Reported

September 20, 2026

Issued

September 23, 2026

Package

faster-hex (crates.io)

Type

INFO Unsound

Categories

• memory-corruption

Keywords

#buffer-overflow #avx2

References

• https://github.com/nervosnetwork/faster-hex/issues/73
• https://github.com/nervosnetwork/faster-hex/pull/72
• https://github.com/nervosnetwork/faster-hex/pull/74

Patched

• >=0.10.1

Unaffected

• <0.3.0

Affected Architectures

• x86
• x86_64

Affected Functions

Version

faster_hex::hex_decode_unchecked

• >=0.3.0, <0.10.1

Description

faster_hex::hex_decode_unchecked is a safe function. On x86 and x86_64, when AVX2 is enabled, hex_decode_avx2 keeps decoding while dst.len() >= 32 and then loads 32 bytes from src and 32 bytes from src[32..]. Nothing in that loop requires src to contain 64 bytes.

A safe caller can pass a short src and a dst of at least 32 bytes. The AVX2 load then reads outside src.

The function has had this shape since 0.3.0. 0.10.0 is the last release with the unbounded AVX2 read. The maintainer agreed to this advisory and shipped the backport in 0.10.1 (pull request 74): the AVX2 loop now requires src.len() >= 64 before those loads. 1.0.0 rewrites hex_decode_unchecked so the decoded length is min(src.len() / 2, dst.len()). Both are covered by >= 0.10.1. Other architectures take the scalar fallback and are outside this advisory.

Advisory available under CC0-1.0 license.