Summary
vm2 leaks absolute host filesystem paths to sandbox code via error stack formatting
Advisory details
Summary
Attacker-controlled code can trigger a host-realm syntax error and read its stack through the vm2 bridge. Host-realm stack formatting bypasses the sandbox-side redaction, so the returned value exposes absolute paths from vm2, Node.js internals, and the embedding application. Default VM and NodeVM configurations are affected without requiring special options.
PoC
A default-configured VM runs attacker-supplied code. Calling eval with deliberately malformed source makes the host-side source transformer throw a SyntaxError; reading .stack on the caught error exposes the host call stack to the sandbox.
const { VM } = require("vm2");
console.log(new VM().run(`
var s; try { eval("@@@ catch") } catch (e) { s = e.stack }
s;
`));
Observed output
SyntaxError: Unexpected character '@'
at makeNiceSyntaxError (.../lib/transformer.js:41:16)
at transformer (.../lib/transformer.js:116:8)
at Object.transformAndCheck (.../lib/vm.js:76:14)
at Object.apply (.../lib/setup-sandbox.js:2585:16)
at VM.run (.../lib/vm.js:529:16)
The stack string returned to the sandbox contains absolute host paths for lib/transformer.js, lib/vm.js, lib/setup-sandbox.js, Node internals, and the embedding application's own source file.
Impact
Sandboxed code running under the default new VM() or new NodeVM() configuration can read absolute filesystem paths of the embedding application's source tree plus host function names, bypassing the host-path redaction added for GHSA-v27g-jcqj-v8rw. On a multi-tenant code-runner this discloses deployment layout such as /home/app/... or /var/task/..., useful for fingerprinting and for chaining into further attacks. The bridge forwards .stack reads to the host-realm formatter (lib/bridge.js:1482), so the sandbox-side stack redaction never runs; the leak does not require special configuration and persists when string eval is disabled, because the host-side transformer throws before eval is handled. Information disclosure only, no code execution.
References
- https://github.com/advisories/GHSA-x6m4-chr9-cg97
- https://github.com/patriksimek/vm2/security/advisories/GHSA-x6m4-chr9-cg97
- https://nvd.nist.gov/vuln/detail/CVE-2026-92936
- https://github.com/patriksimek/vm2/commit/5d5ba5343af571c688a3a81cefad2de25b94180a
- https://github.com/patriksimek/vm2/releases/tag/v3.11.7
- https://www.vulncheck.com/advisories/vm2-3.11.0-before-3.11.7-information-disclosure-via-error-stack
Related vulnerabilities
All Supply chain →- MEDIUMCVE-2026-75597
pyLoad: Unauthenticated access to /web/<path:filename> bypasses authentication on sensitive templates and leaks internal error details via exception attribute typo
- HIGHCVE-2026-106106
Quasar Framework: SSR/SSG dev error page discloses the full shell environment and its </script> escape is bypassable
- MEDIUMCVE-2026-102904
JupyterLab: Argument injection in JupyterLab extension uninstall exposes server-readable files and internal URLs
- MEDIUMCVE-2026-85709
lightrag-hku: Sensitive Information Exposure Through Raw Exception Messages in API Error Responses
- MEDIUMCVE-2026-73555
vLLM: Unauthenticated Internal Path and Username Disclosure via Validation Error Messages
- MEDIUMCVE-2026-69127
Kirby: System path exposure from error messages in the REST API