Heap Buffer Overflow in Spotify Voyager HNSW Index Loading

Executive Summary

This vulnerability is a heap out-of-bounds write issue in the deserialization process of HNSW vector indexes used by hnswlib and Spotify Voyager. The loadIndex function does not verify that the element count in a serialized HNSW index is within the stored capacity. An attacker can craft an index file with a smaller capacity but larger element count, causing the loader to write beyond the allocated buffer during deserialization.

Useful 0 Not Useful 0

Detection Guidance

To detect this vulnerability, inspect HNSW index files for mismatches between stored capacity (max_elements_) and element count (cur_element_count). Check if the loader validates these values before deserialization. Review logs for crashes during index loading or memory corruption errors.

Useful 0 Not Useful 0

Impact Analysis

This vulnerability can lead to memory corruption, application crashes, or potential arbitrary code execution. It is particularly dangerous when untrusted HNSW index files are loaded, such as in retrieval-augmented generation pipelines, multi-tenant search services, or tools handling user-uploaded index files.

Useful 0 Not Useful 0

Mitigation Strategies

Update hnswlib and Spotify Voyager to patched versions that enforce cur_element_count <= max_elements_ checks. Avoid loading untrusted index files. Implement input validation for index files in your application. Monitor for crashes or memory corruption during index loading.

Useful 0 Not Useful 0