node-tar: Process crash via PAX numeric path type confusionGHSA-w8wr-v893-vjvp

Summary

A crafted 2.5KB tar archive crashes any Node.js process that extracts it. The PAX header parser coerces all-digit path values to JavaScript numbers, which causes an uncaught TypeError when downstream code calls .split('/') on the numeric value. Error handlers and strict: false cannot intercept the crash.

Details

In pax.ts line 180, parseKV converts PAX values matching /^[0-9]+$/ to numbers via +v. This applies to all fields including path and linkpath. When a PAX header sets path to an all-digit string like "12345", the value becomes the number 12345.

This number flows through Header -> ReadEntry -> Unpack.CHECKPATH, where normalizeWindowsPath(entry.path).split('/') throws a TypeError because numbers don't have .split().

The throw is synchronous during event emission and bypasses all error handling:

• strict: false does not help
• 'error' event handlers do not catch it
• 'warn' handlers do not catch it
• The TypeError propagates through the event emitter stack as an uncaughtException

Directory, SymbolicLink, and Link type entries reach CHECKPATH and crash. File type entries crash earlier in Header constructor at this.path.slice(-1), but that throw is caught and emitted as a warning only.

PoC

Create a tar archive with a PAX extended header containing an all-digit path:

PAX header body: "18 path=12345\n"
Entry type: Directory (type '5')

Extract it:

const tar = require('tar');

// All of these crash with TypeError: t.split is not a function
tar.extract({ file: 'malicious.tar', cwd: '/tmp/test' });

// Error handlers don't help:
tar.extract({ file: 'malicious.tar', cwd: '/tmp/test', strict: false })
  .on('error', (err) => { /* never reached */ })
  .on('warn', (code, msg) => { /* never reached */ });

The archive is ~2.5KB. The crash is deterministic on every attempt.

Impact

Denial of service. Any application or tool that extracts untrusted tar archives crashes from a single small file. This includes npm (which uses node-tar to extract packages), CI/CD pipelines, file upload processors, and backup tools. The crash cannot be caught by application-level error handling.