Wickr account cloning vulnerability

USB backup attacks against Android applications

 Android allows users to make online backups of their application data. This happens silently based on settings created by the applications' developers (http://developer.android.com/google/backup/index.html). According to the documentation, the data is held in "cloud storage" (the exact definition will vary based on the OEM), allowing users to restore their data when they get a new phone.

 But the backup function also allows users to make backups over USB using the "adb backup" command. This command will backup data from all applications that have not explicitly set the allowBackup flag to false in their manifest. The data that is retrieved is taken from the sandbox of the applications, in escence giving users access to the contents of their applications' sandboxes without being root. This is something that the shell user can not normally do. For an excellent writeup as to how to access, edit and restore data using adb backup by Nikolay Elenkov see here: http://nelenkov.blogspot.co.uk/2012/06/unpacking-android-backups.html

 As far as an attack vector it is often contrived and unrealistic. Essentially an attacker would need to have physical access to an unlocked phone. If they have this then they could do all manner of nasty things, like replace applications, install trojans or just ringing their premium rate phone line.

 For a number of applications however, the vector is more reasonable. The value of the assets is high and they can only be accessed from the sandbox. Below are three applications that can be compromised using the ADB backup restore functionality:


MyWicker (com.mywickr.wickr) - 1.9.8.8 Beta
F-Secure Freedome VPN (com.fsecure.freedome.vpn.security.privacy.android) - 1.04
OpenVPN(net.openvpn.openvpn) - 1.1.12

OpenVPN / F-Secure Freedome VPN

Both the OpenVPN and F-Secure Freedome were found to allow backups over ADB

According to Android documentation:

android:allowbackup
Whether to allow the application to participate in the backup and restore infrastructure. If this attribute is set to false, no backup or restore of the application will ever be performed, even by a full-system backup that would otherwise cause all application data to be saved via adb. The default value of this attribute is true.

Performing a backup on the Freedome app we can see the following:

androidvm@androidvm:/tmp$ adb backup com.fsecure.freedome.vpn.security.privacy.android
Now unlock your device and confirm the backup operation.
androidvm@androidvm:/tmp$ dd if=backup.ab bs=1 skip=24| openssl zlib -d > backup.tar
8726+0 records in
8726+0 records out
8726 bytes (8.7 kB) copied, 0.0134311 s, 650 kB/s
androidvm@androidvm:/tmp$ tar -xaf backup.tar
androidvm@androidvm:/tmp$ cd apps/com.fsecure.freedome.vpn.security.privacy.android/
androidvm@androidvm:/tmp/apps/com.fsecure.freedome.vpn.security.privacy.android$ ls -la
total 24
drwxrwxr-x 5 androidvm androidvm 4096 Jan 14 16:15 .
drwxrwxr-x 3 androidvm androidvm 4096 Jan 14 16:15 ..
drwxrwxr-x 2 androidvm androidvm 4096 Jan 14 16:15 f
-rw------- 1 androidvm androidvm 1506 Jan 14 16:15 _manifest
drwxrwxr-x 2 androidvm androidvm 4096 Jan 14 16:15 r
drwxrwxr-x 2 androidvm androidvm 4096 Jan 14 16:15 sp
androidvm@androidvm:/tmp/apps/com.fsecure.freedome.vpn.security.privacy.android$ ls f/
clp-beta-prov.client.crt  clp-beta-prov.client.key  clp-beta-prov.servers.conf  clp-ca-server.pem  pvpn.conf

We can see that there are a number of files available for us either to just read, or to alter and restore.

For Freedome, we are able to overwrite the configuration and key files used by the application. This would mean that every time the user then ran this application, they would connect to our malicious VPN. This would allow us to MITM the user's traffic.

The OpenVPN looked almost identical in terms of what we could access and change. This would give us access to their private VPN keys and configuration, letting us connect to the user's VPN. It would also allow us to overwrite the VPN settings as per Freedome and have them connect to our malicious VPN without them realising.

Wickr

Wickr was found to also allow backups by not explicitly disabling it

MyWickr allows users to select the "remember me" option. If this is selected (the user must accept a warning), then they can log in to the app without entering their password. When they choose this option, two extra files are created in the sandbox (pcc.wic and pcd.wic). These can be accessed using the ADB backup.

androidvm@androidvm:/tmp/com.mywickr.wickr$ ls
db  f  _manifest  r  sp
androidvm@androidvm:/tmp/com.mywickr.wickr$ ls f/
icu  magic.mgc  pcc.wic  pcd.wic

This would allow an attacker to clone an account by performing and ADB backup on the victim device and then an ADB restore on to a different device. A safe guard against this is that Wickr's functionality relies on the Android device ID for decryption. Unfortunately, Wickr store the device ID when the app is first run in a shared preferences file. As this is copied as part of the backup, the security control doesn't work.

It is therefore possible for an attacker to clone a Wickr account on to a different device using the ADB backup/restore functionality. Below is a video demo-ing a wickr account on the device to the left being cloned on to a device on the right.

Communication with Wickr

19-01-2014 - Emailed Wickr with details of vulnerability
12-02-2014 - Wickr replied accepting issue as a valid vulnerability
15-04-2014 - Wickr permitted vulnerability to be disclosed

Android to Arduino: the code

In a follow up to the Android to Arduino post, here's the code used. The code currently takes a command from an Android app (a button press), and sends a value to the Arduino over serial. In the mean time the arduino is sending a "nop" string back to the app every second. Not much, but enough to prove input & output over serial.
Video of it working here
Arduino code:

 #include  String inData;
Servo myservo;
int pos=0;

// the setup routine runs once when you press reset:
void setup() {
  // initialize serial communication at 115200 bits per second:
  Serial.begin(115200);
  myservo.attach(14);
}

// the loop routine runs over and over again forever:
void loop() {
      while (Serial.available() > 0)
    {
        char recieved = Serial.read();
        inData += recieved; 

        // Process message when new line character is recieved
        if (recieved == '\n')
        {
            Serial.print("Arduino Received: ");
            Serial.print(inData);
           
            Serial.print("\n");
            
            inData.trim();
            int dataAsInt=inData.toInt();
            if(dataAsInt>0 && dataAsInt<360 buffer="" clear="" dataasint="" delay="" indata="" myservo.write="" nop="" pre="" recieved="" serial.println="">

The Java app is based on the example from https://github.com/mik3y/usb-serial-for-android/tree/master/UsbSerialExamples but has a button with an onClick function with mSerialIoManager.writeAsync("10\n".getBytes());

Linking Android to Arduino

I've started our latest project (to be unveiled soon). It requires Arduino to receive commands from an Android device. It needs to be as light as possible, so ideally needs to be powered from the USB connection.

I've seen a number of solutions using Android ADK and special Arduinos, but this seems rather unnecessary. The problem seems to be the USB communication on either side, requiring USB sheilds (http://www.freetronics.com/products/usbdroid#.Upz8jx8hG1E). Alternatively there's Bluetooth (http://blog.arduino.cc/2013/07/18/how-to-control-arduino-board-using-an-android-phone/), but that would mean an external power source.

So instead, just use a Serial interface. There are Java libraries that can be used on Android. Everything including examples can be found here: https://github.com/mik3y/usb-serial-for-android

 I've hooked up my arduino Uno to an SIII using an A to B USB cable and OTG cable. It all works nicely. Stay tuned for what we're going to use it for ;-)

How Many Apps are on my phone?

I've seen several people posting about how they have "over a hundred apps" on their phone. I don't think this is accurate. There are a lot of apps installed by default that the user doesn't see (and can't uninstall). Interestingly, I couldn't find a good way to display the total number of packages (apps), installed on a device using the GUI. You can list apps using Settings->Apps but this is incomplete.

Instead using a USB cable and ADB it's possible to get a clearer picture.

So the total number is more like over 300

automated odex to dex script

When grabbing a new Android device, most applications' dex files have been optimised for the device. A few people have written a guide to using smali and baksmali in this situation (http://forum.xda-developers.com/showthread.php?t=1208320), but here's a handy script to do it all. You'll need smali.jar and baksmali.jar in the local directory when you run it.

echo "+------------------------------+"
echo "+                              +"
echo "+       mass de-odexer         +"
echo "+         TROTMASTER           +"
echo "+                              +"
echo "+------------------------------+"

mkdir BOOTCLASSPATH 2>/dev/null
mkdir tobedeodexed 2>/dev/null
mkdir deodexed  2>/dev/null
#clear out the previous contents COMMENT THIS OUT IF YOU ARE USING THE SAME BOOTCLASSPATH etc.
rm BOOTCLASSPATH/* 2>/dev/null
rm tobedeodexed/*  2>/dev/null
rm deodexed/* 2>/dev/null

#get bootclasspath files TODO:get only .jar files
echo "pulling boot class path files..."
adb pull /system/framework/ BOOTCLASSPATH/ > /dev/null 2>&1

#get odexes
echo "pulling .odex files..."
for i in $(adb shell ls /system/app/*.odex | sed -e 's/odex./odex/'); do adb pull $i tobedeodexed/$(echo $i|cut -d"/" -f 4)>/dev/null 2>&1;done

#Now run smali then baksmali to change odex to dex. may need -a to change api level
for i in $(ls tobedeodexed/*.odex);do rm -R out/ 2>/dev/null; echo baksmali-ing $(echo $i|cut -d"/" -f 2);java -Xmx1024m -jar baksmali.jar  -x $i -d BOOTCLASSPATH/; echo smali-ing $(echo $i|cut -d"/" -f 2);java -Xmx1024m -jar smali.jar -o deodexed/$(echo $i| cut -d "/" -f 2).dex out;done

Settting the phone number in an android emulator

So this is another short post to help answer a question that the Internet seems to faff around with:
  "How do I set the phone number in an Android emulator?"

This was important for me as I have been producing a proof of concept app that relies on using the phone number as a way of identifying you. Rather hard if all the emulators are on the same number.

So as a result it appears that the following information can be used:
 - The number of the phone ends in the four digits relating to the port number that the emulator is listening on. So for instance when it's run on port 5554:

To change, you can run ./emulator -avd -port
There doesn't seem to be a way to set the port when using Eclipse's AVD manager, but they will at least start on incrementing ports from 5554 onwards so no chance of a collision.

android adb - daemon still not runningerror: cannot connect to daemon

For people getting:

* daemon not running. starting it now on port 5037 *
* daemon started successfully *
** daemon still not runningerror: cannot connect to daemon

Check your iptables!
if you are whitelisting INPUT on local interface, you'll get this error. You need to allow at least port 5037 through on lo.

Or if you're happy with allowing all input on lo:

sudo iptables -I INPUT -i lo -j ACCEPT

hope that saves you the hour I just wasted :-)

Android webview security loadDataWithBaseURL

A quick lesson in web views in Android. Normally they are a big risky area, be it adding JavaScript interfaces (http://labs.mwrinfosecurity.com/blog/2012/04/30/building-android-javajavascript-bridges/), or just generally enabling javascript and opening yourself up to webkit exploits (which given how little the OEMs update their firmware, is not hard to do).

So another interesting bit to concern yourself with is found on the following page's code:
http://www.androidsnippets.com/webview-with-custom-html-and-local-images

The code is as follows:

/**
 * This code loads a custom HTML from a string which references a local image
 * - for this to work, simply place the image in the directory /assets/
 */
 
public void loadHTML() {
    final String mimeType = "text/html";
    final String encoding = "utf-8";
    final String html = "

Header
Custom HTML



";
    
    WebView wv = (WebView) findViewById(R.id.wv1);
    wv.loadDataWithBaseURL("fake://not/needed", html, mimeType, encoding, "");

The problem is with the loadDataWithBaseURL's first argument. This defines how relative links on the page should be dealt with. Normally setting it to null or "" is enough, although this obviously produces invalid links. A better way is to use "file://android_asset". The problem with "fake:" is that if any other app registers for this custom url scheme, they will receive the link and start one of their activities.

Not a bad issue, unless the url has sensitive data in it right? Well as a malware writer, this is could still be useful. Imagine if this screen is part of the Facebook app and has a relative link on it, or has JavaScript that changes the window.location to a relative page. All the malware needs to do is dress the pop up screen in nice Facebook attire and ask the user to re-enter their creds. It doesn't even require the malware app to request any permissions from the user.

To fix it, just set the baseurl to null, or include a custom webview client (setWebViewClient)

SQLite injection - beyond SELECT

I work a lot with mobile and that means whenever I see SQLinjection vulns, chances are that they're in SQLite. This is a massive pain because:
a) no-one really writes guides/tools for SQLite injection
b) It's really limited (as we'll see)

Unfortunately this post is about what is not possible in SQLite injection. Hopefully it will give you a guide to where to focus attacks/defence.

SQL injection as my previous post mentioned is a subset of regular SQL. There's no command execution, file reading/writing or anything particularly cool. You can't even stack queries unless the developer is using less orthodox functions (like rawquery). In essence it normally boils down to information disclosure (i.e. reading from tables that you shouldn't).

Select Statements in Mobile SQLite

With Android you (as the attacker) will hopefully be able to inject on 'projection' variable, that is the part that goes in the "SELECT ____ FROM". This obviously allows you to redirect the statement to go to anywhere else. The ideal test is to try projection=" * FROM SQLITE_MASTER; -- ". If you can get to the master table, you can get anywhere else (unless the programmer has some custom code ready to black list you out of having fun).
Failing this you could try on the 'selection' variable and try a union statement. (see my previous post for more details

This week has been spent looking at the other commands (insert,update,delete). The goal for the attacker is to change values in other tables in the .db file. The big issue here is the table to be queried is selected before your variables get included.

Inject into INSERT statements

insert into TABLENAME(field1,field2) values (val1,val2)
 
Here you can maybe control field1 & 2 and val1 & 2. The problem is that you can't jump tables. SQLite doesn't permit UNIONs or JOINs in anything other than select statements. You can use CASE to hide SELECT statments in the middle of this but that doesn't get you much in the way of altering data powers. You also can't use table.field in the query. Essentially you are stuck altering the data within that table.

It is therefore possible for SQL injection through inference of values being changed with CASE statements to read data from a table, but it's not possible to jump to and attack other tables.

 Inject into UPDATE or DELETE statements

UPDATE table SET (val1 = val2) WHERE condition
DELETE FROM table WHERE condition

Similarly the update query permits field names (val1), updated values (val2) and where clauses (condition), but no changing of the table name. The delete function is similarly limited.

I'm scratching my head finding a more useful attack vector that somehow nests more useful things inside statements. All comments are welcome. In the meantime, this should make for some useful reading: http://www.sqlite.org/syntaxdiagrams.html

Hacking Android dynamic Broadcast Receivers

An interesting point was raised this week when the subject of attacking IPC endpoints in Android applications was being discussed.
Attacking Endpoints is a method by which a malicious application on an Android device can communicate with, and misuse functions of other applications installed on the device. This can lead to anything from information disclosure to code execution.

To find these end points it's normally a matter of checking the application's manifest file (which can be recovered from the apk)

adb shell "pm list packages"
adb shell "pm path com.example.apptotest"
adb pull /data/app/com.example.apptotest-1.apk
aapt d xmltree com.example.apptotest-1.apk AndroidManifest.xml

This process (which can be sped up with mercury), gives you a whole list of text which allows you to identify which of these end points are exported either on purpose or otherwise, and which have permissions.

This gives us a list of things to look at in code, but I believe we are missing a step here. This assumes that these are the only end points we can talk to. Obviously some require permissions that we as a malicious app might have ourselves by just asking the user for it, but for this instance the malware will have zero permissions.

It is possible to register a broadcast receiver dynamically, typically by doing the following:
registerReceiver(BroadcastReceiver myreceiver, new IntentFilter("android.some.THING"))

This is fine, and often required to listen out for an event on a device. There are 2 things that can go badly wrong here though. First if the broadcast filter is listening for an intent that isn't a restricted broadcast then the malicious app can start sending them without needing any permissions. Secondly if the receiver goes on to use the extras sent with the broadcast, or to start running other functions, then the malware has a whole new entry point to your application. Of course it would require the malware to send the broadcast at the right moment, but hell just have it send the broadcast every second and run all the time.

To protect your broadcast, you need to register permissions when you declare it (details here), secondly don't trust the data that is being sent with it. It's passing through a trust zone to arrive in your app, so validate it! Details on Android permissions can be found here.

What is Android application licensing?

I ran into this permission this morning:

com.android.vending.CHECK_LICENSE

What it means is that the app will check with Google play (aka android marketplace) that the app was bought by the user that is now running it. The application holds the public key and uses this to check the "check_license" response from Google.

Is it possible to mess with this process? Well it would depend on where the public key is stored and how Google implements the process (including obviously security of transmission).

One thing that immediately stands out is this following piece of advice from Android themselves:

A typical implementation would extract some or all fields from the license response and store the data locally to a persistent store, such as through SharedPreferences storage, to ensure that the data is accessible across application invocations and device power cycles. For example, a Policy would maintain the timestamp of the last successful license check, the retry count, the license validity period, and similar information in a persistent store, rather than resetting the values each time the application is launched.

Sharedpreferences is only secured by using the sandbox model. This means any rooted devices would permit the user to alter these values simple by editing the .xml file.

What Google have done to allow developers to defend against this is to create AESObfuscator. This encrypts the data using the Android ID as the key. Obviously this is still recoverable by the user, but certainly makes it take a little longer to perform.

I can foresee two attacks coming out of this.
 #1 - Applications on 3rd party app stores stating they can "unlock" licensed applications. They would require a r00ted device to work and could do all sorts of nefarious things in the background. Because they would not be allowed on Google Play, they would have no checking and could contain malware.

#2 - Applications implementing their own obfuscation/ trusting license info input. Their is a long history of Android applications assuming the best about files in their own sandbox and implementing their own logic and functions when using the data contained. It allows attackers who can overcome the sandbox (say by running on a rooted device) to inject data that is then used in the context of the app.

Cross App Request Forgery in Android - CARF

An interesting thing about Android application security is that it is all very "What if?". Most scenarios I hear involve malicious apps. Malicious apps get installed by unsuspecting users and run in the guise of some innocent game whilst maliciously harvesting or installing other bits and pieces. The usual goal is to recover important data from another application or to use premium rate services or as botnets.
Normally the idea of malware specifically targeting other apps on a user's device would seem ludicrous but given the nature of some of the apps coming out on to the market place (sorry, Google Play), this seems more and more likely a vector in these more extreme cases.
This article has two points: The flaw in the root attack theory, and a potentially unthought of and certainly overlooked attack vector, the cross app request forgery (CARF)

Attacking an app with root.

I've often heard of a scenario in which malware elevates to root, and then attacks the flaws in the Manifest or file system of a specific app. Often android manifest writers overlook the fact that just because the action isn't exported it doesn't mean that it can't be run by other users. Root users may run all actions (including your precious providers), whether or not they are exported, unless they are explicitly not exported (exported=0x0).
But this is a bad argument for an attack vector. If you have root, why go after the app? You can access all the files, allowing you to grab the db files and everything else. If the db is encrypted, switch out the binary and trick the user into giving out the information you need to decrypt it. When you're done, switch the apps back, send off the data and disappear. This is far simpler to write (and portable to attack any app), than trying to pick apart and exploit specific vulnerabilities in others' apps.

Protecting against root?

When it comes to protecting against this vector there are three answers, keep the information off of the device (best), use the secure element, or encrypt information with a decent user entered password.
Keeping information off the phone means interacting with a server. Ideally the user sends up a hashed password, the server checks and responds with the information, allowing authentication and no chance for offline brute force attacks.
Secondly, the secure element is a purpose built secure vault with lock-out and no public bypasses. This is limited by having to partner with either a device (for an on-device secure element), or a network provider (if it's on the SIM).
Finally there's user passwords. These are used to encrypt the data on the file system. This is obviously limited as no one likes typing in long passwords on phones, and still vulnerable to social engineering.

CARF-ing

The inability of many developers to control cross app communications securely leads to some unusual attack vectors. Normally these vectors are too complicated for the minimal rewards achieved for the amount of work, but given the nature of more high risk apps turning up on the market, the rewards are increasing.
An app may export several actions but it's careful only show up the login screen to outside attackers. Unable to login, the vector is closed right?
Well maybe. I was rather surprised on this one. Several apps developers are assuming that this is game over for the attacker and stop the protection there. I assume the point of this is that the phone user is using one app, passes some info across to the secure app at which point there is a prompt for password, limiting what information gets passed to only that which the user has permitted.
The issue is that if the user is already logged in, they are not prompted for their password. So if a user session exists, the doors are open to any app sending through information in this window. This could be done by an app without root permissions, so far more likely to make it through app store checks.
Either the app could try to indirectly monitor for when the app is in use, such as reading the log files, or just blindly hit the app every 5 minutes.

Protection

 Firstly Exported=0x0 is your friend. Everything should be explicitly not exported unless absolutely required.
 Secondly, always require user interaction before doing anything. Even validating the input before asking the user to confirm it allows for foreign arguments to be processed in your app. One mistake in this area of code means a vulnerability that could be exploited. If the user is using the phone, having one simple confirmation pop up won't hurt user experience, and scuppers the attack.