ghsa

### Summary A Server-Side Request Forgery (SSRF) vulnerability exists in the `MediaConnector` class within the vLLM project's multimodal feature set. The `load_from_url` and `load_from_url_async` methods fetch and process media from user-provided URLs without adequate restrictions on the target hosts. This allows an attacker to coerce the vLLM server into making arbitrary requests to internal network resources. This vulnerability is particularly critical in containerized environments like `llm-d`, where a compromised vLLM pod could be used to scan the internal network, interact with other pods, and potentially cause denial of service or access sensitive data. For example, an attacker could make the vLLM pod send malicious requests to an internal `llm-d` management endpoint, leading to system instability by falsely reporting metrics like the KV cache state. ### Vulnerability Details The core of the vulnerability lies in the `MediaConnector.load_from_url` method and its asynchronous ...

## Summary ## A Server-Side Request Forgery (SSRF) vulnerability in the chat API allows any authenticated user to force the server to make arbitrary HTTP requests to internal and external networks. This can lead to the exposure of sensitive internal services, reconnaissance of the internal network, or interaction with third-party services. The same mechanism also allows for a Local File Inclusion (LFI) vulnerability, enabling users to read arbitrary files from the server's filesystem. ## Details ## The vulnerability exists in the _process_request function within src/llamafactory/api/chat.py. This function is responsible for processing incoming multimodal content, including images, videos, and audio provided via URLs. The function checks if the provided URL is a base64 data URI or a local file path (os.path.isfile). If neither is true, it falls back to treating the URL as a web URI and makes a direct HTTP GET request using requests.get(url, stream=True).raw without any validation or...

### Summary A resource-exhaustion (denial-of-service) vulnerability exists in multiple endpoints of the OpenAI-Compatible Server due to the ability to specify Jinja templates via the `chat_template` and `chat_template_kwargs` parameters. If an attacker can supply these parameters to the API, they can cause a service outage by exhausting CPU and/or memory resources. ### Details When using an LLM as a chat model, the conversation history must be rendered into a text input for the model. In `hf/transformer`, this rendering is performed using a Jinja template. The OpenAI-Compatible Server launched by vllm serve exposes a `chat_template` parameter that lets users specify that template. In addition, the server accepts a `chat_template_kwargs` parameter to pass extra keyword arguments to the rendering function. Because Jinja templates support programming-language-like constructs (loops, nested iterations, etc.), a crafted template can consume extremely large amounts of CPU and memory and ...

### Impact This is a critical network security vulnerability for Akka.Remote **users who have SSL / TLS enabled** on their Akka.Remote connections and were expecting certificate-based authentication to be enforced on all peers attempting to join the network. In all versions of Akka.Remote from v1.2.0 to v1.5.51, TLS could be enabled via our `akka.remote.dot-netty.tcp` transport and this would correctly enforce private key validation on the server-side of inbound connections. Akka.Remote, however, never asked the outbound-connecting client to present ITS certificate - therefore it's possible for untrusted parties to connect to a private key'd Akka.NET cluster and begin communicating with it **without any certificate**. The issue here is that for certificate-based authentication to work properly, ensuring that all members of the Akka.Remote network are secured with the same private key, Akka.Remote needed to implement mutual TLS. This was not the case before Akka.NET v1.5.52. If you...

## Summary `Rack::Multipart::Parser` can accumulate unbounded data when a multipart part’s header block never terminates with the required blank line (`CRLFCRLF`). The parser keeps appending incoming bytes to memory without a size cap, allowing a remote attacker to exhaust memory and cause a denial of service (DoS). ## Details While reading multipart headers, the parser waits for `CRLFCRLF` using: ```ruby @sbuf.scan_until(/(.*?\r\n)\r\n/m) ``` If the terminator never appears, it continues appending data (`@sbuf.concat(content)`) indefinitely. There is no limit on accumulated header bytes, so a single malformed part can consume memory proportional to the request body size. ## Impact Attackers can send incomplete multipart headers to trigger high memory use, leading to process termination (OOM) or severe slowdown. The effect scales with request size limits and concurrency. All applications handling multipart uploads may be affected. ## Mitigation * Upgrade to a patched Rack vers...

## Summary `Rack::Multipart::Parser` stores non-file form fields (parts without a `filename`) entirely in memory as Ruby `String` objects. A single large text field in a multipart/form-data request (hundreds of megabytes or more) can consume equivalent process memory, potentially leading to out-of-memory (OOM) conditions and denial of service (DoS). ## Details During multipart parsing, file parts are streamed to temporary files, but non-file parts are buffered into memory: ```ruby body = String.new # non-file → in-RAM buffer @mime_parts[mime_index].body << content ``` There is no size limit on these in-memory buffers. As a result, any large text field—while technically valid—will be loaded fully into process memory before being added to `params`. ## Impact Attackers can send large non-file fields to trigger excessive memory usage. Impact scales with request size and concurrency, potentially leading to worker crashes or severe garbage-collection overhead. All Rack applications p...

## Summary `Rack::Multipart::Parser` buffers the entire multipart **preamble** (bytes before the first boundary) in memory without any size limit. A client can send a large preamble followed by a valid boundary, causing significant memory use and potential process termination due to out-of-memory (OOM) conditions. ## Details While searching for the first boundary, the parser appends incoming data into a shared buffer (`@sbuf.concat(content)`) and scans for the boundary pattern: ```ruby @sbuf.scan_until(@body_regex) ``` If the boundary is not yet found, the parser continues buffering data indefinitely. There is no trimming or size cap on the preamble, allowing attackers to send arbitrary amounts of data before the first boundary. ## Impact Remote attackers can trigger large transient memory spikes by including a long preamble in multipart/form-data requests. The impact scales with allowed request sizes and concurrency, potentially causing worker crashes or severe slowdown due to ...

### Summary The API key support in vLLM performed validation using a method that was vulnerable to a timing attack. This could potentially allow an attacker to discover a valid API key using an approach more efficient than brute force. ### Details https://github.com/vllm-project/vllm/blob/4b946d693e0af15740e9ca9c0e059d5f333b1083/vllm/entrypoints/openai/api_server.py#L1270-L1274 API key validation used a string comparison that will take longer the more characters the provided API key gets correct. Data analysis across many attempts can allow an attacker to determine when it finds the next correct character in the key sequence. ### Impact Deployments relying on vLLM's built-in API key validation are vulnerable to authentication bypass using this technique.

The email parsing library incorrectly handles quoted local-parts containing @. This leads to misrouting of email recipients, where the parser extracts and routes to an unintended domain instead of the RFC-compliant target. Payload: `"xclow3n@gmail.com x"@internal.domain` Using the following code to send mail ``` const nodemailer = require("nodemailer"); let transporter = nodemailer.createTransport({ service: "gmail", auth: { user: "", pass: "", }, }); let mailOptions = { from: '"Test Sender" <your_email@gmail.com>', to: "\"xclow3n@gmail.com x\"@internal.domain", subject: "Hello from Nodemailer", text: "This is a test email sent using Gmail SMTP and Nodemailer!", }; transporter.sendMail(mailOptions, (error, info) => { if (error) { return console.log("Error: ", error); } console.log("Message sent: %s", info.messageId); }); (async () => { const parser = await import("@sparser/email-address-parser"); const { EmailAddress, ParsingOptions } = parse...

### Summary A remote code execution vulnerability in python-socketio versions prior to 5.14.0 allows attackers to execute arbitrary Python code through malicious pickle deserialization in multi-server deployments on which the attacker previously gained access to the message queue that the servers use for internal communications. ### Details When Socket.IO servers are configured to use a message queue backend such as Redis for inter-server communication, messages sent between the servers are encoded using the `pickle` Python module. When a server receives one of these messages through the message queue, it assumes it is trusted and immediately deserializes it. The vulnerability stems from deserialization of messages using Python's `pickle.loads()` function. Having previously obtained access to the message queue, the attacker can send a python-socketio server a crafted pickle payload that executes arbitrary code during deserialization via Python's `__reduce__` method. ### Impact This ...