GHSA-cg4g-m8jx-vjv2

## Summary

`is_url_safe` in v1.0.3 contains an SSRF bypass. `remove_at_symbol_in_string` is applied to the raw URL string **before** `new URL()` parses it. This strips the `@` that separates userinfo from host, corrupting the hostname so internal IPs are never checked.

## Vulnerability

In `helpers.ts`, `is_url_safe` does:

```ts u = remove_at_symbol_in_string(u); // strips ALL '@' from the raw string // ... const parsed = new URL(u); const hostname = parsed.hostname; // resolved from the corrupted string ```

### What happens step by step

Input: `http://evil.com@127.0.0.1/`

1. `remove_at_symbol_in_string` → `http://evil.com127.0.0.1/` 2. `new URL(...)` → `hostname = "evil.com127.0.0.1"` 3. Not a bare IP, not IPv6 → passes all IP checks 4. `is_hostname_resolve_to_internal_ip("evil.com127.0.0.1")` → NXDOMAIN → returns false 5. **Result: `true` (safe)** — but any HTTP client using the *original* URL connects to `127.0.0.1`

### Proof of Concept

```js import nock from 'nock'; import { got } from 'got'; import { is_url_safe } from 'dssrf';

// Simulate an internal server at 10.0.0.1 that returns secret data nock('http://10.0.0.1:80').persist().get('/').reply(200, 'SECRET_DATA');

const BYPASS_URL = 'http://2@10.0.0.1/'; const PLAIN_URL = 'http://10.0.0.1/';

// dssrf should block both — it only blocks the plain one console.log('--- dssrf validator ---'); console.log(`is_url_safe('${PLAIN_URL}') =`, await is_url_safe(PLAIN_URL), '← correctly blocked'); console.log(`is_url_safe('${BYPASS_URL}') =`, await is_url_safe(BYPASS_URL), '← ⚠️ BYPASSED (should be false)');

// HTTP client with the bypass URL — gets SECRET_DATA back from 10.0.0.1 console.log('\n--- HTTP client ---'); try { const res = await got(BYPASS_URL, { retry: { limit: 0 } }); console.log(`got('${BYPASS_URL}') response:`, res.body, '← ⚠️ VULNERABLE'); } catch (e) { console.log(`got('${BYPASS_URL}') blocked:`, e.message); } ```

## Root Cause

`@` in a URL separates `userinfo` (credentials) from `host`. Stripping it from the raw string before parsing destroys that boundary. The fix is to **reject any URL that contains a userinfo component** after parsing.

## Suggested Fix

Remove the `remove_at_symbol_in_string` call from `is_url_safe` and add a userinfo check after `new URL()`:

```ts const parsed = new URL(u);

// Reject userinfo — '@' in authority is a classic SSRF bypass vector if (parsed.username !== "" || parsed.password !== "") { return false; } ```

A working patch verified against 15 vectors (all internal IPv4 ranges, IMDS, IPv6 via userinfo, and legitimate public URLs) is ready to submit as a PR.

## Impact

- **Affected version**: 1.0.3 (latest) - **Bypasses**: all internal IPv4 ranges, IPv6 loopback/ULA/link-local, AWS IMDS (`169.254.169.254`), any internal hostname via userinfo prefix - **Note**: The GHSA-8p33-q827-ghj5 advisory patched version (`1.0.3`) should be updated since this vector was not covered by that fix

Users are strongly advised to upgrade to dssrf 1.0.4