vfs: MemoryProvider.setReadOnly() does not prevent writes through existing file descriptors

[Archkon]

Version

latest main branch

Platform

7.1.2-arch3-1

Subsystem

vfs

What steps will reproduce the bug?

const vfs = require('node:vfs');

const provider = new vfs.MemoryProvider();
const virtualFs = vfs.create(provider, {
  emitExperimentalWarning: false,
});

virtualFs.writeFileSync('/file.txt', 'old');

const fd = virtualFs.openSync('/file.txt', 'r+');

provider.setReadOnly();

const data = Buffer.from('new');
virtualFs.writeSync(fd, data, 0, data.length, 0);

console.log(virtualFs.readFileSync('/file.txt', 'utf8')); // "new"

virtualFs.closeSync(fd);

Run with:

node --experimental-vfs reproduction.js

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

It reproduces consistently.

The writable file descriptor must be opened before setReadOnly() is called. Opening a new writable descriptor after setReadOnly() correctly throws EROFS.

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

Once setReadOnly() has been called, subsequent writes through any VirtualFileSystem backed by that provider should throw an error with code === 'EROFS', including writes through file descriptors that were opened before the provider became read-only.

This follows the current documentation for MemoryProvider.setReadOnly():

Subsequent writes through any VirtualFileSystem using this provider throw EROFS.

ftruncateSync() through an existing writable descriptor should similarly throw EROFS.

What do you see instead?

writeSync() succeeds and modifies the file even though:

virtualFs.readonly === true

ftruncateSync() also succeeds through a descriptor opened before setReadOnly().

Additional information

MemoryProvider.openSync() checks provider.readonly when a writable handle is opened. However, an existing MemoryFileHandle does not retain or re-check the provider's read-only state.

File-descriptor operations such as VirtualFileSystem.writeSync() and ftruncateSync() operate directly on the stored handle, so they bypass the provider's current readonly value.

Path-based writes made after setReadOnly(), such as writeFileSync(), already throw EROFS as expected.