CVEs

KubeAI is an AI inference operator for kubernetes. Prior to 0.23.2, the ollamaStartupProbeScript() function in internal/modelcontroller/engine_ollama.go constructs a shell command string using fmt.Sprintf with unsanitized model URL components (ref, modelParam). This shell command is executed via bash -c as a Kubernetes startup probe. An attacker who can create or update Model custom resources can inject arbitrary shell commands that execute inside model server pods. This vulnerability is fixed in 0.23.2.

Electron is a framework for writing cross-platform desktop applications using JavaScript, HTML and CSS. From 33.0.0-alpha.1 to before 39.8.5, 40.8.5, 41.1.0, and 42.0.0-alpha.5, apps that use offscreen rendering with GPU shared textures may be vulnerable to a use-after-free. Under certain conditions, the release() callback provided on a paint event texture can outlive its backing native state, and invoking it after that point dereferences freed memory in the main process, which may lead to a crash or memory corruption. Apps are only affected if they use offscreen rendering with webPreferences.offscreen: { useSharedTexture: true }. Apps that do not enable shared-texture offscreen rendering are not affected. To mitigate this issue, ensure texture.release() is called promptly after the texture has been consumed, before the texture object becomes unreachable. This vulnerability is fixed in 39.8.5, 40.8.5, 41.1.0, and 42.0.0-alpha.5.

vLLM is an inference and serving engine for large language models (LLMs). From 0.1.0 to before 0.19.0, a Denial of Service vulnerability exists in the vLLM OpenAI-compatible API server. Due to the lack of an upper bound validation on the n parameter in the ChatCompletionRequest and CompletionRequest Pydantic models, an unauthenticated attacker can send a single HTTP request with an astronomically large n value. This completely blocks the Python asyncio event loop and causes immediate Out-Of-Memory crashes by allocating millions of request object copies in the heap before the request even reaches the scheduling queue. This vulnerability is fixed in 0.19.0.

vLLM is an inference and serving engine for large language models (LLMs). From 0.7.0 to before 0.19.0, the VideoMediaIO.load_base64() method at vllm/multimodal/media/video.py splits video/jpeg data URLs by comma to extract individual JPEG frames, but does not enforce a frame count limit. The num_frames parameter (default: 32), which is enforced by the load_bytes() code path, is completely bypassed in the video/jpeg base64 path. An attacker can send a single API request containing thousands of comma-separated base64-encoded JPEG frames, causing the server to decode all frames into memory and crash with OOM. This vulnerability is fixed in 0.19.0.

vLLM is an inference and serving engine for large language models (LLMs). From 0.16.0 to before 0.19.0, a server-side request forgery (SSRF) vulnerability in download_bytes_from_url allows any actor who can control batch input JSON to make the vLLM batch runner issue arbitrary HTTP/HTTPS requests from the server, without any URL validation or domain restrictions. This can be used to target internal services (e.g. cloud metadata endpoints or internal HTTP APIs) reachable from the vLLM host. This vulnerability is fixed in 0.19.0.

Memory Corruption when handling power management requests with improperly sized input/output buffers.

Transient DOS when receiving a service data frame with excessive length during device matching over a neighborhood awareness network protocol connection.

Memory Corruption when using deprecated DMABUF IOCTL calls to manage video memory.

Memory Corruption when accessing an output buffer without validating its size during IOCTL processing in a camera sensor driver.

Memory Corruption when accessing an output buffer without validating its size during IOCTL processing in a camera sensor driver.

Memory Corruption when accessing an output buffer without validating its size during IOCTL processing.

Memory Corruption when processing auxiliary sensor input/output control commands with insufficient buffer size validation.

Memory Corruption when accessing an output buffer without validating its size during IOCTL processing.

Memory Corruption when sending IOCTL requests with invalid buffer sizes during memcpy operations.

Memory Corruption when retrieving output buffer with insufficient size validation.

Transient DOS when processing nonstandard FILS Discovery Frames with out-of-range action sizes during initial scans.

Cryptographic issue while copying data to a destination buffer without validating its size.

Memory corruption when decoding corrupted satellite data files with invalid signature offsets.

Memory corruption while processing a frame request from user.

Memory corruption while preprocessing IOCTL request in JPEG driver.

Memory corruption when buffer copy operation fails due to integer overflow during attestation report generation.

Memory Corruption when accessing freed memory due to concurrent fence deregistration and signal handling.

Twitch Studio version 0.114.8 and prior contain a privilege escalation vulnerability in its privileged helper tool that allows local attackers to execute arbitrary code as root by exploiting an unprotected XPC service. Attackers can invoke the installFromPath:toPath:withReply: method to overwrite system files and privileged binaries, achieving full system compromise. Twitch Studio was discontinued in May 2024.

OpenEXR provides the specification and reference implementation of the EXR file format, an image storage format for the motion picture industry. From 3.2.0 to before 3.2.7, 3.3.9, and 3.4.9, the DWA lossy decoder constructs temporary per-component block pointers using signed 32-bit arithmetic. For a large enough width, the calculation overflows and later decoder stores operate on a wrapped pointer outside the allocated rowBlock backing store. This vulnerability is fixed in 3.2.7, 3.3.9, and 3.4.9.

OpenEXR provides the specification and reference implementation of the EXR file format, an image storage format for the motion picture industry. From 3.1.0 to before 3.2.7, 3.3.9, and 3.4.9, internal_exr_undo_piz() advances the working wavelet pointer with signed 32-bit arithmetic. Because nx, ny, and wcount are int, a crafted EXR file can make this product overflow and wrap. The next channel then decodes from an incorrect address. The wavelet decode path operates in place, so this yields both out-of-bounds reads and out-of-bounds writes. This vulnerability is fixed in 3.2.7, 3.3.9, and 3.4.9.

Lupa integrates the runtimes of Lua or LuaJIT2 into CPython. In 2.6 and earlier, attribute_filter is not consistently applied when attributes are accessed through built-in functions like getattr and setattr. This allows an attacker to bypass the intended restrictions and eventually achieve arbitrary code execution.

A vulnerability was detected in code-projects Online FIR System 1.0. Affected by this issue is some unknown functionality of the file /complaints.sql of the component SQL Database Backup File Handler. The manipulation results in insecure storage of sensitive information. The attack may be performed from remote. The exploit is now public and may be used.

Pi-hole Admin Interface is a web interface for managing Pi-hole, a network-level ad and internet tracker blocking application. From 6.0 to before 6.5, the formatInfo() function in queries.js renders data.upstream, data.client.ip, and data.ede.text into HTML without escaping when a user expands a query row in the Query Log, enabling stored HTML injection. JavaScript execution is blocked by the server's CSP (script-src 'self'). The same fields are properly escaped in the table view (rowCallback), confirming the omission was an oversight. This vulnerability is fixed in 6.5.

OpenEXR provides the specification and reference implementation of the EXR file format, an image storage format for the motion picture industry. From 3.2.0 to before 3.2.7, 3.3.9, and 3.4.9, a signed integer overflow exists in undo_pxr24_impl() in src/lib/OpenEXRCore/internal_pxr24.c at line 377. The expression (uint64_t)(w * 3) computes w * 3 as a signed 32-bit integer before casting to uint64_t. When w is large, this multiplication constitutes undefined behavior under the C standard. On tested builds (clang/gcc without sanitizers), two's-complement wraparound commonly occurs, and for specific values of w the wrapped result is a small positive integer, which may allow the subsequent bounds check to pass incorrectly. If the check is bypassed, the decoding loop proceeds to write pixel data through dout, potentially extending far beyond the allocated output buffer. This vulnerability is fixed in 3.2.7, 3.3.9, and 3.4.9.

OpenEXR provides the specification and reference implementation of the EXR file format, an image storage format for the motion picture industry. From 3.2.0 to before 3.2.7, 3.3.9, and 3.4.9, a misaligned memory write vulnerability exists in LossyDctDecoder_execute() in src/lib/OpenEXRCore/internal_dwa_decoder.h:749. When decoding a DWA or DWAB-compressed EXR file containing a FLOAT-type channel, the decoder performs an in-place HALF→FLOAT conversion by casting an unaligned uint8_t * row pointer to float * and writing through it. Because the row buffer may not be 4-byte aligned, this constitutes undefined behavior under the C standard and crashes immediately on architectures that enforce alignment (ARM, RISC-V, etc.). On x86 it is silently tolerated at runtime but remains exploitable via compiler optimizations that assume aligned access. This vulnerability is fixed in 3.2.7, 3.3.9, and 3.4.9.

OpenEXR provides the specification and reference implementation of the EXR file format, an image storage format for the motion picture industry. From 3.4.0 to before 3.4.9, a missing bounds check on the dataWindow attribute in EXR file headers allows an attacker to trigger a signed integer overflow in generic_unpack(). By setting dataWindow.min.x to a large negative value, OpenEXRCore computes an enormous image width, which is later used in a signed integer multiplication that overflows, causing the process to terminate with SIGILL via UBSan. This vulnerability is fixed in 3.4.9.

A flaw was found in tar. A remote attacker could exploit this vulnerability by crafting a malicious archive, leading to hidden file injection with fully attacker-controlled content. This bypasses pre-extraction inspection mechanisms, potentially allowing an attacker to introduce malicious files onto a system without detection.

Vim is an open source, command line text editor. Prior to version 9.2.0276, a modeline sandbox bypass in Vim allows arbitrary OS command execution when a user opens a crafted file. The `complete`, `guitabtooltip` and `printheader` options are missing the `P_MLE` flag, allowing a modeline to be executed. Additionally, the `mapset()` function lacks a `check_secure()` call, allowing it to be abused from sandboxed expressions. Commit 9.2.0276 fixes the issue.

A security vulnerability has been detected in code-projects Online FIR System 1.0. Affected by this vulnerability is an unknown functionality of the file /Login/checklogin.php of the component Login. The manipulation of the argument email/password leads to sql injection. The attack is possible to be carried out remotely. The exploit has been disclosed publicly and may be used.

SandboxJS is a JavaScript sandboxing library. Prior to 0.8.36, a scope modification vulnerability exists in @nyariv/sandboxjs. The vulnerability allows untrusted sandboxed code to leak internal interpreter objects through the new operator, exposing sandbox scope objects in the scope hierarchy to untrusted code; an unexpected and undesired exploit. While this could allow modifying scopes inside the sandbox, code evaluation remains sandboxed and prototypes remain protected throughout the execution. This vulnerability is fixed in 0.8.36.

SandboxJS is a JavaScript sandboxing library. Prior to 0.8.36, the @nyariv/sandboxjs parser contains unbounded recursion in the restOfExp function and the lispify/lispifyExpr call chain. An attacker can crash any Node.js process that parses untrusted input by supplying deeply nested expressions (e.g., ~2000 nested parentheses), causing a RangeError: Maximum call stack size exceeded that terminates the process. This vulnerability is fixed in 0.8.36.

SandboxJS is a JavaScript sandboxing library. Prior to 0.8.36, SandboxJS blocks direct assignment to global objects (for example Math.random = ...), but this protection can be bypassed through an exposed callable constructor path: this.constructor.call(target, attackerObject). Because this.constructor resolves to the internal SandboxGlobal function and Function.prototype.call is allowed, attacker code can write arbitrary properties into host global objects and persist those mutations across sandbox instances in the same process. This vulnerability is fixed in 0.8.36.

Fedify is a TypeScript library for building federated server apps powered by ActivityPub. Prior to 1.9.6, 1.10.5, 2.0.8, and 2.1.1, @fedify/fedify follows HTTP redirects recursively in its remote document loader and authenticated document loader without enforcing a maximum redirect count or visited-URL loop detection. An attacker who controls a remote ActivityPub key or actor URL can force a server using Fedify to make repeated outbound requests from a single inbound request, leading to resource consumption and denial of service. This vulnerability is fixed in 1.9.6, 1.10.5, 2.0.8, and 2.1.1.

Pi-hole is a Linux network-level advertisement and Internet tracker blocking application. Version 6.4 has a local privilege-escalation vulnerability allows code execution as root from the low-privilege pihole account. Important context: the pihole account uses nologin, so this is not a direct interactive-login issue. However, nologin does not prevent code from running as UID pihole if a Pi-hole component is compromised. In that realistic post-compromise scenario, attacker-controlled content in /etc/pihole/versions is sourced by root-run Pi-hole scripts, leading to root code execution. This vulnerability is fixed in 6.4.1.

curl_cffi is the a Python binding for curl. Prior to 0.15.0, curl_cffi does not restrict requests to internal IP ranges, and follows redirects automatically via the underlying libcurl. Because of this, an attacker-controlled URL can redirect requests to internal services such as cloud metadata endpoints. In addition, curl_cffi’s TLS impersonation feature can make these requests appear as legitimate browser traffic, which may bypass certain network controls. This vulnerability is fixed in 0.15.0.

Distribution is a toolkit to pack, ship, store, and deliver container content. Prior to 3.1.0, in pull-through cache mode, distribution discovers token auth endpoints by parsing WWW-Authenticate challenges returned by the configured upstream registry. The realm URL from a bearer challenge is used without validating that it matches the upstream registry host. As a result, an attacker-controlled upstream (or an attacker with MitM position to the upstream) can cause distribution to send the configured upstream credentials via basic auth to an attacker-controlled realm URL. This vulnerability is fixed in 3.1.0.

Homarr is an open-source dashboard. Prior to 1.57.0, a DOM-based Cross-Site Scripting (XSS) vulnerability has been discovered in Homarr's /auth/login page. The application improperly trusts a URL parameter (callbackUrl), which is passed to redirect and router.push. An attacker can craft a malicious link that, when opened by an authenticated user, performs a client-side redirect and executes arbitrary JavaScript in the context of their browser. This could lead to credential theft, internal network pivoting, and unauthorized actions performed on behalf of the victim. This vulnerability is fixed in 1.57.0.

Improper Neutralization of Input During Web Page Generation ('Cross-site Scripting') vulnerability in David Lingren Media LIbrary Assistant allows Stored XSS.This issue affects Media LIbrary Assistant: from n/a through 3.34.

Pi-hole Admin Interface is a web interface for managing Pi-hole, a network-level ad and internet tracker blocking application. From 6.0 to before 6.5, configuration values from the /api/config endpoint are placed directly into HTML value="" attributes without escaping in settings-advanced.js, enabling HTML attribute injection. A double quote in any config value breaks out of the attribute context. JavaScript execution is blocked by the server's CSP (script-src 'self'), but injected attributes can alter element styling for UI redressing. The primary attack vector is importing a malicious teleporter backup, which bypasses per-field server-side validation. This vulnerability is fixed in 6.5.

Pi-hole Admin Interface is a web interface for managing Pi-hole, a network-level ad and internet tracker blocking application. From 6.0 to before 6.5, client hostnames and IP addresses from the FTL database are rendered into the DOM without escaping in network.js (Network page) and charts.js/index.js (Dashboard chart tooltips). While upstream validation in dnsmasq and FTL blocks HTML characters via normal DHCP/DNS paths, the web UI performs no output escaping — an inconsistency with other fields in the same file that are properly escaped. This vulnerability is fixed in 6.5.

Pi-hole Admin Interface is a web interface for managing Pi-hole, a network-level ad and internet tracker blocking application. From 6.0 to before 6.5, a reflected DOM-based XSS vulnerability in taillog.js allows an unauthenticated attacker to inject arbitrary HTML into the Pi-hole admin interface by crafting a malicious URL. The file query parameter is interpolated into an innerHTML assignment without escaping. Because the Content-Security-Policy is missing the form-action directive, injected <form> elements can exfiltrate credentials to an external origin. This vulnerability is fixed in 6.5.

Improper Neutralization of Special Elements used in an SQL Command ('SQL Injection') vulnerability in David Lingren Media LIbrary Assistant allows SQL Injection.This issue affects Media LIbrary Assistant: from n/a through 3.34.

Homarr is an open-source dashboard. Prior to 1.57.0, the user registration endpoint (/api/trpc/user.register) is vulnerable to a race condition that allows an attacker to create multiple user accounts from a single-use invite token. The registration flow performs three sequential database operations without a transaction: CHECK, CREATE, and DELETE. Because these operations are not atomic, concurrent requests can all pass the validation step (1) before any of them reaches the deletion step (3). This allows multiple accounts to be registered using a single invite token that was intended to be single-use. This vulnerability is fixed in 1.57.0.

GLPI is a free asset and IT management software package. From 10.0.0 to before 10.0.24 and 11.0.6, an authenticated user can perform a SQL injection via the logs export feature. This vulnerability is fixed in 10.0.24 and 11.0.6.