Thursday, 10 September 2015

.NET MVC ReDoS (Denial of Service) Vulnerability - CVE-2015-2526 (MS15-101)

submit to reddit Vote on Hacker News Share
Microsoft released a security bulletin (MS15-101) describing a .NET MVC Denial of Service vulnerability (CVE-2015-2526) that I reported back in April. This blog post analyses the vulnerability in details, starting from the theory and then providing a PoC exploit against a MVC web application developed with Visual Studio 2013.
For those of you who want to see the bug, you can directly skip to the last part of this post or watch the video directly... ;-)

A bit of theory

The .NET framework (4.5 tested version) uses backtracking regular expression matcher when performing a match against an expression. Backtracking is based on the NFA (non-deterministic finite automata) algorithm engine which is designed to validate all input states. By providing an “evil” regex expression – an expression for which the engine can be forced to calculate an exponential number of states - it is possible to force the engine to calculate an exponential number of states, leading to a condition defined such as “catastrophic backtracking” aka ReDoS.

The vulnerability

In .NET Framework (4.5), “evil” regular expressions are used by default in three classes (EmailAddressAttribute, PhoneAttribute, UrlAttribute) which are part of System.CompontentModel.DataAnnotations .NET library.

These classes provide the default validation mechanism for email address, phone number and URL input types in web forms. Furthermore, these three classes do not enforce a regex match timeout.

The following screen shots show the evil regex and the lack of match timeout:

EmailAddressAttribute Source code 

PhoneAttribute Source Code

UrlAttribute Source Code

As a consequence, an attacker can craft a malicious payload to force the .NET regex engine to perform a large number of computations and cause a Denial of Service against the targeted controller (e.g. login form) which uses default validation mechanism provided by .NET framework.

The Denial of Service condition is only specific to the target class controller (e.g. login form, registration form, contact form, etc.). Users can still potentially navigate the site but they are prevented from using parts of it.

As an example, the .NET email address regex is analyzed. Its regex expression is considered an “evil” regex, due to its complexity, repetition, nesting and recursion. The regex is reported in the screen shot below. The software RegexBuddy was used to analyze it.

The theory of the attack is demonstrated below, with the help of RegexBuddy and its built-in debugger (set for C# - .NET 2.0-4.5) - with payload (in the table below) which will never match the above regex:


An extract of the last 26 operations (stopped by RegexBuddy) can be found below, from the Debugger view:

This shows the “catastrophic backtracking” condition reached by the matcher. In this case, RegexBuddy stops calculations after one million steps, however, the vulnerable class – EmailAddressAttribute - does not enforce a match timeout and therefore the .NET regex engine continues to compute steps, leading the w3wp.exe process (IIS Worker Pool) on the web server to reach a 99% CPU starvation condition for an extended amount of time, which can last various hours to days, depending on the payload used.

The payload can be constructed in different ways, providing the attacker with the capability to bypass IDS/IPS signature based controls. The attacker can set scripts to automatically attack vulnerable forms on a regular time basis.

The exploit

The exploitation consists in sending a crafted HTTP POST request against a web form using a vulnerable class (e.g. EmailAddressAttribute). As an example, the attack is demonstrated against a .NET MVC web application developed with the Visual Studio 2013. The application provides a login form which uses the default email address validation mechanism in .NET framework. The screen shot below shows the login page:

An attacker can bypass client-side validation in .NET by sending the request via script or proxy and manipulating the request, as shown below:

POST /Account/Login HTTP/1.1
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:36.0) Gecko/20100101 Firefox/36.0
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
Cookie: __RequestVerificationToken=FkLGrc6-XD2IBVU9g1nPycs0GTu3jWiK2QEyvR8IsowXAJU3C5fHlHvQvwGgB0VcN1FTa_hB9KZ6Pi8SeI5EKpvz_EeOqD7y_FnipWJWqOU1
Connection: keep-alive
Content-Type: application/x-www-form-urlencoded
Content-Length: 239


Below, an extract of the source code used for the validation of the EmailAddress field:

AccountModelView.cs - use of [EmailAddress] default class in .NET

   public class LoginViewModel
        [Display(Name = "Email")]
        public string Email { get; set; }

AccountController.cs – ModelState is validated when  the POST request occurs

// POST: /Account/Login
        public async Task<ActionResult> Login(LoginViewModel model, string returnUrl)
            if (!ModelState.IsValid)
                return View(model);

            // This doesn't count login failures towards account lockout
            // To enable password failures to trigger account lockout, change to shouldLockout: true
            var result = await SignInManager.PasswordSignInAsync(model.Email, model.Password, model.RememberMe, shouldLockout: false);
            switch (result)

The table below shows the DoS condition on the web server, after the request has been issued.

Following the request, the Denial of Service occurs against the /Account/Login controller class. At this stage, no other users can use /Account/Login form controller class, while the w3wp.exe process is at 99% CPU starvation.

The w3wp.exe process needs to be terminated in order to recover the application from the attack. After few manual recoveries, the application becomes unusable, and the server needs to be restarted.

Below a video that demonstrates the attack in action:

The table below includes valid and tested attack patterns which result in a successful ReDoS attack against .NET applications:

Malicious Payload


Monday, 27 April 2015

Pwning a thin client in less than two minutes

submit to reddit Vote on Hacker News Share

Have you ever encountered a zero client or a thin client? It looks something like this...

If yes, keep reading below, if not, then if you encounter one, you know what you can do if you read below...

The model above is a T520, produced by HP - this model and other similar models are typically employed to support a medium/large VDI (Virtual Desktop Infrastructure) enterprise.

These clients run a Linux-based HP ThinPro OS by default and I had a chance to play with image version T6X44017 in particular, which is fun to play with it, since you can get a root shell in a very short time without knowing any password...

Normally, HP ThinPro OS interface is configured in a kiosk mode, as the concept of a thin/zero client is based on using a thick client to connect to another resource. For this purpose, a standard user does not need to authenticate to the thin client per se and would just need to perform a connection - e.g. VMware Horizon View. The user will eventually authenticate through the connection.

The point of this blog post is to demonstrate that a malicious actor can compromise such thin clients in a trivial and quick way provided physical access, a standard prerequisite in an attack against a kiosk.

During my testing, I have tried to harden as much as possible the thin client, with the following options:

I did not set the "Allow user to lock screen" to simulate a scenario where users can use any thin-clients (kiosk style). However, I have also noticed that the default password for the account "user" is "user", so if you find an environment where they enforce account lockout, you can try that password directly (it is very often unchanged...)

I also set a password for the administrator's view, so when a user attempts to switch to the admin view, a password would be required.

In this scenario, the standard user does not know this password and should only be able to use a single VMware Horizon View pre-configured connection, as shown below:

However, I have found out that unless there is user lockout enabled, then it is possible to get a root shell following the steps below:

- Select the connection profile, and edit the profile (if it auto-starts, then you should cancel the connection)

- A new window with a form is presented to you - fill the server field with dummy data and then click on the General options

- Perform the attack to "escape" from the ThinPro Control Center kiosk by entering under "Command Line Arguments": && xterm

- Click Ok to save the new "VMware Horizon View" profile

- Click on new "VMware Horizon View" profile and the connection will timeout/fail, as dummy data was entered. However, when you close/cancel the window, the xterm window will be spawn

- You have bash shell access to the HP Zero client

The user id is user, so no root yet.

However, if you check sudo -ll you see that by default, the account user can perform a lot of commands as root without the need to enter a password. The output of sudo is included in this pastebin and an exerpt below:


Sudoers entry:
    RunAsUsers: root
NOPASSWD: /usr/bin/hpobl


The most interesting command I have found is: /usr/bin/hpobl

This command allows access to the HP Easy Setup Wizard panel. Through this one, it is possible to change settings. However, in this scenario, you do not know any password, so you must find a way to get a root shell from this Wizard panel.

By going directly to the last step "Thank you" (all the previous steps can be ignored for the lolz), then click on the link - this will spawn Firefox to load the link you just clicked.

At this stage, you have launched HP Easy Setup Wizard as root and Firefox process is also launched as root. One elegant option to get a shell of Firefox is to set Firefox's external mail handler - Edit / Preferences / Applications / Mailto - and point it to /usr/bin/xterm

Then you just need to point a tab to: and you will then be gratified with a root shell:

Here is a video of the entire attack which takes less than two minutes, in respect of the title of this blog post (in reality, the title was " less that one minute", but couldn't do it in that time frame lol...if you do it in less than one minute let me know...):

So what's the catch?

If you are performing a penetration testing against a VDI, look for quick wins as this one... if you are responsible for VDI, then consider that those machines can be compromised very easily - a soft key logger will be enough to get credentials to a Windows domain... also, check what configuration is enforced on the thin-client itself. It might even be more relaxed than the one considered in this scenario.

If you like kiosks and more in particular you like to break them, then you absolutely need to try: . Greetz to Paul Craig, the "self-proclaimed" king of kiosks! ;-)

I haven't tried other HP Thin Pro images yet, but it might be possible that the attack shown in this blog also affects versions earlier than T6X44017.

If you find other ways to bypass HP ThinPro OS, please let me know.