b9917

ggml-org/llama.cppb9917Jul 8, 2026by github-actions[bot]same day as b9916

What changed since b9916

low risk

b9917 adds a security fix for heap-buffer-overflow vulnerabilities in the UGM tokenizer's precompiled_charsmap handling.

Improvements

  • Bounds-checked scanning replaces unsafe strlen when reading the UGM precompiled_charsmap, preventing heap overflows from malicious T5/UGM GGUF files.
  • Minimum 4-byte size validation is now enforced before reading xcda_blob_size, with checks moved into the load path.

Bug fixes

  • Fixed out-of-bounds reads in the UGM tokenizer that could be triggered by crafted T5/UGM GGUF files.

Upgrade notes

This is a security patch with no API or config changes; upgrading is a drop-in replacement. Users loading T5/UGM GGUF models from untrusted sources should upgrade promptly, as the fix closes heap-buffer-overflow paths. No migration steps are required.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

Fixes out-of-bounds reads in the UGM tokenizer to prevent heap-buffer-overflow vulnerabilities in T5/UGM GGUF file parsing.

Key Highlights

  • Security fix: Heap-buffer-overflow in UGM tokenizer
  • Replaced unsafe strlen with bounds-checked strnlen
  • Added validation for minimum size (4 bytes) before reading xcda_blob_size

Release Notes (excerpt)

<details open> fix: OOB reads in UGM tokenizer (precompiled_charsmap handling) (#18750) * fix: OOB reads in UGM tokenizer (precompiled_charsmap handling) - Validate minimum size (4 bytes) before reading xcda_blob_size - Use strnlen with bounds check instead of unsafe strlen Both issues allow heap-buffer-overflow from malicious T5/UGM GGUF files. * Replace unsafe strnlen() with a bounds-checked loop that scans for \0 within the remaining array size. * move bounds checks to load…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories