Input Validation for Security
Filed under: Development, Security
Validating input is an important step for reducing risk to our applications. It might not eliminate the risk, and for that reason we should consider what exactly we are doing with input validation.
Should you be looking for every attack possible?
Should you create a list of every known malicious payload?
When you think about input validation are you focusing on things like Cross-site Scripting, SQL Injection, or XXE, just to name a few? How feasible is it to attempt to block all these different vulnerabilities with input validation? Are each of these even a concern for your application?
I think that input validation is important. I think it can help reduce the ability for many of these vulnerabilities. I also think that our expectation should be aligned with what we can be doing with input validation. We shouldn’t overlook these vulnerabilities, but instead realize appropriate limitations. All of these vulnerabilities have a counterpart, such as escaping, output encoding, parser configurations, etc. that will complete the appropriate mitigation.
If we can’t, or shouldn’t, block all vulnerabilities with input validation, what should we focus on?
Start with focusing on what is acceptable data. It might be counter-intuitive, but malicious data could be acceptable data from a data requirements statement. We define our acceptable data with specific constraints. These typically fall under the following categories:
* Range – What are the bounds of the data? ex. Age can only be between 0 and 150.
* Type – What type of data is it? Integer, Datetime, String.
* Length – How many characters should be allow?
* Format – Is there a specific format? Ie. SSN or Account Number
As noted, these are not specifically targeting any vulnerability. They are narrowing the capabilities. If you verify that a value is an Integer, it is hard to get typical injection exploits. This is similar to custom format requirements. Limiting the length also restricts malicious payloads. A state abbreviation field with a length of 2 is much more difficult to exploit.
The most difficult type is the string. Here you may have more complexity and depending on your purpose, might actually have more specific attacks you might look for. Maybe you allow some HTML or markup in the field. In that case, you may have more advanced input validation to remove malicious HTML or events.
There is nothing wrong with using libraries that will help look for malicious attack payloads during your input validation. However, the point here is to not spend so much time focusing on blocking EVERYTHING when that is not necessary to get the product moving forward. Understand the limitation of that input validation and ensure that the complimenting controls like output encoding are properly engaged where they need to be.
The final point I want to make on input validation is where it should happen. There are two options: on the client, or on the server. Client validation is used for immediate feedback to the user, but it should never be used for security.
It is too easy to bypass client-side validation routines, so all validation should also be checked on the server. The user doesn’t have the ability to bypass controls once the data is on the server. Be careful with how you try to validate things on the client directly.
Like anything we do with security, understand the context and reasoning behind the control. Don’t get so caught up in trying to block every single attack that you never release. There is a good chance something will get through your input validation. That is why it is important to have other controls in place at the point of impact. Input validation limits the amount of bad traffic that can get to the important functions, but the functions still may need to do additional processes to be truly secure.
XXE DoS and .Net
External XML Entity (XXE) vulnerabilities can be more than just a risk of remote code execution (RCE), information leakage, or server side request forgery (SSRF). A denial of service (DoS) attack is commonly overlooked. However, given a mis-configured XML parser, it may be possible for an attacker to cause a denial of service attack and block your application’s resources. This would limit the ability for a user to access the expected application when needed.
In most cases, the parser can be configured to just ignore any entities, by disabling DTD parsing. As a matter of fact, many of the common parsers do this by default. If the DTD is not processed, then even the denial of service risk should be removed.
For this post, I want to talk about if DTDs are parsed and focus specifically on the denial of service aspect. One of the properties that becomes important when working with .Net and XML is the MaxCharactersFromEntities property.
The purpose of this property is to limit how long the value of an entity can be. This is important because often times in a DoS attempt, the use of expanding entities can cause a very large request with very few actual lines of data. The following is an example of what a DoS attack might look like in an entity.
<?xml version="1.0" encoding="UTF-8"?> <!DOCTYPE foo [ <!ELEMENT foo ANY > <!ENTITY dos 'dos' > <!ENTITY dos1 '&dos;&dos;&dos;&dos;&dos;&dos;&dos;&dos;&dos;&dos;&dos;&dos;' > <!ENTITY dos2 '&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;&dos1;' > <!ENTITY dos3 '&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;&dos2;' > <!ENTITY dos4 '&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;&dos3;' > <!ENTITY dos5 '&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;&dos4;' > <!ENTITY dos6 '&dos5;&dos5;&dos5;&dos5;&dos5;&dos5;&dos5;&dos5;&dos5;&dos5;&dos5;&dos5;' >]>
Notice in the above example we have multiple entities that each reference the previous one multiple times. This results in a very large string being created when dos6 is actually referenced in the XML code. This would probably not be large enough to actually cause a denial of service, but you can see how quickly this becomes a very large value.
To help protect the XML parser and the application the MaxCharactersFromEntities helps limit how large this expansion can get. Once it reaches the max amount, it will throw a System.XmlXmlException: ‘The input document has exceeded a limit set by MaxCharactersFromEntities’ exception.
The Microsoft documentation (linked above) states that the default value is 0. This means that it is undefined and there is no limit in place. Through my testing, it appears that this is true for ASP.Net Framework versions up to 4.5.1. In 4.5.2 and above, as well as .Net Core, the default value for this property is 10,000,000. This is most likely a small enough value to protect against denial of service with the XmlReader object.
XSS in Script Tag
Filed under: Development, Security, Testing
Cross-site scripting is a pretty common vulnerability, even with many of the new advances in UI frameworks. One of the first things we mention when discussing the vulnerability is to understand the context. Is it HTML, Attribute, JavaScript, etc.? This understanding helps us better understand the types of characters that can be used to expose the vulnerability.
In this post, I want to take a quick look at placing data within a <script> tag. In particular, I want to look at how embedded <script> tags are processed. Let’s use a simple web page as our example.
<html> <head> </head> <body> <script> var x = "<a href=test.html>test</a>"; </script> </body> </html>
The above example works as we expect. When you load the page, nothing is displayed. The link tag embedded in the variable is rated as a string, not parsed as a link tag. What happens, though, when we embed a <script> tag?
<html> <head> </head> <body> <script> var x = "<script>alert(9)</script>"; </script> </body> </html>
In the above snippet, actually nothing happens on the screen. Meaning that the alert box does not actually trigger. This often misleads people into thinking the code is not vulnerable to cross-site scripting. if the link tag is not processed, why would the script tag be. In many situations, the understanding is that we need to break out of the (“) delimiter to start writing our own JavaScript commands. For example, if I submitted a payload of (test”;alert(9);t = “). This type of payload would break out of the x variable and add new JavaScript commands. Of course, this doesn’t work if the (“) character is properly encoded to not allow breaking out.
Going back to our previous example, we may have overlooked something very simple. It wasn’t that the script wasn’t executing because it wasn’t being parsed. Instead, it wasn’t executing because our JavaScript was bad. Our issue was that we were attempting to open a <script> within a <script>. What if we modify our value to the following:
<html> <head> </head> <body> <script> var x = "</script><script>alert(9)</script>"; </script> </body> </html>
In the above code, we are first closing out the original <script> tag and then we are starting a new one. This removes the embedded nuance and when the page is loaded, the alert box will appear.
This technique works in many places where a user can control the text returned within the <script> element. Of course, the important remediation step is to make sure that data is properly encoded when returned to the browser. By default, Content Security Policy may not be an immediate solution since this situation would indicate that inline scripts are allowed. However, if you are limiting the use of inline scripts to ones with a registered nonce would help prevent this technique. This reference shows setting the nonce (https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Content-Security-Policy/script-src).
When testing our applications, it is important to focus on the lack of output encoding and less on the ability to fully exploit a situation. Our secure coding standards should identify the types of encoding that should be applied to outputs. If the encodings are not properly implemented then we are citing a violation of our standards.
Does the End of an Iteration Change Your View of Risk?
Filed under: Development, Security, Testing
You have been working hard for the past few weeks or months on the latest round of features for your flagship product. You are excited. The team is excited. Then a security test identifies a vulnerability. Balloons deflate and everyone starts to scramble.
Take a breath.
Not all vulnerabilities are created equal and the risk that each presents is vastly different. The organization should already have a process for triaging security findings. That process should be assessing the risk of the finding to determine its impact on the application, organization, and your customers. Some of these flaws will need immediate attention. Some may require holding up the release. Some may pose a lower risk and can wait.
Take the time to analyze the situation.
If an item is severe and poses great risk, by all means, stop what you are doing and fix it. But, what happens when the risk is fairly low. When I say risk, I include in that the ability for it to be exploited. The difficulty to exploit can be a critical factor in what decision you make.
When does the risk of remediation override the risk of waiting until the next iteration?
There are some instances where the risk to remediate so late in the iteration may actually be higher than waiting until the next iteration to resolve the actual issue. But all security vulnerabilities need to be fixed, you say? This is not an attempt to get out of doing work or not resolve issues. However, I believe there are situations where the risk of the exploit is less than the risk of trying to fix it in a chaotic, last minute manner.
I can’t count the number of times I have seen issues arise that appeared to be simple fixes. The bug was not very serious and could only be exploited in a very limited way. For example, the bug required the user’s machine to be compromised to enable exploitation. The fix, however, ended up taking more than a week due to some complications. When the bug appeared 2 days before code freeze there were many discussions on performing a fix, and potentially holding up the release, and moving the remediation to the next iteration.
When we take the time to analyze the risk and exposure of the finding, it is possible to make an educated decision as to which risk is better for the organization and the customers. In this situation, the assumption is that the user’s system would need to be compromised for the exploit to happen. If that is true, the application is already vulnerable to password sniffing or other attacks that would make this specific exploit a waste of time.
Forcing a fix at this point in the game increases the chances of introducing another vulnerability, possibly more severe than the one that we are trying to fix. Was that risk worth it?
Timing can have an affect on our judgement when it comes to resolving security issues. It should not be used as an escape goat or reason not to fix things. When analyzing the risk of an item, make sure you are also considering how that may affect the environment as a whole. This risk may not be directly with the flaw, but indirectly due to how it is fixed. There is no hard and fast rule, exactly the reason why we use a risk based approach.
Engage your information security office or enterprise risk teams to help with the analysis. They may be able to provide a different point of view or insight you may have overlooked.
Follow Us
