Finding the cause of minor contagious metadata corruption in exiv2
A bug in the exiv2 library causes minor metadata corruption due to data races under concurrent access. The author traced the issue to unsynchronized shared state and proposed a patch adding proper locking.
Background
- Exiv2 is a widely used open-source C++ library for reading, writing, and manipulating image metadata (EXIF, IPTC, XMP). It is relied upon by photo management tools, digital forensics software, and many Linux distributions.
- The article describes a subtle, hard-to-reproduce bug in Exiv2's PNG metadata handling: writing a small amount of new data could corrupt metadata in nearby files, and only on certain systems. The root cause was a memory reuse issue (a dangling pointer into a recycled buffer) triggered by a standard library optimization.
- This matters because metadata corruption is silent — it can degrade photo archives or forensic evidence without warning. The bug also illustrates how hard "minor contagious" bugs are to find: they spread corruption to *other* files, not just the one being edited.
- Prior context: the author is a maintainer of Exiv2. This post is a deep-dive postmortem of a low-level C++ debugging journey, tracking a data race between object destruction and memory allocation behavior in libstdc++.