Understanding the Android clearTextTrafficPermitted Flag

Introduction

The cleartextTrafficPermitted flag is one of the options in Android’s Network Security Configuration file. The online documentation (https://developer.android.com/training/articles/security-config) explains that from Android 9 (API level 28) and higher, it will be set by default to false and it is intended to prevent insecure communication attempts using clear-text HTTP originating from Android applications.

OK, so what does this actually mean? Will all apps that run on the Android 9 platform with a default configuration be protected? Do these flags instruct the system to prohibit any kind of non-secure TCP communications or does it only apply to HTTP? What about 3rd party libraries such as OkHttp, async-http-client, dubbo? What about WebView controllers? What about native libraries issuing network requests? To answer all these questions, we need to understand at which level this configuration is being enforced. Otherwise, we may fall into a false sense of security.

But how does it actually work? What component and at which level enforces these restrictions? Is it a kernel module such as SELinux or AppArmor? It is Linux containers? Is it the Android Runtime environment? Is it a compiler flag that acts at the time of Ahead-Of-Time or Just-in-time compilation to strip the insecure code parts?

In this article, we will try to understand how this flag works by example and how it can help us to make a more secure application. We will also learn by example which use-cases aren’t being protected by the flag.

The importance of TLS

Just before we begin, a short recap on why we need to protect all application communication with TLS.

Potential threat agents include:

  • Wi-fi layer attackers – Numerous router level vulnerabilities, wi-fi hacking, and results of our private research show that tens of millions of routers could be compromised. As such, even home wi-fi cannot be considered as a secure by default. The app must maintain security at the network layer to provide secure transport for app-server communications.
  • WAN layer attackers – ISPs logging network traffic or APT and Government level actors monitoring communications at central network hops.

Compromised communications can lead to a number of possible risks including:

  • Hijacking secrets from client-side libraries via server-initiated redirects (We will talk about this more in the next blog post)
  • HTTPS connections can be downgraded to HTTP even if a server does not support HTTP
  • Open redirects to HTTP, both client-side initiated and server-side triggered (for example, the HTTP Desync Attack)
  • HTML injected resources (img) with http:// link that trigger HTTP communication
  • In-app user-provided clear-text HTTP link to app’s domain
  • Cache poisoning resulting in clear-text HTTP redirects

Creating a test-bed

To check how applications actually behave, we created an Android testing application and a python server for the application to communicate with. The test included trying different target APIs for the application and checking for each target API if the app could issue non-secure outgoing HTTP requests using different libraries as well as the default Android WebView controller.

For each library, three requests were sent: standard HTTPS, standard HTTP, and an HTTPS request to a URL which redirects the user agent to HTTP to simulate server-side-issued redirect attacks.

Testing notes

  1. Test results for HTTPS to HTTP redirects were the same as for HTTP. If in normal conditions the library prohibits HTTP connections it will also prohibit redirects to HTTP.
  2. Differences between minimal, target, and compile SDK versions from the manifest file:
    1. Minimal API – minimal API support required for the app to work;
    2. Compile API – IDE compiler used to compile the app
    3. Target API – the API level on which the application should run correctly as it was fully tested by the developers.
  3. Minimum SDK has no impact on how the cleartextTrafficPermitted flag is being enforced; only the API target level affects it.
  4. Most of the security issues arise on outdated Android OS versions. However, these versions are running on a large number of devices according to Google:

The results

This section will go through the results for the different libraries tested.

OkHttp

OkHttp is the most popular Android HTTP library and is maintained by Square: https://square.github.io/okhttp/. For the test the latest 3.* and 4.* versions were used. The results for both versions were the same.

The following table shows the results of using our testing app with different target SDK configurations and without the clearTextTrafficPermited flag set (i.e. the default configuration) on multiple Android devices of different OS versions.

 Android 5 API 21Android 6 API 23Android 7.1 API 25Android 8 API 26Android 9 API 28Android 10 API 29
target API 21-27 HTTP works HTTP works HTTP works HTTP works HTTP works HTTP works
target API 28+ HTTP works HTTP works HTTP works HTTP worksHTTP doesn’t workHTTP doesn’t work

The following table shows the results of using our testing app with different target SDK configurations and with the clearTextTrafficPermited flag set to “false” on multiple Android devices of different OS versions.

<base-config cleartextTrafficPermitted=“false”>


 Android 5 API 21Android 6 API 23Android 7.1 API 25Android 8 API 26Android 9 API 28Android 10 API 29
target API 21-27 HTTP works HTTP worksHTTP doesn’t workHTTP doesn’t workHTTP doesn’t workHTTP doesn’t work
target API 28+ HTTP works HTTP worksHTTP doesn’t workHTTP doesn’t workHTTP doesn’t workHTTP doesn’t work

The following table shows the results of using our testing app with different target SDK configurations and with the clearTextTrafficPermited flag set to “true” on multiple Android devices of different OS versions.

<base-config cleartextTrafficPermitted=“true”>


 Android 5 API 21Android 6 API 23Android 7.1 API 25Android 8 API 26Android 9 API 28Android 10 API 29
target API 21-27 HTTP works HTTP works HTTP works HTTP works HTTP works HTTP works
target API 28+ HTTP works HTTP works HTTP works HTTP works HTTP works HTTP works

When the network security policy is being correctly enforced, the OkHttp library won’t issue the request and will throw the following exception:

We can find multiple instances of checking the flag in OkHttp’s source code (https://github.com/square/okhttp) and can manually verify at which cases the check is being done. However, understanding the libraries internal logic may be required to identify the edge cases in which the flag could be ignored. The screenshot shows the source code for the relevant part of the connection chain:

Other Popular Network Libraries for Android

This section shows results for async-http-client library but the results were the same for multiple other libraries that were tested. For the most part, libraries which use okhttp “behind the scenes” support the flag but other popular Android libraries which don’t use okhttp (like async-http-client and many others) don’t support the flag at all.

The following table shows the results of using our testing app with different target SDK configurations with the clearTextTrafficPermited flag set to “false” on multiple Android devices of different OS versions. Since the flag is unsupported, the behavior is in fact the same as if the flag wasn’t set or if the flag was set to “true”.

<base-config cleartextTrafficPermitted=“false”>


 
Android 5 API 21

Android 6 API 23


Android 7.1 API 25

Android 8 API 26

Android 9 API 28

Android 10 API 29
target API 21-27 HTTP works HTTP works HTTP works HTTP works HTTP works HTTP works
target API 28+ HTTP works HTTP works HTTP works HTTP works HTTP works HTTP works

Android WebView

In this section we tested the default Android WebView . This test showed that in fact the WebView controller behaves differently to the OkHttp library in the context of enforcing the configuration flag. In particular:

  1. When the application’s target API is 25 or less, the “cleartextTrafficPermitted=”false”” is being ignored across all tested Android OS versions.
  2. When HTTP redirect occurs, the WebView behaves differently – it forwards the request to the device’s default browser (external) instead of rendering it on its own. This is secure behavior since the browser doesn’t send any sensitive cookies because it runs in a different context to the application’s WebView.

The following table shows the results of using WebView in our testing app with different target SDK configurations and without a value for the clearTextTrafficPermited flag set (default configuration) on multiple Android devices of different OS versions.

 Android 5 API 21Android 6 API 23Android 7.1 API 258 API 269 API 2810 API 29
target API 23-27HTTP worksHTTP worksHTTP worksHTTP worksHTTP worksHTTP works
target API 28+HTTP worksHTTP worksHTTP worksHTTP worksHTTP doesn’t workHTTP doesn’t work

The following table shows the results of using WebView in our testing app with different target SDK configurations and with the clearTextTrafficPermited flag set to “false” on multiple Android devices of different OS versions.

 Android 5 API 21Android 6 API 23Android 7.1 API 258 API 269 API 2810 API 29
target API 23-25HTTP worksHTTP worksHTTP worksHTTP worksHTTP worksHTTP works
target API 26-27HTTP worksHTTP worksHTTP doesn’t workHTTP doesn’t workHTTP doesn’t workHTTP doesn’t work
target API 28+HTTP worksHTTP worksHTTP doesn’t workHTTP doesn’t workHTTP doesn’t workHTTP doesn’t work

When the network policy is being correctly enforced, the WebView will show the following image

Configuration recipes

In this post, we won’t cover all configuration options, but only the main configuration options that should be understood. Full information can be found at https://developer.android.com/training/articles/security-config

To configure the clearTextTrafficPermited flag, we need to set the network security config at the manifest file level:

<application

    android:networkSecurityConfig=“@xml/network_security_config”>

We then need to populate the network_security_config.xml file with the relevant settings.

We have included below some examples of the network config file (network_security_config.xml) and the actual meaning of the settings:

  1. Disables all insecure outgoing connections – should be a default choice

<?xml version=“1.0” encoding=“utf-8”?>

 

<network-security-config>

    <base-config cleartextTrafficPermitted=“true”>

    </base-config>

</network-security-config>


  1. Allows HTTP traffic only to appdomain.com domain. All other domains will only be allowed via HTTPS.

<?xml version=“1.0” encoding=“utf-8”?>

 

<network-security-config>

    <base-config cleartextTrafficPermitted=“false”>

    </base-config>

    <domain-config cleartextTrafficPermitted=“true”>

            <domain includeSubdomains=“true”>appdomain.com</domain>

    </domain-config>

</network-security-config>


  1. Forbids any insecure traffic to appdomain.com but allows insecure connection for all other domains (base-config will be default):

<?xml version=“1.0” encoding=“utf-8” ?>

<network-security-config>

        <domain-config cleartextTrafficPermitted=“false”>

               <domain includeSubdomains=“true”>appdomain.com</domain>

        </domain-config>

</network-security-config>


  1. Forbids any insecure traffic to appdomain.com; allow explicitly all insecure connection for all other domains:

<?xml version=“1.0” encoding=“utf-8”?>

 

<network-security-config>

    <base-config cleartextTrafficPermitted=“true”>

    </base-config>

    <domain-config cleartextTrafficPermitted=“false”>

            <domain includeSubdomains=“true”>appdomain.com</domain>

    </domain-config>

</network-security-config>


Results Quick Q&A

In this section we will summarize some of the key questions which we have answered so far in our research:

  • What happens when the app communicates using HTTP and makes use of a framework/library which does not support this manifest flag
    • No restrictions will be in place
  • At which level does the manifest restrict clear-text communications?
    • Each library reads the flag from configuration flags and enforces the restrictions but only if it supports the flag.
  • What happens when native code issues a clear-text network request? For example, it sends an HTTP request or sends binary data (such as protobuf)? Will it be protected by this manifest configuration?
    • Only if the library supports the flag. In most cases these libraries won’t support the flag. The native code libraries and non-HTML libraries mostly have no flag awareness and support. You should rely only on how you configured the library. These configs either can be done in a “config” file or at the actual network request object properties and methods.
  • What happens when these libraries issue requests using HTTP and nonstandard port?
    • Generally, if the library supports the flag it will block all outgoing cleartext HTTP communication regardless of the port.

Result analysis and wrap-up

Multiple tests show that manifest flag clearTextTrafficPermited is not being universally enforced by either the Android OS or at the SDK level. IN practice, each client-side library must implement support for the flag on its own and the actual implementation may heavily differ from one library to another.

Certain libraries such as OkHttp (or Volley which uses OkHttp behind the scenes) have full support for this flag according to their documentation. However, many other HTTP libraries which do not use OkHttp do not support it at all and the presence of the flag has no impact on the app’s behavior.

The Android WebView component has support for the flag but actual results show inconsistencies between the WebView and OkHttp implementations. In fact, Android WebView puts more clients at risk compared to OkHttp (refer to the WebView result table) as it does not enforce the flag unless target API is 26+.

Regarding other clients, such as binary communication libraries, etc, do not assume that these libraries will respect the flag. Instead, the source code of each such library should be checked for the code part that enforces the restrictions on outgoing insecure communications. The default assumption for all client-side network libraries should be that they don’t support the flag unless the library explicitly states that it does.

Despite the lack of actual system-wide level protection, the flag still offers a greater level of security in situations where it is supported by the client library and either the application’s target SDK is set to 28+ or the application is running on Android 7.1 or higher.

Recommendations:
  1. Activate the clearTextTrafficPermited protection by setting it to “false” for all outgoing connections.
  2. Set the target API level to 28+.
  3. Check actual support for the flag across different libraries in use in your application.
  4. Be aware that the flag enforcement has its restrictions.
  5. Enforce other supplementary transport security measures, such as:
    • Using secure response headers such as HSTS.
    • Using explicit request object configurations to prohibit HTTP.
    • Implementing custom interception layers for libraries being used which will block clear-text URI schemes.
    • Removing any HTTP:// links from the code and server’s configuration.

Angular Template Injection without Quote Characters

Introduction

When you’re trying to detect XSS in an Angular application (AngularJS or Angular >2, the version doesn’t matter for this post), you’ll probably try the following possibilities:

  • You enter a basic XSS payload such as <script>alert(0);</script>
  • You find out that the developer encoded the output properly and the < > characters turn into &lt; and &gt; 
  • Your next attempt will be a client-side template injection: {{1338-1}}
  • The application displays 1337
  • From here on, it will be straight-forward: you’ll find a payload that matches the Angular version of the app and pop that alert to get a Proof of Concept.

However, in some cases, the output sanitation applies not only to characters like > and < but also encodes single and double quote characters ( ‘ will turn into &apos;, etc.)

This means that if you try to insert a payload that has quotes in it, for example:

{{constructor.constructor('alert(0)')()}}

It will turn into

{{constructor.constructor(&apos;alert(0)&apos;)()}}

At this point, your expression will trigger a syntax error and the payload will fail. 

So, what can we do about it?

When trying to avoid quotes in a payload (because it contains strings) the easiest way is to use functions that generates strings. In JavaScript, we will need to use the String object, specifically, the String.fromCharCode() method

The issue in this case (Angular applications) is that we cannot just call String.fromCharCode()in an Angular expression:

{{constructor.constructor(String.formCharCode(...))()}}

It will not work because Angular limits us to only using the properties and variables that are accessible under Angular’s scope. (More about the scope here: https://github.com/angular/angular.js/wiki/Understanding-Scopes). If Angular wasn’t stopping us from accessing objects outside the scope, we wouldn’t need to access the Stringobject. As attacker, we could just inject {{window.alert(0)}}

So what do we do when we don’t have access to the Stringobject but we still need to call fromCharCode() to generate our “alert(0)” string?

We can use this payload:

{{constructor.constructor(valueOf.name.constructor.fromCharCode(...))()}}

Let’s break it down

  • We used valueOf.name because it returns a string value. Right now, it doesn’t matter what string is returned. As long as it’s a string variable – we’re good to go. In the same way, we could use any of the following possibilities as they all generate strings:
  • Now that we have a string variable in our payload, let’s look at its constructor:
  • The constructor itself is a String object! Now, all we have to do is to add a dot character and access all of the available functions in the String object including fromCharCode

By getting to the String object in this indirect way, even though the object itself isn’t in Angular’s scope, we can create an Angular, Client-Side Template Injection payload without needing single or double quotes

And here’s our alert 🙂 without quotes.

Acknowledgements

Big shout-out goes to the folks from PortSwigger and Tomer Haddad


A Taxonomy on Brute Force Attacks

Brute force attack is a well-known technique of trial and error attempts used by attackers to gain access to unauthorized data. It can be leveraged against servers as an online attack and also against files as a local attack.

The common denominator of all these types is that the same pattern is almost always the same:

In most cases the attacker would have to know two major keys from the diagram – let’s take for example a common scenario where the attacker tries a brute-force attack on the login interface of an applicaiton in order to find the correct credentials of a certain account. The known keys would be:

  1. Something I know – In this case, the username of the account that the attacker wants to hijack.
  2. The result – On each attempt, the attacker receives an indicator of failed/successful result.

The well-known brute-force methods are:

Read more

Secure Development Lifecycle for Open Source Usage

Preface How do we adjust the SDL (Security Development Lifecycle) process for the growing use of open source in internal/external systems we develop and maintain? This is a question I hear a lot lately from our customers in some recent SDL projects we (AppSec Labs) carried out for our customers. After we did some research, […]

Hacking Android Apps Through Exposed Components

In almost every Android application, developers expose activities without sufficient protections. Exposing activities can lead to various attacks. For example, an attacker or a malicious app installed on the same device, can call those exposed activities to invoke internal pages of the application. Calling internal pages puts the application at risk of phishing by manipulating […]

Guidelines for pentesting security systems

How do you perform a security test? Ever wanted to perform a penetration test on a security system (website, application, mobile and etc.) and just didn’t know where to start? These days, companies are very concerned about the security in their applications due to the increase in the number of disclosed vulnerabilities and exploits which can […]

Password Autocomplete vulnerability and a workaround solution

Until recently, it was trivial for developers to disable the “save you password” feature implemented by all major browsers. However, in the last years, browser vendors have begun to actively discourage and prevent applications from disabling this feature. Their case is that the safest practice for users is to use a password manager, instead of having their passwords lying around on digital or physical support, where they can be exposed or stolen. Since it’s a client-side issue, they claim that the option should be given to users (and not to the developers) to disable this feature by configuring the browser itself.

Although this may be partly true, it does not take into account highly sensitive applications, which might be used on a shared computer, and which do not want to rely on the browser being properly configured (with autocomplete disabled). If this is your case, you should keep on reading.

It is now a real challenge to find a workaround that will work across all major browsers. So we came up with the following trick which detects the user’s browser version and acts accordingly:

Read more