ffi: refCallback()/unrefCallback() crash (SIGSEGV) when called on a garbage-collected callback #64880

Version

main

Platform

N/A

Subsystem

ffi

What steps will reproduce the bug?

repro.c

#include <stdint.h>

int32_t noop(int32_t x) { return x; }

Compile it by running

$ cc -shared -o repro.so repro.c # or .dylib on macOS

repro.js

import { DynamicLibrary, suffix } from 'node:ffi';

const lib = new DynamicLibrary(`./repro.${suffix}`);

let cb = (x) => x + 1;
const ptr = lib.registerCallback({ arguments: ['i32'], return: 'i32' }, cb);

// Allow GC to collect the callback backing function
lib.unrefCallback(ptr);
cb = null;
globalThis.gc();

// Second unref (or refCallback) on the now-empty weak handle → SIGSEGV
lib.unrefCallback(ptr);

lib.close();

Run it using

$ node --experimental-ffi --expose-gc --no-warnings repro.js

How often does it reproduce? Is there a required condition?

Always

What is the expected behavior? Why is that the expected behavior?

unrefCallback/refCallback on a collected callback should throw ERR_INVALID_ARG_VALUE("Callback not found").

What do you see instead?

segmentation fault (core dumped)

The process crashes with SIGSEGV inside v8::PersistentBase::SetWeak() (or ClearWeak() for refCallback) because the weak handle slot is already cleared after GC.

Additional information

No response