CVE-2024-21825: 
Homebrew vulnerability analysis and mitigation

A heap-based buffer overflow vulnerability was discovered in the GGUF library GGUF_TYPE_ARRAY/GGUF_TYPE_STRING parsing functionality of llama.cpp Commit 18c2e17. The vulnerability was identified and disclosed in January 2024, affecting the LLaMA.cpp implementation which relies on the ggml library for tensor operations and the GGUF file format for storing LLM model representations (Talos Report).

The vulnerability exists in the GGUF file parsing functionality, specifically in the gguf_init_from_file function. The issue occurs during the parsing of GGUF_TYPE_ARRAY/GGUF_TYPE_STRING elements, where an integer overflow can occur during memory allocation. The multiplication between kv->value.arr.n (an arbitrary uint64_t value) and sizeof(struct gguf_str) (16 bytes) can lead to allocating less space than required, potentially resulting in a heap-based buffer overflow. The vulnerability has been assigned a CVSSv3 score of 8.8 (CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H) and is classified under CWE-190 (Integer Overflow or Wraparound) (Talos Report).

The vulnerability can lead to code execution when a specially crafted .gguf file is processed by the affected system. An attacker can trigger this vulnerability by providing a malicious file, potentially leading to unauthorized code execution with the privileges of the application processing the file (Talos Report).

The vulnerability has been fixed in a vendor patch released on January 29, 2024. While Databricks independently reported this vulnerability concurrently with Cisco Talos's discovery, users are advised to update to the latest version of llama.cpp that includes the security fix (Talos Report).