vllm-project vllm version 0.6.0 contains a vulnerability in the AsyncEngineRPCServer() RPC server entrypoints. The core functionality run_server_loop() calls the function _make_handler_coro(), which directly uses cloudpickle.loads() on received messages without any sanitization. This can result in remote code execution by deserializing malicious pickle data.

vLLM, an inference and serving engine for large language models (LLMs), has an issue in versions 0.6.5 through 0.8.4 that ONLY impacts environments using the `PyNcclPipe` KV cache transfer integration with the V0 engine. No other configurations are affected. vLLM supports the use of the `PyNcclPipe` class to establish a peer-to-peer communication domain for data transmission between distributed nodes. The GPU-side KV-Cache transmission is implemented through the `PyNcclCommunicator` class, while CPU-side control message passing is handled via the `send_obj` and `recv_obj` methods on the CPU side.​ The intention was that this interface should only be exposed to a private network using the IP address specified by the `--kv-ip` CLI parameter. The vLLM documentation covers how this must be limited to a secured network. The default and intentional behavior from PyTorch is that the `TCPStore` interface listens on ALL interfaces, regardless of what IP address is provided. The IP address given was only used as a client-side address to use. vLLM was fixed to use a workaround to force the `TCPStore` instance to bind its socket to a specified private interface. As of version 0.8.5, vLLM limits the `TCPStore` socket to the private interface as configured.

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Versions starting from 0.6.5 and prior to 0.8.5, having vLLM integration with mooncake, are vulnerable to remote code execution due to using pickle based serialization over unsecured ZeroMQ sockets. The vulnerable sockets were set to listen on all network interfaces, increasing the likelihood that an attacker is able to reach the vulnerable ZeroMQ sockets to carry out an attack. vLLM instances that do not make use of the mooncake integration are not vulnerable. This issue has been patched in version 0.8.5.

vLLM is an inference and serving engine for large language models. In a multi-node vLLM deployment using the V0 engine, vLLM uses ZeroMQ for some multi-node communication purposes. The secondary vLLM hosts open a `SUB` ZeroMQ socket and connect to an `XPUB` socket on the primary vLLM host. When data is received on this `SUB` socket, it is deserialized with `pickle`. This is unsafe, as it can be abused to execute code on a remote machine. Since the vulnerability exists in a client that connects to the primary vLLM host, this vulnerability serves as an escalation point. If the primary vLLM host is compromised, this vulnerability could be used to compromise the rest of the hosts in the vLLM deployment. Attackers could also use other means to exploit the vulnerability without requiring access to the primary vLLM host. One example would be the use of ARP cache poisoning to redirect traffic to a malicious endpoint used to deliver a payload with arbitrary code to execute on the target machine. Note that this issue only affects the V0 engine, which has been off by default since v0.8.0. Further, the issue only applies to a deployment using tensor parallelism across multiple hosts, which we do not expect to be a common deployment pattern. Since V0 is has been off by default since v0.8.0 and the fix is fairly invasive, the maintainers of vLLM have decided not to fix this issue. Instead, the maintainers recommend that users ensure their environment is on a secure network in case this pattern is in use. The V1 engine is not affected by this issue.

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. When vLLM is configured to use Mooncake, unsafe deserialization exposed directly over ZMQ/TCP on all network interfaces will allow attackers to execute remote code on distributed hosts. This is a remote code execution vulnerability impacting any deployments using Mooncake to distribute KV across distributed hosts. This vulnerability is fixed in 0.8.0.

vLLM is a library for LLM inference and serving. vllm/model_executor/weight_utils.py implements hf_model_weights_iterator to load the model checkpoint, which is downloaded from huggingface. It uses the torch.load function and the weights_only parameter defaults to False. When torch.load loads malicious pickle data, it will execute arbitrary code during unpickling. This vulnerability is fixed in v0.7.0.

The REST API in CyberArk Password Vault Web Access before 9.9.5 and 10.x before 10.1 allows remote attackers to execute arbitrary code via a serialized .NET object in an Authorization HTTP header.

In Apache Solr versions 5.0.0 to 5.5.5 and 6.0.0 to 6.6.5, the Config API allows to configure the JMX server via an HTTP POST request. By pointing it to a malicious RMI server, an attacker could take advantage of Solr's unsafe deserialization to trigger remote code execution on the Solr side.

In Apache Batik 1.x before 1.10, when deserializing subclass of `AbstractDocument`, the class takes a string from the inputStream as the class name which then use it to call the no-arg constructor of the class. Fix was to check the class type before calling newInstance in deserialization.

Versions of Superset prior to 0.23 used an unsafe load method from the pickle library to deserialize data leading to possible remote code execution. Note Superset 0.23 was released prior to any Superset release under the Apache Software Foundation.

Unauthenticated RCE is possible when JMeter is used in distributed mode (-r or -R command line options). Attacker can establish a RMI connection to a jmeter-server using RemoteJMeterEngine and proceed with an attack using untrusted data deserialization. This only affect tests running in Distributed mode. Note that versions before 4.0 are not able to encrypt traffic between the nodes, nor authenticate the participating nodes so upgrade to JMeter 5.1 is also advised.

In Apache Ignite before 2.4.8 and 2.5.x before 2.5.3, the serialization mechanism does not have a list of classes allowed for serialization/deserialization, which makes it possible to run arbitrary code when 3-rd party vulnerable classes are present in Ignite classpath. The vulnerability can be exploited if the one sends a specially prepared form of a serialized object to GridClientJdkMarshaller deserialization endpoint.

Mescius ActiveReports.NET TypeResolutionService Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Mescius ActiveReports.NET. Interaction with this library is required to exploit this vulnerability but attack vectors may vary depending on the implementation. The specific flaw exists within the TypeResolutionService class. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of the current process. Was ZDI-CAN-25397.

Mescius ActiveReports.NET ReadValue Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Mescius ActiveReports.NET. Interaction with this library is required to exploit this vulnerability but attack vectors may vary depending on the implementation. The specific flaw exists within the implementation of the ReadValue method. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of the current process. Was ZDI-CAN-25246.

Buck parser-cache command loads/saves state using Java serialized object. If the state information is maliciously crafted, deserializing it could lead to code execution. This issue affects Buck versions prior to v2018.06.25.01.

The EjbObjectInputStream class in Apache TomEE before 1.7.4 and 7.x before 7.0.0-M3 allows remote attackers to execute arbitrary code via a crafted serialized object.

FasterXML jackson-databind before 2.7.9.3, 2.8.x before 2.8.11.1 and 2.9.x before 2.9.5 allows unauthenticated remote code execution because of an incomplete fix for the CVE-2017-7525 deserialization flaw. This is exploitable by sending maliciously crafted JSON input to the readValue method of the ObjectMapper, bypassing a blacklist that is ineffective if the c3p0 libraries are available in the classpath.

IBM Websphere MQ JMS 7.0.1, 7.1, 7.5, 8.0, and 9.0 client provides classes that deserialize objects from untrusted sources which could allow a malicious user to execute arbitrary Java code by adding vulnerable classes to the classpath. IBM Reference #: 1983457.

The newsletters-lite plugin before 4.6.8.6 for WordPress has PHP object injection.

DomPDF before version 2.0.0 is vulnerable to PHAR deserialization due to a lack of checking on the protocol before passing it into the file_get_contents() function. An attacker who can upload files of any type to the server can pass in the phar:// protocol to unserialize the uploaded file and instantiate arbitrary PHP objects. This can lead to remote code execution, especially when DOMPdf is used with frameworks with documented POP chains like Laravel or vulnerable developer code.

In Pydio before 8.2.2, an attack is possible via PHP Object Injection because a user is allowed to use the $phpserial$a:0:{} syntax to store a preference. An attacker either needs a "public link" of a file, or access to any unprivileged user account for creation of such a link.

A remote code execution vulnerability exists in invoke-ai/invokeai versions 5.3.1 through 5.4.2 via the /api/v2/models/install API. The vulnerability arises from unsafe deserialization of model files using torch.load without proper validation. Attackers can exploit this by embedding malicious code in model files, which is executed upon loading. This issue is fixed in version 5.4.3.

A vulnerability in infiniflow/ragflow versions v0.12.0 allows for remote code execution. The RPC server in RagFlow uses a hard-coded AuthKey 'authkey=b'infiniflow-token4kevinhu'' which can be easily fetched by attackers to join the group communication without restrictions. Additionally, the server processes incoming data using pickle deserialization via `pickle.loads()` on `connection.recv()`, making it vulnerable to remote code execution. This issue is fixed in version 0.14.0.

A remote code execution vulnerability exists in open-mmlab/mmdetection version v3.3.0. The vulnerability is due to the use of the `pickle.loads()` function in the `all_reduce_dict()` distributed training API without proper sanitization. This allows an attacker to execute arbitrary code by broadcasting a malicious payload to the distributed training network.

SAS Web Infrastructure Platform before 9.4M6 allows remote attackers to execute arbitrary code via a Java deserialization variant.

A vulnerability in the h2oai/h2o-3 REST API versions 3.46.0.4 allows unauthenticated remote attackers to execute arbitrary code via deserialization of untrusted data. The vulnerability exists in the endpoints POST /99/ImportSQLTable and POST /3/SaveToHiveTable, where user-controlled JDBC URLs are passed to DriverManager.getConnection, leading to deserialization if a MySQL or PostgreSQL driver is available in the classpath. This issue is fixed in version 3.47.0.

Red Hat JBoss A-MQ 6.x; BPM Suite (BPMS) 6.x; BRMS 6.x and 5.x; Data Grid (JDG) 6.x; Data Virtualization (JDV) 6.x and 5.x; Enterprise Application Platform 6.x, 5.x, and 4.3.x; Fuse 6.x; Fuse Service Works (FSW) 6.x; Operations Network (JBoss ON) 3.x; Portal 6.x; SOA Platform (SOA-P) 5.x; Web Server (JWS) 3.x; Red Hat OpenShift/xPAAS 3.x; and Red Hat Subscription Asset Manager 1.3 allow remote attackers to execute arbitrary commands via a crafted serialized Java object, related to the Apache Commons Collections (ACC) library.

The patreon-connect plugin before 1.2.2 for WordPress has Object Injection.

FasterXML jackson-databind 2.x before 2.9.8 might allow attackers to have unspecified impact by leveraging failure to block the jboss-common-core class from polymorphic deserialization.

IBM WebSphere Application Server Liberty OpenID Connect could allow a remote attacker to execute arbitrary code on the system, caused by improper deserialization. By sending a specially-crafted request to the RP service, an attacker could exploit this vulnerability to execute arbitrary code. IBM X-Force ID: 150999.

Vulnerability in the Oracle WebLogic Server component of Oracle Fusion Middleware (subcomponent: WLS Core Components). Supported versions that are affected are 10.3.6.0, 12.1.3.0 and 12.2.1.3. Easily exploitable vulnerability allows unauthenticated attacker with network access via T3 to compromise Oracle WebLogic Server. Successful attacks of this vulnerability can result in takeover of Oracle WebLogic Server. CVSS 3.0 Base Score 9.8 (Confidentiality, Integrity and Availability impacts). CVSS Vector: (CVSS:3.0/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H).

Allegra loadFieldMatch Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Allegra. Although authentication is required to exploit this vulnerability, product implements a registration mechanism that can be used to create a user with a sufficient privilege level. The specific flaw exists within the loadFieldMatch method. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of LOCAL SERVICE. Was ZDI-CAN-22506.

Allegra renderFieldMatch Deserialization of Unstrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Allegra. Although authentication is required to exploit this vulnerability, product implements a registration mechanism that can be used to create a user with a sufficient privilege level. The specific flaw exists within the renderFieldMatch method. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of LOCAL SERVICE. Was ZDI-CAN-22505.

Voltronic Power ViewPower Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Voltronic Power ViewPower. Authentication is not required to exploit this vulnerability. The specific flaw exists within the RMI interface, which listens on TCP port 51099 by default. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of SYSTEM. Was ZDI-CAN-22012.

Voltronic Power ViewPower Pro Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Voltronic Power ViewPower Pro. Authentication is not required to exploit this vulnerability. The specific flaw exists within the RMI interface, which listens on TCP port 41009 by default. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of SYSTEM. Was ZDI-CAN-21012.

The MyScript SDK before 1.3 for Android might allow attackers to execute arbitrary code by leveraging a finalize method in a Serializable class that improperly passes an attacker-controlled pointer to a native function.

An issue was discovered in TCPDF before 6.2.22. Attackers can trigger deserialization of arbitrary data via the phar:// wrapper.

Adobe ColdFusion versions July 12 release (2018.0.0.310739), Update 6 and earlier, and Update 14 and earlier have a deserialization of untrusted data vulnerability. Successful exploitation could lead to arbitrary code execution.

Adobe ColdFusion versions July 12 release (2018.0.0.310739), Update 6 and earlier, and Update 14 and earlier have a deserialization of untrusted data vulnerability. Successful exploitation could lead to arbitrary code execution.

A Java deserialization vulnerability in Cisco Unity Express (CUE) could allow an unauthenticated, remote attacker to execute arbitrary shell commands with the privileges of the root user. The vulnerability is due to insecure deserialization of user-supplied content by the affected software. An attacker could exploit this vulnerability by sending a malicious serialized Java object to the listening Java Remote Method Invocation (RMI) service. A successful exploit could allow the attacker to execute arbitrary commands on the device with root privileges.

Insecure deserialization of a specially crafted serialized object, in CA Release Automation 6.5 and earlier, allows attackers to potentially execute arbitrary code.

Adobe ColdFusion versions July 12 release (2018.0.0.310739), Update 6 and earlier, and Update 14 and earlier have a deserialization of untrusted data vulnerability. Successful exploitation could lead to arbitrary code execution.

An issue was discovered in EthereumJ 1.8.2. There is Unsafe Deserialization in ois.readObject in mine/Ethash.java and decoder.readObject in crypto/ECKey.java. When a node syncs and mines a new block, arbitrary OS commands can be run on the server.

This vulnerability allows remote attackers to execute arbitrary code on affected installations of SolarWinds Network Performance Monitor 2020.2.1. Authentication is not required to exploit this vulnerability. The specific flaw exists within the SolarWinds.Serialization library. The issue results from the lack of proper validation of user-supplied data, which can result in deserialization of untrusted data. An attacker can leverage this vulnerability to execute code in the context of SYSTEM. Was ZDI-CAN-12213.

IBM WebSphere Application Server 7.0, 8.0, 8.5, and 9.0 could allow remote attackers to execute arbitrary Java code through an administrative client class with a serialized object from untrusted sources. IBM X-Force ID: 152533.

In WordPress before 4.9.9 and 5.x before 5.0.1, contributors could conduct PHP object injection attacks via crafted metadata in a wp.getMediaItem XMLRPC call. This is caused by mishandling of serialized data at phar:// URLs in the wp_get_attachment_thumb_file function in wp-includes/post.php.

FasterXML jackson-databind 2.x before 2.9.8 might allow attackers to have unspecified impact by leveraging failure to block the openjpa class from polymorphic deserialization.

FasterXML jackson-databind 2.x before 2.9.8 might allow attackers to have unspecified impact by leveraging failure to block the axis2-transport-jms class from polymorphic deserialization.

IBM Partner Engagement Manager 2.0 could allow a remote attacker to execute arbitrary code on the system, caused by an unsafe deserialization flaw. By sending specially-crafted data, an attacker could exploit this vulnerability to execute arbitrary code on the system. IBM X-Force ID: 203091.

In Apache Ignite 2.3 or earlier, the serialization mechanism does not have a list of classes allowed for serialization/deserialization, which makes it possible to run arbitrary code when 3-rd party vulnerable classes are present in Ignite classpath. The vulnerability can be exploited if the one sends a specially prepared form of a serialized object to one of the deserialization endpoints of some Ignite components - discovery SPI, Ignite persistence, Memcached endpoint, socket steamer.