# GHSA-662Q-V5RM-78GM: Unauthenticated Stored XSS in the Contact module

- Database: GitHub
- Severity: CVSS 9.3 (Critical)
- Published: 2026-08-30
- Modified: 2026-09-22

## Tracked as

- [CVE-2026-94565](https://cve.now/CVE-2026-94565.md): NukeViet unauthenticated stored XSS in contact module

## Affected versions

- nukeviet/nukeviet (Packagist): 3.x, 4.x

## Description

### Summary
A stored-XSS vulnerability in the Contact module allows any unauthenticated user to inject malicious Javascript.

### Details
NukeViet's central input filter (`NukeViet\Core\Request::filterTags`), applied to virtually all `POST` data, is a denylist HTML sanitizer. Its most important rule strips any attribute whose name matches `/^on/i` (event handlers). This rule is bypassable through an **HTML parser differential**: the filter tokenizes a tag's attributes by splitting on ASCII space (`0x20`) only, whereas browsers also separate attributes on `/`. By inserting a `/` immediately before an event-handler name, the attacker makes the filter read the attribute name as `/onerror` — which does **not** start with `on`, so the denylist misses it — while the browser parses it as `onerror` and fires it.

The payload survives the entire filter chain unmodified and is stored and served **raw** (no output encoding) into the administration panel, which sends **no Content-Security-Policy**:

```
INPUT  : <img src=x /onerror=alert(document.domain)>
STORED : <img src="x" /onerror="alert(document.domain)" />
ADMIN  : <td colspan="2"><img src="x" /onerror="alert(document.domain)" /></td>
```

Because the handler executes in the **same origin** as the admin panel, an unauthenticated attacker who submits a crafted contact message obtains arbitrary JavaScript execution in the session of any administrator who opens it — enabling administrator **account takeover**.

### PoC
#### Minimal payload

```html
<img src=x /onerror=alert(document.domain)>
```

`alert(document.domain)` displays the site's real origin (e.g. the target host), proving **same-origin** execution.

#### Manual reproduction (fresh 4.6.00 install, default config)

1. As an anonymous visitor open `/contact/`; set the **message** field to the payload above.
2. Solve the CAPTCHA and submit (form returns `status: ok`).
3. Log in to the admin panel; open **Admin → Contact** and view the message
   (`admin/index.php?nv=contact&op=view&id=<id>`).
4. `onerror` fires on view; `alert()` shows the panel's own origin.

### Impact
An unauthenticated attacker achieves arbitrary JavaScript execution in the **same origin** as the NukeViet admin panel, running in the session of any administrator who views the malicious contact message. Possible attacks:
1. Super admin account persistence
2. RCE via admin functions

### Mitigation
1. **Tokenize attributes the way the HTML spec does.** In `filterTags()`, treat `/` (and all HTML    whitespace: tab, LF, FF, CR, space) as attribute separators, so `/onerror` is parsed as the token `onerror` and caught by the existing denylist. Equivalently, strip leading `/` and other non-name characters from the attribute name before the `^on` test.
2. **Do not preserve non-conforming attribute names.** Reject/normalize any attribute whose name is not `[A-Za-z][A-Za-z0-9-]*` after normalization, rather than emitting `/onerror` verbatim.
3. **Prefer an allowlist of attributes** over the `^on` denylist; drop anything not explicitly permitted for the element.

## References

- https://github.com/nukeviet/nukeviet/security/advisories/GHSA-662q-v5rm-78gm

Page: https://cve.now/GHSA-662Q-V5RM-78GM
