0.1-inf-cpu-py310

Graviton

v1.7-pt-graviton-ec2-2.0.1-inf-cpu-py310 v1.5-pt-graviton-sagemaker-2.0.1-inf-cpu-py310

Neuron

1.13.1-neuron-py310-sdk2.13.2-ubuntu20.04 1.13.1-neuronx-py310-sdk2.13.2-ubuntu20.04 1.13.1-neuronx-py310-sdk2.13.2-ubuntu20.04

The full DLC image URI details can be found at: https://github.com/aws/deep-learning-containers/blob/master/availableimages.md#available-deep-learning-containers-images

References https://github.com/pytorch/serve/blob/b3eced56b4d9d5d3b8597aa506a0bcf954d291bc/docs/configuration.md?plain=1#L296 https://github.com/pytorch/serve/pull/2534 https://github.com/pytorch/serve/releases/tag/v0.8.2 https://github.com/aws/deep-learning-containers/blob/master/availableimages.md#available-deep-learning-containers-images

Credit We would like to thank Oligo Security for responsibly disclosing this issue and working with us on its resolution. If you have any questions or comments about this advisory, we ask that you contact AWS/Amazon Security via our vulnerability reporting page](https://aws.amazon.com/security/vulnerability-reporting)) or directly via email to aws-security@amazon.com. Please do not create a public GitHub issue.

Impact TorchServe's check on allowedurls configuration can be by-passed if the URL contains characters such as ".." but it does not prevent the model from being downloaded into the model store. Once a file is downloaded, it can be referenced without providing a URL the second time, which effectively bypasses the allowedurls security check. Customers using PyTorch inference Deep Learning Containers (DLC) through Amazon SageMaker and EKS are not affected.

Patches This issue in TorchServe has been fixed by validating the URL without characters such as ".." before downloading: #3082.

TorchServe release 0.11.0 includes the fix to address this vulnerability.

References #3082 TorchServe release v0.11.0

Thank Kroll Cyber Risk for for responsibly disclosing this issue.

If you have any questions or comments about this advisory, we ask that you contact AWS Security via our vulnerability reporting page or directly via email to aws-security@amazon.com. Please do not create a public GitHub issue.

Withdrawn Advisory This advisory has been withdrawn because it describes known functionality of PyTorch. This link is maintained to preserve external references.

Original Description A deserialization vulnerability exists in the Pytorch RPC framework (torch.distributed.rpc) in pytorch/pytorch versions <=2.3.1. The vulnerability arises from the lack of security verification during the deserialization process of PythonUDF objects in pytorch/torch/distributed/rpc/internal.py. This flaw allows an attacker to execute arbitrary code remotely by sending a malicious serialized PythonUDF object, leading to remote code execution (RCE) on the master node.

Impact The two gRPC ports 7070 and 7071, are not bound to localhost by default, so when TorchServe is launched, these two interfaces are bound to all interfaces. Customers using PyTorch inference Deep Learning Containers (DLC) through Amazon SageMaker and EKS are not affected.

Patches This issue in TorchServe has been fixed in #3083.

TorchServe release 0.11.0 includes the fix to address this vulnerability.

References #3083 TorchServe release v0.11.0

Thank Kroll Cyber Risk for for responsibly disclosing this issue.

If you have any questions or comments about this advisory, we ask that you contact AWS Security via our vulnerability reporting page or directly via email to aws-security@amazon.com. Please do not create a public GitHub issue.

An issue in the component torch.linalg.lu of pytorch v2.8.0 allows attackers to cause a Denial of Service (DoS) when performing a slice operation.

A Name Error occurs in pytorch v2.7.0 when a PyTorch model consists of torch.cummin and is compiled by Inductor, leading to a Denial of Service (DoS).

A syntax error in the component proxytensor.py of pytorch v2.7.0 allows attackers to cause a Denial of Service (DoS).

pytorch v2.8.0 was discovered to display unexpected behavior when the components torch.rot90 and torch.randnlike are used together.

A buffer overflow occurs in pytorch v2.7.0 when a PyTorch model consists of torch.nn.Conv2d, torch.nn.functional.hardshrink, and torch.Tensor.view-torch.mv() and is compiled by Inductor, leading to a Denial of Service (DoS).

An issue in pytorch v2.7.0 can lead to a Denial of Service (DoS) when a PyTorch model consists of torch.Tensor.tosparse() and torch.Tensor.todense() and is compiled by Inductor.

PyTorch torchvision through 0.28.0, fixed in commit 4e05dc2, contains an out-of-bounds heap read vulnerability in the GIF decoder's readfromtensor callback that passes unclamped length to memcpy. Attackers can supply malicious or truncated GIF files to cause denial of service via segmentation fault or disclose adjacent heap memory contents.

In the latest version of pytorch/serve, the script 'uploadresultstos3.sh' references the S3 bucket 'benchmarkai-metrics-prod' without ensuring its ownership or confirming its accessibility. This could lead to potential security vulnerabilities or unauthorized access to the bucket if it is not properly secured or claimed by the appropriate entity. The issue may result in data breaches, exposure of proprietary information, or unauthorized modifications to stored data.

Pytorch before v2.2.0 has an Out-of-bounds Read vulnerability via the component torch/csrc/jit/mobile/flatbufferloader.cpp.

A vulnerability classified as problematic has been found in PyTorch 2.6.0. Affected is the function torch.jit.jitmodulefromflatbuffer. The manipulation leads to memory corruption. Local access is required to approach this attack. The exploit has been disclosed to the public and may be used.

A vulnerability classified as critical has been found in PyTorch 2.6.0. This affects the function torch.jit.script. The manipulation leads to memory corruption. It is possible to launch the attack on the local host. The exploit has been disclosed to the public and may be used.

A vulnerability classified as critical was found in PyTorch 2.6.0. This vulnerability affects the function torch.lstmcell. The manipulation leads to memory corruption. The attack needs to be approached locally. The exploit has been disclosed to the public and may be used.

In PyTorch through 2.6.0, when eager is used, nn.PairwiseDistance(p=2) produces incorrect results.

In PyTorch before 2.7.0, bitwiserightshift produces incorrect output for certain out-of-bounds values of the "other" argument.

In PyTorch before 2.7.0, when inductor is used, nn.Fold has an assertion error.

In PyTorch before 2.7.0, when torch.compile is used, FractionalMaxPool2d has inconsistent results.

PyTorch before 3.7.0 has a bernoullip decompose function in decompositions.py even though it lacks full consistency with the eager CPU implementation, negatively affecting nn.Dropout1d, nn.Dropout2d, and nn.Dropout3d for fallbackrandom=True.

pytorch v2.8.0 was discovered to contain an integer overflow in the component torch.nantonum-.long().

Impact Using the model/workflow management API, there is a chance of uploading potentially harmful archives that contain files that are extracted to any location on the filesystem that is within the process permissions. Leveraging this issue could aid third-party actors in hiding harmful code in open-source/public models, which can be downloaded from the internet, and take advantage of machines running Torchserve.

Patches The ZipSlip issue in TorchServe has been fixed by validating the paths of files contained within a zip archive before extracting them: https://github.com/pytorch/serve/pull/2634

TorchServe release 0.9.0 includes fixes to address the ZipSlip vulnerability: https://github.com/pytorch/serve/releases/tag/v0.9.0

References https://github.com/pytorch/serve/pull/2634 https://github.com/pytorch/serve/releases/tag/v0.9.0

Credit We would like to thank Oligo Security for responsibly disclosing this issue.

If you have any questions or comments about this advisory, we ask that you contact AWS Security via our vulnerability reporting page or directly via email to aws-security@amazon.com. Please do not create a public GitHub issue.

A vulnerability was found in PyTorch 2.6.0. It has been declared as critical. Affected by this vulnerability is the function torch.nn.utils.rnn.padpackedsequence. The manipulation leads to memory corruption. Local access is required to approach this attack. The exploit has been disclosed to the public and may be used.

A vulnerability was found in PyTorch 2.6.0. It has been rated as critical. Affected by this issue is the function torch.nn.utils.rnn.unpacksequence. The manipulation leads to memory corruption. Attacking locally is a requirement. The exploit has been disclosed to the public and may be used.

A vulnerability, which was classified as problematic, has been found in PyTorch 2.6.0. This issue affects the function torch.cuda.memory.cachingallocatordelete of the file c10/cuda/CUDACachingAllocator.cpp. The manipulation leads to memory corruption. An attack has to be approached locally. The exploit has been disclosed to the public and may be used.

A vulnerability, which was classified as problematic, was found in PyTorch 2.6.0. Affected is the function torch.nn.functional.ctcloss of the file aten/src/ATen/native/LossCTC.cpp. The manipulation leads to denial of service. An attack has to be approached locally. The exploit has been disclosed to the public and may be used. The name of the patch is 46fc5d8e360127361211cb237d5f9eef0223e567. It is recommended to apply a patch to fix this issue.

An issue was discovered in PyTorch v2.5 and v2.7.1. Omission of profiler.stop() can cause torch.profiler.profile (PythonTracer) to crash or hang during finalization, leading to a Denial of Service (DoS).

A vulnerability was found in PyTorch 2.6.0+cu124. It has been rated as problematic. Affected by this issue is the function nnqSigmoid of the component Quantized Sigmoid Module. The manipulation of the argument scale/zeropoint leads to improper initialization. The attack needs to be approached locally. The complexity of an attack is rather high. The exploitation is known to be difficult. The exploit has been disclosed to the public and may be used.

A vulnerability was found in PyTorch 2.6.0+cu124. It has been declared as critical. Affected by this vulnerability is the function torch.ops.profiler.callendcallbacksonjitfut of the component Tuple Handler. The manipulation of the argument None leads to memory corruption. The attack can be launched remotely. The complexity of an attack is rather high. The exploitation appears to be difficult.