Fix NPZ buffer lifetime to prevent use-after-free

[gagandhakrey]

Purpose

Fix a lifetime issue in NpzFile that could result in a dangling pointer when loading NPZ data from an in-memory buffer.

Issue

NpzFile only stores the raw memory address of the input buffer:

self._data = _get_pointer(data)

The original data object is not retained after load_npz() returns. As a result, the backing buffer may be garbage-collected while NpzFile continues to hold a pointer to its memory.

Subsequent accesses through __getitem__ use this stale pointer to read NPZ entries, resulting in a classic use-after-free.

Example:

data = spdl.io.load_npz(f.read())
array = data["x"]

After load_npz() returns, the temporary buffer can be freed even though NpzFile still references its address.

Fix

Retain a strong reference to the original input buffer for the lifetime of the NpzFile instance by storing it as an instance attribute.

This ensures the underlying memory remains valid until the NpzFile object is destroyed, preventing dangling pointers during subsequent array access.

Post-fix Impact

• Eliminates a potential use-after-free when accessing NPZ entries.
• Prevents undefined behavior caused by reading freed memory.
• Ensures returned NumPy arrays always reference valid backing storage.
• Preserves the existing API and runtime behavior with negligible memory overhead beyond retaining the original input buffer for the lifetime of the NpzFile.

[meta-cla]

Hi @gagandhakrey!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

[meta-cla]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

[mthrok]

Hi @gagandhakrey

Thanks for the PR. It looks good. I'm away from my laptop for a week or so, so I'll merge it when I come back.

Meanwhile if you can add a unit test, that'll be great.

[meta-codesync]

This pull request has been imported. If you are a Meta employee, you can view this in D114881820. (Because this pull request was imported automatically, there will not be any future comments.)

[mthrok]

IIUC, this _reuse_freed_memory takes advantage of the implementation detail of CPython which reuses pointers of deallocated objects, so if the pointer to the temporary is not retained by npz object, then _reuse_freed_memory can overrides the data and the following assert_array_equal will fail. and the _reuse_freed_memory increases the chance of pointer being reused by repeating the allocation 2000 times.

This is a bit elaborated logic that uses internal details of the CPython interpreter. Can you add a comment of the intention here so that it's easier for future maintainer to get what is happening?

[gagandhakrey]

yes, exactly added comments explaining the cpython allocator reuse and the intent. thanks for pointing it out

[mthrok]

Thank you for providing the unit test. I left one comment. Otherwise looks good.

[gagandhakrey]

Thank you for providing the unit test. I left one comment. Otherwise looks good.

Thank you @mthrok for your review ,I have addressed that comment

[mthrok]

@gagandhakrey Thank you for your contribution. Please feel free to open an issue/PR if you should encounter any issue.