INCIDENT MAP

Browse / AI Framework and Platform Flaws / CVE-2026-92933

vm2: util.getCallSites() bypasses GHSA-v27g-jcqj-v8rw host-frame redaction, leaks host call stack

CVE-2026-92933 · 2026-10-05 moderate

Class
AI Framework and Platform Flaws (filed by a model that read the advisory)
Technique
Information exposure
Affected
vm2 (npm)
Gets past
Earlier patch
CVSS
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:L/I:N/A:N
EPSS
0.4% chance of exploitation in 30 days

How it works, in short

Sandboxed code reaches the host util module's getCallSites, exposing host file paths and stack frames that an earlier redaction fix was meant to hide. (Our summary, written by a model. The source text is below.)

Details

Summary

NodeVM exposes the host util module to the sandbox through an unfiltered shallow copy (Object.assign({}, util)). On Node.js >= 22.9 this hands sandboxed code util.getCallSites(), a programmatic stack-introspection API that returns the host process's full call stack — absolute file paths, function names, and line numbers — including vm2 bridge internals and the embedding application's entrypoint. This bypasses the host-frame redaction established in GHSA-v27g-jcqj-v8rw, which only covers the Error.prepareStackTrace channel.

Details

util.getCallSites() produces its data host-side and never passes through that formatter, so frames such as lib/bridge.js @apply (the bridge apply trap), lib/nodevm.js @run, the embedder's own entry file, and node:internal/* frames reach the sandbox verbatim as data properties (scriptName, functionName, lineNumber, columnNumber, scriptId). A repository-wide grep (source, docs/ATTACKS.md, CHANGELOG, tests) shows no occurrence of getCallSites; the member was never considered.

PoC

Prerequisites: Node.js >= 22.9 (verified on v22.23.1); run npm ci --no-audit --no-fund at the repository root.

One-line reproducer (prints /workspace/repo/lib/bridge.js, i.e. a vm2-internal host path):

node -e "const {NodeVM}=require('/workspace/repo'); console.log(new NodeVM({require:{builtin:['util']}}).run(\"module.exports = require('util').getCallSites(4)[0].scriptName\"))"

Observed frames inside the sandbox, with both require: { builtin: ['util'] } and require: { builtin: ['*'] }:

/workspace/repo/lib/bridge.js @apply:1664
vm.js @:5
/workspace/repo/lib/bridge.js @apply:1664
/workspace/repo/lib/nodevm.js @run:563
/workspace/out/<embedder entrypoint>.js @main:29
node:internal/modules/cjs/loader @:1781
node:internal/modules/cjs/loader @:1913
... (8 node:internal/* frames in total)

Impact

Information disclosure. Any NodeVM configuration that allows the util (or sys) builtin — including the wildcard builtin: ['*'], both typical configurations from the README — lets untrusted sandboxed code programmatically read the host call stack: the vm2 installation path, the embedding application's file layout and entrypoint, and internal function names and line numbers. This is the same information category redacted by GHSA-v27g-jcqj-v8rw (Defense Invariant #5 in docs/ATTACKS.md) and breaks defense-in-depth assumptions of embedders that rely on host frames being invisible to the sandbox. The API returns only strings/numbers (no host object references); no escalation to code execution was identified.

Suggested fix: filter members in defaultBuiltinLoaderUtil (allowlist, or at minimum drop getCallSites), apply the same treatment to the generic loader path used by sys, and extend the GHSA-v27g regression tests to cover programmatic stack-introspection APIs.

References

Source: GitHub Advisory Database, used under CC BY 4.0.