fix: add bounds checking in normalizeSkinningWeights to prevent heap OOB write

[YLChen-007]

Summary

Fix a heap buffer overflow vulnerability in normalizeSkinningWeights (resolves #9877).

Problem

The normalize lambda in libs/gltfio/src/ResourceLoader.cpp trusts the glTF accessor's count field without validating it against the actual buffer size. A maliciously crafted .gltf file can specify an arbitrarily large count with a small buffer, causing out-of-bounds heap writes during skinning weight normalization.

This leads to:

• Denial of Service via segfault
• Potential code execution via heap memory corruption

Any application using Filament's gltfio with normalizeSkinningWeights = true (the default in gltf_viewer, Android ModelViewer, Web viewer, etc.) is vulnerable when loading untrusted glTF files.

Fix

This patch adds bounds checking to the normalize lambda:

1. Null pointer validation: Check buffer_view, buffer, and data pointers before dereferencing
2. Offset validation: Ensure totalOffset (viewOffset + accessorOffset) does not exceed buffer size
3. Safe count computation: Calculate the maximum number of elements that fit within the buffer based on available bytes and stride
4. Count clamping: Clamp data->count to the safe maximum with a LOG(WARNING) message

The fix gracefully handles malformed inputs by clamping the iteration count rather than crashing.

Testing

Verified with a standalone harness mirroring the normalizeSkinningWeights logic:

Test Case	Input	Expected	Result
Exploit (poc.gltf)	count=1000000, buffer=16 bytes	No crash, warning logged	✅ Pass
Control (control.gltf)	count=1, buffer=16 bytes	Normal processing	✅ Pass
Unpatched baseline	Exploit → segfault (exit 139)	Crash confirmed	✅ Confirmed

Changes

• libs/gltfio/src/ResourceLoader.cpp: Added bounds checking in normalizeSkinningWeights normalize lambda (+38 lines, -2 lines)

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.