GHSA-rm67-g9ch-vxff

Affected software

• Product: Poweradmin (web front-end for PowerDNS)
• Version tested: master, commit 7f28c3a97 (also reachable in the 4.x release line — the record-edit code is the same)
• Component: record editing (web UI)
• Vulnerability type: Broken Access Control / Insecure Direct Object Reference

Severity

Any authenticated user who owns at least one zone can tamper with records in every other zone on the server. In a shared/multi-tenant Poweradmin install this is effectively a cross-tenant DNS takeover.

Summary

When you save a record edit, Poweradmin checks whether you're allowed to touch the record by looking at a zone id you send in the POST body, but it then applies the change to a record id you also send in the POST body. Nothing checks that the record id actually belongs to that zone id. So you point the permission check at a zone you legitimately own, point the record id at somebody else's record, and the update goes through.

A low-privilege "Editor" who only owns attacker.example was able to overwrite a record living in victim.example (owned by the admin), even though the same account gets a flat "you do not have permission to access this zone" if it tries to open that zone directly.

Where the bug is

The core mistake is in RecordManager::editRecord() in lib/Domain/Service/Dns/RecordManager.php:

// line 298 – ownership is checked against the zone id from the request
$user_is_zone_owner = UserManager::verifyUserIsOwnerZoneId($this->db, $record['zid']);
...
// line 346 – but the update targets the record id from the request
$this->backendProvider->editRecord(
    $record['rid'], $name, $record['type'], $content, $validatedTtl, $validatedPrio, $record['disabled']
);

$record['zid'] and $record['rid'] are both taken straight from the submitted form. The permission gate uses one, the write uses the other, and they're never tied together.

The backend write has no zone scoping at all — lib/Infrastructure/Service/SqlDnsBackendProvider.php line 235:

$stmt = $this->db->prepare(
    "UPDATE $recordsTable SET name = ?, type = ?, content = ?, ttl = ?, prio = ?, disabled = ? WHERE id = ?"
);
$stmt->execute([$name, $type, $content, $ttl, $prio, $disabled, $recordId]);

It's WHERE id = ? on the record id and nothing else, so any record in the database is fair game.

The validation layer doesn't save it either. DnsRecordValidationService::validateRecord() (lib/Domain/Service/DnsRecordValidationService.php, around line 97) uses the zone id only to look up the zone name for hostname validation. It never verifies the record id lives in that zone.

There are two request paths that reach this same sink:

1. Multi-record save — EditController::saveRecords() (lib/Application/Controller/EditController.php, ~line 571) loops over $_POST['record'] and hands each entry's rid/zid to editRecord() verbatim. The controller's run() only permission-checks the zone id in the URL (~line 204), not the zone id inside each record row.
2. Single-record save — EditRecordController::saveRecord() reads $_POST['rid'] and forwards the whole $_POST (including the attacker-chosen zid) into editRecord().

For comparison, the delete path does it correctly. DeleteRecordController (~line 101) derives the zone id from the record id on the server side with getZoneIdFromRecordId() and checks permission against that. EditCommentController also uses the URL zone id for both the check and the write. So this really is an oversight in the edit path, not a deliberate design choice — the safe pattern already exists elsewhere in the same codebase.

Proof of concept

Setup (two accounts, two zones):

• attacker — a normal user on the "Editor" permission template (zone_content_edit_own_as_client, i.e. can only edit records in zones it owns). Owns attacker.example, which is zone/domain id 3.
• admin — owns victim.example, which is zone/domain id 2. That zone contains the record www.victim.example A 1.1.1.1 with record id 5.

Steps:

1. Log in as attacker. Confirm the access control is actually there: browse to /zones/2/edit (victim.example). You get "You do not have permission to access this zone." Good — you genuinely can't edit that zone the normal way.

2. Open your own zone's edit page, /zones/3/edit, and start a save (change any field and submit) with your intercepting proxy on. Catch the POST /zones/3/edit request. It already carries a valid CSRF _token and a valid serial for your zone, so don't touch those.

3. Edit the record block in that request so it targets the victim's record while keeping your own zone id:

POST /zones/3/edit HTTP/1.1
Host: localhost:8080
Cookie: PHPSESSID=<your attacker session>
Content-Type: application/x-www-form-urlencoded

_token=<the valid token from step 2>&serial=<the valid serial from step 2>&commit=1
&record[5][rid]=5&record[5][zid]=3
&record[5][name]=pwned&record[5][type]=A&record[5][content]=6.6.6.6
&record[5][ttl]=3600&record[5][prio]=0

The important part: record[5][rid]=5 is the victim's record (in zone 2), and record[5][zid]=3 is your own zone (so the ownership check passes).

1. Send it. You get HTTP 200.

2. Check the result. Record 5, which lives in the admin's zone, is now overwritten:

sqlite> SELECT id, domain_id, name, type, content FROM records WHERE id = 5;
5|2|pwned.attacker.example|A|6.6.6.6

It's still domain_id = 2 (the victim's zone) but its content is now attacker-controlled. You changed a record in a zone you were explicitly denied access to in step 1.

Two practical notes if you're reproducing this:

• The serial field is Poweradmin's optimistic-lock. It's checked against your own zone's current SOA serial and it bumps by one after every successful save, so grab a fresh one each time (the intercept-and-modify approach in step 2 handles this for you automatically).
• Record ids are sequential integers, so in a real attack you'd just walk rid = 1, 2, 3, … to hit every record in every zone on the box.

Impact

With one owned zone, an attacker can rewrite content, type, ttl and the disabled flag of any record anywhere in the installation. Concretely that means:

• Repointing or corrupting other customers' / other departments' DNS records (integrity).
• Silently disabling records by setting disabled = 1 (availability).
• A clean, resolvable hijack (victim's name kept intact) when the attacker owns a parent/ancestor zone of the target name — HostnameValidator::normalizeRecordName leaves the name alone if it already ends in the attacker's zone suffix, which is exactly the case in delegated-subzone setups (attacker owns example.com, victim owns delegated sub.example.com).

In the plain sibling-zone case the edited record's name gets forced under the attacker's zone suffix, which still destroys the original record — so at minimum it's cross-zone record deletion, and at most it's full record hijack.

The attacker's own CSRF token and own zone serial satisfy every form-level check, so nothing about the request looks abnormal to the app.

Suggested fix

Stop trusting the zid in the request. In RecordManager::editRecord() (and the EditRecordController path), look the zone up from the record id on the server side and use that for both the ownership check and the hostname normalization — the same thing deleteRecord() already does:

$details = $recordRepository->getRecordDetailsFromRecordId($record['rid']);
if ($details === null) {
    // record doesn't exist – reject
    return false;
}
$realZid = $details['zid'];
// use $realZid for verifyUserIsOwnerZoneId() and for zone-name normalization

Reject the edit if the record doesn't exist, and never let the client decide which zone a record "belongs" to.

Disclosure

Credit: Found by Saif Salah (https://saifsalah.github.io/) Found during a security review of Poweradmin.