vllm-project vllm version v0.6.2 contains a vulnerability in the MessageQueue.dequeue() API function. The function uses pickle.loads to parse received sockets directly, leading to a remote code execution vulnerability. An attacker can exploit this by sending a malicious payload to the MessageQueue, causing the victim's machine to execute arbitrary code.

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, 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 an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.

vLLM is an inference and serving engine for large language models (LLMs). From 0.8.3 to before 0.14.1, when an invalid image is sent to vLLM's multimodal endpoint, PIL throws an error. vLLM returns this error to the client, leaking a heap address. With this leak, we reduce ASLR from 4 billion guesses to ~8 guesses. This vulnerability can be chained a heap overflow with JPEG2000 decoder in OpenCV/FFmpeg to achieve remote code execution. This vulnerability is fixed in 0.14.1.

vLLM is an inference and serving engine for large language models (LLMs). From versions 0.10.2 to before 0.11.1, a memory corruption vulnerability could lead to a crash (denial-of-service) and potentially remote code execution (RCE), exists in the Completions API endpoint. When processing user-supplied prompt embeddings, the endpoint loads serialized tensors using torch.load() without sufficient validation. Due to a change introduced in PyTorch 2.8.0, sparse tensor integrity checks are disabled by default. As a result, maliciously crafted tensors can bypass internal bounds checks and trigger an out-of-bounds memory write during the call to to_dense(). This memory corruption can crash vLLM and potentially lead to code execution on the server hosting vLLM. This issue has been patched in version 0.11.1.

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.

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.

D-Link DIR-846W A1 FW100A43 was discovered to contain a remote command execution (RCE) vulnerability via the tomography_ping_address parameter in /HNAP1/ interface.

In applications using jfinal 4.9.08 and below, there is a deserialization vulnerability when using redis,may be vulnerable to remote code execute

The WLS Security component in Oracle WebLogic Server 10.3.6.0, 12.1.2.0, 12.1.3.0, and 12.2.1.0 allows remote attackers to execute arbitrary commands via a crafted serialized Java object in T3 protocol traffic to TCP port 7001, related to oracle_common/modules/com.bea.core.apache.commons.collections.jar. NOTE: the scope of this CVE is limited to the WebLogic Server product.

A vulnerability in the web-based management allows an unauthenticated remote attacker to inject arbitrary system commands and gain full system control. Those commands are executed with root privileges. The vulnerability is located in the user request handling of the web-based management.

An unauthenticated command-execution vulnerability exists in TerraMaster TOS through 4.2.06 via shell metacharacters in the Event parameter in include/makecvs.php during CSV creation.

KLog Server 2.4.1 allows OS command injection via shell metacharacters in the actions/authenticate.php user parameter.

QSAN SANOS factory reset function does not filter special parameters. Remote attackers can use this vulnerability to inject and execute arbitrary commands without permissions. The referred vulnerability has been solved with the updated version of QSAN SANOS v2.1.0.

The QSAN SANOS setting page does not filter special parameters. Remote attackers can use this vulnerability to inject and execute arbitrary commands without permissions. The referred vulnerability has been solved with the updated version of QSAN SANOS v2.1.0.

tdpServer on TP-Link Archer A7 AC1750 devices before 201029 allows remote attackers to execute arbitrary code via the slave_mac parameter. NOTE: this issue exists because of an incomplete fix for CVE-2020-10882 in which shell quotes are mishandled.

TinyWeb is a web server (HTTP, HTTPS) written in Delphi for Win32. TinyWeb HTTP Server before version 1.98 is vulnerable to OS command injection via CGI ISINDEX-style query parameters. The query parameters are passed as command-line arguments to the CGI executable via Windows CreateProcess(). An unauthenticated remote attacker can execute arbitrary commands on the server by injecting Windows shell metacharacters into HTTP requests. This vulnerability is fixed in 1.98.

All versions of package geojson2kml are vulnerable to Command Injection via the index.js file. PoC: var a =require("geojson2kml"); a("./","& touch JHU",function(){})

Dasan Networks - W-Web versions 1.22-1.27 - CWE-78: Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection')

Laravel Reverb provides a real-time WebSocket communication backend for Laravel applications. In versions 1.6.3 and below, Reverb passes data from the Redis channel directly into PHP’s unserialize() function without restricting which classes can be instantiated, which leaves users vulnerable to Remote Code Execution. The exploitability of this vulnerability is increased because Redis servers are commonly deployed without authentication, but only affects Laravel Reverb when horizontal scaling is enabled (REVERB_SCALING_ENABLED=true). This issue has been fixed in version 1.7.0. As a workaround, require a strong password for Redis access and ensure the service is only accessible via a private network or local loopback, and/or set REVERB_SCALING_ENABLED=false to bypass the vulnerable logic entirely (if the environment uses only one Reverb node).

Zavio CF7500, CF7300, CF7201, CF7501, CB3211, CB3212, CB5220, CB6231, B8520, B8220, and CD321 IP Cameras with firmware version M2.1.6.05 has a command injection vulnerability in their implementation of their binaries and handling of network requests.

Deserialization of untrusted data in Azure SDK allows an unauthorized attacker to execute code over a network.

There are command injection vulnerabilities in the underlying Soft AP Daemon service that could lead to unauthenticated remote code execution by sending specially crafted packets destined to the PAPI (Aruba's Access Point management protocol) UDP port (8211). Successful exploitation of these vulnerabilities result in the ability to execute arbitrary code as a privileged user on the underlying operating system.

Tenda AC6 US_AC6V1.0BR_V15.03.05.16_multi_TD01.bin function 'sub_3A1D0' contains a command execution vulnerability.

LG Simple Editor readVideoInfo Command Injection Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of LG Simple Editor. Authentication is not required to exploit this vulnerability. The specific flaw exists within the readVideoInfo method. The issue results from the lack of proper validation of a user-supplied string before using it to execute a system call. An attacker can leverage this vulnerability to execute code in the context of SYSTEM. . Was ZDI-CAN-19953.

WordPress before 5.5.2 mishandles deserialization requests in wp-includes/Requests/Utility/FilteredIterator.php.

knplabs/knp-snappy is a PHP library allowing thumbnail, snapshot or PDF generation from a url or a html page. ## Issue On March 17th the vulnerability CVE-2023-28115 was disclosed, allowing an attacker to gain remote code execution through PHAR deserialization. Version 1.4.2 added a check `if (\strpos($filename, 'phar://') === 0)` in the `prepareOutput` function to resolve this CVE, however if the user is able to control the second parameter of the `generateFromHtml()` function of Snappy, it will then be passed as the `$filename` parameter in the `prepareOutput()` function. In the original vulnerability, a file name with a `phar://` wrapper could be sent to the `fileExists()` function, equivalent to the `file_exists()` PHP function. This allowed users to trigger a deserialization on arbitrary PHAR files. To fix this issue, the string is now passed to the `strpos()` function and if it starts with `phar://`, an exception is raised. However, PHP wrappers being case insensitive, this patch can be bypassed using `PHAR://` instead of `phar://`. A successful exploitation of this vulnerability allows executing arbitrary code and accessing the underlying filesystem. The attacker must be able to upload a file and the server must be running a PHP version prior to 8. This issue has been addressed in commit `d3b742d61a` which has been included in version 1.4.3. Users are advised to upgrade. Users unable to upgrade should ensure that only trusted users may submit data to the `AbstractGenerator->generate(...)` function.

Tenda AC6 US_AC6V1.0BR_V15.03.05.16_multi_TD01.bin function 'sub_ADD50' contains a command execution vulnerability. In the "formSetIptv" function, obtaining the "list" and "vlanId" fields, unfiltered passing these two fields as parameters to the "sub_ADD50" function to execute commands.

phpPgAdmin 7.14.4 and earlier is vulnerable to deserialization of untrusted data which may lead to remote code execution because user-controlled data is directly passed to the PHP 'unserialize()' function in multiple places. An example is the functionality to manage tables in 'tables.php' where the 'ma[]' POST parameter is deserialized.

An issue was discovered on CDATA 72408A, 9008A, 9016A, 92408A, 92416A, 9288, 97016, 97024P, 97028P, 97042P, 97084P, 97168P, FD1002S, FD1104, FD1104B, FD1104S, FD1104SN, FD1108S, FD1204S-R2, FD1204SN, FD1204SN-R2, FD1208S-R2, FD1216S-R1, FD1608GS, FD1608SN, FD1616GS, FD1616SN, and FD8000 devices. One can escape from a shell and acquire root privileges by leveraging the TFTP download configuration.

There is a command injection vulnerability in the underlying Central Communications service that could lead to unauthenticated remote code execution by sending specially crafted packets destined to the PAPI (Aruba's Access Point management protocol) UDP port (8211). Successful exploitation of this vulnerability results in the ability to execute arbitrary code as a privileged user on the underlying operating system.

LG Simple Editor createThumbnailByMovie Command Injection Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of LG Simple Editor. Authentication is not required to exploit this vulnerability. The specific flaw exists within the createThumbnailByMovie method. The issue results from the lack of proper validation of a user-supplied string before using it to execute a system call. An attacker can leverage this vulnerability to execute code in the context of SYSTEM. . Was ZDI-CAN-19978.

All versions of package corenlp-js-interface are vulnerable to Command Injection via the main function.

SmartNode SN200 (aka SN200) 3.21.2-23021 allows unauthenticated OS Command Injection.

An issue was discovered in the add function in Shenzhim AAPTJS 1.3.1 which allows attackers to execute arbitrary code via the filePath parameter.

weblogic-framework is a tool for detecting weblogic vulnerabilities. Versions 0.2.3 and prior do not verify the returned data packets, and there is a deserialization vulnerability which may lead to remote code execution. When weblogic-framework gets the command echo, it directly deserializes the data returned by the server without verifying it. At the same time, the classloader loads a lot of deserialization calls. In this case, the malicious serialized data returned by the server will cause remote code execution. Version 0.2.4 contains a patch for this issue.

Multiple vulnerabilities in the Java deserialization function that is used by Cisco Security Manager could allow an unauthenticated, remote attacker to execute arbitrary commands on an affected device. These vulnerabilities are due to insecure deserialization of user-supplied content by the affected software. An attacker could exploit these vulnerabilities by sending a malicious serialized Java object to a specific listener on an affected system. A successful exploit could allow the attacker to execute arbitrary commands on the device with the privileges of NT AUTHORITY\SYSTEM on the Windows target host. Cisco has not released software updates that address these vulnerabilities.

systeminformation is an npm package that provides system and OS information library for node.js. In systeminformation before version 4.26.2 there is a command injection vulnerability. Problem was fixed in version 4.26.2 with a shell string sanitation fix.

Spreecommerce versions prior to 0.50.x contain a remote command execution vulnerability in the API's search functionality. Improper input sanitation allows attackers to inject arbitrary shell commands via the search[instance_eval] parameter, which is dynamically invoked using Ruby’s send method. This flaw enables unauthenticated attackers to execute commands on the server.

Addressed remote code execution vulnerability in cgi_api.php that allowed escalation of privileges in Western Digital My Cloud NAS devices prior to 5.04.114.

Zoho ManageEngine OpManager before 12.5.329 allows unauthenticated Remote Code Execution due to a general bypass in the deserialization class.

Command Injection in PPGo_Jobs v2.8.0 allows remote attackers to execute arbitrary code via the 'AjaxRun()' function.

find-exec is a utility to discover available shell commands. Versions prior to 1.0.3 did not properly escape user input and are vulnerable to Command Injection via an attacker controlled parameter. As a result, attackers may run malicious shell commands in the context of the running process. This issue has been addressed in version 1.0.3. users are advised to upgrade. Users unable to upgrade should ensure that all input passed to find-exec comes from a trusted source.

Improper Authentication vulnerability in Genians Genian NAC V4.0, Genians Genian NAC V5.0, Genians Genian NAC Suite V5.0, Genians Genian ZTNA allows Authentication Abuse.This issue affects Genian NAC V4.0: from V4.0.0 through V4.0.155; Genian NAC V5.0: from V5.0.0 through V5.0.42 (Revision 117460); Genian NAC Suite V5.0: from V5.0.0 through V5.0.54; Genian ZTNA: from V6.0.0 through V6.0.15.

The affected Cognex product, the In-Sight OPC Server versions v5.7.4 (96) and prior, deserializes untrusted data, which could allow a remote attacker access to system level permission commands and local privilege escalation.

Tenda AC6 US_AC6V1.0BR_V15.03.05.16_multi_TD01.bin function 'sub_ADF3C' contains a command execution vulnerability. In the "formSetIptv" function, obtaining the "list" and "vlanId" fields, unfiltered passing these two fields as parameters to the "sub_ADF3C" function to execute commands.

An issue was discovered on Western Digital My Cloud NAS devices before 5.04.114. They allow remote code execution with resultant escalation of privileges.

A vulnerability was identified in Sangfor Operation and Maintenance Management System up to 3.0.12. Affected by this issue is the function SessionController of the file /isomp-protocol/protocol/session of the component SSH Protocol Handler. The manipulation of the argument keypassword leads to os command injection. It is possible to initiate the attack remotely. The exploit is publicly available and might be used. The vendor was contacted early about this disclosure but did not respond in any way.